Abstract Disulfidptosis, a novel form of disulfide stress-induced cell death involved in tumor progression, hasn’t be well defined the function in tumor progression. And the clinical impacts of disulfidptosis-related genes (DRGs) in pancreatic adenocarcinoma (PAAD) remain largely unclear. In this study, we identified two distinct disulfidptosis subtypes and found that multilayer DRG alterations were associated with prognosis and TME infiltration characteristics. A three-gene prognostic signature was constructed to predict prognosis, and its clinical significance was characterized in the TCGA-PAAD cohort. The disulfidptosis signature was significantly correlated with prognosis, molecular subtype, CD8 T-cell infiltration, response to immune checkpoint inhibitors and chemotherapeutic drug sensitivity, and its predictive capability in PAAD patients was validated in multiple cohorts. Meanwhile, two anti-PD-L1 immunotherapy cohorts confirmed that low-risk patients exhibited substantially enhanced clinical response and treatment advantages. Furthermore, the expression patterns of DRGs were validated by quantitative real-time PCR. The expression and prognostic predictive capability of GLUT1 were verified by 87 PAAD patients from our cohort. These findings may help us understand the roles of DRGs in PAAD and the molecular characterization of disulfidptosis subtypes. The disulfidptosis signature could be a promising biomarker for prognosis, molecular subtypes, TME infiltration characteristics and immunotherapy efficacy. Supplementary Information The online version contains supplementary material available at 10.1007/s12672-025-02053-w. Keywords: Disulfidptosis, Pancreatic adenocarcinoma, Molecular subtypes, Tumor microenvironment, Immunotherapy Introductions Pancreatic adenocarcinoma (PAAD) is one of the deadliest malignancies, with a 5-year survival rate of ~ 11% [[28]1]. Approximately 62,210 new cases and 49,830 deaths from PAAD occurred in the USA in 2022 [[29]1]. The mortality of PAAD is increasing in both sexes, and PAAD is expected to become the second leading cause of cancer-related death by 2030 after lung cancer [[30]2]. Many factors, such as difficult early diagnosis, high tumor heterogeneity and pervasive therapeutic resistance, result in a dismal prognosis; nevertheless, efficient treatment remains limited [[31]3]. Immunotherapy is another important approach for PAAD in addition to surgery and chemotherapy, but its application in PAAD is very limited, which is associated with the strong immunosuppressive tumor microenvironment (TME) and unique biological behavior of PAAD [[32]4, [33]5]. Recent insights into the biology and genomics of PAAD have informed the genetic classification and evolutionary patterns of tumor progression and may provide effective treatment options to overcome the defects in its treatment [[34]6, [35]7]. However, due to the complex molecular profile and tumor heterogeneity, most genetic alterations of PAAD remain poorly characterized, limiting their clinical translation [[36]7]. Therefore, a novel biomarker is urgently needed for prognostic stratification and therapeutic effects to enhance the prognosis of patients with PAAD. The cystine/glutamate antiporter solute carrier family 7 member 11 (SLC7A11, also commonly known as xCT) promotes cystine uptake for glutathione biosynthesis, which plays a crucial role in protecting against cancer cell oxidative stress and ferroptosis [[37]8, [38]9]. However, it was reported that SLC7A11 significantly induced cell death under glucose starvation conditions [[39]10, [40]11]. Recently, this type of cell death has been demonstrated as a novel form of regulated cell death distinct from known mechanisms of cell death, which was named “disulfidptosis” [[41]12]. In cells with high expression of SLC7A11 under glucose starvation, high cystine uptake coupled with nicotinamide adenine dinucleotide phosphate (NADPH) depletion triggered disulfide stress, leading to aberrant disulfide bonding among actin cytoskeleton proteins, actin network collapse, and ultimately cell death [[42]12]. Disulfidptosis is mainly found in cancer cells with high expression of SLC7A11; coincidentally, the expression of SLC7A11 is upregulated in multiple human cancers, including PAAD [[43]9, [44]13, [45]14]. These findings suggested that induction of disulfidptosis in SLC7A11-high cancer cells may be an effective strategy for cancer therapy. However, the role of disulfidptosis in cancer progression is not well defined, and the comprehensive role of these DRGs in PAAD phenotyping and the tumor microenvironment (TME) is unclear. In this study, pancancer analysis was first utilized to systematically analyze the 15 DRGs in 33 cancer types. Then, 644 PAAD samples were divided into two different disulfidptosis subtypes based on the expression levels of DRGs, and the prognosis and TME cell infiltration differences between the two subtypes were evaluated. In addition, a disulfidptosis signature was constructed to predict prognosis and characterize the immune landscape of PAAD. The findings demonstrated that the disulfidptosis signature is a powerful prognostic marker. Materials and methods Pancancer analysis The Cancer Genome Atlas (TCGA) pancancer data, which includes transcription expression data, clinical data, immune subtypes, stemness scores based on DNA methylation (DNA stemness score, DNAss) and mRNA (RNA stemness score, RNAss), were obtained from the Xena platform [[46]15]. TME analysis utilized expression data of immune and stromal scores from ESTIMATE [[47]16]. Spearman correlation analysis was used to examine the links between the DRGs and scores. ANOVA assessed the correlation between DRGs levels and the six immune subtypes [[48]17]. Moreover, correlations between DRGs expression and RNAss/DNAss were assessed by Spearman correlation analysis [[49]18]. The pancancer analysis include 33 TCGA cancer types including pancreatic adenocarcinoma (PAAD). NCI-60 analysis The drug sensitivity of the DRGs was evaluated using the The National Cancer Institute (NCI)-60 database, available through the CellMiner interface ([50]https://discover.nci.nih.gov/cellminer/). To examine the correlations between DRGs and drug sensitivity, Pearson correlations were performed between DRG expression and corresponding GI50 [[51]15]. 