KRAS Mutation Allele Frequency Dynamics in Plasma Extracellular Vesicles: Association with Survival in Localized Pancreatic Adenocarcinoma

Abstract

Background:

Pancreatic ductal adenocarcinoma (PDAC) lacks consistent biomarkers to monitor treatment response and predict survival. Metabolically active extracellular vesicles (EVs) carrying tumor-specific KRAS mutations offer promise as disease-specific biomarkers.

Methods:

Informed by genomic profiling of tumor tissue, plasma samples were prospectively collected from 44 patients, with confirmed KRAS mutated PDAC, undergoing neoadjuvant therapy (NAT) followed by surgery between 2019–2021. Samples were obtained at diagnosis, post-NAT, and 1-month post-surgery. EVs were isolated using lipid nanoprobe technology, and EV-associated KRAS mutations were detected using droplet-digital PCR. Patients were grouped based on temporal changes in EV-associated KRAS mutation allele frequency (MAF): ND (no KRAS detected), DD (decreasing MAF), and ID (increasing MAF).

Results:

Among 44 patients, 29 (65.9%) were ND, 8 (18.2%) DD, and 7 (15.9%) ID. Detectable EV-associated KRAS MAF was found in 21%, 30%, and 50% of patients with stages I, II, and III PDAC. No significant differences were noted in demographic or clinical variables (p>0.05). The ND group had the longest restricted mean disease-free survival (rmDFS: 31.2months), followed by DD (27.8months) and ID (9.8months; p=0.010). Similarly, restricted mean overall survival (rmOS) was longest in the ND (40.3months), followed by DD (35.7months) and ID (17.7months; p=0.012). On multivariable analysis, increasing EV-KRAS MAF (ID group) independently predicted inferior rmDFS (HR:6.14; p=0.001) and rmOS (HR:6.95; p=0.002).

Conclusion:

Temporal increase of EV-KRAS MAF is a significant predictor of reduced DFS and OS in PDAC. Integrating EV-KRAS mutation allele frequency dynamics analysis with current biomarkers like CA19–9 could improve treatment monitoring and survival prognostication.

INTRODUCTION

Pancreatic ductal adenocarcinoma (PDAC) is a highly aggressive malignancy with a 5-year survival rate of 12%.[1] Early recurrence—within 6 months after curative resection—occurs in up to one-third of patients, raising concerns about undetected micro-metastasis across all stages.[2,3] In response, neoadjuvant therapy (NAT) is increasingly used to treat systemic disease early, downstaging tumors, and enhancing surgical eligibility and outcomes.[4–6]

Biomarkers are essential for evaluating response to NAT and predicting prognosis. Clinically used PDAC biomarkers primarily include CA19–9 and carcinoembryonic antigen (CEA). CA19–9, the most commonly utilized biomarker, is associated with improved survival when levels are low at diagnosis or normalize following NAT—except in non-secretors.[7–9] However, CA19–9 lacks specificity and can yield false results in benign conditions or in Lewis antigen-negative individuals.[10–12] Additionally, it is not tumor-specific.[13] Emerging biomarkers include circulating tumor cells (CTCs), circulating tumor DNA (ctDNA), and extracellular vesicles (EVs), investigated through tumor-informed and tumor-naive liquid biopsy approaches.[14] As a unique form of intercellular communication, EVs play a significant role in the natural history of cancer from initiation to metastasis, generating considerable interest in EVs compared to other tumor-specific biomarkers.[15] EVs contain large fragments of double-stranded DNA (dsDNA), as well as RNA, proteins, lipids, and metabolites, all of which are being evaluated both individually and in combination as potential biomarkers.[16–18] KRAS mutation plays a pivotal role as the principal driver and the most commonly mutated gene in PDAC, present in over 90% of cases.[19, 20] These mutations typically occur as single-point mutations at residues G12, G13, or Q61, with G12 mutations accounting for 98% of KRAS mutations.[21] At G12, eight different amino acid substitutions have been identified, with G12D, G12V, and G12R being the most common. Accumulating evidence suggests that changes in KRAS mutant allele frequency (MAF) may reflect treatment response. A decrease in circulating KRAS mutations during NAT is linked to better outcomes, while persistence or increase suggests resistance.[22] We hypothesized that both the presence and temporal changes in EV-KRAS MAF could serve as prognostic indicators of treatment response and survival in PDAC. This study evaluated these dynamics across NAT and surgery with the primary aim of assessing their association with disease-free survival (DFS) and overall survival (OS). We further investigated the prognostic value of integrating EV-KRAS MAF dynamics with CA19–9 responses in patients who underwent NAT and curative resection.

