Introduction

Endometrial cancer (EC) is the sixth most common cancer in women worldwide, with over 400,000 new cases per year; moreover, the incidence and mortality of this cancer are increasing [1]. Although prognostic indicators such as histologic subtype, patient age, tumor grade, stage, and lymphovascular space invasion (LVSI) are instrumental in guiding treatment, their use in risk stratification is challenging, primarily because of variability in interpretation [2, 3]. These challenges have highlighted the need for enhanced surgical and therapeutic planning strategies.

The Cancer Genome Atlas (TCGA) has made significant strides in identifying four molecular subtypes of EC: POLE (ultramutated), microsatellite instability-high (MSI-H) (hypermutated), copy number low (CN-L), and copy number high (CN-H) [4]. The Proactive Molecular Risk Classifier for Endometrial Cancer (ProMisE) algorithm was developed to bridge the gap between research and clinical practice [5]. This approach uses mutational analysis of the POLE exonuclease domain, along with immunohistochemical (IHC) assessment of mismatch repair (MMR) proteins (MLH1, MSH2, MSH6, and PMS2) and p53 to categorize ECs [6]. The ProMisE system, now integrated into the National Comprehensive Cancer Network guidelines and the World Health Organization Classification of Female Genital Tumors, delineates four molecular EC subtypes [7, 8]. These subtypes are similar, although not identical, to those described in TCGA [9]. Notably, in TCGA’s original findings [4], 92% of ECs categorized as the CN-H molecular subtype were found to have TP53 mutations. This finding indicates that the ProMisE classification may not capture a small fraction of CN-H ECs that do not harbor TP53 mutations.

Here, we introduce a novel next-generation sequencing (NGS)-based molecular classification system to address these limitations. This approach streamlines the identification of the four TCGA subtypes, potentially enhancing the precision and efficiency of EC categorization.

The 2023 revision of the EC staging system by the International Federation of Gynecology and Obstetrics underscores this paradigm shift, emphasizing the incorporation of molecular characteristics for a nuanced understanding of the disease [5]. Our system offers a refined framework for risk assessment aligned with the 2023 FIGO revision, which underscores the molecular characteristics of EC staging.

Pioneering this advanced application, our institution employs NGS-based comprehensive genomic profiling for accurate early diagnosis and treatment planning. This study retrospectively assessed 200 patients with EC using an innovative NGS-based molecular classification system. This system is explicitly designed to align with TCGA molecular classification, focusing on genomic alterations such as mutations in POLE/TP53, tumor mutational burden (TMB), MSI status, and copy number alteration (CNA) status. Our primary aim was to validate the efficacy of this NGS-based approach for providing a comprehensive and nuanced molecular classification of EC. Additionally, we compared the NGS-based classification with the ProMisE to illustrate the potential of NGS-based molecular classification to revolutionize EC management strategies, ensuring a precision medicine approach firmly grounded in the foundational work established by TCGA.

Materials and methods

Study design

This study included 200 patients with EC who underwent surgery, including hysterectomy, as initial treatment at the Keio University Hospital. Written informed consent was obtained from all participants.

NGS-based cancer gene panel test

In January 2022, we implemented the Rapid-Neo, an in-house clinical sequencing system [10]. We analyzed tumor DNAs extracted from formalin-fixed paraffin-embedded tissues using a next-generation sequencer (NextSeq 550; Illumina, San Diego, CA, USA), targeting 145 cancer-related genes.

Defining molecular subtypes in EC using NGS

The molecular classification of EC was based on NGS data from high-quality samples, all of which had at least 20% tumor purity. The process for determining the molecular subtypes was hierarchically structured. Tumors were classified as a POLE molecular subtype if they exhibited a known mutation in the POLE exonuclease domain [11]. For cases where a POLE mutation was detected at an uncharacterized hotspot, such as V411M, the sample was categorized as POLE if TMB-H was present. The MSI-H molecular subtype was designated for tumors with an MSIsensor score ≥12. Moreover, tumors with a CNA count <35 were classified as CN-L, indicating a low degree of CNAs, whereas those with a CNA count ≥35 were classified as CN-H, indicating a high degree of CNAs. Importantly, if tumors had already been classified as either POLE or MSI-H subtypes, this classification was maintained regardless of their CNA count.

