RNA-seq data (TPM) for 730 para-cancerous tissue samples and 10,363 pan-cancer samples were obtained from the Cancer Genome Atlas (TCGA) database (https://portal.gdc.cancer.gov/repository). For PAAD, 4 para-cancerous tissue samples and 179 cancer samples were downloaded from the TCGA Knowledge Base. Furthermore, 167 normal samples were retrieved from University of California Santa Cruz Xena (UCSC Xena, http://xena.ucsc.edu/) take into account sample heterogeneity.

Gene alteration data including alteration frequency, possible mutation sites, and alteration type were obtained from the cBioPortal database (https://www.cbioportal.org/). Details of the gene alteration types, such as amplification, missense mutation, and deep deletion of the genes were also examined.

Immune cell infiltration and target gene expressions in PAAD was assessed using R (v 3.6.3, The R Foundation for Statistical Computing, Vienna, Austria). The Tumor Immune Estimation Resource (TIMER) database (https://cistrome.shinyapps.io/timer/) was used to examine correlations between immune cell infiltration and gene copy number variations, and also between immune cell infiltration and gene expression levels. The Tumor-Immune System Interactions Database (TISIDB) (http://cis.Hku.hk/TISIDB/) was used to analyze gene expression in different immune cell subtypes, as well as correlations with lymphocytes and immunostimulators. The correlation between 70 common checkpoints and target genes was analyzed with R software (v 3.6.3).

Univariate and multivariate Cox regression analyses were used to identify appropriate genes for constructing a nomogram. The parameters included T stage, N stage, pathological stage, radiation therapy, residual tumor, histological grade, anatomic neoplasm subdivision, and the expression of EIF2S1, EIF2S2, and EIF2S3. Receiver Operating Characteristic (ROC) curves, Kaplan-Meier survival curves, calibration curves, and Decision Curve Analysis (DCA) diagrams were used to evaluate the predictive ability of the prognostic model for patient survival.

A scRNA-seq data source, tumor immune single-cell hub (TISCH) (http://tisch.comp-genomics.org/home/), focuses on the tumor microenviroment (TME) and provides cell-type specific annotations at the single-cell level, thereby enabling research on TME in various types of malignancies. The TISCH was used to analyze PAAD scRNA data obtained from the CRA001160 and GSE111672 databases. EIF2Ss expression in various cell types were shown in T-SNE projection. The enrichment of cells in various pathways was also shown in TISCH.

Gene Set Enrichment Analysis (GSEA) is a previously established gene collection that shows statistically significant and continuous changes between two biological states using a computerized method (https://www.gsea-msigdb.org/gsea/index.jsp). This was used to find all genes that have been linked to EIF2Ss gene expression in previous research and to investigate for possible survival differences between high and low expression groups. Gene set permutation was conducted 1000 times for each analysis. Highly correlated genes were illustrated by heatmap using the R package (version 3.3.3, ggplot2). The genes and correlated genes were enriched using GSEA.

The human pancreatic cancer cell lines SW1990 and Capan1 were purchased from the American Type Culture Collection (Manassas, VA, USA). Short Tandem Repeat (STR) profiling was used to authenticate the cell lines, and all cells tested free of mycoplasma. Both cell lines were cultured in an incubator at 37 °C with 5% CO${}_2$ incubator and grown in Dulbecco’s Modified Eagle Medium (DMEM) (Sigma-Aldrich, St. Louis, MO, USA) containing 1% penicillin (Sigma-Aldrich), 1% streptomycin (Sigma-Aldrich), and 10% fetal bovine serum (Hyclone; Cytiva, Logan, UT, USA). The total RNA of pancreatic cancer cells was extracted by TRIzol Reagent (Invitrogen, Carlsbad, CA, USA). The reverse transcription of total RNA to DNA was completed by the TaKaRa PrimeScript RT Reagent Kit (TaKaRa, Kusatsu, Japan). SYBR Green Real-Time PCR Master Mixes (Yeasen Biotechnology, Shanghai, China) was usedto detect the expression of EIF2S1, EIF2S2, and EIF2S3. The GAPDH served as an internal reference control.

