An endoplasmic reticulum stress-related signature could robustly predict prognosis and closely associate with response to immunotherapy in pancreatic ductal adenocarcinoma

Figure 1 depicts our analysis workflow. First, we obtained 768 ERS-related genes from GeneCards (https://​www.​genecards.​org/​). The E-MTAB-6134 cohort (n = 288) from the ArrayExpress database (https://​www.​ebi.​ac.​uk/​biostudies/​arrayexpress) included in this study for survival analyses was used as the training set. The RNA-seq data and clinical information of 178 pancreatic cancer samples and 167 normal samples in TCGA and Genotype-Tissue Expression (GTEx) were downloaded from the UCSC Xena Database (https://​xenabrowser.​net/​datapages/​). The GSE21501 (n = 102), GSE28735 (n = 42) and GSE62452 dataset (n = 66) from Gene Expression Omnibus (GEO, http://​www.​ncbi.​nlm.​nih.​gov/​geo) were retrieved and used as a validation set to validate the ERS-related signature. GSE41368 and GSE62165 dataset were used to investigate the expression of ERS-related gene in tumor and normal tissues of PDAC patients.

To identify the differentially expressed genes (DEGs) among the 768 ERS-related genes, we normalized the raw read counts and screened out DEGs between tumor and normal tissues in the TCGA-PDAC cohort by the R package “DEseq2”, using the criteria |log2-fold change (FC)| > 1.5 and false discovery rate (FDR) < 0.05 (Table S1). The protein–protein interaction (PPI) network for DEGs were analyzed by R package “igraph” and the Search Tool for the Retrieval of Interacting Genes (STRING, https://​cn.​string-db.​org/​).

To identify the relevant pathways in the ERS-related signature, Gene Ontology (GO, http://​www.​geneontology.​org) and Kyoto Encyclopedia of Genes and Genomes (KEGG, http://​www.​kegg.​jp/​) pathway enrichment analyses were conducted via the “clusterProfiler” and “GOplot” R packages. An adjusted p value < 0.05 was set as the threshold for statistical significance.

To identify the 163 ERS-related genes associated with prognosis in PDAC patients, univariate Cox regression analysis was performed (p < 0.01). LASSO regression analysis was applied to reduce the dimensionality of candidate genes according to the best penalty factor (λ). Then, the prognostic ERS-related genes were used to construct the ERS-related signature, and the risk scores were generated in a multivariate Cox regression model with the following formula: RiskScore = \({\sum }_{i=1}^{n}(\text{expression of signature gene i} \times \text{coefficient } \upbeta \text{ i})\)

Considering both clinical information and the risk score, a nomogram was drawn with the relevant R package to predict the 1-, 3-, and 5-year survival rates. The predictive performance of the nomogram was identified using the consistency index and calibration curves. The “ggDCA” R package was used to perform decision curve analysis (DCA) to verify the accuracy of the prognostic signature.

Gene set enrichment analysis (GSEA) was conducted to identify the pathways that were significantly different between the ERS-related high- and low-risk groups. The risk score was used as the phenotype label. The nominal p value and normalized enrichment score (NES) were evaluated to sort the pathways enriched in each phenotype. Gene set permutations for each analysis were executed 1000 times. An absolute value of the standardized NES > 1 and a nominal p value less than 0.05 were regarded as the threshold of statistical significance (Table S2).

The immune infiltration scores were measured by TIMER, CIBERSORT, xCell and ssGSEA. Spearman correlation analysis was applied to evaluated the correlation between risk score and immune cells. ssGSEA was applied to calculate the scores of infiltrating immune cells and to evaluate the activity of immune-related pathways. Wilcoxon test was carried out to assess the correlation between the ERS-related risk score and the expression of 40 immune checkpoint genes.

The immunotherapeutic signatures and the activity of each step of the cancer-immunity cycle was evaluated via ssGSEA algorithm. The “ggcor” R package was used to analyze the correlation of risk score with immunotherapeutic signatures and cancer-immunity cycles. The Tumor Immune Dysfunction and Exclusion (TIDE) algorithm was used to predict the immune responses of the ERS-related signature to anti-PD-1 and anti-CTLA-4 treatment (Jiang et al. 2018). The expression levels of the marker of IFN-γ pathway, m6A regulator genes, 148 immunomodulators including chemokines, MHC, receptors, immunoinhibits and immunostimulators in ERS-related risk groups were explored by Wilcoxon test.

