Systematic Multiplatform Discovery of Methylation Markers Enables Non‐Invasive Early Detection of Endometrial Cancer

ABSTRACT

Endometrial cancer, the most common gynecological malignancy with an annual increase of 1%–3%, lacks suitable noninvasive diagnostic tools, as current methods like hysteroscopy and biopsy are invasive and impractical for routine screening. We conducted a comprehensive, multilayered marker discovery workflow integrating whole‐genome bisulfite sequencing and targeted methylation panels in tumor and control tissues, then prioritized and optimized candidates for detection in cervical exfoliated cells. Using a two‐stage design, we built and tested a quantitative methylation–specific PCR (qMSP) model in ThinPrep Cytology Test (TCT) samples, with 148 samples for discovery/model construction and an independent cohort of 80 TCT samples for validation. We identified a three‐gene methylation panel—ZNF626, GRIA4, and SPDYA—that demonstrated high accuracy for early endometrial cancer detection from cervical cytology. In the validation cohort, the model showed strong performance across menopausal subgroups: in premenopausal women, sensitivity was 90.91% and specificity for benign endometrial disease was 92.59%; in postmenopausal women, sensitivity and specificity were 96.55% and 84.62%, respectively. Notably, the approach achieved a 92.86% detection rate for stage I endometrial cancer. These results support a robust, noninvasive diagnostic strategy that leverages simple cervical cytology sampling to enable early detection, facilitate clinical decision‐making, and potentially improve outcomes for patients at risk of endometrial cancer.

We identified key endometrial cancer DNA methylation biomarkers (ZNF626, GRIA4, SPDYA) from cervical cells. The model achieved high endometrial cancer sensitivity and detected 92.86% of Stage I endometrial cancers using TCT samples.

Abbreviations

BE

benign endometrium

CC

cervical cancer

DMR

Differential methylation region

EC

Uterine Corpus Endometrial Cancer

NE

normal endometrium

OC

ovarian cancer

PBL

peripheral blood leukocytes

qMSP

quantitative methylation‐specific polymerase chain reaction

TCT

Thinprep Cytologic Test

1. Introduction

Endometrial cancer (EC) is a prevalent malignancy of the female reproductive system, predominantly affecting peri‐ and postmenopausal women. Over the past two decades, its incidence has steadily increased, with a notable shift toward younger onset, likely reflecting rising life expectancy and lifestyle factors. In 2022, 420,368 new cases were reported worldwide [1, 2]. In many European and North American countries, EC is now the most frequently diagnosed gynecologic malignancy [1]. According to the National Cancer Center of China, there were approximately 77,700 new EC cases and 13,500 deaths in China in 2022, corresponding to an incidence of 11.25 per 100,000 and ranking second among gynecologic cancers after cervical cancer [3]. For patients with stage IA disease, surgery alone is often curative [4]. In contrast, advanced disease typically necessitates multimodal therapy, including surgery, radiotherapy, and chemotherapy, yet outcomes remain suboptimal [5]. It underscore a central imperative: earlier detection and timely intervention are critical to improving survival and quality of life [6].

Current approaches to preliminary detection of endometrial cancer (EC) typically draw on clinical symptoms (e.g., abnormal uterine bleeding and lower abdominal pain), transvaginal ultrasonography, and, in some cases, serum tumor markers such as CA125, CA19‐9, CA15‐3, and HE4. Definitive diagnosis, however, still requires histopathological examination of tissue obtained via diagnostic curettage or hysteroscopy. These approaches have notable limitations: symptom‐based triage lacks specificity; there is no standardized threshold for endometrial thickness in ultrasounds; and high‐level guidelines do not endorse any tumor markers for early detection. Moreover, both diagnostic curettage and hysteroscopy are invasive. Consequently, there remains a clear unmet clinical need for a convenient, accurate method to support earlier detection prior to undertaking invasive diagnostic procedures.

