Comprehensive analysis of anoikis-related gene signature in ulcerative colitis using machine learning algorithms

Peng Liu, Chunyan Sun, Xiaojuan Wang, Bing Han, Fuchun Sun, Yanbing Liu, Xin ZengFrontiers in Medicine 2025 · DOI

TL;DR

Researchers have identified a five-gene "anoikis" signature (PDK4, CEACAM6, CFB, CX3CL1, and HLA-DMA) that accurately identifies ulcerative colitis (UC) and correlates with specific immune cell infiltration. By combining three machine learning algorithms with wet-lab validation, the study reveals how programmed cell death triggered by cell-matrix detachment plays a pivotal role in gut inflammation and patient stratification.

What Problem Does This Paper Address?

Ulcerative colitis (UC) is a grueling chronic condition characterized by persistent inflammation of the colon. While we know it involves a haywire immune response, its exact origins remain idiopathic. A major challenge in treating UC is the lack of precise diagnostic biomarkers and a limited understanding of the molecular "off-switches" that fail in the intestinal lining.

The authors focus on anoikis—a specific type of programmed cell death that occurs when a cell detaches from the extracellular matrix (ECM). While well-studied in cancer (where cells evolve to resist anoikis to metastasize), its role in the inflammatory landscape of the gut has been a "black box." If intestinal epithelial cells are dying prematurely or incorrectly via anoikis pathways, it could compromise the gut barrier and trigger the massive immune infiltration seen in UC.

Methods

The study employed a multi-stage bioinformatics pipeline integrated with experimental validation:

  1. Feature Selection via Machine Learning: To find the most important genes among 58 anoikis-related candidates, the team used an ensemble of three algorithms:
  2. LASSO Cox regression: For variable selection and regularization.
  3. Random Forest (RF): To rank genes by importance scores.
  4. Support Vector Machine (SVM): To identify features through recursive elimination.
  5. Immune Profiling: They used ssGSEA (single-sample Gene Set Enrichment Analysis) to quantify the presence of 23 different immune cell types in UC samples.
  6. Unsupervised Clustering: The researchers used the identified gene signature to see if UC patients naturally fall into distinct biological subgroups (subtypes).
  7. Wet-Lab Validation: Crucially, they validated their findings using a DSS-induced colitis mouse model, performing qPCR to confirm that these genes actually change expression in a living system during active inflammation.

Datasets

  • Discovery/Integration: Three datasets from the Gene Expression Omnibus (GEO): GSE75214, GSE92415, and GSE16879.
  • Gene Sets: 58 anoikis-related genes (ARGs) identified via the GSEA database.
  • Regulatory Mapping: The RegNetwork database was used to predict upstream transcription factors and miRNAs.

Key Insights and Ideas

  • The "Core Five" Signature: Only five genes survived the rigorous three-algorithm filter: PDK4, CEACAM6, CFB, CX3CL1, and HLA-DMA. These genes aren't just random markers; they are functional players in metabolism, complement activation, and immune signaling.
  • The PDK4 Outlier: Interestingly, while four of the signature genes were upregulated in UC and positively correlated with immune cells, PDK4 was downregulated and negatively correlated with all immune cells. This suggests PDK4 might normally act as a metabolic stabilizer that is lost during the progression of UC.
  • Patient Stratification: The signature successfully split UC patients into two clusters (A and B). Cluster A showed high expression of the signature genes and was heavily associated with "hot" immune pathways (like TNF-α and IL-6 signaling), whereas Cluster B was more associated with metabolic pathways like bile acid metabolism. This suggests we shouldn't treat all UC patients as a monolithic group.
  • Regulatory Hubs: The study identified CTCF and TP53 as potential "master regulators" of this anoikis signature, providing new targets for future drug development.

