High cyclin E1 protein, but not gene amplification, is prognostic for basal-like breast cancer.

Basal-like breast cancer (BLBC) has a greater overlap in molecular features with high-grade serous ovarian cancer (HGSOC) than with other breast cancer subtypes. Similarities include BRCA1 mutation, high frequency of TP53 mutation, and amplification of CCNE1 (encoding the cyclin E1 protein) in 6-34% of cases, and these features can be used to group patients for targeted therapies in clinical trials. In HGSOC, we previously reported two subsets with high levels of cyclin E1: those in which CCNE1 is amplified, have intact homologous recombination (HR), and very poor prognosis; and a CCNE1 non-amplified subset, with more prevalent HR defects. Here, we investigate whether similar subsets are identifiable in BLBC that may allow alignment of patient grouping in clinical trials of agents targeting cyclin E1 overexpression. We examined cyclin E1 protein and CCNE1 amplification in a cohort of 76 BLBCs and validated the findings in additional breast cancer datasets. Compared to HGSOC, CCNE1 amplified BLBC had a lower level of amplification (3.5 versus 5.2 copies) and lower relative cyclin E1 protein, a lack of correlation of amplification with expression, and no association with polyploidy. BLBC with elevated cyclin E1 protein also had prevalent HR defects, and high-level expression of the cyclin E1 deubiquitinase ubiquitin-specific protease 28 (USP28). Using a meta-analysis across multiple studies, we determined that cyclin E1 protein overexpression but not amplification is prognostic in BLBC, while both cyclin E1 overexpression and amplification are prognostic in HGSOC. Overall CCNE1 gene amplification is not equivalent between BLBC and HGSOC. However, high cyclin E1 protein expression can co-occur with HR defects in both BLBC and HGSOC, and is associated with poor prognosis in BLBC.

Introduction

Basal‐like breast cancer (BLBC) and high‐grade serous ovarian cancer (HGSOC) are both aggressive diseases with high frequency of TP53 mutation, BRCA1 mutation, and CCNE1 gene amplification at 19q12 [1]. Overall, BLBC is more similar to HGSOC than to other breast cancer subtypes based on the comparison of copy number gains and losses, and the comparison of gene network activation by analysis of transcriptomes [1]. There is also a significant co‐occurrence of these tumours in patients [2].

While the mainstay of treatment for BLBC and HGSOC is chemotherapy, the high frequency of BRCA1 mutation and other homologous recombination (HR) defects has led to the introduction of PARP (poly adenosine diphosphate‐ribose polymerase) inhibitors as a maintenance therapy in both cancers [3, 4]. It is now common for clinical trials to group HGSOC and BLBC patients in order to access larger cohorts when testing new drug combinations with PARP inhibitors or PARP inhibitors in pre‐treated patients (e.g. clinical trials NCT01623349, NCT00679783, NCT00892736, EMBRACE). CCNE1 amplification is also being explored as a therapeutic biomarker for HGSOC and BLBC. This is because cells with very high cyclin E1 expression undergo replication stress and genomic instability, and these cancer cells can be forced into mitotic catastrophe and cell death through inhibition of the Chk1 and Wee1 cell cycle checkpoints [5]. Consequently, CCNE1 amplification has been used as a biomarker to group patients with solid tumours for clinical trials of Chk1 inhibitor prexasertib (NCT02873975) and Wee1 inhibitor adavosertib (NCT03253679), with results of these trials pending. An alternative way to target the oncogenic activity of cyclin E1 is via the inhibition of its partner kinase, CDK2. Pre‐clinical studies have demonstrated the potential of CDK2 inhibitors in combination with other therapies for treatment of CCNE1‐amplified HGSOC and cyclin E1 overexpressing BLBC [6, 7].

With the increasing use of biomarker‐driven treatment for BLBC and HGSOC, it is imperative to ensure that each biomarker associates with similar biology and therapeutic vulnerabilities in these cancers. Recently, we described separate subsets of HGSOC, one where CCNE1 amplification and BRCA1/BRCA2 mutation were mutually exclusive (the 19q12 amplified subset), and another where BRCA1/BRCA2 mutation co‐occurred with high cyclin E1 protein overexpression (the high cyclin E1 protein subset) [8]. We subsequently investigated a patient cohort that is highly enriched for BLBC and identified the co‐occurrence of BRCA1 mutation and high cyclin E1 protein [7]. However, an unresolved question is whether CCNE1 amplification and cyclin E1 protein expression in BLBC demarcate patients into subgroups that are similar to those that have been detected in HGSOC. This has important implications for the selection of BLBC patients to receive therapies such as PARP inhibitors, cell cycle checkpoint inhibitors, and CDK2 inhibitors.

