TCGA molecular subtypes in endometriosis-associated ovarian cancer: a systematic review and meta-analysis

Endometrioid ovarian cancer (ENOC) represents 10% to 15% of ovarian cancer, exhibiting a relatively favourable prognosis in ovarian cancer [ 1 ]. Clear cell ovarian cancer (CCOC) constitutes 1% to 25% of ovarian cancer, with variations observed among different racial and regional populations [ 2 ]. CCOC often has a worse prognosis and is associated with chemotherapy resistance [ 2–4 ]. The lack of effective screening strategies has resulted in persistently high rates of late-stage disease diagnosis, leading to a poor prognosis. Therefore, more effective and precise treatment strategies are needed to be developed [ 5 ]. Certain studies have shown that endometriosis may be a precursor to ENOC and CCOC. The occurrence of ENOC and CCOC is also closely related to endometriosis [ 6–8 ]. Therefore, it is also referred to as endometriosis-associated ovarian cancer (EAOC). In 2013, the Cancer Genome Atlas (TCGA) classified molecular subtypes of endometrial cancer (EC) as POLE ultramutated, microsatellite instability (MSI), copy number low (CNL) and copy number high (CNH). This classification is valuable for diagnosis, treatment and predicting prognosis. The frequency of common gene mutations in subgroups was reported. The POLE ultramutated subtype exhibited a higher frequency of PTEN mutation, the MSI subtype displayed elevated KRAS mutation, while the CNH subtype predominantly harboured mutation in TP53 , FBXW7 and PPP2R1A , and the CNL subgroup demonstrated distinctive CTNNB1 mutation frequency [ 9 ]. Subsequently, the TransPORTEC subtypes [ 10 ] and the proactive molecular risk classifier for endometrial cancer (ProMisE) subtypes [ 11 ] revealed the four molecular subtypes as POLE proofreading-mutant/POLE mutation (POLEmut), MSI/mismatch repair deficient (MMRd), no specific molecular profile (NSMP)/p53 wild type (p53wt) and p53-mutant/p53 abnormal (p53abn). However, unlike high-grade serous ovarian cancer (HGSOC), EAOC lacks significant prognostic markers such as breast cancer susceptibility gene (BRCA) mutation or homologous recombination repair deficiency. Therefore, it is crucial to understand the molecular characteristics of EAOC. Several studies have indicated a high molecular landscape similarity between EAOC and EC [ 12 , 13 ], with recent investigations revealing analogous mutation frequencies in key genes, including ARID1A , PTEN , PIK3CA , TP53 , CTNNB1 and KRAS [ 14 ]. Based on the TCGA molecular subtypes of EC, this systematic review and meta-analysis aim to investigate the prevalence of TCGA molecular subtypes, staging and prognosis of ENOC and CCOC.

The Preferred Reporting Items for Systematic Reviews and Meta-Analyses Protocol (PRISMA) guideline was applied for reporting this review [ 15 ] ( Supplementary File 1 ). All stages of the review process underwent independent screening by two reviewers (L.Y. and J.C.), with any discrepancies resolved through discussion. In cases where disagreement persisted, the third reviewer (Q.Z.) intervened to facilitate consensus. PubMed, Embase and Web of Science were systematically searched from January 2013 to December 2023 using predefined keywords. The search strategy employed terms such as ‘Ovarian Neoplasms’ AND ‘Endometrioid Carcinoma/Clear Cell’ AND ((‘p53’) OR (‘POLE’) OR ((‘DNA Mismatch Repair’) OR (‘Microsatellite Instability’) OR (‘TCGA’) OR (‘molecular’)) ( Supplementary File 2 ). The exclusion criteria included the inability to obtain complete data for four TCGA subtypes; the research subject is not ENOC or CCOC; reviews and conference abstracts. Two reviewers (L.Y. and J.C.) independently extracted relevant information from each included study, encompassing details of the article such as title, the first author’s name, journal, publication year, country of origin, publication type, corresponding information and any pertinent notes. The prognostic data were extracted from tables and figures within the included articles. The total sample size, number of patients, International Federation of Gynecology