Establishment and validation of a prognostic model for pancreatic cancer utilizing genes of tumor-associated neutrophils

Abstract

Background

Dysregulation in chemotaxis and activation of neutrophils may trigger cancer. Nevertheless, the function of neutrophils and their mechanism during the prognosis of pancreatic cancer (PC) remain unclear.

Methods

From the databases of The Cancer Genome Atlas (TGCA) and GeneCards, the genes of tumor-associated neutrophils (TANs) were screened out leveraging the differential expression analysis. The constructed prognostic model for PC was analyzed through the least absolute shrinkage and selection operator (LASSO) regression and Cox univariate and multivariate regression. The dataset (GSE62452) provided by the Gene Expression Omnibus (GEO) database served as the validation cohort, and its potential mechanistic pathways and biological functions were analyzed leveraging Kyoto Encyclopedia of Genes and Genomes (KEGG) and Gene Ontology (GO). The immune infiltration analysis, as well as the drug sensitivity prediction, were conducted. The gene expression in the prognostic model was checked through online databases, including Human Protein Atlas (HPA) and Tumor Immune Single-Cell Hub (TISCH). Finally, the genes were experimentally confirmed by immunohistochemistry (IHC) and qPCR in combination with clinical samples from patients with PC.

Results

AIM2, PSCA, IL18BP, and LIPE were four genes of TANs closely linked to the prognosis of PC. A model for the prognosis of PC was established utilizing these 4 genes. The AUC (area under the receiver operating characteristic curve (ROC curve)) values of this model for forecasting the survival rates of patients at 1, 2, and 3 years were 0.774, 0.841, and 0.953, respectively. The immune infiltration analysis revealed that resting dendritic cells (resting DCs), naive B cells, CD8T cells, as well as follicular helper T cells, were highly infiltrated among patients suffering from PC. According to the drug sensitivity analysis, the high-risk pancreatic ductal adenocarcinoma (PAAD) group had a significantly higher inhibitory concentration 50 (IC50) for dactolisib, docetaxel, gemcitabine, and ulixertinib compared to the group by patients with low-risk PAAD. The results of IHC and qPCR confirmed those of bioinformatics analysis.

Conclusion

New TANs-related biomarkers have been found that effectively forecast the prognosis of patients suffering from PAAD.

Supplementary Information

The online version contains supplementary material available at 10.1007/s12672-026-05432-z.

Introduction

The incidence rate and death rate of pancreatic cancer (PC) in recent years have climbed around the globe, and pancreatic ductal adenocarcinoma (PAAD) constitutes roughly 95% of the rates of incidence and death [1]. Statistics from the American Cancer Society (ACS) in 2024 manifest that among the malignant tumors statistically analyzed, PC ranks 10th and 3rd in terms of incidence rate and mortality rate, respectively [2]. The overall 1-year and 5-year mortality rates are 24% and 6%, respectively [3]. Additionally, PC has an insidious onset and is highly prone to metastases in the early stage. For most patients with PC at initial diagnosis, the disease has progressed to locally advanced disease or developed distant metastases [4], and only 10%~20% of them qualify for surgery [5]. Due to the extremely high malignancy of PC, most patients are confronted with a high risk of recurrence. Targeted therapies represented by olaparib and chemotherapy have to some extent prolonged the survival of patients in early and intermediate stages [6, 7]. A study by Gaonan Tian et al. demonstrated that the ARRB2 gene is significantly associated with prognosis in pancreatic cancer patients, with low expression indicating a poorer prognosis [8].Nevertheless, for the majority of patients with advanced PC, the benefits of palliative therapy remain limited [6]. Therefore, it has become urgent to explore new indicators for the prognosis of PC.

As short-lived granulocytes originating in hematopoietic stem cells, becoming mature within the bone marrow, and eventually entering the bloodstream, neutrophils play a pivotal role in the body’s innate immune system [9]. Neutrophils have a high extravasation ability and recruit dendritic cells and T cells by secreting chemokines such as CCL3 and CXCL12 [10]. Within the tumor microenvironment (TME), neutrophils release arginase to inhibit lymphocyte proliferation, eventually leading to an immunosuppressive state [9]. In tumors, neutrophils have a prolonged lifespan and exhibit functions of promoting tumors, activating typical molecular pathways, comprising vascular endothelial growth factor (VEGF) signaling, to facilitate angiogenesis [11]. Neutrophils, in tumors, also secrete matrix metalloproteinases (MMPs) to remodel the matrix, facilitating tumor invasion and metastasis [12]. Clinical studies have demonstrated that the neutrophil/lymphocyte ratio (NLR), serving as an indicator for systemic inflammatory responses, is significantly associated with poor prognoses of patients suffering from breast cancer [13], suggesting its pivotal role in cancer.

Although genes of tumor-associated neutrophils (TANs) have been studied in the prognoses of various tumors, their role and mechanism in the prognosis of PC have not been specifically elucidated. Currently, systemic therapy for advanced PC mainly relies on chemotherapy, and targeted therapy is only applicable to a few patients with specific molecular subtypes. It is an urgent need to discover new genes for the prognosis of PC to assist in early diagnosis, prognosis, and treatment of this disease. Therefore, based on the deficiencies of previous studies and the severe situation, the genes of TANs related to the prognosis of PC have been screened out through the methods of bioinformatics, and a prognostic model has been established. This study may provide further evidence for exploring the pathogenesis of PC, forecasting clinical prognoses, and exploring and discovering new chemotherapy and targeted medications. Figure 1 below shows the technical roadmap.

Fig. 1.

