The Impact of Immune Microenvironment on the Prognosis of Pancreatic Ductal Adenocarcinoma Based on Multi-Omics Analysis

Bing Yang, Mingyao Zhou, Yunzi Wu, Yuanyuan Ma, Qin Tan, Wei Yuan, Jie MaFrontiers in Immunology 2021 · DOI

TL;DR

Researchers have developed a novel risk-score model for pancreatic ductal adenocarcinoma (PDAC) that leverages the infiltration levels of two specific immune cell types: CD4+ naive T cells (CD4+TN) and common lymphoid progenitors (CLP). This model significantly improves prognostic prediction accuracy and can identify patients with poorer outcomes and those who might benefit from immunotherapy, offering a more accessible alternative to complex molecular subtyping.

What Problem Does This Paper Address?

Pancreatic ductal adenocarcinoma (PDAC) is notoriously aggressive, with a very low survival rate and limited treatment options. Current prognostic tools for PDAC are insufficient, lacking the precision needed for effective patient management. While the tumor microenvironment (TME), particularly immune cell infiltration, is known to influence cancer progression, its specific role in PDAC prognosis has been unclear, with conflicting findings. Existing methods for assessing TME impact often rely on gene expression signatures or complex molecular subtypes, which can be costly and difficult to implement clinically. This paper aims to address this gap by developing a simpler, more economical, and clinically applicable prognostic model based directly on the abundance of specific infiltrating immune cells.

Methods

The study employed a multi-omics approach. First, the xCell algorithm was used to computationally infer the proportions of 64 different cell types, including 36 immune cell types, from bulk RNA sequencing data of PDAC tumors. This allowed for the identification of immune cell populations associated with patient survival.

Next, univariate and multivariate Cox regression analyses were performed to identify immune cell types that were independent predictors of prognosis. Two key cell types, CD4+ naive T cells (CD4+TN) and common lymphoid progenitors (CLP), were selected.

A Cox proportional hazards model was then constructed using the levels of CD4+TN and CLP to generate a risk-score for each patient. This risk-score was designed to quantify the overall impact of these immune cells on prognosis.

Further analyses included:

  • Kaplan-Meier (KM) survival analysis to compare survival between high and low risk-score groups.
  • Receiver Operating Characteristic (ROC) curves to assess the model's predictive accuracy over time.
  • Transcriptome and whole-exome sequencing (WES) analysis from The Cancer Genome Atlas (TCGA) to investigate the molecular mechanisms underlying the risk-score, focusing on differentially expressed genes (DEGs), copy number variations (CNVs), and tumor mutation burden (TMB).
  • Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses to understand the functional implications of DEGs.
  • Protein-Protein Interaction (PPI) network analysis to identify key molecular interactions.
  • Multiplex immunofluorescence staining was used for validation of CD4+TN and CLP levels in patient samples.

Datasets

The study utilized several publicly available and newly collected datasets:

  • TCGA-PDAC dataset: A primary dataset for model development and discovery, comprising RNA sequencing, mutation, and CNV data from 143 PDAC patients (136 after filtering).
  • Gene Expression Omnibus (GEO) datasets: Specifically, GSE71729, GSE102238, and GSE57495 were used for validation of the risk-score and findings related to CCL19.
  • Genotype-Tissue Expression (GTEx) database: Used for obtaining normal pancreatic tissue gene expression data.
  • Peking2020 cohort: A newly collected clinical validation cohort of 40 PDAC patient samples, used for validating the risk-score and assessing immune cell infiltration via multiplex immunofluorescence.

Key Insights and Ideas

  • Immune cells as direct prognostic markers: The study highlights that specific immune cell populations (CD4+TN and CLP) can serve as direct and powerful indicators of PDAC prognosis, moving beyond indirect gene expression markers.
  • Synergistic prognostic power: Combining the levels of CD4+TN and CLP in a Cox model creates a risk-score that is significantly more sensitive and accurate for predicting overall survival than considering each cell type individually.
  • Beyond "cold" tumor classification: While PDAC is often considered a "cold" tumor, this research demonstrates that the presence and balance of specific immune cells within the TME can still be highly informative about patient outcomes.
  • CCL19 as a key regulator: The study identifies CCL19 as a critical factor potentially influencing the infiltration of CD4+TN and CLP, and consequently, the risk-score and prognosis. Its copy number deletion and low expression are linked to higher risk.
  • Clinical applicability: The developed risk-score model is shown to be robust and can be validated using more accessible techniques like multiplex immunofluorescence, making it potentially more translatable to clinical practice than complex genomic analyses.

