Plasma circulating tumor deoxyribonucleic acid methylation enables noninvasive disease stratification beyond prostate-specific antigen in prostate cancer

Abstract

Background

Accurate discrimination between localized and metastatic prostate cancer (PCa) remains a critical unmet need in clinical practice, particularly in patients with equivocal imaging findings or biologically high-risk disease. Circulating tumor DNA (ctDNA) methylation represents a stable epigenetic signal that may reflect tumor burden and disease state. We aimed to evaluate whether plasma ctDNA methylation markers can accurately discriminate disease state beyond conventional clinical parameters.

Methods

A total of 174 patients were included, comprising localized PCa (n = 73), metastatic hormone-sensitive PCa (mHSPC, n = 16), and metastatic castration-resistant PCa (mCRPC, n = 85). Candidate methylation markers (C2orf88 and HAPLN3) were identified using a machine learning–based approach and quantified in plasma ctDNA. Associations with metastatic disease were evaluated using logistic regression analyses, and diagnostic performance was assessed using receiver operating characteristic analyses.

Results

Plasma methylation levels of C2orf88 and HAPLN3 increased significantly with advancing disease stage (P < 0.05). In multivariable analysis, C2orf88 (OR 3.47, 95% CI 1.95–6.15; P < 0.001) and HAPLN3 (OR 2.82, 95% CI 1.65–4.82; P < 0.001) were independently associated with metastatic disease. The clinical model demonstrated modest discrimination (AUC range, 0.537–0.828), whereas methylation-based models achieved superior performance (AUC range, 0.807–0.843). The integrated two-marker signature provided the highest discriminatory accuracy (AUC = 0.913), indicating substantial incremental value over conventional clinical assessment.

Conclusions

A two-marker ctDNA methylation signature enables accurate discrimination between localized and metastatic PCa. This plasma-based epigenetic biomarker captures clinically meaningful disease biology and offers a scalable, noninvasive approach for refined disease stratification.

1. Introduction

Prostate cancer (PCa) exhibits marked biological heterogeneity, ranging from indolent localized tumors to aggressive metastatic disease.^1,2 Accurate discrimination between localized and metastatic PCa is critical for clinical decision-making, as treatment strategies and prognostic expectations differ substantially between these disease states. However, conventional clinical parameters such as prostate-specific antigen (PSA), radiographic imaging, and histopathologic features often fail to fully capture the underlying biological aggressiveness of the tumor.^{3, 4, 5}

Liquid biopsy approaches have emerged as powerful tools for noninvasive characterization of tumor biology. Circulating tumor cells (CTCs) have been extensively studied as biomarkers reflecting tumor burden and metastatic potential ^{6, 7, 8}. In our previous work, we demonstrated that CTC-based multigene expression signatures could effectively distinguish metastatic prostate cancer from localized disease, supporting the concept that circulating tumor-derived molecular information reflects disease progression and tumor biology.⁹ Despite their biological relevance, CTC-based assays have practical limitations due to the rarity of CTCs in circulation, technical complexity of isolation, and variability in assay reproducibility.^6,10 These limitations highlight the need for complementary liquid biopsy biomarkers that are more abundant, analytically robust, and easier to implement in clinical settings.

Circulating tumor DNA (ctDNA) has emerged as an attractive alternative biomarker that reflects tumor-derived molecular alterations.^{11, 12, 13, 14} Among ctDNA-based approaches, DNA methylation profiling offers several advantages, including high tissue specificity, relative stability, and greater abundance compared with mutation-based signals.^{15, 16, 17} Epigenetic alterations are known to occur early in tumorigenesis and evolve with disease progression, suggesting that ctDNA methylation burden may serve as a surrogate marker of tumor biology and disease state.^{18, 19, 20} Despite these advantages, the composition of clinically useful methylation marker panels and their ability to distinguish localized from metastatic disease remain to be defined.

In this study, we evaluated plasma ctDNA methylation burden at two candidate loci, C2orf88 and HAPLN3, identified through an AI-assisted discovery approach. We investigated whether ctDNA methylation burden reflects disease state and whether a simplified methylation signature can discriminate localized from metastatic prostate cancer.

