The interplay between tumor-intrinsic KRT6A and host-derived immune-nutritional status (PNI) defines prognosis in resectable PDAC

Abstract

Current anatomic staging inadequately predicts outcomes in resectable pancreatic ductal adenocarcinoma (PDAC). This study integrates tumor-intrinsic KRT6A expression with host-derived prognostic nutritional index (PNI) to develop a novel prognostic model. This retrospective study enrolled 105 patients who underwent pancreaticoduodenectomy for histologically confirmed PDAC between January 2019 and December 2024. KRT6A expression was quantified by immunohistochemistry using computer-assisted image analysis. PNI was calculated as serum albumin (g/L) + 5 × lymphocyte count (10⁹/L). The primary endpoint was disease-free survival (DFS). Univariate and multivariate Cox regression analyses were performed to identify independent prognostic factors. A prognostic nomogram (KSI-Nomo) integrating KRT6A and PNI was constructed and validated using time-dependent ROC curves, calibration plots, and decision curve analysis. Risk stratification was performed based on nomogram total points. KRT6A expression was significantly elevated in poorly differentiated tumors and functionally promoted PDAC cell migration and invasion in vitro. Multivariate analysis identified KRT6A expression (HR = 6.337, 95% CI: 1.733–9.540, P = 0.034) and PNI (HR = 0.953, 95% CI: 0.245–1.702, P = 0.014) as the sole independent prognostic factors for DFS, outperforming conventional inflammatory indices (NLR, PLR, SII) and TNM staging. The KRT6A-PNI nomogram demonstrated excellent discriminative accuracy with AUC values of 0.838 (1-year), 0.836 (2-year), and 0.969 (3-year). Calibration curves showed good agreement between predicted and observed survival probabilities. Decision curve analysis confirmed superior net clinical benefit compared to treat-all or treat-none strategies. Risk stratification identified three distinct prognostic groups: low-risk (30% of patients, 3-year DFS rate 52.3%), intermediate-risk (40%, 38.6%), and high-risk (30%, 12.4%) (log-rank P < 0.001). The KRT6A-PNI model provides superior risk stratification for resectable PDAC, enabling personalized treatment decisions.

Introduction

Pancreatic ductal adenocarcinoma (PDAC) remains one of the most lethal malignancies with a 5-year overall survival (OS) rate below 10%, primarily due to late diagnosis, early metastasis, and remarkable resistance to conventional therapies. Despite surgical resection offering the only potential for cure, even among patients undergoing radical surgery, prognosis varies significantly, with median survival ranging from 12 to 36 months^1–3. Current staging systems based solely on anatomic extent (TNM classification) inadequately capture the biological heterogeneity of PDAC, particularly the immunosuppressive tumor microenvironment (TME) that critically dictates therapeutic response and clinical outcomes⁴.

Recently, keratin 6 A (KRT6A), a type II intermediate filament protein traditionally associated with epithelial differentiation and wound healing, has emerged as a critical molecular determinant in pancreatic cancer pathobiology⁵. Transcriptomic profiling by Moffitt et al.⁶ identified KRT6A as a cardinal marker of the basal-like molecular subtype of PDAC, which is characterized by squamous differentiation features, aggressive clinical behavior, and resistance to conventional chemotherapy. Subsequent immunohistochemical studies have validated that KRT6A protein expression is significantly elevated in poorly differentiated tumors and serves as an independent prognostic biomarker for reduced overall survival⁷.

Beyond tumor-autonomous mechanisms, the host systemic response plays an equally critical role in determining outcomes. Emerging data emphasize that the balance between tumor burden and host defense capacity—encompassing immune competence and nutritional reserves—is a pivotal determinant of survival in PDAC⁸. The prognostic nutritional index (PNI), calculated from serum albumin and peripheral lymphocyte count, integrates these dimensions, reflecting both inflammatory burden and adaptive immune function. Low PNI, indicative of malnutrition and lymphopenia, is linked to impaired antitumor immunity, increased postoperative complications, and worse survival across multiple cancers, including PDAC. Unlike static inflammation ratios (e.g., NLR, PLR), PNI captures a more comprehensive state of host frailty, making it particularly relevant in a disease characterized by profound cachexia and immune suppression^9–11.

Despite these advances, a critical gap persists in integrating tumor-intrinsic aggression with host systemic resilience. Most prognostic models evaluate either molecular features within the tumor or systemic markers in isolation, neglecting the dynamic interplay between the two^12,13. We hypothesized that combining KRT6A (representing tumor ‘attack’ potential) with PNI (reflecting host ‘defense’ capacity) would provide superior risk stratification than either factor alone, capturing the essential tumor-host balance that governs clinical outcomes.

Herein, we present a retrospective cohort study establishing a novel KRT6A-PNI prognostic model. By integrating immunohistochemistry-quantified KRT6A expression with preoperative PNI, we developed and validated a clinically practical nomogram for individualized survival prediction in resectable PDAC. This model bridges tumor biology and host physiology, offering a biologically grounded, easily implementable tool for preoperative risk assessment, adjuvant therapy personalization, and patient selection for neoadjuvant or immunomodulatory trials.

