ECPPF (E2F1, CCNA2, POLE, PPP2R1A, FBXW7) stratification: Profiling high-risk subtypes of histomorphologically low-risk and treatment-insensitive endometrioid endometrial cancer

Abstract

In endometrial cancer, occult high-risk subtypes (rooted in histomorphologically low-risk disease) with insensitivity to adjuvant therapies impede improvements in therapeutic efficacy. Therefore, we aimed to assess the ability of molecular high-risk (MHR) and low-risk (MLR) ECPPF (E2F1, CCNA2, POLE, PPP2R1A, FBXW7) stratification to profile recurrence in early, low-risk endometrioid endometrial cancer (EEC) and insensitivity to platinum-based chemotherapy or radiotherapy (or both) in high-risk EEC. Using The Cancer Genome Atlas endometrial cancer database, we identified 192 EEC cases with available DNA sequencing and RNA expression data. Molecular parameters were integrated with clinicopathologic risk factors and adverse surveillance events. MHR was defined as high (-H) CCNA2 or E2F1 log₂ expression (≥2.75), PPP2R1A mutations (-mu), or FBXW7mu; MLR was defined as low (-L) CCNA2 and E2F1 log₂ expression (<2.75). We assessed 164 cases, plus another 28 with POLEmu for favorable-outcomes comparisons. MHR and MLR had significantly different progression-free survival (PFS) rates (P < .001), independent of traditional risk factors (eg, TP53mu), except for stage IV disease. PFS of CCNA2-L/E2F1-L paralleled that of POLEmu. ECPPF status stratified responses to adjuvant therapy in stage III-IV EEC (P < .01) and profiled stage I, grade 1–2 cases with risk of recurrence (P < .001). MHR was associated with CTNNB1mu-linked treatment failures (P < .001). Expression of homologous recombination repair (HR) and cell cycle genes was significantly elevated in CCNA2-H/E2F1-H compared with CCNA2-L/E2F1-L (P<1.0E-10), suggesting that HR deficiencies may underlie the favorable PFS in MLR. HRmu were detected in 20.7%. No treatment failures were observed in high-grade or advanced EEC with HRmu (P = .02). Favorable PFS in clinically high-risk EEC was associated with HRmu and MLR ECPPF (P < .001). In summary, MLR ECPPF and HRmu were associated with therapeutic efficacy in EEC. MHR ECPPF was associated with low-risk, early-stage recurrences and insensitivity to adjuvant therapies.

Introduction

The advent of surgical staging and adjuvant platinum-based chemotherapy (PbCT) in the management of endometrial cancer (EC) occurred 30 years ago, and a plethora of publications addressing their virtues (or lack thereof) have appeared since then. The American Cancer Society reports provide sobering comparisons of estimated cancer statistics for 1991 and 2021—with 33,300 and 66,570 new cases and 5,500 and 12,940 deaths, respectively [1, 2]. From 2008 through 2015, the age-adjusted EC mortality rate was reported to have a 1.9% per annum increase, further stressing the need to characterize therapeutic deficiencies and formulate resolutions [3]. Enhanced knowledge is needed to identify and target vulnerabilities within the oncogenic signaling pathways of EC.

Considering the morphologic similarities to ovarian cancer and high prevalence of TP53 variants in serous EC (SEC), the adaptation of PbCT from high-grade serous ovarian cancer appears reasonable. However, the efficacy of PbCT is generally predicated on deficiencies in homologous recombination repair (HR) [4], and genomic and epigenomic assessments have shown disparities in HR deficiencies of endometrioid EC (EEC) and SEC compared with high-grade serous ovarian cancer [5, 6]. The frequency of HR variants and HR deficiency is less than 20% in SEC and EEC [7, 8], and unsurprisingly, the responses to PbCT are suboptimal for SEC and high-grade EEC [9–12]. In contrast, low-grade EEC generally appears sensitive to PbCT [12–14].

Impaired apoptosis, cell cycle dysregulation, and enhanced DNA damage repair (DDR) are dominant characteristics associated with radioresistance and chemoresistance in solid tumors [15–17]. Our previous studies integrating EC molecular aberrations identified common characteristics among high-risk EC—specifically, CCNA2-E2F1-CIP2A axis dysregulation, coupled with PPP2R1A and FBXW7 variants, were associated with insensitivity to adjuvant therapy and a poor prognosis [18]. Overexpressed CCNA2 binds to E2F1 and converts it from an apoptotic regulator to a potent transcription activator [19]. FOXM1, CIP2A, and multiple cell cycle and HR genes harbor E2F1 activation sites [20–22]; FOXM1 reportedly induces several HR genes, and CIP2A is the nexus to the PI3K-AKT pathway [23]. Upregulation of CIP2A or the presence of variants in PPP2R1A or FBXW7 impede the proteasomal degradation of FOXM1 and multiple cell cycle proteins [24, 25]. The activation of FOXM1 and cell cycle genes, combined with the suppressed degradation of their corresponding proteins, portends overexpression of HR pathway components and insensitivity to DNA-damaging agents in HR-proficient EC [15–17, 20–25].