The technical roadmap

Methods and materials

Data collection and analysis

The clinical data and coding genes of 178 participants suffering from PAAD were collected from The Cancer Genome Atlas (TCGA) database (https://portal.gdc.cancer.gov), including 178 samples of PAAD and 4 samples of non-cancer adjacency. A total of 3,352 TANs were derived from the GeneCards database (https://www.genecards.org/). Genes were retrieved from GeneCards using the search term ‘tumor-associated neutrophils.’ Only genes with a GeneCards Relevance Score > 1.0 were retained, as this threshold reflects a minimum level of curated evidence for gene-disease association. Then, after the intersection of the above two datasets, the R package edgeR was employed for investigating the differences of PAAD with TANs, and a p < 0.05 and a |log2 (FC)| ≥1 indicated statistical significance. Moreover, an external validation set (GSE62452) was provided by the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/), including 65 samples of PAAD. Volcano plots and heatmaps were produced leveraging the R package ggplot2 to visualize the TANs differentially expressed in PAAD.

Establishment of the model for risk assessment

The participants suffering from PC were randomly classified into a training set and a validation set at a 1:1 ratio. Univariate Cox regression was first applied to identify genes significantly associated with overall survival (p < 0.05). The resulting candidate genes were then subjected to LASSO-penalized Cox regression using the R package glmnet. To prevent overfitting, 10-fold cross-validation was performed, and the optimal regularization parameter lambda was selected as lambda.min (the value minimizing cross-validated partial likelihood deviance). Genes with non-zero coefficients at the optimal lambda were retained for the final multivariate Cox model. A model for risk assessment was built using the prognostic genes of TANs, and the formula was as follows: (β1×LIPE expression) + (β2×AIM2 expression) + (β3×PSCA expression) + (β4×IL18BP expression). Subsequently, a prognostic model was established based on the expression levels of the genes of these TANs as well as the regression coefficients of them. According to the risk median, the participants were separated into 2 groups: a high-risk PAAD group and a low-risk PAAD group. The 2 R packages, survival and survminer, assisted in evaluating the survival discrepancies between these 2 groups through the Kaplan-Meier survival curve. Additionally, the receiver operating characteristic curve (ROC curve) was produced leveraging the R package timeROC to examine this prognostic model’s performance.

Verification of the model for risk assessment

To additionally confirm the predictive accuracy of this prognostic model utilizing the genes of TANs, the dataset (GSE62452) cohort was selected as an external validation set. Per-patient risk score was obtained leveraging the pre-established formula. Then, according to the risk median, the participants were separated into 2 groups: high-risk PAAD participants and low-risk PAAD participants. Subsequently, the discrepancies in OS between these 2 groups were analyzed through the Kaplan-Meier survival curve, with the log-rank test evaluating the statistical significance. Moreover, time-dependent ROC curves of survival predictions for patients at 1, 3, and 5 years were produced to compute the AUC (area under the ROC curve) values for quantifying the discrimination ability of this prognostic model at different years. Therefore, the robustness and clinical utility of this prognostic model were evaluated from a comprehensive perspective.

Nomogram construction

The risk scores and other clinical pathological factors were combined, and the analyses of Cox univariate and multivariate regression were conducted in the training set of PAAD. Through the analyses, the independent prognostic predictors were determined according to the risk scores derived from the clinical pathological factors, comprising age, sex, and tumor-node-metastasis (TNM) stage. The nomogram was produced utilizing the R package, rms. Then, discrimination and calibration of the nomogram were assessed across the entire PAAD dataset using ROC curves and calibration curves, with calibration curves adopted to analyze and evaluate the clinical utility of the nomogram.

Enrichment analysis

To investigate the latent biological mechanisms and roles of the selected TANs in PAAD, the enrichment analysis on these TANs was carried out leveraging Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG). Pathways or functions demonstrating a corrected p < 0.05 were defined as significant enrichment. The R package clusterProfiler assisted in performing the functional enrichment analysis.

Immune infiltration analysis

To examine the prevalence of immune cells of various types in high-risk and low-risk populations, the proportions and infiltration levels of different immune cells were estimated utilizing the CIBERSORT (https://cibersort.stanford.edu/). CIBERSORT was run in absolute mode with 1,000 permutations; a significance threshold of p < 0.05 was applied to filter reliable deconvolution results. Additionally, the Pearson correlation analysis was implemented, and differences between the 2 groups were visualized through the R packages, including ggplot2, ggExtra, and ggpubr.

Drug sensitivity analysis

The R package oncoPredict aided in calculating the half-maximal inhibitory concentration 50 (IC50) for anti-cancer medications and the expected response of patients with PAAD to anti-cancer medications. To assess the differences in IC50 of drugs for conventional chemotherapy or targeted drugs between the 2 groups, the Pearson correlation analysis was carried out between the genes of TANs and these drugs. The genes or drugs with a p < 0.05 were screened out. Additionally, the R package ggplot2 was adopted to visualize the results.

Validation of prognostic genes

The database, Tumor Immune Single Cell Hub (TISCH) (http://tisch.comp-genomics.org), was utilized for comprehensively analyzing the heterogeneity of the TME, including various datasets and cell types. Therefore, the dataset of single-cell RNA sequencing (CRA001160) provided by the TISCH database was adopted to analyze the expression levels of the identified genes of TANs. The immunohistochemistry (IHC) data and the Human Protein Atlas (HPA) (https://www.proteinatlas.org/) were cross-compared to provisionally determine the protein expression of the prognostic genes in PAAD tissues.