262 FDA-approved and clinical trial drugs were included in the correlation analysis. Collection of data on PAAD patients Transcriptome data and the relevant clinical data of PAAD were obtained from TCGA and the Gene Expression Omnibus (GEO). Six GEO cohorts ([52]GSE21501, [53]GSE62452, [54]GSE28735, [55]GSE85916, [56]GSE71729 and [57]GSE57495) and TCGA-PAAD cohort were obtained for subsequent analyses (Table S1). The PAAD patients were diagnosed by surgery and postoperative pathological diagnosis. Raw “cel” files were retrieved from the GEO dataset, and background adjustment along with quantitative normalization were performed. The seven datasets were merged, and batch effects were eliminated by performing the “combat” algorithm [[58]15]. Samples without survival information or died in 30 days were excluded, 644 PAAD patients were included for subsequent analyses. Moreover, data of 90 PAAD samples from the Pancreatic Cancer‑AU (PACA-AU) were sourced from the International Cancer Gene Consortium (ICGC), and 288 PAAD samples from E-MTAB-6134 dataset were obtained from ArrayExpress database; these data were used for external validation. For TCGA-PAAD cohort only include 4 normal samples, 167 normal pancreas sample data were obtained from the Genotype-Tissue Expression (GTEx) project to analysis the expression profiles of 15 DRGs between PAAD and normal samples. Unsupervised clustering for DRGs R package “ConsensusClusterPlus” performed consensus unsupervised clustering analysis to categorize patients to different subtypes of disulfidptosis, according to DRGs expression. The consensus clustering algorithm confirmed the ideal number of stable subtypes, with 1000 repetitions were conducted to confirm the classification stability [[59]15]. Functional and pathway enrichment analysis “GSVA” R package was used for conducting gene set variation analysis (GSVA) to investigate the biological differences among disulfidptosis subtypes [[60]15]. The Molecular Signatures Database (MSigDB) database provided the “c2.cp.kegg.v2022.1.Hs.symbols” hallmark genes used for GSVA. In addition, The R package “GSEA” (version 4.2.3) was applied to assess the risk groups and investigate potential cellular pathways. Tumor microenvironment (TME) immune infiltration analysis To investigate the immune cell infiltration, CIBERSORT [[61]19] algorithm was used to evaluated the relative proportions of immune cells. Single-sample gene set enrichment analysis (ssGSEA) was conducted, to estimate the relative abundance of infiltrating immune cells. Each sample showed the relative abundance of infiltrating immune cells via ssGSEA. The ESTIMATE evaluated the immune and stromal scores. In addition, multiple methods including TIMER, CIBERSORT, CIBERSORT-ABS, XCELL, MCP-COUNTER, QUANTISEQ and EPIC algorithms were utilized to analyze the abundances of immune cells [[62]15]. Construction and validation of the prognostic signature According to the 15 DRGs expression profile and corresponding clinical data, a three-gene prognostic model was created by the least absolute shrinkage and selection operator (LASSO) and stepwise multivariate Cox regression analysis in all PAAD patients. Each sample’s risk score was determined by summing the expression levels of each gene and multiplying the corresponding coefficients. Furthermore, the formula was also applied in multiple cohorts to validate the signature. The “rtsne” and “ggplot2” packages were used to perform principal component analysis (PCA) for evaluating classification ability. The R package “survminer” support survival analysis. The “survivalROC” R package plotted the receiver operating characteristic (ROC) curve, evaluating the specificity and sensitivity. In addition, prognostic signature nomogram was plotted by the “rms” package. The calibration curve was also plotted to verify the consistency between the OS rates predicted by the nomogram and those observed. Finally, we utilized decision curve analysis (DCA) to analyze the net benefits of various predictors using the “rmda” package [[63]15]. Mutation and drug susceptibility analysis To examine somatic mutations in PAAD patients across two risk groups, we used the mutation annotation format (MAF) from the TCGA and analyzed it with R package “maftools”. We assessed each patient’s tumor mutation burden (TMB) score, and assessed its correlation with risk score was evaluated in both groups. To further examine the differences in drug sensitivity, the half-maximal inhibitory concentration (IC50) values of common PAAD chemotherapeutic drugs was performed using the “pRRophetic” package. To explore the direct predictive value and correlations of the risk score for the anti-PD-L1 therapy response, the metastatic urothelial carcinoma cohort [[64]20] and metastatic melanoma cohort [65]GSE78220 [[66]21] were included. Cell lines and tissue samples Human pancreatic ductal epithelial cells HPDE6-C7 and PAAD cells (AsPC-1, BxPC-3, CFPAC-1 and PANC-1) were obtained from certified manufacturer and cultured in a complete growth medium according to the instructions. 87 PAAD paraffin-embedded samples (10 with adjacent non-tumor tissues) were obtained from the Department of Hepatopancreatobiliary Surgery, Chongqing General Hospital (Chongqing, China) (from April 2012 to August 2015). Organ donors provided the normal pancreatic tissues. PAAD patients did not receive chemotherapy or radiotherapy before surgeries. Written informed consent and ethics approval are stated in the declaration section. Quantitative real-time PCR (qRT‒PCR) RNA extraction and reverse transcription were conducted according to kits’ instruction, as our former work [[67]15]. qRT–PCR was implemented on the CFX96 (Bio-Rad) using TB Green Premix (Takara). We calculate the relative expression levels and normalized to GAPDH levels, using the 2-ΔΔCT method [[68]22]. Primers summaried in Table S2. Immunohistochemistry Immunohistochemistry (IHC) was conducted according to the DAB plus kit instructions (ZSGB-Bio, Beijing, China), as which in our previous work [[69]23]. The anti-GLUT1 antibody was purchased from Proteintech (Chicago, USA). We calculated the IHC scores based on the percentage of positive stained (0, 0%; 1, < 10%; 2, 10–50%; 3, > 50%) and the intensity of staining (0, none; 1, weak; 2, moderate; 3, strong). IHC Score = percentage score*(intensity score + 1). Statistical analysis R (version 4.1.0) was used in our computational and statistical analyses. Wilcoxon tests examined difference between two groups. In comparisons between groups, P < 0.05 was identified as statistically significant. *P < 0.05; **P < 0.01; ***P < 0.001. Results Disulfidptosis-related gene (DRG) expression across cancers We curated the catalog of 15 genes (SLC7A11, SLC3A2, RPN1, NCKAP1, BRK1, WASF2, ABI2, CYFIP1, RAC1, NUBPL, NDUFA11, OXSM, NDUFS1, GYS1 and SLC2A1) that function closely with disulfidptosis [[70]12]. Based on TCGA pancancer data, we examined the intrinsic expression profile of 15 DRGs in all thirty-three cancer types. The DRGs expression varied greatly from one to others, with SLC2A1 having the highest heterogeneity in pancancer (Fig. [71]1A). The expression of 15 DRGs varied between tumor and normal tissues (Fig. [72]1B and Fig. S1). SLC2A1 expression showed the greatest intertumor heterogeneity, with very low levels in some tumors (KICH, KIRP, and PRAD), whereas LUSC expressed high level of SLC2A1. (Fig. [73]1B). Interestingly, LUSC also has a high expression of SLC7A11, indicating the therapeutic potential of inducing disulfidptosis by using glucose transporter (GLUT) inhibitors in the LUSC [[74]12]. According to Spearman correlation, NCKAP1 and CYFIP1 showed the highest level of correlation (r = 0.46, P = 0.0001), indicating some potential interaction between the two genes (Fig. [75]1C). Different cancer types have different prognostic significance for the same gene (Fig. [76]1D), in ACC and KICH, SLC7A11 was negatively associated with OS, while in OV and READ, it was positively associated with OS (Fig. S2). By inducing