1 Introduction

Pancreatic ductal adenocarcinoma (PAAD) is marked with a poor 5-year survival rate of about 10%, rendering it extremely aggressive and deadly [1]. Due to the highly aggressive nature of pancreatic cancer, many patients are diagnosed at an advanced stage when the opportunity for surgical intervention has been missed. Consequently, chemotherapy remains the primary treatment for pancreatic cancer [2]. Recently, the development of more effective treatment has opened up new treatment strategies. For instance, the use of pre- or perioperative medical nutrition, including immunonutrition, has decreased the rate of postoperative infections and improved the survival [3]. The therapeutic strategies, including immune-checkpoint inhibitors, tyrosine kinase inhibitors, and agents targeting metabolic pathways or the tumor microenvironment, are currently under investigation [4, 5]. The tumor microenvironment (TME) is inherently complex and heterogeneous, necessitating the exploration of combination therapies that have shown promising results. To optimize these therapeutic approaches, it is imperative to deepen our understanding of the clinical, immunological, and molecular aspects of cancer. Moreover, identifying predictive and prognostic biomarkers is crucial for personalized medicine and can significantly improve patient outcomes.

The tumor-node-metastasis (TNM) staging system, established by the American Joint Committee on Cancer (AJCC), serves as a foundational tool for cancer prognosis. However, its predictive power is not uniform across all patients [6]. Emerging molecular markers are outperforming traditional tumor parameters in predicting cancer risk. Immunoscores that assess in situ immune cell infiltration within tumors, as well as aberrant DNA and mRNA levels, offer more precise prognostic information. The RMH score, in particular, shows potential as a versatile prognostic tool applicable across various cancer types and clinical scenarios [7]. A better understanding of the molecular landscape of pancreatic cancer would enable the development of novel therapeutic strategies to improve clinical outcomes and facilitate the stratification of patients into prognostic groups to guide personalized treatment.

Mitochondria possess a remarkable capacity for change and adaptation, playing a crucial role in cellular processes that significantly impact metabolic activities within cells. There has been a growing emphasis on the importance of mitochondria in cancer development, stress response, and maintaining homeostasis. In recent years, the essential role of mitochondria in cancer initiation, tumor evolution, and metastasis is gaining increasing attention [8]. Mitochondria are central to numerous metabolic processes, including glucose breakdown, fatty acid β-oxidation, and amino acid metabolism, all of which are implicated in cancer [9]. The abnormal accumulation of certain substances in mitochondria, including fumarate, succinic acid, and 2-hydroxyglutarate (2-HG), significantly influences malignant transformation [10]. Moreover, mitochondria contribute to the generation of reactive oxygen species (ROS), which are implicated in the progression towards cancer [11]. Additionally, mitochondrial metabolism strongly influences the immune cell maturation and functionality such as macrophages and T cells, thereby exerting a critical impact on the TME [12, 13]. Alterations in mitochondrial function have been found to affect the efficacy of immunotherapy [14]. These studies' results strongly suggest that mitochondrial metabolic processes are essential to the dynamics of tumorigenesis and cancer progression.

Restructuring of metabolic pathways within cells is widely recognized as a hallmark of cancerous cells [15]. Pancreatic cancer cells undergo significant changes in their metabolic profiles, showing increased activities in fatty acid and cholesterol synthesis, as well as elevated cholesterol absorption [16,17,18]. Disruptions in mitochondrial-related and other metabolic pathways may facilitate the onset and progression of cancer, as well as aiding the cancer cells in evading immune surveillance [19]. A recent study revealed that genes associated with glycolysis, a metabolic process, have the ability to predict cancer outcomes accurately and reflect the immune system's status [20]. Therefore, it is plausible that key genes involved in mitochondrial processes could influence the predictive outcomes of PAAD and serve as viable targets for therapeutic intervention.

In the study, the Cancer Genome Atlas (TCGA) database was used to explore mitochondrial metabolism-related genes (MMRGs) in PAAD. Our analysis involved examining gene expression levels, mutations, responses to chemotherapy drugs, and their correlation with the infiltration of immune cells within TME using bioinformatics techniques. Additionally, a predictive algorithm was developed to assist in assessing patients' survival probabilities. The findings could serve as a significant predictive tool and a fresh viewpoint for personalized treatment of pancreatic adenocarcinoma.