METHODS

The study was performed under an IRB-approved protocol (STUDY20020051) developed at the University of Pittsburgh Medical Center (UPMC) and Carnegie Mellon University. Patients with PDAC prospectively enrolled in a biorepository study (STUDY19060127) at UPMC between 2019 and 2021 were included in this study. All adult patients with tissue biopsy-confirmed PDAC and known KRAS mutation who received NAT (including chemotherapy alone or chemotherapy and radiotherapy) followed by curative intent surgery were included. Patients who died within 90 days of surgery (due to surgical complications) or lacked plasma samples at three predefined time points were excluded from the analysis. Patients enrolled in the protocol after 2021 were not included in this analysis to allow for a 3-year follow-up period after potentially curative resection.

Plasma samples were prospectively collected from patients at three time points during the treatment course: (1) at diagnosis, prior to receiving any cancer therapy; (2) on the day of surgery, marking the completion of NAT but before surgical resection; and (3) one month after curative resection. These plasma samples were analyzed to determine EV-KRAS MAF, as described below. Patients were categorized based on longitudinal changes in these samples into three groups:

1. No Detection (ND): Patients with no detectable EV-KRAS MAF at any of the three time points.

2. Decreasing Detection (DD): Patients with detectable EV-KRAS MAF, either at diagnosis or after completion of NAT, that decreased to non-detectable levels after curative resection.

3. Increasing Detection (ID): Patients with detectable EV-KRAS MAF that exhibited increasing levels after completing NAT or curative resection.

The ID and DD cohorts were heterogeneous because of the various possible permutations at the three time points. For patients with fluctuating detection, such as those initially undetectable at diagnosis, detectable at the end of NAT, and non-detectable one-month post-surgery—categorization was based on the final time point, i.e., one-month post-surgery. The one-month postoperative timepoint was selected as a critical juncture because, at this time, the patient has completed neoadjuvant therapy, aimed at eradicating micrometastatic or distant disease, and has undergone surgical resection of the primary tumor. At this stage, residual tumor burden is theoretically absent, making it a key moment for evaluating the presence of any remaining occult disease. A complete logic table showing the various scenarios of temporal changes in KRAS MAF and their classification into three defined groups is provided in Supplemental Table 1.

Data Collection

Electronic medical records were reviewed to collect demographic data [age, sex, Charlson Comorbidity Index (CCI) [23], and ECOG performance status], disease [tumor size by CT and EUS, CA19–9 secretor status, baseline CA19–9 level, tumor resectability per NCCN criteria [24], KRAS mutation type, and CAP response score [25]], and treatment characteristics [neoadjuvant chemotherapy (NAC), number of NAC cycles, receipt of radiotherapy, surgical modality, adjuvant therapy, number of adjuvant, and total treatment cycles]. Neoadjuvant therapy (NAT) included all treatments before curative surgery, and adjuvant therapy was defined as systemic treatments initiated after curative resection, with a planned duration based on standard regimens (e.g., 6 months for gemcitabine-based and oxaliplatin-based protocols). Therapy may have been discontinued earlier due to recurrence or adverse effects. Clinical records were also queried for biochemical (post-NAT CA19–9, postoperative CA19–9, and CA19–9 change), radiologic (post-NAT CT tumor size), and pathologic variables [pathological tumor size, lymph node (LN) positivity and ratio, tumor grade, lymphovascular invasion, perineural invasion, margin status (positive if <1 mm from invasive carcinoma), and AJCC 8th edition stage [26]]. Optimal CA19–9 response was defined as >50% reduction and normalization (<37 U/mL) in secretors [9]. Survival metrics included overall survival (OS) and disease-free survival (DFS), calculated from surgery using the restricted mean (rm) survival.

Plasma Collection

Peripheral blood was prospectively collected at three time points in EDTA tubes, transported cold, centrifuged at 1,000 × g for 10 minutes, and plasma stored at −80 °C.

