Association of a novel endometrial cancer biomarker panel with prognostic risk, platinum insensitivity, and targetable therapeutic options

During the past decade, the age-adjusted mortality rate for endometrial cancer (EC) increased 1.9% annually with TP53 mutant (TP53-mu) EC disproportionally represented in advanced disease and deaths. Therefore, we aimed to assess pivotal molecular parameters that differentiate clinical outcomes in high- and low-risk EC. Using the Cancer Genome Atlas, we analyzed EC specimens with available DNA sequences and quantitative gene-specific RNA expression data. After polymerase ɛ (POLE)-mutant specimens were excluded, differential gene-specific mutations and mRNA expressions were annotated and integrated. Consequent to TP53-mu failure to induce p21, derepression of multiple oncogenes harboring promoter p21 repressive sites was observed, including CCNA2 and FOXM1 (P < .001 compared with TP53 wild type [TP53-wt]). TP53-wt EC with high CCNA2 expression (CCNA2-H) had a targeted transcriptomic profile similar to that of TP53-mu EC, suggesting CCNA2 is a seminal determinant for both TP53-wt and TP53-mu EC. CCNA2 enhances E2F1 function, upregulating FOXM1 and CIP2A, as observed in TP53-mu and CCNA2-H TP53-wt EC (P < .001). CIP2A inhibits protein phosphatase 2A, leading to AKT inactivation of GSK3β and restricted oncoprotein degradation; PPP2R1A and FBXW7 mutations yield similar results. Upregulation of FOXM1 and failed degradation of FOXM1 is evidenced by marked upregulation of multiple homologous recombination genes (P < .001). Integrating these molecular aberrations generated a molecular biomarker panel with significant prognostic discrimination (P = 5.8×10−7); adjusting for age, histology, grade, myometrial invasion, TP53 status, and stage, only CCNA2-H/E2F1-H (P = .0003), FBXW7-mu/PPP2R1A-mu (P = .0002), and stage (P = .017) were significant. The generated prognostic molecular classification system identifies dissimilar signaling aberrations potentially amenable to targetable therapeutic options.


Introduction
The American Cancer Society (ACS) predicted 61,880 new cases and 12,160 deaths that would be attributable to endometrial cancer (EC) in 2019 [1]. In 2018, the ACS reported an alarming 1.9% annual increase during the decade in age-adjusted mortality for EC [2]-a trajectory needing reversal. Standard treatment for high-risk EC is definitive surgery followed by systemic platinum-based chemotherapy (PbCT) or radiotherapy, or both. Sensitivity to PbCT positively correlates with deficiencies in the homologous recombination (HR) pathway [3]. However, the majority of ECs are HR proficient; thus, tailored molecular-based therapy needs to be developed, which requires identifying molecular profiles that harbor targetable aberrations.
Integrating the above generic TP53 mechanistic information with data available from the EC literature, we developed a working schematic (Fig 1A) for comparing the mRNA expression between TP53-mu and TP53-wt EC for numerous genes that impact cell-cycle dynamics, apoptosis, and DNA-damage repair. We identified the seminal role of CCNA2 in 1) integrating the TP53-p21-CDE/CHR and PI3K-AKT-FBW7 pathways and 2) combining with E2F1 overexpression and mutations in FBXW7 and PPP2R1A in determining outcomes of both TP53-mu and TP53-wt EC. An untoward commonality included induction of FOXM1 or failed degradation of FOXM1, or both, which portends enhanced HR gene expression and potential insensitivity to chemotherapy.

The Cancer Genome Atlas
We obtained and analyzed TCGA (www.cancergenome.nih.gov) data as previously described [22]. TCGA contains comprehensive genomic information including copy number variation, single-nucleotide polymorphisms, miRNA expression, gene expression, and DNA methylation data, as well as clinical and outcome information. Data from TCGA were downloaded,

PLOS ONE
normalized, formatted, and organized for integration and analysis with other biological datasets in accordance with TCGA data-sharing agreements. Somatic mutations and gene expression data were recorded. All data collection and processing, including the consenting process, were performed after approval by each of the participating institution's local institutional review board/ethics committee and in accordance with TCGA Human Subjects Protection and Data Access Policies, adopted by the National Cancer Institute and the National Human Genome Research Institute.

