Using cell-free DNA tumor methylation score to detect and monitor cholangiocarcinoma

Abstract

Background:

Cholangiocarcinoma (CCA) is characterized by late or unclear diagnosis, high recurrence rates, and limited options for noninvasive disease monitoring. The widely used serum biomarker CA 19-9 has suboptimal sensitivity and specificity, underscoring the need for more accurate and reliable quantitative measures of tumor burden. Cell-free DNA (cfDNA) methylation represents a promising approach for longitudinal disease monitoring.

Methods:

In this single-center cohort study, we evaluated a cfDNA methylation-based Tumor Methylation Score (TMS) for detecting and monitoring CCA. Twenty-five patients with histologically confirmed CCA and twelve controls with benign hepatobiliary conditions were enrolled, yielding 100 plasma samples and 26 tissue samples. cfDNA methylation was quantified across 537 cancer-associated CpG loci by qPCR method, and TMS was calculated from the difference in methylation in paired plasma and buffy coat samples. For tissue samples, TMS was calculated by quantifying methylation in tumor and background liver separately. Response to treatment was assessed using blinded radiographic review. Tissue-based validation was performed using The Cancer Genome Atlas and 4 additional GEO cohorts. Finaly, cellular pathways associated with the panel genes were explored using the Ingenuity Pathway Analysis (IPA).

Results:

TMS was significantly elevated in CCA samples compared with controls in both tissue (P=0.007) and plasma (P<0.0001). Plasma TMS distinguished CCA from controls with a sensitivity of 74.3% and specificity of 80% (AUC=0.823), while CA 19-9 showed no discriminatory value. TMS decreased consistently following curative resection. Longitudinal changes in TMS (ΔTMS) robustly distinguished radiographic progression from non-progression during systemic therapy (AUC=0.968), outperforming ΔCA 19-9. Tissue-based analyses across independent cohorts confirmed cancer specificity and association with adverse outcomes. IPA identified HOXA9 as a significantly altered gene implicated in CCA carcinogenesis.

Conclusions:

Quantitative cfDNA methylation monitoring using TMS is feasible and accurately reflects dynamic tumor burden in CCA. TMS may serve as a complementary adjunct to imaging and CA 19-9 for diagnosis, prognosis, and longitudinal disease monitoring, warranting validation in larger prospective studies.

INTRODUCTION

Cholangiocarcinoma (CCA) is a malignant neoplasm arising from biliary tract epithelial cells and is classified into intrahepatic (iCCA) and extrahepatic (eCCA) subtypes^1,2. A major risk factor for CCA is chronic inflammation of the biliary tree, with major contributors including primary sclerosing cholangitis (PSC), liver fluke infections, and chronic liver diseases such as cirrhosis and viral hepatitis^3,4. The incidence of CCA increases with age, with a median age at diagnosis of 65 years^5,6.

CCA is most commonly diagnosed at advanced stages of disease. Historically, untreated advanced disease was associated with a median survival of 3-6 months; however, systemic therapy now confers a median survival of approximately 11-12 months^7,8. In carefully selected patients with unresectable intrahepatic disease without systemic involvement, liver transplantation may offer a curative option^9,10. Following curative-intent surgery or during systemic therapy, monitoring tumor eradication, treatment response, and recurrence is essential. This process is commonly supported by tumor markers such as carbohydrate antigen 19-9 (CA 19-9)¹¹. Even after curative-intent surgery, recurrence rates in cholangiocarcinoma remain high, with more than 60% of patients experiencing disease relapse within two years^12,13. Therefore, reliable, quantitative, and noninvasive biomarkers for dynamic monitoring of tumor eradication, disease burden, and recurrence are urgently needed. For patients with advanced disease, systemic chemotherapy, immunotherapy, and locoregional radiotherapy are the primary therapeutic options¹⁴. In all these settings, a reliable biomarker reflecting disease burden is essential for monitoring treatment response and guiding therapeutic decision making. Currently, CA 19-9 remains the most widely used biomarker for monitoring CCA; however, its clinical utility is limited by several factors. In pooled analysis, the sensitivity and specificity of CA 19-9 range from 60 to 78% and 50 to 91%, respectively^15-19. Additionally, CA 19-9 is not expressed in a substantial subset of patients^20,21 and may be falsely elevated in benign biliary conditions²². While other biomarkers have been investigated for CCA detection and monitoring, they lack validation in large cohorts^23,24.

