Prognostic value and immune infiltration of novel markers TNRC6C/AMPD1 in pancreatic cancer microenvironment

Abstract

Background

Pancreatic cancer (PC) is a highly lethal malignancy with limited treatment options. Identifying novel prognostic biomarkers and therapeutic targets is crucial for improving patient outcomes.

Methods

A comprehensive bioinformatics analysis was conducted on the Gene Expression Omnibus (GEO, GSE79668, GSE183795) and The Cancer Genome Atlas- Pancreatic Adenocarcinoma (TCGA-PAAD) datasets to identify prognostic biomarkers. The prognostic value of these biomarkers was validated through survival analysis and a Cox proportional hazards model (Cox model). A clinical phenotypic prediction model was constructed using AMPD1 and TNRC6C expression levels, with logistic regression models being built for their combination. The nomogram was constructed to visually represent the model's predictive power. Additionally, immune infiltration and single-cell analyses were performed to explore the underlying mechanisms. Functional experiments were conducted to validate the effects of these biomarkers on PC cell behavior.

Results

Adenosine Monophosphate Deaminase 1 (AMPD1) and Trinucleotide Repeat Containing Adaptor 6C (TNRC6C) were identified as key prognostic biomarkers for PC. High expression of these genes was associated with improved patient survival. Furthermore, AMPD1 and TNRC6C were found to be positively correlated with various immune cells, suggesting their potential role in modulating the tumor immune microenvironment. Functional experiments confirmed that these genes inhibited cancer cell proliferation, migration, invasion, and promoted apoptosis. The prognostic model based on AMPD1 and TNRC6C expression showed significant predictive accuracy, suggesting its potential clinical utility.

Conclusion

This study highlights the prognostic significance of AMPD1 and TNRC6C in PC. These findings provide potential new therapeutic targets for PC and warrant further investigation. The developed clinical prediction model further supports their potential utility as biomarkers for patient stratification and prognosis.

Highlights

• •Identified AMPD1 and TNRC6C as key prognostic factors for pancreatic cancer.
• •Constructed and validated a PC prognostic model based on AMPD1 and TNRC6C.
• •Established a PC diagnostic model using AMPD1 and TNRC6C with good performance.
• •Revealed links between the prognostic model, key factors and tumor immune microenvironment.
• •Elucidated anti-tumor roles of AMPD1/TNRC6C and miRNA-mediated regulatory mechanism.

1. Introduction

Pancreatic ductal adenocarcinoma, a prevalent exocrine cell tumor of the pancreas, ranks as the seventh leading cause of cancer-related mortality globally [1,2]. The five-year disease-free survival rate for individuals diagnosed with pancreatic ductal adenocarcinoma remains notably low, consistently under 10 % over the past several decades [3,4]. Despite advancements in diagnosis, systematic treatments, and radiotherapy methods leading to improvements in survival rates for pancreatic cancer (PC) patients, the incidence of PC and associated fatalities has continued to rise steadily [1,5].

With the rapid advancement of bioinformatics and high-throughput sequencing technologies, machine learning has become increasingly applied in cancer prognosis prediction. By integrating large-scale gene expression data with clinical information, machine learning enables the identification of potential biomarkers and the development of robust prognostic models, providing valuable insights for clinical decision-making [6,7]. In prognostic factor selection, the Least Absolute Shrinkage and Selection Operator (LASSO) regression has been widely employed for its ability to effectively reduce the dimensionality of high-throughput data and select key variables. By introducing L1 regularization, LASSO regression can perform both variable selection and parameter estimation, making it well-suited for addressing multicollinearity issues [8]. Additionally, Random Forest (RF), an ensemble learning method based on decision trees, demonstrates strong classification and regression capabilities. It effectively handles complex, nonlinear relationships and ranks variable importance, proving to be a powerful tool in cancer prognosis studies [9,10]. Cox proportional hazards model (Cox model) remain a cornerstone in survival analysis, frequently used in medical research. This semi-parametric model estimates the relationship between survival time and multiple covariates, allowing for individualized risk scoring and survival prediction [11,12]. In recent years, hybrid approaches that combine machine learning algorithms with Cox models have gained prominence. These methods first utilize LASSO or RF to screen key prognostic factors and then incorporate them into Cox models to enhance prediction accuracy and robustness [13].

In this study, we aim to identify key prognostic factors for PC using LASSO regression and RF, followed by constructing a Cox survival model to assess their predictive power. The model's performance will be validated across multiple datasets to ensure its robustness and accuracy. Additionally, we will explore the potential role of the selected factors in the tumor microenvironment. This approach not only provides insights into the molecular mechanisms underlying PC but also offers potential clinical applications for personalized treatment and prognostic evaluation.

2. Materials and methods

2.1. Screening of key prognostic factors for PC

To construct a survival prediction model for PC, we performed LASSO regression analysis using the glmnet package and combined it with RF analysis using the randomForest package to identify key prognostic genes in the GSE79668 dataset. Subsequently, we validated the expression of these key genes in external datasets GSE183795 and The Cancer Genome Atlas- Pancreatic Adenocarcinoma (TCGA-PAAD) and assessed their relationship with patient survival. Additionally, correlation analysis using the stats package was conducted to evaluate the interrelationship between the selected genes. Detailed information on the dataset and clinical data is provided in Tables S1 and S2.