In vitro verification of the genes for the prognostic model for PC

Source of the samples of PC tissues

All pancreatic cancer tissue samples in this study were clinical specimens surgically resected at Yueyang Hospital Affiliated to Hunan Normal University between January 2022 and December 2024. A total of 12 pancreatic cancer tissue samples were included in the study, all of which obtained informed consent from the subjects or their families. Sample collection was approved by the Ethics Committee of Yueyang Hospital Affiliated to Hunan Normal University (approval number: 2025001).

Inclusion and exclusion criteria

Inclusion criteria

1. Patients satisfying the diagnosis criteria for PC and diagnosed as PC by postoperative pathological diagnosis;

2. Patients who had not received any systemic anti-tumor treatment before the operation, including chemotherapy, radiotherapy, and targeted immunotherapy;

3. Patients without a medical history of other malignant tumors;

4. Patients with complete clinicopathological features.

Exclusion criteria

1. Patients suffering from PC with metastases to other parts of the body or having a history of malignant tumors in other parts of the body;

2. Patients who had received systemic anti-tumor treatment before the operation, including chemotherapy, radiotherapy, and targeted immunotherapy;

3.  Patients whose postoperative pathology revealed benign lesions such as inflammatory masses or chronic pancreatitis were excluded;

4.  Patients with incomplete clinicopathological features.

IHC results of paraffin sections

The IHC results were semi-quantitatively evaluated based on the staining intensity and staining range within each field of view. (1) Staining intensity: no staining (0 point), light yellow staining (1 point), brownish yellow staining (2 points), and brown staining (3 points). (2) Staining range: 0%~25% (1 point), 26%~50% (2 points), 51%~75% (3 points), and > 75% (4 points). The final score was calculated as the sum of the staining intensity score and the staining range score. The expression level was determined as follows: A total score of < 4 was considered low level, and high level was described by a total score of ≥ 4.

RNA extraction and real-time fluorescent quantitative PCR

RNA was extracted from pancreatic cancer and adjacent non-cancerous tissues. Reverse transcription was performed using a qPCR RT kit. The human GAPDH gene served as the reference gene, with primers as shown in Table 1. Relative expression values of TANs were calculated using the 2^−ΔΔCt method.

Table 1.

The primers used in qRT-PCR of the TANs risk model genes

Genes	Forward primer sequence	Reverse Primer sequence
hAIM2	TGGCAAAACGTCTTCAGGAGG	AGCTTGACTTAGTGGCTTTGG
hIL18BP	ATGAGACACAACTGGACACCA	GCCAGGTCACTTCCAATGC
hLIPE	TCAGTGTCTAGGTCAGACTGG	AGGCTTCTGTTGGGTATTGGA
hPSCA	CCTAACGCAAGTCTGACCATGTATG	TGCAGGCGGATCTGTGTCAATA
hGAPDH	CCATGGGGAAGGTGAAGGTC	GAAGGGGTCATTGATGGCAAC

Statistical analysis

The R (version 4.3.3) and the SPSS (version 26) were applied to the statistical analysis. OS comparison was conducted leveraging the Kaplan-Meier survival analysis, with the significance assessed through the log-rank test. The discrepancies among categorical variables were analyzed leveraging the chi-square test, while the t-test was utilized for analyzing the differences among continuous variables. If the sample size was no more than 40 or the theoretical frequency was below 1, the Fisher’s exact test was adopted. ImageJ (version 1.54) was employed to process the IHC images and compute the average value of integrated optical density (IOD). GraphPad Prism (version 10) was used for plotting, with p-values of < 0.05, < 0.01, < 0.001, and < 0.0001 denoted by *, **, ***, and****, respectively. In all statistical analyses, the p-value of < 0.05 (two-tailed test) was statistically significant.

Results

Identification of the differentially expressed genes (DEGs) of TANs

The clinical data of 178 patients suffering from PAAD from the TCGA database and the encoding genes of TANs were obtained. The data of gene expression of PAAD samples provided by the TCGA database and normal tissues were examined, with 2912 hotspot genes screened out. In the GeneCards database, 3,352 genes of TANs in total were identified, and 696 DEGs were screened out, comprising 50 up-regulated genes and 646 down-regulated genes (Fig. 2A and B).

Fig. 2.

Characterization of differentially expressed TANs. A Volcano plot of DEGs in PAAD. B Venn diagram of DEGs and TANs

Establishment of the prognostic model

In total, 178 patients suffering from PAAD were at random divided into 2 groups at a 1:1 ratio, with 89 patients in each group (the training set vs. the validation set). As illustrated in Fig. 3A–D, the analysis of Cox univariate regression was conducted on the DEGs of TANs, with 27 of these genes related to prognosis. Through LASSO regression, 6 of these 27 genes related to prognosis were further identified, and ultimately, 4 best prognostic genes, including LIPE, AIM2, PSCA, and IL18BP, were determined through the analysis of Cox multivariate regression. These 4 genes were used to establish the model for risk prediction. The risk score was computed with the formula as follows: (-0.76252×LIPE expression) + (0.23993×AIM2 expression) + (0.17740×PSCA expression) + (0.33835×IL18BP expression). The risk score was obtained in the training set leveraging the formula above, and based on the risk median, participants were separated into a group by high-risk PAAD and a group by low-risk PAAD. In the training set, the Kaplan-Meier survival curve suggested that participants with a higher risk score had a significantly lower survival rate, as illustrated in Fig. 3C. The expression of 4 prognostic genes of TANs in the 2 groups was described using box plots and heatmaps (Fig. 3D and E). The AUCs of OS at 1, 2, and 3 years were 0.774, 0.841, and 0.953, respectively (Fig. 3F). Additionally, the risk plots indicated a good association of the rise in risk score with the survival of participants with PC (Fig. 3G and H). These results demonstrated that in participants with a lower risk score, LIPE increased. In contrast, AIM2, PSCA, and IL18BP levels were higher in the group by patients with high-risk PAAD.