EEC provides an intriguing model for examining molecular characteristics, including genomic and transcriptomic anomalies, across a spectrum of grades, stages, depths of myometrial invasion (MI), and clinical outcomes. In this study, we assessed the ability of the recently reported E2F1, CCNA2, POLE mutations (-mu), PPP2R1Amu, and FBXW7mu molecular stratification system [18] (hereafter termed ECPPF) to characterize patients with suboptimal outcomes after EEC treatment. We assessed ECPPF in the setting of numerous clinicopathologic parameters, HR and prevalent EC-associated variants, and HR, cell cycle, and related gene expression patterns.

Methods

Study population

The original report describing the prognostic stratification of EC by ECPPF [18] included 239 cases with annotated clinicopathologic, DNA sequencing, RNA expression, and surveillance data that were abstracted from the initial public release of The Cancer Genome Atlas (TCGA) EC database [5]. The prior report was substantially leveraged by the sizeable SEC cohort [18]. The current study (N = 192) was based on the same data set but excluded serous histology to focus on assessment of ECPPF stratification in EEC as a function of grade, depth of MI, stage, expression of select genes, and specific, relevant variants. TCGA required surgical staging, but adjuvant therapy was unknown for approximately half the cases. Nevertheless, PbCT and radiotherapy were the predominant adjuvant therapies during the period of patient accrual; 98% of patients receiving chemotherapy received PbCT [18].

TCGA classification defined 4 subgroups: 1) ultramutated (POLEmu, exonuclease domain mu); 2) hypermutated (microsatellite unstable); 3) copy number low (CNL) or no specific molecular profile (NSMP); and 4) copy number high (CNH; TP53mu/serous-like) [18]. Microsatellite instability (MSI), CNL (NSMP), and CNH were integrated in the EEC ECPPF stratification; CNH was defined as TP53mu, and CNL (NSMP) was estimated as 192 –(POLEmu + MSI + TP53mu). POLEmu were excluded from analyses but served as a metric of favorable clinical outcome.

The cancer genome atlas

We downloaded data from the TCGA EEC database [5]. Data were normalized, formatted, and organized for integration and analysis, as described previously [18, 26].

All data collection and processing, including the consenting process, were performed after approval was obtained from each contributing institution’s local institutional review board or ethics committee and in accordance with the TCGA Human Subjects Protection and Data Access Policies, adopted by the National Cancer Institute and the National Human Genome Research Institute.

Variant analysis

Only validated variants (or TCGA level 3 variants) were used for analysis. Variant information was extracted from exome sequencing data obtained with the Illumina Genome Analyzer GAIIx or HiSeq 2000 sequencing platforms (Illumina, Inc). Silent variants were excluded from the analysis, and only frame-shift insertions and deletions, in-frame insertions or deletions, and missense, nonsense, nonstop, and splice-site variants were included in the study. Two hundred thirty-nine patients from the TCGA EC cohort were reported to have somatic variants in 18,388 unique genes. For our analysis, the number of variants for each selected gene and for each patient were recorded.

Gene expression

Normalized and log-transformed gene expression data from the TCGA EC database were downloaded as level 3 RNA-sequenced data. As previously described, data were collected by Illumina RNA sequencer HiSeq 2000 platforms and annotated with the hg19 version of the human genome.

Gene expression analyses were performed with R statistical packages for computing and graphics (The R Foundation) [27] and the Bioconductor open-source bioinformatics software package [28].

Statistical analysis

Data are descriptively summarized with frequencies and percentages for categorical variables and mean (SD) or median (IQR) for continuous variables. Comparisons of gene expression levels between cohorts was evaluated with the 2-sample t test. Correlations between expression levels were assessed with the Pearson correlation coefficient. Follow-up was calculated from the date of surgery to the date of first documented progression or last follow-up. Progression-free survival (PFS) was estimated with the Kaplan-Meier method. Cox proportional hazards models were fit to evaluate the association between assessed molecular parameters and the risk of progression; associations were summarized with the hazard ratio, and the corresponding 95% CIs were estimated from the models. All calculated P values less than .05 were considered statistically significant. Data were analyzed with SAS software (version 9.4; SAS Institute Inc).

Results

Study population demographic characteristics

DNA sequencing and RNA expression data were annotated for 192 EEC cases in the TCGA EC database, which facilitated assessment of molecular aberrations, clinicopathologic risk factors, and adverse surveillance events [5]. POLEmu were detected in 28 cases (14.6%). These mutations served as a marker of favorable prognoses [5], and cases with POLEmu were excluded from the integrative assessments. The remaining 164 cases constituted our main study cohort (median age, 61 years). In this cohort (Table 1), 124 patients had stage I disease (21 were grade [G] 3); 10 had stage II disease (3 were G3); 23 had stage III disease (6 were G3); and 7 had stage IV disease (6 were G3). PPP2R1Amu were present in 15 patients (9.1%), FBXW7mu were present in 9 (5.5%), and TP53mu were present in 21 (12.8%). When categorizing TP53mu prevalence by tumor grade, 1 of 63 G1 cases had TP53mu, as did 8 of 65 G2 cases and 12 of 36 G3 cases. When assessing variants in HR genes, including ATM, ATR, BRCA1, BRCA2, PALB2, CDK12, BARD1, NBN, CHK1, and Rad51, 1 or more variants were identified in 34 patients (20.7%). Fifty-eight patients (35.4%) had high microsatellite instability (MSI-H) and 85 (51.8%) had CNL (microsatellite stable; NSMP). Recurrences (time from surgery date) were documented for 31 patients (18.9%); for the other 133 patients, the median duration of follow-up was 28.7 months (IQR, 15.9–46.8 months).