2. Methods

2.1. Study population and patient cohort

This study evaluated plasma samples obtained from a total of 174 individuals, including patients with benign prostatic hyperplasia and prostate cancer across different disease states. The prostate cancer cohort comprised localized prostate cancer (n = 73), metastatic hormone-sensitive prostate cancer (mHSPC, n = 16), and metastatic castration-resistant prostate cancer (mCRPC, n = 85). All samples were collected prior to initiation of systemic therapy or definitive treatment. Clinical and pathological characteristics of the study population are summarized in Table 1. The study protocol was approved by the institutional review board, and all participants provided written informed consent.

Table 1.

Baseline clinicopathologic characteristics of patients with benign prostatic hyperplasia, localized prostate cancer, and metastatic prostate cancer.

	Localized stage	Metastatic stage
	T2-4 (N = 73)	mHSPC (N = 16)	mCRPC (N = 85)
Age, median (IQR), years	70.0 (66.0–76.0)	76.0 (71.0–78.0)	76.0 (72.0–82.0)
PSA, median (IQR), ng/mL	9.3 (5.1–23.1)	199.2 (13.4–928.2)	35.6 (7.3–156.0)
Gleason score (%)
6	22 (30.1)	0 (0.0)	0 (0.0)
7	23 (31.5)	3 (18.8)	6 (7.1)
≥8	28 (38.4)	13 (81.2)	79 (92.9)
Node metastasis (%)
Yes	6 (8.2)	12 (75.0)	77 (90.6)
No	67 (91.8)	4 (25.0)	8 (9.4)
Bone metastasis (%)
Yes	0 (0.0)	15 (93.8)	83 (97.6)
No	73 (100.0)	1 (6.2)	2 (2.4)
Visceral metastasis (%)
Yes	0 (0.0)	10 (62.5)	59 (69.4)
No	73 (100.0)	6 (37.5)	26 (30.6)

mCRPC, metastatic castration-resistant prostate cancer; mHSPC, metastatic hormone-sensitive prostate cancer; PSA, prostate-specific antigen.

2.2. Identification of candidate methylation markers

Candidate methylation loci were selected through an artificial intelligence–assisted discovery workflow integrating publicly available prostate cancer methylome datasets, including the Cancer Genome Atlas (TCGA). DNA methylation data from these datasets were generated using the Infinium HumanMethylation450 BeadChip platform (Cat. No. WG-314-1002, Illumina), which interrogates over 450,000 CpG sites across the genome according to the manufacturer's standard protocol. Genome-wide methylation profiles were analyzed to identify CpG regions demonstrating tumor-specific hypermethylation with minimal background signal in normal tissues and peripheral blood. Machine learning–based feature selection algorithms were applied to prioritize markers with high discriminatory potential, leading to the selection of two loci, C2orf88 and HAPLN3, for downstream validation.

2.3. Plasma processing and ctDNA isolation

Peripheral blood samples were collected in EDTA-containing tubes and processed within a standardized time window. Plasma was separated by centrifugation and stored at −80°C until analysis. Cell-free DNA was extracted from 3 ml of plasma using a magnetic bead–based isolation method, MagMAX Cell-Free DNA Isolation Kit.

(Cat. No. A29319; Thermo Fisher Scientific, Waltham, MA, USA) according to the manufacturer's protocol. DNA concentration and quality were assessed prior to downstream analysis.

2.4. Bisulfite conversion and methylation assessment

Extracted cell-free DNA underwent enzymatic digestion (Acil restriction enzyme) followed by bisulfite conversion using the EZ DNA Methylation-Lightning Kit (Cat. No. D5030; Zymo Research, Irvine, CA) to enable methylation-specific detection. Methylation levels at target loci were quantified using a methylation-sensitive high-resolution melting–based assay optimized for low-input DNA. The resulting melting profiles were analyzed to derive methylation height values representing relative methylation burden at each locus.