Here, we aimed to determine the similarities and differences between BLBC and HGSOC in terms of cyclin E1 expression, CCNE1 amplification, and patient survival. Accordingly, we assessed the overlap between BLBC and HGSOC subsets in terms of the relative expression of cyclin E1, CCNE1 amplification, homologous recombination deficiency (HRD), and their association with outcome.

Materials and methods

Patient cohorts

Molecular and clinical data were obtained for breast cancer patients from the Kathleen Cunningham Foundation Consortium for research into Familial Breast cancer (kConFab), as described in Ref. [7], and for HGSOC from the Australian Ovarian Cancer Study (AOCS), as described in Ref. [8]. For both studies, ethics board approval was obtained at all institutions for patient recruitment, sample collection, and research studies. Written informed consent was obtained from all participants for participation in research studies. Both cohorts are described in Supplementary materials and methods, including the definition of BLBC for the kConFab cohort. Summary details of BLBC and breast cancers other than BLBC (non‐BLBC) are shown in Table 1.

Table 1

Patient and tumour characteristics of BLBC and non‐BLBC

	BLBC		Non‐BLBC
Age, years	Median	Range	Median	Range
	40	19–74	43.5	29–73

EGFR, epidermal growth factor receptor; ER, oestrogen receptor; PR, progesterone receptor; HER2, human epidermal growth factor receptor 2.

BLBC was identified from PAM50‐defined subsets of The Cancer Genome Atlas (TCGA) [9] and Molecular Taxonomy of Breast Cancer International Consortium (METABRIC) [10], and HGSOC from the TCGA dataset [11]. Data for protein expression, gene amplification, and patient survival were downloaded via cBioPortal [12]. Cut‐offs were ≥0 z‐score on reverse phase protein array (RPPA) for high protein expression [13], and a Genomic Identification of Significant Targets in Cancer (GISTIC) score of 2 for gene amplification [14]. GISTIC scores of CCNE1 amplification in cell lines were downloaded via cBioPortal [12] from the Cancer Cell Line Encyclopaedia [15]. Gene expression and ASCAT (Allele‐Specific Copy number Analysis of Tumours) estimates of ploidy [16] from HGSOC and BLBC of the TCGA cohorts were accessed through COSMIC [17]. Homologous recombination deficiency‐loss of heterozygosity (HRD‐LOH) score is a measure of LOH through analysis of allele loss, and scores for TCGA samples were derived from Ref. [18].

Cyclin E1 analysis

Cyclin E1 immunohistochemistry (IHC) is described in Supplementary materials and methods and supplementary material, Figure S1. Detection of CCNE1 amplification is described in Supplementary materials and methods.

Meta‐analyses

Publications reporting cyclin E1 protein expression, CCNE1 amplification, and overall survival data were identified through searches with the terms ‘cyclin E’, ‘cyclin E1’, ‘CCNE1’ ‘survival’, ‘breast cancer’, and ‘ovarian cancer’ of the PubMed database, Google Scholar, and SCOPUS. To be included, studies needed to be able to report overall survival at 50 months for either HGSOC or BLBC, have 40 patients or more, and report the HGSOC and BLBC subgroups (defined by cytokeratin [CK] 5/6/14 staining or PAM50 status) as distinct entities within ovarian cancer and breast cancer, respectively. Study identification is shown with PRISMA diagrams in supplementary material, Figure S2. Bias within each meta‐analysis was assessed by inspection of FUNNEL plots of treatment effects from individual studies plotted against a measure of study size [19]. The prognostic power of cyclin E1 expression and CCNE1 amplification were assessed by meta‐analysis using Revman 5.0 software [20].

Cell line culture

Cell lines were obtained from ATCC and cultured in RPMI 1640, 5–10% foetal calf serum, and insulin (10 μg/ml). All cell lines were authenticated by STR profiling (CellBank Australia) and cultured for less than 6 months after authentication. Bona fide HGSOC and BLBC cell lines were classified as described [21, 22]. The derivation of cyclin E1 overexpressing cell lines, ploidy analysis, and colony‐forming assays is described in Supplementary materials and methods.

Statistical analysis

Statistical analyses were performed using Prism Software™ version 9.3.1 [23]. Statistical tests, P value calculation, and data presentation are described in detail in Supplementary materials and methods.

Results

High cyclin E1 protein expression and amplification are frequent in BLBC

We explored the relationship between cyclin E1 protein expression and survival in BLBC by quantifying cyclin E1 expression across the kConFab familial breast cancer cohort [7]. Cyclin E1 expression was significantly higher in BLBC compared to non‐BLBC cases (p < 0.0001) (Figure 1A). We validated these findings in the PAM50‐defined BLBC cohort of TCGA using RPPA data [1]. This showed that cyclin E1 protein is significantly higher in BLBC compared to other PAM50‐defined breast cancer subtypes including luminal A, luminal B, and HER2 enriched (p < 0.0001) (Figure 1B).