and Obstetrics (FIGO) stage in each TCGA subtype and the prognosis data were analysed. The collected data were all from the original study and not modified. About the comparison between ENOC and CCOC, the PICO of this review was defined as [ 15 ] P (population) = patients diagnosed with EAOC; I (intervention or risk factor) = ENOC; C (comparator) = CCOC; O (outcome) = prevalence of the TCGA subtypes. Regarding the survival data analysis, PICO was P = patients diagnosed with ENOC/CCOC; I = POLEmut/MMRd/p53abn subtype; C = NSMP subtype (reference category); O = outcomes. The Joanna Briggs Institute appraisal checklist for studies reporting prevalence data was employed to evaluate the risk of bias [ 16 ]. Responses were categorized as ‘Yes’, ‘No’ or ‘Unclear or Not/Applicable’. Two reviewers (L.Y. and J.C.) assessed the risk of bias assessment, with any discrepancies resolved through discussion. In cases where consensus could not be reached, the third reviewer (Q.Z.) intervened to facilitate resolution. The Stata 14MP (StataCorp LLC, College Station, TX, USA) was employed for data analysis. Prevalence analysis of each TCGA subtype in ENOC and CCOC was conducted based on the number of ENOC and CCOC patients in the included studies. Statistical heterogeneity was quantified using the inconsistency index I 2 . To evaluate the prognosis data of progression-free survival (PFS) and disease-free survival (DFS) in the p53abn subtype of both ENOC and CCOC, hazard ratios (HR) with corresponding 95% confidence interval (CI) were employed, with the NSMP subtype serving as the reference category. The Engauge Digitizer version 12.1 (Engauge Digitizer Project, 2021, December 6) was employed to extract the survival data of the Kaplan–Meier (K-M) curve from the included articles. Subsequently, a calculation spreadsheet was used to calculate the HR [ 17 ].

There were 6 articles with 1,133 ENOC patients [ 18–23 ] and 4 studies with 377 CCOC patients in this research [ 21 , 24–26 ]. Upon subsequent rescreening, 46 studies related to ENOC and 20 studies about CCOC were eligible for inclusion. Following the review process based on predefined exclusion criteria, six articles were ultimately selected for inclusion in the analysis of ENOC. Similarly, in CCOC, adhering to the same exclusion criteria, four articles were suitable for inclusion in this review. The PRISMA flow diagram of the selection process is summarized ( Figure 1(A) and (B) ). PRISMA 2020 flow diagram in EAOC. (A) PRISMA 2020 flow diagram in ENOC. (B) PRISMA 2020 flow diagram in CCOC. *PRISMA: Preferred Reporting Items for Systematic Reviews and Meta-Analyses; TCGA: the Cancer Genome Atlas; ENOC: endometrioid ovarian carcinoma; CCOC: clear cell ovarian carcinoma. The characteristics of the included studies are presented in Tables 1 and 2 . The sample size of ENOC ranged from 36 to 511 across the included studies [ 18–23 ], while for CCOC, the range was from 72 to 115 [ 21 , 24–26 ] . The POLEmut subtype was assessed by sequencing in all studies [ 18–26 ]. Regarding the MMRd and p53abn subtypes, evaluation was performed using immunohistochemistry in six studies [ 18 , 20–24 ], with three additional studies using sequencing methods as detailed in the provided articles [ 19 , 25 , 26 ]. Consequently, the ProMisE subtypes were employed in this review. Characteristics of the included studies within ENOC. *ENOC: endometrioid ovarian carcinoma; TCGA: the Cancer Genome Atlas; POLEmut: POLE mutation; MMRd: mismatch repair deficient; p53abn: p53 abnormal; NSMP: no specific molecular profile; IHC: immunohistochemistry. Characteristics of the included studies within CCOC. *CCOC: clear cell ovarian carcinoma; TCGA: the Cancer Genome Atlas; POLEmut: POLE mutation; MMRd: mismatch repair deficient; p53abn: p53 abnormal; NSMP: no specific molecular profile; IHC: immunohistochemistry. The data we ultimately included in the analysis were the prevalence of each TCGA subtype in ENOC and CCOC, the prevalence of each TCGA subtype with FIGO stage > I in ENOC ( Table 3 ), and the prognosis data. There was a lack of consistent prognostic indicators in