Improving MS clinical care with an innovative lab-developed test

Multiple sclerosis (MS) is an autoimmune, neuroinflammatory disease of the central nervous system that exhibits a complex disease course and variable symptoms and manifestations. The heterogeneity in the clinical disease course of MS makes diagnosis and prognosis difficult, with no validated clinical tests to track disease activity or progression. To address this, Octave Bioscience (Menlo Park, CA) undertook a broad biomarker discovery approach to identify markers associated with clinical and radiographic endpoints in MS. The study included measurement of over 1,100 proteins using multiple Olink Target 96 panels and resulted in the identification of a 21-protein signature that could classify MS disease activity status with over 90% accuracy (area under the curve, AUC = 0.93).

This represented a substantial improvement compared to the established single biomarker, neurofilament light (NfL), which had an AUC of just 0.717. Moreover, the proteins in the model represented multiple pathways known to be significant in MS pathophysiology, emphasizing the biological relevance of the individual biomarkers.

Development and technical validation of a custom panel for MS disease activity monitoring

Building on the biomarker discovery study, the Octave team embarked on the development of a Laboratory Developed Test (LDT) with clinical utility, based on the 21 proteins identified. They designed a custom PEA panel developed at Olink, ensuring a smooth transition without the need to switch proteomics technologies.

The Multiple Sclerosis Disease Activity (MSDA) panel underwent rigorous technical validation, with fifty-one plates processed using two different production lots over the course of a year. The evaluation criteria included accuracy, precision, sensitivity, assay interference, sample stability, and sample reanalysis across different equipment, reagents, location, and personnel. Of the 21 evaluated biomarkers, 18 met the acceptability criteria and were included in the final algorithm (APLP1, CCL20, CD6, CDCP1, CNTN2, CXCL13, CXCL9,FLRT2, GFAP, IL-12β, MOG, NfL, OPG, OPN, PRTG, SERPINA9, TNFRSF10A, and TNFSF13B). The assay accuracy was confirmed by mixing samples from four affected patients at different dilutions, and the percent recovery for each biomarker was determined to be within established limits. The expected Disease Activity Scores were calculated and compared to the observed scores, with a measured r² correlation of 0.997, well above the established acceptability criterion of >0.85.

Clinical validation of the MSDA panel

After the successful analytical validation of the panel, the crucial next phase was the clinical validation of the MSDA protein signature in extended cohort studies. As described in an article published in Clinical Immunology, the 18-protein MSDA test was validated based on associations between algorithm scores and clinical/ radiographic assessments using serum samples from 614 MS patients. The multi-protein model was trained based on presence/absence of gadoliniumpositive (Gd+) lesions and was also strongly associated with new/enlarging T2 lesions, and active versus stable disease. The data obtained showed that the test was clinically validated, significantly outperforming the top-performing single-protein model and providing a quantitative tool to enhance the care of MS patients.

Real-world application of the MSDA panel

The full technical and clinical validation of the MSDA PEA panel opened up its practical application with multiple sclerosis patients. Dr. Taylor Gonyou, DO (Multiple Sclerosis Fellow, Michigan Institute for Neurologic Disorders) was the first physician to report on real-world testing in the clinic, as shown in the following video clip, recorded at the ACTRIMS Forum 2023 and showcased on NeurologyLive.

Subsequent studies have further demonstrated the clinical utility of the panel. A team from the University of Pittsburgh extended the association of the protein biomarkers beyond purely physiological endpoints (lesion counts, brain atrophy etc.) and assessed protein levels in association with patient-reported disabilities. Taking data from two independent clinical cohorts, they applied machine learning to derive models to predict a Patient Determined Disease Steps (PDDS) score. A combined feature input of routine clinical factors and the panel proteins consistently outperformed base models in predicting severe versus mild/moderate PDDS disability across multiple ML approaches, with LASSO analysis achieving the best area under the curve with an impressive accuracy of AUC = 0.91. Interestingly, when NfL or GFAP (the currently used single protein biomarkers for MS) were used in combined models with the clinical features, they produced no improvement on the clinical features alone mode, emphasizing the power of multi-protein signatures compared to single biomarkers and illustrating how the initial broad biomarker discovery phase of the project enabled a more powerful and clinically useful panel to be developed.

A recent study undertaken by Octave Bioscience examined longitudinal changes in the selected proteins in relation to specific clinical (physical, cognitive, and radiographic) outcomes, with samples compared at baseline and at the end of the clinical follow-up period (average 5.4 years). This enabled the team to develop a series of both cross-sectional and longitudinal models for protein associations with multiple outcomes. Two observations of particular interest were that increased NfL was associated with greater cognitive decline and that the baseline levels of 7 proteins showed predictive associations with longitudinal outcomes – GFAP and PRTG showed particularly robust associations after multiple testing with brain imaging and cognitive score metrics.

The findings were also of significant interest in terms of MS disease biology. The biomarkers comprising the MSDA panel cover 4 main disease pathway scores (immunomodulation, neuroinflammation, myelin biology, and neuroaxonal integrity) and the findings in this latest study indicated that the four MS pathways are differentially associated with distinct phenotypical and pathological changes. As the real-world use of this panel continues, the data suggests that its clinical utility may well be even more expansive than originally envisaged.