Figure 1

High protein expression of cyclin E1 and CCNE1 amplification are frequent in BLBC. (A) Cyclin E1 protein expression in BLBC compared to non‐BLBC in the kConFab cohort, analysed by Mann–Whitney test. (B) Cyclin E1 protein expression by RPPA in 273 TCGA breast cancer cases categorised into PAM50 defined subtypes: luminal A, basal‐like, HER2 enriched, and luminal B. Data analysed by one‐way ANOVA with Tukey test for multiple comparisons. (C) 19q12/INSR ratio (ISH) in BLBC compared to non‐BLBC in the kConFab cohort, analysed by Mann–Whitney test. Data in (A)–(C) are box and whisker plots with error bars of minimum to maximum, where the box extends from the 25th to the 75th percentile, and the line in the middle of the box indicates the median. (D) Relative 19q12 amplification in the BLBC and non‐BLBC subsets of the kConFab cohort, as defined by 19q12/INSR ratio of ≥3, and/or 19q12 score >6. Analysed by Fisher's exact test. (E) Relative CCNE1 amplification across the molecular subtypes of breast cancer, including BLBC, based on amplification (GISTIC = 2) in the TCGA breast cancer dataset. (F) Relative CCNE1 amplification across the molecular subtypes of breast cancer, including BLBC, based on amplification (GISTIC = 2) in the METABRIC dataset.

Amplification of the CCNE1 gene located at 19q12 is a major contributor to high cyclin E1 expression in HGSOC [8, 24] and therefore we assessed whether 19q12 amplification is associated with BLBC. Compared to non‐BLBC breast cancers, BLBC had significantly higher copy number at the CCNE1 gene locus, as measured by the ratio of 19q12 to chromosome 19 control probe, insulin receptor (INSR) (p = 0.0009) (Figure 1C). Twenty‐five percent of BLBC patients in the kConFab cohort had 19q12 amplification, compared to 7.4% of non‐BLBC (p = 0.0006, Fisher's exact test, Figure 1D). BLBC had the highest rate of CCNE1 amplification in samples from the TCGA and METABRIC datasets (Figure 1E,F).

Meta‐analysis identifies that amplification and cyclin E1 protein expression are both prognostic for HGSOC, but only cyclin E1 protein expression is prognostic for BLBC

Cyclin E1 protein and CCNE1 amplification have been reported as prognostic markers in subtypes of breast cancer [25, 26, 27] and in HGSOC, which has many similarities to BLBC. The kConFab, TCGA, and Metabric BLBC cohorts are small and hence underpowered for survival analyses, and the kConFab cohort also has inherent bias as it is highly enriched for mutation carriers for BRCA1 and BRCA2. To address these limitations, we combined these cohorts in meta‐analyses, along with other published cohorts, where additional studies were identified via the listed search terms (Figure 2A). In parallel, we examined studies of cyclin E1 protein and CCNE1 amplification in HGSOC, identified by the same search terms.

Figure 2

Meta‐analysis identifies that only cyclin E1 protein expression is prognostic for BLBC, but both CCNE1 amplification and cyclin E1 protein expression are prognostic for HGSOC. (A) Schematic of meta‐analysis of cyclin E1 protein expression and CCNE1 amplification in BLBC and HGSOC. (B) Amplification rates in studies of overall survival of BLBC (n = 3) and HGSOC (n = 5), analysed by t‐test. Error bars represent standard deviation (SD). (C–F) Meta‐analyses performed on studies using random‐forest analysis of a dichotomous value of overall survival at 50 months. (C) Meta‐analysis of association of high cyclin E1 protein with overall survival in HGSOC. (D) Meta‐analysis of association of CCNE1 amplification with overall survival in HGSOC. (E) Meta‐analysis of association of high cyclin E1 protein with overall survival in BLBC. (F) Meta‐analysis of association of CCNE1 amplification with overall survival in BLBC. Data from studies indicated with (*) are shown in supplementary material, Figure S4.

We identified three BLBC cyclin E1 protein studies [7, 9, 28], three BLBC 19q12/CCNE1 amplification studies [7, 9, 10], four HGSOC cyclin E1 protein studies [8, 29, 30, 31], and five HGSOC 19q12/CCNE1 amplification studies [8, 29, 31, 32, 33], which were suitable for inclusion in the meta‐analyses (PRISMA diagram, supplementary material, Figure S2). In these studies, there was a similar rate of CCNE1 amplification across BLBC and HGSOC (Figure 2B), although this was highly variable. This variability could have arisen because CCNE1 amplification can be measured by either estimates of copy number from exome/genome sequencing or by 19q12 ISH, and this may give rise to different sensitivity in the detection of amplification. Each meta‐analysis was weighted to patient number and analysed using a random effects model.