the included studies, for ENOC, prognosis data encompassed assessments of PFS in three studies [ 19 , 20 , 22 ] and DFS in two studies [ 18 , 21 ]. For CCOC, prognosis data comprised the assessments of DFS in two studies [ 21 , 25 ]. Prevalence of the TCGA subtypes in ENOC of FIGO stage > I. *TCGA: the Cancer Genome Atlas; ENOC: endometrioid ovarian carcinoma; FIGO: International Federation of Gynecology and Obstetrics; POLEmut: POLE mutation; MMRd: mismatch repair deficient; p53abn: p53 abnormal; NSMP: no specific molecular profile. The risk of bias assessment of all included studies is illustrated in Figure 2 . Notably, all studies demonstrated a low risk of bias in adequately describing the study subjects and settings in detail. Furthermore, each article employed valid and standard measures, for appropriate statistical analyses, and managed the low response rate appropriately. However, it is noteworthy that all studies exhibited limitations related to sample sizes and insufficient coverage of the identified sample. The risk of bias using the Joanna Briggs appraisal checklist for prevalent studies. The prevalence of the TCGA subtypes in ENOC was assessed, with POLEmut = 3.9% (95% CI: 2.7%–5.2%, I 2 = 3.489%, p  = 0.394), MMRd = 12.1% (95% CI: 8.8%–15.9%, I 2 = 62.406%, p  = 0.021), p53abn = 15.7% (95% CI: 9.3%–23.4%, I 2 = 88.859%, p  = 0.000), NSMP = 67.0% (95% CI: 61.0%–72.7%, I 2 = 72.347%, p  = 0.003) ( Figure 3(A)–(D) ). Prevalence of each TCGA subtype in EAOC. (A) The prevalence of POLEmut in ENOC. (B) The prevalence of MMRd in ENOC. (C) The prevalence of p53abn in ENOC. (D) The prevalence of NSMP in ENOC. (E) The prevalence of POLEmut in CCOC. (F) The prevalence of MMRd in CCOC. (G) The prevalence of p53abn in CCOC. (H) The prevalence of NSMP in CCOC. *TCGA: the Cancer Genome Atlas; EAOC: endometriosis-associated ovarian cancer; ENOC: endometrioid ovarian carcinoma; CCOC: clear cell ovarian carcinoma; POLEmut: POLE mutation; MMRd: mismatch repair deficient; p53abn: p53 abnormal; NSMP: no specific molecular profile. The distribution of the TCGA subtypes in CCOC was assessed as follows, POLEmut = 1.4% (95% CI: 0.1%–3.9%, I 2 = 48.443%, p  = 0.121), MMRd = 3.8% (95% CI: 2.0%–6.1%, I 2 = 0%, p  = 0.441), p53abn = 13.1% (95% CI: 8.0%–19.3%, I 2 = 61.893%, p  = 0.049), NSMP = 81.2% (95% CI: 73.0%–88.2%, I 2 = 72.433%, p  = 0.012) ( Figure 3(E)–(H) ). ENOC had a higher frequency of the POLEmut subtype (OR = 2.29, 95% CI: 1.03–5.11, p  = 0.043) and the MMRd subtype (OR = 3.54, 95% CI: 2.05–6.11, p  = 0.000) than CCOC; ENOC had a lower frequency of the NSMP subtype (OR = 0.55, 95% CI: 0.41–0.73, p  = 0.000) and the p53abn subtype (OR = 0.97, 95% CI: 0.67–1.42, p  = 0.893) ( Figure 4 ). OR for the comparison of each TCGA subtype between ENOC and CCOC. *OR: odds ratio; TCGA: the Cancer Genome Atlas; ENOC: endometrioid ovarian carcinoma; CCOC: clear cell ovarian carcinoma. The presence of the TCGA subtypes in ENOC FIGO stage > I were POLEmut = 31.2% (95% CI: 16.8%–47.3%, I 2 = 0%, p  = 0.697), MMRd = 48.4% (95% CI: 16.3%–81.2%, I 2 = 91.050%, p  = 0.000), p53abn = 52.0% (95% CI: 33.5%–70.3%, I 2 = 66.725%, p  = 0.029), NSMP = 28.4% (95% CI: 16.3%–42.2%, I 2 = 90.890%, p  = 0.000) ( Figure 5(A)–(D) ). Prevalence of the TCGA subtypes in ENOC of FIGO stage > I. (A) The prevalence of POLEmut in ENOC of FIGO stage > I. (B) The prevalence of MMRd in ENOC of FIGO stage > I. (C) The prevalence of p53abn in ENOC of FIGO stage > I. (D) The prevalence of NSMP in ENOC of FIGO stage > I. *TCGA: the Cancer Genome Atlas; ENOC: endometrioid ovarian carcinoma; FIGO: International Federation of Gynecology and Obstetrics; POLEmut: POLE mutation; MMRd: mismatch repair deficient; p53abn: p53 abnormal; NSMP: no specific molecular profile. The POLEmut subtype survival data achieved 100% in ENOC. However, for the MMRd subtype, the survival meta-analysis reached 100% in two studies [ 18 , 19 ], resulting in the HR of the MMRd subtype not being analysed. Subsequently, the survival data for the p53abn subtype were analysed, with the NSMP subtype serving as the reference category. Eventually, the