Detailed NGS protocol

Detailed procedures for DNA extraction, library preparation, and sequencing, including the amount of DNA required, DNA integrity number (DIN) scores, and the specifics of the cancer gene panel test, are provided [10]. Supplementary Table S1 lists all 145 genes analyzed.

Bioinformatic analysis

Bioinformatic analysis was conducted using the PleSSision pipeline (Mitsubishi Electric Software, Tokyo, Japan) for genome annotation and curation. As previously described, we identified cancer-specific gene alterations, including single nucleotide variants (SNVs), indels, and CNAs [10]. The pathogenicity of gene alterations in 145 cancer-related genes was evaluated using multiple databases: COSMIC (Catalogue of Somatic Mutations in Cancer, https://cancer.sanger.ac.uk/cosmic), ClinVar (https://www.ncbi.nlm.nih.gov/clinvar/), CIViC (Clinical Interpretation of Variants in Cancer, https://civicdb.org/home), CKB (Clinical Knowledgebase, https://ckb.jax.org/), and OncoKB (MSK’s Precision Oncology Knowledge Base, https://www.oncokb.org/).

TMB was defined as the number of nonsynonymous and synonymous mutations per megabase in the tumor genome, with TMB-high (TMB-H) defined as ≥10 mut/Mb. MSI status was determined by analyzing 95 intronic homopolymer repeat loci with an MSIsensor score obtained via principal component analysis. In addition to TMB and MSI evaluations, gene copy numbers were determined by comparing the median values of sequencing reads covering the target genes against control samples. CNAs were categorized as “gain” for more than four-fold increases and “loss” for less than two-fold decreases. We used the k-means clustering algorithm to categorize the CNAs objectively based on their counts across samples. This method facilitated the identification of a cutoff value of 35.3, distinguishing samples with significant alteration burden from those with minimal changes (Supplementary Fig. S1).

Confirmation of germline variants

For the MSI-H subtype of EC, further steps were undertaken to distinguish between somatic and germline mutations when pathogenic mutations in MMR genes (MLH1, MSH2, MSH6, and PMS2) were identified in tissue genetic profiles. Genomic DNA was extracted from patient-derived peripheral blood mononuclear cells and subjected to germline testing at Kazusa Research Institute. The patients were diagnosed with Lynch syndrome when these mutations were detected as germline mutations.

Immunohistochemistry

We evaluated the expression of the MMR proteins (MSH2, MLH1, MSH6, and PMS2) and p53 using immunohistochemistry (IHC). The tissues were formalin-fixed, paraffin-embedded tissue blocks. They were sliced from the block to a thickness of 4 μm. We used antibodies against the MutS protein homolog 2 (MSH2, 1:50; Dako, Glostrup, Denmark), MutL protein homolog 1 (MLH1, 1:50; Dako), MutS protein homolog 6 (MSH6, 1:50; Dako), PMS1 homolog 2 (PMS2, 1:40; Dako), and anti-human p53 protein mouse monoclonal antibody (DO-7, 1:200; Dako). The processed IHC slides were evaluated by the pathologists Y.M., M.K., and R.K. Cases with the loss of expression of at least one MLH1, MSH2, MSH6, or PMS2 in the tumor cells were defined as having deficient MMR. Proficient MMR was defined as positive nuclear staining for all MMR proteins.

MLH1 methylation analysis

The promoter regions of MLH1 were analyzed for methylation using MethyLight, a real-time polymerase chain reaction (PCR)-based method, as proposed by Takamatsu et al. [12]. DNA from the liquid-based cytology samples was subjected to bisulfite treatment using an EZ DNA Methylation-Gold kit (D5005, ZYMO RESEARCH). Methylation positivity was defined as a cutoff percentage of methylated reference volume of >25%.

MSI testing using PCR

We conducted a PCR assay to evaluate MSI by analyzing the nucleotide count in five mononucleotide microsatellite loci (NR-21, BAT-25, MONO-27, NR-24, and BAT-26) within the tumor DNA. Instability at a locus was determined based on predefined criteria (indicating a deviation in nucleotide count) and comparison with the established baseline values for these loci. Tumors demonstrating instability at ≥2 microsatellite loci were classified as MSI-H. Conversely, tumors with <2 unstable loci were categorized as microsatellite stable.

