Ferroptosis-related biomarkers for adamantinomatous craniopharyngioma treatment: conclusions from machine learning techniques

Yingqing Feng, Zhen Zhang, Jiahao Tang, Yan Chen, Dan Hu, Xinwei Huang, Fangping LiFrontiers in Endocrinology 2024 · DOI

TL;DR

This study leverages machine learning to identify novel ferroptosis-related biomarkers for adamantinomatous craniopharyngioma (ACP), a challenging brain tumor. By analyzing gene expression data and applying advanced algorithms, the researchers pinpointed a set of genes (CASP8, KRT16, KRT19, TP63, and GOT1) that show promise for improving ACP diagnosis and potentially guiding treatment strategies.

What Problem Does This Paper Address?

Adamantinomatous craniopharyngioma (ACP) is a rare but persistent epithelial neoplasm that is difficult to treat completely and has a high recurrence rate. Current treatment options often lead to significant complications, including hypothalamic dysfunction and endocrine deficiencies, impacting patient quality of life. While the genetic underpinnings of ACP, such as CTNNB1 mutations, are being explored, there's a critical need for more effective diagnostic markers and therapeutic targets. This paper addresses this gap by investigating the role of ferroptosis, a distinct form of programmed cell death, in ACP and aims to identify ferroptosis-related genes that can serve as diagnostic biomarkers.

Methods

The researchers employed a multi-pronged approach combining bioinformatics analysis and machine learning.

  • Data Acquisition and Preprocessing: Gene expression profiles for ACP and normal brain tissue samples were downloaded from the Gene Expression Omnibus (GEO) database (GSE68015 and GSE94349). These datasets were merged and normalized to account for batch effects using the sva R package.
  • Differential Gene Expression Analysis: The limma R package was used to identify differentially expressed genes (DEGs) between ACP and normal samples, with a stringent threshold of |LogFC| > 2 and P < 0.05.
  • Ferroptosis Gene Identification: A comprehensive list of ferroptosis-related factors was compiled from the FerrDb V2 and GeneCard databases.
  • Identification of Differentially Expressed Ferroptosis-Related Genes (DEFRGs): The intersection of DEGs and ferroptosis-related factors was determined to identify DEFRGs.
  • Enrichment Analysis: Gene Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG), and Disease Ontology (DO) enrichment analyses were performed to understand the biological functions and pathways associated with the DEGs. Gene Set Enrichment Analysis (GSEA) and Gene Set Variation Analysis (GSVA) were also utilized.
  • Machine Learning for Biomarker Screening: Two powerful machine learning algorithms, Least Absolute Shrinkage and Selection Operator (LASSO) and Support Vector Machine-Recursive Feature Elimination (SVM-RFE), were applied to screen for DEFRGs with high discriminative power. These algorithms are known for their ability to perform feature selection and reduce overfitting.
  • Validation and Correlation Analysis: The expression levels of the identified key DEFRGs were validated using quantitative polymerase chain reaction (q-PCR) and enzyme-linked immunosorbent assay (ELISA) on patient tissue samples. Correlation analysis was performed to understand the relationships between the identified genes and ferroptosis markers.
  • Diagnostic Model Development: A nomogram was constructed to predict the risk of disease based on the expression levels of the key DEFRGs.
  • Subtype Analysis: Consensus clustering analysis was employed to stratify ACP samples into distinct subtypes based on the expression of the identified DEFRGs.

Datasets

  • Gene Expression Omnibus (GEO) Database:
    • GSE68015: Contains 15 ACP samples and 16 normal brain tissue samples.
  • GSE94349: Contains 9 ACP and 17 normal brain tissue samples.
  • Patient Tissue Samples: 35 ACP patients and 10 normal brain tissue samples collected from The Seventh Affiliated Hospital of Sun Yat-sen University were used for validation experiments (q-PCR and ELISA).

Key Insights and Ideas

  • Ferroptosis as a Key Player in ACP: The study strongly suggests that ferroptosis is a significant biological process involved in the development and progression of ACP, opening new avenues for therapeutic intervention.
  • Machine Learning for Precision Biomarker Discovery: The application of LASSO and SVM-RFE algorithms proved highly effective in sifting through a large number of DEGs to pinpoint a small, highly discriminative set of ferroptosis-related genes. This highlights the power of machine learning in complex biological data analysis.
  • Identification of a Promising Diagnostic Panel: The identification of CASP8, KRT16, KRT19, TP63, and GOT1 as potential diagnostic biomarkers for ACP is a significant finding. Their differential expression and correlation with ferroptosis pathways provide a strong foundation for future diagnostic tools.
  • Protective vs. Risk Factors: The nomogram analysis revealed that CASP8, KRT16, KRT19, and TP63 act as protective factors against ACP risk, while GOT1 and TFAP2C are associated with increased risk. This distinction is crucial for understanding disease mechanisms and developing targeted therapies.
  • Subtype Stratification: Consensus clustering identified two distinct isoforms of ACP based on DEFRG expression, suggesting potential heterogeneity within the disease that could influence treatment response.

Key Results

  • A total of 69 DEFRGs were identified between ACP and normal samples.
  • Machine learning algorithms (LASSO and SVM-RFE) successfully narrowed down these DEFRGs to a core set of 6 overlapped genes: CASP8, GOT1, KRT16, KRT19, TFAP2C, and TP63.
  • Expression Levels:
    • CASP8, KRT16, KRT19, and TP63 were significantly upregulated in ACP patients compared to normal controls.
  • GOT1 was significantly downregulated in ACP patients.
  • TFAP2C showed no significant difference in expression between ACP and normal groups in the validation phase.
  • The ROC curves for the six diagnostic markers showed an AUC of 1, indicating high accuracy in distinguishing between ACP and normal samples.
  • Correlation analysis revealed that GOT1 was negatively correlated with other DEFRGs, while KRT16, KRT19, and TFAP2C showed positive correlations.
  • Consensus clustering analysis successfully stratified ACP samples into two distinct clusters.

Limitations and Future Work

  • Sample Size: While the study utilized publicly available datasets and a reasonable number of patient samples for validation, ACP is a rare disease. The authors acknowledge that future studies should focus on in vivo and in vitro experiments to further validate these findings.
  • TFAP2C: The lack of significant differential expression for TFAP2C in the validation phase, despite its selection by machine learning, warrants further investigation.
  • Mechanism Elucidation: While the study identifies potential biomarkers and links them to ferroptosis, further research is needed to fully elucidate the precise molecular mechanisms by which these genes influence ACP development and ferroptosis.
  • Papillary Craniopharyngioma (PCP): The study focused on adamantinomatous craniopharyngioma (ACP). Future work could explore the role of ferroptosis and these identified biomarkers in papillary craniopharyngioma (PCP).
  • Clinical Application: The authors aim to use these predicted targets for medical diagnosis and treatment of ACP, suggesting that clinical trials and further validation are necessary steps.

Summary updated

Abstract

Ferroptosis-related genes, including CASP8, KRT16, KRT19, TP63, and GOT1 were the potential markers for ACP, which lays the theoretical foundation for ACP diagnosis.
BibTeX
@inproceedings{feng2024ferroptosis,
  author = {Yingqing Feng and Zhen Zhang and Jiahao Tang and Yan Chen and Dan Hu and Xinwei Huang and Fangping Li},
  title = {{Ferroptosis-related biomarkers for adamantinomatous craniopharyngioma treatment: conclusions from machine learning techniques}},
  booktitle = {Frontiers in Endocrinology},
  year = {2024},
  doi = {10.3389/fendo.2024.1362278},
}
Figure