Materials and methods

Patient population and study design

This retrospective study consecutively enrolled patients who underwent curative-intent pancreaticoduodenectomy for histologically confirmed PDAC at our hospital between January 2019 and December 2024.

Inclusion criteria: (1) Age range (29–81 years old); (2) Postoperative pathology confirmed pancreatic ductal adenocarcinoma; (3) Preoperatively evaluated as resectable according to the 2024 National Comprehensive Cancer Network(NCCN) guidelines¹⁴; (4) Availability of formalin-fixed paraffin-embedded (FFPE) tumor tissue for IHC analysis; (5) No contraindication in preoperative evaluation; (6) Informed consent of patients and their families; (7) Complete clinicopathological data and follow-up information.

Exclusion criteria: (1) Borderline resectable or unresectable pancreatic head cancer according to NCCN Guidelines (2022); (2) Postoperative histopathological diagnosis other than pancreatic ductal adenocarcinoma; (3) Unavailability of formalin-fixed paraffin-embedded (FFPE) tumor tissue for immunohistochemical analysis; (4) Incomplete clinicopathological data or loss to follow-up.

The study protocol adhered to the Declaration of Helsinki and received formal approval from the Ethics Committee of Beijing Chaoyang Hospital (Approval No.2024-D-512). Written informed consent was secured from all participants and their legal representatives prior to data inclusion. For cell line experiments, all procedures complied with institutional biosafety guidelines.

Data collection and calculation of clinical indicators

The demographic profiles of the enrolled patients were documented, including age and gender distribution. The initial clinical presentations upon admission, preoperative comorbidities and medical histories (smoking and diabetes mellitus), preoperative interventional procedures for biliary drainage, such as Endoscopic Retrograde Cholangiopancreatography (ERCP) with stent placement and Percutaneous Transhepatic Biliary Drainage (PTBD), were also collected as baseline clinical characteristics. Peripheral blood parameters obtained within 7 days preoperatively were extracted from electronic medical records. This included full blood counts (neutrophils, lymphocytes, monocytes, and platelets), serum biochemical markers (albumin, total bilirubin, and γ-glutamyl transferase), and tumor marker levels (CA19-9). The following inflammation-based prognostic scores were calculated: Neutrophil-to-Lymphocyte Ratio (NLR)=Absolute neutrophil count/Absolute lymphocyte count; Platelet-to-Lymphocyte Ratio (PLR)=Absolute platelet count/Absolute lymphocyte count; Lymphocyte-to-Monocyte Ratio (LMR)=Absolute lymphocyte count/Absolute monocyte count; Systemic Immune-Inflammation Index (SII)=Platelet count×Neutrophil count/Lymphocyte count; Prognostic Nutritional Index (PNI)=Serum albumin (g/L) + 5 × Lymphocyte count (10⁹/L).

Intraoperative variables were extracted from surgical records, including the duration of surgery, the volume of intraoperative blood loss, and the requirement for blood transfusion during the procedure. Postoperative pathological reports were reviewed to extract key tumor characteristics, including tumor size, tumor differentiation, lymph node metastasis, R0/R1 resection, TNM stage.

Postoperative short-term outcomes were monitored. The occurrence of various surgical complications including specific types of fistulas, infections, hemorrhage, and gastrointestinal issues, postoperative adjuvant chemotherapy and postoperative hospital stay was also documented.

Follow-up and survival endpoints

Standardized follow-up protocols were conducted to monitor patient outcomes. The primary survival endpoint for this study was defined as Disease-Free Survival (DFS). DFS was calculated from the date of surgery to the date of tumor recurrence or death from any cause, whichever occurred first. In cases where no recurrence or death was recorded, patients were censored at the date of last follow-up. All subsequent univariate and multivariate analyses were performed with DFS as the endpoint. The survival durations and survival rates at predefined time intervals were analyzed accordingly.

Immunohistochemistry and KRT6A quantification

FFPE tissue Sect. (4 μm thick) were deparaffinized and subjected to antigen retrieval in citrate buffer (pH 6.0) using microwave heating. Endogenous peroxidase activity was blocked with 3% hydrogen peroxide. Sections were incubated overnight at 4 °C with rabbit monoclonal anti-KRT6A antibody (2 µg/mL dilution; Clone: KRT6/3997R, Thermo Fisher) after serum blocking. Immunodetection was performed using a polymer-based horseradish peroxidase (HRP) detection system (Dako EnVision+) with 3,3’-diaminobenzidine (DAB) as chromogen. Sections were counterstained with hematoxylin. Negative controls replaced primary antibody with PBS.

KRT6A immunoreactivity was quantified using computer-assisted image analysis (ImageJ). Following antigen retrieval and DAB chromogen development, whole-slide images were captured at 10x objective magnification. The average optical density (AOD) was calculated as the mean DAB intensity within the tumor region, normalized to the total tissue area. Five representative high-power fields (HPFs) were randomly selected from each tumor section, and the mean AOD value was determined for statistical analysis. Higher AOD values indicate stronger KRT6A protein expression. AOD was used as the continuous variable for all subsequent survival analyses and nomogram construction.

