Background
Pancreatic cancer has a low 5-year survival rate and is one of the cancers that have high mortality [
1]. There were 458,918 new pancreatic cancer cases and 432,242 deaths due to it, reported globally in 2018 [
2]. Adenocarcinoma is the most common type of pancreatic cancer and accounts for > 90% of the diagnosed pancreatic cancer cases. Although adjuvant chemotherapy and other multimodal treatments have been developed, surgery is still the most effective method for treating this disease [
3]. Despite advances in the treatment of pancreatic adenocarcinoma (PAAD), the 5-year survival rate remains only 9% [
4]. To tackle this, there is an urgent need to identify the prognostic biomarkers of PAAD. This would aid clinicians to predict the clinical outcomes promptly and accurately as well as initiate a protocol for a personalized treatment regimen.
Ferroptosis, a type of cell death, plays a vital role in inhibiting tumorigenesis by removing cells that either has a deficiency or overabundance of key nutrients or cells damaged by environmental pressure [
5]. Unlike autophagy and apoptosis, ferroptosis is an iron (Fe) and reactive oxygen species (ROS)-dependent form of cell death, [
6]. It regulates cell death through the overproduction of phospholipid hydroperoxides in a mechanism different from that of autophagy and apoptosis. It is induced in abnormal cells due to loss of the selective permeability of the plasma membrane and oxidative stress caused by intense membrane lipid peroxidation [
7]. Ferroptosis plays an important regulatory role in the occurrence and progression of tumors and provides a promising therapeutic strategy for PAAD [
8,
9].
Recently, research has focused on the immune infiltration and its role in cancer. Previous studies have demonstrated that infiltrating immune cells can be isolated from tumors suggesting that tumor immune infiltration is a crucial biological process of cancers [
10]. Further, many studies indicate that ferroptosis regulators play an important role in immune environment of cancers including breast cancer [
11], hepatocellular carcinoma [
12], and others. However, the ferroptosis related genes of tumor immune infiltration in PAAD remain largely unknown. Based on the essential roles of the ferroptosis related genes within the tumor immune infiltration in carcinogenesis and progression could have major potential as biomarkers in PAAD.
With the development of next-generation sequencing technology, gene transcription profiles in PAAD can be better understood. Based on The Cancer Genome Atlas (TCGA) and the comprehensive Gene Expression Omnibus (GEO) databases, many genes be found contribute to the development of PAAD. Recent studies have shown that CA9, CXCL9, and GIMAP7 genes specifically regulate the expression of FoxO1, thereby regulating immune infiltration in PAAD [
13]. Further, FOXP4-AS1 exerts oncogenic activity in PAAD [
14] and COL11A1 as an immune infiltrates correlated prognosticator in pancreatic adenocarcinoma [
15].
In this study, we identified ferroptosis-associated genes and constructed molecular subtypes of PAAD based on TCGA and GEO databases. Finally, we established a 3-gene signature prognostic model and verified its ability to predict the prognostic risk and immune infiltrates of PAAD. This serves as a potentially effective tool for prognostic risk prediction in patients with pancreatic cancer.
Methods
Data source and processing
Expression data and corresponding clinical follow-up information from the TCGA-PAAD data set were downloaded using the UCSC genome browser database. GSE62452 chip data sets with survival time were selected from Gene Expression Omnibus (GEO) database. The GSE62452 data set is Microarray gene-expression profiles of 69 pancreatic tumors and 61 adjacent non-tumor tissue from patients with pancreatic ductal adenocarcinoma. ([HuGene-1_0-st] Affymetrix Human Gene 1.0 ST Array [transcript (gene) version]). Henceforth, samples that lacked clinical follow-up information were removed and the expression of multiple gene symbols was considered the median value.
Molecular subtype identification using the non-negative matrix factorization algorithm
Firstly, 60 ferroptosis-related genes were retrieved from the literature [
15‐
18] (Table
1). Next, 58 ferroptosis-related genes with gene expression data were matched with the TCGA-PAAD data set, and PAAD samples were clustered by non-negative matrix factorization (NMF). The standard "Lee" was selected in the NMF method, and ten iterations were performed. The cluster number ‘k’ was set at 2–10, the average contour width of the common member matrix was determined by the R package "NMF", and the samples were divided into three categories.