Recent advancements in cancer molecular biomarkers are promising, as researchers have identified many specific changes in EC, including gene mutations, microsatellite instability (MSI), copy number changes, mRNA and miRNA expression patterns, and DNA methylation [7, 8]. Huang et al. identified three gene methylation markers, BHLHE22, CDO1, and CELF4, using TCGA and a custom methylation database for early detection of EC in cervical smears [9]. Kong et al. then used qPCR methods to detect the methylation levels of CDO1 and CELF4 genes in cervical exfoliated cell specimens from postmenopausal women in the Chinese population, and the results showed that these two biomarkers used in combination for early detection of EC had a sensitivity and specificity of 87.5% and 90.8%, respectively (AUC 0.89) [10]. For a biomarker or diagnostic model, it must demonstrate a sufficiently high specificity, thereby minimizing the number of women who test positive but ultimately undergo unnecessary hysteroscopic biopsy or diagnostic curettage. Additionally, factors such as the cost of the test, turnaround time, and the accessibility of the testing platform are also critical determinants of whether a method can be feasibly implemented in real‐world clinical practice.

This study seeks to mine EC tissue methylation sequencing data to identify specific DNA methylation markers, which will then be validated using cervical exfoliated cell specimens obtained through the ThinPrep Cytologic Test (TCT) from EC patients and other gynecological diseases in the Chinese population. The goal is to develop a high‐performance and cost‐effective diagnostic assay for the early detection of endometrial cancer.

2. Material and Methods

2.1. Clinical Participants

Clinical participants were from the Gynecological Oncology Department of Peking University Cancer Hospital & Institute. This study was approved by The Ethics Committee of Peking University Cancer Hospital. The ethical approval numbers are as follows:

The Ethics Committee of Peking University Cancer Hospital 2024YJZ37.

All procedures were in accordance with the ethical standards of the responsible ethics committee and with the Helsinki Declaration of 1964 and later versions. Informed consent was obtained from all patients before sample collection.

Inclusion criteria:

1. Suspected endometrial disease as recommended by the “Guidelines for Endometrial Cancer Screening” or clinical suggestion for hysteroscopy and biopsy.

2. Patients had not received prior treatment for endometrial disease.

3. Patients were willing to undergo testing with signed informed consent

4. All samples included in this study were collected prior to any intrauterine manipulation or intervention, such as endoscopic endometrial biopsy, pipelle biopsy, or diagnostic curettage (D&C) in order to avoid any procedure‐related bias.

5. Patient enrollment was determined through comprehensive evaluation by experienced medical professionals, considering both clinical presentations and diagnostic findings (abnormal bleeding, medical history, and imaging observations of Endometrial thickness, biomarker CA125). Experienced gynecologists assessed each patient comprehensively to ensure unbiased inclusion.

In addition, participants with potential cervical pathology (including those presenting with symptoms common to cervical cancer) were excluded if they tested positive for high‐risk HPV.

After inclusion of the participants and TCT samples collection, the reference detection (gold standard detection methods) would be performed. A qualified clinician performed hysteroscopy and conducted biopsy based on patient conditions. Surgically removed endometrial tissues were used for histopathology evaluation. Each specimen was independently assessed by at least two pathologists. Staging was performed according to the NCCN guidelines for uterine neoplasms, and surgical pathological staging was performed according to the International Federation of Gynecology and Obstetrics (FIGO) 2009 classification. Patients with inconsistent pathological results between the endometrial biopsy and hysterectomy specimens were excluded from the study.

3. Statistical Analysis

ROC analysis was conducted using R package ggplot2. The effectiveness of the diagnostic method was evaluated by conducting a diagnostic test assessment, which involved comparing it to the gold standard and calculating 95% confidence intervals. Sensitivity was defined as the ratio of accurately identified positive cases of endometrial cancer to the total number of endometrial cancer cases identified by the gold standard, whereas specificity was defined as the ratio of accurately identified negative cases among all control cases identified by the gold standard. The sensitivity and specificity were calculated by clinical calculator1 (http://vassarstats.net/clin1.html). The calculation formulas are as follows:

$$\text{sensitivity}=\frac{\mathrm{TP}}{\mathrm{TP}+\mathrm{FN}}\times 100\%$$

$$\text{specificity}=\frac{\mathrm{TN}}{\mathrm{FP}+\mathrm{TN}}\times 100\%$$

TP, true positive, positive in both methylation test and reference detection. FP, false positive, positive in methylation test, negative in reference detection. FN, false negative, negative in methylation test, positive in reference detection. TN, true negative, negative in both methylation test and reference detection. The details of the experimental and bioinformatic methods are presented in the Doc S1 [11, 12, 13].