2 Materials and methods

2.1 Data availability and clinical information

RNA transcriptome data and clinical record of individuals with PAAD and healthy individuals were obtained from the TCGA and GTEx databases through the UCSC Xena database (https://xena.ucsc.edu/). The cohorts named ICGC-PACA-CA and ICGC-PAAD-US, linked to PAAD, were extracted from the International Cancer Genome Consortium (ICGC, https://dcc.icgc.org/). Genes related to mitochondrial metabolism were sourced from the Molecular Signatures Database (MSigDB) and a previous publication [21]. Patients were excluded from this study based on the following criteria: (1) patients had not detailed clinical parameters e.g. sex, age, grade, TNM stage and survival time; (2) patients had not complete information of genome data. Finally, 179 patients obtained from TCGA database with a total of 196 PAAD patients.

2.2 Gene expression differential profiling

The differentially expressed genes (DEGs) of TCGA PAAD patients and GTEx normal subjects were assessed using three R packages: DESeq2 (v4.41.12), edgeR (v3.40.2), and limma (v3.54.2). A significance level of adjusted P-value < 0.05 and |log2 FC|> 2 was utilized as the threshold for DEGs selected. The final upregulated and downregulated genes were determined by the overlap of DEGs identified by the three R packages. The overlap of the DEGs with gene sets from MsigDB was visualized using a Venn diagram for analysis.

2.3 Risk model construction

A cohort of 179 PAAD patients with complete survival data from TCGA served as the training dataset for identifying prognostic genes. Initially, in a univariate analysis, the Cox model and Kaplan–Meier analysis were used to identify MMRGs signatures associated with overall survival (OS). This analysis was conducted with the survival R package (version 3.3.1) and P-value < 0.05. Subsequently, we applied LASSO regression, followed by a stepwise method with My.stepwise (version 0.1.0) packages to select variables for the prognostic model. The ultimate risk model was developed using multivariate Cox regression analysis with the survival R package (version 3.3.1). Each patient received an individual risk score calculated based on the formula of prognostic model, where the regression coefficients, gene expression levels of MMRGs, and total gene counts were integrated. Subsequently, using median risk scores, PAAD patients were stratified into high and low risk categories for further analysis.

2.4 Risk model evaluation and validation

TCGA-PAAD dataset for training and two external validation sets, ICGC-PACA-CA (n = 186) and ICGC-PAAD-US (n = 112) were utilized to predict outcomes for individuals with PAAD through the prognostic index. With the tinyarray R package (version 2.2.9), we executed an analysis to plot risk score curves, stratify patients into high or low risk groups, and study the expression patterns of MMRGs within these stratified groups. Utilizing the survival (version 3.3.1) and survminer (version 0.4.9) packages in R [22], we generated survival curves at 1, 3, and 5 years and computed their respective AUC values. The timeROC R package (version 0.4) was then used to visually represent these survival analyses over time.

Then the risk score and other clinical factors were analyzed with univariate Cox and multivariate Cox model to define the independently prognostic indictor, both P-value < 0.05. A nomogram was formulated to predict risk using the rms R package (version 6.0). Decision Curve Analysis (DCA) was conducted to determine the clinical utility of the nomogram with the ggDCA R package (version 1).

2.5 Gene set variation analysis (GSVA)

GSVA was used to assess the functional enrichment of a gene set by assigning a comprehensive score, converting gene-level changes into pathway-level alterations, and subsequently determining the biological relevance of the sample, which was employed to compute the enrichment scores for gene sets corresponding to KEGG pathways [23].

2.6 Mutation landscape analysis

Maftools R package (version 2.14.0) was carried out to visualize of genetic mutations [24], which the mutation and clinical information sourced from The Pan-Cancer Atlas (PanCanAtlas, https://gdc.cancer.gov/about-data/publications/pancanatlas). A waterfall plot was used to display, the results of mutation rates between patients classified as high-risk and low-risk. Enrichr provides access to a diverse array of gene set libraries for analysis and can be downloaded from (https://maayanlab.cloud/Enrichr/), focusing on the biological processes associated with Gene Ontology (GO) analyses.