EV Isolation and KRAS Mutation Analysis

Extracellular vesicles (EVs) were isolated from thawed plasma using a modified lipid nanoprobe method [27]. EV-DNA was extracted (Quick-DNA Microprep Plus Kit), quantified (Qubit Fluorometer), and pre-amplified using KRAS wild-type blocker PCR with Q5 polymerase. KRAS mutation detection and quantification were performed using single-plex droplet digital PCR (ddPCR; QX200 AutoDG, Bio-Rad) with validated assays for KRAS p.G12A, p.G12S, p.G12C, p.G12V, p.G12R, and p.G12D. The assays were validated using synthetic or genomic DNA controls. Mutations were deemed positive upon detection of 6-FAM–positive droplets (further details regarding the methods are available in the Supplemental File: Methods (Extended)).

Statistical Analysis

Comparisons across the ND, DD, and ID groups were made using t-tests, Mann-Whitney U tests, Fisher’s exact test, or chi-square test, as appropriate. Kaplan-Meier survival estimates were used to evaluate the association between longitudinal changes in EV-KRAS MAF, OS, and DFS. The log-rank test was used to compare the restricted mean overall survival (rmOS) and restricted mean disease-free survival (rmDFS). Multivariable Cox proportional hazards modeling (forward selection) identified independent survival predictors (entry threshold, p <0.2 retention,n p <0.05). The predictive accuracy was assessed using Harrell’s C-statistic. Analyses were performed using STATA v18 (StataCorp, TX, USA), and significance was set at p <0.05.

RESULTS

Among the cohort (mean age 66.5±8.2 years, 36% females), 29 patients (65.9%) were classified as ND, 8 (18.2%) as DD and 7 (15.9%) as ID based on EV-KRAS MAF dynamics. Baseline demographics, including age, sex, CCI, and ECOG performance status, were comparable across the groups (all p >0.05, Table 1). No differences were observed in baseline disease characteristics, including tumor size, CA19–9 secretor status, CA19–9 level at diagnosis, and tumor resectability (all p >0.05, Table 1).

Table 1:

Demographics, disease factors and pre-operative treatment variables

Variable		Non-detected (n=29)	Increase (n=7)	Decrease (n=8)	p-value
Age		65.0 (61.5, 72.4)	63.3 (59.0, 71.5)	69.5 (64.0, 71.0)	0.900
Sex	Male	17 (59%)	6 (86%)	5 (62%)	0.410
	Female	12 (41%)	1 (14%)	3 (38%)
CCI- Age adjusted		5.0 (4.0, 6.0)	4.0 (4.0, 5.0)	4.5 (4.0, 5.0)	0.710
ECOG		0.0 (0.0, 1.0)	0.0 (0.0, 1.0)	1.0 (0.0, 1.5)	0.140
Resectability at diagnosis	Resectable	16 (55%)	2 (29%)	4 (50%)	0.550
	Borderline	12 (41%)	4 (57%)	4 (50%)
	Locally advanced	1 (3%)	1 (14%)	0 (0%)
CA19–9 secretor	Secretors	20 (69%)	5 (71%)	5 (62%)	0.920
	Non Secretors	9 (31%)	2 (29%)	3 (38%)
CA19–9 (Pre-NAT)		143.8 (38.1, 668.7)	83.2 (8.7, 446.0)	83.0 (63.2, 176.6)	0.760
CT tumor size (Pre-chemo)(cm)		2.9 (2.4, 3.4)	3.7 (2.7, 3.8)	2.5 (1.7, 3.6)	0.120
EUS tumor size (Diagnosis)(cm)		2.6 (2.1, 3.0)	2.8 (2.2, 3.6)	2.8 (2.0, 3.3)	0.820
NAC Type Groups	Gem/Abraxane	7 (24%)	0 (0%)	2 (25%)	0.020
	FOLFIRINOX	19 (66%)	3 (43%)	6 (75%)
	Adaptive GA/FX	3 (10%)	4 (57%)	0 (0%)
NAC Number of cycles		5.0 (3.0, 6.0)	6.0 (4.0, 7.0)	6.0 (3.5, 8.5)	0.350
NART	No	27 (93%)	7 (100%)	7 (88%)	0.630
	Yes	2 (7%)	0 (0%)	1 (12%)
Mode of surgery	Open	13 (45%)	4 (57%)	5 (62%)	0.620
	Robotic	16 (55%)	3 (43%)	3 (38%)
Vessel Resection	No	17 (59%)	2 (29%)	5 (62%)	0.320
	Yes	12 (41%)	5 (71%)	3 (38%)
Clavien Dindo ≥3	No	26 (90%)	5 (71%)	8 (100%)	0.210
	Yes	3 (10%)	2 (29%)	0 (0%)

Abbreviations: CCI, Charlson comorbidity index; ECOG, Eastern Cooperative Oncology Group; CA19–9, Carbohydrate antigen 19–9; NAT, Neoadjuvant therapy (chemotherapy and/or radiation); CT, Computed tomography; EUS, Endoscopic ultrasound; NAC, neoadjuvant chemotherapy; NART, neoadjuvant radiotherapy.

