Serial circulating tumor DNA as a biomarker for monitoring and prognostication in patients with pancreatic cancer

Abstract

BACKGROUND

Despite emerging evidence supporting circulating tumor DNA (ctDNA) as a potential biomarker for pancreatic ductal adenocarcinoma (PDAC), its clinical efficacy for serial monitoring has not been thoroughly explored. We hypothesized that serial monitoring of ctDNA provides clinically relevant information on treatment response and prognosis in patients with PDAC.

AIM

To investigate the feasibility of using longitudinal ctDNA monitoring to predict treatment response and prognostic outcomes in patients with PDAC.

METHODS

This prospective observational study enrolled patients with PDAC confirmed histologically by biopsy. Blood samples were collected at baseline and during follow-up for ctDNA analysis using droplet digital polymerase chain reaction targeting KRAS G12/G13 mutations. Patients with PDAC were divided into resection and chemotherapy groups for further analysis. Radiologic responses were compared with changes in ctDNA (ΔctDNA) and carbohydrate antigen 19-9. Survival outcomes were analyzed based on the baseline ctDNA levels and clearance status.

RESULTS

Of the 200 enrolled patients, 168 were eligible for ctDNA detection rate analysis using the droplet digital polymerase chain reaction and 139 underwent serial ctDNA monitoring (34 resection group, 105 chemotherapy group). The overall ctDNA detection rate was 80.4%. Higher rates were observed in advanced disease stages (P = 0.004) and liver metastasis (P = 0.031). In the resection group, ΔctDNA did not differ significantly between the recurrence and non-recurrence groups. However, in the chemotherapy group, ΔctDNA showed significant differences among response groups (P < 0.001), demonstrating a moderate discriminative ability (area under the curve = 0.630), which was comparable to that of carbohydrate antigen 19-9 (area under the curve = 0.642). Preoperative and postoperative ctDNA detection was not associated with recurrence-free survival or overall survival (OS). However, among patients with metastatic disease, higher baseline ctDNA levels (> 0.418) were associated with shorter progression-free survival (4.7 months vs 7.5 months, P = 0.024) and OS (9.7 months vs 15.9 months, P = 0.039). In addition, ctDNA clearance at 8 weeks after chemotherapy was associated with improved OS (18.8 months vs 11.4 months, P = 0.031).

CONCLUSION

Serial ctDNA monitoring is a promising biomarker for treatment response and prognosis of PDAC, particularly in advanced disease.

Key Words: Circulating tumor DNA; Pancreatic ductal adenocarcinoma; Liquid biopsy; Serial monitoring; Treatment response; Prognosis

Core Tip: This study highlights the emerging role of circulating tumor DNA (ctDNA) as a clinically relevant biomarker in pancreatic ductal adenocarcinoma. ctDNA offers a minimally invasive approach for serial disease monitoring and may provide complementary information to conventional imaging and tumor markers for treatment assessment and prognostication. In the setting of limited effective biomarkers for this malignancy, ctDNA represents a promising tool for improving clinical management.

INTRODUCTION

Pancreatic ductal adenocarcinoma (PDAC) remains an aggressive malignancy with a poor prognosis, with an estimated five-year survival rate of approximately 13%[1]. The primary curative treatment for PDAC is surgical resection; however, even among patients who undergo resection, the recurrence rate remains high[2]. The majority of PDAC cases are diagnosed at advanced stages, necessitating systemic chemotherapy as a primary treatment option[3]. Despite advances in systemic chemotherapy, the median overall survival for metastatic pancreatic cancer (MPC) remains less than one year[4,5].

Currently, carbohydrate antigen 19-9 (CA19-9) is the most widely used biomarker in PDAC for disease monitoring and prognostication. However, its clinical utility is limited by insufficient sensitivity and specificity, as well as false-negative results in Lewis antigen-negative individuals[6]. Moreover, CA19-9 levels may not reliably reflect real-time tumor burden or early treatment response[7]. Therefore, there is a critical need to develop biomarkers that can accurately and rapidly predict treatment response and prognosis. Circulating tumor DNA (ctDNA) has recently emerged as a promising solution to address these significant gaps[8].

Current research on ctDNA in PDAC has yielded promising results, particularly in the contexts of minimal residual disease assessment, treatment monitoring, and prognostic prediction[9]. ctDNA has been identified as a biomarker capable of detecting minimal residual disease, allowing early intervention in patients at risk of recurrence[10,11]. Additionally, ctDNA monitoring may predict treatment responses, offering insight into the efficacy of chemotherapy regimens[12]. Furthermore, the prognostic value of ctDNA has been highlighted, with its presence in plasma correlating with poor outcomes[13-15]. Despite its potential, the clinical utility of serial ctDNA monitoring has not been thoroughly explored. In the present study, we comprehensively investigated the clinical role of serial ctDNA monitoring - including detection rate, response monitoring, and prognostic assessment - in patients with PDAC across all disease stages.