2.7 Immune cell infiltration estimation and immune subtype analysis

Previously established signatures detailing the classification of immune cells and their related roles were referenced for this study [25]. The ESTIMATE algorithm was employed to assess TME by quantifying both stromal and immune components, indicated by their respective stromal and immune scores. GSVA R package with the single-sample Gene Set Enrichment Analysis (ssGSEA) method were used to estimate the immune cell infiltration and the activation status. Additionally, the immunedeconv (version 2.1.3) and CIBERSORT (version 0.1.0) R packages integrated XCELL, TIMER, QUANTISEQ, and CIBERSORT algorithms to investigate immune cell infiltration levels [26].

The Spearman rank-order correlation method, implemented in the psych R package (version 2.3.3) was used to assess the relationship association the risk score with immune cell infiltration and P < 0.05. The ImmuneSubtypeClassifier R package (version 0.1.0) was conducted to classify TCGA-PAAD patients into six immune subtypes: C1 (Wound Healing), C2 (IFN-γ Dominant), C3 (Inflammatory), C4 (Lymphocyte Depleted), C5 (Immunologically Quiet), and C6 (TGF-β Dominant).

2.8 Drug response prediction

The Genomics of Drug Sensitivity in Cancer 2 (GDSC2, https://www.cancerrxgene.org/) stands as the largest pharmacogenomics database, within this repository, the PharmacoDB database holds a vast collection of pharmacogenomics data, including details on 805 types of cells and their responses to 198 antitumor drugs. The drug response predictions for TCGA-PAAD patients were made using the oncoPredict R package (version 0.2) [27]. The psych R package (v2.3.3) was utilized to analyze the relationship between drug response and risk score and the results were visualized with the ggstatsplot (version 0.0.6) and ggpubr (version 0.6.0) R packages.

2.9 Statistical analysis

R software (version 4.2.3) was employed for all statistical analyses. The Student’s t-test were carried out to compare the group mean, and the Pearson method were evaluated the correlations. The log-rank test was utilized for Kaplan–Meier survival analysis. A significance threshold of P < 0.05.

3 Results

3.1 Differential expression analysis

The flow diagram of the study process was shown in Fig. 1. Firstly, 179 tumor samples from PAAD patients and 171 control individuals were chosen from the integrated data of TCGA and GTEx. Whole-genome expressions were analyzed using three distinct bioinformatics tools: DESeq2, edgeR, and limma packages using R, with significance determined at an adjusted p-value less than 0.05 and |log2FC| greater than 2 as the criteria. A total of 3156 genes exhibiting altered expression levels were detected, with 1985 genes increased and 1171 genes decreased (Table 1). Subsequently, the prognostic value of the DEGs was assessed via univariate Cox regression, resulting in the identification of 1150 DEGs (Table S1) and 681 DEGs (Table S2) associated with prognostic value through Kaplan–Meier survival analysis. Through the alignment of DEGs with MMRGs gene sourced from MSigDB, compilation of 36 MMRGs was established (Fig. 2A, Table S3).

Fig. 1
figure 1

The general workflow of the current study. MMRGs: mitochondrial metabolism-related genes; OS: overall survival

Full size image
Table 1 The numbers of differential expression gene analysis with three R packages
Full size table
Fig. 2
figure 2

Screening of MMRGs-related DEGs in PAAD. A An intersection of prognostic MMRGs and DEGs was conducted based on their significance difference analysis, resulting in a Venn diagram with the jvenn online tool. B, C LASSO coefficient profiles were created to identify five MMRGs-related DEGs. D, E Univariate and multivariate Cox regression analyses of the association connection DEGs with OS, and a forest plot illustrating the results

Full size image

3.2 Risk model construction, evaluation, and validation

The optimal candidate genes for the Cox model were screened using LASSO regression and My.stepwise (version 0.1.0) packages (as depicted in Fig. 2B and C). The final MMRGs risk model consisted of five specific genes: HPGD, ABCB4, PLD4, KCNN4, and CDA. As illustrated in Fig. 2D and E, the correlation between these genetic risk factors and the overall survival was confirmed. Subsequently, each patient's risk score was calculated by applying the following formula: riskscore = 0.1362892*HPGD-0.4482712*ABCB4-0.3109010*PLD4 + 0.1787457*KCNN4 + 0.1545474*CDA. The TCGA-PAAD training cohort patients categorized into high-risk and low-risk subgroups using the median risk score. High-risk patients exhibited reduced expression of ABCB4 and PLD4, while heightened expression for HPGD, KCNN4, and CDA were revealed (Fig. 3A). Kaplan–Meier survival analysis showed that the high-risk patients experienced more poorer outcomes compared to those in the low-risk group (Fig. 3B). Additionally, The AUC values from ROC analyses for the 1st, 3rd, and 5th year were calculated as 0.73, 0.85, and 0.85, respectively (Fig. 3C).

