Cyclin E Overexpression in Human Mammary Epithelial Cells Promotes Epithelial Cancer-Specific Copy Number Alterations.

Cyclin E, a key cell cycle regulatory protein, has been linked to oncogenesis when dysregulated. We have previously shown that overexpression of cyclin E causes replication stress, leading to failure to complete replication at specific chromosomal loci during S phase of the cell cycle. This in turn promotes chromosomal damage during anaphase. Here we show that non-transformed human mammary epithelial cell clones that survive such aberrant mitoses have a specific and reproducible pattern of chromosomal Copy Number Alterations (CNAs) that we have characterized and termed the cyclin E CNA signature. Using a number of computational approaches, we show that this signature resembles one specific CNA pattern enriched in differentiated epithelial-like tumors of the breast and ovary. Analysis of the CNA profile of these clones provides a potential mechanism for cyclin E-mediated oncogenesis.

Introduction

There are several lines of evidence from different models indicating that cyclin E deregulation causes replication stress in S phase and chromosome segregation errors in M phase, leading to genomic instability and cancer (Bartkova et al., 2006, Bortner and Rosenberg, 1997, Porter et al., 1997, Smith et al., 2006, Spruck et al., 1999). However, the mechanistic link between cyclin E-mediated genomic instability and tumorigenesis remains poorly understood. We hypothesized that cyclin E deregulation promotes oncogenesis by creating a specific pattern of Copy Number Alterations (CNAs).

In our previous work, we demonstrated that transient overexpression of cyclin E in non-transformed human mammary epithelial cells (HME1) causes failure to replicate specific genomic regions and resulting genomic instability (Teixeira et al., 2015). However, the degree of DNA damage incurred impeded clonal expansion of these cells, presumably due to p53-dependent checkpoint activation, and therefore prevented us from studying the effects of cyclin E overexpression on the genomic landscape during the early stages of cancer development. To circumvent this problem, an HME1 derivative cell line (HME1-E6) constitutively expressing the human papilloma virus (HPV) E6 protein, which promotes proteasomal degradation of p53 (Thomas et al., 1999), was constructed. As shown (Figure S1A), p53 levels are reduced in HME1-E6 cells relative to the parental cell line HME1. To test whether reduction of p53 levels confers resistance to DNA damage, we treated HME1 and HME1-E6 lines with the topoisomerase inhibitor etoposide and determined the level of apoptosis based on the appearance of cleaved PARP. Cleaved PARP was apparent in HME1 cells after 24 h, whereas HME1-E6 exhibited a similar level of cleaved PARP only after 48 h, suggesting that HME1-E6 cells are more resistant to DNA damage (Figure S1B). Using a second marker for apoptosis, Annexin V staining, similar results were obtained. Approximately 80% of HME1 cells stained positively after 24 h of etoposide treatment compared with 58% of HME1-E6 (Figure S1C) These data indicate that HME1-E6 is more resistant to DNA damage and therefore more likely to sustain clonal expansion after acute expression of cyclin E. This was confirmed when, after overexpressing cyclin E, we were able to obtain single-cell-derived populations bearing chromosomal deletions in regions defined as under-replicated in our previous study (Teixeira et al., 2015) (Figure 1A).

Figure 1

Chromosomal Analysis of HME1-E6 Clonal Populations of Transient Expression of Cyclin E

(A) Heatmap of CNAs determined by qPCR at some of loci identified in previous study (Teixeira et al., 2015). Forty-two cyclin E clones were analyzed and compared with the average of seven control clones. Blue, CNA gains; red CNA losses. Normalization to locus encoding α-tubulin. Intensity of color indicates magnitude of CNA.

(B) Transient cyclin E overexpression leads to characteristic CNA signature. CMA analysis of 12 cyclin E and 7 control clones. Data shown are based on CNA call analysis. Bottom plot shows gains (+1) and losses (−1) across 22 chromosomes for all CNAs with a p value ≤0.05. Upper Manhattan plot shows significance as -log(p).

Transient Cyclin E Expression Produces a Specific CNA Signature in HME1-E6 Cells

To determine whether cyclin E expression produces a specific genome-wide CNA signature, Chromosomal Microarray (CMA) analysis was carried out on DNA from individual clones. CNA call analysis of the CMA data from 12 cyclin E-overexpressing clones and 7 control clones (expressing GFP) was performed to generate a cyclin E-mediated CNA profile. The gains and losses are represented as a non-quantitative histogram, with significance indicated using a Manhattan plot, in Figure 1B. The normalized array data used to produce Figure 1B are shown in Table S1. As can be seen, cyclin E expression causes highly significant copy number increases and decreases relative to controls.