Cell culture, RNA interference, and western blot analysis

Human PDAC cell line PANC-1 (ATCC, Manassas, VA) was maintained in DMEM (Gibco, USA) supplemented with 10% FBS (Gibco) and 1% penicillin-streptomycin (Gibco) at 37 °C with 5% CO2. Cell identity was authenticated by STR profiling and confirmed Mycoplasma-negative.

For stable knockdown, PANC-1 cells were transduced with lentiviral particles carrying KRT6A-targeting shRNA (shKRT6A, target sequence: 5’-CAACTTCTTGAGAGCTCTCTA-3’) or scrambled control shRNA (shCtrl) at MOI 20 with 8 µg/mL polybrene (Sigma-Aldrich). After 48 h, cells were selected with 1.5 µg/mL puromycin (Sigma-Aldrich) for 10-14 days to establish stable knockdown cell lines. Efficiency was confirmed by Western blot.

For stable KRT6A overexpression, full-length human KRT6A cDNA (NM_005554) was cloned into pLVX-puro vector (Takara Bio, formerly Clontech) and packaged in HEK293T cells transfected with psPAX2/pMD2.G using PEI (Polysciences). PANC-1 cells were transduced at MOI 20 with 8 µg/mL polybrene and selected with 1.5 µg/mL puromycin for 10-14 days. Efficiency was confirmed by Western blot.

Total protein was extracted using RIPA buffer (Beyotime, China) with protease inhibitors (Roche, Switzerland). Proteins (30 µg/lane) were separated by 10% SDS-PAGE and transferred to PVDF membranes (Millipore). After blocking with 5% non-fat milk in TBST, membranes were incubated overnight at 4 °C with primary antibodies: KRT6A (1:1000; Abcam), E-cadherin (1:1000; CST), Vimentin (1:1000; CST), and GAPDH (1:5000; Proteintech, China). Following TBST washes, HRP-conjugated secondary antibodies (1:5000; Jackson ImmunoResearch) were applied for 1 h at room temperature. Bands were visualized by ECL (Bio-Rad) and quantified using ImageJ (NIH).

Cell migration and invasion assays

Wound healing assay. Cells (5 × 10⁵/well) were seeded in 6-well plates until 90–100% confluence. A sterile 200µL pipette tip was used to create a linear wound, followed by PBS wash and supplementation with serum-free DMEM. Wound closure was imaged at 0 h and 24 h using an inverted microscope (Olympus IX73, ×100 magnification). Migration rate was calculated as: (Wound area at 0 h−Wound area at 24 h)/Wound area at 0 h×100%. Experiments were performed in triplicate with three independent repetitions.

Transwell migration assay. 2 × 10⁴ cells in 200µL serum-free DMEM were seeded into Transwell upper chambers (8 μm pore size; Corning) with 600µL 20% FBS-DMEM in the lower chamber. After 24 h of incubation at 37 °C, non-migrated cells were removed from the upper surface with cotton swabs. Migrated cells on the lower surface were fixed with 4% paraformaldehyde for 20 min and stained with 0.1% crystal violet for 15 min. Cell counts were averaged from five random fields.

Transwell invasion assay. Upper chambers were pre-coated with 50µL Matrigel (Corning, 1:8 dilution in serum-free DMEM) for 2 h at 37 °C. 4 × 10⁴ cells were seeded per well and incubated for 24 h. Subsequent steps followed the migration protocol described above. All experiments were conducted in triplicate with three independent repetitions.

Statistical analysis and model construction

Continuous variables were tested for normality using the Shapiro-Wilk test and expressed as mean±standard deviation (SD) or median (interquartile range, IQR) accordingly. Categorical variables were presented as frequencies (%). Intergroup comparisons were performed using Student’s t-test or Mann-Whitney U test for continuous variables, and Chi-square or Fisher’s exact test for categorical variables. The correlation between continuous KRT6A AOD and tumor differentiation grade was assessed using Spearman’s rank correlation, with differences among well, moderately, and poorly differentiated subgroups compared by one-way ANOVA followed by Tukey’s post-hoc test.

Overall survival (OS) was defined as the interval from surgery to death from any cause or last follow-up, while disease-free survival (DFS) was defined from surgery to tumor recurrence (radiological or histological) or death. Survival curves were generated using the Kaplan-Meier method and compared by log-rank test. Univariate Cox proportional hazards regression identified significant prognostic variables; those with P < 0.05 were entered into multivariate Cox regression using backward stepwise selection (likelihood ratio test) to determine independent prognostic factors, expressed as hazard ratios (HRs) with 95% confidence intervals (CIs). To test the local-global immune interaction hypothesis, multiplicative interaction terms (KRT6A × NLR, KRT6A × PLR) were introduced into Cox models, with biological interaction on an additive scale quantified by calculating the relative excess risk due to interaction (RERI), attributable proportion (AP), and synergy index (S).