HR of the p53abn subtype through several subgroup analyses was assessed as DFS (HR = 3.25, 95% CI: 1.46–7.21, p  = 0.004); PFS (HR = 4.11, 95% CI: 2.86–5.92, p  = 0.000) ( Figure 6(A) and (B) ). Prognosis of the p53abn subtype in EAOC of DFS/PFS. (A) The HR of DFS in ENOC. (B) The HR of PFS in ENOC. (C) The HR of DFS in CCOC. *p53abn: p53 abnormal; EAOC: endometriosis-associated ovarian cancer; DFS: disease-free survival; PFS: progression-free survival; HR: hazard ratios; ENOC: endometrioid ovarian carcinoma; CCOC: clear cell ovarian carcinoma. In CCOC, the survival data of POLEmut subtype [ 21 , 25 ] and MMRd subtype [ 21 ] were 100%. As in ENOC, the p53abn subtype was assessed and the NSMP subtype was defined as a reference. The DFS of the p53abn subtype in CCOC was evaluated (HR = 5.52, 95% CI: 3.43–8.90, p  = 0.000) ( Figure 6(C) ).

The objective of this review was to evaluate the prevalence of TCGA subtypes in ENOC and CCOC, the prevalence of FIGO stage > I in each subtype, the outcomes associated with the p53abn subtype in EAOC and the comparative analysis of TCGA subtypes prevalence between ENOC and CCOC. DNA polymerase ε (DNA polymerase epsilon, POLE), crucial for DNA replication and synthesis, plays a key role in reducing spontaneous mutations and cancer risk [ 27 , 28 ]. The POLEmut subtype of EC is associated with ultramutation, favourable prognosis and responsiveness to immunotherapy [ 29 ]. Based on the prognostic analysis in the included studies, the POLEmut subtype had the best outcome during the follow-up period, in which PFS/DFS/5-year recurrence-free survival (RFS) were 100%. The DNA mismatch repair (MMR) system is instrumental in maintaining and repairing DNA replication, and MMRd can lead to MSI and subsequent cancer development [ 30 ]. Survival analysis revealed the outcomes for the MMRd subtype similar to the NSMP subtype, which was between the POLEmut and the p53abn subtypes. The TP53 gene, a tumour suppressor gene, is frequently mutated in cancer and is an indicator of poor survival [ 31 ]. Notably, the p53abn subtype had the poorest outcome. Recent research has suggested that CTNNB1 and TP53 mutations are mutually exclusive in ENOC, with better prognosis observed in CTNNB1 mutation cases [ 32 ]. The NSMP subtype was defined without the POLEmut/MMRd/p53abn characteristics and is also referred to as p53wt/microsatellite stable. This subtype constitutes the largest proportion of patients, which needs further investigation into its molecular characteristics to guide clinical practice. Comparative analysis between ENOC and CCOC revealed more pronounced molecular subtype distribution in ENOC, possibly due to sample size limitations in CCOC studies. Further research is warranted to elucidate the molecular subtype characteristics of CCOC. The p53abn subtype had the highest proportion of FIGO stage > I and the worst outcome. However, according to these findings, the prognosis of the POLEmut subtype and the MMRd subtype was not as poor as the true staging, the POLEmut subtype and the MMRd subtype had more than half the proportion of middle and advanced cases, but the overall survival (OS) and DFS were 100% [ 18 ]. In CCOC, the POLEmut subtype accounted for about a third of middle and advanced cases, but the DFS and OS were 100% [ 25 ]. Therefore, molecular subtypes hold significant promise in guiding staging and prognostic assessment. As is widely acknowledged, several factors such as advanced stage, surgery, chemoresistance, metastasis and TP53 mutation were considered prognostic predictors [ 32 , 33 ]. Notably, CCOC was associated with chemoresistance and worse survival outcomes. However, the prognosis was suboptimal in both advanced ENOC and CCOC [ 34 ]. This article aims to explore the TCGA molecular subtype characteristics of EC in EAOC to develop the precise treatment strategies. Several studies have confirmed the clonal relationship between synchronous EAOC and EC [ 12 , 35 ]. Common mutations shared by EAOC and EC include PTEN , KRAS , ARID1A and CTNNB1 [ 34 , 36 ]. There is a lack of molecular characteristics corresponding to the EAOC