SW1990 and Capan1 pancreatic cancer cells were washed twice with PBS, then cell lysis buffer (Beyotime, Shanghai, China) supplemented with protease and phosphatase inhibitors (Beyotime, Shanghai, China) was added, and the dish was placed on ice for 30 minutes. Cells were scraped by a pre-cooled cell scraper and the suspension transferred to a 1.5 mL EP tube and centrifuged at 14,000 g for 15 minutes at 4 °C. The cell lysate supernatant was immediately transferred to a new 1.5 mL EP tube and the relative protein concentration measured with a BCA Protein Assay kit (Beyotime, Shanghai, China). A small amount of cell lysate supernatant was taken as input. The corresponding antibody was then added to the remaining cell lysate supernatant and incubated for 24 hours at 4 °C. Next, 50 µL protein A/G magnetic beads (Beyotime, Shanghai, China) were incubated with the complex overnight at 4 °C and then washed with lysis buffer. Finally, the input, IgG, and IP fractions were subjected to western blot analysis. Following electrophoresis with 10% SDS-PAGE, the proteins were transferred onto polyvinylidene difluoride (PVDF) membranes. The membranes were then treated with the following primary antibodies: c-Myc (1:1000, Proteintech, Chicago, IL, USA), EIF2S1 (1:1000, Proteintech), EIF2S2 (1:1000, Proteintech), EIF2S3 (1:1000, Proteintech), and glyceraldehyde-3-phosphate dehydrogenase (GAPDH) (1:1000, Proteintech). These antibodies were treated with the PVDF membrane overnight, washed, and then incubated with a Horseradish Peroxidase (HRP)-labeled secondary antibody solution for 1 hour. Finally, enhanced chemiluminescence (ECL) luminescent solution was added to the surface of the PVDF membranes, to visualize the immunoblot bands in a pre-cooled LAS4000 luminescent imager (GE, Boston, MA, USA)..

Pancreatic cancer cells were harvested, lysed with cell lysis buffer, and then quantified and labeled as the input sample. The IP sample was separated by SDS-PAGE, and LC-MS/MS analysis was then performed (Shanghai Genechem Co., Ltd.). Briefly, the bands from SDS-PAGE were divided into 1 mm${}^3$ for each gel slice. The proteins were then dissolved, and the peptide samples were desalted and resuspended. Easy nLC chromatography was subsequently performed, with 0.1% formic acid aqueous solution used as buffer A and 0.1% formic acid acetonitrile aqueous solution used as buffer B. A nanoliter flow rate Easy nLC system was used to separate the experimental samples, which were then subjected to mass spectrometry analysis using a Q Active Plus mass spectrometer (Thermo Fisher Scientific, MA, USA). Finally, the original graph file obtained from Q Active Plus was converted, processed, and analyzed.

Data were analyzed using R software version 3.6.3 (The R Foundation for Statistical Computing, Vienna, Austria), SPSS 20.0 (IBM Corp. Released 2011. IBM SPSS Statistics for Windows, Version 20.0. IBM Corp., Armonk, NY, USA), and GraphPad Prism 8.0.0 (GraphPad Prism version 8.0.0 for Windows, GraphPad Software, Boston, MA, USA, https://www.graphpad.com/). Student’s t-test or one-way ANOVA were used to compare groups, and a p-value 0.05 was deemed statistically significant.

The expression of EIF2Ss and their prognostic value in many different cancer types were examined using TCGA data. The expression of EIF2S1, EIF2S2, and EIF2S3 were elevated in tumor tissue compared to the corresponding normal tissue (Supplementary Fig. 1A–C). These genes were also consistently overexpressed in PAAD (Fig. 1A–C). Kaplan-Meier analysis revealed that high expression of EIF2S1, EIF2S2 and EIF2S3 in PAAD patients was associated with poor OS (Fig. 1D–F), disease-specific survival (DSS) (Supplementary Fig. 1D), and progress-free interval (PFI) (Supplementary Fig. 1E). The area under the curve (AUC) for ROC analysis of EIF2S1, EIF2S2 and EIF2S3 expression were 0.926, 0.971, and 0.881, respectively (Fig. 1G–I). Moreover, time-dependent ROC curves for EIF2Ss showed AUCs of 0.614, 0.646, and 0.754 for EIF2S1 (Supplementary Fig. 1F), 0.674, 0.629, and 0.620 for EIF2S2 (Supplementary Fig. 1F), and 0.586, 0.655, and 0.760 for EIF2S3 (Supplementary Fig. 1F).

Fig. 1.