Fig. 3
figure 3

Assessment of the risk score model in PAAD. A The distribution of risk scores and patients, along with the expression profiles of genes within the TCGA-PAAD cohort. B Survival plots were created for patients categorized as high risk and low risk using Kaplan–Meier survival analysis. C ROC curves were plotted to evaluate the model's predictive performance over 1, 3, and 5 years using the timeROC (v0.4) R package

Full size image

Moreover, the predictive capability of the risk model was validated  using two separate datasets from the ICGC database. The results showed that the patients identified as high risk exhibited significantly poorer prognoses, with similar expression profiles observed in the training dataset (Fig. 4A–D). Integrating the above data, the MMRGs signature emerged as a promising predictive biomarker for individuals with PAAD. Subsequently, the risk score may serve as an independent prognostic indicator for overall survival through both univariate and multivariate Cox regression analyses (Fig. 5D, E).

Fig. 4
figure 4

The validation of the risk model in PAAD was broadened to external cohorts: A, B ICGC-PACA-CA and C, D ICGC-PAAD-US. Kaplan–Meier analysis was constructed to create survival curves that stratified patients based on high and low risk, along with, the distribution of survival, expression in the subgroup patients

Full size image
Fig. 5
figure 5

A nomogram was developed which integrated clinical parameters alongside the calculated risk score. A To evaluate its precision within TCGA-PAAD patients, the nomogram plot was generated based on clinical factors and the risk score. B Calibration plot of the nomogram in TCGA-PAAD patients, calibration plots were drawn using the rms (v6.6–0) R package. C A decision curve analysis (DCA) was used to compare the predictive performance of the nomogram against conventional medical factors such as age, N stage, T stage, and tumor grade. The DCA curve was drawn using the ggDCA (v1.2) R package. D, E The results from univariate and multivariate Cox regression analyses provided substantial evidence supporting the risk score as an independent prognostic factor

Full size image

Exploring the determinants impacting the long-term prognosis of individuals under medical treatment and care is paramount. Forecasting the survival of TCGA-PAAD patients, incorporating prognostic factors like age, N-level, and T-level, elucidated the relationship between survival and risk score (Fig. 5A). Notably, the risk score emerged as the significant influence among those scrutinized in the research. Calibration plots revealed that the developed nomogram closely emulated the performance of an ideal model when forecasting patient outcomes within the TCGA-PAAD cohort (Fig. 5B). Additionally, decision curve analysis showcased the nomogram outperformed individual clinical factors in predictive accuracy for 1-, 3-, and 5-year outcomes (Fig. 5C).

Furthermore, the study utilized GSVA to reveal the characteristics that differentiated high-risk from low-risk patients. The identified significant differential KEGG pathways were presented in Fig. 6A and Table S4. Notably, the analysis highlighted four key KEGG pathways significantly enriched in the high-risk group, all closely linked to cancer development and progression (Fig. 6B).

Fig. 6
figure 6

Comparing the molecular characteristics between high- and low-risk patients. A Firstly, the volcano plot displayed the differentially enriched KEGG pathways in high- versus low-risk PAAD patients using ggplot2 R package. B The bar plot showcasing the top 4 positively and negatively enriched KEGG pathways in high-risk patients

Full size image

3.3 Mutational landscape and Variations in drug responsiveness comparing the high-risk cohort to the low-risk cohort

Mutational landscape analysis revealed that the top five mutated genes in both groups were KRAS, TP53, SMAD4, CDKN2A, and TTN. Notably, significantly higher mutation levels were observed in seven genes—TP53, KRAS, CDKN2A, BTBD11, COL6A2, PCDH9, and USH2A—within the high-risk patients versus the low-risk patients (Fig. 7A, B). The examination of mutation rates suggested that the high-risk cohort exhibited a high prevalence of KRAS gene mutations, while the TP53 gene mutation was most prevalent in the low-risk group (Fig. 7C, D). Furthermore, the primary focus of the top 10 GO biological process terms, depicted in Fig. 7E, centered around cell proliferation and programmed cell death. These findings indicate that the molecular characteristics and mutation patterns can help define PAAD patients with distinct profiles, particularly emphasizing the role of mutational molecular risk groups.