disulfideptosis, DRGs may affect tumor sensitivity to antineoplastic agents. Using the NCI-60 panel, we explored the association between DRG expression levels and drug sensitivity. Figure [77]1E summarized the top 16 gene-drug pairs ranked by Pearson correlation coefficient, and Table S3 recorded the whole analysis results of drug sensitivity. CYFIP1 and NCKAP1 showed the strongest correlation with drug sensitivity; The CYFIP1 gene was associated with 71-drug sensitivity, while NCKAP1 was associated with 80-drug sensitivity. Eight drugs correlated with CYFIP1 had a correlation coefficient over 0.5 (including bendamustine, valrubicin, XK-469, epirubicin, teniposide, etoposide, BN-2629 and carmustine). Many of the drugs were reported as antitumor drugs, like bendamustine to lymphoma, and valrubicin is used to treat bladder cancer. It appears that some tumor-targeted therapies are affected by disulfidptosis. Fig. 1. [78]Fig. 1 [79]Open in a new tab Pancancer analysis of 15 DRGs. A The DRGs expression in pancancer TCGA data. B Heatmap of DRGs comparing tumor and normal samples across 19 tumor types with over four normal samples. C Gene expression correlations among the 15 DRGs (Spearman correlation test). D Forest plots showed hazard ratios of DRGs for tumors. E Scatter plots showed correlations between DRG expression and drug sensitivity (top 16 by P value) To understand associations between DRGs and immune components, TCGA pancancer data was examined to see how DRGs correlate with immune infiltrates. Six types of immune infiltrated identified in human tumors: C1 (wound healing), C2 (INF-γ dominant), C3 (inflammatory), C4 (lymphocyte depleted), C5 (immunologically quiet), and C6 (TGFβ dominant) [[80]24]. As shown in Fig. S3A, 15 DRGs expression levels were different across different ISs (all P < 0.001). Furthermore, TME analysis revealed that the DRGs were associated with immune and stromal cell infiltration across various tumor types (Fig. S3B and C). These findings implied that the disulfidptosis process may affect the TME and tumor immunity. Finally, tumor stemness feature analysis of the 15 DRGs was performed. As tumors progress, the tumor cells may gradually lose their differentiated characteristics and gain stem-like properties. The stemness of tumors can be assessed using RNAss and DNAss [[81]18]. The DRGs displayed different levels of correlation with RNAss and DNAss in distinct cancer types (Fig. S4A and B). The five disulfidptosis suppressor genes, including NUBPL, NDUFA11, OXSM, NDUFS1 and GYS1, were positively correlated with DNAss and RNAss in most types, which suggested that disulfidptosis may cause tumor heterogeneity and promote tumor progression. In addition, NCKAP1 and CYFIP1 were negatively correlated with RNAss in thymoma (THYM), while they showed a noticeably positive relationship with DNAss in THYM. These inconsistant results proposed that DNAss and RNAss might identify unique cancer cell populations with diverse characteristics or stemness degrees. Genetic changes and expression variations of DRGs in PAAD Firstly, we assessed the landscape of mutation profiles in the TCGA-PAAD cohort. Somatic mutation analysis of the 15 DRGs showed a low mutation frequency in the PAAD cohort (Fig. S5A). Among the 173 samples, only four patients had DRG mutations. Next, we explored copy number variation (CNV) in these DRGs and revealed prevalent CNV in all DRGs. Among them, WASF2, NDUFA11 and GYS1 had widespread CNV decreases, while NDUFS1, ABI2, NCKAP1 and OXSM showed CNV increases (Fig. [82]2A). The types of CNV in DRGs and their respective positions on the chromosome are shown in Fig. [83]2B. We conducted a deeper analysis of the links between CNV and DRG expression levels in PAAD, finding that most DRG expression levels did not correlate with CNV variations. For instance, some PRGs with CNV gain, such as NDUFS1, ABI2, NCKAP1 and OXSM, showed downregulated mRNA expression, while WASF2 with high frequencies of CNV loss was upregulated in tumor samples (Fig. [84]2A and C). Thus, CNV is not the sole factor influencing DRG expression. Other factors, including the DNA methylation [[85]25] and transcription factors [[86]26], can also affect gene expression. Moreover, Spearman correlation analysis indicated the associations of the 15 DRGs (Fig. S5B). The relationship between DRG expression levels and clinical stages, grades, and immune subtypes was also examined in PAAD. Figure [87]2D and E indicated that expression level of ABI2, RAC1, NDUFA11 and SLC2A1 were different in different clinical stages, NCKAP1, CYFIP1, RAC1 and SLC2A1 were expressed differently in distinct grades. Moreover, RPN1, RAC1, OXSM, GYS1 and SLC2A1 were differentially expressed among immune subtypes (Fig. S5C). The Pearson correlation revealed that the DRGs had distinct levels of relationhip with tumor stemness and TME in PAAD (Fig. S6). Our results showed a significant difference in both DRG levels and genetic landscape between PAAD and normal samples, indicating the potential role of DRGs in PAAD tumorigenesis. Fig. 2. [88]Fig. 2 [89]Open in a new tab DRG genetic and transcriptional alterations in PAAD. A Frequencies of copy number variation (CNV) gain and loss of DRGs. B The types of CNV in DRGs and their respective positions on the chromosome. C Wilcoxon rank sum test showed expression patterns of 15 DRGs in PAAD and normal tissues basd on TCGA-GTEx data (*** P < 0.001). D, E The correlation across DRGs and tumor stages (D), grades (E). ANOVA was used. *P < 0.05; **P < 0.01; ***P < 0.001 Identification subtypes of disulfidptosis in PAAD We combined seven PAAD cohorts (TCGA-PAAD, [90]GSE21501, [91]GSE62452, [92]GSE28735, [93]GSE85916, [94]GSE57495 and [95]GSE71729) and removed batch effects to explore the expression patterns of DRG related to tumorigenesis. Kaplan–Meier and Cox regression analyses confirmed the prognostic significance of 15 DRGs in 644 PAAD patients. (Fig. S7, Table S4). Furthermore, the network delineated the landscape of DRG interactions, regulatory connections, and their prognostic value in patients with PAAD (Fig. [96]3A). Then We conducted unsupervised clustering to categorize PAAD patients according to the expression of 15 DRGs, finding that k = 2 was optimal for dividing the cohort into two groups (Fig. [97]3B–D and Fig. S8). Cluster An included 344 cases, and Cluster B included 300 cases. PCA revealed that DRG expression levels significantly differentiated the two clusters (Fig. [98]3E). Survival analysis showed patients of subtype B had a longer OS than subtype A patients (log-rank test, P < 0.001, Fig. [99]3F). As expected, most DRGs showed differential expression between groups (Fig. [100]3G). TCGA-PAAD validated clustering repeatability with larger cohort, identifying two clusters via consensus clustering (Fig. S9A-H). Kaplan–Meier curves indicated longer survival in subtype B patients (Fig. S9I). PCA confirmed two distinct subtypes (Fig. S9J), further proving the two disulfidptosis clusters in PAAD. Fig. 3. [101]Fig. 3 [102]Open in a new tab Identification of disulfidptosis-related subtypes. A DRGs interaction network in PAAD. B, C The consensus CDF curves for k values 2 