A prognostic nomogram (KSI-Nomo) integrating independent predictors was constructed using the rms package in R software (version 4.2.0). Model discrimination was quantified using Harrell’s concordance index (C-index), and calibration was assessed by comparing predicted versus observed survival probabilities at 1, 2, and 3 years using calibration plots with bootstrap resampling (n = 1000). Time-dependent ROC curves were plotted using the timeROC package to evaluate predictive accuracy versus TNM stage alone. Decision curve analysis (DCA) was performed using the rmda package to quantify net clinical benefit across different threshold probabilities. Based on total nomogram points, patients were stratified into low-risk (bottom 30%), intermediate-risk (middle 40%), and high-risk (top 30%) groups. All analyses were conducted using SPSS 26.0 (IBM Corp.), R software (version 4.2.0), and GraphPad Prism 9.0. A two-tailed P < 0.05 was considered statistically significant.

Results

Baseline clinicopathological characteristics

A total of 105 patients were enrolled in this study cohort, comprising 68 males and 37 females (male-to-female ratio 1.84:1), with a mean age of 67.5 ± 10.2 years. Initial clinical presentations included abdominal pain (n = 54), atypical gastrointestinal symptoms (n = 5), incidental detection during physical examination (n = 23), and jaundice (n = 23). All the jaundiced patients received preoperative biliary drainage, including endoscopic retrograde cholangiopancreatography (ERCP) with stent placement (n = 6) and percutaneous transhepatic biliary drainage (PTBD) (n = 17).

All patients underwent successful surgery, with postoperative pathological examination confirming PDAC and no perioperative mortality. Intraoperative blood loss was 400 mL (IQR 200–2000), with 36 patients requiring blood transfusion. The median operative time was 8 h (IQR 5–16). The median tumor size was 2.5 cm (range 0.5–8.5), and 56 patients had lymph node metastasis. Tumor differentiation was classified as moderately in 41 cases, poorly in 26 cases, and well-differentiated in 38 cases. Among the 91 patients (86.7%) who achieved R0 radical resection, positive margins were identified at the distal bile duct margin (n = 5), pancreatic transection margin (n = 6), and proximal bile duct margin (n = 3). Postoperative complications occurred in 30 patients (28.6%), including biochemical leak (n = 8), Grade B pancreatic fistula (n = 6), Grade C pancreatic fistula (n = 4), delayed gastric emptying (n = 13), intra-abdominal infection (n = 6), intra-abdominal hemorrhage (n = 4), biliary fistula (n = 2), pulmonary infection (n = 1), and gastrointestinal bleeding (n = 1). The median postoperative hospital stay was 15 days (IQR 8–92).

KRT6A expression is associated with tumor dedifferentiation and enhanced invasive capacity

To elucidate the clinical significance of KRT6A in PDAC, we performed immunohistochemistry (IHC) staining on a cohort of patient samples. Representative images demonstrated that KRT6A expression was significantly elevated in poorly differentiated tumors compared to moderately and well-differentiated tumors (Fig. 1A). Quantitative analysis of the average optical density (AOD) confirmed a stepwise increase in KRT6A expression levels as the degree of tumor differentiation decreased (Fig. 1D).

Fig. 1.

KRT6A expression correlates with differentiation and promotes PDAC cell migration and invasion in vitro. A Representative immunohistochemistry (IHC) images showing KRT6A protein expression in pancreatic ductal adenocarcinoma (PDAC) tissues. Upper panel: poorly differentiated tumors; middle panel: moderately differentiated tumors; lower panel: well-differentiated tumors. Scale bar = 100 μm. B Representative images of Transwell migration assays (upper panel) and invasion assays (lower panel) in the Ctrl, OE and KD groups (from left to right). Scale bar = 1000 μm. C Representative images of wound healing (scratch) assays at 0 h and 24 h in the Ctrl, OE, and KD groups. The dashed blue lines indicate the scratch edges. Scale bar = 1000 μm. D Quantitative analysis of the average optical density (AOD) of KRT6A in IHC samples according to differentiation grade (Well: n = 38, Moderately: n = 41, Poorly: n = 26). E Quantification of migrated cells from the Transwell migration assays (n = 5 independent experiments). F Quantification of invaded cells from the Transwell invasion assays (n = 5 independent experiments). G Quantitative analysis of migration rate based on the wound healing assays. Data are presented as mean ± SD. Statistical significance was determined using one-way ANOVA with Tukey’s post hoc test. *P < 0.05, **P < 0.01, ***P < 0.001.

Next, we investigated the functional role of KRT6A in PANC-1 cell lines. Three experimental groups were established: control group (Ctrl, transduced with control plasmids), knockdown group (KD, transduced with KRT6A-specific shRNAs), and overexpression group (OE, transduced with KRT6A overexpression plasmids). Transwell migration and invasion assays were conducted to assess cell motility. Representative images revealed that OE group significantly enhanced cell migration and invasion compared to the Ctrl group (Fig. 1B). Conversely, KD group markedly suppressed both migratory and invasive capabilities. Quantification of the migrated and invaded cells corroborated these observations, showing a significant increase in the OE group and a significant decrease in KD group (Fig. 2E and F).

Fig. 2.