High cyclin E1 protein expression and CCNE1 amplification were both associated with greater risk of cancer death in HGSOC (Figure 2C,D). The risk of death was twice as high in patients with high cyclin E1 protein expression, with an odds ratio of 2.00 (95% CI 1.40, 2.87; p = 0.0001). Similarly, the risk was also increased for those patients with CCNE1 amplification with an odds ratio of 2.41 (95% CI 1.49, 3.90; p = 0.0003). The studies across CCNE1‐amplified ovarian cancer displayed some heterogeneity (I 2 = 43%), although this was not significant, and FUNNEL plots did not show significant deviations overall in the meta‐analyses (supplementary material, Figure S3).

In BLBC, only high cyclin E1 protein, and not CCNE1 amplification, was associated with increased risk of death (Figure 2E,F). The risk of death was 2.97‐fold higher in patients with high cyclin E1 protein expression (95% CI 1.19, 7.40; p = 0.02), whereas the risk of death for CCNE1 amplification was not significantly increased (odds ratio 1.47; 95% CI 0.71, 3.01; p = 0.30). Individual Kaplan–Meier curves for the kConFab, TCGA, and Metabric cohorts are shown in supplementary material, Figure S4.

‐amplified HGSOC expresses higher cyclin E1 protein than ‐amplified BLBC

The meta‐analysis showed that CCNE1 amplification is prognostic of poor overall survival for HGSOC but not for BLBC. This was despite a similar rate of CCNE1 amplification across BLBC (10–25%) and HGSOC (6–34%) (Figure 2B).

To explore why CCNE1 amplification has different prognostic value in these cancers, we examined how the level of protein expression was related to CCNE1 amplification in BLBC and HGSOC using data from the kConFab and AOCS cohorts [7, 8]. These cohorts were stained by the same protocol and used the same positive control, allowing their direct comparison. Protein expression was 2.8‐fold higher in 19q12 amplified HGSOC than in 19q12 amplified BLBC (p < 0.0001, Figure 3A). As 19q12 amplified HGSOC had significantly more cyclin E1 protein than 19q12 amplified BLBC, we assessed the degree of 19q12 amplification in each subset. We found that the median 19q12/INSR ratio in 19q12 amplified HGSOC was 5.2, which was significantly higher than the median ratio of 3.5 in BLBC (p = 0.0065, Figure 3B). In the TCGA dataset, we were unable to compare the degree of amplification, as all putative amplified cases are binned into a single amplification category of GISTIC = 2, and there was no significant difference between HGSOC and BLBC in cyclin E1 protein expression measured by RPPA (p = 0.11, Figure 3C).

Figure 3

CCNE1 undergoes low range amplification in BLBC, leading to moderate cyclin E1 protein expression. (A) Cyclin E1 protein expression in 19q12 amplified BLBC from kConFab and 19q12 amplified HGSOC from the AOCS. Analysed by t‐test, error bars represent standard deviation (SD). (B) The relative 19q12/INSR ratio of the amplified subsets of HGSOC and BLBC. Analysed using a non‐parametric Mann–Whitney test, error bars represent SD. (C) Cyclin E1 protein expression by RPPA in amplified BLBC and HGSOC from TCGA. Analysed by t‐test, error bars represent SD. (D) Scatter plot of cyclin E1 protein expression versus 19q12/INSR ratio amplification status in BLBC subset of the kConFab cohort. r = Spearman coefficient for correlation analysis. Interrupted lines indicate cut‐offs for high expression. (E) Relative cyclin E1 protein expression compared to copy number status in BLBC from TCGA. Analysed by one‐way‐ANOVA with Tukey test for multiple comparisons, error bars represent SD. (F) Western blots of HGSOC and BLBC cell lines with different degrees of CCNE1 gene amplification. Amplification levels derived from [15]. *** = High amplification (GISTIC = 2), * = CCNE1 gain (GISTIC = 1), − = unamplified or allelic loss (GISTIC = −1 or 0).

We had previously observed that cyclin E1 protein expression is correlated positively with CCNE1 amplification in HGSOC [8]. By contrast, there was no correlation between cyclin E1 protein expression and CCNE1 amplification (19q12/INSR ratio) across the kConFab BLBC cohort (r = 0.049, p < 0.67, Figure 3D). Several BLBC showed low levels of amplification (19q12/INSR score between 3 and 6) associated with either low or high cyclin E1 protein expression. However, 3/76 (4%) BLBC showed a 19q12/INSR ratio of ≥6, indicative of high‐level amplification. These three cases also had high cyclin E1 protein.