Expression of EIF2Ss in Pancreatic adenocarcinoma (PAAD) and their prognostic and diagnostic value. (A–C) Differential expression of EIF2S1, EIF2S2, and EIF2S3 in PAAD tumor and normal tissue. (D–F) OS curves for PAAD patients with low (n = 89) or high (n = 90) expression levels of EIF2S1, EIF2S2, and EIF2S3. (G–I) ROC curves indicate the diagnostic value of EIF2S1, EIF2S2, and EIF2S3 expression in PAAD. ***p 0.001. PAAD, pancreatic adenocarcinoma; EIF2S1, eukaryotic translation initiation factor 2 subunit alpha; EIF2S2, eukaryotic translation initiation factor 2 subunit beta; EIF2S3, eukaryotic translation initiation factor 2 subunit gamma; OS, overall survival; ROC, receiver operating characteristic; AUC, area under the curve; CI, confidence interval.

We next examined the clinicopathological characteristics of PAAD groups with differentially expressed EIF2Ss. Patients with high expression of EIF2S1 presented with T3/T4 stage, stage Ⅱ/Ⅲ/Ⅳ, PD, and G2/G3/G4 grades (Supplementary Fig. 2A–D). Patients with high EIF2S2 expression presented with PD and G2/G3/G4 grades (Supplementary Fig. 2E,F). Patients with high expression of EIF2S3 presented with T3/ T4 stage, female gender, residual tumor R1 and R2, and G2/G3/G4 grade (Supplementary Fig. 2G–J). Possible associations between OS with clinical characteristics and EIF2Ss were also investigated. High expression of EIF2S1 in patients with T1/T2, N0, no radiotherapy, male, residual tumor R0, G1/G2 grade, diabetes history and no family history of cancer was associated with poor OS (Supplementary Fig. 3A). Low EIF2S2 expression in PAAD patients with T1/T2, M0, P0, no radiation, PD, male, white, R1/R2 residual tumor, G1/G2 grade, pancreatic cancer head, no alcohol history, and no cancer family history of cancer was associated with good OS (Supplementary Fig. 3B). Furthermore, poor OS was observed with high expression of EIF2S3 in pancreatic cancer patients with N0 stage, no radiation therapy, less than or equal to 65 years old, residual tumor of R1/R2, G1/G2 grade, no smoker, and history of diabetes (Supplementary Fig. 3C).

Data on EIF2Ss gene alterations were obtained from the cBioPortal database. The distibution of alteration types is shown in Fig. 2A–C. EIF2S1 and EIF2S2 showed amplification and deep deletions (Fig. 2D,E), whereas EIF2S3 showed missense mutations and deep deletions (Fig. 2F).

Fig. 2.

EIF2Ss gene alterations in PAAD. (A–C) The location of gene alteration types in EIF2S1, EIF2S2 and EIF2S3. (D–F) Gene alteration types and proportions in EIF2S1, EIF2S2, and EIF2S3.

PAADs with high levels of EIF2Ss expression were infiltrated with Th2 cells (Fig. 3A–C). In contrast, PAADs with a low level of EIF2Ss expression were infiltrated with pDC (Fig. 3A–C). High levels of EIF2S1 and EIF2S3 expression were associated with infiltration by T helper cells and Th2 cells (Fig. 3A,C). Low EIF2S2 expression was associated with infiltration by CD8+ T cells, cytotoxic cells, iDC, mast cells, NK cells, pDC, T cells, TFH, Th17 cells, and Th2 cells (Fig. 3B). The relationship between immune cell infiltration and EIF2Ss expression was also investigated using the TIMER database. EIF2S1, EIF2S2, and EIF2S3 expression were all positively associated with CD8+ T cells (Fig. 3D–F). In addition, EIF2S1 expression was positively associated with B cells, macrophages, neutrophils, and DCs (Fig. 3D). EIF2S3 expression was positively associated with B cells, neutrophils, and DCs (Fig. 3F).

Fig. 3.

Immune infiltration and EIF2Ss in PAAD. (A–C) Immune cells enrichment in PAAD with low or high expression of EIF2S1, EIF2S2, and EIF2S3. (D–F) Immune cell enrichment in copy number variations (CNVs) of EIF2S1, EIF2S2, and EIF2S3. *p 0.05, **p 0.01 and ***p 0.001.