Statistical analyses

Statistical analyses were performed using JMP version 17.0 (SAS Institute, New York, USA). Categorical variables were compared using Fisher’s exact test. The Wilcoxon, Kruskal–Wallis, and Steel–Dwass tests were used for nonparametric comparisons of continuous variables when appropriate. The k-means clustering method was applied to categorize the samples based on CNA counts. Cumulative survival was estimated using the Kaplan–Meier method, and differences in OS or PFS among the four subtypes were analyzed using the log-rank test. The effects of variables on PFS were determined via multivariate analyses using the Cox proportional hazards model with the JMP software. Statistical significance was set at p < 0.05.

Results

Clinical and pathological characteristics

Table 1 summarizes the clinical characteristics and pathological data, including FIGO 2008 stage distribution, histological types, and other relevant factors, for all 200 patients with EC. Our cohort had a median age of 55 (range, 29–85) years and a median body mass index of 22.2 (range, 15.0–50.8) kg/m². All patients underwent surgery, including hysterectomy (extended total hysterectomy in most cases, or semi-radical or radical hysterectomy depending on the degree of cervical involvement). Bilateral salpingo-oophorectomy was performed in 97.0% (194/200) of cases, while ovary-sparing surgery was performed in 3.0% (6/200). Minimally invasive surgeries (conventional laparoscopic or robotic surgery) were performed for 51.5% (103/200) of the cases. Comprehensive pelvic or combined pelvic and para-aortic lymphadenectomies were performed for 127 patients. For 10 patients, only a lymph node biopsy was conducted, primarily because of advanced cancer or low patient tolerance. Lymph node dissection was omitted for 63 cases, typically in patients with endometrioid carcinoma G1/2 without myometrial invasion. LVSI was observed in 36.5% of patients, and histologically proven nodal metastases occurred in 18.5%. Postoperative adjuvant therapy was administered to 22 of the 37 patients with intermediate-risk profiles and 78 of the 79 patients with high-risk profiles. All postoperative adjuvant therapy consisted of chemotherapy except for one case of radiation therapy and one case of both radiation therapy and chemotherapy. The median follow-up duration was 1139.5 days, with a recurrence rate of 9.5%.

Table 1 Clinical and pathological characteristics of 200 patients with endometrial cancer.
Full size table

Characteristics of the patients by NGS-based molecular classification

In this study, 200 ECs were subjected to NGS for molecular classification. The specimens were primarily obtained by hysterectomy (94.0%, 188/200), followed by endometrial curettage (5.0%, 10/200) and hysteroscopic transcervical resection (1.0%, 2/200). Clinical sequencing of initial diagnostic specimens identified the CN-L subtype as the most prevalent (104/200, 52.0%), followed by MSI-H (38/200, 19.0%), POLE (33/200, 16.5%), and CN-H (25/200, 12.5%) subtypes (Fig. 1a). The clinical characteristics and pathological data obtained by molecular classification are summarized in Table 2. A significantly higher proportion of patients with the CN-H subtype had FIGO 2008 stages III–IV, compared with those with other molecular subtypes (p = 0.006). The distribution of histological types varied among the molecular subtypes, with high-grade histology being more prevalent for the CN-H subtype than for the other subtypes (p < 0.001). The proportion of LVSI positivity was significantly higher for the MSI-H and CN-H subtypes than for the other subtypes (p < 0.001) as was the rate of retroperitoneal lymph node metastasis (p = 0.009). The recurrence rate was significantly higher for the CN-H subtype than for the other subtypes (p < 0.001). The rate of progression was significantly lower for the POLE subtype and higher for the CN-H subtype than for the other subtypes (p < 0.001). No association was found between age and molecular subtype.