Values depicted as median (IQR) or occurrence (percentage)

p-value < 0.05 was considered significant

For treatment variables, patients received distinct NAC regimens, with the ID cohort demonstrating a higher proportion (57%) of crossover adaptive treatments (gemcitabine-based to 5-Flurouracil-based or vice versa) (p =0.020). However, the number of neoadjuvant cycles, rate of receiving neoadjuvant radiation, surgical modalities, vascular resections, and incidence of serious postoperative complications were comparable among the groups (all p >0.05, Table 1).

Post-NAT assessments revealed no differences in CA19–9 levels, postoperative CA19–9 levels, or the percent change in CA19–9 from diagnosis to postoperative measurements (all p >0.05, Table 2). Among the thirty CA19–9 secretors (excluding 14 non-secretors), an optimal CA19–9 response was observed in 6 of 20 patients (30%) in the ND cohort, 3 of 5 patients (60%) in the DD cohort, and 2 of 5 patients (40%) in the ID cohort (p =0.551).

Table 2:

Oncologic outcomes among the three defined cohorts

Factor		Non-detected (n=29)	Increase (n=7)	Decrease (n=8)	p-value
CA 19–9 (Post- NAT)		45.3 (16.0, 84.0)	11.0 (7.0, 57.0)	18.5 (14.9, 49.0)	0.500
CA 19–9 (Post-op)		17.6 (9.6, 25.4)	29.7 (4.0, 272.0)	11.4 (8.8, 34.4)	0.890
CA19–9 percentage change *		91.9 (78.2, 96.3)	78.7 (39.0, 88.4)	86.6 (58.8, 89.7)	0.110
CT tumor size (Post NAT) (cm)		2.0 (1.6, 2.4)	2.6 (2.3, 2.9)	1.7 (1.3, 2.2)	0.034
CT tumor size percentage change		28.8 (4.0, 34.6)	17.8 (14.8, 26.3)	26.0 (11.8, 38.9)	0.850
KRAS mutation type	G12C	1 (3%)	0 (0%)	0 (0%)	0.029
	G12D	4 (14%)	2 (29%)	6 (75%)
	G12R	15 (52%)	2 (29%)	0 (0%)
	G12V	9 (31%)	3 (43%)	2 (25%)
Pathological tumor size (cm)		2.1 (1.7, 2.6)	2.0 (1.8, 3.2)	2.0 (1.1, 2.5)	0.760
Tumor grade	G2	22 (76%)	3 (43%)	6 (75%)	0.380
	G3	6 (21%)	4 (57%)	2 (25%)
	G4	1 (3%)	0 (0%)	0 (0%)
Lymph node positive		1.0 (0.0, 5.0)	4.0 (3.0, 5.0)	4.5 (1.0, 9.0)	0.190
Lymph node ratio		0.0 (0.0, 0.1)	0.1 (0.1, 0.2)	0.2 (0.0, 0.3)	0.120
Total lymph nodes		32.0 (28.0, 37.0)	34.0 (33.0, 42.0)	27.5 (24.5, 36.0)	0.340
Margin status	Negative	18 (62%)	4 (57%)	4 (50%)	0.820
	Positive	11 (38%)	3 (43%)	4 (50%)
NAT response	Not reported	1 (3%)	0 (0%)	0 (0%)	0.910
	Near-complete	1 (3%)	0 (0%)	1 (12%)
	No response	8 (28%)	2 (29%)	2 (25%)
	Partial	19 (66%)	5 (71%)	5 (62%)
Stage	I	13 (45%)	1 (14%)	2 (25%)	0.340
	II	7 (24%)	1 (14%)	2 (25%)
	III	9 (31%)	5 (71%)	4 (50%)
Recurrence	No	12 (41%)	1 (14%)	3 (38%)	0.410
	Yes	17 (59%)	6 (86%)	5 (62%)
Adjuvant therapy		27 (93%)	5 (71%)	7 (88%)	0.270
Adj therapy cycles		4.0 (3.0, 6.0)	2.5 (1.5, 3.5)	4.0 (2.2, 5.5)	0.310
Total Cycles		9.0 (6.0, 11.0)	7.0 (6.0, 8.0)	9.8 (8.5, 11.5)	0.072

Abbreviations: CA 19–9, Carbohydrate antigen 19–9; NAT, Neoadjuvant therapy (chemotherapy and/or radiation); CT, Computed tomography; EUS, Endoscopic ultrasound; NAC, neoadjuvant chemotherapy; NART, neoadjuvant radiotherapy.