2.2. Construction and Evaluation of the PC prognostic model

Using Adenosine Monophosphate Deaminase 1 (AMPD1) and Trinucleotide Repeat Containing Adaptor 6C (TNRC6C) as key variables, we constructed a Cox model in the GSE79668 dataset and generated a risk score. The risk score was calculated based on the following formula:

$$RiskScore=\sum _{i=1}^{n}Coe{f}_i\ast x_i$$

where Coefᵢ represents the regression coefficient of gene i derived from the multivariate Cox model, and Expᵢ denotes the normalized expression level of gene i. Samples were divided into high-risk and low-risk groups based on the median risk score, and then arranged in ascending order of risk score. Using ggplot2, a trend plot of the risk scores was generated, with high-risk and low-risk groups distinguished by blue and red colors, and the optimal cutoff point marked with a black dashed line. Survival time scatter plots were created using ggplot2 to illustrate individual survival status, and gene expression heatmaps were generated with ggplot2 and pheatmap to show the overall risk distribution of the patients. Subsequently, time-dependent receiver operating characteristic (ROC) analysis was performed using timeROC to evaluate the predictive performance of the risk score at different time points, and three-colored ROC curves were drawn with AUC values annotated. To assess the model's accuracy and predictive performance, a confusion matrix was calculated, and classification metrics, including accuracy, precision, and recall, were extracted and visualized with bar plots using ggplot2. Finally, a Cox proportional hazards nomogram was constructed using regplot, with 1-year, 2-year, and 3-year time points chosen as key clinical milestones, displaying the predictive effectiveness of the risk score on survival probabilities.

2.3. Establishment of diagnostic model

We constructed a diagnostic model for PC based on the expression levels of AMPD1 and TNRC6C. First, transcript-level transcripts per million (TPM) expression data for normal pancreatic tissues were obtained from the GTEx portal (https://www.gtexportal.org/home/) and combined with the TPM matrix from the TCGA-PAAD dataset. The merged data were log-transformed and then normalized using the normalizeBetweenArrays function to enhance comparability across samples. To account for platform-specific batch effects while retaining group labels, the ComBat function was applied for batch correction. Based on the grouping information, expression data for AMPD1 and TNRC6C were extracted for downstream analysis.

Logistic regression models for AMPD1, TNRC6C, and their combination were then built using the glm function in R. Subsequently, the pROC package was used to plot ROC curves to evaluate the predictive performance of each model in the test set. During feature selection, stepwise regression analysis was performed, and the optimal variable combination was identified by calculating the Akaike Information Criterion (AIC) using the stepAIC function from the MASS package. To assess multicollinearity between variables in the model, the Variance Inflation Factor (VIF) was calculated using the car package. In addition, a nomogram was constructed using the rms package to visually display the predictive power of the model.

2.4. Analysis of the immune microenvironment

To gain insights into the immune microenvironment of PC patients, we performed single-sample Gene Set Enrichment Analysis (ssGSEA) using the GSVA package and an immune-related dataset [14] to evaluate the enrichment of immune-related gene sets in each sample of the GSE79668 dataset. Then, patients were divided into high- and low-risk groups based on their calculated risk scores. To systematically compare the immune cell infiltration profiles between these two groups, we employed the pheatmap package to generate heatmaps. In addition, correlation analysis was performed to explore the relationship between AMPD1 and TNRC6C expression and immune cell infiltration. Correlation coefficients were calculated to evaluate the correlation between AMPD1 and TNRC6C expression and the infiltration scores of different immune cell types, further elucidating the potential role of these two genes in the immune microenvironment. Furthermore, we employed correlation analysis to quantify the relationships between various immune cell types. Subsequently, we utilized the ggplot2 package to generate heatmaps, visually representing the interaction network among immune cells.

Utilizing DISCO (https://disco.bii.a-star.edu.sg/) and TISCH2 (http://tisch.comp-genomics.org/) web tools, we conducted a comprehensive analysis of single-cell RNA sequencing (scRNA-seq) data from PC to further elucidate the relationship between AMPD1 and TNRC6C and the tumor microenvironment.

2.5. Gene Set Enrichment Analysis (GSEA) enrichment analysis

For AMPD1 and TNRC6C, we conducted correlation analysis to identify other genes closely related to these two genes and calculated their correlation scores. Subsequently, we performed single-gene GSEA enrichment analysis using the clusterProfiler package to identify significant pathways related to AMPD1 and TNRC6C in the HALLMARK and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways (h.all.v2023.1. Hs.symbols.gmt and c2. cp.kegg.v2023.1. Hs.symbols.gmt).

2.6. GO analysis

In this study, we extracted genes significantly correlated with AMPD1 and TNRC6C (P < 0.05) and compiled them into gene lists. The enrichGO method was employed to conduct Gene Ontology (GO) analysis on genes associated with AMPD1 and TNRC6C. This analysis aimed to identify biological processes, cellular components, and molecular functions related to these prognostic factors, providing deeper insights into the biological significance of AMPD1 and TNRC6C in PC. The analysis results were visualized using GOplot to display the significance and relevance of different GO terms.