MATERIALS AND METHODS

Patient enrolment

This prospective observational study was conducted at Seoul National University Bundang Hospital and enrolled patients with PDAC between May 2019 and April 2022. Inclusion criteria were adults (≥ 18 years) with a confirmed PDAC diagnosis who provided written informed consent. PDAC was histologically confirmed by biopsy in all enrolled patients. Patients with an Eastern Cooperative Oncology Group performance status ≥ 2 were excluded.

Sample collection

Patients were categorized into resection and chemotherapy groups based on treatment approach. The resection group included patients undergoing upfront surgery or resection following neoadjuvant chemotherapy, whereas the chemotherapy group received palliative treatment only. Peripheral blood samples were collected at predefined time points. Baseline ctDNA sampling was performed within one week prior to surgery or initiation of chemotherapy. In the resection group, ctDNA was additionally measured within one week after surgery and serially monitored during adjuvant chemotherapy, with radiological assessments conducted every 2-3 months until recurrence (Figure 1A). For patients receiving neoadjuvant therapy, ctDNA sampling was performed before and after surgery, as well as during both neoadjuvant and adjuvant treatment phases (Figure 1B). In the chemotherapy group, ctDNA sampling was performed at each radiological assessment, conducted every 2-3 months, until disease progression (Figure 1C). Patients without detectable ctDNA at both baseline and first follow-up were excluded from serial monitoring. Serum CA19-9 levels were measured concurrently, when available, using an electrochemiluminescence immunoassay on an automated analyzer (cobas e801; Roche Diagnostics, Indianapolis, IN, United States) in the clinical laboratory of our institution.

Figure 1 Study timeline of patients in resection and chemotherapy groups, indicating treatment phases and assessment points. A: Study timeline for patients undergoing upfront surgery; B: Study timeline for patients undergoing resection after neoadjuvant chemotherapy; C: Study timeline for patients receiving palliative chemotherapy; D: Flow diagram showing patient enrolment and allocation in this study. ctDNA: Circulating tumor DNA; f/u: Follow up.

Extraction of ctDNA

Quantitative ctDNA analysis was performed using droplet digital polymerase chain reaction (ddPCR). Blood samples were collected and immediately processed to isolate plasma within 2 hours. ctDNA was extracted from plasma using a multiplex kit (QIAamp^® Circulating Nucleic Acid Kit, QIAGEN, Inc., Germantown, MD, United States). Specific primers targeting KRAS G12/G13 (G12A, G12C, G12D, G12R, G12S, G12V, and G13D) were used for amplification using the QX200™ Droplet Digital™ PCR System (Bio-Rad, Inc., Hercules, CA, United States). The quantification data obtained from ddPCR were analyzed using the QuantaSoft software version 1.7.4.0917 (Bio-Rad, Inc., Hercules, CA, United States). To ensure accuracy and reliability, appropriate quality control was implemented throughout ctDNA analysis. Only data from ddPCR runs generating more than 10000 droplets were included in the analysis to ensure data reliability. To minimize inter-assay variability, a negative control was included in all experiments, and assay reproducibility was verified every three months by testing the same reference sample to confirm consistency of the results. In addition, the performance of the ddPCR droplet reader was evaluated and instrument calibration was performed annually using Droplet Digital Spectral Calibration. The results are reported as the variant allele fraction (VAF).

Outcome measurements

Detection performance was evaluated in patients with ≥ 2 blood samples, including baseline. ctDNA positivity was defined as detection in ≥ 1 sample. Detection rates were compared by disease stage and metastatic patterns. Additionally, to assess the sensitivity of ctDNA detection, a comparative analysis of detection rate was performed using pre-existing whole exome sequencing (WES). WES data were generated using pancreatic cancer tissue samples, and sequenced on the Illumina NovaSeq 6000 platform (Theragen Bio, Seongnam, South Korea) with an average depth of 200 ×.

The predictive values of ctDNA and CA19-9 for treatment response were evaluated in patients with sequential follow-up. The difference in ctDNA and CA19-9 levels between two consecutive time points was calculated and denoted as ΔctDNA and ΔCA19-9, respectively. Infinite values were replaced with group maximum values; zero-to-zero changes were set to -100.

In the resection group, patients were monitored for recurrence after resection, and statistical analyses were performed to compare the ΔctDNA and ΔCA19-9 levels between patients with and without recurrence. In the chemotherapy group, ΔctDNA and ΔCA19-9 levels were compared with radiological response according to the Response Evaluation Criteria in Solid Tumors version 1.1.