We then compared cyclin E1 protein expression in TCGA BLBC cases classed as having different copy number states. This showed that BLBC with CCNE1 gene amplification does not have significantly higher cyclin E1 protein expression than BLBC with diploid CCNE1 (p < 0.17, Figure 3E). We next examined the protein expression of BLBC and HGSOC in cell lines with different reported levels of amplification. Most BLBC cell lines had no amplification or low‐level gene gain of copy number and we did not observe a substantial difference in protein expression dependent on copy number (Figure 3F). Cyclin E1 expression in these models was comparable to expression in the HGSOC model CaOv3 which lacks CCNE1 amplification. One BLBC cell line with high‐level CCNE1 amplification (HCC1806) showed comparable cyclin E1 protein expression to the HGSOC CCNE1‐amplified cell line, OvCa3 (Figure 3F). Overall, it appears that while cyclin E1 protein expression is frequently high in BLBC, it does not necessarily associate with CCNE1 amplification.

High cyclin E1 protein is associated with high USP28 protein in BLBC

As CCNE1 amplification was not associated with high cyclin E1 in BLBC, we investigated other mechanisms that could be leading to higher cyclin E1 protein. We previously showed that patients with BRCA1 mutation, a BLBC‐enriched population, have significantly higher expression of ubiquitin‐specific protease 28 (USP28) [7]. USP28 is a deubiquitinase that stabilises cyclin E1 [34], and USP28 is also high in cyclin E1 high HGSOC that does not have CCNE1 amplification [8].

We examined the expression of USP28 in BLBC and its relationship to cyclin E1 protein expression in the BLBC and non‐BLBC patient subgroups. USP28 protein expression was assessed by immunostaining of formalin‐fixed paraffin‐embedded samples and H‐score. USP28 was significantly higher in BLBC than non‐BLBC in the kConFab cohort (p < 0.0001, Figure 4A). We also found that USP28 was significantly higher in cancers that were positive for basal markers CK5 and CK14, although not in cancers that were positive for the basal marker EGFR (epidermal growth factor receptor) (Figure 4B). We then examined the TCGA breast cancer cohort and found that USP28 mRNA was significantly higher in BLBC than in luminal A and luminal B cancers, although HER2‐enriched breast cancers had similar USP28 mRNA expression levels to BLBC (Figure 4C).

Figure 4

USP28 is high in BLBC, and correlates with high cyclin E1 protein. (A) USP28 protein expression in BLBC compared to non‐BLBC in the kConFab cohort, analysed by Mann–Whitney test. (B) USP28 protein expression in breast cancers from the kConFab cohort that are positive or negative for the basal markers, CK5, CK14, and epidermal growth factor receptor (EGFR). Analysed by Mann–Whitney test. NS = non‐significant. (C) USP28 mRNA expression as normalised transcripts per million (TPM) in TCGA breast cancer cases categorised into PAM50 defined subtypes: luminal A, basal‐like, HER2 enriched and luminal B. Data analysed by one‐way ANOVA with Tukey test for multiple comparisons. Data in (A)–(C) are box and whisker plots with error bars of minimum to maximum, where the box extends from the 25th to the 75th percentile, and the line in the middle of the box indicates the median. (D) Scatter plot of cyclin E1 protein expression versus USP28 protein expression in BLBC subset of the kConFab cohort. (E) Scatter plot of cyclin E1 protein expression versus USP28 protein expression in non‐BLBC subset of the kConFab cohort. r = Spearman coefficient for correlation analysis.

We examined the correlation between expression of cyclin E1 and USP28 protein in BLBC and non‐BLBC in the kConFab cohort. Cyclin E1 and USP28 protein expression showed a moderate and significant correlation in BLBC (r = 0.31, p < 0.0073, Figure 4D) but had only a trend towards a poor correlation in non‐BLBC (r = 0.16, p < 0.061, Figure 4E). Overall, these data suggest that USP28 is associated with the basal‐like phenotype, and that cyclin E1 protein is possibly stabilised via USP28‐mediated deubiquitination in BLBC.

Moderate expression of cyclin E1 is sufficient to increase survival advantage in a BLBC cell line

We had observed that CCNE1‐amplified BLBC did not express cyclin E1 at the same level as in CCNE1‐amplified HGSOC, and that that the overall expression of cyclin E1 protein in BLBC was lower than HGSOC regardless of amplification. This led us to speculate that very high expression of cyclin E1 that is commensurate with the expression seen in CCNE1‐amplified HGSOC may not lead to a growth or survival advantage for some BLBC cells. We tested this hypothesis by overexpressing cyclin E1 at different levels in the BLBC cell line MDA‐MB‐468, which has moderate endogenous expression of cyclin E1 (Figure 3F). Using retroviral infection, we overexpressed cyclin E1.V5‐fusion protein using the pMSCV vector, which co‐expresses GFP using an IRES (internal ribosome entry site) sequence. Populations expressing different levels of cyclin E1.V5 were selected by flow cytometry and validated by western blot (Figure 5A). Moderate overexpression of cyclin E1 was commensurate with non‐genomic overexpression seen in ovarian cancer CaOv3 cells, whereas high overexpression corresponded to the levels seen with CCNE1 amplification in the HGSOC cell line, OvCa3 (Figure 5A). The third population, expressing ‘extreme’ levels of cyclin E1, was >2‐fold higher than observed with amplification in OvCa3 cells.