2.7. Cell culture

Shanghai Cell Bank (Shanghai, China) provided the normal human pancreatic ductal epithelial cells HPDE6-C7. The human PC cell lines (Panc 04.03 and PK-59) were purchased from Nanjing Cobioer (Nanjing, China). The BxPC-3 and PANC-1 cells were purchased from Procell Life Science&Technology (Wuhan, China). The BxPC-3, Panc 04.03, and PK-59 cells were cultured in RPMI-1640 and contained with 1 % penicillin-streptomycin (P/S) solution and 10 % fetal bovine serum. In Dulbecco's Modified Eagle Medium (Procell) and contained with 1 % P/S solution and 10 % FBS, HPDE6-C7 and PANC-1 cells were cultured. At 37 °C under a humidified atmosphere with 5 % CO₂, all cells were incubated.

2.8. Cell transfection

The overexpression vector, AMPD1 and TNRC6C overexpression vector were constructed by RiboBio (Guangzhou, China). According to the manufacturer's instructions, Lipofectamine 2000 (Invitrogen, USA) was used for transfection. Transfection efficiency was checked by Western blot and quantitative real-time polymerase chain reaction (qRT-PCR).

2.9. qRT-PCR

From PC cells, total RNA was extracted using TRIzol Reagent (Invitrogen). PrimeScript RT Reagent Kit (TaKaRa, Japan) was applied to reverse transcribed into complementary DNA. SYBR Green PCR Kit (Qiagen) was used for qRT-PCR through a LightCycler 480 II (Roche, Switzerland). RiboBio provided primers [AMPD1, forward (F): 5′-AACTCCCAGCTGAAGAGAAACAAAT-3′, reverse (R): 5′-TCTGTGGAGGTGGACAGAGTC-3’; TNRC6C, F: 5′-CCGTTGCTTGGTCCAGTTTC-3′, R: 5′-GGGTAGAAGGGGAGGATGGT-3’; GAPDH (internal control), F: 5′- GACAGTCAGCCGCATCTTCT-3′, R: 5′-CCCGTTCTCAGCCATGTAGT-3’]. According to the 2^−ΔΔCt method, the expression levels of mRNA were counted.

2.10. Western blot

To extract total proteins from cells and tissues, RIPA buffer (Sigma-Aldrich, USA) was utilized. Following this, 40 μg of protein samples were separated using SDS-PAGE and transferred onto PVDF membranes (Millipore, USA). The membranes were blocked with 5 % non-fat milk and then incubated overnight at 4 °C with primary antibodies: anti-AMPD1 (19780-1-AP, Proteintech, China), anti-TNRC6C (CSB-PA862070LA01HU, Cusabio, China), and anti-GAPDH (60004-1-Ig, Proteintech, China). Afterward, the membranes were treated with secondary antibodies (zsbio) for 2 h at room temperature. Protein bands were visualized using ECL reagent (Nanjing KeyGEN Biotech Co., Ltd., China).

2.11. Cell counting kit-8 (CCK-8) assay

PC cells (5 × 10³ cells per well) were plated in 96-well plates and subsequently treated with 10 μl of CCK-8 solution (servicebio). Following a 2-h incubation at 37 °C, the optical density (OD) at 450 nm was assessed using a spectrophotometer (Thermo Fisher, USA). Cell viability was measured at 24, 48, 72, and 96 h post-seeding.

2.12. Flow cytometry

After transfection, PC cells were re-suspended in 1 × binding buffer. Subsequently, cells were treated with 5 μl Annexin V-PE and 5 μl 7-AAD (Meilunebio, China) for 15 min. Finally, flow cytometry (BD Biosciences, USA) was taken to analyze apoptotic cells.

2.13. Transwell assay

For the invasion assay, PC cells were transfected and then seeded in serum-free medium (100 μl) into the upper chamber of a Transwell system (Corning, USA) coated with Matrigel. The lower chamber contained 600 μl of medium with 10 % FBS. After 24 h of incubation, invasive cells were fixed with methanol and stained with 0.1 % crystal violet. The number of invasive cells was counted under a microscope.

For the migration assay, transwell chambers were used without Matrigel, and all other experimental procedures followed the same protocol as the invasion assay.

2.14. Network construction and miRNA prediction

We constructed the protein-protein interaction (PPI) network of AMPD1 and TNRC6C using the MANIA platform (https://genemania.org/) to explore their interaction. To investigate the upstream regulatory mechanisms of TNRC6C, we first analyzed the differentially expressed miRNAs in the PC miRNA datasets GSE24279 and GSE32678 using the GEO2R platform (https://www.ncbi.nlm.nih.gov/geo/geo2r/). The expression profiles of these miRNAs were then visualized using volcano plots, which were generated with the ggplot2 package in R. Next, we predicted the upstream targeting miRNAs of TNRC6C using the multiMiR package. Since the two datasets contained miRNAs from different species, we first filtered for human-derived miRNAs. Upregulated miRNAs (P < 0.05, logFC >0) were then intersected using a Venn diagram to identify candidates potentially upregulating TNRC6C. Finally, we combined the GSEA enrichment analysis of AMPD1 and used Cytoscape to construct a miRNA-mRNA-pathway network. Key pathways identified from the GSEA results were highlighted and displayed in the network diagram.