Values depicted as median (IQR) or occurrence (percentage)

p-value < 0.05 was considered significant

^*CA19–9 percentage change calculated from diagnosis to post-op

^**CT tumor size percentage change calculated from diagnosis to post NAT

The post-NAT CT tumor size was significantly smaller in the DD cohort (1.7 cm) than in the ND (2.0 cm) and ID (2.6 cm, p =0.034) cohorts. However, no significant differences were noted in the percentage change in CT tumor size from diagnosis to post-NAT. Pathological outcomes, including tumor size, LN positivity, LN ratio, tumor grade, perineural invasion, lymphovascular invasion, and margin status, were similar between the three groups (all p >0.05, Table 2). Although the pathologic tumor stage did not differ significantly across groups, detection of EV-KRAS mutants correlated with PDAC stage, where 50%, 30%, and 21% of cases for stages III, II, and I PDAC had detectable EV-KRAS mutants, respectively.

Patients in the ID group received fewer adjuvant and total systemic therapy cycles (neoadjuvant + adjuvant) compared to the ND and DD groups, although this difference was not statistically significant (p >0.05, Table 2).

Survival analysis using Kaplan-Meier estimate (Figures 1A and 1B) revealed significant differences: the ND cohort exhibited the longest rmDFS [31.2 months (95% CI: 24.1–38.3)] compared to the DD [27.8 months (95% CI: 12.4–43.1)] and ID cohorts [9.8 months (95% CI: 0.1–19.7)] (p =0.010, Figure 1A). Similarly, rmOS was superior in ND [40.3 months (95% CI: 34.2–46.4)] relative to DD [35.7 months (95% CI: 22.1–49.2)] and ID [17.7 months (95% CI: 6.4–28.9)] cohorts (p =0.012, Figure 1B).

Figure 1:

Kaplan-Meier survival estimate curve depicting DFS (1A) and OS (1B) in the three groups: EV-KRAS mutation non-detected (ND), the decreasing EV-KRAS MAF detection (DD), and the increasing EV-KRAS MAF detection (ID).

Further analyses stratified by temporal changes in EV-KRAS MAF and optimal CA19–9 response (>50% reduction and normalization post-NAT) demonstrated distinct prognostic implications (Figure 2A and 2B). Notably, patients in the ND cohort with optimal CA19–9 response (n =6; 30%) achieved the longest DFS and OS (median not reached), outperforming all the other subgroups. Interestingly, even among patients who achieved an optimal CA19–9 response following NAT, those in the ID group had significantly shorter disease-free intervals after surgical resection, with an rmDFS of 7.9 months (95% CI, 4.5–11.3 months; p < 0.05). Furthermore, patients in the ID group who also exhibited a suboptimal CA19–9 response to NAT had the poorest rmDFS in the entire cohort at only 3.2 months (95% CI, 1.3 to 5.0 months; p <0.05; Figure 2A).

Figure 2:

Kaplan-Meier survival estimate curve showing the combined impact of EV-KRAS MAF and CA19–9 on DFS (2A) and OS (2B) in the three patient cohorts

Univariable and multivariable proportional hazard analyses were performed to identify the independent predictors of OS and DFS (Supplemental Tables 2 and 3 and Tables 3A and 3B). The prognostic significance of EV-KRAS MAF was evident in both univariable and multivariable analyses for OS and DFS. On univariable analysis, ID was strongly associated with worse OS (HR: 4.53, 95% CI: 1.52–13.42, p = 0.007) and DFS (HR: 3.91, 95% CI: 1.50–10.18, p = 0.005) (Supplemental Tables 2 and 3). These findings persisted in the most parsimonious multivariable models, which adjusted for key confounders, including surgical margin status and LN involvement. ID remained significantly associated with inferior OS (HR: 6.95, 95% CI: 2.00–24.16, p= 0.002) and DFS (HR: 6.14, 95% CI: 2.04–18.41, p= 0.001) (Tables 3A and 3B, respectively). These models showed satisfactory predictive performance, as indicated by Harrell’s C-values of 0.840 for OS and 0.794 for DFS, confirming their robustness in stratifying survival probabilities.