Fig. 3.

Construction of risk model risk model based on the prognostic TANs in PAAD. A Univariate Cox regression analysis of the TANs. B multivariate Cox regression analysis of the TANs. C K-M curves of survival outcome between high- and low-risk score groups in the training set. D The boxplot of the prognostic TANs in the training set. E The heatmaps of the prognostic TANs in the training set. F ROC curves for survival predictive performance at 1-, 2-, and 3-year in the training set. G, H The risk score distribution and survival status of PAAD patients in the training set

Subsequently, the predictive precision of the risk model was confirmed in the validation set. As shown in Fig. 4A, B and a significant difference in the rate of OS was noted between these 2 groups. The AUCs of the OS at 1, 2, and 3 years were 0.767, 0.763, and 0.834, respectively, suggesting that the risk model utilizing the genes of TANs had a good predictive accuracy. Figure 4C–F display risk distribution maps, box plots of gene expression, and risk heatmaps between the two validation-set groups. AIM2, IL18BP, and LIPE levels in the low-risk group were significantly higher than those in the high-risk group.

Fig. 4.

Construction of risk model risk model based on the prognostic TANs in PAAD. A K-M curves of survival outcome between high- and low-risk score groups in the testing set. B ROC curves for survival predictive performance at 1-, 2-, and 3-year in the testing set. C, D The risk score distribution and survival status of PAAD patients in the testing set. E The boxplot of the prognostic TANs in the testing set. F The risk heatmap of the prognostic TANs in the testing set. G Univariate Cox analysis and H multivariate Cox analysis shows the correlation of OS and risk score, and clinicopathological factors

Independent analysis of the prognostic features of the genes of TANs

The analyses of Cox univariate and multivariate regression of risk scores and clinical features were performed in the TCGA-PAAD dataset to validate the independence of the prognosis features related to the genes of TANs in PAAD. As shown in Fig. 4G and H, the analyses of univariate and multivariate independent prognosis suggested that the risk score might be an independent predictor (p < 0.001). Whereafter, a new nomogram was established according to clinically pathological features and risk scores to forecast the survival probability of patients suffering from PAAD at 1, 2, 3, 4, and 5 years (Fig. 5A). This nomogram exhibited a good predictive performance. Figure 5B shows the calibration curve of this nomogram. The stratified subgroup analyses were implemented to examine the prognostic value of the risk score using the genes of TANs under different clinical pathological conditions. Additionally, as shown in Fig. 5C–G, the higher risk scores were not significantly correlated with the clinically adverse pathological features, comprising sex, age, lymph node metastasis, tumor stage (T stage), as well as distant metastasis. The Kaplan-Meier survival curve is shown in Fig. 5H. These results demonstrated that the risk score obtained from the genes of TANs might stand as an independent indicator for the prognosis of patients suffering from PAAD, effectively forecasting the OS rate of patients with PAAD.

Fig. 5.

The independence of the TANs prognostic signatures for PAAD. A A nomogram based on the TANs and clinicopathological factors. B Calibration curve of the nomogram. The box plots of correlation between risk scores and clinicopathological characteristics, including C Age; D Sex; E T stage; F N stage; G M stage. H The KM curve of clinicopathological characteristics

Verification of the risk model using the GEO dataset

To examine the predictive accuracy of the risk model and verify its reliability, the same method was applied to the GEO dataset (GSE62452) cohort that was serving as an external validation set. The analysis of Cox multivariate regression of these two cohorts showed that the prognostic features of the genes of TANs were independent (Fig. 6A). The survival analysis suggested that participants with low risk scores had significantly higher OS rates than those with high risk scores (Fig. 6B), in line with the findings in the TCGA dataset. In the GSE62452 dataset, the AUCs of survival prediction at 1, 2, and 3 years were 0.657, 0.622, and 0.774, respectively (Fig. 6C). Figure 6D–G visually display the distribution of the risk scores for the patients and the expression levels of 4 genes in the external validation set. The results demonstrated that the risk model performed well in predicting the prognosis of patients with PAAD.

Fig. 6.

Validation of risk model based on GEO cohort. A Multivariate Cox regression analysis in GEO cohort. B K-M curve of survival between high- and low-risk score groups. C The AUC of the risk model in the GSE62452. D, E The risk score distribution and survival status of PAAD patients in the GSE62452. F The boxplots of the prognostic TANs in the GSE62452. G The risk heatmap of the prognostic TANs in the GSE62452. Functional enrichment analysis of the differentially expressed TANs in the different risk groups. H The results of GO enrichment analysis. I The results of KEGG enrichment analysis

Functional enrichment analysis of the DEGs of TANs

The functional enrichment analysis was carried out to investigate and reveal the potential biological functions and the underlying mechanisms of the genes of TANs. The enrichment analysis through GO indicated that the DEGs were highly enriched in biological processes, comprising monocyte differentiation, leukocyte migration, and leukocyte-mediated immunity, and in cellular components, comprising the extracellular side of the plasma membrane, membrane rafts, and membrane microdomains. The DEGs also demonstrated high enrichment in molecular functions, including binding of cytokine receptors, cytokine activity, and cytokine binding (Fig. 6H). Moreover, the enrichment analysis by KEGG demonstrated that the DEGs were highly enriched in interactions of cytokines with cytokine receptors, cytotoxicity mediated by natural killer (NK) cells, as well as interactions of viral proteins with cytokines and their receptors (Fig. 6I). The findings suggested that the genes of TANs might be crucial in the metastases of PAAD.