Survival curves for patients with PDAC. A Overall survival (OS) curve. Numbers at risk at 0, 12, 24, 36 months are 105, 89, 72, 58. B Disease-free survival(DFS) curve. Numbers at risk at 0, 12, 24, 36 months are 105, 75, 58, 42.

Furthermore, wound healing assays were performed to evaluate cell migration over time. The scratch width in the OE group was notably narrower than in the Ctrl group at the 24 h points, indicating accelerated wound closure. In contrast, the KD group exhibited significantly impaired wound healing compared to the Ctrl group (Fig. 2C). Quantitative analysis of the migration rate confirmed that KRT6A overexpression promoted cell migration, whereas its depletion inhibited this process (Fig. 2G).

Collectively, these findings demonstrate that KRT6A is associated with poor tumor differentiation and plays a critical role in promoting the migratory and invasive potential of PDAC cells in vitro.

Survival outcomes of the study cohort

The follow-up was up to February 2026, with a median follow-up time of 36 months (ranged from 12 to 85 months; interquartile range 24–48 months). The overall median survival time and disease-free survival time of patients was 33 months and 28 months respectively, and the overall survival rates and disease-free survival rates of 1-, 2-, and 3-year were 85.4%, 62.6%, 47.5% and 70.9%, 53.1%, 38.6% (Fig. 2).

Independent prognostic factors: integration of tumor-intrinsic KRT6A and host-derived PNI

Univariate Cox regression analysis identified several inflammatory indices and clinicopathological variables associated with disease-free survival, including PLR (χ² = 11.038, P = 0.004), SII (χ² = 3.908, P = 0.048), PNI (χ² = 11.031, P = 0.001), tumor differentiation (χ² = 4.562, P = 0.033), and KRT6A Expression (χ² = 12.400, P < 0.001) (Table 1). Multivariate Cox regression analysis using backward stepwise selection revealed that KRT6A expression (HR = 6.337, 95% CI: 1.733–9.540, P = 0.034) and PNI (HR = 0.953, 95% CI: 0.245–1.702, P = 0.014) were the only independent prognostic factors for disease-free survival. Notably, other inflammatory indices (PLR, SII) and tumor differentiation lost statistical significance in the multivariate model after adjusting for KRT6A expression and PNI, suggesting that the combination of tumor-intrinsic KRT6A status and host immune-nutritional reserve (PNI) captures the essential prognostic information in resectable PDAC.

Table 1.

Cox Regression Analyses of Prognostic Factors for Disease-Free Survival.

Factors	Univariate analysis		Multivariate analysis
	χ2 value	P value	HR value(95% CI)	Wald	P value
Gender	2.206	0.137
Age	2.755	0.097
Smoking History	0.427	0.808
Diabetes History	0.697	0.706
Biliary drainage	0.095	0.757
NEUT (×10⁹/L)	2.646	0.104
LYM (×10⁹/L)	2.868	0.090
MONO (×10⁹/L)	0.347	0.556
PLT (×10⁹/L)	1.967	0.161
ALB (g/L)	0.005	0.944
CA19-9 (U/mL)	0.001	0.970
GGT(U/L)	2.551	0.110
TB(mg/dL)	1.493	0.222
LMR	1.367	0.242
NLR	3.415	0.065
PLR	11.038	0.004	1.013 (0.797–5.468)	3.627	0.163
SII	3.908	0.048	1.002 (0.918–3.540)	0.018	0.894
PNI	11.031	0.001	0.953 (0.245–1.702)	5.993	0.014
Hemorrhage(ml)	0.112	0.737
Intraoperative blood transfusion	0.023	0.880
Operative time(h)	0.052	0.819
Tumor diameter(cm)	0.006	0.937
Tumor differentiation	4.562	0.033	0.996 (0.990–1.002)	1.955	0.162
Lymph node metastasis	0.573	0.449
R0/R1 resection	1.270	0.260
TNM stage	2.011	0.570
Postoperative chemotherapy	0.212	0.645
Postoperative complications	0.784	0.376
KRT6A expression	12.400	0.000	6.337 (1.733–9.540)	4.479	0.034

Development of the KRT6A-PNI prognostic nomogram

Based on the two independent predictors identified in multivariate analysis, we constructed a prognostic nomogram (designated as KSI-Nomo) to provide individualized prediction of 1-, 2-, and 3-year disease-free survival probabilities (Fig. 3A). The nomogram integrates KRT6A expression (ranging from 0.1 to 1.0) and PNI values (ranging from 20 to 60), assigning weighted points to each variable. Higher KRT6A expression contributes up to 100 points, while lower PNI values contribute up to 70 points, yielding a total point range from 0 to 160. By summing the individual points and locating the total points on the scale, clinicians can directly read the predicted DFS probabilities at each time point.

Fig. 3.

Development of the KRT6A-PNI prognostic nomogram and its discriminative performance. A Nomogram for predicting 1-, 2-, and 3-year disease-free survival probabilities based on KRT6A expression and PNI. To use the nomogram, locate the patient’s KRT6A expression value and PNI on the respective axes, draw upward to determine the points for each variable, sum the points to obtain the total points, and draw downward to read the predicted survival probabilities. B Forest plot of multivariate Cox regression analysis showing hazard ratios (HR) with 95% confidence intervals for independent prognostic factors. KRT6A expression (HR = 6.337) was identified as a risk factor, while PNI (HR = 0.953) served as a protective factor. C Time-dependent ROC curves for the nomogram at 1, 2, and 3 years, demonstrating progressive improvement in predictive accuracy over time.