4. Results

4.1. Abundant Genomic Loci Are Hypermethylated in EC Tissues

As outlined in Figure 1, comprehensive interrogation of multiple datasets revealed numerous DMRs, subsequently distilled to a curated set of 156 (Table S1). Notably, this DMR set includes genes previously implicated in endometrial cancer (EC) or cancer epigenetics, such as CDO1, CELF4, BHLHE22, ADCYAP1, HAND2, POU4F3, HOXA9, HS3ST2, AJAP1, GALR1, and SOX11, et al. Collectively, the findings delineate a widespread hypermethylation landscape in EC that converges on biologically plausible targets.

FIGURE 1.

The workflow of this study. WGBS, Whole Genome Bisulfite Sequencing. TCGA, The Cancer Genome Atlas. DMR, Differential Methylation Region; EC, Endometrial Cancer; BE, Benign Endometrium; NE, Normal Endometrium; NAT, Normal Tissue Adjacent to the Tumor; WBC, White Blood Cells; qMSP, quantitative Methylation‐Specific Polymerase chain reaction; TCT, Thinprep Cytologic Test.

4.2. Candidate Methylation Markers Identification From Tissue and TCT Samples

After 156 DMRs were selected, we designed the primers and probes to perform the qMSP templated by the DMRs, positive methylation control, negative methylation control, and no template control (NTC). Since there are 49 markers that exhibited poor amplification of positive control or unexpected significant amplification of negative control, we excluded them (16 markers were abandoned for poor PC amplification, 33 markers were abandoned for significant NC amplification) and selected the remaining 107 markers for the subsequent qMSP test conducted on tissue samples. The Ct values of quality control for 156 markers were presented in Table S1.

The qMSP test for 107 markers was conducted on 8 EC tissues and 8 BE tissues, with the detail of Ct values presented in Table S2. The top 29 markers with the largest methylation level differences between EC tissues and BE tissues were selected as candidates. We presented the methylation levels of these 29 candidate markers across three high‐throughput datasets: in‐house tissue samples tested by WGBS or panel‐targeted methylation sequencing (Figure 2A) and the TCGA UCEC dataset (Figure 2B). These findings suggest consistent hypermethylation in these genomic regions across various platforms and cohorts. Their sensitivity in 8 EC tissues and specificity in 8 BE tissues by qMSP were presented in Figure 2C.

FIGURE 2.

The identification of 29 candidate methylation markers. (A) The methylation levels of 29 markers tested by WGBS (left panel) or panel‐targeted sequencing (right panel) derived from in‐house EC tissues and control samples. (B) The methylation level of 29 markers in EC and control tissue samples derived from TCGA database. (C) The sensitivity and specificity of 29 markers in 8 EC and 8 BE tissues based on qMSP. (D) The sensitivity and specificity of 29 markers in 8 EC and 8 normal TCT samples based on qMSP.

Following this, the qMSP for the 29 candidate methylation markers was performed on 16 TCT samples (8 samples from EC patients and 8 samples from healthy individuals). The ΔCt values of the qMSP assay were presented in Table S3. Their sensitivities in 8 EC patients' TCT samples and specificities in 8 healthy individuals' TCT samples were presented in Figure 2D. We found that compared to tissue samples, the sensitivity of some markers in TCT samples was significantly reduced; this may be due to the small proportion of endometrial exfoliated cells in TCT samples and the complex signaling background of exfoliated cells from the end of the cervix. We further identified the top 12 candidate markers with the largest methylation level differences between the TCT samples from EC and BE.

4.3. Construction of Endometrial Cancer Diagnostic Model

Next, further verification for the 12 candidate markers was conducted using qMSP with 148 TCT samples, including 58 cases of endometrial cancer, 66 cases of benign uterine diseases, 10 cases of cervical cancer, and 14 cases of ovarian cancer. Besides, there were three additional atypical hyperplasia (AH) patients representing precancerous lesions. The clinical characteristics of the participants are presented in Table 1.

TABLE 1.

Clinical characteristics of participants involved in model construction and independent validation.