Correlation analysis was performed to evaluate synchronized and desynchronized biomarker dynamics between ΔctDNA and ΔCA19-9 at each sampling time point relative to the previous time point. Synchronized patterns were defined as concordant changes in both biomarkers, in which ΔctDNA and ΔCA19-9 either increased or decreased in the same direction. In contrast, desynchronized patterns were defined as discordant changes, including cases where ΔctDNA levels increased while ΔCA19-9 levels decreased, or vice versa. Furthermore, the efficacy of ΔctDNA for monitoring treatment response was assessed in patients with consistently normal CA19-9 levels. To assess the potential of ΔctDNA for the early detection of disease progression, a specific cohort from the chemotherapy group was selected according to predefined criteria: Patients exhibiting radiologically stable disease (SD) with no increase in CA19-9, but with increasing ΔctDNA.

In the resection group, the preoperative and postoperative ctDNA levels were analyzed to compare recurrence-free survival (RFS) and overall survival (OS) based on the presence of ctDNA. In the chemotherapy group, baseline ctDNA levels before chemotherapy were used as prognostic biomarkers. Patients were stratified into two groups based on the median baseline ctDNA values. Additionally, ctDNA clearance at multiple time points, specifically at 8 weeks and 2 weeks, was evaluated. Progression-free survival (PFS) and OS were compared between the stratified groups. Multivariable regression analysis was conducted to adjust for potential confounders, including sex, age, stage, chemotherapy regimen, and second-line chemotherapy administration as covariates. RFS was defined as the time from surgical resection to documented disease recurrence, PFS as the time from treatment initiation to radiologic disease progression, and OS as the time from diagnosis to death from any cause.

Statistical analysis

The study size was determined by the number of eligible patients diagnosed and enrolled during the study period. χ² tests were used to analyze ctDNA detection patterns; Mann-Whitney U and Kruskal-Wallis tests compared biomarker distributions. Receiver operating characteristic (ROC) curve analysis evaluated predictive performance. Survival analyses were conducted using the Kaplan-Meier method and log-rank tests. Multivariable analyses were performed using logistic and Cox regression models, adjusting for sex, age, disease stage, and treatment variables. Missing data were minimal and were not imputed. Statistical significance was defined as P < 0.05. All analyses were performed using R version 4.3.1.

RESULTS

Baseline characteristics of patients

A total of 200 patients were enrolled, including 58 in the resection group and 142 in the chemotherapy group. After exclusions, 139 patients underwent serial ctDNA testing: 34 in the resection group and 105 in the chemotherapy group (Figure 1D). The resection group had a median age of 62 years (52.9% female), with 38.2% having resectable pancreatic cancer (RPC) and 61.8% having borderline resectable or locally advanced pancreatic cancer (BRPC/LAPC). The chemotherapy group had a median age of 66 years (36.2% female), including 32.4% with BRPC/LAPC and 67.6% with MPC (Table 1). The median follow-up duration was 14.9 months [interquartile range (IQR): 9.8-21.9]. Missing data were minimal (< 5%) for all key variables.

Table 1 Baseline characteristics of the patients and disease status, n (%)/median (range).

Variable	Resection group, n = 34	Chemotherapy group, n = 105	All patients, n = 139
Age	62 (40-83)	66 (37-82)	65 (37-83)
Sex
Male	16 (47.1)	67 (63.8)	83 (59.7)
Female	18 (52.9)	38 (36.2)	56 (40.3)
Tumor location
Head	19 (55.9)	38 (36.2)	57 (41.0)
Body	9 (26.5)	29 (27.6)	38 (27.3)
Tale	6 (17.6)	25 (23.8)	31 (22.3)
Unknown or multiple	0 (0.0)	13 (12.4)	13 (9.4)
Stage
RPC	13 (38.2)	0 (0.0)	13 (9.4)
BRPC/LAPC	21 (61.8)	34 (32.4)	55 (39.5)
MPC	0 (0.0)	71 (67.6)	71 (51.1)
Metastasis location
Liver	0 (0.0)	56 (53.3)	54 (38.8)
Lung	0 (0.0)	14 (13.3)	14 (10.1)
Peritoneum	0 (0.0)	19 (18.1)	19 (13.7)
Other organs	0 (0.0)	13 (12.4)	13 (9.4)
CA19-9
> 37 U/mL	25 (73.5)	87 (82.9)	112 (80.6)
< 37 U/mL	9 (26.5)	18 (17.1)	27 (19.4)

RPC: Resectable pancreatic cancer; BRPC: Borderline resectable pancreatic cancer; LAPC: Locally advanced pancreatic cancer; MPC: Metastatic pancreatic cancer; CA19-9: Carbohydrate antigen 19-9.