to 7. D Consensus clustering and matrix heatmap of 15 DRGs( k = 2). E PCA revealed two distinct DRGs expression subtypes. F Kaplan‒Meier curves for the two disulfidptosis subtypes. G Boxplot of DRG expression patterns between two disulfidptosis clusters. CDF: cumulative distribution function Characteristics of biological function and TME in the disulfidptosis subtypes GSVA analysis indicated that Cluster A was enriched in cancer-related pathways, such as pancreatic cancer, thyroid cancer, bladder cancer, cell cycle and P53 signaling pathway, while Cluster B considerably manifested in some amino acid biosynthesis processes, such as threonine and glycine serine metabolism, tryptophan metabolism and bile acid biosynthesis (Fig. [103]4A). Subsequently, CIBERSORT algorithm was performed to investigate the links across the disulfidptosis subtypes and immune infiltration. Figure [104]4B indicated variations in infiltration of 22 immune cell types in each PAAD sample, implying that this is a fundamental characteristic reflecting individual differences. We observed great differences in most infiltrating immune cell types between the two subtypes (Fig. [105]4C). Subtype B, which has a better prognosis, showed higher infiltration of CD8 + T cells, naive B cells, gamma delta T cells, follicular helper T cells, monocytes, resting mast cells and eosinophils. In contrast, subtype A had more infiltrated activated CD4 + memory T cells, M0 macrophages, neutrophils and activated NK cells. The ssGSEA algorithm was used to calculate enrichment score of immune cell infiltration to confirm these differences between two clusters. Consistently, cluster B had higher activated CD8 + T cells, activated B cells, follicular helper T cells and eosinophils, whereas subtype A had more activated CD4 + T cells, type 2 helper T cells and neutrophils (Fig. [106]4D). In addition, we analyzed the TME scores (immune, stromal, and ESTIMATE score) between the two clusters, and the results demonstrated higher TME scores in patients with subtype B (Fig. [107]4E). To predict responses to immune checkpoint blockade (ICB) therapies, we analyzed immune checkpoint gene expression in two clusters. Subtyple B showed significantly higher level of CTLA-4 and PD-1, while subtype A has higher PD-L1 level (Fig. [108]4F). Combined, these results suggested that DRGs had a critical impact on PAAD progression. Fig. 4. [109]Fig. 4 [110]Open in a new tab Biological characteristics and TME cell infiltration in disulfidptosis subtypes of PAAD. A GSVA analysis showed activated (red) and inhibited (blue) pathways between. B CIBERSORT revealed immune cell proportions in every PAAD patient. C CIBERSORT assessed infiltrating immune cell abundance in two subtypes. D ssGSEA evaluated infiltrating immune cell abundance in two subtypes. E Correlations between the subtypes and TME score. F Expression differences of PD-L1, PD-1 and CTLA-4 in the two subtypes Construction of the disulfidptosis prognostic signature The LASSO regression algorithm was used to screen these 15 DRGs, constructing a disulfidptosis signature, and four genes were retained according to the minimum partial likelihood deviance (Fig. S10A and B). Moreover, three genes were identified using multivariate Cox proportional hazards regression analysis, and a risk score was derived based on their risk coefficients and these genes (Table S5). Interestingly, all three genes were identified as independent predictive factors (Fig. [111]5A). The risk score was calculated as: (0.222 * expression level of WASF2 + (-0.298) * expression level of NUBPL + 0.175 * expression level of SLC2A1). The PAAD samples were divided into two groups (high-risk and low-risk) by the median score. The risk score and the distribution of survival status of every PAAD patients are shown in Fig. [112]5B and C. Heatmap illustrated the expression of the three genes in the two groups (Fig. [113]5D). PCA results showed that the patients in different groups were distributed in two directions (Fig. [114]5E). The Kaplan‒Meier curve indicated that high-risk group patients had significantly poorer overall survival than in low-risk group (Fig. [115]5F, P < 0.001). ROC analysis was performed to determine whether survival predictions made with the prognostic signature were accurate (Fig. [116]5G). Next, the process of constructing the prognostic model and the distribution of PAAD patients in different subtypes was illustrated in a Sankey diagram (Fig. [117]5H). Correlation analyses showed that a low-risk score was significantly linked to survival status, whereas a high-risk score was correlated with death status in PAAD patients (Fig. [118]5I, J). Furthermore, Cluster B with a better prognosis was significantly associated with survival status (Fig. [119]5K), and the risk score of Cluster B was significantly lower than that of Cluster A (Fig. [120]5L). The results showed that the risk score is highly correlated with prognosis and may help predict disulfidptosis clusters in PAAD. Fig. 5. [121]Fig. 5 [122]Open in a new tab Development of a prognostic disulfidptosis-related gene signature. A Three DRGs (WASF2, NUBPL and SLC2A1) were identified to construct a risk signature by multivariate Cox regression analysis. B, C Risk scores and survival statuses distribution in PAAD patients. D Heatmap showed the expression of three model genes across risk groups. E PCA plot of the PAAD series. F Kaplan‒Meier curves of high-risk and low-risk PAAD patients. G AUC of ROC curves cinfirmed the risk score prognostic accuracy. H Alluvial diagram of subtype distributions by risk score and outcome. I, J Correlation of risk score and survival outcome. K Relationships between disulfidptosis clusters and survival outcome. L Differences in risk scores between the two disulfidptosis groups Characteristics of biological function and TME cell infiltration in the prognostic signature GSVA showed that the high-risk group was enriched in cancer-related pathways, like thyroid and bladder cancer, as well as Notch and Hedgehog signaling. In contrast, low-risk group was enriched in amino acid metabolism pathways, including glycine, serine and threonine metabolism, tryptophan, beta-alanine and histidine metabolism (Fig. [123]6A and Table S6). Spearman correlation analysis using ssGSEA revealed that the risk score was inversely related to activated B cells, activated CD8 T cells, eosinophils, mast cells and monocytes, while it was positively associated with activated CD4 T cells, activated dendritic cells, natural killer cells, neutrophils and Th17 cells (Fig. [124]6B and Fig. [125]6C). CIBERSORT algorithm indicated that higher levels of CD8 T cells, follicular helper T cells, gamma delta T cells, monocytes, M1 macrophages and mast cells in the low-risk group (Fig. S11A and B). Scatter plots confirmed similar trends between risk score and immune cell infiltration (Fig. S11C-L). Comparative analysis assessed differences in immune functions, including APC coinhibition, APC costimulation, cytolytic activity, inflammation promotion, MHC class I, parainflammation, type I IFN response and type II IFN response across groups (Fig. [126]6D). The higher TME scores in the