Cell-free DNA (cfDNA) has emerged as a promising tool for tumor detection and disease monitoring across various cancers^25,26. As a minimally invasive, blood-derived method, cfDNA allows for dynamic and frequent monitoring throughout treatment²⁷. Furthermore, cfDNA overcomes key limitations of tissue biopsies, including sampling bias and intratumoral heterogeneity^28-30. Most cfDNA assays in CCA have focused on mutation-based detection, either through tumor-informed approaches requiring prior tumor sequencing or tumor-agnostic panels targeting recurrent genomic alterations. However, the mutational landscape of CCA is heterogeneous, and actionable mutations are not universally present, limiting the applicability of mutation-based cfDNA assays in routine clinical practice^31,32. Differentiating tumor-derived circulating DNA from DNA derived from normal tissue can be achieved using various computational or bioinformatics techniques³³. Most commercially available approaches require a priori knowledge of tumor-specific genetic alterations; in contrast, emerging methodologies leverage DNA methylation patterns to detect the presence and burden of cancer.

DNA methylation represents an alternative and complementary approach to cfDNA analysis. DNA methylation is an epigenetic mechanism that regulates gene expression. CpG islands, regions of the genome enriched with cytosine and guanine bases, undergo methylation, which alters the binding affinity of histones and transcription factors and subsequently impacts transcription³⁴. Methylation patterns are cell-type specific, reflect both oncogenic transformation and tissue of origin, and are highly prevalent across cancer genomes. Distinct methylation patterns in cfDNA have been observed in different etiologies of CCA³⁵. Typically, promoter methylation leads to gene silencing, with tumor suppressor genes being hypermethylated and proto-oncogenes hypomethylated in cancerous tissue³⁶. The unique methylation patterns in cfDNA allow for differentiation between normal and tumor-derived DNA. Large scale methylation-based assays have demonstrated the ability to detect multiple cancer types simultaneously, including CCA, in multi-cancer early detection settings, such as the GRAIL Galleri test. However, this assay is a binary test primarily designed for cancer signal detection rather than quantitative monitoring of disease burden or recurrence in patients with established malignancy^37,38.

In this study, we quantified a cfDNA methylation panel termed the Tumor Methylation Score (TMS) to evaluate its performance in detecting and dynamically monitoring tumor burden change, with the goal of improving early recurrence detection and treatment response monitoring beyond the limitations of CA 19-9 and imaging. By focusing on dynamic disease monitoring rather than early detection, this study seeks to define a potential role for cfDNA methylation quantification as an adjunct to imaging and existing biomarkers in the management of patients with cholangiocarcinoma.

METHODS

Study Population and Design

This single-center prospective cohort study included 25 patients with histologically confirmed CCA undergoing tumor resection (N=7), systemic therapy (N=9), or combined surgical and systemic treatment (N=9), as well as 12 control patients with benign hepatobiliary masses (N=7) or PSC (N=5). A total of 100 blood samples were collected across multiple time points, including 75 CCA and 25 control, respectively. A total of 26 tissue samples were collected, including 22 from CCA patients and 4 from controls. Informed consent was obtained from all participants before enrollment, and the study was approved by the Institutional Review Board (IRB202500304) at the University of Florida Health. All research procedures were conducted with strict adherence to the Declaration of Helsinki. Comprehensive demographic data were collected and are summarized in Table 1.

Table 1.