Fig. 1: Molecular classification of ECs using an NGS-based approach and distribution of TMB, MSI, and CNAs.
figure 1

a A hierarchical NGS-based approach to classify 200 EC cases into the POLE, MSI-H, CN-L, and CN-H subtypes. Criteria include POLE mutations (33 cases, including 6 with concurrent MSI-H), MSI-H (38 cases, classified by underlying cause), CN-L (104 cases, CNA count <35), and CN-H (25 cases, CNA count ≥35). b The flowchart highlights the molecular diversity in the EC cohort. the distribution of TMB, c MSIsensor score, and (d) CNA count across the four endometrial cancer subtypes: POLE, MSI-H, CN-L, and CN-H. The plots highlight the POLE subtype’s high TMB, the MSI-H subtype’s elevated MSI score, and the increased CNA count in the CN-H subtype. Significant differences (p < 0.001) in these genomic features among the subtypes are observed, revealing the distinct molecular characteristics of each subtype. e Venn diagram illustrating the overlap among cases with TMB-H (red circle, TMB ≥ 10), MSI-H (green circle, MSIsensor score ≥12), and CN-H (blue circle, CNA count ≥35). The overlapping regions indicate cases that meet both criteria. CNA copy number alteration, CN-H copy number high, CN-L copy number low, EC endometrial cancer, MSI microsatellite instability, TMB tumor mutational burden, MSI microsatellite instability, MSI-H microsatellite instability-high, NGS next-generation sequencing.

Full size image
Table 2 Clinical characteristics and pathological findings by molecular subtype in endometrial cancer.
Full size table

Molecular profiling and genomic variability in the EC subtypes

In our analysis of 200 EC cases, significant differences were observed in the TMB, MSIsensor scores, and CNA counts across the molecular subtypes (Fig. 1b–d). The POLE subtype had the highest mean TMB (154.1 mutations/Mb, range: 34.7–536.7), indicative of a hypermutated profile. Significant differences in TMB were observed on comparing the POLE subtype with the MSI-H (p < 0.001), CN-L (p < 0.001), and CN-H (p < 0.001) subtypes. In contrast, the MSI-H subtype, having a mean TMB of 29.6 mutations/Mb (range: 8.0–92.0), showed the highest mean MSIsensor score (47.69, range: 17.4–83.3), significantly different from those of both the CN-L (p < 0.001) and CN-H (p < 0.001) subtypes. The CN-L subtype had a mean TMB of 5.1 mutations/Mb (range: 0.7–25.0) and a relatively low MSIsensor score (mean: 1.9, range: 0.0–9.5). The CN-H subtype, associated with the high-grade histological type, had a mean TMB of 4.4 mutations/Mb (range: 1.5–8.5) and a low MSIsensor score (mean: 1.9, range: 0.0–11.1). It also showed the highest mean CNA count (58.4, range: 35–93), indicating high genomic instability and significantly different CNA counts, compared with the other subtypes (all p < 0.001). The accompanying violin plots visually represent these molecular distinctions, providing a comprehensive view of the genomic landscape across the different EC subtypes (Fig. 1b–d). Among 44 MSI-H cases, 2 cases (5%) showed CN-H, and 98% of MSI-H cases were TMB-High (Fig. 1e).

Figure 2a shows a heatmap based on NGS data capturing the somatic mutation spectrum across the 200 EC cases categorized into four molecular subtypes. The heatmap includes mutations in key genes, such as POLE, MMR genes, and TP53, as well as 20 additional representative genes. The clinical and pathological parameters included TMB, MSIsensor score, CNA count, and pathological tumor content rate. MMR and p53 IHC were integral to our analysis, with MMR IHC performed for all POLE/MSI-H subtypes and others with TMB ≥ 8 or available clinical results. MLH1 methylation data were included for all MSI-H cases and others based on the MMR IHC results. This provided a comprehensive view of the molecular diversity of EC.

