Identification of distinct immune signatures in inclusion body myositis by peripheral blood immunophenotyping using machine learning models

Abstract

Abstract Objective Inclusion body myositis (IBM) is a progressive late‐onset muscle disease characterised by preferential weakness of quadriceps femoris and finger flexors, with elusive causes involving immune, degenerative, genetic and age‐related factors. Overlapping with normal muscle ageing makes diagnosis and prognosis problematic. Methods We characterised peripheral blood leucocytes in 81 IBM patients and 45 healthy controls using flow cytometry. Using a random forest classifier, we identified immune changes in IBM compared to HC. K‐means clustering and the random forest one‐versus‐rest model classified patients into three immunophenotypic clusters. Functional outcome measures including mTUG, 2MWT, IBM‐FRS, EAT‐10, knee extension and grip strength were assessed across clusters. Results The random forest model achieved a 94% AUC ROC with 82.76% specificity and 100% sensitivity. Significant differences were found in IBM patients, including increased CD8+ T‐bet+ cells, CD4+ T cells skewed towards a Th1 phenotype and altered γδ T cell repertoire with a reduced proportion of Vγ9+Vδ2+ cells. IBM patients formed three clusters: (i) activated and inflammatory CD8+ and CD4+ T‐cell profile and the highest proportion of anti‐cN1A‐positive patients in cluster 1; (ii) limited inflammation in cluster 2; (iii) highly differentiated, pro‐inflammatory T‐cell profile in cluster 3. Additionally, no significant differences in patients' age and gender were detected between immunophenotype clusters; however, worsening trends were detected with several functional outcomes. Conclusion These findings unveil distinct immune profiles in IBM, shedding light on underlying pathological mechanisms for potential immunoregulatory therapeutic development.

Keywords

• AI
• IBM
• inflammatory myopathies
• machine learning
• random forest