Immune infiltration analysis

The CIBERSORT algorithm assisted in evaluating the infiltration pattern of immune cells in patients with PAAD, focusing on the 2 subgroups by high and low risks. The CIBERSORT analysis revealed that the original B cells and CD8 T cells were highly infiltrated in the group by patients with low-risk PAAD, and dendritic cells were highly infiltrated in the group by patients with high-risk PAAD (Fig. 7A). The dumbbell diagram was established for exploring the association of the single gene of TANs with immune cells. The diagram indicated that naive B cells and memory B cells showed a positive correlation with AIM2, and γ-δ T cells and follicular helper T cells were negatively correlated with AIM2. T cell regulators manifested a positive correlation with IL18BP, and γ-δ T cells, follicular helper T cells, as well as CD4 memory T cells, were negatively correlated with IL18BP. CD8 T cells and naive B cells demonstrated a positive correlation with LIPE, and eosinophils and dendritic cells showed a negative correlation with LIPE. Plasma cells and activated NK cells were positively correlated with PSCA, and CD4 memory activated T cells and resting NK cells showed a negative correlation with PSCA (Fig. 7B–E). The correlation graphs were generated to verify the correlation of immune cells, comprising naive B cells, resting dendritic cells (resting DCs), CD8 cells, and follicular helper T cells, with different risk scores (Fig. 7F–I). In general, the results indicated that the risk model using the genes of TANs was correlated with the immune infiltration of patients with PAAD.

Fig. 7.

Immune infiltration analysis of PAAD patients in risk groups. A Box diagram of immune infiltration. Prognostic gene and immune cell dumbbell map. B AIM2 and immune cell dumbbell diagram. C IL18BP and immune cell dumbbell diagram. D LIPE and immune cell dumbbell diagram. E PSCA and immune cell dumbbell diagram. Graphs of correlation between immune cells and risk score. F The graph of correlation between B. cells. naive and risk score. G The graph of correlation between Dendritic cells resting and risk score. H The graph of correlation between T.cells.CD8 and risk score. I The graph of correlation between T. cells. follicular. helper and risk score

Drug sensitivity analysis

The drug sensitivity analysis was carried out to evaluate the responses of the 2 subgroups to chemotherapy and targeted drugs. During the analysis, a total of 198 small-molecule compounds with significantly different responses were identified among the 2 subgroups. Among the top 6 compounds with the lowest p-values, 4 — dactolisib, docetaxel, gemcitabine, and ulixertinib — showed the most significant differences between the two subgroups. Among the 2 subgroups (Fig. 8A–F). These results indicated that the IC50 values of patients suffering from high-risk PAAD for dactolisib, docetaxel, gemcitabine, and ulixertinib were considerably higher compared to those of patients with low-risk PAAD. The correlation analyses of prognostic genes with chemotherapy drugs revealed that the prognostic genes of TANs were correlated with these 4 drugs (Supplementary Fig. S1–S2). Based on the findings above, these 4 small-molecule compounds showed potential for treating PAAD in the preliminary study. Nevertheless, their efficacy and safety need to be verified through further research. This study tries to offer some exploratory elicitation of a molecular targeted chemotherapy to patients with PAAD.

Fig. 8.

Analysis of drug sensitivity of PAAD patients in the different risk groups. A Dactolisib_1057 (B) Docetaxel_1007 (C) Docetaxel_1819 (D) Gemcitabine_1190 (E) Ulixertinib_1908 (F) Ulixertinib_2047

Verification of the prognostic genes of TANs

According to the data provided by the HPA website, in the risk model, AIM2 was expressed lowly in the PC tissues and expressed highly in the adjacent-to-tumor tissues, LIPE was expressed moderately in the PC tissues and expressed lowly in the adjacent-to-tumor tissues, IL18BP was expressed lowly in both the PC tissues and the adjacent-to-tumor tissues, and PSCA was expressed lowly or unmarked in the PC tissues and the adjacent-to-tumor tissues (Fig. 9A–H). Additionally, the dataset of single-cell RNA sequencing (CRA001160) provided by the TISCH database was leveraged to analyze the expression of the 4 prognostic genes of TANs in PAAD. This dataset contained 32 cell clusters and intermediate cells of 12 types. Figure 9I–Q present the proportions and expression levels of these 4 genes, AIM2, IL18BP, LIPE, and PSCA, in cells of each type in the dataset. The results demonstrated that the prognostic genes of TANs might be crucial in the TME of PAAD.

Fig. 9.

Validation of prognostic genes of TANs. Based on the HPA database, the immunohistochemical results for A, B AIM2, C, D IL18BP, E, F LIPE and G, H PSCA. I Annotations for all cell types in CRA001160 and the proportion of each cell type. J, L, N, P Proportions of AIM2, IL18BP, ILPE and PSCA. K, M, O, Q Expressions of AIM2, IL18BP, ILPE and PSCA

Clinicopathological features of the patients

To additionally validate the reliability of the genes of TANs used for the model, their expression levels were verified. A cohort of 12 eligible participants was enrolled, comprising 6 males (average age: 60.8 years) and 6 females (average age: 74.5 years). Regarding tumor location, 8 patients had tumors located at the head of the pancreas, and 4 at the tail of the pancreas. As for the degree of differentiation, 11 participants had moderately differentiated PC, and 1 had well-differentiated PC. Based on the TNM stage from the American Joint Committee on Cancer (AJCC), 10 participants were in stage I and II, while 2 were in stage III and IV. The clinicopathological features of the participants are illustrated in Table 2.