Fig. 2: Comprehensive analysis of genomic variations in endometrial carcinoma subtypes.
figure 2

a Landscape of genomic variability in endometrial cancers by molecular subtype using NGS. Heatmap illustrates typical somatic mutations in cancer-related genes across endometrial carcinomas stratified by molecular subtype. Each column represents an individual tumor, and the rows highlight specific gene mutations. The upper section of the graph summarizes the histopathological and clinical data for each case. The right-hand bar graph shows each subtype’s mutation frequency and gene distribution. Mutation types and clinicopathological features are color-coded, as detailed in the legend. b Distribution of MSI status in POLE endometrial carcinoma. The figure shows the MSIsensor values, with segments indicating values < 12 (light blue) and ≥12 (vibrant orange). The middle ring depicts MSI-PCR results, delineating negative (mid-tone blue) outcomes from positive (clear red) ones. The outer ring details the MMR IHC status using color coding consistent with other figures to represent dMMR (distinct yellow), pMMR (stable blue), and partial loss (light red). c Causal factors behind MSI-H status in endometrial carcinoma subtypes. This figure illustrates the distribution of MSI-H causes, including MLH1 methylation (dark and muted purple), somatic MMR gene variants (dark blue), Lynch syndrome factors (muted green), and unknown etiologies (soft and distinct green). This chart visually correlates MSI-H causes and their respective molecular and genetic markers. dMMR deficient mismatch repair, IHC immunohistochemical, MMR mismatch repair, MSI microsatellite instability, MSI-H microsatellite instability-high, NGS next-generation sequencing, PCR polymerase chain reaction.

Full size image

The POLE subtype (16.5% of cases) showed the mutations P286R and V411L in 51.5% (17/33) and 27.3% (9/33) of cases, respectively. Detailed information on POLE mutations in all POLE subtypes is provided in Table 3. The mutation frequencies in the 20 representative genes, such as PTEN (97%), ARID1A (91%), and PIK3CA (91%), underlay the distinct genomic characteristics of the subtypes. Regarding the MSI-H subtype, MLH1 methylation was positive in 21 of the 38 cases (55%). The mutation frequencies of the representative genes in MSI-H cases were notable for PTEN (84%), ARID1A (95%), and PIK3CA (61%).

Table 3 POLE exonuclease domain mutations in the POLE subtype.
Full size table

In contrast, the CN-H subtype, characterized predominantly by TP53 mutations (56%), did not show a consistent pattern of mutation frequency across other genes (Fig. 2a). Aberrant p53 protein expressions were observed in 93% (13 of 14) of the cases harboring TP53 mutations. The CN-L subtype had a more heterogeneous mutational profile and a relatively lower frequency of mutations in the representative 20 genes than the other subtypes, indicating the involvement of diverse molecular pathways. A higher number of cases with the CN-L subtype showed mutations in PTEN and ARID1A, with PTEN mutations occurring in 72% and ARID1A mutations in 47% of the CN-L cases. TP53 mutations were identified in 10 of the 104 CN-L cases (10%). Aberrant p53 protein expressions were observed in 90% (9 of 10) of those cases with TP53 mutations.

Concurrent MSI-H in the POLE subtype

Six of the 33 (18.2%) POLE molecular subtype cases, representing 16.5% of the sequenced ECs, were MSI-H. Among these, five cases were confirmed as MSI-H through PCR testing. MMR IHC analysis revealed deficient MMR in 1 case and partial loss of MMR protein expression in 3 cases. Case 128, with the POLE subtype, was distinguished not only by the presence of a POLE p.A463V mutation but also the two PMS2 mutations p.R211* and p.P780Rfs*4. IHC findings further complicated the picture, revealing a loss of PMS2 and partial loss of MSH2 and MSH6 protein expression (Supplementary Fig. S2a), despite the absence of mutations in MSH2 and MSH6. MLH1 methylation was observed in two cases, indicating complexity and heterogeneity of the POLE subtype for MSI status. Detailed visualization of the MSI status of the POLE subtype is presented in Fig. 2b.

Elucidation of MSI-H etiology and Lynch syndrome diagnosis in MSI-H EC

In our cohort of EC cases identified as MSI-H, a comprehensive approach was employed to identify the causes of high microsatellite instability. We identified the factors underlying MSI-H in these tumors through an integrated analysis combining MLH1 methylation status, MMR gene mutation assessment, MMR IHC results, and genetic testing to distinguish somatic from germline origins.

