Ovarian cancer is a heterogeneous disease displaying complex genomic alterations, and consequently, it has been difficult to determine the most relevant copy number alterations with the scale of studies to date. We obtained genome-wide copy number alteration (CNA) data from four different SNP array platforms, with a final data set of 398 ovarian tumours, mostly of the serous histological subtype. Frequent CNA aberrations targeted many thousands of genes. However, high-level amplicons and homozygous deletions enabled filtering of this list to the most relevant. The large data set enabled refinement of minimal regions and identification of rare amplicons such as at 1p34 and 20q11. We performed a novel co-occurrence analysis to assess cooperation and exclusivity of CNAs and analysed their relationship to patient outcome. Positive associations were identified between gains on 19 and 20q, gain of 20q and loss of X, and between several regions of loss, particularly 17q. We found weak correlations of CNA at genomic loci such as 19q12 with clinical outcome. We also assessed genomic instability measures and found a correlation of the number of higher amplitude gains with poorer overall survival. By assembling the largest collection of ovarian copy number data to date, we have been able to identify the most frequent aberrations and their interactions.
Citation: Gorringe KL, George J, Anglesio MS, Ramakrishna M, Etemadmoghadam D, Cowin P, et al. (2010) Copy Number Analysis Identifies Novel Interactions Between Genomic Loci in Ovarian Cancer. PLoS ONE 5(9): e11408. https://doi.org/10.1371/journal.pone.0011408
Editor: I. King Jordan, Georgia Institute of Technology, United States of America
Received: February 11, 2010; Accepted: April 16, 2010; Published: September 10, 2010
Copyright: © 2010 Gorringe et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: This work was supported by the National Health and Medical Research Council of Australia; the Victorian Breast Cancer Research Consortium (VBCRC), Australia; and the Department of Defense (DOD), United States of America. JG is supported by an Australian Postgraduate Award. MR is supported by a Cancer Council of Victoria Postgraduate Scholarship. This research was also supported by a Victorian Life Sciences Computation Initiative (VLSCI) grant on its Peak Computing Facility at the University of Melbourne and at Victorian Partnership for Advanced Computing (VPAC). Australian Ovarian Cancer Study (AOCS) was supported by the United States Army Medical Research and Materiel Command under DAMD17-01-1-0729, The Cancer Council Victoria, Queensland Cancer Fund, The Cancer Council New South Wales, The Cancer Council South Australia, The Cancer Foundation of Western Australia, The Cancer Council Tasmania, and the National Health and Medical Research Council of Australia (NHMRC). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Epithelial ovarian cancer (EOC) is one of the deadliest malignancies, with high recurrence and poor survival rates . The genetic aberrations observed in EOC are highly complex, comprising frequent aneuploidy and multiply rearranged chromosomes , . The heterogeneity of copy number alterations (CNA) observed in EOC has made it difficult for small studies to be able to accurately identify the true frequency of the less common CNAs or to reproducibly identify CNAs that correlate with clinical parameters. A small sample size also makes it difficult to identify CNAs that co-exist or are mutually exclusive, which is a prerequisite to identify any common pathways that may be deregulated in EOC through alterations in gene copy number. The paradigm for mutually exclusive aberrations targeting the same pathway was set in colorectal tumours for APC and CTNNB1 mutations  and extended in other examples such as exclusivity of BRAF and KRAS mutations . Conversely, other genetic aberrations are more frequently observed in the same tumour than would have been expected by chance, suggesting a co-operative effect, for example the significant association of 11q13 and 8p12 amplicons in breast cancer . In ovarian cancer, associations have been found between CCNE1 and 12p amplification , and between MYC and 20q amplification  by fluorescence in situ hybridisation. Few studies have examined co-operativity or complementation of CNA on a genome-wide basis. Losses at 4q and 18q were found to be associated in one study  but this was not replicated in a recent analysis , which identified 7 CNA associations and 6 anti-correlations.
The presence of high level gene amplifications in ovarian cancer has been observed for some time, however most studies have been underpowered in sample size  or genomic resolution ,  to accurately detect the frequency and target of these events. Similarly, few robust associations of CNA with clinical parameters such as survival have been identified , . The detection of these CNA is relevant not only to the identification of tumour subgroups and the pathways affected in the tumours, but also to the targeting of molecular therapies in ovarian cancer. In this study we have brought together a large cohort of single nucleotide polymorphism (SNP) mapping array data to robustly annotate CNAs in serous and endometrioid ovarian cancers in order to identify the genes targeted by these genetic events and how these correlate with clinical parameters. In addition, we have assessed the interaction of CNA by evaluating their associations and anti-associations.
Materials and Methods
Peter MacCallum Cancer Centre (PMCC) data set: Tissue samples and DNA extraction
All samples were collected with the patient's informed consent and the study was approved by all the participating hospital Human Research Ethics Committees. Patients with ovarian cancer were identified through four primary sources between 1992 and 2006: a) 53 at hospitals in Southampton, UK, b) 141 through the Australian Ovarian Cancer Study, including 20 from the Westmead Gynaecological Oncology Tissue Bank, c) 15 through the PMCC Tissue Bank (Melbourne, Australia) and d) 41 from Jikei University (Tokyo, Japan). Pathology review was conducted from either formalin fixed, paraffin embedded tissue and/or fresh-frozen sections adjacent to the tissue from which DNA was extracted (n = 141) or through examination of the original diagnostic pathology reports (n = 109) (Table 1, Table S1).
All tissue samples were collected as fresh frozen material. A representative haematoxylin and eosin stained section was assessed and samples with >80% epithelial cells were used directly for DNA extraction from the whole tissue. For the remainder, needle or laser dissection was performed using 10 µm sections to obtain high percentage tumour epithelial cell component. DNA was extracted as previously described , . Normal DNA extracted from blood lymphocytes was available for 106 patients.
The Cancer Genome Atlas (TCGA) data set: Tissue samples and DNA extraction
Samples were collected as fresh frozen material from hospitals in the USA (n = 163). Tumour samples were assessed to be >80% of epithelial cells prior to DNA extraction from the whole tissue, as outlined . Normal DNA extracted from blood lymphocytes was available for 161 patients. The results published here are in part based upon data generated by The Cancer Genome Atlas pilot project established by the NCI and NHGRI. Information about TCGA and the investigators and institutions who constitute the TCGA research network can be found at http://cancergenome.nih.gov.
Copy number arrays
Samples were processed as previously described for Affymetrix Mapping arrays a) n = 108 50 K XbaI , GSE 13813 b) n = 27 250 K StyI arrays c) n = 32 500 K arrays (250 K StyI and 250 K NspI, ) d) n = 83 SNP6.0 (1.8 M probe sets , , GSE19539). When available, matching normal DNA was also analysed on the same array platform and in the same batch. TCGA SNP6.0 CEL files for 163 samples were downloaded from the Data Portal (http://tcga-data.nci.nih.gov/tcga/homepage.htm).
Data pre-processing and analysis
All SNP Mapping arrays were first normalized using methods available in the R package “aroma.affymetrix” , including techniques to remove systematic biases introduced due to allelic cross talk, PCR fragment length bias and differences in GC content. DNA copy number was estimated probe set-wise by comparing the normalized signal from a tumour sample to data from normal lymphocyte DNA from the same patient, if available. On tumour samples for which matched normal tissue was not available, the average signal from all the normals generated in the same lab was used as reference. Quality control steps are described in Methods S1. Only the included samples are summarised in Table 1.
The circular binary segmentation method was used to segment the copy normalized data , . Any probe sets within a CNA that was present in >5% of normal samples were excluded from the tumour analysis prior to segmentation to remove common copy number polymorphisms (CNP). Segments with fewer than 10 probe sets (SNP6) or 5 probe sets (500 K) were merged with the adjacent segment of closest copy number as previous QPCR analysis suggested that aberrations represented by few probes on these platforms may not be reliable . In addition, we used Genomic Identification of Significant Targets in Cancer (GISTIC), which is a method that aggregates data over different tumours to try to differentiate between driver and passenger aberrations, combining prevalence and amplitude . This technique was performed using a web-based interface (http://genepattern.broadinstitute.org) with CNA thresholds of ±0.3, a minimum of 10 markers and a q-value threshold of 0.25.