Table 1
The clinical statistical information of the samples
OS | | | |
0 | 39 | 43 | 16 |
1 | 49 | 42 | 50 |
Grade | | | |
G1 | 16 | 11 | |
G2 | 48 | 44 | |
G3 | 23 | 27 | |
G4 | 1 | 3 | |
Gender | | | |
Male | 39 | 40 | |
Female | 49 | 45 | |
Age | | | |
≤ 48 | 6 | 6 | |
> 48 | 82 | 79 | |
M_stage | | | |
M0 | 38 | 38 | |
M1 | 5 | 0 | |
MX | 45 | 44 | |
N_stage | | | |
N0 | 25 | 24 | |
N1 | 63 | 61 | |
T_stage | | | |
T1 | 3 | 3 | |
T2 | 12 | 11 | |
T3 | 72 | 69 | |
T4 | 1 | 2 | |
Stage | | | |
Stage i | 10 | 9 | |
Stage ii | 74 | 72 | |
Stage iii | 1 | 2 | |
Stage iv | 3 | 2 | |
Identification and functional analysis of differentially expressed genes (DEGs)
The limma_3.42.2 package [
19] was used to analyze the differentially expressed genes (DEGs) in cluster 1, cluster 2, and cluster 3 among the molecular subtypes, based on the threshold false discovery rate (FDR) < 0.05 and |log2FC|> 0.5 filters. The DEGs shared by the three clusters were identified, and the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis and gene ontology (GO) functional enrichment analysis were performed on them through the R package clusterProfiler (v3.16.1) (Additional file
1).
Comparative analysis of immune scores among molecular subtypes
The single-sample gene set enrichment analysis (ssGSEA) method of the GSVA (Gene Set Variation Analysis) package was used to identify the immune score-based relationships among the molecular subtypes in the TCGA-PAAD data set. The scores of 28 immune cells were assessed [
20] and then the differences in immune scores among the molecular subtypes were compared.
Training set and internal test set construction
A total of 173 samples in the TCGA-PAAD data set were divided into a training set and a test set. To prevent the random allocation bias from affecting the stability of subsequent modeling, all samples were put back into random grouping, two hundred times in advance. Using the training data set, a univariate Cox proportional risk regression model was constructed for ferroptosis-related genes (n = 60) and survival data was constructed using the coxph function with survival R package. Since 58 ferroptosis-related genes had expression profile data in our data set, only these were selected for the univariate Cox regression analysis, and P < 0.05 was selected as the threshold for filtering. The R software package “glmnet_4.10–1” [
21] was used to carry out the LASSO Cox regression analysis. We first analyzed the changing trajectory of each independent variable, and later used the fivefold cross-validation to build a model and analyze the confidence interval under each lambda. The target genes were selected by multivariate Cox regression analysis, and a prognostic Kaplan–Meier (KM) curve was established.
3-Gene signature robustness in different data sets
The risk scores of each sample were calculated separately based on the expression level of the sample. KM-curve analysis showed significant differences between the high and low expression groups. Furthermore, we used the R software package timeROC_0.4 [
22] to conduct ROC analysis of the prognostic classification of the risk score and analyze the prognostic classification efficiency at 1-year, 3-years, and 5-years. The model and the survival coefficient, developed using the training dataset, were adopted to evaluate the entire TCGA-PAAD data set, calculate the risk score of each sample and establish the risk score distribution of the samples. The independent GSE62452 data set was used to analyze the robustness of the model.
Univariate and multivariate analyses of the 3-gene signature
To identify the independence of the 3-gene signature model in clinical applications, we performed Cox regression analysis on the TCGA-PAAD training dataset. Based on the results of univariate and multivariate analyses, we used the TCGA-PAAD training dataset to construct a histogram. In addition, corrected curves were used to analyze the prediction accuracy of nomogram at 1, 3, and 5 years.
