Multivariate protein biomarker models more accurately predict multiple sclerosis MRI disease activity compared to serum levels of neurofilament light chain alone
ECTRIMS Online Library. Chitnis T. Sep 13, 2019; 278535; P1333
Tanuja Chitnis
Tanuja Chitnis
Contributions
Abstract

Abstract: P1333

Type: Poster Sessions

Abstract Category: Pathology and pathogenesis of MS - Biomarkers

T. Chitnis1, H. Yano1, S. Saxena1, H. Lokhande1, N. Sattarnezhad1, M.C. Manieri1, A. Paul1, F. Saleh1, M. Collins1, B. Glanz1, C. Guttmann1, R. Bakshi1, F. Qureshi2, M. Becich2, R. Osan2, V. Gehman2, H. Weiner1

1Brigham & Women's Hospital, Harvard Medical School, Boston, MA, 2Octave Bioscience, Menlo Park, CA, United States

Introduction: Serum levels of neurofilament light chain (sNfL) are associated with neurodegeneration in Multiple Sclerosis (MS) and correlate with measurements of disease activity (DA), including the presence of gadolinium enhancing (GAD+) lesions. MS is a complex disease. Many inflammatory and immune-modulated biological pathways associated with neurodegeneration may impact MS pathophysiology. Investigating these pathways, as represented by protein biomarker expression, can provide deeper insights and reveal stronger correlations to radiographic DA than sNfL alone.
Aims: To compare the performance of multivariate protein biomarker models with sNfL individually to classify samples from subjects with and without GAD+ lesions on brain MRI from the Comprehensive Longitudinal Investigation of Multiple Sclerosis at Brigham and Women's Hospital (CLIMB) study.
Methods: Serum samples from two longitudinal timepoints (median interval 11.3 months), one with zero GAD+ lesions and one with ≥ 1 GAD+ lesions, each drawn within close proximity (median interval 0 days) to a contrast-enhanced brain MRI scan, from 29 subjects with MS were measured for 1104 proteins including sNfL using Proximity Extension Assays (PEA) from Olink. Univariate/multivariate machine learning-driven biostatistical techniques were then applied to the data. Cross-validation and regularization methods were used to minimize overfitting and ensure generalizability for predicting DA of new samples. Accuracy and area under the curve (AUC) were selected as the key metrics for comparison.
Results: Performance of sNfL as a single biomarker used for classification (of lesion presence) achieved 0.617 ± 0.064 Accuracy and 0.686 ± 0.070 AUC. A multivariate classifier that consisted of 10 biomarkers (including sNfL) improved performance with 0.910 ± 0.038 Accuracy and 0.938 ± 0.034 AUC. Furthermore, sNfL alone was unable to distinguish samples with 0 vs. 1 lesion (p = 0.138) while multivariate biomarker models were able to at a statistically significant level (p < 0.01). Biomarkers that were identified as important features in multivariate classifiers were investigated to establish causation using biological pathway and network models.
Conclusions: Multivariate protein biomarker models representing several biological pathways predicted radiographic DA with greater statistical significance than sNfL alone. Analysis of 260 additional samples is underway to confirm the observed performance of the multivariate models.
Disclosure: Tanuja Chitnis has served on advisory boards for Biogen, Novartis, and Sanofi-Genzyme; has participated in clinical trials sponsored by Sanofi-Genzyme and Novartis; has received research support from the Department of Defense, National MS Society, Guthy Jackson Charitable Foundation, Novartis, Octave, Serono and Verily
Hajime Yano has received research support has received the research grant from Yoshida Scholarship Foundation, Japan
Shrishti Saxena has received research support from Octave, Serono and Verily
Hrishikesh Lokhande has received research support from Serono and Verily
Neda Sattarnezhad has received research support from Serono and Verily
Maria Claudia Manieri has no financial conflicts of interest to disclose
Anu Paul has no financial conflicts of interest to disclose
Fermisk Saleh has no financial conflicts of interest to disclose
Mikaela Collins has no financial conflicts of interest to disclose
Bonnie Glanz has received research support from Serono and Verily
Charles Guttmann has no financial conflicts of interest to disclose
Rohit Bakshi has received consulting fees from Bayer, Biogen, Celgene, EMD Serono, Genentech, Guerbet, Sanofi-Genzyme, and Shire and research support from EMD Serono and Sanofi-Genzyme
Ferhan Qureshi, Michael Becich, Remus Osan, and Victor Gehman are employees of Octave Bioscience.
Howard Weiner reports grants from National Institutes of Health, grants from National Multiple Sclerosis Society, grants from Verily, grants from EMD Serono, grants from Biogen, grants from Teva Pharmaceuticals, grants from Sanofi, grants from Novartis, grants and personal fees from Genentech, Inc, grants and personal fees from Tilos Therapeutics, personal fees from Tiziana Life Sciences, personal fees from IM Therapeutics, personal fees from MedDay Pharmaceuticals, personal fees from vTv Therapeutics, outside the submitted work.

By clicking “Accept Terms & all Cookies” or by continuing to browse, you agree to the storing of third-party cookies on your device to enhance your user experience and agree to the user terms and conditions of this learning management system (LMS).

Cookie Settings
Accept Terms & all Cookies