For hierarchical clustering, all tumours were assessed for the presence (“1”) or absence (“0”) of each GISTIC peak alteration (n = 89), where any overlap was considered as presence. Hierarchical clustering using average Euclidean clustering of the samples (n = 398) was performed using Partek Genomics Suite v.6.4 (Partek Inc., St. Louis, MO).
Association between regions of aberrations
We undertook the analysis of association on the TCGA data set (for which we re-ran GISTIC) and then on the remaining samples. Two different methods were used to compute associations between regions of gain and loss. GISTIC results were summarized as a matrix X with tumours as rows and regions of aberrations as columns. For each tumour (t) and focal region of aberration (i), the measurement X[t,i] was 1 if the aberration was present for that tumour and 0 otherwise. A Poisson log-linear model was fit to the contingency table describing the aberration status. Statistical significance of the association was computed using a score test that yields a standard normal z-statistic . This is equivalent to the square-root of the usual Pearson test statistic for independence, signed according to the direction of the association. The Benjamini and Hochberg method was used to correct for multiple testing .
Association between regions of aberrations was also tested using the Monte Carlo permutation test. Briefly, every column in matrix X was permuted independently (maintaining the number of entries in the columns to be the same). A score for association was computed using the permuted matrix as described for the parametric test above. The average rank obtained for every pair of regions from a large number of permutations was used to estimate the false discovery rate and the number of times a test statistic greater than or above the original test statistic was used to compute the p-value. Using a 5% false discovery rate the methods selected >98% of the same pairs of regions. We chose to use the first method described for region selection but both are reported.
Analysis of expression correlations between associated copy number aberrations
We posited that the correlation between regions of aberrations should result in correlation of mRNA levels of the genes within the region. Affymetrix U133A array data was obtained for all samples from TCGA. For all associated regions above, four Pearson correlation tests were performed for the genes in the regions: a) correlation of copy number between Gene X in Region A and Gene Y in Region B, b) correlation between copy number and expression of Gene X in Region A, c) correlation between copy number and expression of Gene Y in Region B and d) correlation of expression between Gene X and Gene Y. All four tests had to be significant at p<0.05.
The Cox proportional hazard model was used to compute the association between regions of aberration detected by GISTIC and overall or progression free survival, correcting for multiple testing using the Benjamini-Hochberg method. To compute the survival association with pairs of regions, samples were classified into four groups based on the aberration status of the pairs of regions. Similarly, for the genomic measures, samples were binned into one of four groups based on data quartiles for each measure. Survival association with the groups thus identified was computed using the Cox proportional hazard model.
Integration of copy number alterations from 398 ovarian carcinomas
We compiled high resolution copy number data from nearly 400 ovarian cancer samples representing two histological subtypes, serous and endometrioid (Table 1), 270 of which had matching normal lymphocyte DNA data. Data was compiled from multiple sources: high quality Affymetrix SNP6.0 Mapping Array “CEL” files were sourced through The Cancer Genome Atlas (TCGA, 157 cases) or were obtained at the Peter MacCallum Cancer Centre (83 cases ) SNP Mapping array data derived from lower resolution Affymetrix platforms including 108 cases assayed on 50 K XbaI arrays , 27 cases on 500 K arrays  and 23 cases on 250 K StyI arrays obtained from Japan, were also included. Extensive quality control criteria were applied to all data sets (see Methods S1). Following normalisation of each data set, copy number alterations (CNA) were detected by circular binary segmentation . We evaluated a number of possibilities for combining the datasets including cohort-specific thresholds (see Methods S1), however this made little difference to the final CNA pattern and a standard threshold of +/− 0.3 (log2) was applied universally as previously described by us  and others .
Comparison between the five sources of data showed a remarkable consistency of CNA across the genome, indicating a high degree of non-randomness to the CNA and equally importantly, an absence of significant array batch effects (Figure S1). The exception was the Japanese data set, which appeared to show a reduced number of alterations. However, a genome-wide test was conducted to identify regions aberrant at different frequencies between different platforms and could not identify any statistically significant regions after multiple testing correction.
We assessed the possibility of molecular subgroups within the combined cohort defined by copy number using hierarchical clustering (Figure S1). Only a single group of samples was distinguishable; these had few CNAs and tended to be low-grade samples or the Japanese samples, for which grade information was mostly not available. There were no other distinct clusters or major groupings attributable to histological subtype or grade. In particular, the high grade serous and high grade endometrioid were evenly integrated, which is consistent with the previously observed similarity of these subtypes as assessed using immunohistochemical markers  and gene expression profiles .
In order to identify the most relevant CNAs we performed a number of complementary analyses as each method used has strengths and weaknesses that may be complemented by the other. Firstly, GISTIC was applied to all 240 SNP6 samples to identify “focal” and “broad” peaks (as defined in ) (Figure 1, Table S2). However, GISTIC cannot readily integrate samples from different platforms. We therefore elected to use a second complementary method to GISTIC: an overall frequency approach that would integrate segmented copy number data independent of platform for analysing our entire 398 sample cohort. As expected, the most significant regions of copy number gain predicted by both GISTIC and overall frequency were located on 3q (63% of samples with CN gain) and 8q (62% samples with CN gain) (Figure 1). Other frequent gains were observed on 20q (47%) and 12p (39%). The most frequent regions of loss identified in this study (chromosomes X, 8p, 22q, 17, 4q, 19p and 16, >40%) are consistent with previous studies by us  and others , . To select the most relevant genes, we firstly report those in regions of gain and loss with at least 30% frequency or in GISTIC peaks and then identified genes that were also targeted by higher amplitude events even if this was at a lower frequency (Table S2). Since there is no clear consensus on what constitutes a “high-level” amplification, we report regions with frequent gains at log2 ratios of >0.6 (in 40 or more samples,10%+), >0.8 (5%+) and >1 (2.5%+). For losses, we considered homozygous deletions (log2 ratios of <−1) present in at least 4 samples. The list of genes was prioritised taking into account the frequency of high-amplitude CNA and the overlap with GISTIC (Tables 2 and 3). Specific regions of gain are shown in Figures S2, S3, S4, S5, S6, and S7.
Gains (A) and losses (D) in 240 samples on SNP6 arrays analysed by GISTIC. Gains (B) and losses (C) in 398 samples on various array platforms. Sample segments were overlapped in Partek Genomics Suite v 6.4, creating a data point for each segment defined by copy number breakpoints, and then plotted by sample number.
In using this flexible approach we found that some regions were only clearly identified by one or the other method. By including a range of higher amplitude CN thresholds and the peaks predicted by GISTIC, additional regions were identified such as gains on chromosomes 1, 6p, 11q, 19 and losses on 5q, 6q26, 10q23, 13q and 18q22. In addition, on high resolution platforms such as the SNP6 array, GISTIC tended to identify very small regions, potentially missing relevant genes. For example, on 3q26 there were two closely spaced peaks of significance in the GISTIC profile (Figure S2). The highest of these, by a very narrow margin (–log q value 93.88 vs. 93.43), does not intersect with any genes, while the other peak overlaps with MECOM (MDS/EVI1); there is good evidence for this gene being an oncogene in ovarian cancer . Thus, relying on GISTIC alone would annotate the 3q26 region as having no genes of interest. In contrast, using a frequency approach, the maximum frequency at all copy number thresholds encompasses MECOM.
Similarly, there were other regions for which using a frequency approach missed genes or gave conflicting data. For example, on 19q12, each copy number threshold identified a slightly different region of peak frequency, variously identifying CCNE1, C19ORF2 or no gene in the peak (Figure S3). In contrast, the ability of GISTIC to integrate the amplitude of gain across all samples clearly identified CCNE1 as the gene in the peak. There is good evidence that CCNE1 is the correct call since Cyclin E is a key cell cycle protein and its amplification and over expression has been previously identified as a key driver of patient response to chemotherapy in serous ovarian carcinoma . Major conclusions arising from our analysis of individual deletions and amplicons, including insights into potential driver genes, are provided in the Discussion.