The IC50 was evaluated via the Genomics of Drug Sensitivity in Cancer (GDSC, https://​www.​cancerrxgene.​org) database and the chemotherapy response was predicted with the “oncoPredict” package of R. Six chemotherapy drugs (dasatinib, paclitaxel, BI.2536, gemcitabine, lapatinib and docetaxel) were selected to predict the IC50 values in patients. The Wilcoxon rank-sum test was used to analyze the difference between the two risk groups. Tumor stemness score was analyzed by one-class logistic regression (OCLR) via performing “limma” R package.

The human PDAC cell line AsPC-1 was purchased from Procell and verified through short tandem repeat (STR) sequence identification. AsPC-1 cell was cultured in RPMI 1640 medium, supplemented with 10% bovine serum, 1% penicillin–streptomycin (Gibco, CA, USA) and maintained at 37 °C in a humidified incubator with 5% CO₂.

5 FFPE samples were collected from patients diagnosed with PDAC from January 2021 to December 2022 at the Eighth Affiliated Hospital of Sun Yat-Sen University. The samples of surgically resected cancerous and normal tissues were preserved after formalin fixation and paraffin embedding treatment.

HMOX1 and TGFB1 specific siRNA and negative control siRNA were purchased from GenePharma, Suzhou, China. The siRNAs were transfected into cells using jetPRIME transfection reagent (Polyplus Transfection, Illkirch, France) following the guideline. The sequences of siRNAs are shown in Table S3.

Total RNA was extracted from cells using Total RNA Kit II (Genebase Bioscience, China). Total RNA extracted from FFPE tissues was using FFPE RNA Kit (Genebase Bioscience, China) following the manufacturer’s instruction. q-PCR assay was performed using SYBR Green Pro Taq HS Premix (Accurate Biology, Shenzhen, China) and carried out in the LightCycler 480 (Roche, Basel, Switzerland). The relevant primers used are shown in Table S4.

Cell viability was measured using Cell Counting Kit-8 (CCK-8, Dojindo, Kumamoto, Japan) following the manufacturer’s instructions. Cells after indicated treatment were seeded in 96-well plates at 5000 cells/well and cultured. Then, the CCK-8 solution (10μL) was added to each well at 48 h. Next, 1.5 h after CCK-8 administration, the absorbance was measured at 450 nm wavelength with spectrophotometer.

Cells after indicated treatment were reseeded in 12-well plates at density of 500 cells/well and cultured for 2 weeks. Then fixed with 4% paraformaldehyde, stained with crystal violet and counted the number of the colonies formed to quantification.

RNA was harvested using Rneasy mini plus kit (Qiagen). 1.4 ug of total RNA was used for the construction of sequencing libraries. After total RNA was extracted, RNA libraries for RNA-seq were prepared using Hieff NGS™ MaxUp Dual-mode mRNA Library Prep Kit for Illumina^® following manufacturer's protocols. The resulting library was sequenced using DNBSEQ-T7 by Sangon Biotech Co. (Shanghai, China).

As outlined in Fig. 2a, we obtained 4801 DEGs (Fig. 2b) by analyzing TCGA data between PDAC tumor and normal samples. We overlapped these 4801 DEGs with 768 ERS-related genes from the GeneCards database and 20,213 genes from the E-MTAB-6134 training set. Subsequently, 163 differentially expressed ERS-related genes from the intersection of above 3 gene lists were selected (Fig. 2c). Among them, 116 genes were upregulated and 47 genes were downregulated. Finally, we uploaded ERS-related DEGs to the STRING website with combined score > 0.5 to explore the interaction of the potential protein (Fig. 2d).

To investigate the functional annotations of the 163 DEGs, we performed GO and KEGG enrichment analyses. The results demonstrated that terms such as calcium ion transport and response to endoplasmic reticulum stress were notably enriched, consistent with the characteristics of ERS-related genes (Fig. S1A). Simultaneously, KEGG pathway enrichment analysis indicated that the DEGs were involved in the phagosome and HIF-1 signaling pathways (Fig. S1B).

