AI Insight
This study identified five genes (PCOLCE, KAZALD1, PDE3A, CRIP1, and ID1) associated with osteoarthritis that overlap with a predefined set of genes related to N4-acetylcytidine (ac4C) RNA modification. Using machine learning analysis of multiple gene expression datasets, researchers developed classification models that achieved high diagnostic accuracy (mean AUC of 0.910) in distinguishing osteoarthritis samples from controls. Four of the five genes showed increased expression in laboratory-treated mouse cartilage cells, though the authors emphasize these findings are exploratory and do not establish direct ac4C modification or clinical validity.
Why it matters
This research suggests potential new molecular markers for osteoarthritis classification and points toward a possible role of RNA modifications in the disease. The identified gene signature could inform future diagnostic approaches or therapeutic targets, though substantial additional validation would be required before clinical application.
Understand the Science
by Tianyang Li, Jinpeng Wei, Hua Wu, Ming Zhang
Objective
To identify osteoarthritis (OA) associated transcripts overlapping a predefined N4-acetylcytidine (ac4C) related gene set and evaluate their potential as an exploratory classification signature and basis for expression based clustering.
Methods
Five GEO datasets were analyzed using differential expression, functional enrichment, weighted gene co-expression network analysis, immune-signature scoring, machine learning, SHAP interpretation, consensus clustering, and gene set variation analysis. The predefined 2,135-gene set was derived from a published ac4C-RIP-seq comparison between wild-type and NAT10-deficient HeLa cells and was used only for candidate filtering. Twelve algorithms were combined into 113 two-stage feature-selection/classification pipelines, which were ranked by the mean area under the receiver operating characteristic curve (AUC) across the development cohort and two external evaluation cohorts. Five retained genes were assessed by qRT-PCR in IL-1β-treated primary mouse chondrocytes with three biological replicates per group.
Results
Among 441 differentially expressed genes, eight overlapped the ac4C related set. Three pipelines shared the highest mean AUC of 0.910. The representative glmBoost–Naive Bayes pipeline achieved AUCs of 0.883 (95% CI, 0.783–0.959), 0.980 (95% CI, 0.880–1.000), and 0.867 (95% CI, 0.600–1.000) in the development cohort, GSE114007, and GSE169077, respectively. Because the two secondary cohorts contributed to pipeline ranking, these estimates represent exploratory evaluation rather than independent validation. Ultimately, five genes were retained, including PCOLCE, KAZALD1, PDE3A, CRIP1, and ID1. Kazald1, Pde3a, Crip1, and Id1 showed nominally significant increases after interleukin-1βtreatment, whereas Pcolce did not. Immune signature differences and the two cluster solution were exploratory.
Conclusions
A five gene OA associated transcriptomic signature linked to a predefined ac4C related gene set was identified. These findings are hypothesis generating and do not establish direct ac4C modification, independent clinical validity, or reproducible molecular subtypes.