Tissue samples
PAAD tissues were derived from surgically resected specimens and snap-frozen in liquid nitrogen until RNA extraction. None of the patients received chemotherapy or radiation therapy before surgery. All patients signed informed consent forms provided by the Eastern Hepatobiliary Surgery Hospital. This study was approved by the Ethics Committee of the Eastern Hepatobiliary Surgery Hospital.
RNA isolation and RT-qPCR analysis
RNA was extracted from tissues using the TRIzol reagent (Invitrogen, Carlsbad, CA, USA) and was reverse-transcribed into cDNA using the QuantiTect Reverse Transcription Kit (Qiagen, Valencia, CA, USA). Quantitative PCR (qPCR) uses real-time fluorescence to measure the quantity of DNA present at each cycle during a PCR. Real-time qPCR analyses were quantified with SYBR-Green (Takara, Otsu, Shiga, Japan), and expression levels were normalized to GAPDH levels.
Immunohistochemistry
Immunohistochemistry was performed by two-step method according to the instructions (PV-9000; ZSGB-BIO, Beijing, China). Pancreatic cancer samples were fixed in 10% formalin, embedded in paraffin, and processed into 5-µm sequential sections. The samples were de-waxed with ethanol and blocked to inhibit the endogenous peroxidase activity. After this, samples were heated in a microwave for antigen retrieval, cooled to room temperature, and blocked using goat serum for 30 min at 37 °C. The samples were incubated overnight at 4 °C with rabbit anti-ALOX5 (ab169755), anti-ALOX12 (ab211506), and anti-CISD1 (ab203096) (Abcam, USA) (1:200), followed by incubation with horseradish peroxidase-coupled goat anti-rabbit secondary antibody (PV-9000; ZSGB-BIO, Beijing, China) at 37 °C for 30 min. The samples were then stained with 3,3′-Diaminobenzidine (DAB). Cell nuclei were stained blue with hematoxylin. The sections were then dehydrated, cleared with xylene, and mounted. ALOX5, ALOX12, and CISD1 expressions were determined by immunohistochemistry (IHC) using the streptavidin peroxidase method, with adjacent tissues serving as the controls. The experimental procedure was performed as per the manufacturer’s instructions. Image-Pro Plus 6.0 Software (Media Cybernetics, USA) was used to analyze protein expression and perform statistical analysis of the results obtained by IHC.
Cell culture and transfection
The human PAAD cell line T3M4 and Panc 02.03 were provided by the National Collection Authenticated Cell Cultures (Shanghai, China). The T3M4 cell lines are derived from the metastatic lymph node tissue of human pancreatic cancer and are epithelial-like cells. The T3M4 cell lines were cultured in DMEM (Dulbecco’ modified eagle medium) (Gibco, Grand Island, NY, USA). Supplemented with 10% fetal bovine serum (Invitrogen, San Diego, CA, USA) at 37 °C under 5% CO2 in a humidified incubator. The Panc 02.03 cell lines are derived from human primary pancreatic cancer and are epithelial-like cells. The Panc 02.03 cell lines were cultured in RPMI-1640 (Gibco, Grand Island, NY, USA) with 15% fetal bovine serum at 37 °C under 5% CO2 in a humidified incubator. Si-ALOX5 (No: CAT#: SR319325) was purchased from Origene (Beijing, China). Transfection was performed using Lipofectamine 3000 reagent (No. L3000015, Invitrogen, China) according to the instructions and cell transfection efficiency was 82%.The human ALOX12 and CISD coding sequences were cloned into the pEZ-M03 Vector.
Cell viability assays
The si-ALOX5 transfected in T3M4 cell lines, and ALOX12 or CISD transfected in Panc 02.03 cell lines. Forty-eight hours post infection, the cells were collected and seeded into 96-well plates at a concentration of 2000 cells per well. Cell viability was detected by Cell Counting Kit-8 assay (CCK-8, Dojindo, Japan) according to the manufacturer’s protocol after 48 h. Te absorbance at 450 nm was measured using an automatic microplate reader (BioTek, Winooski, VT, USA). All Cell Counting Kit-8 assay were performed in five times.