Table 3A:

Multivariable proportional hazard for overall survival from the date of surgery

Variable		Hazard Ratio	95% CI	p-value
Positive (R1) Margin		4.49	1.67–12.07	0.002
Lymph node ratio
	>0 LNR < 20%	2.04	0.58–7.21	0.267
	≥ 20%	6.55	1.70–25.18	0.003
EV-KRAS MAF
	Decrease (DD)	0.64	0.18–2.25	0.483
	Increase (ID)	6.95	2.00–24.16	0.002

EV-KRAS reference: not detected. Margin reference: R0 (no R2 resection in this series). Lymph node ratio reference: 0 (negative lymph node)

The global test for the proportional-hazards assumption showed no significant violations (p = 0.4265), confirming the model’s validity. Among 44 subjects and 625 comparison pairs, the analysis yielded a Harrell’s C value of 0.8400. Somers’ D, reflecting the strength of association between the model’s predictions and observed outcomes, was 0.6800, demonstrating a strong predictive ability. These findings suggest that the model performs well in discriminating between individuals with different survival probabilities.

Table 3B:

Multivariable proportional hazard for disease free survival from the date of surgery

Variable		Hazard Ratio	95% CI	p-value
Pathologic Stage AJCC
	II	4.01	1.30–12.41	0.016
	III	4.95	1.75–14.00	0.003
Positive (R1) Margin		4.47	1.84–10.86	0.001
EV-KRAS MAF
	Decrease (DD)	0.55	0.18–1.67	0.290
	Increase (ID)	6.14	2.04–18.41	0.001

Pathological Stage reference: stage AJCC stage 1; EV-KRAS reference: not detected. Margin reference: R0 (no R2 resection in this series.

The global test for the proportional-hazards assumption showed no significant violations (p = 0.105), confirming the model’s validity. Among 44 subjects and 780 comparison pairs, the analysis yielded a Harrell’s C value of 0.794, and a Somers’D of 0.5872 reflecting a moderate-to-strong predictive ability of the model.

DISCUSSION

This study identified EV-KRAS MAF as a promising biomarker with significant prognostic value in patients with localized PDAC. Temporal changes in EV-KRAS MAF and its correlation with survival offer important insights into disease biology and treatment response. Patients with “No Detection” (ND) of EV-KRAS MAF had the longest DFS and OS, especially when paired with an optimal CA19–9 response. In contrast, patients with “Increasing Detection” (ID) of KRAS MAF exhibited poor survival, indicating that rising levels reflect either aggressive tumor biology or therapy resistance. These findings were supported by multivariable analyses, which showed that increasing EV-KRAS MAF dynamics independently predicted poor DFS and OS.

EVs are secreted by metabolically active cells and carry genetic materials, including DNA fragments with mutant KRAS, providing insights into tumor behavior, treatment response, and disease progression. The temporal dynamics of EV-KRAS MAF reflect a shift in tumor burden and microenvironmental interactions during therapy [28]. Tumor-derived EVs are known to facilitate metastasis by conditioning distant organs through immune modulation, stromal remodeling, and vascular permeability, processes collectively termed pre-metastatic niche formation.[29] Through the delivery of oncogenic cargo, EVs may promote micrometastatic dissemination before clinical recurrence is apparent.[30] In this context, rising EV-KRAS MAF could serve as a real-time indicator of increasing systemic disease burden and biologically aggressive tumor behavior.