Mutation analysis
Somatic mutation detection, calling, annotation, and validation from TCGA have been described [23]. Somatic mutation information resulting from exome sequencing with the Illumina Genome Analyzer DNA Sequencing GAIIx or HiSeq 2000 platforms (Illumina Inc) was downloaded and formatted for analysis. Mutation information was downloaded as level 3 or validated somatic mutations.
Of the 239 cancerous endometrial tumors included, we identified 18,388 unique genes with 138,838 validated somatic mutations, including frame-shift insertions and deletions; in-frame insertions or deletions; and missense, nonsense, nonstop, and splice-site mutations. Silent mutations were excluded from the analysis. The number of mutations for each selected gene was recorded for each patient.

Gene expression
Gene expression data were downloaded from TCGA data repository as level 3 RNA sequence data [4] created by Illumina RNA Sequencer HiSeq 2000 platforms (Illumina Inc) and annotated with the HG-19 version of the human genome. Normalized and log-transformed gene expression data from these endometrial tumors were available for analysis. Analyses were performed with R statistical packages (R Foundation) for statistical computing and graphics [24] and bioconductor packages as open-source software for bioinformatics [25]. For the front end, we used Biometric Research Branch Array Tools, an integrated package for visualization and statistical analysis that uses Excel (Microsoft Corp) [26].

Cell lines and in vitro assessments
As PbCT is the predominant adjuvant therapy for high-risk EC, which are frequently insensitive to therapy [2,[27][28][29][30], we chose cell lines recognized as platinum insensitive with identified mutational anomalies associated with adverse clinical outcomes in the study population. ARK2, a uterine serous carcinoma (USC) (type II) derived cell line, harbors mutant TP53 and wt FBXW7 and PPP2R1A (personal communication with A. Santin, Yale University) [31]. HEC-1B cells (endometrioid endometrial carcinomas [EEC]; type I) have mutations in TP53, FBXW7, and PPP2R1A [32]. Both cell lines were cultured in Dulbecco's Modified Eagle's Medium containing 10% fetal bovine serum, 100 mcg/mL streptomycin, 100 units/mL penicillin, and 2 mM L-glutamine. Cells were maintained in an incubator at 37˚C in an atmosphere containing 5% CO 2 . Carboplatin and panobinostat (HDAC10 inhibitor) were purchased from ApexBio.

Real-time polymerase chain reaction
Total RNA was isolated using RNeasy Plus MiniK (Qiagen). cDNA was synthesized using a Reverse Transcription Kit (Applied Biosystems). Real-time polymerase chain reaction (PCR) was performed using the SYBR Green PCR Master Mix (ThermoFisher Scientific) on the LightCycler 480 (Roche Molecular Systems Inc). The sequences of primers for the analyzed genes are detailed in S1 Table. Western blot analysis ARK-2 cells were treated with panobinostat at 10 nM. After incubation for 3, 6, 12, and 24 hours, cell lysates were collected for protein expression analyses and compared with untreated (time = 0) controls. Expression of p21, FOXM1, acetylated-H3, and GAPDH were measured by Western blot. Antibodies used in this study were P21 (Cell Signaling Technology, 2947), FOXM1 (Cell Signaling Technology, 5436), acetyl-H3 (Millipore, 06-599), and GAPDH (Sigma-Aldrich, G8795).

MTT assay and synergy assessment
Three thousand cells per well were seeded in triplicate in 96-well plates and the cells treated with increasing concentrations of panobinostat and carboplatin for 72 hours, respectively. MTT-based CellTiter 96 Aqueous One Solution Cell Proliferation Assay (Promega Corp) was performed (per manual) to assess half-maximum inhibitory concentration. Constant-ratio studies were performed to investigate the combinatory effect of carboplatin with panobinostat in HEC-1B and ARK-2 cell lines [33].

Statistical analysis
For each candidate gene surveyed, TCGA-quantitated expression levels of the corresponding mRNA were annotated for the 239 specimens. Comparisons between groups were evaluated with the χ 2 test for nominal variables and the 2-sample t test for continuous variables. Correlations were quantified by using Pearson correlation coefficients. All calculated P values were 2-sided.