2.15. Prediction of transcription factors (TFs) and correlation analysis

The promoter region sequence of AMPD1 was retrieved from the UCSC Genome Browser website (https://genome.ucsc.edu). TFs potentially regulating the AMPD1 promoter region were predicted using the AnimalTFDB4 website (https://guolab.wchscu.cn/AnimalTFDB4). In the TCGA-PAAD cancer patient dataset, spearman correlation analysis was performed using the cor function to assess the relationship between the expression of all genes and AMPD1. Additionally, we performed a Venn diagram analysis to identify the TFs upstream of AMPD1 and selected genes that are inversely correlated with AMPD1.

2.16. Statistical analysis

Data analysis was performed using R software and GraphPad Prism, with results presented as mean ± standard deviation (Mean ± SD). Student's t-test was used for comparisons between two groups, while one-way ANOVA was applied for multiple group comparisons. A p-value <0.05 was considered statistically significant.

3. Results

3.1. Screening of key PC prognostic factors

The analysis process of this study is shown in Fig. 1. In order to construct a PC survival prediction model, we performed LASSO and RF analysis on GSE79668. The results are shown in Fig. 2A and B. After comparing the two models, two genes were finally screened out: AMPD1 and TNRC6C (Fig. 2C). As shown in Fig. 2D, in the GSE79668 data set, the survival time of patients with high expression of AMPD1 and TNRC6C was significantly increased compared with patients with low expression. Subsequently, we verified it in the data sets GSE183795 and TCGA-PAAD, and the results of the survival analysis were consistent with the results in the GSE79668 data set (Fig. 2E and F). Correlation analysis of AMPD1 and TNRC6C was performed in the data sets GSE79668, GSE183795 and TCGA-PAAD respectively, and the results showed that there was a significant positive correlation between the two (Fig. 2G).

Fig. 1.

Flowchart for research.

Fig. 2.

Construction and Validation of Survival Prediction Models (A) LASSO regression analysis was performed on the GSE79668 datasets, displaying the coefficient paths of different features. (B) RF analysis was conducted on the GSE79668 datasets. (C) Venn analysis of the results from LASSO and RF. (D) The survival curves of patients with high expression of AMPD1 and TNRC6C in the GSE79668 datasets are presented. (E) The survival curves of patients with high expression of AMPD1 and TNRC6C in the GSE183795 datasets are presented. (F) The survival curves of patients with high expression of AMPD1 and TNRC6C in the TCGA-PAAD datasets are shown. (G) The correlation analysis results between AMPD1 and TNRC6C in the GSE79668, GSE183795, and TCGA-PAAD datasets.

3.2. Construction and Evaluation of PC prognostic model

In the GSE79668 dataset, we constructed a Cox survival risk model based on AMPD1 and TNRC6C, with the risk score calculated using the formula: Risk Score = (−1.4727 × TNRC6C) + (−0.3751 × AMPD1). As shown in Fig. 3A, the risk score increases with higher risk levels, and the expression levels of TNRC6C and AMPD1 are significantly higher in the low-risk group compared to the high-risk group. Next, we performed ROC curve analysis of the risk score, which demonstrated high predictive sensitivity for 1-year (AUC = 0.75), 2-year (AUC = 0.89), and 3-year (AUC = 0.92) survival outcomes. Additionally, we constructed a clinical prognostic nomogram based on the risk score (Fig. 3D), where the risk score was linearly mapped to Points (0–6 points corresponding to 0–98 points), allowing for the prediction of 1-year survival probability: at a total score of 0, the 1-year mortality risk is 6.34 % (survival rate: 93.66 %), while at a total score of 120, the mortality risk increases to 98.53 % (survival rate: 1.47 %), indicating that patients with High risk scores have extremely poor prognoses. These findings suggest that the risk score serves as an effective tool for predicting patient survival outcomes. According to the results of the confusion matrix, the overall accuracy of the model was 85.71 %, and the 95 % confidence interval was (72.76 %, 94.06 %) (Fig. 3C). Subsequently, survival analysis of the risk score also showed that the survival rate of the high-risk group was significantly lower than that of the low-risk score group (Fig. 3E).

Fig. 3.

Cox Survival Risk Model and External Validation (A) Risk score distribution, risk assessment scatter plot, and gene expression heatmap. (B) Time-dependent ROC curves. (C) Model evaluation parameters are displayed. (D) Clinical prognostic nomogram. (E) Survival analysis results of risk scores indicate that the survival of the high-risk group is lower than that of the low-risk score group. (F–G) External validation of the Cox survival prediction model in the TCGA-PAAD datasets. (H–I) External validation of the Cox survival prediction model in the GSE183795 datasets. ∗∗∗P < 0.001.