Cell migration and invasion assays
The si-ALOX5 transfected in T3M4 cell lines, and ALOX12 or CISD transfected in Panc 02.03 cell lines. Forty-eight hours post infection, the cells were collected. For the migration assay, 800 μl DMEM with 20% serum was added to the lower chamber of a Transwell plate (Corning, NY, USA), and 1.5 × 105 cells were added in the upper chamber. The cells were harvested, resuspended in serum-free media and placed into the upper chamber of a Transwell membrane filter (Corning, NY, USA) for the migration assays or in the upper chamber of a transwell membrane filter coated with Matrigel (Corning) for the invasion assays.The T3M4 and Panc 02.03 cells were added in the upper chamber. After incubation for 24 h at 37 °C, the Transwell chamber was removed. The cells were stained with methanol and 0.1% crystal violet, imaged, and the relative cell density was measured by ImageJ (National Institute of Health, USA). ImageJ software was used to analyze and calculate the migration and invasion area of the cells. The migration and invasion index (%) depicts a proportion of an area where cells have invaded in percentage and is calculated as epithelium area divided by the total area of invaded cells area. The area describes the overall area (μm2) of invaded cells. Evaluation of invasive capacity was performed by counting invading cells under a microscope (40 × 10). Five random fields of view were analyzed for each chamber. All cell migration and invasion assays were performed in five times.
Statistical analysis
All data were analyzed using the SPSS 21.0 statistical software program (IBM Corporation, Armonk, NY, USA). Graphs were generated with GraphPad Prism 8.0 software (GraphPad Software, Inc., San Diego, CA, USA). Student’s t-tests were performed. For a two-tailed t-test, P < 0.05 was essential for considering the results to be statistically significant.
Discussion
Ferroptosis is an iron oxide-dependent form of regulated cell death (RCD). It is the link that connects metabolism, redox biology, and human health. Accumulating evidence indicates that ferroptosis can be triggered to treat cancer to eradicate aggressive malignancies that are resistant to conventional therapies [
25]. It is characterized by the accumulation of ROS and lipid peroxidation products to lethal levels. Although ferroptosis plays an important role in maintaining the survival of normal cells and tissues, it has been increasingly recognized that some carcinogenic pathways are related to ferroptosis, making cancer cells vulnerable to ferroptosis death [
26,
27]. In recent years, ferroptosis death were reported to be associated with antitumor immunity, which is introduced in the follow. Wang et al. reported that CD8 + T cells induce ferroptosis in tumor cells in vivo which shown the direct evidence of the connection between ferroptosis and antitumor immunity [
28]. Tumor cells undergoing ferroptosis might conceivably function as arachidonic acid (AA) donors for the transcellular biosynthesis of eicosanoids, thereby participating in the generation of biologically active immunomodulatory AA metabolites that affect antitumor immunity [
29]. In pancreatic cancer, the process of iron death significantly promotes disease progression and may be a potential strategy for inhibiting pancreatic cancer development. For example, GOT1 inhibition promotes pancreatic cancer cell death by ferroptosis. In this study, we identified a ferroptosis related gene signature for predicting the prognosis of pancreatic adenocarcinoma patients.