Fig. 7
figure 7

Comparing mutational landscapes between high- and low-risk groups. A, B The top 5 mutated genes in low-risk group, and the top 7 mutated genes in high-risk group. C, D A forest plot was used to emphasize the significant differences in gene mutations between high- and low-risk patients by employing Fisher's test. E An enrichment analysis of significantly different mutated genes was conducted through Enrichr to explore the top 10 GO biological process terms

Full size image

In order to delve deeper into the variations in drug responsiveness between the two risk groups, an analysis of the IC50 values for 198 chemotherapy drugs or inhibitors in both cohorts was conducted to assess their efficacy. Figure 8A, B showcased 10 representative drugs from this analysis. Spearman correlation analysis revealed a positive correlation between the risk scores and drugs like BMS.754807, Sabutoclax, GSK2578215A, Venetoclax, and AZD8055, while showing a negative correlation with Trametinib, Acetalax, SCH772984, ERK_6604, and Selumetinib. Interestingly, drugs that exhibited a positive correlation displayed significantly lower IC50 values in the low-risk group, indicating more sensitive in the low-risk group, whereas a negative correlation suggested more effective in the high-risk group.

Fig. 8
figure 8

Exploration of the relationship between drug sensitivity and risk score was conducted using two different statistical analyses. A Spearman correlation between drug IC50 values and risk scores. B Wilcoxon tests utilized to evaluate variances in drug IC50 values

Full size image

3.4 Immune characteristics of high- and low-risk groups

Initially, we compared immune cell infiltration levels between the two groups, evaluating stromal, immune, and estimate scores. The results revealed significantly higher scores in all categories for the low-risk group (p < 0.001) (Fig. 9A). Using the CIBERSORT algorithm, we examined immune cell infiltration, which demonstrated higher levels of B cells, CD8 T cells, and tumor-infiltrating lymphocytes (TILs) in the low-risk patients, while high-risk patients showed elevated levels of immature dendritic cells (iDCs), macrophages, neutrophils, Th2 cells, and regulatory T cells (Tregs) (Fig. 9B, C). Moreover, we observed reduced cytolytic activity, diminished HLA presentation, impaired T cell co-stimulation, and a weakened type II IFN response in the high-risk group, all indicative of an immunosuppressive TME in these patients. Additionally, the correlation between immune cell infiltration and risk score was investigated utilizing numerous algorithms. The results demonstrated that the risk score was negatively correlated with B cells and CD4 + T-cells (Fig. 9D). Furthermore, patients with PAAD were categorized into C1-C6 immune subtypes. Notably, the high-risk group exclusively demonstrated the presence of C4 (Lymphocyte Depleted), characterized by an enhanced macrophage signature and a strong M2 response, indicating an immunosuppressed state. The absence of C5 (Immunologically Quiet) in both groups point to the aggressive nature of PAAD (Fig. 9E).

Fig. 9
figure 9

Distinguishing the immune profiles between high- and low-risk groups. A Comparing stromal, immune, and estimate scores across the two groups using the Wilcoxon test. B, C Comparing the immune infiltration and the immune function scores between these groups using Wilcoxon test. The gene set enrichment was performed using the GSVA (v1.46.0) R pack, the box plots were drawn using the ggpubr (v0.6.0) R package. D The infiltration levels of immune cells with the immunedeconv and CIBERSORT R packages. The spearman correlation between the infiltration ratio and risk score was performed using the psych (v2.3.3) R package. E The distribution of immune subtypes in PAAD patients using the ImmuneSubtypeClassifier (v0.1.0) R package and plotted using the ggplot2 R package

Full size image

3.5 Expression validation of risk genes in PAAD tissues

Analysis within the TCGA-GTEx cohort revealed a significant increase in mRNA levels of HPGD, ABCB4, PLD4, KCNN4, and CDA genes in tumor samples, as opposed to normal tissues (Fig. 10A). Furthermore, in the TCGA cohort, the Kaplan–Meier curve demonstrated a positive association between the OS of PAAD patients and the gene expression levels of ABCB4 and PLD4, while a negative association within HPGD, KCNN4, and CDA (Fig. 10B).