2.5. Quantification of methylation burden

Methylation burden was defined as the peak fluorescence intensity within the target melting temperature range following derivative curve transformation of fluorescence signals. For each marker, methylation height values were calculated and used for downstream statistical analyses. A combined methylation score was calculated by summing normalized marker values to evaluate panel performance.

2.6. Logistic regression analysis

Logistic regression analyses were performed to evaluate associations between clinical variables (serum PSA, age, and Gleason score) and methylation markers with metastatic prostate cancer. Both univariable and multivariable models were constructed, and odds ratios with 95% confidence intervals were estimated.

2.7. Diagnostic performance and ROC analysis

Receiver operating characteristic (ROC) analyses were conducted to evaluate the discriminatory performance of clinical variables, methylation markers, and integrated models in distinguishing localized from metastatic prostate cancer. Area under the curve (AUC) values were calculated to assess model performance. Comparisons between models were performed using standard statistical methods.

2.8. Statistical analysis

Group comparisons were performed using nonparametric statistical tests as appropriate. Logistic regression analyses were conducted to assess the association between clinical variables, methylation markers, and metastatic prostate cancer, and odds ratios with 95% confidence intervals were estimated. All statistical analyses were performed using SPSS software (version 25.0; IBM Corp., Armonk, NY, USA) and R statistical software. Statistical significance was defined as a two-sided P-value < 0.05.

3. Results

3.1. Study cohort and analytical workflow

A total of 174 patients were included, comprising localized prostate cancer (n = 73), metastatic hormone-sensitive prostate cancer (mHSPC, n = 16), and metastatic castration-resistant prostate cancer (mCRPC, n = 85). The overall study design and analytical workflow are illustrated in Fig. 1. Baseline clinical characteristics are summarized in Table 1. Median PSA levels were substantially higher in metastatic disease compared with localized prostate cancer (199.2 and 35.6 vs. 9.3 ng/mL). High-grade tumors (Gleason score ≥8) were more prevalent in metastatic disease (81.2% in mHSPC and 92.9% in mCRPC) than in localized disease (38.4%), indicating a more aggressive disease phenotype.

Figure 1.

Study design and workflow of plasma-based ctDNA methylation analysis. Plasma samples were collected from patients with localized prostate cancer (LPC, n = 73), metastatic hormone-sensitive prostate cancer (mHSPC, n = 16), and metastatic castration-resistant prostate cancer (mCRPC, n = 85), totaling 174 samples. Plasma-derived cell-free DNA was analyzed for ctDNA methylation, and candidate markers were identified using a machine learning–assisted approach. ctDNA, circulating tumor DNA.

3.2. Plasma ctDNA methylation burden reflects disease progression

Quantitative analysis of plasma ctDNA methylation revealed a marked and consistent increase with advancing disease state. Both C2orf88 and HAPLN3 demonstrated significantly elevated methylation levels in metastatic prostate cancer compared with localized disease (P < 0.001), with a clear separation between localized and metastatic groups. Notably, methylation levels were comparable between mHSPC and mCRPC, indicating that ctDNA methylation burden reaches a plateau once systemic disease is established (Fig. 2). The combined methylation score further enhanced this distinction, providing the most pronounced separation between localized and metastatic disease states. These findings indicate that plasma ctDNA methylation captures stage-specific epigenetic alterations and serves as a reliable indicator of metastatic disease biology.

Figure 2.

Stage-dependent alterations in plasma ctDNA methylation burden across the spectrum of prostate cancer disease states. Plasma methylation levels of C2orf88, HAPLN3, and the combined score increase with disease stage across BPH, localized prostate cancer, mHSPC, and mCRPC. Methylation burden is significantly elevated in metastatic disease, with the combined score clearly discriminating disease states (P < 0.001). ctDNA, circulating tumor DNA; mCRPC, metastatic castration-resistant prostate cancer; mHSPC, metastatic hormone-sensitive prostate cancer; PCa, prostate cancer.