Features	Model construction				Independent validation
	EC	AH	BE	Other cancers	EC	BE
Number	58	3	66	24	40	40
Age, mean (SD)	54.26 (9.09)	57.33 (9.1)	50.52 (12.81)	53.96 (8.47)	55.15 (11.1)	46.13 (12.78)
FIGO Stage, n (%)
I	41 (70.69%)				28 (70%)
II	5 (8.62%)				5 (12.5%)
III	7 (12.07%)				5 (12.5%)
IV	1 (1.72%)				1 (2.5%)
Undetermined	4 (6.9%)				1 (2.5%)
Histologic subtypes, n (%)
Endometroid carcinoma	51 (87.93%)				34 (85%)
FIGO Grade, n (%)
Type I‐G1	18 (31.03%)				12 (30%)
Type I‐G2	21 (36.21%)				17 (42.5%)
Type I‐G3	12 (20.69%)				5 (12.5%)
Serous carcinoma					2 (5%)
Clear cell carcinoma	3 (5.17%)				1 (2.5%)
Mixed type	1 (1.72%)
Uterine carcinosarcomas	3 (5.17%)				2 (5%)
Undetermined					1 (2.5%)

Abbreviation: AH, atypical hyperplasia.

Figure 3A illustrates the ΔCt values for 12 candidate markers across 148 TCT samples. The original data of these qMSP assays were displayed in Tables S4 and S5. The box‐whisker plots indicate that most candidate markers showed low methylation levels in ovarian cancer, distinguishing EC from ovarian cancer, possibly because ovarian cancer cells shed into the cervical area at a lesser rate. For cervical cancer, there were only three markers (GRM8, SPDYA_2, NR1H2) exhibiting significantly lower methylation levels relative to EC, aiding the distinction between EC and cervical cancer, with SPDYA_2 demonstrating a most excellent specificity (Figure 3B). Thus, we intended to select SPDYA_2 as one of the EC‐specific methylation markers.

FIGURE 3.

Best combination analysis from 12 markers in 148 TCT samples. (A) The methylation level (Δ Ct value) of 12 markers in 58 endometrial cancer, 66 uterine benign diseases, 10 cervical cancer, and 14 ovarian cancer TCT samples based on qMSP. (B) The best 3 markers to distinguish endometrial cancer from cervical cancer. (C) The area under the curve (AUC) values, sensitivities, and specificities of different numbers of markers used in combination. (D) The top 5 three‐marker combinations according to the number of correct detection cases based on the interpretation logic of “single positive indicating positive”. (E) Complementary analysis for EC detection with the selected three markers (SPDYA_2, ZNF626, GRIA4).

Considering that combining multiple markers could improve EC detection rates, we analyzed the performance of different numbers of marker combinations. The results indicated that when three markers were combined, the comprehensive performance reached an optimal level, with the highest AUC value and proper sensitivity and specificity (Figure 3C). Therefore, we plan to build a model with a combination of three markers.

To ensure that the EC‐specific methylation marker, SPDYA_2, was included in the diagnostic model, we exhaustively generated all possible combinations of three markers containing SPDYA_2. Each combination was assessed using a “single positive indicating positive” criterion to diagnose. We calculated the number of correct detection cases for each combination, selecting the top five combinations sorted by the numbers in descending order (Figure 3D). The combination 1 (SPDYA_2, ZNF626, GRIA4) and combination 2 (SPDYA_2, ZNF626, CCDC181) yielded the highest case number of 110 (Figure 3D). Concerning GRIA4's better differentiation between endometrial cancer and benign disease or ovarian cancer than CCDC181, we selected combination 1 for further exploration. Additionally, the complementary analysis for EC detection demonstrated that ZNF626, GRIA4, and SPDYA_2 provide mutual enhancement in the detection of endometrial cancer samples. Any combination of two markers can detect the majority of cancer cases, while the use of all three markers significantly improves overall sensitivity (Figure 3E).