E-MTAB-6134 was used as the training dataset to develop the prognostic signature. First, 23 of the above 163 differentially expressed ERS-related genes were found to be closely related to the prognosis of PDAC patients by univariate Cox regression analysis. Among them, 16 genes were correlated with increased risk with hazard ratios (HRs) > 1; the remaining 7 genes were considered protective with HRs < 1 (Fig. S2). Subsequently, LASSO regression analysis was applied to those 23 genes based on the optimum λ value and 16 candidate genes were screened out (Fig. 3a, b). Finally, by performing multivariate Cox regression analysis, a prognostic model comprising eight genes, HMOX1, TGFB1, JSRP1, GAPDH, ERN2, CHRNE, CD74 and CAV1, was generated to predict the survival outcomes of PDAC patients (Fig. 3c). In addition, eight genes included in ERS-related signature were significantly correlated with each other (Fig. 3d).

The risk score was calculated as follows: Risk score = 0.240632 × TGFB1 expression + 0.390215 × JSRP1 expression + 0.311506 × HMOX1 expression + 0.645877 × GAPDH expression + 0.406487 × CHRNE expression + 0.231012 × CAV1 expression − 0.41471 × CD74 expression − 0.20355 × ERN2 expression. ERS-related risk scores for PDAC patients in the training and validation datasets were calculated using the above formula. The median risk score of the training dataset was 0.985, which was considered as the cut-point for dichotomizing the samples into high- and low-risk groups in the training set and validation sets. The distribution of patients was significantly different between the high-and low-risk groups, indicating that the risk score had a powerful discriminatory ability (Fig. S3A). Notably, the Kaplan–Meier survival curves revealed that patients in the high-risk group tended to have worse OS outcomes than those in the low-risk group (Fig. 4a and Fig. S4A). In addition, according to the scatter point analysis, with the increase in the ERS-related risk score, patients tended to have increased mortality rates and reduced survival times (Fig. 4b and Fig. S4B). We also applied the ROC curve to evaluate the predictive role of the risk scores regarding the OS outcome. In the E-MTAB-6134, the AUCs for predicting 1-, 3-, and 5-year OS were 0.809, 0.753 and 0.765, respectively (Fig. 4c). While the AUCs for predicting 1-, 3-, 5-year OS in TCGA were 0.734, 0.716 and 0.835 (Fig. 4c), and 0.755, 0.694 and 0.806 in GSE62452 (Fig. S4C). The AUC values were 0.712 for 1 year, 0.728 for 2 years, 0.789 for 3 years in GSE21501 and 0.768 for 1 year, 0.815 for 2 years, and 0.722 for 3 years in GSE28735. Finally, the heatmap showed that the high-risk group highly expressed HMOX1, TGFB1, JSRP1, GAPDH, CHRNE and CAV1, while ERN2 and CD74 were overexpressed in the low-risk group (Fig. S3B). Taken together, these results demonstrated an excellent predictive performance of the ERS-related prognostic signature.

Due to the high correlation between the ERS-related signature and the prognosis of PDAC patients, we evaluated the classic clinical parameters and risk scores by univariate and multivariate Cox regression analyses to identify independent prognostic factors. Through our multivariate Cox regression analysis, grade and resection margin were related to worse survival outcomes, and only N stage and the ERS-related risk score served as independent prognostic factors (Fig. 5a, b). ROC curves were used to evaluate the predictive performance, and the ERS-related risk score with the highest AUC value showed the best discrimination compared with traditional clinicopathological parameters (Fig. 5c). Additionally, the clinical characteristic heatmap showed the tendency of signature gene expression in the two risk groups and were closely correlated with gender, grade, resection margin, T stage and N stage based on the E-MTAB-6134 cohort (Fig. 5d). Next, we investigated the relationship between the ERS-related risk score and clinical characteristics. As shown in Fig. 5e, patients in certain subgroups, such as those with microscopically incomplete resection (R1), a high pathological grade, and an advanced N stage, tended to have higher risk scores, while no differences were obtained with stratification based on gender and T stage. We further confirmed the reliability of the ERS-related prognostic signature. Subgroup analysis revealed that patients with high-risk scores had an unfavorable prognosis compared to those with low-risk scores in the subgroups of gender (male or female), resection margin (R0 or R1), grade (G1/2 or G3), N stage (N0 or N1) and T stage (T1/2 or T3) (Fig. S5).

To further predict the survival of PDAC patients, AJCC stage, resection margin, grade, gender and the ERS-related risk score were incorporated into a nomogram for predicting the 1-, 3- and 5-year survival rates (Fig. 6a). The calibration curve of the nomogram showed that the predicted 1-, 3- and 5-year survival outcomes were highly in accordance with the observed survival rate (Fig. 6b). Significantly, DCA was conducted to determine the clinical performance of the predictive nomogram, which showed that the nomogram provided a greater net benefit (Fig. 6c). Moreover, the nomogram was evaluated using ROC curves. As shown in Fig. 6d, the AUCs of nomogram at 1-, 3- and 5-year was 0.782, 0.774 and 0.778, which indicating well predictive performance of the nomogram.