Relationship of the Cyclin E CNA Signature to Cancer States

To determine whether the cyclin E CNA profile is related to specific cancer states, we carried out an association analysis against cancer cell lines (The Cancer Cell Line Encyclopedia, CCLE [Barretina et al., 2012], and primary samples from the Pan-cancer The Cancer Genome Atlas, TCGA [Lawrence et al., 2014, Wang et al., 2016]) dataset. We defined a cyclin E CNA signature score for each cancer sample by adding all its observed cyclin E-mediated gains and losses weighted by an amount corresponding the log of the inverse of its p value (Figure 1B) so that the most significant CNAs will have a higher contribution to the score (see Transparent Methods). Repeating this procedure for each sample allowed us to define a vector of cyclin E CNA scores for all CCLE and TCGA samples. Then we matched the signature scores against pathway (single-sample Gene Set Enrichment Analysis, GSEA) and protein expression profiles (Reverse Phase Protein Array, RPPA). To quantify the strength of association we used the Information Coefficient, which is an Information theoretic metric that provides a non-linear counterpart of the Pearson correlation coefficient (Joe, 1989, Kim et al., 2016). The goal of this analysis is to find genomic correlates of the cyclin E signature that will be informative, e.g., to assess the characteristics of samples with higher versus lower cyclin E signature scores. Interestingly, significant association was detected between the cyclin E signature and several epithelial-like gene sets from the Molecular Signatures Database (MsigDB [Joe, 1989, Kim et al., 2016, Liberzon, 2014, Liberzon et al., 2015, Liberzon et al., 2011]) (Figure 2, top). These top scoring gene sets include those representing ZEB1 targets, downregulation of the epithelial-mesenchymal transition (EMT), stem cell and metastasis, epithelial cell differentiation, ERBB2, EGFR inhibition sensitivity, and KRAS dependency. Consistent with this picture, the top scoring proteins included claudin, e-cadherin, HER2, and EGFR (Figure 2, bottom).

Figure 2

Pathway and Protein Association Analysis Indicates that Transient Cyclin E Overexpression Produces a Cellular State with Similarities to Differentiated Epithelial-like Tumors

Top 20 pathways and top 10 proteins associated with the cyclin E CNA signature. Top bar in the heatmaps shows the high-to-low sorted amplitude of the cyclin E CNA signature across 958 cell lines in the CCLE dataset. The top scoring gene sets and proteins are strongly consistent with relatively differentiated mammary carcinomas such as ductal and basal and not with less differentiated mesenchymal-like cancers, mammary stem cells, or non-malignant mammary tissue. Statistical signficance of top-scoring features was assessed by an empirical permutation test, and its corresponding null distribution, false discovery rates (FDR) plus bootstrap estimates of the association metric confidence intervals.

We then used the Oncogenic Positioning System (Onco-GPS), a data driven computational analysis that projects and clusters cancer samples onto a 2-dimensional reference map based on their most salient patterns of gene expression behavior (Kim et al., 2017). We took a variant of the global Onco-GPS map introduced in Kim et al. (2017) (Figure 3A) and used it to determine if samples with high cyclin E signature score were enriched in any specific previously determined cancer state. To carry out this analysis, we color-coded each of the CCLE cell lines (Kim et al., 2017). Consistent with the previous analysis, the cyclin E CNA signature was enriched primarily in epithelial-like cancer samples (Figure 3B). This is better appreciated if one compares the map shown in Figure 3B with the same map where the samples are color-coded to represent the enrichment in ZEB1 targets, shown on the left side of Figure 4A, which have higher expression in the top right side of the map, and provide a guide of where the most epithelial cancer states are located. When one compares this pattern (Figure 4A, left) against the pattern of cyclin E CNA signature scores (Figure 3B), one can observe an overlap that is indeed statistically significant when quantified by the Information Coefficient between those two patterns (IC:0.52, p value: 2.62 × 10−6, Figure 2A). The cyclin E CNA signature was much lower in cancers that were less differentiated and had undergone EMT (Figure 4B). A similar trend was observed in the Pan-cancer TCGA samples where the samples that are projected onto the Onco-GPS epithelial states have often higher scores for the cyclin E CNA signature (Figure 5). These results are consistent with the idea that transient cyclin E overexpression in non-transformed mammary epithelial cells produces a CNA signature that appears to be enriched in differentiated epithelial cancer states.