Progression-free survival analysis
Statistical methods for survival data were used to analyze progression-free survival (PFS), defined as the time from surgery to disease recurrence. Patients without evidence of disease at the end of follow-up were treated as censored observations. Comparisons between Kaplan-Meier survival curves were performed with log-rank tests. For association with survival, all clinicopathologic variables were assessed with Cox proportional hazard regression. All variables associated with survival with a univariate P value �.05 were included in an initial multivariate regression model. Those variables with the smallest contributory effect were excluded with a backward elimination technique based on the Akaike information criterion (measure of the quality of the model for a given dataset). Hazard ratios (95% CI) were reported. Analyses were performed using R statistical computing and graphics [24].

Oncogene expression correlation with CCNA2 and E2F1
The overexpression of E2F1 and concomitant TP73 suppression in TP53-mu EC suggested, as previously reported, upregulation of CCNA2, which determines the mode of action of E2F1 [8,34]. Thus, we examined the correlation between reference oncogenes (E2F1 and CCNA2) and multiple direct or downstream targets of E2F1 in TP53-mu and TP53-wt EC (Fig 1C). Correlation coefficients for the reference genes in TP53-mu tumors were similarly positive with regard to cell-cycle genes, but the positivity was substantially higher for CCNA2 than E2F1 for MASTL1, CIP2A, and HR pathway genes. Unexpected were the high positive correlations in TP53-wt tumors between the expressions of CCNA2 and E2F1 targets and HR pathway genes, which paralleled the correlations in TP53-mu tumors. These results suggested a potential role for CCNA2 in the carcinogenesis of both TP53-mu and a subset of TP53-wt tumors.

Comparative expression of oncogenes as a function of TP53-mu and TP53wt CCNA2 expression
The expression of multiple, upregulated oncogenes in TP53-mu EC was assessed in TP53-wt EC with high CCNA2 expression. The upper quartile of annotated CCNA2 mRNA expression levels among TP53-wt specimens (�2.6) was arbitrarily designated as high expression (CCNA2-H). When the expression of multiple CCNA2/E2F1 target and HR-pathway genes in TP53-wt CCNA2-H and TP53-mu EC was assessed, equivalency or higher expression was shown for most assessed genes in TP53-wt CCNA2-H vs TP53-mu specimens (Table 1). Noteworthy was the dramatic upregulation of FOXM1, CIP2A, and multiple HR genes in both TP53-mu and TP53-wt CCNA2-H EC compared with TP53-wt with CCNA2 low expression (CCNA2-L).

Recurrences in traditional low-risk and high-risk EC according to biomarker panel cohorts
Contemporary adjuvant therapy for low-risk EC (stage 1 or 2, grade 1 or 2) is generally limited. These low-risk tumors significantly (P < .0001) stratified according to molecular-panel cohorts. The estimated 5-year PFS for low-risk EC with the low-risk biomarker profile (CCNA2-L/E2F1-L/FBXW7-wt/PPP2R1A-wt) (n = 75) was 92% compared with 31% for the low-risk EC with the high-risk biomarker profile (CCNA2-H/E2F1-H or FBXW7-mu/

Induction of p21 and repression of panel-specific targets
The molecular schematic (Fig 1A) predicts that CDKN1A (p21) induction in TP53-mu tumors would repress multiple oncogenes with downstream suppression of corresponding targets. Histone deacetylase inhibitors (HDACi) have been reported to induce p21 in TP53-mu cell lines [38]. The platinum-insensitive cell lines ARK-2 and HEC-1B were exposed to panobinostat, an HDAC10 inhibitor, and qPCR expression of targeted genes analyzed. Increased expression of CDKN1A (p21) with downregulation of CCNA2, E2F1, CIP2A, FOXM1, and EXO1 was observed in both cell lines (Fig 5A and 5B).
ARK-2 cells were treated with 10 nM panobinostat and protein expression assessed via Western blot. Increased expression of p21 and acetyl-H3 and down-regulation of FOXM1 expression occurred in a time-dependent manner (Fig 5C). The stimulatory effect on p21 and the inhibitory effect of FOXM1 expression in response to panobinostat are consistent with the results observed in real-time PCR analysis (Fig 5A).

Synergism with HDACi and carboplatin in platinum-insensitive cell lines
The downregulation of FOXM1 and HR pathway EXO1 with panobinostat in platinum-insensitive cell lines suggested the potential for HDACi to enhance platinum sensitivity. Synergism occurred in ARK-2 and HEC-1B cell lines exposed to varying concentrations of carboplatin and panobinostat (Fig 5D and 5E). These observations suggested that suppression of FOXM1 and HR pathway components might enhance platinum sensitivity in high-risk HR-proficient EC.