Detection rate of ddPCR

ctDNA detection performance using ddPCR was investigated in 168 patients. ctDNA was detected in 135 of the 168 patients (80.4%) (Figure 2). The analysis of ctDNA demonstrated significant differences across disease stages in both detection rates [RPC: 58.3% (14/24) vs LAPC: 78.1% (50/64) vs MPC: 88.8% (71/80), P = 0.004] and VAF values [RPC: 0.09 (IQR: 0-0.41) vs LAPC: 0.17 (IQR: 0.06-0.49) vs MPC: 0.80 (IQR: 0.18-5.17), P < 0.001] (Figure 3A). Additionally, patients with liver metastases showed both a higher detection rate [94.6% (53/56) vs 76.5% (18/24), P = 0.031] and higher VAF values [2.80 (IQR: 0.36-7.14) vs 0.37 (IQR: 0.06-0.36), P < 0.001] than those without liver metastases. Conversely, neither ctDNA detection rates nor VAF values were significantly different between patients with and without lung metastases [87.5% (14/16) vs 89.1% (57/64), P = 0.791; 1.13 (IQR: 0.12-2.41) vs 0.71 (IQR: 0.21-5.40), P = 0.639] or peritoneal metastases [86.4% (19/22) vs 89.7% (52/58), P = 0.984; 0.30 (IQR: 0.13-1.20) vs 1.81 (IQR: 0.23-6.56), P = 0.062, respectively] (Figure 3B). Multivariable analysis confirmed higher ctDNA levels in patients with liver metastases, demonstrating both increased detection rates [odds ratio (OR) = 9.63, 95% confidence interval (CI): 1.58-58.6, P = 0.014] and higher VAF values (β = 5.92, 95%CI: 1.47-10.37, P = 0.010), while other metastatic sites showed no significant associations (lung: OR = 1.81, 95%CI: 0.27-11.86, P = 0.538, β = 0.43, 95%CI: -5.04 to 4.19, P = 0.854; peritoneum: OR = 2.12, 95%CI: 0.36-12.46, P = 0.407, β = 1.22, 95%CI: -3.25 to 5.68, P = 0.589) (Figure 3C).

Figure 2 Analysis of circulating tumor DNA detection in pancreatic cancer patients using droplet digital polymerase chain reaction. The empty circles indicate undetected circulating tumor DNA, whilst the filled circles represent the detection of circulating tumor DNA. Each color reflects the stages of pancreatic cancer. ctDNA: Circulating tumor DNA; VAF: Variant allele fraction; RPC: Resectable pancreatic cancer; LAPC: Locally advanced pancreatic cancer; BRPC: Borderline resectable pancreatic cancer; MPC: Metastatic pancreatic cancer.

Figure 3 Circulating tumor DNA detection rates and variant allele fraction according to cancer stage and metastatic sites. A: Circulating tumor DNA (ctDNA) detection rates and variant allele fraction (VAF) varied significantly by cancer stage; B: Analysis of ctDNA detection and VAF in metastatic sites showed a significantly higher detection rate in patients with liver metastasis compared to those without. No significant differences were observed in detection rates and VAF for patients with vs without lung metastasis and peritoneal metastasis; C: A multivariable analysis confirmed the persistence of higher ctDNA detection rates and VAF in the liver metastasis group. ^aP < 0.05. RPC: Resectable pancreatic cancer; LAPC: Locally advanced pancreatic cancer; MPC: Metastatic pancreatic cancer; ctDNA: Circulating tumor DNA; VAF: Variant allele fraction; OR: Odds ratio; CI: Confidence interval.

WES and ctDNA analyses were performed on 69 patients. Using WES as the reference standard for the presence of KRAS mutations, ddPCR achieved a ctDNA detection sensitivity of 91.7%, a specificity of 33.3%, and an accuracy of 84.1% (Supplementary Table 1). Concordant positive results were observed in 79.7% (55/69) of the patients, whilst concordant negative results were observed in 4.3% (3/69). Five patients had positive WES results but negative ddPCR results: One with RPC, three with BRPC/LAPC, and one with MPC. Six patients had WES-negative but ddPCR-positive results; three had LAPC and three had MPC.