The Cox survival prediction model was externally validated using the TCGA-PAAD and GSE183795 datasets. The results are presented in Fig. 3F–I. In both TCGA-PAAD and GSE183795, patients with High risk scores exhibited significantly shorter survival times compared to those with Low risk scores. Additionally, the ROC analysis indicated that the risk score of the Cox model demonstrated high sensitivity to survival status.

3.2.1. Establishment of diagnostic model

As shown in Fig. 4A and B, the expression levels of AMPD1 and TNRC6C in PC tissues are significantly lower than those in normal tissues, suggesting their potential as biomarkers in the development and progression of PC. Based on this, we constructed logistic regression models for AMPD1, TNRC6C, and their combination to predict clinical phenotypes of PC. The results showed that the combined model had an AUC of 0.799 in the test set, which was superior to AMPD1 (AUC = 0.716) but slightly lower than the TNRC6C single-gene model (AUC = 0.802). However, in stepwise regression analysis, the combination of AMPD1 and TNRC6C was selected as the optimal feature combination (AIC = 399.47), indicating that their combination offers better stability and explanatory power. The variance inflation factor (VIF = 1.047) also indicated that there was no significant multicollinearity between the two genes. Therefore, we ultimately selected these two genes to construct the diagnostic model. As shown in Fig. 4C, the odds ratios (ORs) for AMPD1 and TNRC6C in the combined model were 0.77047 (p = 0.039) and 0.212 (p < 0.001), respectively, with statistical significance, and the model's C-index was 0.787, demonstrating high predictive accuracy. Further, the nomogram model (Fig. 4D) visually displayed its predictive value. The model score ranged from 76 (corresponding to a normal probability of approximately 90 %) to 118 (corresponding to a PC probability of approximately 90 %), showing good discriminative ability. TNRC6C had a higher weight in risk prediction (β = −1.55, OR = 0.21), but after combining with AMPD1, the overall performance of the model was more robust, suggesting that the combined model constructed by the two genes has higher potential for application in PC risk prediction.

Fig. 4.

Establishment of diagnostic model in TCGA-PAAD and GTEx datasets. (A)The expression levels of AMPD1 and TNRC6C in normal tissues and pancreatic cancer tissues. (B) ROC curves plotted in the test set after constructing single-gene and multi-gene logistic regression prediction models for AMPD1 and TNRC6C in the training set using the glm function. (C) Forest plot generated from the logistic regression model built using AMPD1 and TNRC6C in the training set. (D) Nomogram generated from the prediction model constructed with AMPD1 and TNRC6C in the training set.

3.2.2. Prognostic Model and Key Prognostic Factors Closely Correlate with the immune microenvironment

Immune infiltration analysis was performed on all patients from the GSE79668 dataset using the ssGSEA package and an immune-related dataset (Table S3). Based on risk scores, patients were again categorized into high- and low-risk groups. Fig. 5A shows that, compared to the high-risk group, the low-risk group exhibits significantly lower scores for Activated B cells, Eosinophils, Immature B cells, Mast cells, MDSCs, and Type 1 T helper cells. Fig. 5B demonstrates a significant positive correlation between AMPD1 expression and Activated B cells, Immature B cells, and Natural killer T cells. Similarly, TNRC6C expression was significantly positively correlated with Type 1 T helper cells, T follicular helper cells, Effector memory CD4 T cells, Natural killer cells, and Effector memory CD8 T cells (Fig. 5C). Moreover, the correlation analysis of immune cells revealed a strong and significant correlation among most immune cell types (Fig. 5D). Specifically, adaptive immune subsets like activated CD4+/CD8+ T cells exhibited strong synergy (r = 0.82, p = 3.2e-13), while immunosuppressive cells—regulatory T cells (Tregs) and myeloid-derived suppressor cells (MDSCs)—were highly correlated (r = 0.88, p < 2e-16), indicating dominant immunosuppression. Activated dendritic cells (DCs) and MDSCs showed tight linkage (r = 0.83, p = 1.4e-13), reflecting a balance between antigen presentation and immunosuppression. B cell subsets (activated/immature B cells, r = 0.88) were internally cohesive but weakly antagonistic with NK cells (r = −0.11∼-0.05), suggesting divergent immune pathways.

Fig. 5.

Immune Infiltration Analysis. (A) Boxplot was constructed to visualize gene expression patterns in patients stratified into high-risk and low-risk groups based on a risk score. (B) Analysis of the correlation between AMPD1 gene expression levels and the infiltration of various immune cells. (C) Analysis of the correlation between TNRC6C gene expression levels and the infiltration of various immune cells. (D) Heatmaps illustrating the correlation between immune cell infiltration levels. ∗∗∗∗P < 0.0001, ∗∗∗P < 0.001, ∗∗P < 0.01, ∗P < 0.05, ns: not significant.

To delve deeper into the association between AMPD1 and TNRC6C and the tumor microenvironment, we employed DISCO and TISCH2 web tools to analyze scRNA-seq data from PC. Our analysis revealed a significantly higher expression of TNRC6C in Naive CD4 T cells, NK cells, Memory CD4 T cells, GZMK + CD8 T cells, and KLRB1+ CD8 T cells (Fig. 6A), corroborating our previous findings. Additionally, Fig. 6B demonstrated a significantly higher expression of AMPD1 in plasma cells. As shown in Fig. 6C, Uniform Manifold Approximation and Projection (UMAP) plots were generated to provide a visual representation of the overall distribution of these two genes within the PC cellular landscape.