low-risk group patients were found in the TME difference analysis (Fig. [127]6E). Finally, expression analysis of checkpoint genes showed that 30 immune checkpoint genes expression were different in two risk groups (Fig. [128]6F), implying the potential biomarker value of the risk score for checkpoint-based immunotherapy. Fig. 6. [129]Fig. 6 [130]Open in a new tab Characteristics of biological function and TME cell infiltration in the prognostic signature. A GSVA showed activated (red) and inhibited (blue) pathways in two groups. (B) Correlation of risk score with immune cell infiltration from ssGSEA results. C, D Boxplots of ssGSEA scores for immune cell (C) and immune functions (D) across two groups. E TME score correlations with the two risk groups. (F)Comparison of immune checkpoint gene expression between the groups. The Kruskal–Wallis test was used to assess differences between groups. ns not significant; *P < 0.05, **P < 0.01, ***P < 0.001 Validation of the prognostic signature First, the TCGA-PAAD patients were stratified to high-risk and low-risk groups according to the same risk score formula mentioned above (Table S5). Kaplan‒Meier curves indicated that high risk PAAD patients had noticeably poorer OS than low-risk patients (log-rank test, P = 0.016, Fig. [131]7A). Then, PCA confirmed the classification ability of the risk signature (Fig. [132]7B). ROC curves revealed that the AUC of the risk signature for OS was 0.651 at 5 years (Fig. [133]7C). The risk score and distribution of survival status of each PAAD sample are shown in Fig. [134]7D and E. In addition, heatmap showed the correlations between the risk signature and clinicopathological variables (Fig. [135]7F). We developed a nomogram integrating risk signature and clinicopathological factors to predict the 1-, 3-, and 5-year survival rates of PAAD patients, offering a practical approach for clinical prognosis (Fig. [136]7G). The nomogram’s calibration plots confirmed that the predicted OS rate was close to the actual OS rate at 1, 3, and 5 years, highlighting its robust predictive strength (Fig. [137]7H). Finally, DCA revealed that the nomogram constructed on the risk score provide better clinical practicality in predicting the prognosis of PAAD (F[138]ig. [139]7I). Fig. 7. [140]Fig. 7 [141]Open in a new tab Characteristics of the risk score in the TCGA. A Kaplan‒Meier OS curves for the high-risk and low-risk groups. B PCA for the high-risk and low-risk groups. C ROC curve for survival model prediction. D, E Risk score and survival status distribution in PAAD patients. F Heatmap of model genes and clinicopathological factors in the two groups. G Prognostic nomogram for 1-, 3-, and 5-year survival prediction. H Calibration curves for nomogram predictions of 1-year, 3-year, and 5-year survival. I DCA assessed the clinical practicality of the nomogram for survival prediction We also calculated risk scores to further validate the prognostic performance of the risk score, the E-MTAB-6134 cohort, PACA-AU cohort and five GEO groups ([142]GSE62452, [143]GSE21501, [144]GSE85916, [145]GSE71729 and [146]GSE57495) were enrolled. The patients were stratified into low-risk or high-risk score groups using the same formula (Table S5). Survival analysis demonstrated that the prognosis was notably better in the low-risk group than in the high-risk group in all cohorts (Fig. S12-18A). Analysis of the 5-year prognostic prediction classification efficiencies revealed that the risk score still had relatively high AUC values (Fig. S12-18B), suggesting that the risk score had an excellent ability to predict the survival of PAAD patients. The PCA presenting the variation tendencies, risk scores and patient survival outcomes of the high-risk and low-risk groups are shown in Fig. S12C-E, Fig. S13C-E, Fig. S14C-E, Fig. S15C-E, Fig. S16C-E, Fig. S17C-E and Fig. S18C-E, respectively. Characteristics of pathway enrichment and immune infiltration of the prognostic signature in the TCGA cohort GSEA revealed that the high-risk group had enriched biological functions primarily in cancer-related pathways, including the P53 signaling pathway, pancreatic cancer, pathways in cancer, cell cycle, and Wnt signaling pathways, while the low-risk group was mainly associated with some amino acid biosynthesis processes, including glycine serine and threonine metabolism and primary bile acid biosynthesis (Fig. [147]8A and Table S7). These findings matched the results obtained from all PAAD patients. Next, a bubble chart of immune infiltration based on the TIMER, CIBERSORT, CIBERSORT-ABS, QUANTISEQ, MCP-counter, XCELL and EPIC algorithms is shown in Fig. [148]8B. All the immune cells in the bubble chart were significantly differentially expressed between the two risk groups (all P < 0.05; Fig. S19). The ssGSEA algorithm compared immune functions and immune cells to disclose the differences in CD8 T cells, mast cells, NK cells, pDCs, cytolytic activity, HLA, and type II IFN response between the two groups (Fig. [149]8C, D). Similarly, the CIBERSORT algorithm also showed that the ratio of CD8 + T cells decreased significantly in high-risk group patients (Fig. [150]8E). The low-risk group had a substantially higher TME immune score than the high-risk group (Fig. [151]8F). Moreover, 18 immune checkpoint genes showed differenct expression levels between the two score groups (Fig. [152]8G). These findings indicate that the prognostic signature was correlated with immune cell infiltration to a certain extent. Fig. 8. [153]Fig. 8 [154]Open in a new tab Correlations of the risk score with biological characteristics and immune cell infiltration in the TCGA cohort. A GSEA showed gene enrichment differences in KEGG between two groups. B Bubble chart illustrated immune cell infiltration using various algorithms between two groups. C, D Boxplots of ssGSEA displayed immune cell scores (C) and immune function scores (D) between two groups. E Violin plot of CIBERSORT depicted immune cell infiltration between two groups. F Correlations between two risk groups and TME score. G Immune checkpoint gene expression between two groups. The Kruskal–Wallis test was used to assess differences between groups. ns not significant; *P < 0.05, **P < 0.01, ***P < 0.001 To further investigate the relationship between the risk score and the efficacy of immunotherapy, we used two groups of patients who had received anti-PD-L1 treatment in the IMvigor210 cohort and [155]GSE78220 cohort to test this result. Using the same risk score formula mentioned above (Table S5), the IMvigor210 cohort was divided into high- and low-risk score groups, and patients with low-risk scores substantially outweighed patients with high-risk scores (Fig. S20A). In addition, a low-risk score was associated with a better CR/PR rate and a lower SD/PD rate (Fig. S20B and C). Consistently, patients with a low-risk score had a better prognosis and a better CR/PR rate than the [156]GSE78220 cohort (Fig. S20D-F). These findings provided preliminary evidence that the risk score can be used to predict the efficacy of immunotherapy. Mutation and drug susceptibility analysis Since genetic alterations contribute to tumorigenesis