subtype in the available studies. According to the TCGA subtypes of EC [ 9 ], the POLEmut subtype had a higher PTEN mutation, and the MMRd subtype had a common ARID1A mutation. Studies have shown that ENOC molecular subtypes exhibit a mutational landscape similar to EC, with PTEN mutation being more prevalent in the POLEmut subtype and ARID1A mutation being more common in the MMRd subtype [ 18 , 23 ]. The MMRd subtype of EC has high neoantigen load, making it more sensitive to immunotherapy [ 37 ]. Preliminary results from related clinical research suggest that immunotherapy can improve the oncologic outcomes for EC [ 38 ]. Meanwhile, existing studies indicate that some CCOC are sensitive to immunotherapy [ 39 ]. In CCOC, evaluating MSI status, tumour-infiltrating lymphocytes and programmed cell death 1 (PD-1)/programmed cell death ligand 1 (PD-L1) expression could guide immunotherapy selection [ 40 ]. The ARID1A mutation in CCOC has been correlated with increased tumour immunogenicity [ 41 , 42 ], and the NSMP subtype may be more frequently associated with high PD-L1 positive (combined positive score > 1) [ 25 ], which may benefit from immunotherapy. Therefore, it warrants further exploration of CCOC molecular subtypes combined with molecular characterization of immunotherapy markers. In ENOC, high progesterone receptor (PR) expression has been found to correlate with CTNNB1 mutation and is associated with a favourable prognosis, whereas low PR expression is observed in cases with p53 mutation and carries a poor outcome [ 36 ]. Consequently, investigating the relationship between hormone receptor distribution characteristics and the molecular stratification of ENOC in terms of treatment and prognosis is a worthwhile area of further research. The TCGA molecular subtypes of EC can effectively guide treatment. For instance, low-risk POLEmut ECs in FIGO I-II stages may be spared adjuvant therapy, as recommended by the ESGO-ESTRO-ESP guideline [ 43 ]. Additionally, the FIGO 2023 endometrial cancer staging system has incorporated molecular subtypes results [ 44 ]. Noteworthy findings such as the PORTEC-3 trial of EC indicate the significant improvements in RFS and OS for the p53abn subtype with combination adjuvant chemotherapy compared to radiotherapy [ 45 ]. As EC has done, exploring precision treatment patterns for EAOC based on molecular subtypes can improve the patient outcomes. In the future, molecular subtypes are expected to be recommended for EAOC. Limitations of this meta-analysis include insufficient sample sizes, incomplete staging details and different indicators integrated into the survival analyses of the included studies. Therefore, it did not allow for complete prognostic analysis of the four molecular subtypes. This has also resulted in the staging of data for the four subtypes of CCOC that cannot be adequately assessed. Differences in populations, sample sizes and measurement methods are likely to result in high heterogeneity. Additionally, the incidence of EAOC is low and is not comparable to HGSOC. However, the main focus of current research is HGSOC, and fewer studies have concentrated on EAOC. Large-scale, multi-centre, prospective clinical research in EAOC is lacking, impeding the clinical validation of relevant targeted markers. Moreover, little research has been conducted on the differences in gene mutations among the subtypes. Despite these limitations, this study provides valuable insights into the prevalence and prognostic characteristics of TCGA molecular subtypes in ENOC and CCOC. Future studies focusing on the TCGA molecular subtyping of EAOC are being looked forward to so that molecular mechanisms can be explored and patient prognosis can be further improved.

In conclusion, the TCGA subtypes of EC may exhibit similarities in prognosis between ENOC and CCOC. Some of the molecular subtypes demonstrated the potential of EAOC in immunotherapy or endocrine therapy. More studies will be included in the future to further refine EAOC molecular subtypes. This will lead to more precise treatment strategies for EAOC, as well as risk stratification.