Figure 5

Moderate expression of cyclin E1 is sufficient for a survival benefit in BLBC cell line MDA‐MB‐468. (A) Western blot of cyclin E1 expression in HGSOC cell lines (CaOv3 and OvCa3) and in MDA‐MB‐468 BLBC cell line, with cyclin E1‐V5 overexpression from the pMIG vector at moderate, high, and extreme levels. GAPDH is used as loading control. (B) Cell cycle phase of cyclin E1 overexpressing MDA‐MB‐468 cells determined by propidium iodide staining and flow cytometry. (C) MDA‐MB‐468 cells expressing different levels of cyclin E1 were treated with paclitaxel for 3 weeks, and analysed by colony‐forming assay. (D) Quantitation of colony‐forming assay performed in triplicate and analysed by two‐way ANOVA with Tukey test for multiple comparisons. Error bars represent standard deviation.

We determined the effect that these expression levels had on the cell cycle by performing propidium iodide staining for DNA content. There was a dose‐dependent effect, where the S phase fraction of the cells increased with increased expression of cyclin E1 (Figure 5B). We examined the colony‐forming ability of these cell lines following paclitaxel treatment to determine the effect of cyclin E1 expression on survival. Paclitaxel reduces colony‐forming ability, but cells overexpressing any level of cyclin E1 form more colonies at 2.6 nm paclitaxel (Figure 5C,D). Moderate‐level cyclin E1 overexpression led to a marked increase in colony number at the higher doses of paclitaxel, but further increases in expression to ‘high’ and ‘extreme’ levels did not further increase survival (Figure 5C,D).

Overall, we found that elevated cyclin E1 expression gave a survival advantage in the MDA‐MB‐468 BLBC model, but that very high expression of cyclin E1 does not have additional survival benefit.

Amplification of is associated with increased ploidy in HGSOC but not in BLBC

In HGSOC, the amplification of CCNE1 is specifically associated with an increase in cancer cell ploidy [35] but this has not been determined for BLBC. We assessed the association between cyclin E1 expression and polyploidy in BLBC and HGSOC. In the METABRIC dataset, using ASCAT estimations of ploidy derived from COSMIC, 56% of BLBC and 63% of HGSOC are high ploidy cancers (Figure 6A). When the expression of CCNE1 mRNA is compared between diploid and polyploid cancers for both cancer types, CCNE1 mRNA is similar in both diploid and polyploid BLBC but is significantly higher in the HGSOC polyploid compared to diploid cancers (Figure 6B). We then examined our cyclin E1 overexpressing BLBC cell line for differences in ploidy by measuring >4N content by propidium iodide staining. We observed that moderate overexpression of cyclin E1 did not lead to a significant increase in polyploid cells, but polyploidy was significantly increased in the case of the ‘extreme’ overexpression (Figure 6C). The lack of polyploid cells in BLBC is therefore consistent with the generally lower level of CCNE1 amplification and cyclin E1 expression observed compared to HGSOC.

Figure 6

Increased cyclin E1 expression in the BLBC is not associated with polyploidy but is associated with homologous recombination deficiency (HRD). (A) Non‐genome doubled (NGD) and genome doubled (GD) cancers of BLBC or HGSOC from the METABRIC dataset. (B) Expression of CCNE1 mRNA in NGD and GD cancers of BLBC or HGSOC from the METABRIC dataset. Data shown with floating bars of minimum to maximum, and line is at the mean. Data analysed by Mann–Whitney test. TPM = transcripts per million. (C) Percentage of polyploid cells in cell lines overexpressing cyclin E1. Polyploidy measured by propidium iodide staining and identification of >4N populations on flow cytometry. Performed in triplicate and analysed by one‐way ANOVA with Tukey post‐tests, error bars represent standard deviation (SD). (D) Levels of HRD‐LOH (signature score of cancers with homologous recombination deficiency) in cancer subsets of BLBC and HGSOC from TCGA. Data analysed by one‐way ANOVA and error bars are SD. A dashed line is shown at HRD‐LOH of 15.5 which is the cut‐off for high HRD‐LOH score. (E) Schematic of cyclin E1 expressing subsets of BLBC and HGSOC, and the relationship with USP28 expression, HR deficiency, and patient outcome.

High cyclin E1 protein co‐occurs with homologous recombination deficiency in BLBC and HGSOC, except in HGSOC cases with amplification

CCNE1 amplification is a driver of genomic instability in HGSOC and this is generally exclusive to HRD driven by BRCA1/BRCA2 loss in HGSOC [24]. Our prior study across the kConFab cohort demonstrates that cyclin E1 high BLBC co‐exists with BRCA1 mutations and BRCA1 methylation, suggesting that BLBC often exhibits both HRD and high cyclin E1 protein [7]. As the kConFab cohort is a familial cancer cohort that has natural enrichment for BRCA1 and BRCA2 mutations (Table 1) and may thus have an inherent bias, we investigated further in the TCGA datasets whether BLBC has co‐occurrence of HRD and high cyclin E1 protein.