To better determine the weight of each marker in EC diagnosis, we trained a logistic regression model using the three methylation markers, basing on qMSP data from 124 TCT samples (58 cases of endometrial cancer and 66 cases of uterine benign diseases) out of the previously mentioned 148 TCT samples. A total of 124 TCT samples were randomly divided into two sets, with 83 samples assigned to the training set and 41 samples to the test set. We constructed an EC diagnosis model based on the training set, then calculated a MATS (Methylation Assessment based on TCT Sample) value with the ΔCt value of ZNF626, GRIA4, and SPDYA and their respective weights to assess the EC risk. The formula used was as follows:

$$\text{MATS value}=\exp \left(10.52-0.{21}^{*}\text{SPDYA}-0.{58}^{*}\mathrm{ZNF}626-0.{17}^{*}\text{GRIA4}\right)/\left(1+\exp \left(10.52-0.{21}^{*}\text{SPDYA}-0.{58}^{*}\mathrm{ZNF}626-0.{17}^{*}\text{GRIA4}\right)\right)$$

A threshold of 0.491 for MATS value was applied as the cut off for this model. The performance metrics of the model in the training and test sets are displayed in Table 2, respectively, achieving a sensitivity of 87.18% and specificity of 90.91% in the training set, and a sensitivity of 84.21% and specificity of 95.45% in the test set. The ROC curves of the MATS model in the training set and test set were displayed in Figure S1. Additionally, the model detected 2 out of 3 precancerous lesion cases (atypical hyperplasia, AH) mentioned in Table 1. It suggests the model has a potential ability to detect precancerous conditions at an earlier stage.

TABLE 2.

The performance of the model in training set (83 TCT samples derived from 39 EC cases and 44 BE cases) and test set (41 TCT samples derived from 19 EC cases and 22 BE cases).

	Training set			Test set
		Reference approach detection			Reference approach detection
		EC	BE		EC	BE
Methylation marker prediction	EC	34	4	EC	16	1
	BE	5	40	BE	3	21
		Sensitivity (87.18%)	Specificity (90.91%)		Sensitivity (84.21%)	Specificity (95.45%)

4.4. Independent Validation for the Endometrial Cancer Diagnostic Model

Finally, the model was employed to detect EC in 80 TCT samples which derived from 40 cases of endometrial cancer and 40 cases of benign uterine diseases. The ΔCt values of the qMSP assays performed on 80 TCT samples were presented in Table S6. The ROC curve demonstrated excellent model performance, with a sensitivity of 95% for endometrial cancer, a specificity of 90% for benign diseases, and an AUC value of 0.9525 (Figure S2A). Since the majority of endometrial cancer patients are postmenopausal, we stratified the statistics by menopausal status and calculated sensitivity and specificity separately. The sensitivity of endometrial cancer in postmenopausal women (n = 42) was 96.55% (28/29), while the specificity for benign endometrial disease was 84.62% (11/13). In premenopausal women (n = 38), the sensitivity of endometrial cancer was 90.91% (10/11), and the specificity for benign endometrial disease was 92.59% (25/27). A Sankey diagram depicted the clinical staging of EC samples within the true positive results when detecting independently with the three selected methylation markers and using the model. Each marker contributed to the detection of stage‐I and II endometrial cancer, and utilizing the model significantly improved the detection rate for each stage (Figure S2B). The sensitivities were calculated according to different stages, 92.86% for Stage I (n = 28), 100% for Stage II (n = 5), 100% for Stage III (n = 5), 100% for Stage IV (n = 1), and 100% for unclassified stages (n = 1) (Table 3).

TABLE 3.

Stratified statistics for the model sensitivity based on pre‐ and postmenopausal status in the independent validation set.

Clinical stage	Total			Premenopausal			Postmenopausal
	Total N	Test positive	Sensitivity % (95% CI)	Total N	Test positive	Sensitivity % (95% CI)	Total N	Test positive	Sensitivity % (95% CI)
All	40	38	95% (83.5%–98.62%)	11	10	90.91% (62.27%–98.38%)	29	28	96.55% (82.82%–99.39%)

I	28	26	92.86% (77.35%–98.02%)	8	7	87.5% (52.91%–97.76%)	20	19	95% (76.39%–99.11%)
II	5	5	100% (56.55%–100%)	2	2	100% (34.24%–100%)	3	3	100% (43.85%–100%)
III	5	5	100% (56.55%–100%)	1	1	100% (20.65%–100%)	4	4	100% (51.01%–100%)
IV	1	1	100% (20.65%–100%)	/	/	/	1	1	100% (20.65%–100%)
Unknown	1	1	100% (20.65%–100%)	/	/	/	1	1	100% (20.65%–100%)

Moreover, there were multiple kinds of histologic subtypes being listed in Table 1. There sensitivities of employing the diagnostic model were also summarized, such as 94.12% for endometrioid carcinoma (n = 34), 100% for serous carcinoma (n = 2), 100% for clear cell carcinoma (n = 1), 100% for uterine carcinosarcoma (n = 2), and 100% for unclassified types (n = 1).