Key Results

  • Diagnostic Power: The five-gene signature demonstrated exceptional accuracy, with Area Under the Curve (AUC) values $\ge$ 0.921 for all genes, indicating they are highly reliable biomarkers for distinguishing UC from healthy tissue.
  • Immune Synergy: 22 out of 23 immune cell types were significantly elevated in UC samples compared to controls. CEACAM6, CFB, CX3CL1, and HLA-DMA were strongly linked to the infiltration of activated B cells, macrophages, and Th17 cells.
  • In Vivo Confirmation: In the mouse model, the expression patterns matched the human data perfectly: H2-DMa (the mouse ortholog of HLA-DMA), CX3CL1, and CFB increased, while PDK4 decreased during colitis.

Limitations and Future Work

The authors acknowledge that their primary data comes from public repositories, which can introduce "batch effects" despite correction. Furthermore, the study lacks diversity in terms of the racial and regional backgrounds of the patients.

Future work needs to move beyond correlation. While we know these genes are present and predictive, we don't yet know if knocking them out or overexpressing them can actually cure or cause colitis. The next step will be functional studies to see if modulating the "anoikis threshold" of the gut lining can prevent the barrier breakdown that defines Ulcerative Colitis.

Summary updated

Abstract

Ulcerative colitis (UC) is a chronic inflammatory bowel disease with an idiopathic origin, characterized by persistent mucosal inflammation. Anoikis is a programmed cell death mechanism activated during carcinogenesis to eliminate undetected isolated cells from the extracellular matrix. Although existing evidence indicates that anoikis contributes to the modulation of immune response, the involvement of anoikis-related genes (ARGs) in UC pathogenesis and their interaction with infiltrating immune cells has not been thoroughly explored. The GSE75214, GSE92415, and GSE16879 datasets were acquired and integrated from the GEO database. Additionally, 58 ARGs were identified through the GSEA database. Key anoikis-DEGs in UC were identified using three machine learning algorithms, including least absolute shrinkage and selection operator (LASSO) Cox regression, random forest (RF), and support vector machine (SVM). Receiver operating characteristic (ROC) analysis was utilized to evaluate the diagnostic accuracy of each gene. Subsequently, Single sample GSEA (ssGSEA) was executed to explore the relationships within immune cell infiltration, UC subtypes, and key anoikis-DEGs. Besides, unsupervised cluster analysis was conducted to categorize the UC samples into distinct subgroups, followed by comparing subtype differences. Finally, the upstream regulatory network was constructed and visualized. A comprehensive analysis of the involvement of ARGs in UC was performed, revealing their expression profile, correlation with infiltrating immune cells, and enrichment analyses. We identified five key anoikis-DEGs (<i>PDK4</i>, <i>CEACAM6</i>, <i>CFB</i>, <i>CX3CL1</i>, and <i>HLA-DMA</i>) and demonstrated their high diagnostic accuracy for UC. Moreover, <i>CEACAM6</i>, <i>CFB</i>, <i>CX3CL1</i>, and <i>HLA-DMA</i> exhibited positive associations with infiltrating immune cells in UC, whereas <i>PDK4</i> displayed a negative correlation with all immune cells. Unsupervised cluster analysis enabled the classification of UC patients into two clusters, both of which exhibited distinct gene expression profiles and immune signaling pathways. Further, based upon the upstream regulatory network, TP53, RARB, RXRB, and CTCF potentially exerted regulatory functions. Our analysis identified five key anoikis-DEGs as characteristic biomarkers of UC. These genes were strongly associated with the infiltration of both innate and adaptive immune cells, as well as immune pathways. This study highlights the role of anoikis genes in UC pathophysiology and offers valuable insights for further elucidating UC pathogenesis and individualized therapy.
BibTeX
@inproceedings{liu2025comprehensive,
  author = {Peng Liu and Chunyan Sun and Xiaojuan Wang and Bing Han and Fuchun Sun and Yanbing Liu and Xin Zeng},
  title = {{Comprehensive analysis of anoikis-related gene signature in ulcerative colitis using machine learning algorithms}},
  booktitle = {Frontiers in Medicine},
  year = {2025},
  doi = {10.3389/fmed.2025.1498864},
}
Figure