The forest plot visualization of multivariate analysis clearly demonstrated the opposing effects of the two predictors: KRT6A expression exhibited a hazard ratio of 6.337 (95% CI: 1.733–9.540), indicating a robust risk effect, whereas PNI showed a protective hazard ratio of 0.953 (95% CI: 0.245–1.702) (Fig. 3B). This biological contrast—tumor aggression versus host defense—underlies the nomogram’s prognostic power.

Using bootstrap resampling (n = 1000) for optimism correction, the nomogram demonstrated apparent discriminative ability with AUC values of 0.838 (95% CI: 0.752–0.924) at 1 year, 0.836 (95% CI: 0.741–0.931) at 2 years, and 0.969 (95% CI: 0.936–1.000) at 3 years, demonstrating excellent predictive accuracy that progressively improved throughout the follow-up period (Fig. 3C).

Notably, the substantial increase in AUC at 3 years suggests that the KRT6A-PNI model achieves optimal discrimination for long-term prognosis, likely reflecting the cumulative impact of tumor biology and host immune-nutritional status on disease progression.

Model validation: calibration, clinical utility, and risk stratification

Calibration Assessment: Calibration plots were generated to evaluate the agreement between nomogram-predicted probabilities and actual observed outcomes. At all three time points (1, 2, and 3 years), the calibration curves demonstrated good concordance between predicted and observed survival probabilities, with most data points closely aligned along the ideal 45-degree diagonal line (Fig. 4A). The narrow confidence intervals further supported the reliability of the model’s predictions.

Fig. 4.

Calibration, clinical utility, and risk stratification of the KRT6A-PNI prognostic model. A Calibration plots comparing predicted versus observed survival probabilities at 1, 2, and 3 years. The dashed diagonal line represents perfect calibration, while the solid line with error bars indicates the model’s performance with 95% confidence intervals. B Decision curve analysis (DCA) comparing the net benefit of using the nomogram (blue line) versus treating all patients (red line) or treating no patients (black dashed line) at 1, 2, and 3 years. The nomogram demonstrates superior net clinical benefit across a broad range of threshold probabilities. C Kaplan-Meier disease-free survival curves for patients stratified into low-risk (n = 32), intermediate-risk (n = 42), and high-risk (n = 31) groups. Shaded areas represent 95% confidence intervals. Significant separation between risk groups confirms the model’s robust stratification capability (P < 0.001 by log-rank test).

Clinical Utility Evaluation: Decision curve analysis (DCA) was performed to assess the net clinical benefit of applying the KRT6A-PNI nomogram across different threshold probabilities. Compared to the “treat-all” strategy (red line) and “treat-none” strategy (black dashed line), the nomogram (blue line) provided superior net benefit across a broad range of clinically relevant threshold probabilities at 1, 2, and 3 years (Fig. 4B). This indicates that using the nomogram to guide clinical decision-making would result in better outcomes than either treating all patients aggressively or adopting a watchful waiting approach.

Risk Stratification Performance: To facilitate clinical application, patients were stratified into three distinct risk groups based on total nomogram points: low-risk (bottom 30%, total points < 45), intermediate-risk (middle 40%, total points 45–90), and high-risk (top 30%, total points > 90). Kaplan-Meier analysis revealed significantly separated disease-free survival curves among the three risk groups (log-rank P < 0.001) (Fig. 4C). The median DFS times were 32 months (95% CI: 28–36 months) for the low-risk group, 24 months (95% CI: 18–30 months) for the intermediate-risk group, and 14 months (95% CI: 10–18 months) for the high-risk group. The 3-year DFS rates were 52.3%, 38.6%, and 12.4% for low-, intermediate-, and high-risk groups, respectively. The non-overlapping confidence intervals and clear separation of survival curves confirm the robust stratification capability of the KRT6A-PNI model for identifying patients with distinctly different prognoses.

Discussion

This study establishes a novel prognostic paradigm for resectable pancreatic ductal adenocarcinoma (PDAC) by integrating tumor-intrinsic KRT6A expression with host-derived immune-nutritional status (PNI). Our principal findings are threefold: First, KRT6A expression serves as a robust biomarker of tumor dedifferentiation and aggressive biology, with functional validation demonstrating its causal role in promoting PDAC cell migration and invasion. Second, multivariate analysis identifies KRT6A and PNI as the sole independent prognostic factors, outperforming conventional inflammatory indices (NLR, PLR, SII) and TNM staging. Third, the KRT6A-PNI nomogram demonstrates excellent discriminative accuracy (AUC: 0.838–0.969), calibration, and clinical utility, enabling meaningful risk stratification into three distinct prognostic groups with divergent disease-free survival outcomes. Collectively, these findings substantiate our central hypothesis that the tumor-host interaction—quantified as the balance between malignant potential (KRT6A-driven “attack”) and systemic resilience (PNI-mediated “defense”)—constitutes a fundamental determinant of clinical outcomes in PDAC. This conceptual framework transcends the limitations of anatomy-based staging and single-dimension biomarkers, offering a biologically grounded approach to precision oncology.