5. Discussion

As a prevalent malignancy of the female reproductive system, EC survival is highly correlated with stage, making early detection and treatment essential [14]. Patient and physician groups have begun advocating for methods that can expedite appropriate diagnosis and referral with less invasive approaches. To meet clinical needs, recent studies have explored DNA methylation assays in cervical samples for early EC detection, often combining multiple biomarkers. In a two‐stage, multicenter study, the MPap model (age, BMI, and BHLHE22/CDO1 methylation) detected stage I EC with > 92% sensitivity, outperforming TVS [15]. In postmenopausal women with suspected lesions, dual‐gene methylation (CDO1/CELF4) achieved 87.5% sensitivity and 90.8% specificity; adding TVS raised sensitivity to 100% without improving specificity [10]. A panel of CDH13, CDO1, and ZIC1 yielded an AUC of 0.97% and 93% sensitivity [16]. RASSF1A and HIST1H4F methylation in cervical brushes showed AUCs of 0.938 and 0.951 [17]. The WID‐qEC test (ZSCAN12/GYPC) reported 97.2% sensitivity and 75.8% specificity in a prospective cohort and surpassed TVS on specificity with comparable sensitivity/AUC; versus mutation testing, it achieved similar specificity but higher AUC and sensitivity [18]. The newer WID‐EC test profiles 500 CpGs, enabling multi‐cancer screening (endometrial, cervical, ovarian, breast) from a single liquid‐based cervical cytology sample [19].

In this study, we profiled DNA hypermethylation markers for endometrial cancer (EC) using both tissue and TCT samples. Integrating WGBS, panel‐targeted sequencing data, and TCGA, we identified 156 DMRs, then narrowed these to 29 DMRs which could be tested using qMSP. From these, 12 markers showed consistent performance across tissue and TCT and were taken forward. In 148 TCT samples, all 12 markers differentiated EC from benign disease, but high accuracy required combining at least three markers. Notably, SPDYA displayed high specificity in cervical cancer, highlighting the challenge of distinguishing EC from cervical cancer in TCT due to limited endometrial cell shedding and the predominance of cervical cells. We exhaustively assessed three‐marker panels and identified ZNF626, GRIA4, and SPDYA as the optimal trio. A logistic regression model built on these markers produced a MATS score that, in an independent TCT validation set of 80 samples, achieved 95% sensitivity and 90% specificity. These results establish a strong foundation for large‐scale clinical validation.

In these three markers, ZNF626 remains sparsely studied, especially regarding methylation. In addition, while predicted to regulate DNA‐templated transcription, its mechanism is unclear. Here, we first report ZNF626 hypermethylation in TCT‐derived DNA associated with EC, warranting further investigation. GRIA4, encoding an AMPA‐type glutamate receptor subunit, shows promoter hypermethylation inversely correlated with transcription and linked to improved survival in oropharyngeal squamous cell carcinoma [20]. It is hypermethylated in plasma from metastatic colorectal cancer (CRC) and proposed as a liquid‐biopsy response biomarker [21], and has been repeatedly flagged as a CRC detection marker [22, 23, 24]. We first propose GRIA4 hypermethylation as a non‐invasive EC biomarker detectable in TCT samples, with its mechanistic role in EC carcinogenesis remaining to be clarified.