Figure 3

Onco-GPS Analysis Indicates that Transient Cyclin E Overexpression Produces a State with Similarities to Epithelial-like Cancers

(A) Onco-GPS map showing 15 cancer states (a variant of the map introduced in [Groger et al., 2012]).

(B) Same map as in (A) but with the samples color-coded to show the expression of the cyclin E CNA signature.

Figure 4

Onco-GPS Maps Showing that the Cyclin E CNA Signature Is Consistent with Activation of Epithelial Target Genes and Repression of Gene Sets Associated with the Epithelial-Mesenchymal Transition (EMT)

Same map as in Figure 3A but with samples color-coded to show the expression of an epithelial (ZEB1 targets) (A) and an EMT gene sets (B) (Groger et al., 2012).

Figure 5

Onco-GPA Map of the Cyclin E CNA Signature Projected onto the Pan-cancer TCGA Dataset

Same Onco-GPS map as in Figure 3A but projecting 8,029 samples from the Pan-cancer TCGA dataset. The samples are color coded according to the expression of the cyclin E CNA signature.

Subsets of Clones Bearing the Cyclin E CNA Signature Show Properties Associated with Malignant Transformation

To determine whether the cyclin E CNA signature actually represents a cancer-like state, we tested whether clones analyzed in this study were capable of anchorage-independent growth or were resistant to apoptosis, properties associated with oncogenic transformation. The parental cell line, HME1-E6, and all 7 control clones could not form colonies in soft agar, whereas remarkably 2 out of 12 clones that had been subjected to transient cyclin E overexpression could (Figures 6A and 6B). Clearly not all cells in these clonal populations could form colonies, as compared with the positive control, a metastatic breast cancer-derived cell line. This indicates that additional genetic or epigenetic changes, existing as variants in the population, are necessary to support anchorage-independent growth. To gain more insight into the specific CNAs that might explain the behavior of these two clones, we carried out a genome-wide association study (GWAS) comparing them with the 10 clones incapable of anchorage-independent growth (Figure S2). The most salient characteristic found was amplification of a segment of Chromosome 8 containing the c-Myc oncogene. It is notable that c-Myc overexpression has been shown to be associated with anchorage-independent growth (Lyn-Cook et al., 1990, Telang et al., 1990, Valverius et al., 1990). In addition, copy number losses were detected for segments containing tumor suppressive microRNAs (miRNAs), miR33B, 151A, and 1288 (Gorringe et al., 2009, Xu et al., 2014, Yeh et al., 2016). To determine whether c-Myc copy number gain is reflected by increased c-Myc transcription and activation of downstream targets, we compared c-Myc transcript levels in the 12 clones that had been subjected to transient cyclin E overexpression (Figure 6C) and expression of gene sets known to be under c-Myc control (Figure 6D). Clearly c-Myc is expressed at higher levels in the two clones capable of anchorage-independent growth, and this has biological consequences in that downstream c-Myc targets are also expressed at higher levels in these clones.

Figure 6

Two Clones Subjected to Transient Cyclin E Overexpression Are Capable of Anchorage-Independent Growth

(A) Low- (top) and high- (bottom) magnification images of parental cell line HME1-E6, derivative clones 1 and 10, and positive control breast cancer cell line MDA-MB-231 after plating in soft agar (scale bars, 50 μm).

(B) Quantification of the number of measurable colonies (>50 μm) formed in each cell line. Asterisks indicate statistically significant changes as determined by untailed t test (***) p < 0.001; (****) p < 0.0001. All the values represent the mean ± SD from six independent experiments.

(C) Quantification of c-Myc transcript levels in the 12 clones.

(D) Quantification of transcript levels of genes trans activated by c-Myc. “P” refers to the parental cell lines HME1-E6. Statistical signficance of top-scoring features was assessed by an empirical permutation test, and its corresponding null distribution, false discovery rates (FDR) plus bootstrap estimates of the association metric confidence intervals.