and progression; we analyzed how somatic mutations distribution varied between the two risk score groups from the TCGA-PAAD. The highest top 20 mutation frequency genes were visualized (Fig. [157]9A and B). The top ten mutated genes in the high-risk and low-risk groups were KRAS, TP53, SMAD4, CDKN2A, TTN, MUC16, RNF43, RYR1, TNXB and HECW2. Markedly higher frequencies of KRAS, TP53, and CDKN2A mutations were found in patients of high risk. As 90% of PAAD patients have somatic oncogenic point mutations in KRAS, we also evaluated the links between KRAS mutation and DRG expression, finding a significant correlation with the expression levels of eight DRGs (Fig. S21). Furthermore, a higher TMB was detected in the high-score group than in the low-score group (Fig. [158]9C), suggesting that the high-risk group might benefit from immunotherapy. Correlation analysis revealed that a significant positive correlation also existed between the risk score and TMB (Fig. [159]9D). Finally, we selected chemotherapy drugs currently used for PAAD treatment to detect the sensitivities of patients in the high- and low-risk groups to these drugs. Interestingly, we found that the patients in the low-risk score group had lower IC50 values for camptothecin and rapamycin, while the IC50 values of cisplatin were significantly lower in the patients with high-risk scores (Fig. [160]9E–G). Together, these results suggested that DRGs were associated with drug sensitivity. Fig. 9. [161]Fig. 9 [162]Open in a new tab Mutation and drug susceptibility analysis between the two score groups. A, B Mutation patterns in the top 20 genes of patients with high and low risk. C TMB in different risk score groups. D Spearman correlation analysis of the risk score and TMB. E–G Relationships between risk score and chemotherapy sensitivity Validation of the expression levels of DRGs in cells and tissues To further identify the expression profiles of the three signature DRGs, we analyzed their relative expression levels in the normal cell line HPDE6-C7 and four pancreatic cancer cell lines (AsPC-1, BXPC-3, CFPAC-1 and PANC-1) using qRT‒PCR. The results showed that SLC2A1 and WASF2 were expressed at a higher level in the cancer cell lines than in the normal cell lines, while NUBPL was expressed at a relatively lower level in the cancer cell lines than in the normal controls (Fig. [163]10A–C), which was consistent with the public database results. Moreover, we studied the protein expression levels of DRGs in PAAD using the Human Protein Atlas (HPA) database [[164]27]. Similarly, the protein levels of SLC2A1 and WASF2 were upregulated in PAAD tissues compared with normal pancreatic tissues, while the expression level of NUBPL was downregulated in PAAD compared with normal controls (Fig. S22). SLC2A1, commonly known as glucose transporter 1 (GLUT1), encodes a uniporter protein in humans. As a crucial regulator of glucose metabolism, GLUT1 was identified as a candidate gene for further validation of prognostic prediction, as it had the lowest P value in univariate prognostic analysis (Table S4). Additionally, GLUT1 is upregulated in various types of cancers and indicates a poor prognosis [[165]28]. Nonetheless, the function of GLUT1 in PAAD largely remains unclear. The level of GLUT1 mRNA was higher in PAAD tissues than in the adjacent non-cancerous tissues (Fig. [166]10D and E). The IHC assay further demonstrated that PAAD tissues had a notably higher level of GLUT1 protein than adjacent tissues (Fig. [167]10F). In PAAD cells, GLUT1 is predominantly found in the cytoplasm and membrane. GLUT1 expression levels were determined by IHC staining in 87 paraffin-embedded PAAD samples and confirmed to be overexpressed in PAAD tissue compared to normal pancreatic tissue. (Fig. [168]8G). Survival analysis revealed that patients with high GLUT1 levels had a lower 5-year overall survival rates (P < 0.001, log-rank test, Fig. [169]8H). Additionally, the high expression of GLUT1 was an independent risk prognostic factor to PAAD, according to multivariate Cox regression analysis (HR, 3.649, P < 0.001; Table S8). The prognostic value of GLUT1 was further validated in the E-MEAB-6134 cohort, the PACA-AU cohort and four GEO cohorts (Fig. S23). Finally, we combined the PAAD patients from the five eligible cohorts and performed a meta-analysis (Fig. S24). The results revealed that the prognostic value of GLUT1 was valid (HR = 1.47, 95% CI = 1.25–1.72). Together, these results indicated that the upregulation of GLUT1 promoted the progression of PAAD. Fig. 10. [170]Fig. 10 [171]Open in a new tab Validation of the expression levels of DRGs in cells and tissues. A–C qRT-PCR analysis showed DRG expression in pancreatic cancer cell and normal pancreatic ductal epithelial cells. GAPDH as the internal control. *P < 0.05; **P < 0.01; ***P < 0.001. D, E SLC2A1 expression compared in PAAD and adjacent non-tumorous tissues (ANT) in [172]GSE62452 (D) and [173]GSE28735 (E). F IHC analysis of GLUT1 expression in PAAD and ANT tissues (Scale bar: 100 μm). G IHC staining revealed GLUT1 overexpression in PAAD tissues compared to normal pancreatic tissues (scale bar: 100 μm). H Kaplan‒Meier curves of PAAD patients with low versus high expression levels of GLUT1 Discussion In the past decade, we have gained a deeper understanding of the key variation in genomics in PAAD with the development of the next-generation sequencing [[174]29]. Although previous studies have also identified several molecular subtypes in PAAD, there is still considerable heterogeneity, such as TME infiltration features, within each subtype [[175]30, [176]31]. In addition, molecular heterogeneity affects the clinical outcome of cancer patients; therefore, it is crucial to identify more accurate molecular subtypes for PAAD. Disulfidptosis is a newly identified form of cell death differs from well-established pathways such as apoptosis, necroptosis, pyroptosis, and ferroptosis. Pyroptosis, a form of inflammatory cell death, is characterized by the activation of caspases and the formation of membrane pores by gasdermins, leading to cell lysis and the release of pro-inflammatory cytokines [[177]32, [178]33]. Ferroptosis, defined as an iron-dependent form of regulated cell death, is driven by lipid peroxidation and is distinct in its biochemical and morphological features [[179]34, [180]35]. In contrast to necroptosis, which involves MLKL-mediated disruption of the plasma membrane and is highly inflammatory [[181]36], disulfidptosis does not primarily rely on the same signaling cascades. Instead, disulfidptosis is marked by the rapid depletion of nicotinamide adenine dinucleotide phosphate (NADPH) and abnormal disulfide bond formation in actin cytoskeleton proteins, leading to actin network collapse [[182]12, [183]37]. There are unique metabolic and immune microenvironments in pancreatic adenocarcinoma that significantly influence tumor progression and therapeutic responses. The highly fibrotic and immunosuppressive tumor microenvironment (TME) played a crucial role in modulating the metabolic landscape and immune responses [[184]38, [185]39]. Recent studies have highlighted the metabolic reprogramming