Our observation that KRT6A expression escalates with tumor dedifferentiation aligns with transcriptomic studies identifying KRT6A as a cardinal marker of the basal-like PDAC subtype, which is characterized by squamous differentiation features and chemoresistance¹⁵. Additionally, integrated transcriptome and single-cell RNA sequencing analyses have confirmed KRT6A as one of five GBP4-related prognostic genes (alongside GBP2, MMP7, BCAT1, and SPRR1A), with high-risk groups showing distinct immune cell infiltration patterns and cell cycle pathway enrichment¹⁶. However, the present study extends these findings by demonstrating that KRT6A actively promotes malignant phenotypes in vitro, suggesting a functional role beyond passive structural support. Mechanistically, KRT6A appears to function beyond its canonical role as a cytoskeletal component. Pan-cancer analyses revealed that high KRT6A expression correlates with extensive desmoplasia and the recruitment of immunosuppressive stromal elements, particularly tumor-associated macrophages (TAMs) exhibiting the pro-tumorigenic M2 phenotype¹⁷. This immunosuppressive microenvironment is further reinforced by cancer cell-intrinsic mechanisms; a landmark 2025 study demonstrated that transglutaminase-2 (TGM2) expression in PDAC cells promotes T cell suppression through microtubule-dependent secretion of immunosuppressive cytokines, establishing TGM2 as a potent cell-intrinsic driver of immunosuppression correlating with poor overall survival¹⁸. The spatial distribution and functional state of immune cells within this desmoplastic stroma present substantial challenges for cellular therapies; recent advances in cell-based therapies for PDAC highlight that the heterogeneous TME results in cellular dysfunction and exhaustion upon arrival, limiting effective immune surveillance¹⁹. Furthermore, the crosstalk between TAMs and cancer cells is a key driver of systemic host deterioration, such as cachexia, which itself is a major prognostic determinant^20,21. These findings collectively position KRT6A not merely as a passive differentiation marker, but as an active participant in sculpting an immunosuppressive tumor microenvironment that promotes disease progression. Consistent with this mechanistic insight, the striking hazard ratio of 6.337 for KRT6A in our multivariate model underscores its prognostic dominance, rivaling or exceeding that of lymph node status and resection margin—traditional cornerstones of PDAC staging. This observation challenges the conventional hierarchy of prognostic factors and supports the integration of molecular biomarkers into contemporary risk assessment frameworks.

While KRT6A captures tumor aggression, PNI emerged as the only protective factor in our model  HR=0.953, reinforcing its value as a composite measure of immune competence and nutritional reserve. Unlike static inflammation ratios (NLR, PLR, SII) that lost significance in multivariate analysis, PNI incorporates albumin—a negative acute-phase reactant reflecting both hepatic synthetic function and chronic inflammation. This dual dimensionality enables PNI to capture the “host exhaustion” phenotype characteristic of PDAC-associated cachexia, which cannot be discerned from leukocyte counts alone²². Importantly, the independence of PNI from KRT6A suggests that host resilience and tumor biology exert non-overlapping influences on outcomes, supporting a multiplicative rather than merely additive interaction. From a clinical perspective, this biological dichotomy has immediate translational implications: patients with high KRT6A but preserved PNI may represent a subgroup with localized but biologically aggressive disease amenable to intensified adjuvant therapy, whereas those with dual high-risk features may require novel treatment strategies targeting both tumor and host.

The nomogram developed herein addresses a critical unmet need in PDAC management—the inability of current staging systems to discriminate prognosis among patients with anatomically resectable disease. Notably, the progressive improvement in AUC from 1 year (0.838) to 3 years (0.969) suggests that the KRT6A-PNI model becomes increasingly informative as the biological determinants of recurrence manifest, in contrast to anatomic staging which loses discriminative power over time.Our risk stratification strategy identifies a high-risk cohort (30% of patients) with a 3-year DFS rate of merely 12.4% despite technically successful resection, challenging the current “one-size-fits-all” approach to adjuvant therapy. Conversely, the low-risk group (52.3% 3-year DFS) may represent candidates for treatment de-escalation trials, addressing the overtreatment burden that plagues current practice. The need for such refined stratification is echoed in other recent prognostic models incorporating immune-inflammation and coagulation indicators, or modified inflammation-based indexes^23,24, as well as models for specific populations like non-surgical elderly patients²⁵.