Key Results

  • Prognostic Immune Cells: CD4+ naive T cells (CD4+TN) were positively correlated with survival, while common lymphoid progenitors (CLP) were negatively correlated.
  • Risk-Score Performance: The combined risk-score model significantly improved prognostic prediction sensitivity in the TCGA cohort, with a C-index of 0.706 and AUC values of 0.713 (1-year), 0.667 (2-year), and 0.609 (3-year). This was validated in GEO datasets.
  • Molecular Subtype Association: The risk-score effectively identified molecular subtypes associated with poor prognosis, including basal-like and activated stroma subtypes, with high AUC values (e.g., 0.728 for basal-like).
  • Immunotherapy Sensitivity: Patients with higher T-cell cytotoxicity and those receiving molecular targeted therapy had lower risk-scores, suggesting potential links to immunotherapy response.
  • CCL19 as a Prognostic Factor: Low expression and copy number deletion of CCL19 were associated with higher risk-scores and poorer prognosis. CCL19 was identified as an independent prognostic factor.
  • Validation: The findings were consistently validated across TCGA, GEO, and the independent Peking2020 patient cohort.

Limitations and Future Work

  • Limited Stage Data: The study acknowledges a significant limitation in the scarcity of early-stage (Stage I) and advanced-stage (Stage IV) PDAC samples in both the TCGA and Peking2020 datasets. This may affect the model's accuracy in evaluating prognosis across all disease stages. Future work should focus on expanding the patient population, particularly with early and late-stage samples, for more comprehensive validation.
  • Immunotherapy Cohort Size: While the risk-score predicted immunotherapy sensitivity, the lack of a sufficiently large cohort of PDAC patients who received immunotherapy prevented a detailed assessment of the risk-score's impact in this specific context. Expanding this cohort is crucial for future research.
  • Anatomical Site Discrepancy: While the risk-score was not significantly correlated with tumor location in TCGA, the Peking2020 cohort showed a higher risk-score for tumors in the tail of the pancreas compared to the head. The reasons for this discrepancy require further exploration.
  • Mechanism of CCL19: While CCL19 was identified as a key factor, the precise mechanisms by which its copy number deletion or low expression impacts CD4+TN and CLP infiltration and ultimately the risk-score need further elucidation.
  • Algorithm Bias: The study used xCell as the primary algorithm for immune cell prediction. While they mention using TIMER and MCPCOUNTER for supplementary validation, the reliance on a single algorithm for initial identification could introduce bias. Future work could explore a broader range of computational immunology tools.

Summary updated

Abstract

Pancreatic ductal adenocarcinoma (PDAC) is a malignant tumor characterized by rapid progression, early metastasis, high recurrence, and limited responsiveness to conventional therapies. The 5-year survival rate of PDAC is extremely low (<8%), which lacks effective prognostic evaluation indicators. In this study, we used xCell to analyze infiltrating immune cells in a tumor and through the univariate and multivariate Cox analyses screened out two prognosis-related immune cells, CD4<sup>+</sup>T<sub>N</sub> and common lymphoid progenitor (CLP), which were used to construct a Cox model and figure out the risk-score. It was found that the constructed model could greatly improve the sensitivity of prognostic evaluation, that the higher the risk-score, the worse the prognosis. In addition, the risk-score could also identify molecular subtypes with poor prognosis and immunotherapy sensitivity. Through transcriptome and whole-exome sequencing analysis of PDAC dataset from The Cancer Genome Atlas (TCGA), it was found that copy number deletion and low expression of CCL19 might be crucial factors to affect the risk-score. Lastly, validation of the above findings was confirmed not only in Gene Expression Omnibus (GEO) datasets but also in our PDAC patient samples, Peking2020 cohort.
BibTeX
@inproceedings{yang2021impact,
  author = {Bing Yang and Mingyao Zhou and Yunzi Wu and Yuanyuan Ma and Qin Tan and Wei Yuan and Jie Ma},
  title = {{The Impact of Immune Microenvironment on the Prognosis of Pancreatic Ductal Adenocarcinoma Based on Multi-Omics Analysis}},
  booktitle = {Frontiers in Immunology},
  year = {2021},
  doi = {10.3389/fimmu.2021.769047},
}
Figure