Treatment response monitoring

In the resection group, ctDNA and radiological responses were monitored for 108 events, and CA19-9 and radiological responses were monitored for 166 events. ΔctDNA values showed no significant difference between the recurrence and non-recurrence groups [median ΔctDNA in the recurrence group: -8.96 (IQR: -100 to 1290.79) vs non-recurrence: -68.63 (IQR: -100 to -186.51), P = 0.754]. Conversely, ΔCA19-9 values differed significantly between the groups [median ΔCA19-9 in the recurrence group: 108.22 (IQR: -100 to 900.00) vs non-recurrence: -100 (IQR: -100 to -62.96), P = 0.031]. ROC analysis yielded area under the curve values of 0.528 for ΔctDNA and 0.671 for ΔCA19-9, without a significant difference (P = 0.339) (Figure 4A).

Figure 4 Comparison of changes in circulating tumor DNA and carbohydrate antigen 19-9 for predicting recurrence and treatment response. A: Comparison of changes in circulating tumor DNA (ΔctDNA) and carbohydrate antigen 19-9 (ΔCA19-9) in the resection group. Box plots demonstrated the distribution of ΔctDNA and ΔCA19-9 values between recurrence and non-recurrence groups. The diagnostic performance of ΔctDNA and ΔCA19-9 for predicting tumor recurrence was compared using receiver operating characteristic curves; B: Comparison of ΔctDNA and ΔCA19-9 in the chemotherapy group. Box plots demonstrated the distribution of ΔctDNA and ΔCA19-9 values across different radiological response groups (partial response, stable disease, progressive disease). The diagnostic performance of ΔctDNA and ΔCA19-9 for predicting treatment response was compared using receiver operating characteristic curves. ctDNA: Circulating tumor DNA; CA19-9: Carbohydrate antigen 19-9; AUC: Area under the curve; PR: Partial response; SD: Stable disease; PD: Progressive disease.

During the monitoring of the resection group, recurrence was observed in 11 of 108 cases, of which six occurred without ΔctDNA elevation. Further analysis of these cases revealed that two patients had localized recurrence, two had peritoneal seeding, one had lung metastasis, and one patient had hepatic metastasis. Elevated ΔctDNA levels were observed in 35 events without any clinical evidence of recurrence. Of these, nine were detected in the immediate postoperative period, and four occurred during adjuvant chemotherapy.

In the chemotherapy group, 273 ΔctDNA and 334 ΔCA19-9 events were evaluated. ΔctDNA values showed significant differences across responses [median ΔctDNA in the partial response (PR): -88.66 (IQR: -100 to 100.55) vs SD: -57.75 (IQR: -100 to 384.58) vs progressive disease (PD): 257.96 (IQR: -20.73 to 2219.89), P < 0.001]. Post-hoc tests revealed significant differences between PR vs PD (P < 0.001) and SD vs PD (P < 0.001). Similarly, ΔCA19-9 values differed significantly [median ΔCA19-9 in the PR: -86.36 (IQR: -100 to -18.52) vs SD: -58.25 (IQR: -100 to -12.69) vs PD: 50.1 (IQR: -58.54 to 229.57), P < 0.001]. Significant differences were also found between PR vs PD and SD vs PD in post-hoc tests (P < 0.001). ROC analysis for predicting PD using ctDNA and CA19-9 showed area under the curves of 0.630 and 0.642, respectively, without a significant performance difference (P = 0.820) (Figure 4B).

Among 273 ΔctDNA events, PD occurred in 85 (31.1%). Of these, PD occurred in 74.1% (63/85) with increased ΔctDNA levels, whereas 25.9% (22/85) showed PD despite decreased ΔctDNA levels. In a detailed analysis of these 22 cases, five PD events were identified during the first follow-up evaluation during chemotherapy. The ΔctDNA values at this time point tended to be lower than the relatively high baseline ctDNA values in the chemotherapy-naïve state. In four additional PD events, prior assessments had already shown increasing ctDNA levels, suggesting the potential for early prediction of PD. An examination of 63 events with increased ctDNA levels but no concurrent PD was also conducted. In 10 events, the tumor size increased despite not meeting the PD criteria by Response Evaluation Criteria in Solid Tumors. Furthermore, 14 events showed increased ctDNA levels preceding a later confirmed PD, indicating the possibility of early detection of PD.

In the resection group, 108 events with concurrent ΔctDNA and ΔCA19-9 levels were identified. Among them, 61.1% (66/108) exhibited synchrony, whereas 38.9% (42/108) displayed desynchronization. Recurrence occurred in 10.2% (11/108) of cases, with 72.7% (8/11) predicted by either ΔctDNA or ΔCA19-9. In the chemotherapy group, ΔctDNA and ΔCA19-9 levels were measured simultaneously in 273 events. Among them, 61.9% (169/273) exhibited synchrony and 38.1% (104/273) exhibited desynchronization. PD events were observed in 31.1% (85/273), with 88.2% (75/85) predicted by either ΔctDNA or ΔCA19-9 (Figure 5A).