Patient demographics and medical history

Variable	Cholangiocarcinoma (n=25)	Control (n=12)
Age, y, median (IQR)	64 (12.5)	52 (31)
Sex, n (%)
Male	14 (56)	5 (41)
Female	11 (44)	7 (59)
BMI, kg/m2, median (IQR)	27.64 (8.2)	27.88 (10.59)
Race, n (%)
White	21 (84)	5 (41)
Black	1 (4)	4 (34)
Other	3 (12)	3 (25)
Risk factors, n (%)
Alcohol	13 (52)	1 (9)
Tobacco smoke	20 (80)	3 (25)
Obesity	9 (36)	6 (50)
Drugs	6 (24)	0 (0)
PSC	2 (8)	5 (41)
Other previous cancers	2 (8)	2 (17)
Tumor stage, n (%)
1	7 (28)
2	4 (16)
3	4 (16)
4	8 (32)
NA	2 (8)
Tumor subtype, n (%)
Intra-hepatic	18 (72)
Extra-hepatic	7 (28)

Sample Collection and Processing

Pre-surgical and post-surgical blood samples, as well as tissue samples, were collected from patients. Blood samples were also collected at multiple time points during the course of systemic therapy. Blood was drawn into K2-EDTA tubes and centrifuged at 4°C at 1,000 × g for 15 minutes to separate plasma and buffy coat fractions, which were subsequently stored at −80°C for later analysis. Tissue samples were collected into cryovials in the operation room and stored in −80 for subsequent analysis. Imaging studies, including computed tomography (CT) and magnetic resonance imaging (MRI), were performed per standard of care to assess disease burden in CCA patients. CA 19-9 levels were measured using a sandwich enzyme-linked immunosorbent assay (ELISA) method.

Cell-Free DNA Quantification

Methylated cfDNA was quantified by targeting 537 preselected cancer-specific CpG sites as previously described³⁹. For blood samples, quantitative polymerase chain reaction (qPCR) was used to amplify the selected CpG loci in plasma and buffy coat fractions, and the TMS was calculated as the difference in the number of methylated molecules per 1,000 input genomic equivalents between plasma (representing tumor-derived cfDNA) and buffy coat (representing hematologic somatic DNA). For tissue samples, a similar approach was followed to amplify the selected genomic loci, and since these values were extracted from the tumor or normal tissues, TMS was calculated by summing methylation values within tumor or normal tissue samples separately. Numerically, higher TMS values reflect a greater abundance of methylated cfDNA fragments attributable to non-hematologic sources, consistent with increased tumor-derived cfDNA shedding. TMS values are unitless and are intended for comparative analysis across samples and time points within this assay framework rather than as absolute biological quantities.

Assessing the Response to Treatment

To evaluate response to chemotherapy using TMS, consecutive imaging studies were reviewed by an expert radiologist who was blinded to patient identity and clinical course. Consecutive imaging studies (CT and MRI), performed as part of standard of care and corresponding to TMS blood draw time points, were analyzed. Tumor burden was assessed using three criteria:

1. Change in the size of the primary tumor

2. Change in the size of established metastatic nodules in the abdomen or chest

3. Development of new metastases in the abdomen or chest

If any of these criteria showed progression, the interval between the two consecutive samples was classified as disease progression. If none were present, the interval was considered non-progression. These criteria are detailed in Supplementary e-Table 1. We then calculated the corresponding ∆TMS values for both groups and used these to assess the efficacy of TMS in monitoring tumor burden.

Database Analysis

Publicly available methylation data were obtained from The Cancer Genome Atlas (TCGA⁴⁰) and Gene Expression Omnibus (GEO) databases to provide biological context for the cfDNA methylation panel and to serve as a comparison cohort for tissue-based analyses. These analyses were not intended to evaluate diagnostic performance but rather to examine whether aggregate methylation across the same CpG panel correlated with tumor presence and clinical outcomes in tissue samples. The beta methylation value (BMV) was calculated by summing the beta values of the 537 selected CpG sites for each patient’s tissue sample, similar to the way that we had calculated TMS in our own tissue sample cohort. Since not all cohorts contained all 537 CpG sites, a normalization factor was applied based on the cohort with the fewest overlapping CpG sites. Specifically, GSE156299⁴¹, which contained 451 of the 537 CpG sites, was used as the reference cohort and assigned a fraction number of 1. For all other cohorts, the fraction number was calculated by dividing 451 by the number of CpG sites present (e.g., for TCGA, which included all 537 CpG sites, the fraction number was 451/537 = 0.84). Each patient’s BMV was then calculated by summing the available beta values and multiplying by the corresponding fraction number, enabling consistent comparison and integrated analysis across cohorts.