Associations between CN alterations
The concept of cooperative and mutually exclusive genetic alterations has rarely been examined at the level of CNAs or on a genome-wide scale. We wished to know whether there are any CNAs that cooperate in ovarian tumorigenesis, or that are functionally redundant to each other, for example if they act in the same pathway. To measure this we assessed if there were any CNAs that were more or less likely to be associated with each other, more than by chance, using a statistical analysis. Briefly, we counted the number of samples positive for CNA (e.g. a gain) at region A alone, region B alone, both regions and neither region, and compared the findings to the expected co-occurrence based on the total frequency of CNA at A multiplied by the frequency of B. For example, for a frequency of gain at 20q11 of 68/183 (37%) and at 19q12 of 50/183 (33%), we would expect 12% of samples to have both gains. However, we observe an actual frequency of samples with both alterations that is significantly different from this, i.e. 35/183 (19%, p<0.0001), indicating an increase in co-occurrence above the level of chance and thus possibly cooperating CNAs. The method can also be equally used to detect decreases in co-occurrence. When applying this method genome-wide, we applied a multiple testing correction with a FDR of <5%.
We undertook this analysis first using the TCGA data, as it is most homogenous for grade and subtype, and is high resolution. We repeated GISTIC analysis on this data set alone to obtain 46 peaks of copy number gain and 27 of loss (exclusive of regions of normal copy number variation, or copy number polymorphisms (CNPs)). Samples were identified as being positive or negative for each CNA peak, with gain peaks scored as positive for gains only and loss peaks scored as positive for losses only, and an analysis of association was performed as described in the methods. At a false discovery rate of 5%, 305 pairs of regions of aberration were positively correlated and 18 pairs were negatively correlated (Table S3, Figure 2). Some co-occurring GISTIC peaks were located within the same broad GISTIC region and although the GISTIC analysis indicated that these regions of copy number change were distinct, because they are physically closely linked they may not be independent of each other. As independence is necessary for the association test performed, they were not analysed further. We also excluded those associations in which either peak was a CNP, leaving 98 pairs of regions that were positively correlated, all but 16 of which were located on different chromosome arms (Table 4). 12 pairs of regions were negatively correlated.
(A) Process for identifying associated aberrations (more detail in Methods S1). (B) Summary of significant associations in each data set and those significant in both. As the table progresses, certain associations are filtered out, with the numbers remaining those that pass the filter. Firstly, associated loci that are within the same broad GISTIC intra-chromosomal region are removed and secondly regions that overlap with a CNP are removed. (C) Circos plot. Outer ring indicates the chromosome position of each aberration (coloured bars). The internal purple lines show the significant inter-chromosomal associations (exclusive of those involving a CNP) that have been validated in the second data set.
In order to validate the associations identified using TCGA data, we repeated the association analysis using the same “TCGA GISTIC-defined” regions as above on all other high grade serous and endometrioid samples (n = 183). For this data set, 296 regions were positively correlated and 5 were negatively correlated. Overall, 29 positive associations and no negative were in common between the two datasets (Figure 2). Of these, 14 were associations between two gains, 11 of which were on the same chromosome, and 14 associations were between two losses. None of the loss-loss associations were intra-chromosomal, because all associations of this type were excluded either for being located in the same broad GISTIC region or for being a CNP; indeed, more of the GISTIC peak losses were CNPs (n = 35) compared to the gains (n = 15) likely due to the unmasking effect loss of heterozygosity has on CNP detection in the tumour . There was a single association between a gain and a loss, between an amplicon on 20q11 and loss of Xq. The strongest positive association between gains on different chromosomes was for amplifications on chromosome 19q12 (most likely targeting CCNE1) and at 20q11 (five genes). For losses, the strongest common association was between chromosome 4q and chromosome 17. 17q12 loss was the most promiscuous interactor, with 8 common positive associations.
We identified the genes located in or near positively associated peaks and used gene expression data to evaluate whether any of the genes showed correlation between copy number and expression, and if there were correlation at the level of gene expression across regions (Table S4). We found that the strongest associations across regions involved genes gained on 19q12 or 19p13.11, and genes gained on 20q11. Other positive gene expression associations included CD47 (gained on 3q13.12) with UQCRFS1 or POP4 (both gained on 19q12). CD47 was first identified as an ovarian tumour antigen , however there is no known functional association with either 19q12 partner.
Correlation with clinical parameters and outcome
We used the TCGA clinical data to assess the relationship of copy number and patient outcome using a univariate Cox proportional hazards analysis on the GISTIC peaks (Table S5). Gain on 3q29 was associated with overall survival, however, this correlation was not significant after multiple testing correction. Positive CN associations of 17q12/22q losses and 3q13/19q12 gains were each correlated with overall survival but not progression free survival (Table S5).
Specific patterns of copy number change and genetic instability that correlate with patient outcome, including simplex, sawtooth and firestorm, have been described in breast cancer . The patterns of chromosome aberration in ovarian cancer are difficult to categorise into the groups described by Hicks et al. as most are a combination of sawtooth and firestorm. Therefore, we defined a number of different measures of genome instability and analysed their correlation with patient outcome using the TCGA data set (Table S5). These measures included: the number of copy number changes i.e. gains, losses, higher level gains (>0.6 log2 amplitude) and total number of segments; the percentage of the genome targeted by copy number change (gain, loss and high level gain); and a “Hicks index” as described  for gains, losses and both. The samples were divided into quartiles based on each of these indices and tested for association with clinical outcome using a univariate Cox proportional hazards analysis. Of these measures, only the number of higher amplitude gains (p = 0.019) showed a correlation with progression free survival but not overall survival (Figure S8). The percentage of the genome encompassed in higher level gains was not significant (p = 0.88), suggesting that it is not the proportion of DNA amplified but the number of amplification events that is most important.
Aneuploidy and cytogenetic aberrations have long been recognised as cancer hallmarks. In epithelial cancers, copy number alterations have been shown to be drivers of the cancer phenotype through amplification and over expression of oncogenes like ERBB2 and loss of tumour suppressors such as CDKN2A. Ovarian cancer is both heterogeneous and cytogenetically complex making it difficult to decipher the key genomic regions affected by CNA. Previous studies have generally been underpowered with respect to resolution and/or sample number, at most comprising around 100 cases , , . This study brings together a large collection of ovarian carcinomas profiled for copy number, which we have analysed using both GISTIC and frequency approaches to provide a definitive annotation of driver alterations. Key regions are summarised in Tables 2 and 3 while a more comprehensive catalogue, encompassing the union of both methods is given in Table S2. Because of the large numbers of genes and regions involved, it is not possible to address all in detail, however the regions mentioned below illustrate some of the insights derived from working with this large data set.
We elected to use complementary analytical approaches as each technique has its own strengths and weaknesses: a frequency approach for regions such as 3q26 was better able to identify the likely driver gene, MECOM, whereas for 19q12 the ability of GISTIC to integrate the magnitude of copy number gain for each sample identified CCNE1. Using a tiered frequency approach in concert with GISTIC provided a greater depth of understanding in complex regions for which there is no clear driver. Previous studies have identified an amplification on chromosome 11 in 18% of ovarian cancers, and have proposed that the target gene of this event is EMSY (C11ORF30) . In other cancer types, such as breast cancer, the peak amplification in this region may be different, targeting EMSY and/or CCND1 , . In the data presented here, the main amplicon does not appear to be targeting CCND1, which is >5 Mb outside the peak region (Figure S4). GISTIC identifies a peak encompassing four genes (THRSP, NDUFC2, ALG8 and KCTD21), amplification of which have been shown in breast cancer to correlate with over-expression and poor survival . The most frequently targeted gene by low-level gain is GAB2 (30%). At higher amplitude levels, the amplicon stretches from RSF1 to NARS2, encompassing 11 genes, not including EMSY or PAK1. However, EMSY appears to be targeted in a small proportion of samples (n = 13) by very high level amplification (log2>1), including in two samples that are not highly amplified at the more distal site. Similarly, there are two samples that target CCND1 with very high level amplification that do not cover either more distal region, however the overall frequency of CCND1 high level amplification is very small (n = 6 with log2 >1, 1.5%). The analysis undertaken here has thus found that this apparently simple amplification is in fact more complex, with 3 potentially overlapping targets.