Figure 5 Synchronized and desynchronized changes in circulating tumor DNA and carbohydrate antigen 19-9 and early detection of disease progression. A: Synchronized and desynchronized changes in circulating tumor DNA (ΔctDNA) and carbohydrate antigen 19-9 (ΔCA19-9) levels in the resection and chemotherapy groups. In the resection group, 108 events with concurrent ΔctDNA and ΔCA19-9 measurements were identified, of which 61.1% exhibited synchronized changes and 38.9% exhibited desynchronized changes. Tumor recurrence occurred in 10.2% of events, and 72.7% of recurrent cases were predicted by either ΔctDNA or ΔCA19-9. In the chemotherapy group, ΔctDNA and ΔCA19-9 levels were simultaneously measured in 273 events, with 61.9% showing synchronized changes and 38.1% showing desynchronized changes. Progressive disease (PD) was observed in 31.1% of events, and 88.2% of PD cases were predicted by either biomarker; B: Early detection of PD by ctDNA monitoring. Time to radiological progression was visualized using a swimmer plot in stable disease patients who exhibited increased ΔctDNA levels without corresponding ΔCA19-9 elevation. Each bar represents an individual patient, with the length indicating the follow-up duration. ctDNA: Circulating tumor DNA; CA19-9: Carbohydrate antigen 19-9; PR: Partial response; SD: Stable disease; PD: Progressive disease.

Throughout the follow-up period, 19 patients (13.9%) maintained CA19-9 levels within the normal range, 10 of whom experienced PD events. In seven of these 10 cases, ΔctDNA levels increased at the time of PD. Furthermore, in the remaining nine patients who did not experience PD, the treatment response was continuously monitored through stable ctDNA levels (Supplementary Figure 1). To determine whether ctDNA could predict disease progression prior to changes in radiological responses, an analysis was undertaken. We selected 26 patients in the chemotherapy group who showed radiological SD with an increase in ΔctDNA but no increase in ΔCA19-9. Among them, PD was confirmed by radiological evaluation in 18 patients (69.2%) after a median of 2.6 months (range: 1.2-4.8 months) (Figure 5B).

Prognosis prediction

RFS and OS were analyzed based on preoperative ctDNA positivity in the 23 patients who underwent resection after neoadjuvant chemotherapy. Although the group with preoperatively detected ctDNA had a higher recurrence rate than that of the group with undetected ctDNA, this difference was not statistically significant (undetected ctDNA vs detected ctDNA group: Not reached vs ctDNA-detected group: 20.3 months, P = 0.110) (Figure 6A). There was no difference in OS between the two groups (ctDNA-undetected group: Not reached vs ctDNA-detected group: Not reached; P = 0.450) (Figure 6B). In the survival analysis based on postoperative ctDNA detection, we analyzed 32 patients from the combined cohort of upfront resection and post-chemotherapy resection groups. The analysis revealed no significant differences in either RFS or OS (RFS, ctDNA-undetected group: 21.6 months vs ctDNA-detected group: 19.4 months, P = 0.970; OS, ctDNA-undetected group: Not reached vs ctDNA-detected group: Not reached, P = 0.220) (Figure 6C and D).

Figure 6 Survival outcomes according to preoperative and postoperative circulating tumor DNA status after resection. A and B: Recurrence-free survival (A) and overall survival (B) curves comparing patients with detectable vs undetectable preoperative circulating tumor DNA levels among those who underwent resection after neoadjuvant chemotherapy (n = 23); C and D: Recurrence-free survival (C) and overall survival (D) curves comparing patients with detectable vs undetectable postoperative circulating tumor DNA levels in the combined cohort of upfront resection and post-chemotherapy resection groups (n = 32). ctDNA: Circulating tumor DNA.

When analyzing the prognosis based on baseline ctDNA prior to chemotherapy, we divided the patients into localized and metastatic stages. In the localized stage group, the median baseline ctDNA value was 0.502, and patients were divided into lower and higher groups based on this. There was no statistically significant difference in PFS and OS between the lower and higher groups (PFS: 9.3 months vs 6.1 months, P = 0.200; OS: 17.6 months vs 13.4 months, P = 0.810) (Figure 7A and B). A multivariable analysis showed no significant association between baseline ctDNA and PFS or OS, with hazard ratios (HRs) of 1.43 (95%CI: 0.66-3.08, P = 0.364) for PFS and 1.06 (95%CI: 0.49-2.31, P = 0.884) for OS (Supplementary Table 2). Conversely, in the metastatic stage group, patients with baseline ctDNA below the median ctDNA of 0.418 had significantly longer median PFS (7.5 months vs 4.7 months, P = 0.024) and OS (15.9 months vs 9.7 months, P = 0.039) (Figure 7C and D). A multivariable analysis adjusting for sex, age, disease stage, and type of chemotherapy regimen demonstrated a significant association between the baseline ctDNA level and OS (HR = 1.75, 95%CI: 1.02-3.01, P = 0.044), but not between the baseline ctDNA level with PFS (HR = 1.57, 95%CI: 0.89-2.76, P = 0.119) (Supplementary Table 3).