Discussion
To our knowledge, this is the first report of a classification system for EC that appears to correlate with oncologic outcomes independent of patient age, histology, tumor grade, myometrial invasion, and TP53 mutational status. The discriminatory PFS value of the cohorts in the molecular biomarker panel was predicated on the overexpression of CCNA2 and E2F1 or mutations in FBXW7 or PPP2R1A. These observations constitute a mechanistic commonality regardless of TP53 status that is equally applicable in MSI-H and CNV-L cohorts. Pivotal is the interactive role of CCNA2 and E2F1 in upregulating FOXM1 transcription and inducing CIP2A activation, predictably leading to PP2A inhibition and likely restriction of FOXM1 degradation [19-21, 35, 39, 40]. The latter is likewise anticipated with FBXW7 and PPP2R1A mutations. FOXM1 reportedly induces multiple HR genes such as BRIP1, BRCA2, EXO1, and Rad51 [18], all of which were overexpressed in the poor prognostic molecular biomarker cohorts. The mechanistic molecular distillate from our observations suggests that the overexpression of multiple HR-pathway genes expectedly limits responses in the majority of HR-proficient ECs treated with DNA-damaging agents. The 1.9% annual increase in age-adjusted mortality for EC observed over the past decade warrants reappraisal of contemporary therapeutic algorithms [2]. Our recent institutional assessments coupled with subgroup analyses in select randomized clinical trials suggest that PbCT has suboptimal efficacy for managing high-risk EC [27][28][29][30]. Considering that most EC is HR proficient [4], augmenting HR components, several of which are induced by FOXM1, would presumably enhance DNA-damage repair, yielding insensitivity to DNA-damaging agents such as platinum [18]. This study confirms the marked upregulation of HR components in high-risk EC.
The integrated signaling pathways shown in Fig 1 illustrate the mechanisms that lead to simultaneous upregulation of FOXM1 and downregulation of FOXM1 degradation in TP53mu and TP53-wt with CCNA2-H and/or E2F1-H. The failure of TP53-mu to induce CDKN1A (p21) derepresses FOXM1, and with the upregulation of E2F1, FOXM1 expression is further augmented [18]. The mechanism of action of E2F1 is predicated on CCNA2; high CCNA2 projects a proliferative E2F1 mode [8]. E2F1 and CCNA2 were both upregulated in TP53-mu and a subset of TP53-wt EC. Overexpression of CCNA2 has previously been correlated with TP53 expression, chemoresistance, and poor prognosis in EC [9,10]. We showed for the first time that TP53-wt EC with high CCNA2 expression is associated with molecular aberrations and clinical outcomes similar to TP53-mu EC. The mechanism responsible for high expression of CCNA2 in the subset of TP53-wt EC is unknown.
The prognostic biomarker panel that incorporates CCNA2/E2F1 upregulation and PPP2R1A/FBXW7 mutations is highly discriminatory. Without these molecular aberrations, clinical outcomes are very favorable and appear to be similar to those of POLE-mu tumors. Importantly, the majority of EC is HR proficient, which predicts a high prevalence of platinum insensitivity in biomarker panel-positive patients. Suppressing the induction of FOXM1 or enhancing degradation of FOXM1, or both, thereby downregulating HR components, might potentially facilitate conversion to platinum sensitivity. Exemplary exposure of platinuminsensitive TP53-mu EC cell lines to panobinostat [38], an HDACi currently in clinical trials, resulted in induction of CDKN1A (p21), suppression of CCNA2, CIP2A, FOXM1, and EXO1, and synergism with carboplatin at nM levels of panobinostat.
The strengths of this study include the robustness of TCGA annotated database, which includes specimens obtained at cancer centers dedicated to definitive management of patients with EC. The study is limited by the lack of biomarker-panel validation in a similar, sizeable population having definitive staging, central pathology review, standardized treatment, extended surveillance, and focused molecular analysis. The unavailability of detailed treatment algorithms and reliable long-term disease-specific survival documentation limited correlations of molecular irregularities to PFS and clinicopathologic parameters.
In summary, the integration of CCNA2 and E2F1 overexpression and POLE, PPP2R1A and FBXW7 mutations generated a molecular EC classification that projects prognostic risk, platinum insensitivity, and potential targetable therapeutic options.