Pathway Analysis

Ingenuity Pathway Analysis (IPA) software was used for gene and pathway analysis. RNA sequencing and methylation data from TCGA were integrated to construct and analyze molecular pathways.

Statistical Analysis

Wilcoxon paired tests and Kruskal-Wallis tests with Dunn’s multiple comparisons were used to assess statistical significance. The Youden index was applied to determine the optimal cutoff value for dichotomizing TMS. Receiver operating characteristic (ROC) curve analysis with area under the curve (AUC) estimation, along with Fisher’s exact tests, was used to evaluate test performance. A p-value <0.05 was considered statistically significant. All analyses were performed using GraphPad Prism (version 10.0.2 for Windows, Boston, MA) and RStudio (version 3.6.0)⁴².

RESULTS

TMS in detecting the presence of cancer

TMS values were calculated across all available time points for both tissue and blood samples. TMS values were significantly higher in cancer samples compared to controls, in both tissue (P=0.007) and blood (P < 0.0001) (Figure 1A). A cutoff value of 96.38 correctly classified 57 of 75 cancer blood samples and 20 of 25 control samples, corresponding to a sensitivity of 74.3% and a specificity of 80% (AUC=0.823) (Figure 1B).

Figure 1. Tumor Methylation Score (TMS) in Detecting the Presence and Curative Resection of Cholangiocarcinoma.

(A) Comparison of Tumor Methylation Score (TMS) values between cholangiocarcinoma (CCA) samples and controls in tissue and plasma. TMS values were significantly higher in CCA samples compared with controls in both tissue and blood.

(B) Receiver operating characteristic (ROC) curve showing the performance of plasma TMS for distinguishing CCA from control samples. A cutoff value of 96.38 yielded a sensitivity of 74.32% and specificity of 80% (AUC = 0.823).

(C) Serum CA 19-9 values in CCA and control samples. CA 19-9 failed to significantly differentiate cancer from control samples and showed substantial interpatient variability.

(D) Longitudinal plasma TMS values in patients undergoing curative-intent surgical resection. TMS values decreased consistently within the first postoperative week for all evaluated patients, consistent with curative resection.

We also analyzed TMS across CCA subtypes and observed that TMS was significantly elevated in both iCCA and eCCA compared to controls. However, no significant difference was found between the subtypes themselves (Supplementary e-Figure 1).

We included CA 19-9 values when available from the same samples used for TMS analysis. CA 19-9 failed to significantly differentiate CCA (N=32) from control samples (N=10) in our cohort (P=0.8), and substantial interpatient variability was observed (Figure 1C). No strong correlation was observed between CA 19-9 and TMS values (Supplementary e-Figure 1B).

TMS in detecting curative resection

To evaluate the ability of TMS to detect curative resection, we analyzed seven patients undergoing curative tumor resection. Pre- and post-surgical blood samples were collected from these patients, and TMS values were tracked longitudinally. TMS values significantly decreased in all patients during the first postoperative week following surgery (Figure 1D).

TMS in distinguishing tumor burden changes

To assess whether dynamic changes in cfDNA methylation burden reflect changes in disease status, serial blood samples were collected from patients through the course of their treatment. Patients provided a mean of 5.3 samples, with a mean interval of 25 days between consecutive samples, spanning an average follow-up period of 4 months per patient. Patients received a variety of systemic therapies, including lenvatinib, gemcitabine, cisplatin, pembrolizumab, durvalumab, ivosidenib, and talazoparib.

TMS values for progression and non-progression groups are shown in Figure 2A. The change in TMS (ΔTMS) was calculated for each consecutive sample pair. Comparative analysis demonstrated that ΔTMS significantly distinguished periods of disease progression from non-progression (P=0.0002) (Figure 2B). ROC analysis revealed an AUC of 0.968 (Sensitivity: 100%, Specificity: 85.71%, Likelihood Ratio: 7) for a cutoff value of 325.5 (Figure 2C). Given the limited number of progression events, these findings are best interpreted as evidence of biological association rather than definitive performance characteristics.

Figure 2. TMS Dynamics in Monitoring Changes in Tumor Burden.

(A) Plasma TMS values corresponding to intervals classified as radiographic progression or non-progression during systemic therapy. White squares represent non-progression and gray circles represent progression time-points.