Table 2.

Clinicopathological Features of the Participants

No.	Sex	Age	Tumor location (head or tail of pancreas)	Degree of differentiation (moderately or well differentiated)	T stage	N stage	M stage	Stage
1	Female	79	Head	moderately	3	0	1	Ⅳ
2	Female	88	Head	moderately	3	0	0	ⅡA
3	Male	59	Head	moderately	3	0	0	ⅡA
4	Male	79	Head	well	1	0	0	ⅠA
5	Male	38	Head	moderately	4	0	0	Ⅲ
6	Female	77	Head	moderately	1	0	0	ⅠA
7	Female	66	Tail	moderately	3	1	0	ⅡB
8	Female	60	Tail	moderately	2	0	0	ⅠB
9	Male	51	Head	moderately	3	0	0	ⅡA
10	Male	63	Tail	moderately	2	0	0	ⅠB
11	Male	75	Tail	moderately	2	0	0	ⅠB
12	Female	77	Head	moderately	1	0	0	ⅠA

Expression of the genes of TANs used for the model in clinical PC specimens

AIM2 was highly expressed in 4 cases of PC tissues and 10 cases of adjacent-to-tumor tissues (p < 0.05). IL18BP demonstrated high levels in 3 cases of PC tissues and 9 cases of adjacent-to-tumor tissues (p < 0.05). LIPE was highly expressed in 3 cases of PC tissues and 9 cases of adjacent-to-tumor tissues (p < 0.05). PSCA showed high levels in 6 cases of PC tissues and 1 case of adjacent-to-tumor tissues (p > 0.05). The expression levels of these 4 genes of TANs for the model in PC tissues and adjacent-to-tumor tissues are shown in Table 3. The quantitative analysis was performed to further clarify the expression of the 4 genes in pancreatic tissues. AIM2 expression in PC tissues was lower than that in adjacent-to-tumor tissues (p < 0.05), IL18BP in PC tissues demonstrated a lower expression level compared to that in adjacent-to-tumor tissues (p < 0.01), LIPE in PC tissues was lower in comparison to that in adjacent-to-tumor tissues (p < 0.001), and PSCA showed higher expression in PC tissues compared to that in adjacent-to-tumor tissues (p < 0.001). The expression results of IHC are shown in Fig. 10A–H, and the results of the analyses utilizing ImageJ are shown in Fig. 10I–L. The qPCR results are shown in Fig. 10M–P. The findings are consistent with our previous bioinformatics analyses.

Table 3.

Expression of the genes of TANs used for the model in pancreatic cancer tissues and adjacent-to-tumor tissues

Genes and expression level		AIM2 expression level		IL18BP expression level		LIPE expression level		PSCA expression level
		High	Low	High	Low	High	Low	High	Low
Category	Pancreatic Cancer Tissues	4	8	3	9	3	9	6	6
	Adjacent-to-Tumor Tissues	10	2	9	3	9	3	1	11
χ²		6.171		6.000		6.000		5.042
p		0.036*		0.039*		0.039*		0.069

A p-value of < 0.05 demonstrates a difference with statistical significance

Fig. 10.

Validation of prognostic genes of TANs. The expressions of A, B AIM2, C, D IL18BP, E, F LIPE and G, H PSCA in PAAD and para-cancer tissues. The results of the analyses utilizing ImageJ. I AIM2, J IL18BP, K LIPE, L PSCA. The qPCR results. M AIM2, N IL18BP, O LIPE, P PSCA

Discussion

The incidence of PC continuously increases worldwide, becoming a severe public health concern. Currently, its age-standardized incidence rate is approximately 5.7 per 100,000 [14]. Common driver genes include KRAS, TP53, CDKN2A, and SMAD4, among whom KRAS mutations are found in over 90% of PC cases, continuously activating the MAPK signaling pathway and facilitating the proliferation and survival of cells; TP53 mutations cause a dysregulation of the cycle control of cells and an impaired function of apoptosis; CDKN2A inactivation and SMAD4 deficiency respectively affect the cell cycle and TGF-β signaling pathways [15–18]. A study by Semih Latif Ipek et al. found that rBlmet inhibits the growth of Mia-PaCa-2 cells [19]. Additionally, Rangaraj Kaviyaprabha et al. have demonstrated the potential of hesperetin and emodin as a combination therapy for inhibiting the progression of pancreatic cancer through the C-Met gene [20]. Previous research has suggested that TANs accelerate the invasion and metastasis of tumors, but their role in PC remains unclear [21, 22]. A prognostic model was built in this study for PAAD leveraging the 4 genes of TANs, comprising AIM2, IL18BP, PSCA, and LIPE. The results demonstrated that PAAD might be related to cytokine interactions and cytotoxicity mediated by NK cells, and the genes of TANs used for this model might affect the prognosis of PAAD through epithelial-mesenchymal transition (EMT) and tumor immune escape. The risk model could effectively distinguish the individuals sensitive to chemotherapy, and changes in subsets of immune cells might provide guidelines for immunotherapy. Pancreatic cancer is a molecularly heterogeneous malignancy driven by complex oncogenic mechanisms, and the four-gene TAN-based prognostic model constructed in this study provides a novel tool for predicting patient outcomes in PAAD.