While CA19–9 is routinely used, it is limited by low specificity and inability to detect in non-secretors.[10–12] In contrast, EV-KRAS MAF is tumor-specific and, unlike the non-secretor status in CA19–9, non-detection of EV-KRAS was closely associated with survival. Although both CA19–9 and EV-KRAS MAF demonstrated prognostic value, they were not significantly correlated with each other, reflecting their distinct biological characteristics. CA19–9 is a glycoprotein biomarker influenced by tumor burden as well as non-malignant factors such as biliary obstruction, whereas EV-KRAS MAF directly measures tumor-derived genetic material in circulating EVs.[12,15] Importantly, their combination enhanced risk stratification: patients with non-detectable EV-KRAS MAF and optimal CA19–9 response experienced the longest survival, while those with increasing EV-KRAS MAF and suboptimal CA19–9 response had the shortest rmDFS. Notably, even among patients with optimal CA19–9 response, a group typically associated with a favorable prognosis, those with increasing EV-KRAS MAF had significantly reduced rmDFS. These findings may be explained by more aggressive disease features, including larger post-treatment tumor size and higher AJCC stage in the ID cohort. Additionally, patients in the ND group tended to have fewer positive LN and earlier disease stage, suggesting a lower micrometastatic burden, likely leading to non-detection of EV-KRAS. While differences in LN positivity did not reach statistical significance, this trend supports the hypothesis that EV-KRAS MAF dynamics may reflect systemic tumor activity.

The observation that 66% of patients exhibited non-detectable EV-KRAS MAF at all measured time points, despite having KRAS-mutant tumors, suggests that non-detection may reflect a combination of favorable tumor biology and technical factors. Biologically, patients in this group often had early-stage disease and lower systemic tumor burden, which may limit EV shedding into circulation. Additionally, these tumors may be more responsive to cytotoxic therapies, such as neoadjuvant chemotherapy, suppressing EV release by reducing tumor cell viability and altering vesicle biogenesis. [31] However, technical considerations also warrant attention. Our ddPCR assay has a validated detection threshold of 0.01% MAF. Therefore, low-level KRAS mutations in EV-DNA may fall below this threshold. Overall, these factors suggest that EV-KRAS MAF non-detection may arise from both biological and assay-related mechanisms. Notably, non-detection consistently correlated with significantly improved survival, underscoring its prognostic relevance while highlighting the need for further studies to optimize detection sensitivity and account for tumor heterogeneity.

On the other end, the increased utilization of adaptive chemotherapy regimens ( i.e., chemotherapy treatment switch between different regimens) in the ID group also supports the hypothesis that rising EV-KRAS MAF reflects resistance to NAT and/or a more aggressive tumor phenotype. In our clinical practice, patients demonstrating suboptimal biochemical or radiological responses to an initial chemotherapy regimen are often transitioned to an alternative regimen in an attempt to achieve improved therapeutic outcomes. Although adaptive chemotherapy has been associated with survival benefit in some retrospective studies, it often indicates ineffective initial treatment or inherently aggressive tumors.[32] Additionally, patients in the ID cohort completed fewer adjuvant and total chemotherapy cycles than those in the ND and DD cohorts, a trend attributable to early recurrence in this subgroup, which hindered the initiation or completion of the planned adjuvant therapy.

A major challenge in translating EV-based biomarkers into clinical use has been the need for labor-intensive and costly isolation methods.[33–36] However, our study employed lipid nanoprobe technology, enabling rapid, cost-effective EV isolation.[27] The magnetic lipid nanoprobe utilizes a key structural feature of EVs, the lipid bilayer membrane, for efficient and rapid isolation of EVs using a magnetic stand. This cutting-edge technology enables one-step EV isolation from plasma in as low as 15 minutes. Furthermore, the lipid nanoprobe technology, based on magnetic beads, is highly compatible with automation, supporting high-throughput and reproducible processing of clinical plasma samples. The entire workflow, from EV isolation to ddPCR assay, is also cost-effective, with a per-test cost of approximately $20. The efficient, cost-effective, precise quantification of KRAS mutations with this assay underscores the technical feasibility and clinical applicability of this approach and our study results.