Fig. 10
figure 10

The validation of risk gene expression within PAAD tissues involved two main analyses. A Comparing mRNA levels of risk genes using the Wilcoxon test in the TCGA-GTEx cohort. B Kaplan–Meier survival curves constructed for high- and low-risk patients

Full size image

4 Discussion

Mitochondrial metabolism has been identified as a significant player at various stages of cancer progression, such as the initiation of malignancy, tumor progression, and the response to treatment interventions [28, 29]. Ongoing clinical trials are exploring the efficacy of inhibitors targeting crucial enzymes involved in mitochondrial metabolism, which could lead to advancements in precision medicine for cancer treatment[30]. However, despite ongoing research efforts, the full extent of the impact of mitochondrial metabolism on pancreatic cancer remains inadequately understood. Through the utilization of bioinformatic techniques on the TCGA dataset, this study developed a gene signature (MMRGs) associated with mitochondrial metabolism in PAAD. The predictive capacity of this signature demonstrated improved effectiveness in age, pathological T stage, tumor recurrence, and overall prognosis and received preliminary validation in a PAAD model. Moreover, the genes in the predictive model are closely related to the development of immune cells in PAAD, providing corresponding targets for the development of novel treatments such as immune checkpoint inhibitors.

In the study, the comprehensive MMRGs risk model was established by combining the genes HPGD, ABCB4, PLD4, KCNN4, and CDA. The HPGD gene located on chromosome 4q34-35 encodes the enzyme 15-hydroxyprostaglandin dehydrogenase (HPGD), which is crucial for prostaglandin metabolism, impacting various physiological and pathological processes like inflammation, angiogenesis, and abnormal responses [31]. Studies have linked HPGD to the stemness and progression of cholangiocarcinoma through its targeting by bile exosomal miR-182/183-5p [32]. The ABCB4 gene, on chromosome 7, encodes the multidrug resistance protein 3, also known as the ABCB4 protein, responsible for transporting phospholipids into bile. It is part of a gene signature related to lipid metabolism that can predict prognosis of clear cell renal cell carcinoma [33]. The PLD4 gene encodes a component of the phospholipase D family, linked to phospholipid signaling. Previous research has demonstrated that PLD4, residing at macrophages within colon cancer mesenchymal and lymph nodes, can stimulate M1 macrophages to exhibit anti-tumor effects in colon cancer cells [34]. The KCNN4 gene, situated on 19Q13, encodes the calcium-activated potassium channel K(Ca)3, which play essential roles in various cancers, including papillary thyroid cancer [35], colorectal cancer [36], and pancreatic cancer [37, 38]. The CDA gene encodes cytidine deaminase, a vital enzyme in the pyrimidine salvage pathway for DNA and RNA synthesis. Studies have shown high CDA expression in pancreatic adenocarcinoma, where it significantly contributes to tumor growth [39].

The research study calculated individual risk scores and divided patients into high- and low-risk groups with the median. Analysis using Kaplan–Meier survival plots revealed a stark contrast in prognosis between the two individuals, with those in the high-risk category facing a significantly poorer outlook. Additionally, the ROC curve showed excellent performance, with AUC values of 0.73, 0.85, and 0.85 for 1-year, 3-year, and 5-year intervals, respectively, which was superior to that of the previous models [40, 41]. Furthermore, the study confirmed the robustness and effectiveness of the model as a prognostic biomarker for PAAD across three distinct cohorts. To evaluate the significance of the risk model in terms of patient survival, a nomogram was constructed. Results from the nomogram indicated that the risk score model outperformed various factors such as age, N stage, T stage, grade, and risk score in predicting survival outcomes. In this study, a thorough analysis was conducted on various factors including gene mutations, drug sensitivity potential, and immune infiltration, comparing them among the two groups. Consistent with previous research [42], the TP53 gene mutations exhibited the most significant differences in mutation profiles between the two risk categories, with a notably higher prevalence observed in the high-risk group. Moreover, there was a substantial contrast in the IC50 levels of chemotherapy drugs or inhibitors. Our study showed that the low risk score PAAD patients were more sensitive to the drugs BMS-754807(IGF-type 1 receptor inhibitor), Sabutoclax (Bcl-2 inhibitor), GSK2578215A (leucine-rich repeat kinase 2 (LRRK2) inhibitor), Venetoclax (Bcl-2 inhibitor), AZD8055 (mTORC inhibitor), while the high risk score PAAD patients were more sensitive to the drugs Trametinib and Selumetinib (MEK inhibitors) and SCH772984 and ERK_6604 (ERK inhibitors), both the MEK inhibitors and ERK inhibitors drugs had effective treatment in KRAS mutant pancreatic cancer [43,44,45].