Table 1. Clinicopathologic parameters, TCGA subgroups, and select variants, stratified by molecular high-risk and low-risk endometrioid endometrial cancer.

Characteristica	CCNA2-H/E2F1-H (n = 44)	CCNA2-L/E2F1-L (n = 97)	FBXW7mu/PPP2R1Amu (n = 23)	Total (n = 164)
Parameter stage
I	27 (61.4)	78 (80.4)	19 (82.6)	124 (75.6)
II	5 (11.4)	5 (5.2)	19 (82.6)	10 (6.1)
III	7 (15.9)	12 (12.4)	4 (17.4)	23 (14.0)
IV	5 (11.4)	2 (2.1)	0 (0)	7 (4.3)
Histologic grade
1	7 (15.9)	44 (45.4)	12 (52.2)	63 (38.4)
2	15 (34.1)	42 (43.3)	8 (34.8)	65 (39.6)
3	22 (50.0)	11 (11.3)	3 (13.0)	36 (22.0)
Myometrium invasion
≤50%	33 (75.0)	75 (77.3)	19 (82.6)	127 (77.4)
>50%	11 (25.0)	22 (22.7)	4 (17.4)	37 (22.6)
TCGA classification subgroup
POLEmub	n = 19	n = 9	n = 17	n = 28
MSI-H	23 (52.3)	26 (26.8)	9 (39.1)	58 (35.4)
CNL (NSMP)	9 (20.5)	62 (63.9)	14 (60.9)	85 (51.8)
TP53mu	12 (27.3)	9 (9.3)	0 (0)	21 (12.8)
PIK3CA-AKT-FBXW7 pathway variant
PIK3CA	22 (50.0)	58 (59.8)	13 (56.5)	93 (56.7)
PTEN	32 (72.7)	70 (72.2)	12 (52.2)	114 (69.5)
CTNNB1	18 (40.9)	29 (29.9)	7 (30.4)	54 (32.9)
PPP2R1A	3 (6.8)	12 (12.4)	15 (65.2)c	15 (9.1)c
FBXW7	1 (2.3)	8 (8.2)	9 (39.1)c	9 (5.5)c
DNA damage repair variant
ATR	4 (9.1)	6 (6.2)	1 (4.3)	11 (6.7)d
ATM	3 (6.8)	6 (6.2)	1 (4.3)	10 (6.1)d
BRCA2	4 (9.1)	4 (4.1)	1 (4.3)	9 (5.5)d
BRCA1	1 (2.3)	3 (3.1)	0 (0)	4 (2.4)d
PALB2	1 (2.3)	2 (2.1)	0 (0)	3 (1.8)
CDK12	1 (2.3)	2 (2.1)	0 (0)	3 (1.8)
BARD1	1 (2.3)	1 (1.0)	0 (0)	2 (1.2)
NBN	1 (2.3)	1 (1.0)	0 (0)	2 (1.2)
CHEK1	1 (2.3)	0 (0)	0 (0)	1 (0.6)
Rad51	0 (0)	1 (1.0)	0 (0)	1 (0.6)

Abbreviations: CNL (NSMP), copy number low (no specific molecular profile); -H, high; -L, low; MSI-H, high microsatellite instability; -mu, mutation; TCGA, The Cancer Genome Atlas.

^a Data are reported as number of cases (%).

^b Prevalence of POLEmu cases (excluded from analyses) among high- and low-risk cohorts.

^c A single case harbored PPP2R1Amu and FBXW7mu.

^d Among POLEmu cases, the prevalence of ATR, ATM, BRCA2, and BRCA1 variants was 50.0%, 67.9%, 53.6%, and 35.7%, respectively.

ECPPF stratification of EEC

CCNA2 and E2F1 expression and PPP2R1A and FBXW7 mutations, elements in the CCNA2-E2F1-CIP2A axis and PI3K-AKT pathways, respectively, form the basis of the ECPPF stratification scheme (Fig 1A and 1B). PFS was assessed within strata defined by previously reported [18] ECPPF-discriminating molecular parameters: high (-H) CCNA2 or E2F1 log₂ expression (≥2.75; termed CCNA2-H/E2F1-H) vs low (-L) CCNA2 or E2F1 log₂ expression (<2.75; termed CCNA2-L/E2F1-L) and variants in PPP2R1A or FBXW7 (or both). These ECPPF cohorts showed significantly different clinical outcomes (P < .001) (Fig 1C). PFS for patients with CCNA2-L/E2F1-L and POLEmu tumors appeared nearly equivalent. These results suggest that ECPPF status profiles patients with different EEC responses to standard therapy.

Fig 1. Molecular profile and clinical outcomes.