Fig. 6.

Gene expression profiles in pancreatic cancer based on scRNA-seq data. (A–B) Violin plots showing the expression levels of TNRC6C and AMPD1 in different immune cell subsets based on scRNA-seq data from the DISCO web tool. (C) UMAP visualization of TNRC6C and AMPD1 expression in different immune cell subsets based on scRNA-seq data from the TISCH2 web tool.

3.2.3. Enrichment Analysis of Key Prognostic Factors

Correlation analysis was conducted for AMPD1 and TNRC6C in the GSE79668 dataset, identifying genes significantly correlated with each factor. Subsequently, GSEA enrichment analysis was performed on these genes using the clusterProfiler package for both AMPD1 and TNRC6C (Table S4). As shown in Fig. 7A and B, 14 common pathways were significantly enriched in the HALLMARK dataset and 13 in the KEGG dataset for both genes. Notably, AMPD1 and TNRC6C were significantly enriched in pathways such as HALLMARK_COMPLEMENT, HALLMARK_APOPTOSIS, HALLMARK_ALLOGRAFT_REJECTION, HALLMARK_INTERFERON_GAMMA_RESPONSE, HALLMARK_UNFOLDED_PROTEIN_RESPONSE, HALLMARK_INTERFERON_ALPHA_RESPONSE, HALLMARK_INFLAMMATORY_RESPONSE, KEGG_T_CELL_RECEPTOR_SIGNALING_PATHWAY, KEGG_GRAFT_VERSUS_HOST_DISEASE, KEGG_CYTOKINE_CYTOKINE_RECEPTOR_INTERACTION, KEGG_PRIMARY_IMMUNODEFICIENCY, KEGG_ALLOGRAFT_REJECTION, and KEGG_INTESTINAL_IMMUNE_NETWORK_FOR_IGA_PRODUCTION (Fig. 7C–F).

Fig. 7.

Enrichment Analysis. (A) Enrichment analyses of pathways related to AMPD1 and TNRC6C based on the HALLMARK gene set using single-gene GSEA. (B) Enrichment analysis of pathways related to AMPD1 and TNRC6C based on the KEGG gene set using single-gene GSEA. (C–D) The GSEA plots for AMPD1. (E–F) The GSEA plots for TNRC6C. (G) Chord plot displaying key results of the GO analysis for AMPD1. (H) Chord plot displaying key results of the GO analysis for TNRC6C.

Genes significantly correlated with AMPD1 and TNRC6C (P < 0.05) were extracted and compiled into separate gene lists. GO enrichment analysis was performed on these gene sets using the enrichGO method (Table S5). The results were ranked and visualized using chord plot. As shown in Fig. 7G and H, genes associated with AMPD1 were significantly enriched in processes such as leukocyte-mediated immunity, lymphocyte-mediated immunity, immunoglobulin production, lymphocyte proliferation, leukocyte activation involved in immune response, lymphocyte differentiation, T cell receptor complex, B cell-mediated immunity, regulation of T cell activation, and cytokine receptor binding. Genes associated with TNRC6C were significantly enriched in processes including small GTPase-mediated signal transduction, T cell differentiation, GTPase regulator activity, nucleoside triphosphatase regulator activity, regulation of autophagy, epithelial tube morphogenesis, canonical NF-kappaB signal transduction, immune response-activating signaling pathway, regulation of leukocyte differentiation, immune response-regulating cell surface receptor signaling pathway, regulation of GTPase activity, regulation of T cell activation, positive regulation of lymphocyte activation, and GTPase activator activity.

3.2.4. AMPD1 and TNRC6C inhibit PC Cell Growth, migration, and Invasion While Promoting Apoptosis

Fig. 8A and B shows that the expression levels of AMPD1 and TNRC6C in PC cells were significantly lower than in normal pancreatic cells. Fig. 8, Fig. 9A demonstrates the successful overexpression of AMPD1 and TNRC6C in Panc 04.03 and PANC-1 cells. Functional analysis of these genes revealed that their overexpression inhibited cell viability in Panc 04.03 and PANC-1 cells (Fig. 8, Fig. 9B). Transwell assays showed that cell migration and invasion were suppressed by the overexpression of AMPD1 and TNRC6C (Fig. 8, Fig. 9C). Furthermore, overexpression of these genes promoted apoptosis (Fig. 8, Fig. 9D).

Fig. 8.

The effects of AMPD1 and TNRC6C in Panc 04.03 cells. (A–B) In the normal human pancreatic ductal epithelial cells and PC cell lines, AMPD1 and TNRC6C expressions were tested using qPCR and Western blot. (C) AMPD1 and TNRC6C expression were tested using qRT-PCR. (D) Cell viability was checked via CCK-8 assay. (E) Cell migration and invasion were estimated through Transwell assay. (F) Cell apoptosis was measured by means of flow cytometry. ∗∗∗P < 0.001, ∗∗P < 0.01, ∗P < 0.05.