Notably, SPDYA emerges as a cancer‐type–specific marker in our data: its methylation distinguishes EC from cervical and ovarian cancers, addressing a gap in prior cervical exfoliated‐cell studies that did not establish such discrimination [18, 25, 26, 27] or reported concurrent detection of cervical cancer [19, 28, 29, 30]. Biologically, SPDYA encodes a Speedy/RINGO cell cycle regulator essential for meiosis. Loss of SPDYA causes meiotic defects [31, 32]. Its promoter is hypermethylated in hepatocellular carcinoma [33], and high expression has been observed in colorectal cancer CRC [34]. Consistent with promoter methylation‐mediated repression, we infer lower SPDYA methylation in CRC (and possibly cervical cancer), whereas our findings show SPDYA hypermethylation in EC tissues and TCT samples, supporting it as a specific, non‐invasive diagnostic marker for EC and motivating further work on its expression–methylation dynamics and pathway context across EC, cervical cancer, and ovarian cancer.

Notably, the fact that previously reported EC methylation biomarkers, BHLHE22, CDO1, and CELF4, were not retained in our final three‐marker model does not contradict the findings of the previous study. A plausible explanation is that the CpG sites or genomic regions targeted for these genes in our assay differ from those interrogated previously. In addition, differences in study populations, sample types, experimental platforms, and marker‐selection strategies can all influence which methylated loci emerge as optimal markers. Thus, the divergence in final marker sets likely reflects methodological diversity rather than a fundamental inconsistency in biological signal.

Our results align with existing investigations, which emphasized the critical role of DNA methylation alterations in cancer diagnosis. Traditional methods for detecting endometrial cancer, such as pelvic ultrasound and endometrial sampling, pose certain limitations. Endometrial sampling remains the gold standard for diagnosing endometrial cancer, but its invasive nature is a significant drawback. On the other hand, ultrasonic measurement suffers from limited specificity [35]. A study by Iona Evans and colleagues demonstrated that the non‐invasive WID‐qEC test offers superior sensitivity (90.9%) and specificity (97.3%) for detecting endometrial cancer compared to pelvic ultrasound, which showed a sensitivity of 90.9% but a lower specificity of 79.1% [36]. Moreover, our research included benign lesions and other cancer types as controls, revealing that the SPDYA marker effectively differentiates between EC and other gynecological cancers. Besides, the diagnostic model we developed might also be used to test cervicovaginal self‐samples or urine in the future, further expanding its application.

At this point, we wish to highlight several main advantages demonstrated by our study: (1) rigorous experimental design with foundational discovery and validation, and all cases and controls pathologically confirmed; (2) identification of methylation biomarkers that are distinct from prior reports, conferring novelty and uniqueness; (3) a sampling workflow suitable for primary care settings that requires neither expensive equipment nor highly skilled personnel and does not rely on complex computations or ancillary clinical data (e.g., symptoms, gynecologic ultrasound) to interpret—making it simple and convenient; and (4) potential to inform clinical decision‐making that reduces invasive procedures such as hysteroscopy and dilation and curettage.

In clinical practice, our methylation‐based model may serve as a non‐invasive adjunctive tool in the diagnostic pathway, particularly after an abnormal ultrasound finding. In real clinical scenarios, many women with “suspicious” endometrial changes ultimately undergo unnecessary invasive procedures despite benign pathology. Our model, based on cervical exfoliated cells collected during TCT sampling, could help triage such patients: a positive result would support proceeding to hysteroscopic biopsy, whereas a negative result might justify deferring invasive testing and opting for closer surveillance. Although adding this test could introduce a short turnaround time, we consider it hypothesis‐generating that it may reduce procedure‐related burden and costs by decreasing unnecessary biopsies and enabling testing in lower‐resource settings. However, formal implementation studies and cost‐effectiveness evaluations will be required to substantiate these potential benefits.

There are also some limitations in the study. The clinical validation was conducted by employing the final‐confirmed model on 80 TCT samples. When stratified by menopausal status, the specificity for benign cases was 84.62% (11/13) among postmenopausal participants. The lower specificity in this postmenopausal subgroup is chiefly attributable to the very small number of benign controls (13 cases): just two misclassifications reduced specificity below 90%, indicating that this estimate is unstable and likely to improve with a larger benign sample. We also observed some differences in sensitivity and specificity between the premenopausal and postmenopausal subgroups. However, these differences are more likely driven by sample composition and limited subgroup size than by a true performance gap of the assay between premenopausal and postmenopausal women. Thus, we acknowledge that future studies should aim to include larger and more diverse cohorts from multiple centers to confirm the generalizability of our findings, and should be designed similarly to the clinical trial study published by Ruixiang Zhang et al. [37] The potential impact of confounding factors, such as age, sex, and underlying health conditions, on methylation patterns warrants further investigation in a large‐scale clinical study. Considering the histologic subtypes of EC patients included in this study, a limitation is that poor‐prognosis subtypes (such as serous carcinoma, clear cell carcinoma, and uterine carcinosarcomas) were represented by only one or two cases in the validation cohort, even though all of these cases were correctly identified as positive. Therefore, future large‐scale studies should include more patients with these subtypes to more reliably assess the robustness of the methylation markers across different EC subtypes. In addition, the dynamics of these markers over time should be assessed in future studies, which could provide valuable insights into their role in cancer progression and response to treatment, exploring a potential tool for the surveillance of EC.