Of 12 clones, 4 were found to be more resistant than the parental cell line to apoptosis resulting from treatment with the topoisomerase inhibitor, etoposide (Figure S3). These data indicate that, in an epithelial cell line expressing hTERT and partially compromised for p53 function, CNA alterations produced by transient cyclin E overexpression can promote specific oncogenic properties.

An Unbiased Grouping of Cancer Cell CNA Signatures Reveals a Strong Match with the Cyclin E Signature

To determine whether the cyclin E CNA signature is a global CNA pattern that could have been detected by an unbiased bioinformatics global analysis of CNA alterations, we carried out a Non-negative Matrix Factorization (Brunet et al., 2004, Kim et al., 2017, Tamayo et al., 2007) analysis of the CCLE copy number dataset to define global CNA signatures. For this analysis we define separate amplification and deletion binary features using as threshold a factor of two (half) for amplifications (deletions). This analysis produced 30 distinct CNA signatures, one of which (S7) showed a very strong association with the cyclin E CNA signature (Figure 7). Indeed, many of the cell lines that have high S7 CNA signature have also a strong association with the cyclin E CNA signature and are differentiated epithelial cell lines, mostly of ovarian and breast origin. The association scores for the cyclin E and S7 signatures are shown for the highest scoring cell lines and are shown in Table S2. We then ranked the CNA S7 signature features based on their degree of association with cyclin E CNA signature features (Table S3). Interestingly, we found that the principal driver of CNA signature S7 and cyclin E CNA signature association is a series of deletions in chromosome 18q12-23. These deletions dominate the S7 signature and are also dominant in the cyclin E signature (note the high levels of significance in Figure 1B). These data, taken together, suggest that deletions in the q arm of chromosome 18 are an important feature of differentiated epithelial cancers, particularly of the ovary and breast.

Figure 7

Comparison of the Cyclin E CNA Signature to 30 Global CNA Signatures across 1,030 Cancer-Derived Cell Lines

Top bar in the heat maps shows the high-to-low sorted amplitude of the cyclin E CNA signature across 1,030 cell lines in the CCLE dataset. The heatmap shows the amplitude of each of the 30 global CNA signatures. As can be seen in the figure, signature S7 is the best match. Statistical signficance of top-scoring features was assessed by an empirical permutation test, and its corresponding null distribution, false discovery rates (FDR) plus bootstrap estimates of the association metric confidence intervals.

The Cyclin S7-cyclin E CNA Signature Drives Expected Transcriptional Alterations

The most likely manner in which CNAs can have a biological impact is to alter transcriptional output, where gene expression would be reduced on deleted chromosomal segments and increased on amplified segments. To test this hypothesis on a large dataset, we compared transcriptional amplitude to CNA status for cell lines in the CCLE. We determined whether the total CNA landscape associated with cyclin E overexpression confers the expected transcriptional modulation, by analyzing transcriptional amplitude in amplified and deleted chromosomal regions, respectively (Figure 8). As in Figure 2A, CCLE cell lines are arranged according to their cyclin E CNA signature score from left to right. The top analysis shows the cumulative expression level of genes on amplified chromosomal segments, whereas the bottom analysis shows the cumulative expression level of genes on deleted segments. As can be seen in Figure 8, in cell lines that strongly conform to the cyclin E CNA signature, there is a significant association between higher levels of transcription in amplified regions (IC = 0.329, p value = 0.004, false discovery rate [FDR] = 0.009) and, to an even greater degree, lower levels of transcription in deleted regions (IC = −0.513, p value = 0.0005, FDR = 0.001). In addition, assessment of transcriptional levels of individual genes in these regions also showed that, in most cases, there were strong concordances between deletion or amplification at these loci and reduced or increased transcription, respectively (Figure S4). One can therefore conclude that CNA changes in cells impact cellular biology, at least in part, by driving corresponding transcriptional changes.

Figure 8

The S7 CNA Signature Produces a Parallel Transcriptional Signature

Top bar in the heatmaps shows the high-to-low sorted amplitude of the cyclin E CNA signature across 958 cell lines in the CCLE dataset. The second bar shows combined expression of genes in intervals showing copy number gains in the cyclin E signature, whereas the third bar shows combined expression of gene in the intervals showing copy number losses. Red represents higher levels of transcription, whereas blue represents lower levels. Statistical signficance of top-scoring features was assessed by an empirical permutation test, and its corresponding null distribution, false discovery rates (FDR) plus bootstrap estimates of the association metric confidence intervals.