(B) Change in Tumor Methylation Score (ΔTMS) between consecutive sampling time points, showing a significant difference between progression and non-progression intervals.

(C) ROC curve evaluating ΔTMS for distinguishing radiographic progression from non-progression, with an AUC of 0.968.

(D) Change in CA 19-9 (ΔCA 19-9) across the same intervals, showing no significant association with progression status. (N=11)

(E) Representative patient illustrating longitudinal TMS measurements over time, with a marked increase in TMS preceding radiographic disease progression.

(F-H) Corresponding computed tomography (CT) images from the same patient at sequential time points, confirming radiographic disease progression.

Notably, in one patient, changes in TMS preceded radiographic evidence of progression. As shown in Figure 2E, one patient exhibited a tenfold increase in TMS approximately two months before progression was confirmed by imaging. Corresponding CT scans at each time point are shown in Figures 2F-H. A detailed description of the clinical changes in primary tumor and metastasis size, ΔTMS, ΔCA19–9, and the number of days between consecutive samples are summarized in Supplementary e-Table 2.

ΔCA 19-9 values were also calculated from seven progression and four non-progression time periods, using the same intervals used for ΔTMS analysis. CA 19-9 changes did not show a significant association with either progression or non-progression (Figure 2D) and no strong correlation was observed between ΔTMS and ΔCA 19-9 (Supplementary e-Figure 1C).

Validation in TCGA Cohort

To provide biological context for the cfDNA methylation panel and external validation of tissue-based findings, methylation data from TCGA were examined. The TCGA CCA dataset included 37 tumor tissues and 9 matched non-cancerous background tissues. For each sample, approximately 450,000 CpG sites were analyzed to determine methylation levels. Methylation was expressed as the beta value, calculated as the number of methylated bases divided by the total number of methylatable sites within a CpG island, resulting in a value ranging from 0 to 1. To compute the analogue of TMS for the TCGA dataset, BMV, beta values across the 537-amplicon panel were summed for each sample.

As illustrated in Figure 3A, BMV effectively distinguished CCA tumor tissues from non-cancerous background tissues. Additionally, a significant inverse correlation between BMV and patient survival was identified (Figure 3B). Although not statistically significant, a trend toward higher BMV values with increasing tumor stage was also observed (Figure 3C). To assess whether BMV reflected cancer-specific rather than age-related methylation changes, the relationship between BMV and patient age was analyzed. No correlation was found, reinforcing the specificity of BMV for detecting cancer (Figure 3D).

Figure 3. Tissue-Based Validation of the Methylation Panel in TCGA and Independent Cohorts.

(A) Comparison of beta methylation value (BMV) between cholangiocarcinoma tumor tissues and matched non-cancerous background tissues in the TCGA cohort.

(B) Kaplan-Meier survival analysis stratified by BMV, showing an inverse association between aggregated methylation burden and overall survival.

(C) BMV values across tumor stages, showing a non-significant trend toward higher methylation burden with increasing stage.

(D) Relationship between BMV and patient age, showing no correlation.

(E) Combined analysis of multiple independent cohorts showing significantly higher BMV values in both intrahepatic and extrahepatic CCA tissues compared with cancer-free controls, with no significant difference between CCA subtypes.

Validation in other cohorts

We evaluated the performance of this methylation panel across multiple independent cohorts, summarized in Supplementary e-Table 3. The combined dataset included 152 CCA tissues, comprising 127 iCCA and 25 eCCA cases, along with 72 cancer-free controls and 60 precancerous tissue samples. BMV values were normalized across cohorts to enable integrated analysis.

Several individual datasets provided additional insights. Goeppert et al. examined genetic and epigenetic progression from normal tissue to precursor lesions, intraductal tubulopapillary and papillary neoplasm of the bile duct (ITPN and IPNB, respectively) and invasive CCA. BMV was significantly different between cancerous, precursor, and normal tissues⁴¹ (Supplemental Figure 2A). Similarly, analysis of the Jusakul et al. dataset demonstrated significant differences in BMV between CCA and cancer-free controls⁴³ (Supplemental Figure 2B). A separate study by Goeppert et al. focused specifically on intrahepatic CCA, and BMV again successfully differentiated cancer from non-cancer tissues⁴⁴ (Supplemental Figure 2C).