There are some regions for which neither approach was able to identify the “obvious” oncogene. Surprisingly, on 8q24 both the GISTIC and frequency peaks were located ∼500 kb distally to the two previously implicated oncogenes in ovarian cancer within this locus, MYC and PVT1  (Figure S5). This region is distinct from the locus proximal to MYC on 8q24 that holds alleles conferring an increased predisposition to cancer . There are no genes in the distal peak region, so the reason for its high frequency of amplification is unclear. From UCSC genome information, this region does appear to contain many repeat and structural variation elements, which could potentially explain the peak shift as an artefact of the array measurement , . It is also possible that this region may harbour a long range enhancer for MYC and/or PVT1 or there may be an uncharacterised genetic element in this region that has an additional independent oncogenic effect to that of MYC or PVT1. This peak feature has not been previously appreciated in the somatic genetics of ovarian cancer, and its detection is a consequence of having a large number of samples at high resolution copy number coverage. Recent sequencing analysis of cancer cell lines has found that amplicons can have a highly complex structure at the base-pair level . It is possible that apparently distinct amplicons may have a common origin with subsequent deletion and rearrangement of internal material.
The strength of this study in having a large number of samples provides numerous additional insights to copy number features of ovarian cancer. Firstly, a number of studies in other cancer types have found that low frequency events present in only a few percent of cases may still be very important for the genesis of the case in which they are found, whether these are point mutations  or copy number changes . These types of events are difficult to detect without a large number of samples. Lower frequency but high-amplitude events that may be of relevance to ovarian cancer were readily detected here. The most notable of these as having the highest proportion of high-level amplicons (34/126 of samples with a gain had a log2 value of >1) was on 19q12, targeting CCNE1. The next highest was also on chromosome 19, at 19p13.1, with two apparent peaks (Figure S3). The peak identified by GISTIC contained the gene BRD4 and was the highest by low-level gain frequency. BRD4 appears a likely target, as it is involved in an oncogenic translocation in midline carcinoma . However, at higher amplitude gains, the frequency peak shifts more distal, for example the gene CACNA1A was targeted by the most number of very highly amplified samples on 19p (n = 14 at log2 >1). Again this illustrates why using a single method of gene prioritisation may miss potential oncogenes. GISTIC alone would have missed the second peak that may be important for a low frequency of high-amplitude gain, whereas a peak frequency approach using a single cut-off would also miss potentially important genes. Of course, neither method can unequivocally prove a role for any of the identified genes in ovarian cancer but do provide a rational basis on which to undertake comprehensive functional or nucleotide level mutation analysis.
With a large number of samples we were able to refine regions of gain on other chromosomes. For example, 20q has long been recognised as having a high frequency of gain, yet the proposed target gene(s) vary from study to study. This variability may be a stochastic consequence of low sample numbers in each study, or additionally because there are multiple genes each involved in only a sub-set of cancers or each with a mild oncogenic effect. Our data would support the view that there are multiple drivers at 20q. The most reproducible of the peaks observed between different thresholds and analyses were at approximately 29–30 Mb (20q11.21), 52 Mb (20q13.2) and 62 Mb (20q13.33) (Figure S6). Many samples (166/398) co-amplified more than one region, however, 5.3–6.5% of samples amplified only one of each of the above three regions. The 20q regions encompass previously noted genes including TPX2, BCAS1, ZNF217 and PTK6. Interestingly, some previously characterised oncogenes such as AURKA  were not in peak regions or frequently amplified at a high level.
On chromosome 12p, both GISTIC and the peak frequency of high level amplification clearly identified KRAS as the target for some samples (Figure S6). Interestingly, although activating point mutations in KRAS in ovarian cancer are observed predominantly in low grade serous subtype and rarely in high grade cases, of the 25 samples with an amplification at KRAS of at least 0.8 (log2) in amplitude, all but one were high-grade serous, suggesting an alternative mechanism for activation of this pathway in a subset of high grade serous carcinomas. Elsewhere on 12p, the main frequency peak for low-level gain was located at the distal end, for which oncogenes have not been previously characterised but may include ERC1, EFCAB4B and TSPAN9. ERC1 is a translocation partner of RET in AML .
Chromosome 1 appears to be targeted by several amplifications (Figure S7). There are two amplicons within 5 Mb of each other on 1q, which appear quite distinct, and yet share 85% of samples at a >0.3 (log2) level and 100% at an amplitude threshold of log2 ratio>1. It is possible that the apparent gap between these amplicons may be artefactual and a consequence of poor genome mapping and highly repetitive sequence in this region close to the centromere. There are several interesting genes within the amplicons, including BCL9 (B-cell CLL/lymphoma 9), which has recently been characterised as an oncogene in colon carcinoma and multiple myeloma, with a role in the Wnt pathway . Chromosome 1p is particularly interesting, as the amplicon in this region has not been previously noted in ovarian cancer, most likely because it is a highly focal, low frequency event (3% at log2 >1, 23% at log2 >0.3) that would likely only be detected with a large dataset of high-resolution arrays. Genes within this region include HEYL, BMP8B and MYCL1. HEYL, (Hairy/enhancer of split related with YRPN motif-like protein) is one of a group of transcription factors that act to inhibit gene expression and are thought to be involved in mediating Notch signalling, possibly with a role in epithelial-to-mesenchymal transition . HEYL was found to have increased expression associated with amplification in small cell lung cancer . BMP8B encodes a bone morphogenic protein, which are signalling peptides in the TGF-beta family . MYCL1 is a MYC oncogene homolog also identified as being amplified and overexpressed in small cell lung cancer .
The large data set collected here also enabled us to confirm previous observations that homozygous deletions are common in the genes CSMD1, DMD, NF1, RB1 and PTEN , . Interestingly, frequently co-deleted with DMD is the microRNA hsa-mir-548f-5, a member of a large family of microRNAs predicted to have regulatory functions relevant to cancer . We also note a high frequency of HDs in WWOX, and at several regions on chromosome X, including the novel targets STS (Steroid sulfatase, involved in estrogen synthesis), KLHL4 (Kelch-like 4, with actin binding homology) and SLITRK2 (SLIT and NTRK-like family, member 2, an integral membrane protein). Co-deleted with SLITRK2 were a group of 6 microRNA genes including hsa-mir-888 and hsa-mir-890 among others. None of these microRNAs have been characterised with validated targets.
Although the insights obtained from this data set are highly informative in helping to determine key regions and their likely drivers, the interpretation of biological relevance is restricted to individual gene information, which is often of limited value since it is possible to construe a cancer-related function for most genes. More direct relevance to cancer development may be ascertained through detailed functional analysis in model systems. However, rather than considering genes in isolation, what would be most interesting is to identify the genes that co-operate during tumorigenesis. One way of doing this is to undertake a pathway analysis to link the genes in various copy number aberrations. However, this analysis is generally “noisy” and likely to identify multiple interactions between the many genes affected by copy number change without necessarily picking the key pathways. An alternative method is to locate those CNAs that interact genetically. We first attempted to dissect this using unsupervised hierarchical clustering of copy number data, but were unable to identify separate genetic subgroups or interacting regions, apart from a group with few genetic changes that was biased towards low-grade serous and low-grade endometrioid samples. We undertook a different type of analysis by a novel bioinformatic method that identifies genomic regions that are more or less likely to be associated with each other than would be expected by chance, given their base frequency (Table 4). The method relies on having a large number of samples available, and with a data set as large as assembled here, we were able to perform the analysis on a test set (TCGA SNP6 data) and also a validation set (all other high grade serous and endometrioid samples). This highly stringent analysis identified a number of common positive associations, for example between amplification of CCNE1 and genes on 20q or 3q. A recent study performed a similar analysis in glioblastoma looking for positive associations only, and identified 21 validated inter-chromosomal interactions with aberration frequencies of >10% . In contrast we identified 29 interactions (18 inter-chromosomal). None of the interactions were in common, however, this is unsurprising given the very different genetic landscapes of the two tumour types, in particular the increased genetic complexity of ovarian carcinomas compared to glioblastomas. In comparison to previous ovarian cancer studies, we did not find previous associations such as 19q/12p gains , or any of 13 inter-chromosomal associations in . However, this is also not surprising as the first analysis used low-resolution 10 K SNP array data followed by fluorescence in situ hybridisation of selected loci, while the second performed the analysis on somewhat different GISTIC peaks in many fewer samples and used a Pearson correlation with a low-stringency FDR (0.25).