Fig. 9.

The effects of AMPD1 and TNRC6C in PANC-1 cells. (A) AMPD1 and TNRC6C expression were tested using qRT-PCR. (B) Cell viability was checked via CCK-8 assay. (C) Cell migration and invasion were estimated through Transwell assay. (D) Cell apoptosis was measured by means of flow cytometry. ∗∗∗P < 0.001, ∗∗P < 0.01.

3.2.5. miRNA negatively regulates AMPD1 expression by targeting TNRC6C

PPI network analysis conducted via the MANIA platform (https://genemania.org/) revealed an interaction between AMPD1 and TNRC6C, though not a direct one (Fig. 10A). Upon AMPD1 overexpression, there was no change in TNRC6C protein expression. However, overexpression of TNRC6C led to a significant increase in AMPD1 expression (Fig. 10B). These findings suggest that TNRC6C may regulate AMPD1 indirectly, thereby inhibiting the progression of PC.

Fig. 10.

miRNA negatively regulates AMPD1 expression by targeting TNRC6C. (A) PPI mapping of AMPD1 and TNRC6C. (B) AMPD1 and TNRC6C expression were tested using Western blot. (C) Volcano plots of differentially expressed miRNAs from GSE24279 and GSE32678 datasets. (D) Venn diagram of TNRC6C upstream miRNAs and upregulated miRNAs from GEO datasets. (E) Schematic representation of the putative miRNA–mRNA–pathway interactions. ∗∗∗P < 0.001, ns: not significant.

To further investigate the upstream regulatory mechanisms of TNRC6C, we analyzed PC miRNA datasets GSE24279 and GSE32678 (Fig. 10C). Based on the intersection of the two datasets, we identified 13 upregulated miRNAs that may target TNRC6C, as predicted by the multiMiR package and supported by validated experimental data (Table S6). These miRNAs include hsa-miR-107, hsa-miR-125a-5p, hsa-miR-199a-3p, hsa-miR-199a-5p, hsa-miR-199b-3p, hsa-miR-28-5p, hsa-miR-299-5p, hsa-miR-320b, hsa-miR-320c, hsa-miR-320d, hsa-miR-331-3p, hsa-miR-590-3p, and hsa-miR-876-3p (Fig. 10D). Integrating these results with the GSEA analysis of AMPD1, we proposed a conceptual miRNA–TNRC6C–NR2F6–AMPD1–pathway network to illustrate the potential regulatory mechanism (Fig. 10E). We suggest a potential regulatory axis: Upregulated miRNAs bind to and suppress TNRC6C → suppression of TNRC6C → reduced TNRC6C weakens the miRNA complex's ability to degrade TF mRNAs → increased NR2F6 expression → transcriptional repression of AMPD1 → decreased AMPD1 expression →activation of downstream pathways. Specifically, the downstream immune-related pathways positively associated with AMPD1 expression include Allograft Rejection, IL2–STAT5 Signaling, Interferon Gamma Response, Complement, and Interferon Alpha Response. To validate this mechanism, we identified genes negatively correlated with AMPD1 in TCGA and predicted TFs targeting its promoter region. NR2F6 was found at the intersection. As an oncogenic TF, NR2F6 can repress tumor suppressor genes. Reduced TNRC6C may facilitate NR2F6 upregulation and its inhibitory effect on AMPD1, thereby promoting tumor progression and immune evasion.

4. Discussion

In this study, we screened out AMPD1 and TNRC6C as key prognostic factors for PC from the GSE79668 dataset through LASSO and RF analysis. Through survival analysis, we found that patients with high expression of AMPD1 and TNRC6C had significantly longer survival times compared to those with low expression, which was validated in the GSE183795 and TCGA-PAAD datasets. This indicates that these two genes may play an important role in the prognosis of PC. Further correlation analysis revealed a significant positive correlation between AMPD1 and TNRC6C, suggesting that they may jointly affect the progression of PC through synergistic effects. Additionally, the Cox model we constructed was externally validated in the TCGA-PAAD and GSE183795 datasets, showing that patients with High risk scores had significantly shorter survival times compared to those with Low risk scores. Furthermore, ROC curve analysis indicated that the model had high predictive ability for PC survival status. This further demonstrates the reliability of AMPD1 and TNRC6C as prognostic markers for PC.