Among the 38 cases studied, MLH1 methylation was prevalent in 21 (55%), demonstrating an epigenetic mechanism contributing to MMR dysfunction. Cases in which the MSI-H etiology was determined to be because of somatic MMR gene mutations included three with MLH1 mutation, two with MSH2 mutation, and one with MSH6 mutation. Evaluation of germline mutations led to the identification of Lynch syndrome in nine cases (24% of MSI-H), with genetic breakdown revealing three cases associated with MLH1, two with MSH2, one with MSH6, and three with PMS2. Moreover, two cases remained undetermined after a thorough analysis. Figure 2c shows the results of this extensive molecular profiling, illustrating the diverse etiologies of MSI-H.

Complexities in molecular diagnosis: discrepancies across IHC, genomic mutation, and MSI analysis

Within the 104 CN-L subtype cases, 3 demonstrated discrepancies between IHC findings and genomic mutation analyses, indicating potential diagnostic challenges. Specifically, cases 15 (serous carcinoma, stage IVB) and 53 (endometrioid carcinoma G3, stage IIIA) showed aberrant p53 IHC findings despite the absence of pathogenic mutations in TP53, indicating possible NGS false negatives. Additionally, case 146 (serous carcinoma, stage IVB), also of the CN-L subtype, displayed a p53 wild-type pattern in IHC but was found to have significant TP53 mutation through NGS. This scenario and that of case 189 from the CN-H group (small-cell neuroendocrine carcinoma, stage IVB) indicate potential false positives or challenges in IHC interpretation because of technical issues, such as tissue fixation. Specifically, case 146 had a pathogenic TP53 mutation (p.R306*, VAF 16%), and case 189 had a synonymous mutation (p.T125, VAF 98%), which is located near a splice site; this mutation could potentially affect mRNA splicing and protein configuration.

On evaluating the MSI status, discrepancies were observed between NGS and other testing modalities. Among the MSI-H subtypes (38 cases), case 149 had a partial loss of MLH1/PMS2 on IHC and an MLH1 mutation (p.G98S, VAF 65%) with high TMB, underscoring the value of NGS for heterogeneous cases (Supplementary Fig. S2b). Case 181 also had a high MSIsensor score (18.2) but was MSI-PCR-negative and displayed intact MMR protein expression and low TMB in IHC, indicating a possible NGS false positive.

Moreover, four cases of the CN-L subtype (cases 54, 93, 94, and 152) showed discrepancies between the NGS and MSI-PCR/MMR IHC results, indicating potential NGS false negatives. Cases 93 (TMB 8.5), 94 (TMB 9.9), and 152 (TMB 8.1) were positive for MLH1 methylation and showed MLH1/PMS2 loss on IHC, demonstrating that MSI-H was caused by MLH1 methylation. Case 54, although positive for MSI-PCR with MLH1/PMS2 loss on IHC, did not show MLH1 methylation or gene abnormalities, indicating a possible NGS false-negative result due to factors such as undetected large deletions or epigenetic changes that are difficult to capture with NGS.

Comparative survival outcomes across the EC molecular subtypes according to NGS-based and ProMisE approaches

The Kaplan–Meier curves for progression-free survival (PFS) and overall survival (OS), categorized by molecular classification, are shown in Fig. 3a, b. The POLE cases had the most favorable PFS rates (3-year PFS: 100%), followed by the MSI-H (3-year PFS: 96.8%), CN-L (3-year PFS: 93.8%), and CN-H (3-year PFS: 47.4%) subtypes, showing significant differences in prognosis between the four groups (p < 0.001). There were also significant differences in OS among the four groups (p = 0.006), with the CN-H group having the worst OS rate (3-year OS: 90.5%). Stratified analysis of FIGO 2008 stages I–II showed significant differences among the four groups only for PFS (p < 0.001). However, a stratified analysis of FIGO 2008 stages III–IV showed significant differences in PFS and OS, with the CN-H group having a noticeably worse prognosis than the other groups did (3-year PFS: 14.7%, 3-year OS: 78.6%) (Supplementary Fig. 3a–d).