It is notable that there were many more positive associations identified compared with negative associations. A similar study in breast cancer failed to find any significant negative associations . Positive associations are assumed to identify genes whose activity may be in different but complementary pathways. Negative associations are thought to be redundant aberrations in the same pathway (e.g. BRAF and KRAS mutations). The differences in these association types may influence the process of their selection. To be selected for together above the level of chance, positive associations will act positively together in synergy. Thus, for one to be selected for there is a strong reliance on the other also being present, as each alone will provide little selective advantage, making their detection straightforward. In addition, for some associated CNAs, they may in fact be physically linked through chromosome translocation events and perforce selected together. In contrast, negative associations are acted on individually by selection – each provides a selective advantage when present first of all in the cell, but the eventuality of the second event occurring subsequently will not provide a selective disadvantage, and may be carried along as a passenger event. In the case of point mutations such as in BRAF and KRAS, the probability of one mutation occurring subsequent to the other is extremely small, so they are seldom seen together. However, for large-scale copy number alterations, the cytogenetic instability of ovarian cancer means it may be inevitable that a second event will occur by chance in some cases and be carried along as a passenger. Thus the power to detect negative associations in the current analysis may be low.
Finally, we used the clinical annotations available with the TCGA data to assess the correlation of various copy number features with survival. We identified a number of correlations, including of the number of higher amplitude gains with worse overall survival, however these were generally weak and did not pass multiple testing. It was somewhat concerning that no CNAs were strongly associated with survival, given that several recent studies have identified associations with poor outcome for amplification at genes such as CCNE1 , . One explanation could be that the TCGA sample set is not selected in order to target specific clinical questions compared with other studies. Another possibility is that within this generally high grade serous ovarian cancer cohort, the overall consistency of the common aberrations may mean that there is little “dynamic range” to detect differences associated with copy number, and that variance in outcome may have more to do with constitutive factors specific to the individual such as their immune system or pharmacogenomics.
In this study we have brought together a very large cohort of ovarian cancer cases profiled using high-resolution SNP Mapping arrays to identify the genome-wide frequency of copy number aberrations and their interactions. In addition to the well known frequent events, using this large data set has developed our view of the rare but recurrent events contributing to ovarian cancer, consistent with the “mountains and hills” analogy proposed for other epithelial cancer types . We identified a number of co-occurring aberrations that may co-operate in ovarian tumourigenesis, such as gain of 19q with 20q or 3q, gain of 20q with X loss and loss of 17q with multiple partners. The significance of these aberrations and the validation of the genes underlying the copy number changes remains to be assessed through comprehensive functional and nucleotide level sequence analysis.
Comparison of samples run on different array platforms. A. Overall frequency plot of gains and losses for SNP6 TCGA (n = 157), SNP6 PMCC (n = 83), 500 K PMCC (n = 27), 250 K Japanese (n = 23) and 50 K PMCC (n = 108). All platforms used the same log2 threshold of ± 0.3. B. Hierarchical clustering of samples. Samples were scored as positive (red) or negative (blue) for gains and losses identified in all SNP6 samples by GISTIC. Sample source, histological subtype and grade are indicated in colour at the top. There is no apparent grouping by array source that would suggest a batch effect of the arrays, apart from the Japanese 250 K samples (grade unknown), which tend to have few alterations and cluster with the low-grade endometrioid samples at the left-hand side of the dendrogram.
(0.06 MB PDF)
Gain on 3q. A. Frequency of gain on 3q is shown at various amplitude thresholds. Note the different scales for each threshold. The GISTIC -log q value is plotted above, as is the extent of the broad GISTIC region identified (red bar). GISTIC peaks are indicated by arrows. B. Zoomed in view of 3q26. Each of the minimal peak frequency regions for each CN amplitude is shown by a coloured box. The GISTIC peak is indicated by the red box. Below are shown the genes from the UCSC genome browser.
(0.05 MB PDF)
Gain on chr19. Frequency of gain on chr19 is shown at various amplitude thresholds. Note the different scales for each threshold. The GISTIC -log q value is plotted above, as is the extent of the broad GISTIC region identified (red bar). GISTIC peaks are indicated by arrows. Zoomed in views of 19p13 (B) and 19q12 (C) Each of the minimal peak frequency regions for each CN amplitude is shown by a coloured box. The GISTIC peak is indicated by the red box. Below are shown the genes from the UCSC genome browser.
(0.07 MB PDF)
Gain on 11q. A. Frequency of gain on 11q is shown at various amplitude thresholds. Note the different scales for each threshold. The GISTIC -log q value is plotted above; GISTIC peaks are indicated by arrows. Asterisks show 3 possible amplicons B. Zoomed in view of 11q13–14. Each of the minimal peak frequency regions for each CN amplitude is shown by a coloured box. The GISTIC peak is indicated by the red box. Below are shown the genes from the UCSC genome browser.
(0.05 MB PDF)
Gain on 8q. A. Frequency of gain on 8q is shown at various amplitude thresholds. Note the different scales for each threshold. The GISTIC -log q value is plotted above, as is the extent of the broad GISTIC region identified (red bar). GISTIC peaks are indicated by arrows. B. Zoomed in view of 8q24. Each of the minimal peak frequency regions for each CN amplitude is shown by a coloured box. The GISTIC peak is indicated by the red box. Below are shown the genes from the UCSC genome browser.
(0.05 MB PDF)
Gains on chr20 and chr12. A. Frequency of gain on chr20 is shown at various amplitude thresholds. Note the different scales for each threshold. The GISTIC -log q value is plotted above, as is the extent of the broad GISTIC region identified (red bar). GISTIC peaks are indicated by arrows. Zoomed in views of 20q11 (B), 20q13.2 (C) and 20q13.33 (D). Each of the minimal peak frequency regions for each CN amplitude is shown by a coloured box. The GISTIC peak is indicated by the red box. Below are shown the genes from the UCSC genome browser. E. Gain on 12p. Frequency of gain on chr12 is shown at various amplitude thresholds. Note the different scales for each threshold. The GISTIC -log q value is plotted above, as is the extent of the broad GISTIC region identified (red bar). GISTIC peaks are indicated by arrows. F. Zoomed in view of 12p, with various genes indicated.
(0.10 MB PDF)
Gain on chr1. A. Frequency of gain on chr1 is shown at various amplitude thresholds. Note the different scales for each threshold. The GISTIC -log q value is plotted above, as is the extent of the broad GISTIC region identified (red bar). GISTIC peaks are indicated by arrows. Zoomed in views of 1p34 (B) and 1q21 (C). Each of the minimal peak frequency regions for each CN amplitude is shown by a coloured box. The GISTIC peak is indicated by the red box. Below are shown the genes from the UCSC genome browser.
(0.07 MB PDF)
Survival analysis. A. Kaplan Meier plot of overall survival with samples divided into quartiles based on the number of gains >0.6 (log2). P = 0.045 after a Cox proportional hazard model analysis. 1, 0–18 segments; 2, 19–36 segments; 3, 37–60 segments; 4, >60 segments. B. Kaplan Meier plot of overall survival with residual macroscopic disease as a factor.
(0.08 MB PDF)
Full list of frequent aberrations and genes.
(0.71 MB XLS)
Full list of associated aberrations.
(2.82 MB XLS)
Expression of genes within associated aberrations.