3.3. Distinct methylation landscape across disease states

Heatmap visualization revealed distinct methylation patterns across patient groups, demonstrating progressive enrichment of methylation burden from localized to metastatic disease (Fig. 3). The combined methylation score further highlighted the coordinated behavior of the two-marker signature and supported the presence of a coherent epigenetic gradient across disease states. These observations reinforce the biological relevance of ctDNA methylation as a reflection of systemic tumor biology.

Figure 3.

Differential ctDNA methylation patterns across disease states in prostate cancer. Heatmap of plasma methylation levels for C2orf88, HAPLN3, and combined methylation height across localized prostate cancer (n = 73), and metastatic prostate cancer (n = 101). Columns represent individual patients; color scale indicates relative methylation level. Both markers show a stage-associated increase, with higher levels in metastatic disease. ctDNA, circulating tumor DNA; PCa, prostate cancer.

3.4. Independent association of methylation markers with metastatic disease

Logistic regression analyses demonstrated that Gleason score and ctDNA methylation markers were significantly associated with metastatic prostate cancer (Table 2). In multivariable analysis, C2orf88 methylation (OR 3.47, 95% CI 1.95–6.15, P < 0.001) and HAPLN3 methylation (OR 2.82, 95% CI 1.65–4.82, P < 0.001) remained independently associated with metastatic disease, whereas age and PSA were not significant after adjustment. These findings indicate that methylation markers provide disease-state information beyond conventional clinical parameters.

Table 2.

Clinical and epigenetic factors associated with metastatic prostate cancer.

	Univariate			Multivariate
Variable	OR	95% CI	P value	OR	95% CI	P value
Age	1.09	1.04–1.14	<0.001	1.06	0.99–1.14	0.090
Serum PSA	0.92	0.92–1.86	0.130	0.75	0.41–1.37	0.340
Gleason score (≥8 vs. ≤7)	2.42	1.73–3.39	<0.001	1.74	1.29–2.66	<0.001
C2orf88	4.62	2.80–7.62	<0.001	3.47	1.95–6.15	<0.001
HAPLN3	3.44	2.31–5.13	<0.001	2.82	1.65–4.82	<0.001
Combined methylation score (C2orf88+HAPLN3)	4.74	3.17–7.09	<0.001	–	–	–

PSA, prostate-specific antigen.

3.5. Diagnostic performance of methylation markers

Receiver operating characteristic analyses demonstrated modest discrimination with the clinical model (AUC range, 0.537–0.828). Methylation-based models using individual markers showed improved performance (AUC range, 0.807–0.843). The combined two-marker methylation model achieved the highest discriminatory accuracy (AUC = 0.913), indicating that integration of methylation markers enhances predictive performance compared with clinical variables or individual markers alone (Fig. 4).

Figure 4.

ROC curve analysis demonstrating the discriminatory performance of clinical and ctDNA methylation models for metastatic prostate cancer. (A) The clinical model (PSA, Age, and Gleason score) showed modest discrimination (AUC = 0.537–0.828). (B) The methylation model based on individual markers (C2orf88 or HAPLN3) demonstrated improved performance (AUC = 0.807–0.843). The combined methylation model incorporating both markers achieved the highest discriminatory power (AUC = 0.913), indicating that the integration of multiple methylation markers significantly enhances predictive accuracy compared to individual markers or clinical variables alone. AUC, area under the curve; ctDNA, circulating tumor DNA; PSA, prostate-specific antigen; ROC, receiver operating characteristic.

4. Discussion

This study demonstrates that plasma ctDNA methylation burden reflects disease state and enables reliable discrimination between localized and metastatic prostate cancer. Among the evaluated loci, C2orf88 and HAPLN3 showed consistent stage-associated increases and formed a biologically coherent epigenetic signature. These findings support the concept that circulating tumor-derived methylation signals capture the systemic molecular landscape of tumor progression and may serve as a surrogate marker of disease extent.^21,22 The use of a simplified two-marker panel may facilitate clinical implementation by minimizing assay complexity while maintaining strong discriminatory performance.