that occurs in both PDAC cells and tumor-associated immune cells. This metabolic shift not only supports the aggressive nature of PAAD but also contributes to the immune evasion mechanisms employed by the tumor [[186]40–[187]42]. The unique mechanism of disulfidptosis highlights the importance of metabolic states in regulating cell death pathways, emphasizing the unique role of disulfidptosis in cellular metabolic stress induced cell death, particularly in the context of pancreatic cancer. Considering that disulfidptosis is a complicated multistep cell death process mediated by a series of DRGs, we explored the integrated functions of these DRGs in the molecular classification of PAAD and their characteristics of cell infiltration microenvironment. In this study, pancancer analysis was first performed to comprehensively analyze the 15 DRGs in 33 cancer types, indicating that great heterogeneity of these DRGs was evident in distinct cancer types. These DRGs were critical regulators modulating tumor microenvironment infiltration, cancer stemness and drug resistance. Then, we merged the genomic data of 15 DRGs and identified two distinct disulfidptosis-related subtypes among 644 PAAD samples from seven cohorts. The Survival analysis showed that subtype B had a significant survival advantage. Disulfidptosis Cluster A was mainly enriched in cancer-related pathways, such as pancreatic cancer, thyroid cancer, bladder cancer, the cell cycle and the P53 signaling pathway, while Cluster B was considerably enriched in some amino acid metabolism processes, according to GSVA analysis. Significantly different TME cell infiltration characteristics were observed in the two subtypes. Furthermore, we constructed and validated a disulfidptosis signature based on DRGs for prognostic prediction in PAAD patients. This study may help elucidate the molecular characterization of disulfidptosis subtypes, and the disulfidptosis signature might serve as a predictor for predicting the prognosis and immune landscape of PAAD. As far as we know, there hasn’t been a comprehensive study of DRGs across diverse cancer yet, as well as the association between DRGs and the TME and the clinical impact of PAAD. Here, the 15 DRGs were comprehensively analyzed using TCGA pancancer data. First, the expression and survival analysis proved that the heterogeneity of 15 DRGs is evident across diverse cancer types. Then, drug sensitivity analysis revealed that these DRGs were closely related to chemotherapy resistance. Among them, CYFIP1 and NCKAP1 showed the strongest correlation with drug sensitivity. Of note, our results also provide a new perspective that disulfidptosis may take part in some targeted cancer therapy resistance. Additionally, immune subtype and TME analysis showed that these DRGs were associated with the immune response and TME cell infiltration. Ultimately, we discovered that DRG expression were strongly related to cancer stemness according to cancer stemness indices, providing new perspectives into disulfidptosis. The prognostic signature constructed in this study consisted of three DRGs, among which the expression of WASF2 and SLC2A1 was negatively correlated with the survival risk of PAAD patients, while the level of NUBPL was positively associated with the survival risk. Of note, all three genes were identified as independent prognostic factors. WASF2 (Wiskott-Aldrich syndrome protein family member 2; also known as WAVE2), a member of the Wiskott–Aldrich syndrome protein (WASP) family of actin cytoskeletal regulatory proteins, has been shown to play a crucial role in many types of cancers, including pancreatic, lung, liver, colorectal, prostate, ovary and breast cancer, as well as other hematological malignancies [[188]43]. Yang et al. demonstrated that WASF2 was abnormally expressed in diverse cancer types and significantly associated with OS and progression-free interval (PFI) [[189]44]. In addition, the WASF2 expression level also significantly correlated with the TME [[190]44, [191]45]. In PAAD, high WAVE2 expression is correlated with poor prognosis, and WAVE2 can promote cell motility and invasion by binding to actin cytoskeletal protein alpha-actinin 4 (ACTN4) [[192]46]. Another study reported that monitoring serum levels of WASF2 mRNA may be a useful tool for the early detection of PAAD, as it was highly correlated with the PAAD risk [[193]47]. NUBPL (nucleotide-binding protein-like) is required for assembling human mitochondrial complex I, which is the largest member of the respiratory chain [[194]48]. In addition to its pivotal role in the mitochondrial respiratory chain, NUBPL has been demonstrated to be involved in the development of cancer [[195]49, [196]50]. NUBPL acts as a novel metastasis-related gene that plays an important role in the progression of colorectal carcinoma by inducing epithelial-mesenchymal transition (EMT) through the activation of ERK [[197]50]. As a key regulator of glucose metabolism, GLUT1 has been reported to be upregulated in various types of cancers and to indicate a poor prognosis [[198]28]. Nonetheless, the function of GLUT1 in PAAD largely remains unclear. Previous studies demonstrated that GLUT1 can regulate glucose uptake and maintain glucose metabolism and act as a crucial rate-limiting element in glucose transport in cancer cells [[199]51, [200]52]. Meanwhile, GLUT1 is dysregulated in various cancers to meet the energy requirements for rapid growth and exerts a key role in regulating the differentiation, proliferation and metastasis of cancer cells [[201]51]. Accumulated studies have reported that GLUT1 contributes to aggressive tumor progression in different cancer types, including PAAD [[202]28, [203]53–[204]55]. In this study, GLUT1 was found to be overexpressed in pancreatic cancer cells and tissues, and upregulation of GLUT1 was closely related to the prognosis of PAAD patients in multiple databases and our clinical samples. A recent study reported that GLUT1 inhibitors could induce a novel form of cell death in SLC7A11-high cancer cells, namely, disulfidptosis, and disulfidptosis might mediate the therapeutic effect of GLUT1 inhibitors in the treatment of SLC7A11-high cancers [[205]12]. Excitingly, with the analysis of the three-dimensional crystal structure of GLUT1, it has become possible to design small molecule inhibitors of GLUT1, thereby achieving the goal of inducing disulfidptosis in cancer cells [[206]12]. This has made GLUT1 a therapeutic target that has received much attention. Despite advances in immunotherapy, the efficacy of immunotherapy in patients with PAAD remains far less satisfactory, highlighting the vital role of TME in PAAD progression. The immunosuppressive TME of PAAD is highly heterogeneous and presents challenges for the immunotherapy [[207]4, [208]5]. The TME is a highly complex and dynamic ensemble of cells, of which a variety of immune cells (T cells, B cells, dendritic cells, MDSCs, TAMs) are a major component. More and more researcher focused on the correlation between disulfidptosis and TME [[209]56]. The most significant contribution of this study is the correlation between disulfidptosis subtypes and TME cell