Identification of fibroblast-related genes based on single-cell and machine learning to predict the prognosis and endocrine metabolism of pancreatic cancer

Frontiers in Endocrinology 2023 AI 5 Explanations View Original
Original Paper (PDF)

Unable to display PDF. Download it here or view on PMC.

Plain-English Explanations
Page [1, 2]
The Role of Tumor Fibroblasts in Pancreatic Cancer

Pancreatic adenocarcinoma (PAAD) is highly lethal partly because of its dense surrounding tissue environment called the tumor stroma. A key component of this stroma is cancer-associated fibroblasts - cells that originally support normal tissue structure but become corrupted by cancer to help tumors grow, evade the immune system, and resist treatment.

Single-cell RNA sequencing technology now allows researchers to analyze the gene activity of individual cells within a tumor, revealing the previously hidden complexity of the tumor ecosystem. This technology can distinguish exactly which cells are fibroblasts, cancer cells, or immune cells - and what each type is doing.

By combining single-cell sequencing data with machine learning, this study aimed to identify which fibroblast-related genes are most important for cancer progression and to build a prognostic tool that can predict patient outcomes and metabolic status in pancreatic cancer.

TL;DR: This study used single-cell sequencing and machine learning to identify fibroblast genes that drive pancreatic cancer progression and predict patient prognosis.
Pages 4-4
Single-Cell Analysis Meets Machine Learning

The researchers analyzed the PAAD single-cell RNA dataset GSE165399, applying dimensionality reduction and clustering algorithms to identify 17 distinct cell subpopulations within pancreatic tumors. Each cluster represents a different cell type or functional state within the tumor microenvironment.

Weighted Gene Co-expression Network Analysis (WGCNA) was applied to identify groups of genes that work together and are associated with tumor-driving cell clusters. This revealed the 'brown module' as the gene network most strongly connected to tumor progression.

A Cox proportional hazards regression model was built using genes at the intersection of multiple analytical layers - tumor-associated genes, fibroblast cluster markers, and cancer-driving gene modules. This RiskScore model was validated across multiple independent patient datasets using Kaplan-Meier survival curves and ROC analysis.

TL;DR: Single-cell data from 17 tumor cell clusters was analyzed with WGCNA and Cox regression to build a multi-gene RiskScore for predicting pancreatic cancer patient outcomes.
Pages 7-8
A Ten-Gene Score That Predicts Cancer Outcome

The analysis identified fibroblast cluster C11 as the tumor microenvironment cell population most associated with cancer progression. Genes expressed in C11 fibroblasts showed the strongest overlap with cancer-driving pathways in the brown co-expression module.

The final RiskScore model incorporated 10 key genes: APOL1, BHLHE40, CLMP, GNG12, LOX, LY6E, MYL12B, RND3, SOX4, and the composite RiskScore. Kaplan-Meier analysis confirmed that patients classified as high-risk by RiskScore had significantly shorter survival times.

Validation across multiple independent datasets demonstrated consistent performance, with ROC curves showing strong predictive accuracy. Importantly, the low-risk group identified by RiskScore showed higher endocrine metabolic activity and lower immune cell infiltration in the tumor.

TL;DR: A ten-gene RiskScore built from fibroblast markers accurately predicted survival across multiple datasets, with high-risk patients showing lower endocrine metabolism and more immune suppression.
Page [10, 11]
Fibroblasts Shape Both Immunity and Metabolism in Pancreatic Cancer

The study found a striking connection between the fibroblast-driven RiskScore and endocrine metabolic pathways. Low-risk patients (better prognosis) had higher activity in normal endocrine metabolic processes - a finding with implications for the well-known association between pancreatic cancer and diabetes.

High-risk patients showed increased immune cell infiltration but in a suppressed, dysfunctional state - a pattern associated with immune evasion rather than anti-tumor immunity. This suggests that fibroblasts may actively create an immunosuppressive environment that helps cancer cells escape destruction.

The trajectory analysis of single-cell data showed that fibroblasts (C11 cluster) had low internal heterogeneity, meaning they are a relatively uniform population with consistent pro-tumor functions - making them a potentially attractive therapeutic target.

TL;DR: High-risk fibroblast activity suppresses anti-tumor immunity and disrupts endocrine metabolism, linking tumor biology to both cancer outcomes and diabetes risk in pancreatic cancer.
Page [12, 13]
Clinical Applications of the RiskScore System

The RiskScore based on 10 fibroblast-related genes could be applied to tumor tissue from newly diagnosed pancreatic cancer patients to stratify them into high- and low-risk groups. This could help guide decisions about the intensity of treatment and the appropriateness of clinical trial enrollment.

Several of the 10 genes in the RiskScore have known biological functions that make them attractive therapeutic targets. LOX (lysyl oxidase) promotes fibrosis and tissue stiffening that helps cancer invade, while SOX4 is a transcription factor involved in cancer stem cell maintenance.

The connection between RiskScore and endocrine metabolism suggests this tool could help identify pancreatic cancer patients most at risk of developing diabetes or metabolic complications, enabling earlier preventive interventions in this already vulnerable patient population.

TL;DR: The 10-gene RiskScore could stratify patients for treatment intensity, identify therapeutic targets, and help predict metabolic complications including diabetes risk in pancreatic cancer.
Citation: Open Access, 2023. Available at: PMC10425556.