PSCA, a GPI-anchored glycoprotein involved in cell proliferation and immune signaling, is upregulated in pancreatic cancer tissues. PSCA participates in cell signal transduction, cell cycle regulation, cell proliferation, and immune response, and may disrupt the balance between pro-inflammatory and anti-inflammatory signals [23]. In this study, it was found that PSCA was upregulated in PC tissues. Previous IHC analyses have revealed that PSCA is weakly expressed in benign prostatic hyperplasia and low-grade intraepithelial neoplasia of the prostate but is significantly elevated in high-grade lesions and advanced PC. PSCA may promote tumor progression by activating the PI3K/AKT pathway to upregulate c-Myc [24, 25]. Additionally, PSCA is distributed in the cytoplasm in some breast cancer tissues and is significantly correlated with Her2/neu overexpression [26]. PSCA is also upregulated in various brain tumors, suggesting it may be a potential treatment target. Nonetheless, in the central nervous system, PSCA requires cautious evaluation [27].

IL18BP, a secreted inhibitor of the pro-inflammatory cytokine IL-18, is downregulated in pancreatic cancer and its reduced expression may facilitate immune escape by unleashing downstream inflammatory and checkpoint pathways. IL18BP binds and inhibits the activity of IL-18 to maintain immune balance. In this study, IL18BP was downregulated in PC tissues, in line with the previous findings of the low expression level of IL18BP in patients suffering from head and neck squamous cell carcinoma and breast cancer [28, 29]. This low-level expression may weaken the inhibition to IL-18, triggering the excessive activation of downstream pathways like NF-κB and MAPK, accelerating immune escape [30, 31]. Moreover, the low-level expression of IL18BP in PC may enhance the activity of NLRP3 inflammasome or regulate the PD-1/PD-L1 pathway, thereby inhibiting the functions of immune cells and eventually accelerating the PC progression [32].

AIM2, an intracellular DNA sensor that assembles the inflammasome to trigger pyroptosis, is downregulated in pancreatic cancer. AIM2 recognizes abnormal double-stranded DNA and assembles an inflammasome to facilitate IL-1β and IL-18 secretion and induce pyroptosis, thereby participating in immune responses [33]. In this study, AIM2 was downregulated in PC tissues. Previous research has revealed that the function of AIM2 varies in different cancers. Specifically, AIM2 inhibits colon cancer and liver cancer [34, 35] but accelerates the tumor growth and invasion in skin squamous cell carcinoma [35]. In terms of liver cancer, the low-level expression of AIM2 is linked to poor prognoses. The overexpression of AIM2 inhibits metastases and accelerates cell apoptosis, which is possibly through the regulation for the Notch signaling or the inhibition to the mTOR-S6K1 pathway [36, 37]. In colorectal cancer, low-level expression of AIM2 is related to a raised risk of recurrence, possibly involving the activation of the inflammasome and the cell death mediated by Caspase-1 [38]. As for PC, low-level expression of AIM2 may inhibit pyroptosis and immune surveillance, thereby weakening the killing effect of AIM2 on tumor cells and eventually accelerating the progression of pancreatic tumors [39].

LIPE, a key lipolytic enzyme involved in metabolic reprogramming and tumor microenvironment shaping, is downregulated in pancreatic cancer tissues, playing a significant role in energy balance, weight regulation, and cholesterol metabolism [40, 41]. LIPE has potential in metabolic reprogramming of tumors, fatty acid supply, TME shaping, and immune escape [42]. This study revealed that LIPE was downregulated in PC tissues. Previous research has shown that LIPE is lowly expressed in breast cancer tissues [43]. In head and neck squamous cell carcinoma, high LIPE expression is associated with a higher survival rate and TP53 mutation rate [42]. In terms of cervical cancer, the high-level expression of LIPE is considerably linked to tumor volume and distant metastasis, possibly by inhibiting the miR-195-5p/MAPK pathway, thereby accelerating the progression of cervical tumor [44]. These differences may be related to tumor heterogeneity, sample source, and detection methods, suggesting that further studies on the specific mechanism of LIPE in PC are needed.

The infiltration of naive B cells among tumors is influenced by the tumor type, immune microenvironment, and mechanisms for immune escape, and generally, the infiltration is not significant [45]. During tumor progression, the proportion of naive B cells can increase from early to late stages. Naive B cells account for a higher proportion in liver cancer, which is associated with a better prognosis, suggesting that naive B cells may be an important regulator of inflammatory response and a potential biomarker in the TEM [46]. CD8 T cells stand as important anti-tumor effector cells, and their infiltration degree varies depending on the type of tumor and individual differences. Some research has suggested that high infiltration of CD8 T cells is connected with longer OS and progression-free survival among patients suffering from cancers such as esophageal cancer [47, 48]. Resting DCs have a lower infiltration in tumors, but they activate T cells through antigen presentation, thereby initiating anti-tumor immune responses, and their high infiltration is associated with a higher survival rate [49]. Follicular helper T cells are crucial for the differentiation of B cells, inducing antibody responses, and activating CD8 T cells. Some studies have revealed that the high infiltration of follicular helper T cells is linked to better prognosis among patients with rectal cancer [50].

The enrichment analyses through GO and KEGG were carried out on the genes of TANs. The results suggested that the effects of the genes of TANs in PAAD might result from regulating the interactions of cytokines with their receptors, the cytotoxicity mediated by NK cells, and the interactions between viral proteins and cytokines, as well as their receptors. Previous studies have shown that fatty acid binding protein 5 (FABP5) activates the interactions among cytokine receptors and the signaling pathways for IL-17 and TNF to accelerate the proliferation, invasion, and carcinogenicity of gastric cancer cells [51]. Standing as a crucial component of the innate immune system, NK cells can identify and kill tumor cells, and the activity of NK cells is closely connected with preventing the onset and metastases of tumors [52, 53]. Furthermore, the interactions between viral proteins and cytokines, as well as their receptors, regulate the immune responses of the host to affect the occurrence and progression of tumors [54].

