Splicing-based biomarkers define a robust multigene classifier for relapsing-remitting multiple sclerosis

Background

Alternative splicing (AS) is recognized as a key mechanism in multiple sclerosis (MS). We aimed to construct and validate a multivariate AS-based classifier (MS-Splicing Score, MS-SS) for the discrimination of relapsing-remitting MS (RRMS) patients from healthy controls.

Methods

Three AS events (IFNAR2 exon-8 skipping, NFAT5 exon-2 skipping, PRKCA exon-3∗ inclusion) were selected based on functional and literature evidence. Isoforms were quantified via fluorescent-competitive RT-PCR in peripheral blood RNA from two independent cohorts (Italy: 37 RRMS, 50 controls; USA: 29 RRMS, 20 controls). A logistic regression model was trained to derive the MS-SS, followed by ROC analysis.

Results

The MS-SS distinguished RRMS patients from controls in both cohorts (Italy: p = 0.00083, AUC = 0.71, 95 %CI = 0.59–0.82; USA: p = 0.00074, AUC = 0.77, 95 %CI = 0.63–0.90). In the pooled dataset, the score remained significantly elevated in MS (p = 5.9 × 10⁻⁶, AUC = 0.72, 95 %CI = 0.64–0.81), and a PCA-based refinement improved classification accuracy, yielding an AUC = 0.87 (95 %CI = 0.81–0.94). At the optimal cutoff (Youden's index), the score achieved a sensitivity of 80 % and specificity of 86 %. Supervised rule-based modeling using a logic-learning machine algorithm identified interpretable splicing thresholds and enabled clinical classification at the individual level.

Conclusion

Our study introduces a novel, robust AS-based classifier for RRMS and proposes a strategy for transcriptome-based biomarker development in neuroimmunology. However, the relatively small sample sizes within each cohort may limit the generalizability of these findings, warranting larger validation studies to confirm the clinical utility of this biomarker.