Oxidative stress-related biomarkers in thyroid eye disease: evidence from bioinformatics analysis and experimental validation.

Background: Oxidative stress is a key contributor to the pathogenesis of the autoimmune condition thyroid eye disease (TED). However, its precise molecular mechanisms and reliable biomarkers remain unclear. Bioinformatics enables the identification of differentially expressed genes through transcriptomic analysis. However, distinguishing truly relevant findings from false discoveries remains challenging. Immunohistochemistry helps address this limitation by validating protein expression levels, revealing local immune responses, and linking microscopic tissue changes to clinical manifestations.

Methods: Oxidative stress-related differentially expressed genes (OS-DEGs) were identified. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analyses were performed to explore their biological functions and pathways. Machine learning methods, including LASSO regression and random forest, were used to select key diagnostic genes. Receiver operating characteristic curves assessed their diagnostic performance. A nomogram model was constructed using logistic regression based on selected oxidative stress-related core genes. Single-gene gene set enrichment analysis evaluated the diagnostic potential and functional relevance of these core genes. Expression of three key genes/proteins repeatedly highlighted in multi-omics TED studies was confirmed in 22 orbital tissues by immunohistochemistry with quantitative analysis using automated image tools minimizing operator bias.

Results: Fifty-three OS-DEGs were selected. GO and KEGG enrichment analyses revealed significant involvement of OS-DEGs in cellular responses to oxidative stress, ROS metabolism, and mitochondrial dysfunction, highlighting the role of oxidative damage in TED. Five diagnostic genes (, and ) were identified through machine learning approaches (LASSO regression and random forest), demonstrating strong diagnostic potential with a combined model achieving an area under the curve (AUC) of 0.931. The nomogram model developed using the selected genes showed good predictive performance for TED risk assessment. Immunohistochemical validation confirmed significant upregulation of FOS, MCL1, and ANGPTL7 in TED controls.

Conclusions: To the best of our knowledge, this study is the first to identify three oxidative stress-related genes/proteins as potential biomarkers for TED through bioinformatic analysis of multi-omics data followed by immunohistochemical validation, providing new insights into their roles in the pathogenesis of the disease. These biomarkers could aid in early screening and risk assessment for TED.