We used a genomic estimate of HRD called HRD‐LOH to identify cancers that most likely have defective HR [36]. HRD‐LOH quantitates the LOH regions across the genome that corresponds to damage caused by HRD, and cancers with BRCA1/BRCA2 mutation have high HRD‐LOH (>15.5) [36]. HRD‐LOH is a component of the MyChoiceHRD commercial HRD test (Myriad Genetics, Salt Lake City, UT, USA) which, in some trials, has identified patients who will have better PARP inhibitor responses [37]. Across TCGA datasets, we found high HRD‐LOH in BLBC with CCNE1 amplification/high cyclin E1 (mean HRD‐LOH = 21.7) and non‐amplified BLBC with high cyclin E1 (mean HRD‐LOH = 20.1) (Figure 6D). CCNE1 non‐amplified HGSOC with high cyclin E1 had ~50% high HRD‐LOH (57/113, mean HRD‐LOH = 15.42), and 70% (40/57) of CCNE1‐amplified HGSOC had HRD‐LOH scores less than 15.5 (mean HRD‐LOH = 13.28). Thus, cyclin E1 high cancers have a consistent signature of HRD in BLBC and ~50% of CCNE1 non‐amplified HGSOC also has high HRD‐LOH score.

Discussion

Here, we describe that high expression of cyclin E1 protein by CCNE1 amplification does not frequently occur in BLBC despite the 10–25% rate of CCNE1 amplification in this cancer subtype. CCNE1 amplification also does not provide any prognostic value for overall survival in BLBC, which can be explained by the lack of consistent associated protein expression.

CCNE1 amplification is associated with poor prognosis in a number of gynaecological malignancies, including HGSOC [8], endometrioid endometrial carcinomas [38], and ovarian clear cell carcinoma [39]. There are inconsistent reports on whether there is an association between CCNE1 amplification and poor prognosis in breast cancers. CCNE1 gene amplification was not associated with overall survival in ER+ breast cancer in the Metabric cohort [40], and was not prognostic in a study of 232 node‐negative breast cancers [41]. CCNE1 amplification was associated with poor progression‐free survival in a small cohort of trastuzumab‐treated HER2+ breast cancers (n = 34) [25], but not in a similar study of 185 patients [42]. In a cohort of 54 triple‐negative breast cancers (TNBC) without BRCA1/BRCA2 mutations, CCNE1 amplification detected by targeted exome sequencing was associated with poor disease‐free survival, although not poor overall survival [26]. In the study of Zhao et al on chemotherapy‐treated TNBC disease (n = 59), which includes some BLBC cases, CCNE1 amplification was associated with poor overall survival [27]. Confounding factors across all studies are their small size, the different cut‐offs that may be used for CCNE1 amplification, that there are other potential oncogenic driver genes in the 19q12 amplicon such as URI [43], and that CCNE1 is not always included in the 19q12 amplicon as defined by arrayCGH [44]. Further larger scale studies would considerably improve our understanding of the relationship between CCNE1 amplification and survival across different subtypes of breast cancer, including in BLBC.

A surprising observation was that CCNE1 amplification was not significantly associated with high cyclin E1 protein expression in BLBC. CCNE1 amplification and cyclin E1 protein expression are associated in HGSOC [8], ovarian clear cell carcinoma [39], and endometrial cancer [45]. However, in agreement with our study, an investigation of medullary‐like BLBC found that two of four cancers with CCNE1 amplification lacked cyclin E1 protein expression [46], and in colon adenocarcinoma CCNE1 amplification and cyclin E1 protein expression are not correlated [47]. However, we note that three cases in the kConFab cohort had high‐level CCNE1/INSR ratios of >6, and this co‐occurred with high protein expression of cyclin E1 (Figure 3D). It is worth exploring if adjusting the threshold for CCNE1 amplification could identify a poor prognosis subset of CCNE1‐amplified BLBC, but larger cohorts are required for this to be assessed.