The sensitivity analysis of chemotherapy drugs for PC in this study provides new theoretical support for the clinical drug treatment for PC. Dactolisib is a small-molecule drug that selectively inhibits the PI3K/Akt/mTOR pathway [55]. Dactolisib selectively inhibits the PI3K/Akt/mTOR pathway, thereby effectively inhibiting the proliferation of PC cells. Standing as an antimetabolite drug, gemcitabine inhibits the proliferation of cancer cells and is widely used for PC treatment, especially for advanced or metastatic PC [56]. In the treatment of PC, docetaxel is used as an adjuvant chemotherapy drug, usually combined with drugs such as gemcitabine to enhance the therapeutic effect. A study has shown that docetaxel effectively slows down the growth of PC, especially in some cases of PC with resistance [57]. Ulixertinib is a targeted therapy drug for BRAF mutations [58] and is a BRAF inhibitor mainly used to treat cancer patients with BRAF mutations. BRAF belongs to the RAS/RAF/MEK/ERK pathway that is activated in an abnormal way among many cancers, especially in melanoma, colorectal cancer, and thyroid cancer [59–61]. The application of ulixertinib in PC is relatively limited, mainly due to the infrequently observed BRAF mutations in PC, and the efficacy of ulixertinib for PC still needs to be verified through more clinical trials.

This study has several limitations that should be considered when interpreting the findings and designing future research. To begin with, all clinical data of PAAD cases were obtained from public databases, with a relatively small-scale sample. Therefore, the accuracy of the results might be limited. Future research needs to incorporate broader sample populations and include more clinical cases to enhance the reliability of the findings. Secondly, the expression of 4 related genes of TANs among PC tissues and adjacent-to-tumor tissues was detected only through the IHC experiments, but further functional experiments were not conducted to verify the specific roles of these 4 genes. Therefore, future research needs to further carry out functional experiments to confirm the functions of 4 genes in the progression of PAAD. Furthermore, the potential mechanisms by which these 4 genes of TANs regulate the prognosis of patients with PAAD still need to be further explored and studied.

In summary, this study established a four-gene TAN-based prognostic model (AIM2, IL18BP, PSCA, and LIPE) for PAAD with strong predictive performance (AUC: 0.774, 0.841, and 0.953 at 1, 2, and 3 years) validated in independent cohorts. The model was associated with distinct immune microenvironment profiles and differential drug sensitivity, with findings corroborated by IHC and qPCR in clinical samples. This work provides a promising tool for precision prognosis and individualized treatment in PAAD, with important value for clinical and translational science, pending prospective validation and functional mechanistic studies.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

All authors contributed to the study conception and design. Writing - original draft preparation: [Zhengrong Ou]; Writing - review and editing: [Lijuan Liao]; Conceptualization: [Tingfeng Xu, Qiuli Xie, Pengzhan Deng]; Methodology: [Jie Zhang, Junyou Zhou, Yuanhang Jiang, Weidong Zhu]; Formal analysis and investigation: [Zhengrong Ou]; Funding acquisition: [Wenli Xu, Songqing He]; Resources: [Songqing He]; Supervision: [Weidong Zhu, Wenli Xu, Songqing He], and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.

Funding

This work was supported in part by the National Natural Science Foundation of China (Grant No. 32460144), the Natural Science Foundation of Guangxi (Grant No. 2025GXNSFAA069085), the National Key Research and Development Program (Grant No. 2022YFE0131600), Guangdong Basic and Applied Basic Research Foundation (Grant No. 2023A1515010143), the Joint Project on Regional High-Incidence Diseases Research of Guangxi Natural Science Foundation (Grant No. 2024GXNSFBA010035) and the Guangxi Key Research and Development Program (Grant No. GuikeAB25069034).

Data availability

The data presented in this study are openly available in the Cancer Genome Atlas (TCGA) database at [https://portal.gdc.cancer.gov] (TCGA-PAAD), and the Gene Expression Omnibus (GEO) database at [https://www.ncbi.nlm.nih.gov/geo/] (GSE62452). GeneCards database [https://www.genecards.org/Search/Keyword?queryString=tumor-associated%20neutrophils], TISCH database [https://tisch.compbio.cn/gallery/?cancer=PAAD&celltype=&species=], and HPA database ([https://www.proteinatlas.org/ENSG00000079435-LIPE/cancer/pancreatic+cancer], [https://www.proteinatlas.org/ENSG00000137496-IL18BP/cancer/pancreatic+cancer], [https://www.proteinatlas.org/ENSG00000167653-PSCA/cancer/pancreatic+cancer], [https://www.proteinatlas.org/ENSG00000163568-AIM2/cancer/pancreatic+cancer]). The original contributions presented in the study are included in the article, further inquiries can be directed to the corresponding author.

Declarations

Ethics approval and consent to participate

The patients involved in the database have obtained ethical approval. The sample collection and informed consent of participants were implemented under the approval offered by the ethics committee of Yueyang Hospital affiliated to Hunan Normal University (approval number: 2025001). All procedures performed in this study were in accordance with the ethical standards of the Declaration of Helsinki (as revised in 2013). Informed consent was obtained from all individual participants included in the study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.