Compared to other studies on prognostic markers of PC, the roles of AMPD1 and TNRC6C are more focused on the regulation of the tumor immune microenvironment. In recent years, the role of immune infiltration in tumor progression has received widespread attention, especially the infiltration levels of immune cells such as T cells and macrophages, which are considered to be closely related to the malignant degree of the tumor and patient prognosis [15]. Our immune infiltration and single-cell analyses further validate the roles of AMPD1 and TNRC6C in regulating the tumor microenvironment, particularly their positive correlations with NK T cells, B cells, and CD4/CD8 T cells. These findings suggest that these two genes may influence cancer progression by modulating tumor immune responses. Furthermore, when compared to established biomarkers like CA19-9 and KRAS mutations, TNRC6C and AMPD1 show distinct potential in the context of immune regulation. While CA19-9 remains a widely-used diagnostic marker for tumor burden monitoring, it has limited predictive value for immune contexture and fails to guide immunotherapy personalization [16,17]. KRAS mutations, though driving >90 % of PCs via MAPK pathway activation, face therapeutic resistance challenges [18,19]. In contrast, in our study, the low-risk group exhibited higher infiltration of immune effector cells, including Activated B cells, Immature B cells, Natural killer T cells, Type 1 T helper cells, and Effector memory CD4/CD8 T cells. Both AMPD1 and TNRC6C showed significant positive correlations with these cells, suggesting their protective roles in PC through modulation of anti-tumor immunity. These findings open avenues for integrating TNRC6C and AMPD1 into immunotherapy strategies. Future studies should focus on validating these genes in clinical trials and exploring their roles in combination with existing immunotherapies.

Previous studies have found that AMPD1 has been involved in various cancer-related studies. In lung cancer, immunohistochemistry showed higher AMPD1 protein levels, and qRT-PCR showed that it was upregulated in A549 cells compared with BEAS-2B cells [20]. TNRC6 has been studied in relation to cancer, with several key findings regarding its function. In miRNA biogenesis and function, TNRC6 is a core component of the miRNA-induced silencing complex (miRISC) [21,22]. However, the mechanism of interaction between AMPD1 and TNRC6C in tumors is still unclear.

In this study, we constructed a potential regulatory network involving miRNA–TNRC6C–TF–AMPD1 based on GSEA enrichment results, and proposed the following hypothetical mechanism: Upregulated upstream miRNAs target and suppress TNRC6C expression. As a core component of the miRNA-induced silencing complex, downregulation of TNRC6C weakens the miRNA complex's ability to degrade target mRNAs, such as that of the TF, resulting in increased the TF expression. To support this hypothesis, we identified genes negatively correlated with AMPD1 in the TCGA database and predicted TFs that may regulate the AMPD1 promoter. Cross-analysis revealed NR2F6 as a promising candidate. As a classical pro-oncogenic TF, NR2F6 promotes tumor progression by directly binding to and repressing the transcription of multiple tumor suppressor genes (e.g., interferon-γ and granzyme B) at their promoter regions [23]. It also functions as an immune checkpoint molecule to inhibit T cell-mediated antitumor activity, facilitating immune escape and correlating with poor prognosis [24,25]. It is hypothesized that NR2F6 may also transcriptionally repress AMPD1 expression through similar mechanisms. Further GSEA enrichment analysis showed that AMPD1 expression was positively associated with the activation of immune-related pathways, including Allograft Rejection, IL2–STAT5 Signaling, Interferon Gamma Response, Complement, and Interferon Alpha Response. These pathways are closely linked to the function of B cells (Activated and Immature) and natural killer T cells, supporting the role of AMPD1 in enhancing antitumor immunity and improving prognosis. These findings imply that the proposed regulatory axis may play a pivotal role in shaping the tumor immune microenvironment. However, this hypothesis is primarily based on integrative bioinformatics analysis, partial experimental evidence, and literature mining, and thus, several limitations remain. Therefore, the construction of the miRNA–TNRC6C–NR2F6–AMPD1 axis should be regarded as a theoretical framework. Future studies will be required to experimentally validate the regulatory relationships along this axis and to further elucidate its biological function in cancer immunity and progression.

In summary, this study not only provides new evidence for the prognostic value of AMPD1 and TNRC6C in PC, but also reveals that they may affect tumor progression through immune regulatory mechanisms. Additionally, we constructed a clinical phenotype prediction model that incorporates AMPD1 and TNRC6C, further enhancing the accuracy of prognosis prediction for PC patients. Compared with previous literature, this study is the first to systematically explore the synergistic effects of AMPD1 and TNRC6C in PC and their impact on the tumor immune microenvironment, providing new potential targets for future immunotherapy strategies. To further validate our findings, future studies may consider prospective clinical trials and immunohistochemical (IHC) staining of AMPD1 and TNRC6C in tumor tissues to confirm their expression patterns and clinical relevance.

CRediT authorship contribution statement

Yongting Lan: Writing – review & editing, Writing – original draft. Wenyan Du: Formal analysis, Data curation. Yongfen Ma: Software, Resources. Jingmei Cao: Writing – review & editing, Writing – original draft, Visualization.

Consent to participate

Not applicable.

Data availability statement

The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.

Ethical approval

Not applicable.

Consent for publication

Not applicable.

Funding

This project was funded by Natural Science Foundation of Shandong Province (ZR2022MC174).

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article:

Multimedia component 2

Table S2 Detailed information of clinical data in this study dataset.

Multimedia component 3

Table S3 lmmunoinfiltration analysis results of TNRC6C and AMPD1.

Multimedia component 4

Table S4 GSEA analysis results of TNRC6C and AMPD1

Multimedia component 5

Table S5 GO analysis results of TNRC6C and AMPD1.

Multimedia component 6

Table S6 Prediction results of upregulation of miRNA in PAAD targeting TNRC6C.