Harvard University, Brigham & Women's Hospital

Author Of 1 Presentation

Biomarkers and Bioinformatics Poster Presentation

P0063 - Development of a Custom Multivariate Proteomic Serum Based Assay for Association with Radiographic and Clinical Endpoints in MS (ID 833)

Abstract

Background

Multiple Sclerosis (MS) is a complex and heterogeneous disease. Investigating the biological pathways and cell types involved in MS pathophysiology as represented by protein biomarker expression can help inform the development of tools to monitor disease activity, disease progression, identify early evidence of relapse, and monitor treatment response.

Objectives

To develop a blood based multiplex proteomic assay that associates with clinical and radiographic endpoints in patients with MS. These endpoints include the presence of gadolinium-enhanced (Gd+) lesions, Annualized Relapse Rate (ARR) and clinically defined relapse status (active versus stable).

Methods

Serum samples (n=690 in total) from multiple deeply-phenotyped cohorts (ACP, CLIMB and EPIC) were tested in immunoassays for the measurement of 1196 proteins using Proximity Extension Assays (PEA) from OlinkTM and for 215 proteins using xMAPTM immunoassays from Myriad RBM, Inc. (RBM). Associated radiographic and clinical endpoints at the time of the blood draw were correlated with the protein levels. Twenty-one proteins were selected for inclusion in a custom assay based on their performance in univariate and multivariate statistical models, and replication across independent cohorts. Biological pathway modeling and network analysis were performed to ensure comprehensive representation of MS neurophysiology. Area under the curve (AUC) was selected as the key metric for model performance evaluation.

Results

Multivariate statistical ensembles restricted to the expression levels of the biomarkers selected for the custom assay achieved AUC performance of 0.827 for classification of the presence of Gd+ lesions, 0.802 for classification of clinically defined relapse status, and 0.930 for the classification of patients with Low ARR (≤0.2 relapses) vs High ARR (≥1.0 relapses). A multivariate model utilizing shifts in biomarker expression in longitudinally paired samples achieved the highest observed performance of 0.950 for classification of Gd+ lesion presence. In each case, the multivariate models significantly outperformed (p-value <0.05) the AUC of the highest performing univariate biomarker.

Conclusions

Multivariate models restricted to the 21 selected proteins effectively classified several radiographic and clinical endpoints with stronger performance than any single biomarker. A 21-plex custom assay panel is being developed for further investigation and validation using additional cohorts.

Collapse