The study also discovered that low-risk patients had higher levels of B cells, CD8 T cells, and tumor-infiltrating lymphocytes, suggesting a more favorable immune profile [46]. Conversely, high-risk patients exhibited elevated levels of immature dendritic cells (iDCs), macrophages, neutrophils, Th2 cells, and regulatory T cells (Tregs), indicating an immunosuppressive tumor microenvironment (TME) [47]. The results underscore the MMRGs risk model's promise as a valuable biomarker for predicting molecular characteristics, immune status, and responsiveness to chemotherapy in individuals diagnosed with PAAD.

Mitochondrial metabolic and signaling disruptions play a dual role in cancer: they not only promote tumor growth but also may lead to resistance against conventional therapies [48]. A key aspect of this metabolic reprogramming is the shift to aerobic glycolysis, known as the Warburg effect, which increases ATP production to meet the growing energy demands of the tumor [49]. In pancreatic cancer, this reprogramming is particularly associated with heightened tumor aggressiveness, treatment resistance, and a poorer prognosis [50]. To counteract these effects, potential anti-cancer medications that target mitochondrial functions are under investigation. Currently, clinical studies are exploring the use of antagonists against pyruvate dehydrogenase and α-ketoglutarate dehydrogenase enzymes, focusing on a more relevant target within the TCA cycle for PAAD [51]. Mitochondrial dysfunction significantly alters the tumor microenvironment in pancreatic cancer, impacting immune cell regulation and contributing to the disease's immune evasion mechanisms [52]. It affects metabolic pathways, leading to increased reactive oxygen species (ROS) that in turn trigger the release of inflammatory cytokines such as IL-6 and TNF-α. This shift in the microenvironment can impair the function of T cells and other immune cells [53]. Moreover, mitochondria are implicated in the regulation of immune checkpoint expression, with enhanced aerobic glycolysis in tumor cells potentially increasing the expression of inhibitory receptors PD-1 and LAG-3 on tumor-infiltrating lymphocytes, associated with reduced mitochondrial mass and glucose uptake [54]. These factors may augment the tumor's ability to evade immune responses. The manipulation of mitochondrial dynamics and inhibition of specific metabolic enzymes or transporters represent a promising strategy to potentiate the effectiveness of anti-tumor immunotherapies, particularly by enhancing the activity of CD8 + T cells and B lymphocytes [55, 56]. This approach could offer a novel therapeutic avenue for overcoming resistance and improving patient outcomes. Pan et al. found that venetoclax, an FDA-approved drug, enhances NK cell-mediated killing of tumor cells by inducing mitochondrial apoptosis and inhibiting the BCL2 protein [57]. Synergy with this effect was noted when combining venetoclax with BCL-XL and MCL1 inhibitors, pointing to promising combination therapies for cancer treatment.

This study has limitations. Firstly, the data, sourced from a database, require clinical verification to ensure the reliability of mitochondria-related gene predictions in clinical practice. Secondly, the impact of mitochondrial changes on the PAAD immunophenotype warrants investigation to elucidate the role of these genes in PAAD's immune evasion. Additionally, the study's suggested sensitivity drugs necessitate validation through laboratory and clinical trials.

In the future, personalized treatment strategies, underpinned by multi-omics data integration, represent a promising frontier in cancer research. This approach allows for the development of tailored treatment plans that target the specific genetic and microenvironmental profiles of individual tumors. The potential of such precision medicine to revolutionize pancreatic cancer treatment is immense, offering hope for improved patient outcomes and survival.

Conclusion: We have developed a novel model that can predict prognosis, molecular characteristics, immune status, and sensitivity to chemotherapy drugs, validated through a prognostic risk score. Which may contribute to further understanding of the molecular mechanisms underlying PAAD development. This will enrich the diagnostic and therapeutic strategies for PAAD patients, thus providing a new strategy for targeting mitochondria and immunotherapy.