We combined all CCA and control tissues from these cohorts (Figure 3E) and found BMV significantly different between CCA tissues and cancer-free controls for both iCCA and eCCA subtypes. No significant difference in BMV was observed between iCCA and eCCA.

Gene and pathway analysis

To better understand the biological relevance of the genes in our panel, pathway analysis was conducted on the genes associated with each amplicon. Using IPA software, gene interaction networks were constructed linking each gene to cancer-related functions, such as metastasis and biliary tumor growth. Pathway activation or suppression was then assessed in CCA relative to non-cancerous control tissue using gene expression data from TCGA. Among the pathways related to our set of amplicons, the HOXA9 gene was identified as significantly differentially regulated (p < 0.001). (Figure 4A)

Figure 4. Ingenuity Pathway Analysis for the genes of the cfDNA panel.

(A) Genes from the methylated cfDNA panel were mapped using TCGA CCA data. HOXA9 was significantly differentially regulated (P<0.001) and was associated with multiple metastasis-related genes (e.g., EDNRA, KIT, FLT3, CD36, E-selectin, ERG) with concurrent downregulation of tumor suppressor genes (Caspase 3, Caspase 8, CDKN2A).

(B) Volcano plot showing the pathways with the largest difference between CCA tissue samples and controls. S-100 family signaling, mitotic tubules assembly, collagen degradation, and extracellular matrix organization were the pathways with the most activation, while phase 1 functionalization of compounds, DHCR-24 signaling, and complement cascade were among the most suppressed pathways.

HOXA9, a member of the homeobox transcription factor family, plays an essential role in development, hematopoiesis, and endothelial regulation⁴⁵. Its role in cancer biology is complex and context-dependent, functioning as either a tumor suppressor or oncogene depending on the tumor type and stage of progression⁴⁶. Our analysis suggests that in CCA, HOXA9 serves as a central regulatory node influencing multiple mediators; including EDNRA⁴⁷, KIT⁴⁸, FLT3⁴⁹, CD36⁵⁰, E-selectin⁵¹, ERG⁵², among others; known to be involved in cancer metastasis (Figure 4A). These findings are consistent with studies linking HOXA9 to metastatic progression in hepatocellular carcinoma⁵³ and prostate cancer⁵⁴, further supporting the HOXA9 pathway as a promising therapeutic target across multiple cancer types, including CCA.

A volcano plot summarizing activated and suppressed pathways also identified S-100 family signaling, mitotic spindle assembly, collagen degradation, and extracellular matrix organization as the most activated pathways, while phase I functionalization, DHCR24 signaling, and complement cascade pathways were most suppressed (Figure 4B).

DISCUSSION

In this prospective pilot study, we report the feasibility and potential clinical utility of a cfDNA methylation-based score (TMS) for detecting and longitudinally monitoring cholangiocarcinoma. Across tissue and plasma samples, TMS distinguished patients with CCA from controls, decreased following curative resection, and dynamically reflected changes in tumor burden during systemic therapy. Importantly, longitudinal changes in TMS outperformed the current standard biomarker, CA 19-9, in identifying disease progression and stability.

At baseline, TMS values were significantly higher in both tissue and plasma samples from patients with CCA compared with controls, supporting the ability of this methylation panel to capture cancer-associated epigenetic signals. While the overall diagnostic performance for single time-point detection was moderate, with an AUC of 0.823, this finding should be interpreted in the context of the study’s primary objective. This assay was not designed for screening or early detection in asymptomatic populations, but rather for monitoring disease burden in patients with established malignancy. In this setting, a biomarker that provides consistent longitudinal trends may be more clinically informative than one optimized for binary classification.

Notably, CA 19-9 failed to differentiate cancer from control samples in this cohort and showed substantial interpatient variability, consistent with prior literature highlighting its limited sensitivity, specificity, and dependence on tumor biology and biliary inflammation¹⁵. The lack of correlation between CA 19-9 and TMS suggests that these markers may reflect distinct biological processes and supports the potential value of methylation-based cfDNA quantification as an alternative or complementary approach.