Correlations between the expression levels of EIF2S1, EIF2S2, and EIF2S3 with the expression of 70 common immune checkpoints in PAAD are presented in Fig. 4A. The expression of EIF2S1 was positively associated with CD44, NT5E, CD274, IL2RA, NRP1, ENTPD1, CD86, PDCD1LG2, PVR, TGFBR1, TNFSF4, TNFSF13B, CD80, ULBP1, CD276, HAVCR2, MICB, CSF1R, CD28, RAET1E, ICOS, TIGIT, TNFRSF9, LAIR1, CD96, IL10, CD200R1, CXCR4, TNFSF18, IL10RB, IL6, TNFSF15, IDO1, CD70 and CD244 (Supplementary Fig. 4A). The expression of EIF2S2 was positively associated with NT5E, ULBP1, PVR, CD44, IL10RB and CD276 (Supplementary Fig. 4B). Finally, the expression of EIF2S3 was positively correlated with that of CD44, NT5E, MICB, IL2RA, TNFSF4, CD274, NRP1, PDCD1LG2, TNFSF13, ENTPD1, TNFSF13B, RAET1E, TNFRSF9, TNFRSF15, CD86, TGFBR1, ICOS, TIGIT, CD40, CD96, CD80, CD28, CD276, HAVCR2, CD200R1 and IDO1 (Supplementary Fig. 4C). The expression levels of EIF2S1, EIF2S2, and EIF2S3 in PAAD also correlated with the immune subtypes C1 (wound healing), C2 (IFN-gamma dominant), C3 (Inflammatory), and C6 (TGF-b dominant) (Fig. 4B–D).

Fig. 4.

EIF2Ss correlations with immune checkpoints, subtypes, lymphocytes, and immunostimulators. (A) Correlations between the expression of 70 common immune checkpoints and the expression of EIF2S1, EIF2S2, and EIF2S3 in PAAD. (B–D) Correlations between EIF2S1, EIF2S2, and EIF2S3 expression and immune subtypes.

A heatmap generated with the TISIDB database showed the correlations between EIF2Ss, lymphocytes and immunostimulators in 30 cancer types (Supplementary Fig. 5A–F). In PAAD, most lymphocytes and immunostimulators were positively correlated with the expression of EIF2S1, EIF2S2, and EIF2S3. The expression of EIF2S1 was also positively correlated with an abundance of Act—CD4 cells, Tcm—CD4 cells, Tcm—CD8 cells, and Th2 cells (Supplementary Fig. 5G–J). EIF2S2 and EIF2S3 expression were positively associated with an abundance of CD56dim cells and Act CD4 cells, respectively (Supplementary Fig. 5M,O). EIF2S1 expression was also positively correlated with expression of the IL2RA and NT5E immunostimulators (Supplementary Fig. 5K,L). Finally, the expression of EIF2S2 was positively associated with NT5E (Supplementary Fig. 5N).

Optimization of a prognostic model for patient outcome could improve its accuracy and therefore its clinical value, especially in PAADs with high expression of EIF2Ss. We therefore built a model using univariate and multivariate Cox regression analysis, and incorporated important clinicopathologic features (T stage, N stage, pathological stage, radiation therapy, residual tumor, histologic grade, anatomic neoplasm subdivisions), as well as EIF2Ss (Supplementary Fig. 6A,B). This nomogram was predictive of 1-year, 2-year, and 3-year OS probability (Fig. 5A). Subsequently, the predictive efficacy and value of the nomogram were verified by ROC curves, calibration curves, and DCA diagrams. The AUC of ROC curves for 1-year, 2-year, and 3-year OS were 0.749, 0.824, and 0.807, respectively (Fig. 5B). Furthermore, the calibration curves (Fig. 5C–E) and DCA diagram (Fig. 5F–H) showed good consistency between the prediction and the actual state.

Fig. 5.

Construction of a nomogram for PAAD. (A) The nomogram was used to predict the OS of PAAD patients at 1-year, 2-year, and 3-years. (B) ROC curves indicate the diagnostic value of the nomogram. (C–E) Calibration curves for the nomogram predict OS at 1-year, 2-year, and 3-years. (F–H) Decision Curve Analysis (DCA) diagrams for the nomogram predict OS at 1-year, 2-year, and 3-years.

