Exploitation and Verification of a Stroma- and Metastasis-Associated Risk Prognostic Signature in Pancreatic Adenocarcinoma

Abstract

Pancreatic adenocarcinoma (PAAD), one of the most malignant tumors, not only has abundant mesenchymal components, but is also characterized by an extremely high metastatic risk. The purpose of this study was to construct a model of stroma- and metastasis-associated prognostic signature, aiming to benefit the existing clinical staging system and predict the prognosis of patients. First, stroma-associated genes were screened from the TCGA database with the ESTIMATE algorithm. Subsequently, transcriptomic data from clinical tissues in the RenJi cohort were screened for metastasis-associated genes. Integrating the two sets of genes, we constructed a risk prognostic signature by Cox and LASSO regression analysis. We then obtained a risk score by a quantitative formula and divided all samples into high- and low-risk groups based on the scores. The results demonstrated that patients with high-risk scores have a worse prognosis than those with low-risk scores, both in the TCGA database and in the RenJi cohort. In addition, tumor mutation burden, chemotherapeutic drug sensitivity and immune infiltration analysis also exhibited significant differences between the two groups. In exploring the potential mechanisms of how stromal components affect tumor metastasis, we simulated different matrix stiffness in vitro to explore its effect on EMT key genes in PAAD cells. We found that cancer cells stimulated by high matrix stiffness may trigger EMT and promote PAAD metastasis.

Keywords: metastasis; pancreatic adenocarcinoma; risk prognostic signature; stroma; tumor microenvironment.

Figure 1

Workflow diagram of this study.

Figure 2

Exploration of Stroma- and Metastasis-Associated Genes. (A–C) Simulation of the effect of different matrix stiffness on the expression of key genes in EMT In vitro; (D) Volcano plot of stroma-associated DEGs based on TCGA-PAAD data; (E) Volcano plot of metastasis-associated DEGs based on RenJi cohort transcriptomics data; (F) Venn diagrams for screening stroma- and metastasis-associated genes; (G) Top 15 GO analysis terms for stroma- and metastasis-associated genes; (H) Stroma- and metastasis-associated genes of the top 30 most enriched KEGG pathways. * p < 0.05, ** p < 0.01, *** p < 0.001.

Figure 3

Development of Stroma- and Metastasis-Associated Risk Prognostic Signature. (A) Cross-validation plot for penalty term. (B) LASSO analysis plot of the stroma- and metastasis-associated genes. (C) Forest plot of seven stroma- and metastasis-associated genes. (D) Coefficient of the four elected genes. (E–H) Kaplan–Meier curves of four stroma- and metastasis-associated genes in TCGA database.

Figure 4

The Prognostic Value of Stroma- and Metastasis-Associated Risk Score Model in the Training Group. (A) The heat map illustrated the expression of four genes in each sample. (B) The distribution of low- and high-risk samples. (C) The correlation of risk score, survival status and survival time (D) ROC curve of the risk score. (E,F) Kaplan–Meier curve of overall survival and progression-free survival in low- and high-risk groups. (G,H) The univariate and multivariate Cox regression analysis of clinical characteristics and risk score.

Figure 5

The Prognostic Value of Stroma- and Metastasis-Associated Risk Score Model in the Testing Group. (A) The heat map illustrated the expression of four genes in each sample. (B) The distribution of low- and high-risk samples. (C) The correlation of risk score, survival status and survival time (D) ROC curve of the risk score. (E,F) Kaplan–Meier curve of overall survival and progression-free survival in low- and high-risk groups. (G,H) The univariate and multivariate Cox regression analysis of clinical characteristics and risk score.

Figure 6

Identifying the Predictive Capability of Risk Signatures for Prognosis. (A) Establishment of a nomogram based on the signature to predict the 1-, 2-, and 3-year OS. (B) One-year nomogram calibration curves of the TCGA cohort. (C) Two-year nomogram calibration curves of the TCGA cohort. (D) Three-year nomogram calibration curves of the TCGA cohort. (E–G) The time-dependent ROC of the nomogram based on the OS. (H) The concordance index (c-index) for the risk signature and other clinical characteristics. * p < 0.05, *** p < 0.001.

Figure 7

Correlation of Risk Prognosis Signature with Tumor Mutational Burden. (A,B) The waterfall plots of mutant genes in high- and low-risk groups. (C) The Violin plot of TMB score in high- and low-risk groups. (D) Kaplan–Meier survival curve of high-TMB and low-TMB. (E) Kaplan–Meier survival curve stratified by risk score and TMB. (F) Kaplan–Meier survival curve of KRAS-mutant and KRAS-wild. (G) Kaplan–Meier survival curve stratified by risk score and KRAS. (H) Kaplan–Meier survival curve of TP53-mutant and TP53-wild. (I) Kaplan–Meier survival curve stratified by risk score and TP53.

Figure 8

Correlation of Risk Prognostic Signature and Tumor Immune Microenvironment. (A) Differences in immune function between the high- and low-risk groups; (B) Differences in the proportions of different immune cells in the high- and low-risk groups; (C–F) Correlation between risk scores and key immune cells; (G) Differences in the expression of key molecules of immune checkpoints between the high- and low-risk groups. * p < 0.05, ** p < 0.01, *** p < 0.001.

Figure 9

Correlation of Risk Prognostic Signature and Chemotherapy Drug Sensitivity. (A–J) Differences in sensitivity (IC50) for 10 common chemical drugs between high- and low-risk groups.

Figure 10

Verification of the Stroma- and Metastasis-Associated Risk Prognostic Model in Public Database and RenJi Samples. (A–D) The expression profile of the risk gene signature in TCGA and GTEx datasets (tumor tissue was indicated in red and normal tissue in gray); (E,F) Survival curves regarding high- and low-risk groups in GSE57495 database and RenJi sample; (G) Different TNM stage counts in the high- and low-risk groups in RenJi sample; (H,I) Differences in tumor markers (CA19-9 and CEA) between high- and low-risk groups. * p < 0.05.