Metabolomics Profiling Implicates Altered Lipid Metabolism in Neuromyelitis Optica Spectrum Disorder

Objective: To identify serum biomarkers predictive of neuromyelitis optica spectrum disorder (NMOSD). Background: NMOSD is a rare demyelinating, autoimmune disease. The autoimmune target is the astrocyte water channel aquaoporin-4, and its antibodies (APQ4-IgG) are highly specific for NMOSD, though 10–27% of affected persons will be APQ4-IgG negative. Thus, some affected persons may experience highly variable diagnostic delays, including misdiagnoses. There are opportunities to identify novel NMOSD-specific biomarkers by comparing metabolomic profiles in NMOSD to multiple sclerosis (MS) and unaffected controls. Design/Methods: This study was based on serum samples available through the Accelerated Cure Project, from 24 NMOSD (50% APQ4-IgG positive), 70 relapsing remitting (RR) MS, and 83 unaffected controls (UC) who self-identified as non-Hispanic whites. All NMOSD+RRMS cases were immunomodulatory therapy naïve/free (>90 days), <5 years from first symptom, and <2 years from diagnosis. Untargeted metabolomic profiles were generated, and after quality control and normalized/standardization, there were 952 named biochemical traits for analysis. A supervised machine-learning algorithm, Random forests, identified biochemical traits informative for NMOSD in comparison to MS+UC. Multivariable regression analyses characterized associations adjusting for age, sex, smoking status, and body mass index. Receiver operator curves (ROC) evaluated the predictive capacity of top-ranking metabolites. Results: Random forests determined 4 metabolites as informative for NMOSD vs MS+UC. They included a ceramide and 3 monoacylglycerols. They did not differ between APQ4-IgG positive and negative samples (p>0.2) nor between MS and UC samples (p>0.25). The metabolites were >1 standard deviation higher in NMOSD compared to MS+UC even with adjustment for potential confounders (p: 1×10−5 to 2×10−10), and were highly predictive of NMOSD status (area under ROC >80%) in comparison to MS+UC and versus UC alone. The full parameterized multivariable model was highly predictive (area under ROC=91.5%). Conclusions: We observed a serum biosignature highly predictive of NMOSD, implicating sphingomyelin and triglyceride metabolic processes. Disclosure: Dr. Misicka has nothing to disclose. The institution of Prof. Briggs has received research support from NIH. The institution of Prof. Briggs has received research support from Michael J. Fox Foundation.