Fig. 3: Kaplan–Meier curves for PFS and OS by NGS-based and ProMisE-based molecular subtypes and clinicopathological features and prognostic analysis in the CN-H subtype.
figure 3

a The Kaplan–Meier curves for PFS and (b) OS for all four NGS-based endometrial cancer subtypes: POLE, MSI-H, CN-L, and CN-H. c The Kaplan–Meier curves for PFS and (d) OS for all four ProMisE-based endometrial cancer subtypes: POLE, dMMR, NSMP, and p53mut. The Kaplan–Meier curves for PFS and OS are displayed as stratified prognostic analyses. e This Venn diagram depicts the complex relationship and overlap among 75 cases classified according to high-grade histological type and p53 IHC status in the CN-H molecular subtype. Despite being classified as CN-H, 10 and 12 cases exhibited characteristics associated with low-grade histological type and wild-type p53 status, respectively. f The Kaplan–Meier curve for PFS and (g) OS for the CN-H cases showed that patients with p53 mutant and wild-type status had similar prognoses. CN-H copy number high, CN-L copy number low, MSI-H microsatellite instability-high, OS overall survival, PFS progression-free survival, dMMR deficient mismatch repair, NSMP non-specific molecular profile, OS overall survival, IHC immunohistochemistry.

Full size image

The survival outcomes by ProMisE are provided in Fig. 3c (PFS) and 3d (OS). In terms of PFS, significant differences were observed among the molecular subtypes (p < 0.001), with the POLE subtype showing the most favorable outcomes. However, unlike the NGS-based approach, the ProMisE did not reveal significant differences in OS across the subtypes (p = 0.117).

Clinicopathological features and prognostic analysis of the CN-H subtype

Finally, we investigated the distribution of high-grade histological types, p53 IHC status, and CN-H molecular subtype classifications (Fig. 3e). We identified 55 cases with a high-grade histological type, 36 cases displaying mutant p53 status that was indicative of overexpression or null patterns, and 25 cases with CN-H. Within the CN-H molecular subtype, 13 cases (52%) had the mutant p53 status, and 12 (48%) had the wild-type p53 status on IHC, indicating discrepancies between the NGS-based and ProMisE. There was no CN-H case with p53abn in IHC but no TP53 mutation in NGS. Kaplan–Meier curves for PFS and OS (Fig. 3f, g) indicated no significant difference in prognosis according to the p53 status of the CN-H subtype.

Furthermore, nine cases (36%) had the low-grade histological type and wild-type p53 IHC status despite being classified as CN-H (five and four cases with Stage I and Stages III–IV, respectively). Three of the cases of stage IA without LVSI, which could be underestimated as “low risk of recurrence” by conventional morphological and IHC assessments alone. Of these nine cases, three developed recurrence, including one case of stage IB with LVSI and two cases of stage IVB. Two patients with stage IVB were once assessed as being equivalent to stage IA and underwent fertility preservation therapy (medroxyprogesterone acetate). However, these patients developed distant metastases during fertility preservation treatment. Overall, this subset challenges traditional classification paradigms and indicates the presence of a distinct group within the CN-H category.

Discussion

The landscape of EC management is rapidly evolving with the integration of molecular classifications, as demonstrated in this study. Our analysis of 200 EC cases using NGS offers new insights into the clinical utility of molecular subtyping based on TCGA classification. This study underscores the significance of molecular classification in enhancing prognostic accuracy and guiding therapeutic strategies for EC.

Identifying the pathogenic variants of POLE gene is essential for accurately determining the POLE subtype in EC. Limited access to advanced DNA sequencing methods has historically been a barrier, potentially leading to overtreatment and consequent morbidities [13]. The QPOLE test, developed by Van den Heerik et al., offers an accessible solution by providing a rapid and cost-effective quantitative PCR assay to detect POLE hotspots [14]. However, its limitation of only hotspot mutations underscores the need for broad sequencing methods to ensure a comprehensive molecular classification. In our cohort, identifying non-hotspot mutations such as T357_I358delinsIT and A463V highlights the importance of extensive DNA sequencing. This approach goes beyond conventional hotspots and covers minor POLE exonuclease domain mutations that are crucial for determining a favorable prognosis and mitigating the risk of EC overtreatment.

Our significant finding is the identification of concurrent MSI-H within the POLE subtype, which is a novel insight, given the scarcity of such reports. These findings emphasize the potential of NGS to reveal complex molecular interactions within tumors, which might influence responsiveness to immunotherapies. The presence of concurrent MSI-H in POLE mutation cases, as reported by Devereaux et al., suggests secondary genetic alterations that could affect the efficacy of immune checkpoint inhibitors [15].