Several recent studies have attempted to integrate tumor and host factors in PDAC prognosis. The mGPS (modified Glasgow Prognostic Score) and SII have demonstrated prognostic value, but these indices lack tumor-specific molecular components. Conversely, gene expression signatures (e.g., PurIST, Moffitt subtypes) capture tumor biology but require fresh tissue, complex bioinformatics, and fail to account for host status. The KRT6A-PNI model offers distinct advantages: (i) feasibility, requiring only standard IHC and routine laboratory tests; (ii) biological interpretability, with clear mechanistic links to tumor microenvironment and immune function; (iii) dynamic range, as both KRT6A (continuous AOD values) and PNI (continuous score) allow granular risk discrimination; and (iv) actionability, identifying patients for treatment intensification or novel therapeutic approaches. Notably, our finding that conventional inflammatory ratios lost independent significance contrasts with some reports. We posit that this discrepancy reflects the superior comprehensiveness of PNI in capturing host frailty, as well as the dominant prognostic influence of KRT6A, which may mediate some effects previously attributed to systemic inflammation. The evolving treatment landscape, including the role of neoadjuvant therapy²⁶ and the search for better biological biomarkers to define resectability²⁷, underscores the need for robust, integrative tools like our model. Furthermore, the integration of dynamic biomarkers like circulating tumor DNA into trials represents a future direction for validation and refinement²⁸.

This study has several limitations. First, internal validation only. We employed bootstrap resampling on the same cohort (n = 105) without independent discovery-validation split or external validation. Consequently, performance metrics (AUC 0.838–0.969) may be optimistically biased due to overfitting and require confirmation in multi-institutional cohorts. Second, the retrospective, single-center design introduces selection bias and limits generalizability. Third, while in vitro experiments establish KRT6A as a driver of invasive phenotypes, precise mechanisms(EMT induction, stromal remodeling, or immune modulation) require elucidation in genetically engineered mouse models and organoid systems. Fourth, optimal cut-points for KRT6A and PNI were derived from this cohort; prospective calibration in independent datasets will refine these thresholds. Future research should explore: (i) predictive value for chemotherapy and immunotherapy response; (ii) KRT6A assessment in preoperative biopsies; (iii) longitudinal PNI changes as dynamic indicators; and (iv) therapeutic strategies targeting KRT6A-mediated pathways, potentially with immune checkpoint inhibitors or CAR-M therapies.

As PDAC management evolves toward neoadjuvant and precision therapeutics, biomarkers that capture both malignant potential and systemic resilience will become indispensable. The present study provides compelling evidence that KRT6A and PNI fulfill this need, warranting prospective validation and integration into clinical trials.

Conclusion

The KRT6A-PNI prognostic model bridges the gap between tumor biology and host immunity, offering a practical, biologically grounded tool for individualized risk assessment in resectable PDAC. By capturing the dynamic interplay between malignant aggressiveness (KRT6A-driven) and systemic resilience (PNI-mediated), this model outperforms conventional staging and inflammatory indices, enabling meaningful stratification into distinct risk groups with divergent clinical outcomes. Its feasibility—requiring only standard immunohistochemistry and routine laboratory parameters—facilitates immediate clinical implementation. These findings underscore the necessity of integrating tumor-intrinsic and host-derived factors in contemporary prognostic frameworks. Prospective multi-institutional validation and integration into neoadjuvant therapy trials will be essential to establish its role in guiding treatment intensification and personalized care.

Author contributions

Fangfei Wang: Investigation; Data curation; Writing - original draft; Project administration. Xiangzhou Yang: Investigation; Data curation; Writing - original draft.Xiaodi Dai: Investigation; Data curation. Ren Lang: Data curation; InvestigationXin Zhao: Project administration; Writing - review & editing; Supervision. Shaocheng Lyu: Project administration; Writing - review & editing; Supervision. Qiang He: Project administration, Supervision.

Funding

No Funding.

Data availability

Dataset accessibility is restricted due to stringent adherence to: 1.Legal Compliance——Patient privacy regulations (*e.g.*, China’s Personal Information Protection Law, PIPL) explicitly prohibit public sharing of sensitive health information to mitigate re-identification risks and data misuse. As stated in the manuscript ‘The data are not publicly available due to patient privacy regulations.’ 2.Ethical Mandate——The Ethics Committee of Beijing Chaoyang Hospital (Approval No.: 2024-D-511) enforces compliance with the Declaration of Helsinki (1964), requiring absolute maintenance of patient anonymity and confidentiality. 3.Risk Mitigation——Public dissemination could compromise privacy in this retrospective study involving venous invasion grading—a high-sensitivity oncological parameter. This safeguards against legal/ethical breaches such as unauthorized secondary data exploitation. The de-identified datasets generated during this study are not publicly deposited due to stringent patient privacy protections mandated by Chinese regulations (Personal Information Protection Law, PIPL) and institutional ethical guidelines (§3.2 of Beijing Chaoyang Hospital Ethics Charter). However, qualified researchers may request access by contacting the corresponding author, Dr. Shaocheng Lyu, *via* email at shaocheng0502@163.com.

Declarations

Competing interests

The authors declare no competing interests.

Informed consent

All study participants, or their legal guardian, provided informed written consent prior to study enrollment.

Institutional review board statement

This study was approved by the Ethics Committee of Beijing Chaoyang Hospital (acceptance number: 2024-D-511). All procedures in this study involving human partici pants were performed in accordance with the ethical standards of the institutional research committee and the 1964 Helsinki Declaration.