Serum metabolomic analysis reliably differentiates between multiple sclerosis sub-types
Nicola R. Sibson, James R. Larkin*, Alex M. Dickens*, Julian L. Griffin, Ana Cavey, Lucy Matthews, Benjamin G. Davis, Timothy D.W. Claridge, Jacqueline Palace, Daniel C. Anthony
Poster at Brain (XXVIth International Symposium on Cerebral Blood Flow, Metabolism and Function), Shanghai, China (2013)
Abstract
Introduction: Ascertainment of patient transition from a relapse-remitting (RR) to secondary progressive (SP) phase of multiple sclerosis (MS) is currently difficult. To be clinically certain, a patient may be observed for up to a year. This transition is key because most MS-related disability accrues during the SPMS phase, and effective therapy differs between the two phases. The aim of this work, therefore, was to determine whether a biofluid metabolomics approach, which we have previously demonstrated distinguishes between different brain lesions1, is able to differentiate between sub-types of MS.
Methods: Serum samples from three sets of RRMS and SPMS patients, and age-matched controls, were collected (Set 1: n=15, 38 and 10; Set 2: n=6, 10 and 7; Set 3: n=5, 10 and 7, respectively). For Set 1 serum samples were also collected from 13 primary progressive (PP) MS patients. 1H nuclear magnetic resonance (NMR) spectra were acquired for each sample and partial least squares discriminant analysis was used to identify differences between groups.
Results: Set 1 was used to build significantly predictive models separating control volunteers from SPMS (q2=0.42), RRMS (q2=0.61) and PPMS (q2=0.70) patients, where q2>0.4 is considered significant. Moreover, comparison of RRMS and SPMS patients yielded a significant model (q2=0.45). In contrast, it was not possible to differentiate PPMS from either SP or RRMS.
To confirm these findings, a new set of models were constructed using Set 2. These models were also significant and identified the same metabolites as key in driving the separations. Subsequently, the group membership of each patient in Set 3 was predicted using the models built from Set 2. The models were highly predictive: RRMS vs. SPMS model, sensitivity = 0.9 and specificity = 0.8; RRMS and SPMS vs. control models, sensitivity and specificity = 1.0 (p<0.01).
Conclusions: We have demonstrated for the first time that it is possible to distinguish between RRMS and SPMS patients in a predictive fashion using an objective marker, as well as being able to separate each class of MS patients from a control cohort. The separations are based on the complete metabolomic profile of each patient and are not biased towards one metabolite in particular.
1. FEBS Lett 2004 568:49
