Identification of distinct subgroups in Chinese patients with Behçet's syndrome via cluster analysis of immune cells and clinical features.

OBJECTIVES

This study aimed to use machine learning to explore Behçet's syndrome (BS) heterogeneity by integrating immunocyte subpopulations and clinical characteristics.

METHODS

We prospectively enrolled BS patients and recorded their demographic and clinical characteristics. Various peripheral immune cells were analysed using flow cytometry. Unsupervised machine learning was used to perform cluster analysis based on the clinical manifestations and immune cell subsets. Patients were followed up for one year to evaluate treatment response and remission rates. RNA sequencing was performed in patients with clustered BS and healthy controls.

RESULTS

Unsupervised machine learning categorized 201 BS patients into four clusters with distinct clinical and immunological features. Cluster 1 showed isolated mucocutaneous lesions, low inflammation, and high remission, with transcriptomic enrichment in IFN-γ, IL-6, and JAK-STAT pathways. Cluster 2 featured arthritis, elevated inflammatory levels, and responded well to TNF-α inhibitors, with transcriptomic enrichment in TNF and B-cell activation pathways. Cluster 3 had cardiovascular involvement, reduced CLA+ Tregs, and also responded to TNF-α inhibitors, with transcriptomic enrichment in coagulation, platelet activation, and MAPK pathways. Cluster 4 demonstrated neurological involvement, elevated CD161⁺ Tregs, low remission, and a better response to mycophenolate mofetil, with transcriptomic enrichment in T-cell activation and NF-κB pathways.

CONCLUSION

Unsupervised clustering of BS patients revealed four distinct subtypes with significant clinical and immunological heterogeneity, which may provide a foundation for mechanistic studies and personalized treatment.