Methylation markers were identified using a data-driven approach integrating TCGA methylome datasets with machine learning–based feature selection. This strategy enabled the selection of loci with high tissue specificity and strong discriminatory capacity. Such approaches extend beyond conventional hypothesis-driven biomarker discovery and support the identification of clinically relevant epigenetic signals.^23,24 The resulting two-marker panel illustrates how methylomic information can be translated into a clinically applicable assay.

The distinct methylation signatures identified between localized and metastatic prostate cancer provide biological plausibility for their role as stage-specific biomarkers. Epigenetic alterations are known to accompany tumor progression and may reflect changes in gene regulatory programs associated with metastatic competence ^{25, 26, 27, 28}. Circulating tumor-derived methylation signals therefore likely represent not only tumor presence but also differences in tumor biology related to disease dissemination. These findings support the concept that circulating epigenetic markers can serve as minimally invasive indicators of disease state, capturing biologically meaningful information beyond conventional clinical parameters.^18,29,30

From a clinical perspective, a blood-based methylation assay offers a noninvasive tool that may complement conventional diagnostic approaches. Current clinical parameters such as PSA and imaging provide limited insight into biological aggressiveness, particularly in patients with occult metastatic potential.^3,31,32 The ability to capture disease state using a small number of methylation markers offers practical advantages, including analytical simplicity and potential scalability for routine clinical use. Such biomarkers may improve patient stratification and help identify individuals with biologically advanced disease who may benefit from intensified therapeutic strategies.

Previous studies investigating circulating biomarkers in prostate cancer have largely focused on genomic alterations, including mutations, copy number changes, or circulating tumor cell–based assays, to assess disease burden and progression ^{33, 34, 35}. While these approaches have provided important insights, they are often limited by technical complexity, low detection sensitivity in certain clinical settings, or restricted biological scope.³⁶ In contrast, methylation-based assays offer the advantage of capturing stable epigenetic signals that may more broadly reflect tumor biology.

One technical challenge in utilizing liquid biopsy for localized prostate cancer is that ctDNA levels are often near or below the limit of detection compared to the robust signals found in metastatic disease. However, DNA methylation frequently occurs as an early event in prostate cancer progression and can be detected even when total ctDNA fractions are low. In our study, the significant difference in methylation levels between localized and metastatic groups suggests that while the absolute quantity of ctDNA differs, the specific methylation patterns of C2orf88 and HAPLN3 remain detectable and serve as a surrogate for complementary tools for disease stratification.

4.1. The selection of C2orf88 and HAPLN3 is supported by their roles in the metastatic cascade

C2orf88 has been identified as a functional gene associated with cell cycle regulation and tumor progression.³⁷ Similarly, HAPLN3 (Hyaluronan and proteoglycan link protein 3) is involved in remodeling the extracellular matrix, which is a prerequisite for tumor cell invasion and EMT.³⁸ The independent association of these markers with metastatic disease in our multivariable analysis underscores their role as biological drivers of progression beyond traditional PSA signaling.

Several limitations warrant consideration. The cohort size was modest and external validation in independent populations is needed to confirm generalizability. In addition, although the findings suggest that methylation burden reflects disease state, longitudinal studies are required to determine its predictive value for disease progression and treatment outcomes. Future research integrating methylation profiling with other molecular modalities may further refine risk stratification approaches and enhance our understanding of prostate cancer biology.

5. Conclusion

A simple two-marker ctDNA methylation signature composed of C2orf88 and HAPLN3 reflects disease state and enables accurate discrimination between localized and metastatic prostate cancer. This plasma-based epigenetic biomarker offers a clinically translatable approach for disease stratification beyond PSA alone.

Funding

This work was supported by the National Research Foundation of Korea (NRF) and funded by the Korean government (MSIT) [NRF-2017M3A9A7050614, NRF-2021R1C1C2007117, and RS-2025-00516869].

Conflicts of interest

All the authors declare no potential conflicts of interest to disclose. Given his role as Editorial Board Member of Prostate International, Jae-Seung Chung had no involvement in the peer review of this article and had no access to information regarding its peer review.