A, Integrated schematic showing the CCNA2-E2F1-CIP2A axis and PI3K-AKT-GSK3β-FBW7 pathway. B, ECPPF stratification was used to identify distinct molecular subgroups from a cohort of 192 patients with endometrioid endometrial cancer. The stratification scheme considers POLE mutations (-mu), PPP2R1Amu, FBXW7mu, CCNA2-H/E2F1-H, and CCNA2-L/E2F1-L; the latter 2 are stratified by TP53 status (TP53mu and TP53 wild type [-wt]). Suffixes -H and -L denote high or low gene expression levels, respectively. C, Progression-free survival rates for 4 primary ECPPF molecular cohorts. D, Progression-free survival for ECPPF molecular cohorts with CCNA2-H/E2F1-H and CCNA2-L/E2F1-L, stratified by TP53mu and TP53wt. E, Cox proportional hazards model with hazard ratios for 6 ECPPF molecular cohorts, using CCNA2-L/E2F1-L/TP53wt as the reference. F, A stepwise multivariate analysis considering competing variables with P < .20 in the univariate analysis (age, grade, depth of myometrial invasion, stage, high microsatellite instability, no specific molecular profile, TP53mu, homologous recombination repair mutations, CTNNB1mu, KRASmu, ARID1Amu, PIK3CAmu, PTENmu, PIK3R1mu, CIP2A expression, and ECPPF cohorts).

Because TP53mu has been associated with poorer EC clinical outcomes [5], we evaluated its integration in ECPPF. Notably, TP53mu (n = 21) were not observed with PPP2R1mu and FBXW7mu (n = 23) and appeared to be mutually exclusive variants (P = .01). TP53 status (ie, mutant vs wild type [-wt]) did not appear to further stratify either CCNA2-L/E2F1-L or CCNA2-H/E2F1-H cohorts (Fig 1D). Cox proportional modeling strengthened these observations, suggesting that ECPPF-stratified oncologic outcomes were independent of TP53mu (Fig 1E). A stepwise multivariate analysis considering the variables of age, grade, depth of MI, stage, MSI-H, CNL (NSMP), TP53mu, HRmu, CTNNB1mu, KRASmu, ARID1Amu, PIK3CAmu, PTENmu, PIK3R1mu, CIP2A expression, and ECPPF parameters (all P < .20 in a univariate analysis) showed independent significance only for ECPPF variables and stage IV disease (vs stage I-II disease) (Fig 1F). These data suggest that clinical outcomes in patients with EEC can be stratified by ECPPF.

Cell cycle and DDR gene expression in ECPPF strata

Disparate responses of ECPPF subgroups to adjuvant therapy suggested potentially divergent levels of DDR and cell cycle genes, as annotated within the CCNA2-H/E2F1-H, CCNA2-L/E2F1-L, and PPP2R1Amu/FBXW7mu cohorts (Table 2). We observed striking, significant differences between the CCNA2-H/E2F1-H and CCNA2-L/E2F1-L cohorts in expression of numerous cell cycle, DDR, and historically prognostic genes. Notwithstanding the substantial difference in PFS between the CCNA2-L/E2F1-L and PPP2R1Amu/FBXW7mu cohorts (Fig 1C), their abridged clinicopathologic profiles (Table 1), and transcriptomic profiles (Table 2, cohort B vs cohort C) were remarkably similar. The extremely low log₂ mRNA expression of DDR and cell cycle genes in the CCNA2-L/E2F1-L cases (Table 2) may signal deficiencies in HR function and thus favorable responses to PbCT and radiotherapy [29–31]. In contrast, the high expression of these genes in CCNA2-H/E2F1-H compared with CCNA2-L/E2F1-L suggests enhanced HR proficiency, and in the absence of HRmu, insensitivity of CCNA2-H/E2F1-H to PbCT or radiotherapy (or both) is likely (Fig 1C).

Table 2. Gene log₂ mRNA expression as a function of molecular high- and low-risk endometrioid endometrial cancer cohorts^a.