While various studies have explored the diagnostic potential of EV-associated DNA, miRNA, and proteins, few have investigated the correlations between EV-DNA MAF dynamics and survival.[16,17,37] Previous studies examining longitudinal monitoring of EV-DNA mutation have reported associations with poor prognosis, but there are limited data on the correlation between the temporal change of EV-KRAS MAF and long-term survival in localized PDAC.[38] Bernard et al. evaluated patients with metastatic PDAC, showing that increases in EV-KRAS and ctDNA MAF were associated with worse OS and PFS. [38] In a separate cohort of patients with localized PDAC, in the same study, the authors assessed changes in EV-KRAS MAF after NAT and correlated these dynamics with disease resectability. Building on these observations, the present analysis confirms and extends the prognostic value of EV-KRAS MAF in localized PDAC treated with NAT and curative resection, establishing its predictive potential for long-term outcomes (OS and DFS). The addition of a post-resection time point allowed further stratification based on dynamic EV-KRAS MAF trends, identifying a clear survival difference between patients with high baseline levels at diagnosis that improved with treatment (DD cohort) and those with low baseline levels that increased with treatment (ID cohort). Both studies acknowledged the complementary role of EV-DNA and CA19–9, but our work demonstrates that EV-KRAS dynamics may potentially outperform CA19–9 by enabling more granular patient stratification, particularly in cases with discordant changes between EV-KRAS MAF and CA19–9 levels. Finally, Bernard et al. used ultracentrifugation and multiplex KRAS ddPCR assay from Bio-Rad. This approach has notable limitations: ultracentrifugation is both expensive and inefficient for EV isolation in clinical settings, and multiplex ddPCR assays are prone to false positives, limiting their reliability in detecting low allele frequencies of EV-KRAS mutations in localized PDAC. In contrast, our study utilized a lipid nanoprobe for efficient EV isolation and single-plex KRAS mutation assays with validated sensitivity and specificity for detecting patient-specific KRAS mutations using a tumor-informed approach, which provides accurate, highly sensitive, and highly specific measurements of tumor-derived EV-KRAS MAF, making it a reliable method for translation into clinical diagnostic settings.

Despite the encouraging results, this study has several limitations. These include a small sample size, time points limited to one-month postoperative and analysis restricted to EV-KRAS G12 mutations. In this study, we focused our assay on KRAS G12 mutations, as they account for the vast majority of KRAS alterations in PDAC. Owing to the limited sample size and the necessity of optimizing our detection method, we prioritized this hotspot over less common KRAS variants to achieve maximal sensitivity and specificity. As a result, this study does not capture patients harboring rarer KRAS mutations, representing a limitation that could be addressed in future investigations encompassing a broader mutation profile. Additionally, the observation that EV-KRAS MAF dynamics may serve as a stronger prognostic indicator than CA19–9 responses is based on a very small subset of patients in our cohort: specifically, those who exhibited an optimal CA19–9 response yet had increased detection of EV-KRAS MAF. The limited sample size of this group restricts the statistical power and generalizability of the results. Therefore, while these preliminary findings suggest a potential enhancement in prognostic accuracy by incorporating EV-KRAS dynamics, they should be interpreted with caution because of limitations. Further studies with larger patient populations are necessary to validate these observations and determine the clinical significance of EV-KRAS MAF dynamics among patients with optimal CA19–9 responses.

Future studies with larger cohorts will aim to incorporate additional intermediate time points at two-month intervals, aligning with imaging restaging, and conventional biomarkers such as CA19–9. This expanded framework will enable a more detailed analysis of EV-KRAS MAF dynamics throughout treatment, thereby improving the precision and clinical applicability of these classifications.

In conclusion, this study indicates that EVs carrying tumor-specific KRAS mutations have significant potential as prognostic biomarkers in localized PDAC. By categorizing patients based on temporal changes in EV-KRAS MAF, we identified that non-detection (ND group) was associated with significantly longer survival compared to groups with increasing (ID) or decreasing (DD) MAF detection. Importantly, patients with non-detectable EV-KRAS MAF and an optimal CA19–9 response had the most favorable outcomes, while those with increasing EV-KRAS MAF had the poorest survival, even among CA19–9 responders. These findings highlight the complementary but independent prognostic value of EV-KRAS MAF dynamics, particularly in identifying high-risk patients who may not be captured by CA19–9 alone.

In the context of the current tumor-biomarker literature, our study provides insight into the potential of EV-based biomarkers for the clinical management of PDAC patients.

Supplementary Material

Supplemental Method (Extended)

Supplemental File 1: Supplemental Methods (Extended)

Supplementary Tables

Supplemental File 2:

Supplemental Table 1: Complete classification of clinical cases per temporal changes of EV-KRAS MAF

Supplemental Table 2: Univariate proportional hazard analysis for OS from the date of surgery

Supplemental Table 3: Univariate proportional hazard for DFS from the date of surgery

Synopsis:

Prospective analysis of exosomal vesicle (EV)-associated KRAS mutation allelic frequency (MAF) in plasma samples of patients with pancreatic adenocarcinoma undergoing neoadjuvant therapy and curative resection using lipid nanoprobe technology and ddPCR identified the negative prognostic value of detection and temporal increase of EV-KRAS MAF.

Data access statement:

In compliance with the IRB and patient consents, only de-identified data for this manuscript will be available upon request to the corresponding author.