A key strength of this study lies in the longitudinal analysis of TMS dynamics. Following curative-intent surgical resection, TMS values decreased rapidly in all evaluated patients, suggesting that this assay can sensitively reflect definitive treatment. More importantly, changes in TMS (ΔTMS) distinguished periods of radiographic progression from non-progression during systemic therapy, with an AUC of 0.968. Although derived from a limited number of progression events, these findings support a strong biological association between cfDNA methylation burden and tumor dynamics.

In one illustrative case, a marked increase in TMS preceded radiographic progression by approximately two months. While anecdotal, this observation raises the possibility that cfDNA methylation changes may provide earlier insight into disease activity than imaging alone. Such lead-time advantages could be clinically meaningful, particularly in advanced disease where timely treatment modification is critical. However, these observations should be interpreted cautiously and viewed as hypothesis-generating pending validation in larger cohorts.

To provide biological context and external support for our findings, we performed tissue-based validation using TCGA and multiple independent cohorts. Aggregated beta methylation values across the same CpG panel (BMV) consistently distinguished tumor from non-tumor tissues, correlated inversely with survival and showed a trend toward association with tumor stage, consistent with findings in our own patient cohort. Importantly, BMV did not correlate with patient age, reinforcing that the observed methylation signal reflects cancer-specific rather than age-related epigenetic changes. The consistent performance of this panel across diverse cohorts and disease subtypes supports its biological relevance in cholangiocarcinoma.

Finally, pathway analysis of the genes associated with the amplicon panel identified the HOXA9 pathway as significantly activated in CCA samples from TCGA. HOXA9 and its downstream effectors, including EDNRA, KIT, FLT3, CD36, E-selectin, and ERG, may represent candidate pathways for future therapeutic investigation.

Limitations

This study has several important limitations that warrant careful consideration. First, the cohort size was small and the study was intended as a pilot feasibility analysis rather than a definitive evaluation of diagnostic or prognostic performance. Accordingly, performance metrics, including area under the curve estimates, should be interpreted cautiously and may not reflect real-world performance in larger or more heterogeneous populations. Future studies with expanded cohorts will be required to more precisely define test characteristics.

Second, the methylation panel used in this study was derived from a commercially available 537-amplicon pan-cancer assay and was not specifically optimized for cholangiocarcinoma. While this design enables broad detection of tumor-associated methylation signals, it may limit disease specificity and direct comparison with multi-cancer early detection assays designed for screening applications. Further refinement or disease-specific optimization of methylation targets may enhance performance in future studies.

Finally, our analysis was limited to plasma-derived cfDNA and did not evaluate TMS in other clinically relevant biospecimens, such as bile, ascitic fluid, or biliary brushings, which may provide complementary information in CCA. Larger prospective studies incorporating multiple biofluids, standardized sampling intervals, and extended follow-up will be necessary to fully define the clinical utility of ΔTMS in guiding treatment decisions and disease monitoring.

CONCLUSIONS

In this single-center prospective pilot study, quantitative assessment of cfDNA methylation using a Tumor Methylation Score demonstrated feasibility as a dynamic marker of tumor burden in patients with cholangiocarcinoma. TMS distinguished cancer from non-cancer samples, decreased following curative resection, and longitudinal changes closely reflected radiographic disease progression and stability during systemic therapy. In contrast, the commonly used biomarker CA 19-9 showed limited discriminatory value and did not correlate with disease dynamics.

While this study was not designed to establish screening or early detection performance, the observed associations between TMS dynamics and tumor burden support the biological relevance of cfDNA methylation as a quantitative monitoring tool in established disease. Tissue-based validation across independent cohorts further reinforces the cancer specificity of the selected methylation panel, including involvement of the HOXA9 pathway, and its association with adverse clinical outcomes.

Together, these findings suggest that cfDNA methylation-based monitoring may serve as a complementary adjunct to imaging and existing biomarkers in the longitudinal management of cholangiocarcinoma. Larger prospective studies with extended follow-up, inclusion of earlier disease stages, and standardized sampling strategies will be required to define the clinical utility, optimal thresholds, and role of TMS in routine patient care.

Data Sharing Statement

Deidentified patient data that support the findings of this study are available from the corresponding author upon reasonable request.