infiltration. In the current study, we demonstrated that the characteristics of the TME and the relative abundance of infiltrating immune cells differed significantly between the two disulfidptosis subtypes and different risk scores. Subtype B, with a better prognosis, exhibited increased infiltration of antitumor immune components, such as B cells, CD8 + T cells, follicular helper T cells and eosinophils, suggesting that they play a positive role in PAAD progression. Protumor immune cells, such as activated CD4 + memory T cells, neutrophils and type 2 helper T cells, had higher levels of infiltration in subtype A. In addition, TME score results also showed higher TME scores in patients with subtype B, further suggesting an immunoreactive characterization of subtype B. Nonetheless, subtype A exhibited significantly lower expression of immune checkpoint targets, like PD-1 and CTLA-4, correlating with T cell recognition of tumor cells. This suggested that the poor prognosis of in Cluster A may not only due to the low TME immune infiltration, but also involve other mechanisms, such as activation of cancer-related pathways. Signature DRGs may be involved in the different immune landscape in two subtypes. Genetic and drug-induced inactivation of GLUT1 enhances tumor sensitivity to anti-tumor immunity and complements anti-PD-1 therapy through the TNF-α pathway [[210]57]. Meanwhile, TAMs promote the formation of an immunosuppressive phenotype by increasing metabolic pathways such as glycolysis and fatty acid oxidation [[211]58]. GLUT1 affects tumor immune environment by its role in glycolysis regulation. WAVE2 restraint of mTOR activation is an absolute requirement for maintaining the T cell homeostasis [[212]45]. Additionally, the low-risk group exhibited higher infiltration levels of CD8 + T cells in the whole PAAD cohort or the TCGA-PAAD cohort. As the main effector cells of the antitumor immune response, CD8 + T cells have the potential to activate and differentiate into effector cytotoxic T lymphocytes (CTLs) after stimulation by tumor-associated antigens. Meanwhile, chemokines promote the migration of CTL from peripheral immune organs to the tumor area and then recognize and kill tumor cells [[213]59]. It has been reported that high levels of tumor CD8 + T-cell infiltration are characteristic of an immunogenically warm tumor, suggesting a better response to the immunotherapy [[214]60]. In PAAD, high infiltration of CD8 + lymphocytes were reported to be strongly associated with a better survival [[215]61, [216]62]. Hence, the low-risk-score group of patients with PAAD may benefit greatly from tumor-infiltrating CD8 + T cells. In addition, recent studies have demonstrated that ferroptosis plays a complicated role in tumor-infiltrating CD8 + T cells. It has been reported that CD8 + T cells can induce cancer cell ferroptosis during cancer immunotherapy [[217]63, [218]64]; however, another study revealed that the occurrence of ferroptosis in immune cells, mediated by receptors such as CD36, could impair the antitumor ability of tumor-infiltrating CD8 + T cells [[219]65]. These findings provide new perspectives on the correlation between disulfidptosis, TME cell infiltration and PAAD. However, it is still uncertain whether and how disulfidptosis affects the function of antitumor immune cells and TME infiltration. Further evidence from bench to bedside is necessary to answer this question. Considering the close relationships between DRGs and the clinical outcomes of PAAD patients, a disulfidptosis-related signature was established based on these DRGs. As expected, Cluster B with a better prognosis had a lower risk score, while Cluster A had a high-risk score. Low-risk score patients had better prognosis than those of high-risk score. In addition, a nomogram was shown using the risk score and clinicopathological factors to predict the 1-, 3-, and 5-year survival probabilities of PAAD patients. The enrichment analysis revealed that individuals with high-risk score were primarily associated with carcinogenic pathways, whereas low risk score PAAD patients showed enrichment in metabolism pathways. Comparative analysis showed that the risk score was negatively correlated with B cells, activated CD8 + T cells, and eosinophils, further suggesting the role of disulfidptosis in immune cell infiltration. Importantly, these results were verified in the TCGA cohort. Studies have reported that biomarkers reflecting TME and tumor cell-intrinsic characteristics, such as PD-L1 expression, immune cell infiltration, TMB, and mismatch repair deficiency, strongly affect the therapeutic outcome of anti-PD-1/anti-PD-L1 therapy [[220]66]. In this study, a higher TMB was detected in the high-risk score group than in the low-risk score group. Furthermore, the correlation between the risk signature and immunotherapy efficacy was further analyzed in two groups of patients who had received anti-PD-L1 treatment in the IMvigor210 cohort and [221]GSE78220 cohort. The results showed that patients with a low-risk score had a better prognosis and a better immunotherapy response rate in both cohorts, which provided preliminary evidence that the risk score can be used to predict the efficacy of immunotherapy. This study revealed that DRGs were closely related to patient outcomes, TME cell infiltration and drug resistance in PAAD. Thus, it should be beneficial to future studies on the mechanisms of disulfidptosis. With the improvement of sequencing technology and the lower cost, more and more patients accept postoperative tissue sequence, even personalized organoid construction, which allow risk scores applied in postoperative prediction and drug screening. Combined with fine-needle aspiration, endoscopic ultrasound, sequence on circulating tumor cell (CTC) and exosome would further enhance the clinic translation of risk scores in earlier diagnosis. Recent research showed targeting GLUT1 could amplify anti-PDAC efficacy via aerobic glycolysis regulation [[222]67]. Although we have carried out multiangle and multi-database validations, the current study still had several limitations that need to be considered. First, due to few studies on the role of disulfidptosis in cancers, the information on DRGs provided by the previous study may not be accurate enough, and some unidentified important DRGs may be missing in the disulfidptosis-related gene sets. Second, the prognostic value of the disulfidptosis signature needs to be further verified clinically. Additionally, the results lacked experimental validation in vivo and in vitro. In subsequent studies, we will further confirm the role of the hub genes SLA2A1, WASF2 and NUBPL in disulfidoptosis in PAAD in a series of experiments. Conclusions In conclusion, this study identified two disulfidptosis subtypes with distinct clinical outcomes and characteristics of TME cell infiltration in PAAD. The disulfidptosis signature may act as a promising biomarker for predicting patient prognosis and their response to immunotherapy. Supplementary Information [223]Supplementary material1 (DOCX 34103 KB)^ (33.3MB, docx) [224]Supplementary material2 (XLSX 63 KB)^ (62.5KB, xlsx) Acknowledgements