Figure 7 Survival outcomes according to baseline circulating tumor DNA levels and circulating tumor DNA clearance after chemotherapy. A and B: Survival analysis in localized stage patients stratified by median baseline circulating tumor DNA (ctDNA) level. No significant differences were observed in either median progression-free survival (PFS) or median overall survival between the two groups; C and D: Survival outcomes in metastatic stage patients stratified by median baseline ctDNA level, patients with ctDNA levels below the median showed significantly longer median PFS and compared to those above the median; E and F: Survival analysis according to ctDNA clearance status at 8 weeks after chemotherapy initiation. While PFS showed no significant difference between groups, median overall survival was significantly longer in the clearance group compared to the non-clearance group. ctDNA: Circulating tumor DNA; VAF: Variant allele fraction.

Survival analysis based on ctDNA clearance status at eight weeks post-chemotherapy involved 23 patients with clearance and 74 without clearance. There was no significant difference in PFS between the groups (clearance group: 7.5 months vs non-clearance group: 6.1 months, P = 0.110). However, a significant difference in OS was observed (clearance group: 18.8 months vs non-clearance group: 11.4 months, P = 0.031) (Figure 7E and F). A multivariable analysis adjusting for sex, age, disease stage, and type of chemotherapy regimen revealed that ctDNA clearance at 8 weeks was a significant predictor for both PFS (HR = 0.56; 95%CI: 0.32-1.00; P = 0.048) and OS (HR = 0.59; 95%CI: 0.36-0.99; P = 0.048) (Supplementary Table 4). Additionally, we analyzed the effect of ctDNA clearance at week 2 on the prognosis of 52 patients. Nine patients achieved ctDNA clearance, whereas 43 did not. There were no significant differences in PFS or OS (PFS: Clearance group, 8.2 months vs non-clearance group, 6.1 months, P = 0.440; OS: Clearance group, 12.1 months vs non-clearance group, 13.4 months, P = 0.800) (Supplementary Figure 2).

DISCUSSION

This large-scale study evaluated the clinical efficacy of ctDNA for monitoring the treatment response and predicting the prognosis of patients with PDAC. We demonstrated that ctDNA can be used complementarily with CA19-9 in treatment response monitoring, and provides additional value, including treatment response monitoring in the CA19-9 normal group and the ability to predict early disease progression in some subgroup. Moreover, baseline ctDNA levels and longitudinal changes were closely associated with survival outcomes, suggesting ctDNA may have a role in the prognostic stratification.

The ctDNA detection rate by ddPCR was 80.4% (135/168), which was relatively high compared to previous studies, which reported rates ranging from 12.8% to 88.0%[16-18]. Furthermore, ctDNA detection by ddPCR showed sufficient performance compared to tissue-based WES. In this study, ctDNA-positive patients were defined as those with at least one positive result in either the baseline or first follow-up test, which may explain the relatively high detection rate. The low sensitivity of ctDNA has been a long-standing issue, especially for patients with PDAC, who typically exhibit lower ctDNA shedding rates compared to patients with other cancers, such as breast and colorectal cancers[19]. Given ctDNA’s greater potential for disease monitoring, it is designed to track changes over time rather than rely on a single result, making the double-checking approach a reasonable strategy.

The ctDNA detection rate for patients with MPC was significantly higher than that for patients with localized disease. Furthermore, the likelihood of ctDNA detection in patients with liver metastasis was higher than that in patients without metastasis. Consistent with these findings, previous studies have demonstrated that the tumor burden is positively correlated with ctDNA levels, whereas early-stage disease is often associated with reduced detectability and an increased risk of false-negative results[20-22]. Additionally, organ-specific vascularity and spatial constraints influence macromolecule access to systemic circulation, thereby affecting detection rates[23-25]. Further research is needed, but this quantitative relationship among ctDNA, tumor burden, and metastasis patterns may be used to predict the presence of potential metastatic lesions in the future.