Discussion

Our studies have demonstrated that cyclin E-mediated replication stress, by impairing replication of genomic regions and thereby leading to subsequent deletions and amplifications, promotes a state that has characteristics similar to differentiated epithelial cancers. This is based on a number of criteria, including pathway and protein expression data, as well as direct comparison of CNA signature patterns. The latter analysis has indicated that deletions on the q arms of chromosomes 17 and 18 are the most critical components of the association between the cyclin E signature and the signature characteristic of an un-biased grouping of cancer cells that best matches with the cyclin E signature. It is interesting to note that several tumor suppressor genes have been specifically mapped to this region, possibly explaining the relevance of this signature to tumorigenesis (Dellas et al., 2002, Takei et al., 1998, Thiagalingam et al., 1996, Yokota et al., 1997). We furthermore showed that expression of most genes in this region was reduced in cells containing these deletions, as expected, also likely contributing to the cancer phenotype.

Another interesting finding of this work is that after transient cyclin E expression and clonal expansion, subsets of clones were found capable of anchorage-independent growth and resistant to apoptosis, respectively. Although these properties are not synonymous with transformation, they are components of the cancer phenotype. The most likely driver of anchorage-independent growth in this context is a copy number gain of a segment of chromosome 8 containing c-Myc (Lyn-Cook et al., 1990, Telang et al., 1990, Valverius et al., 1990). In fact, the clearest demonstration that CNAs can have a direct impact on cellular biology came from comparing transcriptional profiles of the clones. The two clones that were capable of anchorage-independent growth and showed CNA gains of a segment that included c-Myc showed increased levels not only of c-Myc transcripts but also of transcripts of known downstream targets of c-Myc. Since only two of twelve clones subjected to cyclin E overexpression contained this CNA, it is too soon to assign it to the overall cyclin E CNA signature or even to be considered statistically significant. Screening a larger number of clones for anchorage-independent growth and analyzing the associated CNAs will likely answer this question.

Our study only characterized one of many possible types of genomic and genetic alterations promoted by the oncoprotein cyclin E, specifically CNAs that could be detected using array technology. However, it is likely that much additional information could be gleaned by carrying out a more detailed analysis of clonal populations that had been subjected to cyclin E overexpression. Recently, a detailed study of the mutational landscape promoted by expression of p21Cip1 in a p53-deficient environment, another situation that promotes genetic and genomic instability, was published (Galanos et al., 2018). By analysis of the end products of the mutational process, the authors were able to deduce that an aberrant repair process was induced under these specific circumstances, suggesting a novel therapeutic target for malignancies with similar characteristics. It will be interesting to determine whether the repair response to damage produced by cyclin E overexpression is altered and whether this information might be used to identify new therapeutic targets.

The most important conclusion of this work is that oncogene generated CNA signatures are likely to be a useful parameter for characterizing tumors and cancer cells. The convergence of a simple in vitro model with patterns derived from the in vivo endpoints of oncogenesis suggests a model where oncoprotein-driven CNA patterns serve as a platform from which further genetic and epigenetic changes favorable for tumor formation can be selected. This hypothesis can be tested by determining the efficacy of the clones described here to form tumors in mouse xenograft models.

Limitations of the Study

In this study we show that transient overexpression of cyclin E in mammary epithelial cells produces a pattern of chromosomal CNAs that resembles those found in a subset of epithelial lineage cancers. This raised the question of whether these CNAs translate into changes in gene expression that would drive cancer biology. To address this, we analyzed how these particular CNAs in a large database of cancer cells translated to changes in gene expression and showed that deletions and amplifications produced decreased and increased expression, respectively, as hypothesized (Figures 8 and S4). We further showed that individual clones in our study that exhibited c-Myc amplification actually expressed elevated levels of c-Myc and transcriptional activation c-Myc downstream targets (Figures 6C and 6D). However, we did not carry out a more global analysis of the relationship between CNAs and gene expression in the individual clones. This study was not carried out at this time because of the small sample size available and therefore insufficient power to draw reliable conclusions. However, we consider elucidating the link between cyclin E-generated CNAs and gene expression in the context of cancer biology to be extremely important and plan to carry out such a study after analyzing a larger number of clones.

Methods

All methods can be found in the accompanying Transparent Methods supplemental file.