Due to the insidious and aggressive nature of PAAD, it is difficult to detect and prevent PAAD at an early stage. At the time of consultation, approximately 80% of patients have locally advanced or metastatic cancer [
30]. Although multimodal therapy has been improved, surgery remains an effective therapeutic strategy for this disease. Even in conjunction with adjuvant therapy, pancreatic surgery can improve 5-year survival by only 20% [
31]. Therefore, prognostic signatures for PAAD patients are urgently needed. With the advancement in the field of bioinformatics and sequencing technology, several potential prognostic evaluation methods for PAAD patients have been developed [
32‐
34]. However, mostly genome or transcriptome parameters are analyzed in these methods, with no consideration of biological processes. Therefore, these models do not analyze the characteristics features of PAAD. Ferroptosis is a significant biological hallmark of tumors and has been demonstrated to be of value in evaluating the prognosis of patients with PAAD [
36]. In this study, ferroptosis-related genes were collected, and gene expression data from public databases such as the TCGA and GEO were used to construct PAAD molecular subtypes based on ferroptosis-related genes. Next, LASSO regression was used to further compress the 10 genes to reduce the number of genes in the risk model. We used fivefold cross-validation to construct a model and analyze the confidence interval under each lambda. Hence, we selected 5 genes at lambda = − 3.75 as target genes and further selected 3 genes (ALOX5, ALOX12, and CISD1) by multivariate Cox regression analysis, and all three genes could significantly improve the performance of distinguishing between the LRG and HRG in the training sample (P < 0.05). The final model based on the 3-gene signature was as follows: RiskScore = 0.289 × ALOX5 + (− 1.359) × ALOX12 + (− 1.053) × CISD1. We found that the constructed 3-gene signature model achieved an accurate prognostic assessment of PAAD samples relative to the other methods. We also compared with previously developed gene-based signatures for PAAD, with two prognostic risk models: 20 and 36 gene signature which result showed our model obtained a more effective result with the reasonable number of genes. The performance of our prognostic risk model was further verified by using validation set data and qPCR and IHC experiments, also with pancreatic cancer cell lines in vitro validation. The functional study shown that expression of ALOX5, ALOX12, and CISD1 can regulate the migration and invasion ability of PAAD cells. Our 3-gene signature model, with fewer genes, is more accurate, reasonable, and efficient as compared to other established models.
The signature we constructed contained three genes, namely, ALOX5, ALOX12, and CISD1, all of which are closely related to tumors genesis and development. 5-Lipoxygenase (ALOX5) is a non-heme iron-containing dioxygenase that catalyzes the peroxidation of polyunsaturated fatty acids, such as arachidonic acid [
37]. ALOX5 is a key enzyme that mediates lipid peroxidation and thereby leads to cell death [
38]. Available evidence shows that lipid peroxidation evokes multiple types of cell death including apoptosis, pyroptosis, and ferroptosis [
39]. It has also been reported that high ALOX5 expression is significantly associated with a poor prognosis in colorectal cancer, gastric cancer, clear cell renal cell carcinoma, papillary thyroid carcinoma, and other tumors [
40‐
43]. ALOX12 gene encodes the enzyme arachidonate 12-lipoxygenase. This enzyme acts on various polyunsaturated fatty acid substrates to produce biologically active lipid intermediates, including eicosanoids and lipoxins. The ALOX12 protein plays an important role in inflammation and oxidation. Abnormal DNA methylation and genetic variation in ALOX12 are associated with various human diseases and pathological phenotypes, such as cardiovascular disease, diabetes, neurodegenerative disease, respiratory disease, cancer, and infection [
44]. Many studies indicate the abnormal expression of ALOX12 in tumors, suggesting that ALOX12 may be a potential marker for many varieties of cancers. Studies in xenotransplantation models show that ALOX12 inactivation reduces the p53-mediated ferroptosis induced by ROS stress, and thus eliminates the p53-dependent tumor growth inhibition [
45]. Compared with pancreatic cancer precursors and normal pancreatic ducts, the expression of ALOX12 in pancreatic cancers is significantly down-regulated and inhibits the proliferation of PAAD cells [
46]. CISD1 (mitoNEET) belongs to a newly discovered class of iron-thionine (2Fe–2S). It is essential for regulating iron and ROS homeostasis in cells and plays a key role in promoting cancer cell proliferation and supporting tumor growth and metastasis [
47]. Additionally, CISD1 is overexpressed in both lung adenocarcinoma and breast cancer [
48,
49].
In this study, we present evidence that ALOX5, ALOX12, and CISD1 may have prognostic value in PAAD. Specifically, our in vitro results suggest that ALOX5 may be an oncogene, while ALOX12 and CISD1 may be tumor suppressor genes in PAAD. The results of our study indicate that the established 3-gene signature model could be an effective prognostic tool for patients with PAAD. However, the limitations associated with this study emphasize the need for additional analysis before the clinical application of this signature. The samples used in our study were obtained retrospectively and the study solely focused on the prognostic value and clinical significance of ferroptosis. To validate our findings for clinical applications, we need to include prospective samples and evaluate the prognostic values of other biological processes characteristic to the development of cancer.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.