Gene	Cohort A: CCNA2-H/E2F1-H (n = 44)	Cohort B: CCNA2-L/E2F1-L (n = 97)	Cohort C: FBXW7mu/PPP2R1Amu (n = 23)	P value, comparison of cohorts
				A to B	B to C
CCNA2-E2F1-CIP2A axis
CCNA2	3.215 (0.787)	1.573 (0.829)	1.910 (0.894)	8.247E-21	.09
E2F1	3.217 (1.055)	1.378 (0.712)	1.796 (0.755)	1.349E-23	.013
CIP2A	2.400 (0.963)	0.895 (0.934)	0.955 (0.994)	5.303E-15	.78
Cell cycle genes
CCNB1	5.297 (0.707)	3.926 (0.686)	4.255 (0.722)	2.473E-20	.043
CCNB2	4.706 (0.697)	3.302 (0.789)	3.579 (0.839)	1.993E-18	.14
CCNE1	3.251 (1.274)	2.117 (1.220)	1.933 (1.013)	1.392E-06	.50
AURKA	3.108 (0.837)	1.644 (0.754)	1.826 (0.699)	6.756E-19	.29
TPX2	4.226 (0.741)	2.653 (0.778)	2.893 (0.761)	2.259E-21	.19
PLK1	4.518 (0.828)	2.823 (0.795)	3.186 (0.878)	4.021E-22	.06
ESPL1	1.971 (0.816)	0.253 (0.844)	0.429 (0.864)	2.011E-21	.38
CHEK1	2.634 (0.698)	1.646 (0.494)	1.881 (0.593)	4.513E-17	.05
FOXM1 and DNA damage repair genes
FOXM1	4.393 (0.735)	2.802 (0.709)	3.024 (0.665)	9.553E-24	.17
BRCA1	1.828 (0.773)	0.714 (0.774)	0.994 (0.738)	6.530E-13	.12
BRCA2	−0.749 (1.233)	−2.288 (1.269)	−2.090 (1.141)	4.141E-10	.50
Rad51	2.209 (0.748)	0.811 (0.779)	1.072 (0.944)	4.500E-18	.17
BRIP1	0.104 (0.893)	−1.148 (0.870)	−0.955 (0.803)	9.950E-13	.33
EXO1	1.487 (0.861)	−0.016 (1.039)	0.445 (0.968)	5.303E-14	.06
MRE11A	1.120 (0.761)	0.672 (0.730)	0.530 (0.796)	.001	.41
Rad50	2.222 (0.780)	2.069 (0.634)	1.840 (0.572)	.22	.12
NBN	2.732 (0.830)	2.387 (0.830)	2.321 (0.790)	.024	.73
ATR	1.700 (0.574)	1.669 (0.504)	1.461 (0.403)	.75	.07
ATM	4.329 (0.708)	4.316 (0.682)	4.212 (0.784)	.92	.53
SKP2	2.916 (0.795)	1.922 (0.751)	1.959 (0.710)	4.500E-11	.83
PARP1	5.431 (0.621)	4.721 (0.795)	4.886 (0.804)	5.925E-07	.38
Historically prognostic markers
ATAD2	2.337 (0.869)	0.833 (0.915)	1.162 (0.754)	5.173E-16	.11
BIRC5	5.883 (0.923)	4.307 (0.780)	4.571 (0.733)	2.581E-19	.14
EZH2	2.993 (0.622)	2.165 (0.628)	2.103 (0.632)	2.407E-11	.67
L1CAM	1.901 (4.754)	0.306 (0.631)	0.396 (1.027)	.0014	.59
STMN	7.121 (1.093)	6.288 (0.951)	6.479 (1.317)	9.617E-06	.43

Abbreviations: -H, high; -L, low; -mu, mutation(s).

^a Gene expression data (normalized and log-transformed) are shown as mean (SD).

The quantitative combined expression (log₂) of CCNA2 and E2F1 (CA2+E2F) was compared with cell cycle and DDR gene expression levels to further weigh the interdependence of these transcriptomic markers. As CA2+E2F expression progressively increased, expression of CIP2A, FOXM1, cell cycle, and HR genes showed a near-parallel increase (Fig 2A). We determined the Pearson correlation coefficients for expression of CA2+E2F and various genes: with CIP2A, r = 0.796; with FOXM1, r = 0.918; with cell cycle genes, r ranged from 0.549 to 0.859; and with HR genes, r ranged from 0.683 to 0.846. Increasing CA2+E2F expression was associated with an elevated prevalence of G3 histology, TP53mu, and recurrences. If we excluded patients with PPP2R1Amu/FBXW7mu, cumulative recurrences sharply increased when CA2+E2F expression exceeded 4.75 (Fig 2B). These results support ECPPF’s potential to profile treatment failure for low-risk tumors and treatment insensitivity for high-risk tumors.

Fig 2. Gene expression and treatment failures.

A, The heat map shows distribution of recurrences, stage, myometrial invasion (MI) >50%, mutation (-mu) in TP53 (TP53mu), homologous recombination repair mutations (HRmu), and high microsatellite instability (MSI-H), according to the increasing quantitative sum of CCNA2 and E2F1 log₂ mRNA expression (CA2+E2F). CIP2A, FOXM1, SKP2, PARP1, and HR and cell cycle gene mRNA expression levels also are shown as a function of increasing expression of CA2+E2F. Group A was defined as the CA2+E2F cohort minus PPP2R1Amu and FBXW7mu cohorts (n = 141); group B was the PPP2R1Amu cohort (n = 14), and group C was the FBXW7mu cohort (n = 9). Patients with POLEmu were excluded from this analysis. B, Cumulative recurrences are shown as a function of the expression sum of CCNA2 and E2F1 (CA2+E2F).

Association of molecular high-risk ECPPF with recurrences in low-risk, stage I EEC

ECPPF’s ability to categorize patients with EEC by clinical outcomes, independent of early stage and histologic grade, suggested that ECPPF might be able to identify early-stage, low-grade cases with heightened risk of occult extrauterine disease. Clinical outcomes were annotated according to stage (and by grade for stage I disease) and were assessed as a function of molecular low-risk ECPPF (MLR; defined as CA2+E2F <4.74) and molecular high-risk ECPPF (MHR; defined as CA2+E2F ≥4.75, PPP2R1Amu, and/or FBXW7mu). Among 103 patients with stage I, G1 or G2 disease, occult extrauterine disease that escaped surgical detection was subsequently documented in 20 patients (19.4%; 5 were MLR, 15 were MHR). When categorizing these patients by grade, 6 of 52 (11.5%) patients with G1 EEC had occult disease, as did 14 of 51 (27.5%) with G2 EEC. Sixty-eight patients (66%) with early-stage, low-grade EEC were categorized as being MLR and 35 (34%) as MHR; the estimated 3-year PFS was significantly different between groups (MLR, 93.0%; MHR, 40.9%; P < .001). The corresponding hazard ratio for MHR (using MLR as the reference) was 10.28 (95% CI, 3.42–30.97; P < .001). For the 85 patients with stage I, G1 or G2, and ≤50% MI disease, who rarely were candidates for adjuvant therapy, ECPPF stratification showed divergent PFS outcomes (P < .001) (Fig 3A). The corresponding hazard ratio (95% CI) for MHR (using MLR as the reference) was 7.50 (95% CI, 2.43–23.13; P < .001).