(0.04 MB XLS)
We would like to acknowledge the contribution of the study nurses and research assistants and would like to thank all of the women who participated in the study. Collaborators and hospitals involved in AOCS can be found at http://www.aocstudy.org/. Data was obtained from the Westmead Gynaecological Oncology Tissue Bank, a member bank of the Australasian Biospecimens Network - Oncology. We would also like to thank the Peak Computing Facility provided by Victorian Life Sciences Computation Initiative (VLSCI) and Victorian Partnership for Advance Computing (VPAC).
Conceived and designed the experiments: KLG JG MSA MR DE PC GKS IC DDB. Performed the experiments: KLG MSA MR DE PC AS LHW SEB. Analyzed the data: KLG JG MSA. Contributed reagents/materials/analysis tools: NY AO MU AOCS. Wrote the paper: KLG JG MSA GKS IC DDB.
- 1. Heintz AP, Odicino F, Maisonneuve P, Quinn MA, Benedet JL, et al. (2006) Carcinoma of the ovary. FIGO 6th Annual Report on the Results of Treatment in Gynecological Cancer. Int J Gynaecol Obstet 95: Suppl 1S161–192.AP HeintzF. OdicinoP. MaisonneuveMA QuinnJL Benedet2006Carcinoma of the ovary. FIGO 6th Annual Report on the Results of Treatment in Gynecological Cancer.Int J Gynaecol Obstet95Suppl 1S161192
- 2. Bayani J, Brenton JD, Macgregor PF, Beheshti B, Albert M, et al. (2002) Parallel analysis of sporadic primary ovarian carcinomas by spectral karyotyping, comparative genomic hybridization, and expression microarrays. Cancer Res 62: 3466–3476.J. BayaniJD BrentonPF MacgregorB. BeheshtiM. Albert2002Parallel analysis of sporadic primary ovarian carcinomas by spectral karyotyping, comparative genomic hybridization, and expression microarrays.Cancer Res6234663476
- 3. Hoglund M, Gisselsson D, Hansen GB, Sall T, Mitelman F (2003) Ovarian carcinoma develops through multiple modes of chromosomal evolution. Cancer Res 63: 3378–3385.M. HoglundD. GisselssonGB HansenT. SallF. Mitelman2003Ovarian carcinoma develops through multiple modes of chromosomal evolution.Cancer Res6333783385
- 4. Sparks AB, Morin PJ, Vogelstein B, Kinzler KW (1998) Mutational analysis of the APC/beta-catenin/Tcf pathway in colorectal cancer. Cancer Res 58: 1130–1134.AB SparksPJ MorinB. VogelsteinKW Kinzler1998Mutational analysis of the APC/beta-catenin/Tcf pathway in colorectal cancer.Cancer Res5811301134
- 5. Rajagopalan H, Bardelli A, Lengauer C, Kinzler KW, Vogelstein B, et al. (2002) Tumorigenesis: RAF/RAS oncogenes and mismatch-repair status. Nature 418: 934.H. RajagopalanA. BardelliC. LengauerKW KinzlerB. Vogelstein2002Tumorigenesis: RAF/RAS oncogenes and mismatch-repair status.Nature418934
- 6. Courjal F, Cuny M, Simony-Lafontaine J, Louason G, Speiser P, et al. (1997) Mapping of DNA amplifications at 15 chromosomal localizations in 1875 breast tumors: definition of phenotypic groups. Cancer Res 57: 4360–4367.F. CourjalM. CunyJ. Simony-LafontaineG. LouasonP. Speiser1997Mapping of DNA amplifications at 15 chromosomal localizations in 1875 breast tumors: definition of phenotypic groups.Cancer Res5743604367
- 7. Nakayama K, Nakayama N, Jinawath N, Salani R, Kurman RJ, et al. (2007) Amplicon profiles in ovarian serous carcinomas. Int J Cancer 120: 2613–2617.K. NakayamaN. NakayamaN. JinawathR. SalaniRJ Kurman2007Amplicon profiles in ovarian serous carcinomas.Int J Cancer12026132617
- 8. Dimova I, Raicheva S, Dimitrov R, Doganov N, Toncheva D (2009) Coexistence of copy number increases of c-Myc, ZNF217, CCND1, ErbB1 and ErbB2 in ovarian cancers. Onkologie 32: 405–410.I. DimovaS. RaichevaR. DimitrovN. DoganovD. Toncheva2009Coexistence of copy number increases of c-Myc, ZNF217, CCND1, ErbB1 and ErbB2 in ovarian cancers.Onkologie32405410
- 9. Lambros MB, Fiegler H, Jones A, Gorman P, Roylance RR, et al. (2005) Analysis of ovarian cancer cell lines using array-based comparative genomic hybridization. J Pathol 205: 29–40.MB LambrosH. FieglerA. JonesP. GormanRR Roylance2005Analysis of ovarian cancer cell lines using array-based comparative genomic hybridization.J Pathol2052940
- 10. Haverty PM, Hon LS, Kaminker JS, Chant J, Zhang Z (2009) High-resolution analysis of copy number alterations and associated expression changes in ovarian tumors. BMC Med Genomics 2: 21.PM HavertyLS HonJS KaminkerJ. ChantZ. Zhang2009High-resolution analysis of copy number alterations and associated expression changes in ovarian tumors.BMC Med Genomics221
- 11. Kiechle M, Jacobsen A, Schwarz-Boeger U, Hedderich J, Pfisterer J, et al. (2001) Comparative genomic hybridization detects genetic imbalances in primary ovarian carcinomas as correlated with grade of differentiation. Cancer 91: 534–540.M. KiechleA. JacobsenU. Schwarz-BoegerJ. HedderichJ. Pfisterer2001Comparative genomic hybridization detects genetic imbalances in primary ovarian carcinomas as correlated with grade of differentiation.Cancer91534540
- 12. Partheen K, Levan K, Osterberg L, Helou K, Horvath G (2004) Analysis of cytogenetic alterations in stage III serous ovarian adenocarcinoma reveals a heterogeneous group regarding survival, surgical outcome, and substage. Genes Chromosomes Cancer 40: 342–348.K. PartheenK. LevanL. OsterbergK. HelouG. Horvath2004Analysis of cytogenetic alterations in stage III serous ovarian adenocarcinoma reveals a heterogeneous group regarding survival, surgical outcome, and substage.Genes Chromosomes Cancer40342348
- 13. Birrer MJ, Johnson ME, Hao K, Wong KK, Park DC, et al. (2007) Whole genome oligonucleotide-based array comparative genomic hybridization analysis identified fibroblast growth factor 1 as a prognostic marker for advanced-stage serous ovarian adenocarcinomas. J Clin Oncol 25: 2281–2287.MJ BirrerME JohnsonK. HaoKK WongDC Park2007Whole genome oligonucleotide-based array comparative genomic hybridization analysis identified fibroblast growth factor 1 as a prognostic marker for advanced-stage serous ovarian adenocarcinomas.J Clin Oncol2522812287
- 14. Etemadmoghadam D, deFazio A, Beroukhim R, Mermel C, George J, et al. (2009) Integrated genome-wide DNA copy number and expression analysis identifies distinct mechanisms of primary chemoresistance in ovarian carcinomas. Clin Cancer Res 15: 1417–1427.D. EtemadmoghadamA. deFazioR. BeroukhimC. MermelJ. George2009Integrated genome-wide DNA copy number and expression analysis identifies distinct mechanisms of primary chemoresistance in ovarian carcinomas.Clin Cancer Res1514171427
- 15. Gorringe KL, Ramakrishna M, Williams LH, Sridhar A, Boyle SE, et al. (2009) Are there any more ovarian tumor suppressor genes? A new perspective using ultra high-resolution copy number and loss of heterozygosity analysis. Genes Chromosomes Cancer 48: 931–942.KL GorringeM. RamakrishnaLH WilliamsA. SridharSE Boyle2009Are there any more ovarian tumor suppressor genes? A new perspective using ultra high-resolution copy number and loss of heterozygosity analysis.Genes Chromosomes Cancer48931942