In conclusion, we developed a three‐markers panel (ZNF626, GRIA4, and SPDYA) for diagnosing EC, demonstrating a sensitivity of 95% and a specificity of 90%, and a sensitivity of 92.86% for stage I cancer. It suggests that these markers could significantly enhance non‐invasive methylation detection for early diagnosis of EC. The successful identification and validation of these markers pave the way for future research aimed at integrating them into clinical practice, ultimately improving patient outcomes through earlier detection.

Author Contributions

Yan Cai: conceptualization, data curation, investigation. Shuchao Chen: data curation, methodology. Zhen Wu: data curation, methodology. Jinxia Guan: software. Jie Zhao: writing – original draft, writing – review and editing. Jianwei Ye: writing – original draft. Baochen Du: project administration. Xiaoliang Han: writing – review and editing. Tong Shu: conceptualization, data curation, resources. Guangpeng Zhou: investigation, supervision. Hong Zheng: conceptualization, project administration, resources, supervision, writing – review and editing.

Funding

This work was supported by Clinical Research Fund For Distinguished Young Scholars of Peking University Cancer Hospital, QNJJ202325.

Ethics Statement

This study was approved by The Ethics Committee of Peking University Cancer Hospital. The ethical approval numbers are as follows: The Ethics Committee of Peking University Cancer Hospital 2024YJZ37. All procedures were in accordance with the ethical standards of the responsible ethics committee and with the Helsinki Declaration of 1964 and later versions.

Consent

All informed consent was obtained from patients participating in the study.

Conflicts of Interest

The authors declare the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: Shuchao Chen, Zhen Wu, Jinxia Guan, Jie Zhao, Jianwei Ye, Baochen Du, Xiaoliang Han, and Guangpeng Zhou are current employees of BioChain (Beijing) Science & Technology Inc. The other authors declare no conflicts of interest.

Supporting information

Figure S1: The ROC curve of the MATS model used for distinguishing endometrial cancer from benign uterine disease in both training and test sets.

Figure S2: The performance evaluation for three‐gene logistic regression diagnostic model in the independent validation (80 TCT samples derived from 40 EC cases and 40 BE cases). (A) The ROC curve of the model in 80 TCT samples for distinguishing EC from BE. (B) The Sankey diagram for each marker and diagnostic model in detection of EC. I‐IV indicate clinical cancer stages I–IV.

Data S1: Supporting Information.

Table S1: Summary of methylation QC test results in marker‐discovery stage (qMSP designed for 156 DMRs).

Table S2: qMSP results of 107 candidate markers on tissue samples in marker‐verification stage (N = 16, including 8 cancer tissues and 8 benign disease tissues).

Table S3: qMSP results of 29 markers on TCT samples in marker‐verification stage (N = 16. TCT samples from 8 cancer patients and 8 normal individuals).

Table S4: qMSP results of 12 markers on 127 TCT samples in model‐construction stage (N = 127, including 58 endometrial cancer patients, 66 benign disease patients and 3 endometrial atypical hyperplasia cases).

Table S5: qMSP results of 12 markers on 24 TCT samples derived from other cancer patients in model‐construction stage (N = 24).

Table S6: qMSP results and MATs value of diagnostic model on 80 TCT samples in independent‐validation stage (N = 80. TCT samples from 40 endometrial cancer patients and 40 benign disease patients).

Data Availability Statement

All data generated or analyzed during this study are included in this published article and the file (Tables S1–, S6).