Fig 3. ECPPF stratification of outcomes in endometrioid endometrial cancer.

A, Progression-free survival for 85 patients with stage I, grade (G) 1 or G2, ≤50% myometrial invasion (MI) disease, stratified by ECPPF as having molecular low risk (MLR; defined as CCNA2 and E2F1 [CA2+E2F] log₂ expression <4.75, PPP2R1A wild type [-wt], and FBXW7wt [n = 56]) or molecular high risk (MHR; defined as CA2+E2F log₂ expression ≥4.75, PPP2R1A mutation [-mu], or FBXW7mu [n = 29]). B, Progression-free survival for 122 patients with stage I-II, G1-G2, <75% MI disease and patients with G3, <50% MI disease, stratified according to ECPPF risk cohorts CA2+E2F <4.75 (n = 75), CA2+E2F ≥4.75 (n = 29), and PPP2R1Amu/FBXW7mu (n = 18). C, Progression-free survival for 28 patients with stage I, G1-G2, ≤50% MI disease and CTNNB1mu, stratified according to MLR (n = 19) and MHR (n = 9) ECPPF. D, Progression-free survival for 42 patients with stage III-IV disease, stage I-II, G2, >75% MI disease, and G3, >50% MI disease, stratified according to MLR (n = 22) and MHR (n = 14). E, Progression-free survival for 30 patients with stage III-IV, stratified according to MLR (n = 16) and MHR (n = 14) ECPPF.

ECPPF stratification similarly discriminated between outcomes when the cohort was expanded to patients with stage I or II, G1 or G2 disease, and <75% MI and patients with G3 and <50% MI (P < .001) (Fig 3B). The corresponding hazard ratio for CA2+E2F ≥4.75 (using CA2+E2F <4.75 as the reference) was 6.34 (95% CI, 1.98–20.35); for PPP2R1Amu/FBXW7mu (using CA2+E2F <4.75 as the reference), the hazard ratio was 8.90 (95% CI, 2.50–31.67; P = .002). These results suggest that ECPPF profiles patients with early-stage, low-risk EEC who have substantial risk of occult extrauterine disease and are candidates for early therapeutic intervention.

Association of MHR ECPPF with compromised survival in CTNNB1mu EEC

CTNNB1mu in low-risk EEC is reportedly associated with compromised survival [32]. CTNNB1mu were identified in 28 patients with early-stage, low-risk EEC (stage I, G1 or G2, ≤50% MI). Seven patients (25%) had documented recurrences, and 6 of these recurrences occurred among the 9 patients in this group with MHR. ECPPF status of patients with early-stage, low-risk CTNNB1mu was associated with differences in survival (P < .001) (Fig 3C). The CTNNB1mu cohort was expanded to include all stage I and II, G1 and G2 cases. Among these 39 cases, 95% of missense mutations were in exons 4 and 5 of the CTNNB1 gene. ECPPF efficiently stratified outcomes (P < .001); the hazard ratio (95% CI) for MHR vs MLR was 19.58 (2.38–161.04; P = .006). These results further underscore the ability of ECPPF to distinguish high-risk patients traditionally thought to have low-risk EEC.

Association of MHR ECPPF with therapeutic insensitivity in patients with high-risk EEC

In the absence of HRmu, the markedly different expression of cell cycle and HR genes in EEC (Table 2) predicts disparate responses to PbCT and radiotherapy [15–17]. Among 40 patients with stage II to IV EEC, 10 therapeutic failures (25%) were documented. Two failures occurred among 21 patients (9.5%) with MLR, and 8 occurred among 19 patients (42.1%) with MHR.

Patients with more advanced EEC (stages III-IV; stages I-II, G2, >75% MI; and stages I-II, G3, >50% MI) invariably receive adjuvant PbCT or radiotherapy (or both). Significant divergence in PFS were observed in this broad at-risk cohort (P = .006) and for patients with stages III to IV disease alone (P = .002) as a function of MLR and MHR ECPPF (Fig 3D and 3E). These results suggest that EEC with MLR ECPPF is responsive to PbCT or radiotherapy (or both), whereas MHR ECPPF likely is insensitive.