- 16. The Cancer Genome Atlas Research Network (2008) Comprehensive genomic characterization defines human glioblastoma genes and core pathways. Nature 455: 1061–1068.The Cancer Genome Atlas Research Network2008Comprehensive genomic characterization defines human glioblastoma genes and core pathways.Nature45510611068
- 17. Gorringe KL, Jacobs S, Thompson ER, Sridhar A, Qiu W, et al. (2007) High-resolution single nucleotide polymorphism array analysis of epithelial ovarian cancer reveals numerous microdeletions and amplifications. Clin Cancer Res 13: 4731–4739.KL GorringeS. JacobsER ThompsonA. SridharW. Qiu2007High-resolution single nucleotide polymorphism array analysis of epithelial ovarian cancer reveals numerous microdeletions and amplifications.Clin Cancer Res1347314739
- 18. Ramakrishna M, Williams LH, Boyle SE, Bearfoot JL, Sridhar A, et al. (2010) Identification of candidate growth promoting genes in ovarian cancer through integrated copy number and expression analysis. PLoS One 5: e9983.M. RamakrishnaLH WilliamsSE BoyleJL BearfootA. Sridhar2010Identification of candidate growth promoting genes in ovarian cancer through integrated copy number and expression analysis.PLoS One5e9983
- 19. Bengtsson H, Irizarry R, Carvalho B, Speed TP (2008) Estimation and assessment of raw copy numbers at the single locus level. Bioinformatics 24: 759–767.H. BengtssonR. IrizarryB. CarvalhoTP Speed2008Estimation and assessment of raw copy numbers at the single locus level.Bioinformatics24759767
- 20. Olshen AB, Venkatraman ES, Lucito R, Wigler M (2004) Circular binary segmentation for the analysis of array-based DNA copy number data. Biostatistics 5: 557–572.AB OlshenES VenkatramanR. LucitoM. Wigler2004Circular binary segmentation for the analysis of array-based DNA copy number data.Biostatistics5557572
- 21. Venkatraman ES, Olshen AB (2007) A faster circular binary segmentation algorithm for the analysis of array CGH data. Bioinformatics 23: 657–663.ES VenkatramanAB Olshen2007A faster circular binary segmentation algorithm for the analysis of array CGH data.Bioinformatics23657663
- 22. Beroukhim R, Getz G, Nghiemphu L, Barretina J, Hsueh T, et al. (2007) Assessing the significance of chromosomal aberrations in cancer: methodology and application to glioma. Proc Natl Acad Sci U S A 104: 20007–20012.R. BeroukhimG. GetzL. NghiemphuJ. BarretinaT. Hsueh2007Assessing the significance of chromosomal aberrations in cancer: methodology and application to glioma.Proc Natl Acad Sci U S A1042000720012
- 23. Smyth GK (2003) Pearson's goodness of fit statistic as a score test statistic. IMS Lecture Notes- Monograph Series, Institute of Mathematical Statistics, Beachwood, Ohio 40: 115–126.GK Smyth2003Pearson's goodness of fit statistic as a score test statistic.IMS Lecture Notes- Monograph Series, Institute of Mathematical Statistics, Beachwood, Ohio40115126
- 24. Benjamini Y, Hochberg Y (1995) Controlling the false discovery rate - a practical and powerful approach to multiple testing. Journal of the Royal Statistical Society Series B-Methodological 57: 289–300.Y. BenjaminiY. Hochberg1995Controlling the false discovery rate - a practical and powerful approach to multiple testing.Journal of the Royal Statistical Society Series B-Methodological57289300
- 25. Gilks CB, Ionescu DN, Kalloger SE, Kobel M, Irving J, et al. (2008) Tumor cell type can be reproducibly diagnosed and is of independent prognostic significance in patients with maximally debulked ovarian carcinoma. Hum Pathol 39: 1239–1251.CB GilksDN IonescuSE KallogerM. KobelJ. Irving2008Tumor cell type can be reproducibly diagnosed and is of independent prognostic significance in patients with maximally debulked ovarian carcinoma.Hum Pathol3912391251
- 26. Tothill RW, Tinker AV, George J, Brown R, Fox SB, et al. (2008) Novel molecular subtypes of serous and endometrioid ovarian cancer linked to clinical outcome. Clin Cancer Res 14: 5198–5208.RW TothillAV TinkerJ. GeorgeR. BrownSB Fox2008Novel molecular subtypes of serous and endometrioid ovarian cancer linked to clinical outcome.Clin Cancer Res1451985208
- 27. Baudis M, Cleary ML (2001) Progenetix.net: an online repository for molecular cytogenetic aberration data. Bioinformatics 17: 1228–1229.M. BaudisML Cleary2001Progenetix.net: an online repository for molecular cytogenetic aberration data.Bioinformatics1712281229
- 28. Nanjundan M, Nakayama Y, Cheng KW, Lahad J, Liu J, et al. (2007) Amplification of MDS1/EVI1 and EVI1, located in the 3q26.2 amplicon, is associated with favorable patient prognosis in ovarian cancer. Cancer Res 67: 3074–3084.M. NanjundanY. NakayamaKW ChengJ. LahadJ. Liu2007Amplification of MDS1/EVI1 and EVI1, located in the 3q26.2 amplicon, is associated with favorable patient prognosis in ovarian cancer.Cancer Res6730743084
- 29. Gorringe KL, Campbell IG (2008) High-resolution copy number arrays in cancer and the problem of normal genome copy number variation. Genes Chromosomes Cancer 47: 933–938.KL GorringeIG Campbell2008High-resolution copy number arrays in cancer and the problem of normal genome copy number variation.Genes Chromosomes Cancer47933938
- 30. Campbell IG, Freemont PS, Foulkes W, Trowsdale J (1992) An ovarian tumor marker with homology to vaccinia virus contains an IgV-like region and multiple transmembrane domains. Cancer Res 52: 5416–5420.IG CampbellPS FreemontW. FoulkesJ. Trowsdale1992An ovarian tumor marker with homology to vaccinia virus contains an IgV-like region and multiple transmembrane domains.Cancer Res5254165420
- 31. Hicks J, Krasnitz A, Lakshmi B, Navin NE, Riggs M, et al. (2006) Novel patterns of genome rearrangement and their association with survival in breast cancer. Genome Res 16: 1465–1479.J. HicksA. KrasnitzB. LakshmiNE NavinM. Riggs2006Novel patterns of genome rearrangement and their association with survival in breast cancer.Genome Res1614651479
- 32. Brown LA, Irving J, Parker R, Kim H, Press JZ, et al. (2006) Amplification of EMSY, a novel oncogene on 11q13, in high grade ovarian surface epithelial carcinomas. Gynecol Oncol 100: 264–270.LA BrownJ. IrvingR. ParkerH. KimJZ Press2006Amplification of EMSY, a novel oncogene on 11q13, in high grade ovarian surface epithelial carcinomas.Gynecol Oncol100264270
- 33. Hughes-Davies L, Huntsman D, Ruas M, Fuks F, Bye J, et al. (2003) EMSY links the BRCA2 pathway to sporadic breast and ovarian cancer. Cell 115: 523–535.L. Hughes-DaviesD. HuntsmanM. RuasF. FuksJ. Bye2003EMSY links the BRCA2 pathway to sporadic breast and ovarian cancer.Cell115523535
- 34. Kadota M, Sato M, Duncan B, Ooshima A, Yang HH, et al. (2009) Identification of novel gene amplifications in breast cancer and coexistence of gene amplification with an activating mutation of PIK3CA. Cancer Res 69: 7357–7365.M. KadotaM. SatoB. DuncanA. OoshimaHH Yang2009Identification of novel gene amplifications in breast cancer and coexistence of gene amplification with an activating mutation of PIK3CA.Cancer Res6973577365