Our findings reveal that the molecular subtypes of EC have distinct clinical characteristics and outcomes. The superior PFS and OS of the patients with the POLE subtype over those of the patients with other subtypes aligns with previous reports of this subtype as a potential marker for a favorable prognosis [4, 5]. Conversely, the CN-H subtype was consistently associated with poor outcomes, indicating the need for aggressive treatment approaches for these patients. A critical aspect of our analysis of the CN-H subgroup was the presence of cases that, despite being classified as CN-H, had characteristics typically associated with a low-grade histological type and wild-type p53 IHC status. This challenges traditional diagnostic paradigms and underscores the complexity of cancer classification, highlighting the potential limitations of relying solely on morphological and IHC evaluations. As observed in the cases in this study, there is the risk of undertreatment (postoperative treatment being omitted or fertility-preserving treatment being opted for) for stage IA cases with low-grade histological type, although they should have been classified as CN-H by NGS-based evaluation. In the present study, the NGS-based approach reflected prognosis more sensitively than the ProMisE, further supporting its clinical utility. This discrepancy underscores the enhanced prognostic precision of the NGS-based approach, which can more accurately stratify patients with EC by both PFS and OS and facilitate better-informed clinical decisions. The risk of undertreatment indicates the existence of a clinically significant subgroup within the CN-H classification that may be overlooked without comprehensive molecular profiling.

In our study, NGS not only facilitated the identification of these subtypes but also provided a more nuanced understanding of their molecular underpinnings. For instance, assessing MMR gene variants through germline testing provides crucial information for diagnosing Lynch syndrome, which is a key aspect in managing EC. Furthermore, the ability of NGS to delineate specific mutation profiles within the EC subtypes provides an opportunity to tailor treatment strategies effectively.

Integrating NGS results with IHC findings is critical to understanding EC comprehensively. The discrepancies observed between NGS-derived molecular subtypes and IHC findings highlight the need to carefully interpret molecular data, particularly when IHC and NGS findings do not align.

Another significant aspect of our study was the ability to categorize EC cases into CN-L and CN-H subtypes based on a cutoff value for CNA counts, despite analyzing a limited panel of 145 genes. This stratification allowed for a sensitive distinction in prognosis, demonstrating the utility of targeted gene panels for revealing clinically relevant molecular insights. The ProMisE system relies on the TP53 mutation status to differentiate between p53 abnormal (p53abn) and non-specific molecular profile (NSMP) groups. However, both TCGA reports and our findings indicate many wild-type p53 cases within the CN-H subgroup. The ProMisE system classified these cases as NSMP, potentially assigning them to an intermediate-risk group and misinforming prognostic expectations. In contrast, our CN-H subgroup analysis revealed no significant differences in PFS and OS between patients with and without p53 mutations, confirming the clinical relevance of our CN-H definition. Our NGS-based approach to CN classification, which does not require specialized analysis such as copy number array, offers significant value by enabling a comprehensive evaluation aligned with TCGA classification.

However, our study has limitations. The retrospective nature and small sample size may limit the generalizability of our findings. Furthermore, one limitation of the NGS-based approach is its potential for false negatives, which could lead to misclassification and affect treatment decisions. To mitigate the risk of false negatives, considering factors such as tumor purity and the integration of NGS results with IHC findings is important. In addition, our study is monocentric and would benefit from external validation. Further research is needed to fully understand the implications of these subtypes in clinical practice, particularly in diverse populations.

In conclusion, our study reaffirms the critical role of molecular classification in EC, highlighting the superiority of NGS over ProMisE. Our findings demonstrate that the NGS-based approach offers more precise stratification and prognostication than by ProMisE, providing a detailed and nuanced method for diagnosis, prognosis, and treatment planning. This enhanced prognostic precision underscores the potential of NGS to guide personalized medicine in EC, for which treatment decisions are increasingly based on molecular characteristics rather than traditional histopathological criteria alone. Future research should focus on expanding the utility of NGS for the molecular classification of EC, exploring new therapeutic targets, and validating these findings in large prospective studies.