We identified DEGs between the high-risk group and low-risk group and performed functional enrichment analyses to further explore and confirm the signaling pathways enriched in the two groups. Based on the GO enrichment analysis, we found the top terms of Biological Process included digestion, tissue homeostasis, humoral immune response, digestive system process and maintenance of gastrointestinal epithelium. For Cellular Component, the top terms included collagen-containing extracellular matrix, tertiary granule lumen and blood microparticle. For Molecular Function, the extracellular matrix structural constituent term was enriched (Fig. 7a). Additionally, we discovered the bile secretion, complement and coagulation cascades, pancreatic secretion and protein digestion and absorption pathways were enriched based on KEGG enrichment analysis (Fig. 7b). Moreover, we investigated the correlation between the expression of ERS-related gene signature and the known signature and found that high-risk and low-risk PDAC patients in E-MTAB-6134 cohort are significantly different in base-excision repair, CD8⁺ T effector, DNA damage response, EMT2, EMT3, immune checkpoint, mismatch repair, nucleotide excision repair, Pan-F TBRs, TME scoreA and TME scoreB signature (Fig. 7c). In addition, the GSEA results revealed that the HIF-1 signaling pathway, ECM receptor interaction pathway and PI3K-AKT signaling pathway were enriched in the high-risk group, while the pancreatic secretion pathway, drug metabolism-cytochrome P450 pathway and tight junction pathway was enriched in the low-risk group (Fig. 8).

To verify the relationship between ERS-related prognostic signature and immune response, TIMER, CIBERSORT, xCell and ssGSEA algorithms were adopted to depict the landscape of infiltrating immune cells in PDAC patients (Fig. 9a). We also examined the correlation of ERS-related risk score and abundance of immune cells in PDAC tumor microenvironment via Spearman correlation analysis. Interestingly, CD8⁺ T cell infiltration had a negative correlation with risk score (Fig. 9b). Furthermore, in order to explore the infiltration of the 22 types of tumor-infiltrating immune cells in E-MTAB-6134 cohort, CIBERSORT algorithm was applied to estimate the levels of immune cell infiltration in each sample. As compared to high-risk group, patients in the low-risk group had higher level of infiltrated CD8⁺ T cells, CD4⁺ memory resting T cells, monocytes and M2 macrophages, while the infiltration levels of plasma cells, regulatory T cells, activated NK cells and M0 macrophages were significantly higher in the high-risk groups (Fig. 9c). In addition, as shown in Fig. 9d, we performed ssGSEA score comparison of immune function and results revealed that the APC costimulation, CCR, MHC class I and parainflammation pathways had higher enrichment scores in the high-risk group than in the low-risk, while the type II IFN response had a higher enrichment score in the low-risk group than that in the high-risk group. Importantly, in accordance with the evaluation of TIMER, CIBERSORT, xCell and ssGSEA algorithms, proportions of CD8⁺ T cells infiltrated in low-risk group were strongly higher than those in high-risk group (Fig. 9e, f).