Association of MLR ECPPF and MHR ECPPF plus HRmu with favorable prognoses

HRmu were detected in 34 patients (20.7%), with 7 (20.6%) harboring more than 1 HRmu. ATMmu, ATRmu, and BRCA2mu were the most prevalent variants (Fig 4A); 94% of mutations occurred in stages I to II EEC and 76% were classified as MSI-H (Fig 4B). No therapeutic failures were documented in the 18 at-risk patients with HRmu (stage I, G3 [n = 13]; stage II [n = 3]; stage III [n = 1]; and stage IV [n = 1]) who frequently received adjuvant treatment; this group included 12 (66%) with MHR ECPPF (Fig 4C). In contrast, recurrences were documented for 5 of 16 patients (31%) with stage I, G1 to G2 disease, who are seldom candidates for adjuvant therapy; 4 had invasion only into the inner third of the myometrium but 4 also were MHR ECPPF. Among 28 at-risk patients with HRmu (likely candidates for adjuvant therapy) and MLR ECPPF (likely untreated cases), 1 therapeutic failure (3.6%) was documented, whereas for the 6 patients with MHR ECPPF (generally managed with adjuvant therapy), 4 recurrences (66%) were documented (P < .001) (Fig 4D). Collectively, these results suggest that MHR ECPPF with HRmu and MLR ECPPF EEC are sensitive to PbCT or radiotherapy (or both). Patients with MHR ECPPF and HRmu, including those with early-stage, low-grade EEC, must be identified so that they can receive definitive treatment.

Fig 4. Homologous recombination repair mutations (HRmu) and ECPPF integration.

A, Prevalence of the 34 most common HRmu (excluding POLEmu) detected among 164 endometrioid endometrial cancers. B, Prevalence of HRmu according to stage, no specific molecular profile (NSMP), TP53mu, myometrial invasion (MI), microsatellite instability (MSI) high (H) or low (L), grade (G), and ECPPF. C, Progression-free survival among 34 HRmu cases stratified according to traditional risk categories: low risk was defined as stage I, G1-G2 (n = 16), and high risk was defined as stage I, G3 and stages II-IV (n = 18). D, Progression-free survival among 34 patients with HRmu, stratified by stage and ECPPF molecular low risk (MLR) and molecular high risk (MHR).

Discussion

This study provides evidence that the ECPPF molecular classifier can profile patients with low-risk stage I, G1 or G2 EEC, including CTNNB1mu, with substantial risk of recurrence who would benefit from adjuvant therapy. ECPPF also profiles subgroups of patients with high-risk and advanced-stage EEC who have distinct responses to PbCT or radiotherapy (or both). Lastly, ECPPF associates with outcomes in a manner independent of traditional risk factors (eg, TP53mu, CTNNB1mu), except for stage IV disease. CCNA2-L/E2F1-L was associated with excellent PFS rates across traditional risk categories, paralleling survival rates associated with POLEmu. In contrast, CCNA2-H/E2F1-H, PPP2R1Amu, and/or FBXW7mu were associated with significantly compromised recurrence-free survival among patients with low- and high-risk EEC, suggesting that patients with MHR ECPPF likely are insensitive to adjuvant therapies (ie, radiotherapy and/or PbCT).

On the basis of our previous study [18], we postulated that treatment failures in MHR ECPPF, predominantly among patients with HR-proficient EEC, would correlate with overexpression of cell cycle and HR regulatory components, which are hallmarks of resistance to radiotherapy and PbCT [16, 17]. The differential expression of multiple cell cycle and HR genes among patients with CCNA2-L/E2F1-L vs CCNA2-H/E2F1-H was substantial (P<1.0E-10). The large disparity in expression potentially suggests that a functional deficiency in HR may exist and is likely caused by attenuated expression of cell cycle and HR genes in CCNA2-L/E2F1-L [29–31]. This mechanism may account for their favorable responses to PbCT and radiotherapy, even among patients with G3 and advanced-stage EEC. Notably, clinicopathologic characteristics and cell cycle and HR expression profiles for PPP2R1Amu and FBXW7mu mirrored those of CCNA2-L/E2F1-L, but PFS rates were strikingly different, suggesting distinctly diverse oncogenic mechanisms. In contrast, the substantially increased expression of cell cycle and HR genes in CCNA2-H/E2F1-H implies enhanced DNA damage response mechanisms, which is consistent with their frequent lack of sensitivity to PbCT or radiotherapy (or both) [16, 17]. Considering that FOXM1 is reported to induce multiple HR genes [21, 23], the overexpression of FOXM1 and lack of FOXM1 degradation (due to CIP2A overexpression) likely are contributing to greater expression of HR genes [18, 26].

In this study, ATM, ATR, BRCA2, and BRCA1 were the most prevalent HRmu and predominantly were observed with stage I to II, low-grade EEC. Almost no recurrences were observed among patients with high-risk, stage I EEC, patients with advanced-stage EEC with HRmu, and patients with stage I, G1 to G2, MLR EEC who were unlikely to receive adjuvant treatment. In contrast, the treatment failure rate was high among patients who were MHR ECPPF and HRwt, regardless of stage or grade. Considering that HRmu EEC appeared sensitive to adjuvant therapies, identification of MLR and MHR ECPPF with HRmu would facilitate selection of patients who are candidates for molecular-based therapeutic approaches. The molecular distillate from this study suggests that treatment regimens with PbCT or radiotherapy (or both) are efficacious for patients with MLR and MHR ECPPF plus HRmu, but they lack acceptable therapeutic efficacy for patients with MHR ECPPF plus HRwt. Patients with MHR ECPPF plus HRwt, including those with stage I disease, appear to be preferable candidates for innovative phase 1 or 2 clinical trials, regardless of traditional clinicopathologic risk factors.