BLBC and HGSOC have similar rates of CCNE1 amplification, but we found that it is only in BLBC that CCNE1 amplification does not correlate with protein expression. BLBC also had lower relative cyclin E1 protein expression compared to HGSOC. We speculate that BLBC may have relatively lower cyclin E1 protein expression and a lack of correlation with gene amplification because of the prevalence of BRCA1 mutation and the related ‘Brcaness’ phenotype in BLBC. ‘Brcaness’ describes tumours with BRCA1/BRCA2‐like properties that most likely have similar HRD. Defects in HR and high genomic instability are synthetically lethal to cancer cells, with the consequence that it is highly unlikely that these aberrations are detected in the same cells. Indeed, CCNE1 amplification occurs in ~20% of HGSOC, and is mutually exclusive with BRCA1/2 mutation [24] and with ‘Brcaness’ that occurs in up to 60% of other HGSOC cases [48]. By contrast, almost all of BLBC shows a HRD signature [49], which could preclude very high cyclin E1 protein expression through synthetic lethality [50]. We note that the MDA‐MB‐468 cell line used to model an example of high expression of cyclin E1 in BLBC has a ‘Brcaness’ phenotype [51] that is associated with a BRCA2 mutation [52], and low homologous repair function [53]. This cell line exhibited a survival advantage in paclitaxel with moderate cyclin E1 protein expression, but there was no further advantage in survival with very high levels of cyclin E1 commensurate with CCNE1 amplification.

While CCNE1 amplification is not associated with poor overall survival of BLBC patients, we show that elevated expression of cyclin E1 protein occurs in ~40% of patients and is associated with poor overall survival. It appears likely that increased cyclin E1 protein in BLBC is due in part to protein stabilisation via the deubiquitinase USP28, rather than via CCNE1 amplification. Our study shows that USP28 is expressed at significantly higher levels in BLBC than non‐BLBC (Figure 4). Prior work has also suggested that USP28 may contribute to the basal phenotype, as the downregulation of USP28 led to decrease of the basal marker CK14, in association with a shift from basal to luminal cellular morphology [54], and in lung cancer the deletion of USP28 reduces the number of lesions with expression of CK5 [55]. We find that USP28 is high in CK5/CK14‐positive BLBC, and it correlates with high cyclin E1 protein. This suggests that USP28‐mediated stabilisation may contribute to high cyclin E1 in BLBC, consistent with our prior observation that USP28 siRNA treatment led to increased cyclin E1 protein expression in a BLBC cell line, MDA‐MB‐468 [7]. Our previous work has also demonstrated that cyclin E1 protein is stabilised by loss of phosphorylation in association with BRCA1 mutation [7]. As BLBC has high‐frequency BRCA1 mutation, it is likely that cyclin E1 protein is stabilised by multiple mechanisms in BLBC.

As part of this study, we have compared BLBC to HGSOC, which has two distinct subgroups with high cyclin E1 expression: CCNE1 amplified and CCNE1 non‐amplified [8]. We conclude that CCNE1 amplification has different associations with protein expression and patient outcome in BLBC and HGSOC, and that cyclin E1 high BLBC are most similar to CCNE1 non‐amplified HGSOC. Therapies and biomarkers developed for CCNE1‐amplified HGSOC, an essentially BRCA1/BRCA2 wild‐type population, are unlikely to be transferable to BLBC. Importantly, the detection of CCNE1 amplification in BLBC via commercial genomic testing platforms is less likely to offer any benefit for therapy matching, unless accompanied by IHC to confirm cyclin E1 protein overexpression.

While BLBC does not resemble CCNE1‐amplified HGSOC, we can conclude that BLBC and CCNE1 non‐amplified HGSOC that overexpress cyclin E1 protein show similarity in their drivers. Common features are moderate/high cyclin E1 protein, BRCA1 mutation, and possible ‘Brcaness’ phenotypes, as well as high expression of the deubiquitinase USP28 [8] (Figure 6E). Consequently, these cancers may have similar therapeutic targets such as combination Wee1 and PARP inhibitors [56, 57], or combination CDK2 inhibitors and PARP inhibitors [7]. Overall, given the differences identified between HGSOC and BLBC subsets, a cautious approach should be taken rather than grouping these cancers for clinical trials on the basis of CCNE1 amplification or cyclin E1 protein expression status alone.

Author contributions statement

CEC and DA conceived and designed the study. kConFab and AOCS collated and provided materials, and updated patient data. DA and CEC collected, analysed and interpreted the data. CL, KF and ZP undertook experiments and analysed the data. CEC and DA extracted data for the meta‐analysis, and performed the analysis. VC advised on design and implementation of the meta‐analysis. PW provided intellectual input into the study design and interpretation. CEC and DA drafted the manuscript. All authors read and approved the final manuscript.

Ethics approval and consent to participate

Ethics board approval for the kConFab study is described in Mann et al (2008) [58], and was obtained for patients' recruitment, sample collection, and research studies. Written informed consent was obtained from all participants as described [58].

Supplementary materials and methods

Figure S1. Distribution of cyclin E1 H‐scores for BLBC and non‐BLBC of the kConFab cohort

Figure S2. PRISMA schematics for identification of studies to include in meta‐analyses

Figure S3. FUNNEL plots of meta‐analyses

Figure S4. Kaplan–Meier analysis of BLBC cases from the kConFab, TCGA, and Metabric cohorts

Click here for additional data file.