Furthermore, we examined the differences in immune checkpoint blockade (ICB) response signature prediction between the two risk groups. The high-risk group scored highly in alcoholism, APM signal, cytokine-cytokine receptor interaction, homologous recombination, microRNAs in cancer, oocyte meiosis, progesterone-mediated oocyte maturation, proteasome, systemic lupus erythematosus and viral carcinogenesis, while scored lowly in IFN-γ signature than those in low-risk group (Fig. S6A). Moreover, except for IFN-γ signature, ERS-related risk score was negatively correlated with ICB-related positive signals (Fig. 10a). The tumor immune cycle is a critical indicator to assess the biological function of the chemokine system and other immunomodulators (Xu et al. 2018). Therefore, we evaluated the variations in the activity of tumor immune steps between two risk groups. As shown in Fig. S6B, high-risk group was highly scored in release of cancer cell antigen (step 1), while low-risk group highly scored in cancer cell antigen expression (step 2), CD4 T cell recruiting (step 4), Treg cell recruiting (step 4) and recognition of cancer cells by T cells (step 6). Similarly, we examined the correlation between risk score and these steps in tumor immune cycle. The results exhibited that risk score had a negative correlation with step 1, while positively correlated with the other tumor immune cycle steps (Fig. 10a). Considering the significant relevance between risk score and CD8⁺ T cells and the crucial role of m6A methylation in regulating the function of CD8⁺ T cells, we next investigated the expression of IFN-γ pathway markers and m6A regulators in two risk groups and most of them were remarkably associated with the ERS-related risk score (Fig. 10b, c). More importantly, to evaluate the immunotherapeutic efficacy between two risk groups, we calculated and compared the TIDE score. Our analysis suggested the risk score was positively correlate with TIDE score, indicating that low-risk PDAC patients may be more sensitive to immunotherapy than PDAC patients in high-risk group. Also, low-risk group had a higher IFNG score, while had a lowly score in the dysfunction (Fig. 10d). In addition, we evaluated the expression of immune checkpoints between the different risk groups. The results showed that immune suppressive molecules, such as PDCD1, TNFSF9, TNFSF4, TNFSF14, TNFRSF8, TNFRSF4, TNFRSF18, TMIGD2, PDCD1LG2, CD86, CD70, CD274 (PD-L1), CD276 (B7-H3), CD244, BTNL2 and ADORA2A, were elevated in the high-risk group, suggesting that patients in the high-risk group have a worse anti-tumor immune response (Fig. 10e). Also, we drawn a heatmap to exhibit the mRNA expression landscape of immune-modulator genes including chemokine, MHC, receptor, immune inhibitor and immune stimulator in two risk groups (Fig. S6C).

Taken together, these results indicated that the ERS-related signature had outstanding potential in evaluating the immunotherapy response. The unfavorable outcomes of high-risk PDAC patients might be due to an immunosuppressive microenvironment and a worse immunotherapy response.

Tumor stemness is known to play an important role in drug resistance. The mRNAsi is an index to describe the expression of stemness-associated genes of tumor cells and range from 0 to 1 (Malta et al. 2018). To investigate the correlation between the ERS-related risk score and drug sensitivity, mRNAsi score was evaluated. Our analysis shows mRNAsi score negatively correlates with risk score (R = − 0.18, p = 0.0019, Fig. 11a). We also specifically compared those stemness genes between the two risk groups. The stemness genes including PROM1, NANOG and IKZF2 were upregulated in low-risk group (Fig. 11b). Furthermore, based on the GDSC database, we predicted the response to common drugs among patients with PDAC by estimating the differences in the IC50 values between the high- and low-risk groups. Dasatinib, paclitaxel, BI.2536, gemcitabine, lapatinib and docetaxel (Zhu et al. 2022) were screened to assess the chemotherapy response of PDAC patients. Notably, the IC50 values of dasatinib and lapatinib in the high-risk group were significantly lower than those in the low-risk group, suggesting that patients in the high-risk group more sensitive to these two drugs (Fig. 11c).

To demonstrate the role of ERS-related genes in PDAC cells, we first analyzed the expression level in tumor and normal tissues of HMOX1 and TGFB1 using the TCGA, GSE41368 and GSE62165 datasets. As expected, HMOX1 and TGFB1 were upregulated in tumor and the related ROC curves showed the great performance of prediction (Fig. S7A, B). Furthermore, compared to the normal tissue, the HMOX1 and TGFB1 expression level in tumor tissue are mostly different in 33 tumor types (Fig. S7C). Then, we verified these findings by analyzing the FFPE tissues from PDAC patients. The results demonstrated that when compared to normal tissues, PDAC had higher levels of HMOX1 and TGFB1 expression (Fig. 12a). Moreover, siRNAs targeting HMOX1 and TGFB1 were transfected into AsPc-1 cells. The efficiency of siRNA knockdown was confirmed via q-PCR (Fig. 12b). As shown in Fig. 12c, compared to the control group, downregulation of HMOX1 and TGFB1 significantly suppressed the viability and of PDAC cells. Consistently, the results of clonogenic assays revealed that HMOX1 or TGFB1 knockdown reduced the potential for colony formation by tumor cells (Fig. 12d). To better understand what signaling pathway affects prognosis of PDAC, we conducted transcription sequencing of the AsPC-1 cells expressing RNAi against HMOX1 and scramble siRNA. Importantly, GSEA analysis revealed that TNF-α signaling via NF-κb pathway was notably enriched, which was worthy of further analysis (Fig. 12e). In summary, these findings suggest that ERS-related gene HMOX1 or TGFB1 may serve as the prognostic marker, which provided some new insights into predict the outcomes of PDAC patients.