- 35. Chin SF, Teschendorff AE, Marioni JC, Wang Y, Barbosa-Morais NL, et al. (2007) High-resolution aCGH and expression profiling identifies a novel genomic subtype of ER negative breast cancer. Genome Biol 8: R215.SF ChinAE TeschendorffJC MarioniY. WangNL Barbosa-Morais2007High-resolution aCGH and expression profiling identifies a novel genomic subtype of ER negative breast cancer.Genome Biol8R215
- 36. Guan Y, Kuo WL, Stilwell JL, Takano H, Lapuk AV, et al. (2007) Amplification of PVT1 contributes to the pathophysiology of ovarian and breast cancer. Clin Cancer Res 13: 5745–5755.Y. GuanWL KuoJL StilwellH. TakanoAV Lapuk2007Amplification of PVT1 contributes to the pathophysiology of ovarian and breast cancer.Clin Cancer Res1357455755
- 37. Haiman CA, Le Marchand L, Yamamato J, Stram DO, Sheng X, et al. (2007) A common genetic risk factor for colorectal and prostate cancer. Nat Genet 39: 954–956.CA HaimanL. Le MarchandJ. YamamatoDO StramX. Sheng2007A common genetic risk factor for colorectal and prostate cancer.Nat Genet39954956
- 38. Stallings RL, Nair P, Maris JM, Catchpoole D, McDermott M, et al. (2006) High-resolution analysis of chromosomal breakpoints and genomic instability identifies PTPRD as a candidate tumor suppressor gene in neuroblastoma. Cancer Res 66: 3673–3680.RL StallingsP. NairJM MarisD. CatchpooleM. McDermott2006High-resolution analysis of chromosomal breakpoints and genomic instability identifies PTPRD as a candidate tumor suppressor gene in neuroblastoma.Cancer Res6636733680
- 39. Campbell PJ, Stephens PJ, Pleasance ED, O'Meara S, Li H, et al. (2008) Identification of somatically acquired rearrangements in cancer using genome-wide massively parallel paired-end sequencing. Nat Genet 40: 722–729.PJ CampbellPJ StephensED PleasanceS. O'MearaH. Li2008Identification of somatically acquired rearrangements in cancer using genome-wide massively parallel paired-end sequencing.Nat Genet40722729
- 40. Sjoblom T, Jones S, Wood LD, Parsons DW, Lin J, et al. (2006) The consensus coding sequences of human breast and colorectal cancers. Science 314: 268–274.T. SjoblomS. JonesLD WoodDW ParsonsJ. Lin2006The consensus coding sequences of human breast and colorectal cancers.Science314268274
- 41. Weir BA, Woo MS, Getz G, Perner S, Ding L, et al. (2007) Characterizing the cancer genome in lung adenocarcinoma. Nature 450: 893–898.BA WeirMS WooG. GetzS. PernerL. Ding2007Characterizing the cancer genome in lung adenocarcinoma.Nature450893898
- 42. French CA, Miyoshi I, Aster JC, Kubonishi I, Kroll TG, et al. (2001) BRD4 bromodomain gene rearrangement in aggressive carcinoma with translocation t(15;19). Am J Pathol 159: 1987–1992.CA FrenchI. MiyoshiJC AsterI. KubonishiTG Kroll2001BRD4 bromodomain gene rearrangement in aggressive carcinoma with translocation t(15;19).Am J Pathol15919871992
- 43. Watanabe T, Imoto I, Katahira T, Hirasawa A, Ishiwata I, et al. (2002) Differentially regulated genes as putative targets of amplifications at 20q in ovarian cancers. Jpn J Cancer Res 93: 1114–1122.T. WatanabeI. ImotoT. KatahiraA. HirasawaI. Ishiwata2002Differentially regulated genes as putative targets of amplifications at 20q in ovarian cancers.Jpn J Cancer Res9311141122
- 44. Gorello P, La Starza R, Brandimarte L, Trisolini SM, Pierini V, et al. (2008) A PDGFRB-positive acute myeloid malignancy with a new t(5;12)(q33;p13.3) involving the ERC1 gene. Leukemia 22: 216–218.P. GorelloR. La StarzaL. BrandimarteSM TrisoliniV. Pierini2008A PDGFRB-positive acute myeloid malignancy with a new t(5;12)(q33;p13.3) involving the ERC1 gene.Leukemia22216218
- 45. Mani M, Carrasco DE, Zhang Y, Takada K, Gatt ME, et al. (2009) BCL9 promotes tumor progression by conferring enhanced proliferative, metastatic, and angiogenic properties to cancer cells. Cancer Res 69: 7577–7586.M. ManiDE CarrascoY. ZhangK. TakadaME Gatt2009BCL9 promotes tumor progression by conferring enhanced proliferative, metastatic, and angiogenic properties to cancer cells.Cancer Res6975777586
- 46. Fischer A, Steidl C, Wagner TU, Lang E, Jakob PM, et al. (2007) Combined loss of Hey1 and HeyL causes congenital heart defects because of impaired epithelial to mesenchymal transition. Circ Res 100: 856–863.A. FischerC. SteidlTU WagnerE. LangPM Jakob2007Combined loss of Hey1 and HeyL causes congenital heart defects because of impaired epithelial to mesenchymal transition.Circ Res100856863
- 47. Henderson LJ, Coe BP, Lee EH, Girard L, Gazdar AF, et al. (2005) Genomic and gene expression profiling of minute alterations of chromosome arm 1p in small-cell lung carcinoma cells. Br J Cancer 92: 1553–1560.LJ HendersonBP CoeEH LeeL. GirardAF Gazdar2005Genomic and gene expression profiling of minute alterations of chromosome arm 1p in small-cell lung carcinoma cells.Br J Cancer9215531560
- 48. Zhao GQ, Deng K, Labosky PA, Liaw L, Hogan BL (1996) The gene encoding bone morphogenetic protein 8B is required for the initiation and maintenance of spermatogenesis in the mouse. Genes Dev 10: 1657–1669.GQ ZhaoK. DengPA LaboskyL. LiawBL Hogan1996The gene encoding bone morphogenetic protein 8B is required for the initiation and maintenance of spermatogenesis in the mouse.Genes Dev1016571669
- 49. Nau MM, Brooks BJ, Battey J, Sausville E, Gazdar AF, et al. (1985) L-myc, a new myc-related gene amplified and expressed in human small cell lung cancer. Nature 318: 69–73.MM NauBJ BrooksJ. BatteyE. SausvilleAF Gazdar1985L-myc, a new myc-related gene amplified and expressed in human small cell lung cancer.Nature3186973
- 50. Sangha N, Wu R, Kuick R, Powers S, Mu D, et al. (2008) Neurofibromin 1 (NF1) defects are common in human ovarian serous carcinomas and co-occur with TP53 mutations. Neoplasia 10: 1372: 1362–1372, following.N. SanghaR. WuR. KuickS. PowersD. Mu2008Neurofibromin 1 (NF1) defects are common in human ovarian serous carcinomas and co-occur with TP53 mutations.Neoplasia10137213621372, following
- 51. Piriyapongsa J, Jordan IK (2007) A family of human microRNA genes from miniature inverted-repeat transposable elements. PLoS One 2: e203.J. PiriyapongsaIK Jordan2007A family of human microRNA genes from miniature inverted-repeat transposable elements.PLoS One2e203
- 52. Bredel M, Scholtens DM, Harsh GR, Bredel C, Chandler JP, et al. (2009) A network model of a cooperative genetic landscape in brain tumors. Jama 302: 261–275.M. BredelDM ScholtensGR HarshC. BredelJP Chandler2009A network model of a cooperative genetic landscape in brain tumors.Jama302261275
- 53. Naylor TL, Greshock J, Wang Y, Colligon T, Yu QC, et al. (2005) High resolution genomic analysis of sporadic breast cancer using array-based comparative genomic hybridization. Breast Cancer Res 7: R1186–1198.TL NaylorJ. GreshockY. WangT. ColligonQC Yu2005High resolution genomic analysis of sporadic breast cancer using array-based comparative genomic hybridization.Breast Cancer Res7R11861198
- 54. Wood LD, Parsons DW, Jones S, Lin J, Sjoblom T, et al. (2007) The genomic landscapes of human breast and colorectal cancers. Science 318: 1108–1113.LD WoodDW ParsonsS. JonesJ. LinT. Sjoblom2007The genomic landscapes of human breast and colorectal cancers.Science31811081113