A novel immunogenomic signature to predict prognosis and reveal immune infiltration characteristics in pancreatic ductal adenocarcinoma

Abstract

Background: The immune response in the tumor microenvironment (TME) plays a crucial role in cancer progression and recurrence. We aimed to develop an immune-related gene (IRG) signature to improve prognostic predictive power and reveal the immune infiltration characteristics of pancreatic ductal adenocarcinoma (PDAC).

Methods: The Cancer Genome Atlas (TCGA) PDAC was used to construct a prognostic model as a training cohort. The International Cancer Genome Consortium (ICGC) and the Gene Expression Omnibus (GEO) databases were set as validation datasets. Prognostic genes were screened by using univariate Cox regression. Then, a novel optimal prognostic model was developed by using least absolute shrinkage and selection operator (LASSO) Cox regression. Cell type identification by estimating the relative subsets of RNA transcripts (CIBERSORT) and estimation of stromal and immune cells in malignant tumors using expression data (ESTIMATE) algorithms were used to characterize tumor immune infiltrating patterns. The tumor immune dysfunction and exclusion (TIDE) algorithm was used to predict immunotherapy responsiveness.

Results: A prognostic signature based on five IRGs (MET, ERAP2, IL20RB, EREG, and SHC2) was constructed in TCGA-PDAC and comprehensively validated in ICGC and GEO cohorts. Multivariate Cox regression analysis demonstrated that this signature had an independent prognostic value. The area under the curve (AUC) values of the receiver operating characteristic (ROC) curve at 1, 3, and 5 years of survival were 0.724, 0.702, and 0.776, respectively. We further demonstrated that our signature has better prognostic performance than recently published ones and is superior to traditional clinical factors such as grade and tumor node metastasis classification (TNM) stage in predicting survival. Moreover, we found higher abundance of CD8+ T cells and lower M2-like macrophages in the low-risk group of TCGA-PDAC, and predicted a higher proportion of immunotherapeutic responders in the low-risk group.

Conclusions: We constructed an optimal prognostic model which had independent prognostic value and was comprehensively validated in external PDAC databases. Additionally, this five-genes signature could predict immune infiltration characteristics. Moreover, the signature helped stratify PDAC patients who might be more responsive to immunotherapy.

Keywords: clinical outcome; immune-related gene; immunotherapy; pancreatic ductal adenocarcinoma.

Figure 1.

Construction of a prognostic signature through comprehensive immunogenomic analysis. (A) Heatmap showing differentially expressed immune-related genes. (B) Volcano plot showing differentially expressed immune-related genes. (C) Forest plot of the hazard ratios showing survival-associated IRGs. (D) LASSO coefficient profiles of the 11 immune genes in TCGA-PDAC. (E) A coefficient profile plot was produced against the log₂(λ) sequence. (F) The expression patterns of different risk groups were analyzed by PCA using the five genes included in this model.

Figure 2.

Evaluation of the prognostic signature. (A) Risk score distribution in PDAC patients. (B) Survival time of PDAC patients in ascending order of the risk score. (C) A heatmap of expression profiles of the five mRNAs. (D) Kaplan–Meier curves of OS stratified by the risk score in the low- and high-risk patients. (E) ROC curves of OS for the risk signature score at 1, 3, and 5 years.

Figure 3.

Validation of the five-genes signature. (A) A meta-analysis was performed using the prognostic results of the five-genes signature in different databases. (B–D) Kaplan–Meier curves were created to estimate OS for high- and low-risk groups from different databases. (E–H) Kaplan–Meier curves were created to estimate the OS for high- and low-risk groups from four independent cohorts.

Figure 4.

Comparison of the five-genes signature with other published signatures and traditional clinical factors. (A) Five-genes signature (Lisig) compared with four signatures (Maosig, Tangsig, Tansig, and Wangsig) published recently. (B) Multivariable analysis for risk score and clinical data. (C) Comparison of sensitivity and specificity of the ROC with traditional clinical factors.

Figure 5.

Relationship between the five-genes signature and immune infiltrating characteristics and immunotherapy response. (A) Significantly enriched pathways in the high-risk group of TCGA-PDAC. (B) Comparison of the 22 types of immune cells in the low- and high-risk groups estimated by the CIBERSORT algorithm. (C) Comparison of the estimate score, stromal score, and immune score in the low- and high-risk groups estimated by the ESIMATE algorithm. (D) Correlation analysis of risk score and abundance of CD8+ T cells by Pearson's correlation test. (E) Correlation analysis of risk score and abundance of M2-like macrophages by Pearson's correlation test. (F) Comparison of the TIDE scores. (G) Correlation analysis of risk score and TIDE scores by Pearson's correlation test. (H) Comparison of the proportion of predicted immunotherapeutic responders.