The ability of ctDNA to monitor treatment response in this study was comparable to that of CA19-9, but was not sufficient for optimal clinical decision-making. Small absolute ctDNA values, particularly in the resection group, may have increased the likelihood of measurement errors, potentially leading to an overemphasis on percentage changes[26]. Furthermore, clonal hematopoietic variants, which are also found in healthy individuals, may have further contributed to erroneous results[27]. Moreover, ctDNA shedding can decline due to chemotherapy effect, potentially leading to false-negative results even during disease progression. For instance, in some cases, ctDNA levels were lower at the first follow-up after chemotherapy despite tumor growth, compared to levels in the chemotherapy-naïve state. Conversely, ctDNA shedding may be absent in patients with extensive chemotherapy exposure, resulting in consistently undetectable ctDNA levels. A deeper understanding of ctDNA characteristics, along with the development of more refined testing and interpretation methods, will enhance clinicians’ ability to interpret this biomarker in clinical practice, ultimately improving its clinical utility.

Desynchrony between ctDNA and CA19-9 was observed in 38.9% (42/108) of the resection group and 38.8% (106/273) of the chemotherapy group, confirming the independence of these two biomarkers, as reported in previous studies[28-30]. Notably, among 19 patients with normal CA19-9 levels, ctDNA predicted PD in 70.0% (7/10), highlighting the value of combining the two biomarkers to significantly enhance sensitivity of predicting disease progression. More importantly, ctDNA offers the advantage of the early detection of disease progression. In cases with discrepancies between ctDNA and CA19-9, isolated increases in ctDNA were associated with an average of 2.6 months earlier detection of disease progression, consistent with previous studies indicating that ctDNA can detect recurrence or occult metastasis in PDAC over three months earlier than imaging[2,9]. Therefore, ctDNA may serve as a valuable alternative or complementary biomarker for CA19-9 in clinical practice.

Survival analysis showed that in the metastatic group, patients with higher baseline ctDNA levels or those who did not achieve ctDNA clearance by week 8 exhibited significantly reduced survival. However, in the resection and localized groups, the survival difference based on ctDNA level or status was not definitive. Previous studies have also reported inconsistent findings regarding the prognostic value of postoperative ctDNA, with some indicating an association with poor prognosis[31-33], while others did not[29]. Additionally, even in the metastatic group, ctDNA measured at the 2-week point did not show significant survival differences based on ctDNA clearance status. This may be due to a transient increase in ctDNA levels caused by apoptosis following chemotherapy, with blood samples collected before ctDNA levels plateaued to accurately reflect tumor burden. Moreover, inter-patient heterogeneity - such as differences in sex, weight, smoking status, and heart disease - as well as intra-patient variations caused by physical activity, inflammation, and diurnal fluctuations can influence ctDNA levels and complicate clinical interpretation[34-36]. Given the limited understanding of ctDNA dynamics and heterogeneity, the optimal timing of ctDNA monitoring after treatment remains a subject of ongoing debate[37-41]. Further research is required to establish the optimal timing of ctDNA assessment.

This study has several limitations. First, the sample size of the resection group was relatively small, which was unavoidable given the inclusion of patients with PDAC across various disease stages, and the study was conducted at a single institution. In addition, patients without detectable ctDNA at baseline and first follow-up were excluded from serial monitoring, which may have introduced selection bias by preferentially including patients with higher tumor burden or increased ctDNA shedding. Finally, we used ddPCR rather than next-generation sequencing, which recently gained recognition for accuracy and broader genetic mutation detection capabilities[13,42]. The KRAS-focused ddPCR approach may not detect tumor-derived DNA in wild-type KRAS cases, and its sensitivity was lower than that of WES. Additionally, the lack of matched leukocyte sequencing precluded formal exclusion of false-positive variants that resulted from clonal hematopoiesis, which should be addressed in future studies using broader mutation panels and paired white blood cell sequencing.

CONCLUSION

In this prospective study, serial ctDNA assessments using the ddPCR demonstrated clinical utility for monitoring treatment responses and predicting the prognosis of patients with PDAC. ctDNA demonstrated comparable performance to CA19-9 in monitoring treatment response, and baseline ctDNA positivity and longitudinal ctDNA changes were significantly associated with survival outcomes, particularly those of advanced-stage disease. These findings collectively support ctDNA as a promising complementary biomarker to CA19-9, potentially enhancing the accuracy of biomarker-based clinical decision making. Additionally, the favorable cost-effectiveness and feasibility of ddPCR-based ctDNA testing compared with next-generation sequencing-based approaches in pancreatic cancer support its potential use for serial monitoring and consideration for future integration into clinical guidelines, although further prospective validation and economic evaluation are required.

ACKNOWLEDGEMENTS

We would like to thank Juhee Kim for administrative support.