Correlations between immune cell infiltration and EIF2Ss expression were further verified by scRNA-Seq data obtained from PAAD_CRA001160 and PAAD_GSE111672 (Fig. 6A,E). The results showed that EIF2S1, EIF2S2, and EIF2S3 were highly expressed in endothelial cells, fibroblasts, malignant cells, and ductal cells (Fig. 6B–D,F–K). Enrichment analysis of EIF2Ss in PAAD_CRA001160 and PAAD_GSE11167 indicated that adipogenesis and interferon-alpha response were mainly associated in endothelial cells; the epithelial-mesenchymal transition was mainly associated in fibroblasts; myc targets, G2M checkpoint, and interferon-alpha response were mainly enriched in malignant cells; oxidative phosphorylation, interferon-alpha response, and hypoxia were mainly enriched in ductal cells (Supplementary Figs. 7,8).

Fig. 6.

ScRNA-Seq analysis of EIF2S1, EIF2S2, and EIF2S3 in PAAD. (A) The t-SNE projection of all cells in scRNA-Seq data of PAAD_CRA001160. (B–D) The expression of EIF2S1, EIF2S2, and EIF2S3 in scRNA-Seq data of PAAD_CRA001160. (E) The t-SNE projection of all cells in scRNA-Seq data of PAAD_GSE111672. (F–H) The expression of EIF2S1, EIF2S2, and EIF2S3 in scRNA-Seq data of PAAD_GSE111672. (I–K) Cells enrichment for EIF2S1, EIF2S2, and EIF2S3 in scRNA-Seq data of PAAD_CRA001160 and PAAD_GSE111672.

The scRNA-Seq data analysis in PAAD showed that myc targets were enriched in malignant cells with EIF2Ss expression. Hence, interactions between EIF2Ss and c-myc were further investigated. The results of GSEA functional enrichment analysis are shown in Fig. 7A–C. The spearman correlation coefficients for EIF2S1, EIF2S2, and EIF2S3 with c-myc were 0.488, 0.255, and 0.4, respectively (Fig. 7D–F). Myc was upregulated in tumor tissue compared with normal tissue (Fig. 7G). Kaplan-Meier analysis showed that high myc expression was associated with poor OS (Fig. 7H). Mass spectrum analysis confirmed that EIF2S1, EIF2S2, and EIF2S3 proteins were co-immunoprecipitated with myc (Fig. 7I). The mass spectrogram showed the EIF2S1 peptide sequence was GVFNVQMEPK (Fig. 7J), the EIF2S2 peptide sequence was EYVTCHTCR (Fig. 7K), the EIF2S3 peptide sequence was VGQEIEVRPGIVSK (Fig. 7L). Co-immunoprecipitation experiments using SW1990 and Capan1 cell lysis protein were used to further confirm the interaction between c-myc and EIF2Ss, c-myc was again found to co-immunoprecipitate with EIF2S1, EIF2S2, and EIF2S3 (Fig. 7M). Western blot analysis also showed that EIF2S1, EIF2S2, and EIF2S3 co-immunoprecipitated with c-myc (Fig. 7N). Moreover, knock down of c-myc was observed to downregulate the expression of EIF2S1, EIF2S2, and EIF2S3 (Fig. 7O). Overexpression of c-myc was observed to upregulate the expression of EIF2S1, EIF2S2, and EIF2S3 (Fig. 7P). In the meanwhile, the RT-qPCR results were also verified this point (Fig. 7Q–R).

Fig. 7.

Correlation of EIF2Ss with c-myc in PAAD. (A–C) GSEA functional enrichment analysis of EIF2Ss. (D–F) Spearman value correlation coefficients for EIF2S1, EIF2S2, and EIF2S3 with c-myc. (G) Upregulated of myc in pancreatic cancer tissue compared with normal tissue. (H) Kaplan–Meier survival analysis for myc expression in PAAD patients. (I) The mass spectrum assay confirmed co-immunoprecipitation of myc with EIF2S1, EIF2S2, and EIF2S3. (J–L) The mass spectrogram of EIF2S1, EIF2S2, and EIF2S3. (M,N) Co-immunoprecipitation of myc with EIF2Ss. (O) Knockdown of c-myc decreased the expression of EIF2S1, EIF2S2, and EIF2S3 (western blot results). (P) Overexpression of c-myc increased the expression of EIF2S1, EIF2S2, and EIF2S3 (western blot results). (Q) Knockdown of c-myc decreased the expression of c-myc, EIF2S1, EIF2S2, and EIF2S3 (RT-qPCR results). (R) Overexpression of c-myc increased the expression of c-myc, EIF2S1, EIF2S2, and EIF2S3 (RT-qPCR results).