Mechanistically, CCNA2, E2F1, PP2A, and FBXW7 are pivotal determinants of cell cycle, DDR, and PI3K-AKT signaling dysregulation [18, 20–26]. With CCNA2 upregulation, E2F1 is reportedly converted to a potent transcription activator of FOXM1, CIP2A, cell cycle, and HR genes [19–22]. A previous study from our group provided a schematic to describe the mechanistic integration of projected CCNA2-E2F1-CIP2A axis targets and signaling pathways [18]. The consequent induction of HR genes by FOXM1, coupled with suppressed degradation of FOXM1 and cell cycle proteins (from CIP2A inhibition of PP2A), is consistent with enhanced DDR and insensitivity to DNA-damaging agents [23, 24]. PPP2R1Amu and FBXW7mu likewise markedly impede degradation of these proteins [24, 25, 33]. Navigation of these mechanisms helps characterize EEC molecular vulnerabilities and allows speculation about potential innovative therapeutic options. Such options may include modulators of the CCNA2-E2F1-CIP2A axis (eg, PRMT5 or CIP2A inhibitors), activators of PPP2R1Amu/PP2A, and inhibitors of downstream FBW7/E3 ubiquitin ligase substrates (eg, DDR elements) [33–36].

Limitations of this study include variation in the duration of short- and long-term surveillance, which thereby limited the survival analyses to progression-free intervals. Long-term PFS is likely underestimated, considering that the median surveillance period for recurrence-free cases was less than 30 months and the lower quartile was 17 months. Nevertheless, most patients had early-stage, low-grade EEC, and an unexpected and sizable number of them had documented recurrences, which allowed us to identify significant, disparate molecular associations. Unfortunately, sites of recurrence were not annotated. Additionally, we could not evaluate PFS in the context of specific therapeutic interventions because the details of adjuvant therapies were not available. However, the submitted specimens were predominantly from gynecologic oncology services at comprehensive cancer centers, where standards of care for EEC are well defined. The predominant adjuvant therapies during patient accrual were PbCT or radiotherapy (or both); 98% of patients receiving chemotherapy received PbCT [5]. Strengths of the study include the size of the population, central pathology review, and the robust genomic and transcriptomic analyses that facilitated integration of specific molecular aberrations with clinicopathologic risk factors and adverse events.

In summary, MLR ECPPF and HRmu were associated with therapeutic efficacy in EEC, and MHR ECPPF was associated with low-risk, early-stage recurrences and insensitivity to current adjuvant therapies. Consequently, patients with MHR ECPPF plus HRwt EEC, including those with stage I disease, appear to be candidates for phase 1 or 2 clinical trials evaluating innovative, molecular, target-specific primary therapies.

Abbreviations and gene expansions

ARID1A

AT-rich interactive domain-containing protein 1A

ATM

ataxia-telangiectasia protein

ATR

ataxia telangiectasia and Rad3-related

BARD1

BRCA1-associated RING domain protein 1

BRCA1

breast cancer type 1 susceptibility protein

BRCA2

breast cancer type 2 susceptibility protein

CA2+E2F

quantitative combined expression (log₂) of CCNA2 and E2F1

CCNA2

cyclin A2

CDK12

cyclin-dependent kinase 12

CHK1

checkpoint kinase 1

CIP2A

cancerous inhibitor of protein phosphatase 2A

CNH

copy number high

CNL

copy number low

CTNNB1

catenin β1

DDR

DNA damage repair

EC

endometrial cancer

EEC

endometrioid endometrial cancer

ECPPF

E2F1, CCNA2, POLE, PPP2R1A, FBXW7

E2F1

E2F transcription factor 1

FBW7 or FBXW7

F-box and WD40 domain protein 7

FOXM1

Forkhead box M1

G

grade

-H

high

HR

homologous recombination repair

KRAS

Kirsten rat sarcoma virus

-L

low

MHR

molecular high-risk ECPPF

MI

myometrial invasion

MLR

molecular low-risk ECPPF

MSI-H

high microsatellite instability

-mu

mutation(s)

NBN

nibrin

NSMP

no specific molecular profile

PALB2

partner and localizer of BRCA2

PbCT

platinum-based chemotherapy

PFS

progression-free survival

PI3K-AKT

phosphatidylinositol 3-kinase−protein kinase B

PIK3CA

phosphatidylinositol-4,5-bisphosphate 3-kinase catalytic subunit α

PIK3R1

phosphatidylinositol 3-kinase regulatory subunit α

POLE

DNA polymerase e, catalytic subunit

PPP2R1A

protein phosphatase 2 scaffold subunit α

PP2A

protein phosphatase 2A

PRMT5

protein arginine methyltransferase 5

PTEN

phosphatase and tensin homologue

Rad51

Rad51 recombinase

SEC

serous endometrial cancer

TCGA

The Cancer Genome Atlas

TP53

tumor protein p53

-wt

wild type

Data Availability

All relevant data are included within the manuscript.

Funding Statement

The study was internally funded. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.