Stratification of primary antiphospholipid syndrome by mechanistic immunophenotype: machine learning identifies...

Primary antiphospholipid syndrome (pAPS) is a clinically heterogeneous disorder. This study aimed to investigate the distribution of immune cell subsets, with a focus on T-bet+CD11c+ B cells, in patients with pAPS. Furthermore, it aimed to utilise machine learning-based clustering to resolve the clinical heterogeneity of pAPS by identifying distinct immunophenotypes and exploring their associations with specific manifestations.

This study involved 54 patients with pAPS, 26 healthy controls (HC), and 16 healthy pregnant (HP) controls. The distribution of 16 T and B cell subsets was analysed using flow cytometry. K-means clustering was applied to stratify patients based on their immune profiles.

Our results indicated a significant increase in the T-bet+CD11c+ B cell population in pAPS patients compared to HCs (median 3.8% vs. 2.4%, p<0.01). More importantly, exploratory machine learning-based clustering resolved the clinical heterogeneity of thrombosis by identifying three distinct immunophenotypes. Notably, Cluster K2 (n=17), characterised by the highest T-bet+CD11c+ B cell levels, was associated with venous thrombosis (41.2%). In contrast, Cluster K1 (n=19), which featured a different immune profile, was associated with the highest rate of arterial thrombosis (31.6%). A positive correlation was also found between T-bet+CD11c+ B cell levels and pathogenic anti-β2GP1 IgG titres in patients with thrombotic events.

Our findings suggest that pAPS is not a monolithic disorder but a heterogeneous syndrome composed of distinct, mechanistically-defined immunophenotypes. Our stratification framework links T-cell-driven and T-bet+CD11c+ B cell-driven pathways to different clinical outcomes and provides novel mechanistic insights into obstetric APS.