Multivariate machine learning analysis of blood-based RNA expression profiles to distinguish relapsing-remitting and progressive forms of multiple sclerosis

PROJECT SUMMARY Multiple sclerosis (MS) is a chronic inflammatory disease mediated by a dysregulated immune system. Diagnosis and monitoring of MS relies on clinical symptoms and examinations outlined in the revised McDonald criteria and supported by appropriate magnetic resonance imaging (MRI) findings or other laboratory tests, such as detecting oligoclonal bands in cerebrospinal fluid and evoked potential testing. MS is classified into phenotypes depending on the patterns of demyelination of the central nervous system, inflammation, and disability progression. 80%-90% of patients will develop relapsing-remitting MS (RRMS), where symptoms develop over a few days or months and then greatly improve or remit entirely. Up to two-thirds of patients with RRMS advance to secondary progressive MS (SPMS) within 10-15 years, and up to 90% of RRMS patients will transition to SPMS within 20-25 years. SPMS patients typically experience a steady disease progression with or without relapses. Should relapses occur in SPMS, they typically do not fully remit. Early treatment with disease- modifying therapies (DMTs) has been shown to slow or prevent the transition of RRMS to SPMS. In addition to RRMS and SPMS, approximately 15% of patients will develop a primary progressive course of disease (PPMS) where disability continuously accumulates without evidence of remission. Disability in MS accrues predominantly in the progressive forms of the disease, creating a substantial care burden. Currently, more than twenty therapies are approved for RRMS, while only one treatment is approved for PPMS. Treating SPMS patients with approved RRMS DMTs remains an area of active investigation and debate. Difficulties in identifying the correct MS phenotype can lead to patients receiving or remaining on ineffective therapies, resulting in unnecessary costs and the potential for adverse reactions. As new therapies are introduced, early disease phenotype classification may represent therapeutic intervention opportunities. MS- related costs can exceed $50,000 annually, and mischaracterization of MS can be a significant cost burden since certain RRMS therapies lack evidence of slowing SPMS or PPMS. Identifying actionable biomarkers would provide clinicians with additional information for diagnosis, prognosis, clinical subtyping, and therapy selection. We hypothesize that differences in gene expression, particularly long non-coding RNAs (lncRNAs), reflect changes in a patient’s immune system at different stages of disease progression, and measuring these changes in whole blood can aid physicians in the differential diagnosis of neurodegenerative diseases, such as MS. Furthermore, applying multivariate approaches including machine learning (ML) enables the identification of RNAs signatures to uncover the transcriptional networks involved in disease development and progression. We have generated preliminary data supporting the hypothesis that measuring diverse RNA biotypes in whole blood may produce a more discriminatory test for RRMS, PPMS, and SPMS patients. Elucidation of RNAs as actionable biomarkers for MS subtypes allows early indications of unregulated, potentially destructive autoimmune processes. Identifying these changes before they become diagnostically apparent through physical examination or radiographic means permits intervention before irreversible tissue damage has occurred. In the proposed study, we will verify candidate RNAs in larger, independent cohorts of MS subtypes versus healthy and disease controls. ML classifiers will be developed with an optimal number of RNA biomarkers to distinguish MS subtypes with the greatest overall accuracy.