Comparing associations found between unannotated and annotated metabolomics in a study of behavioral outcomes in children with cerebral malaria

Abstract

OPS 59: The Exposome: progress in methods and applications, Room 412, Floor 4, August 28, 2019, 1:30 PM - 3:00 PM Background: Untargeted metabolomics analyses include unannotated feature sets in which a subset of features are annotated. In a cohort of preschool (<5 years old; n=99) Ugandan children with cerebral malaria (CM), the cerebrospinal fluid (CSF) metabolome was measured during hospitalization. CM is the most severe form of malaria and has been associated with long-term behavioral impairment. Child Behavior Checklist (CBCL) scores were measured in children with CM 12 months after the children’s hospitalization to analyze behavioral outcomes. Our objectives are to detect CSF metabolome features associated with CBCL scores, and to determine the overlap between analyses using only annotated features and the full feature set. Methods: CSF metabolomic features were measured using liquid chromatography mass spectrometry (LC-MS). We assessed full feature sets separately from annotated metabolites with 80% detection across samples. We performed single-chemical analyses using false discovery rates for multiple comparisons. To accommodate the correlation among the components, we included a weighted quantile sum regression with a random subset ensemble step (WQSRS). Covariates included subject age, sex, socioeconomic status, z-scores of both weight and height, and retinopathy status. Results: In the unannotated feature set, we found 3 features significantly related to CBCL scores. In the annotated features set, 12 metabolites had significant linear associations with CBCL scores. In the mixtures WQSRS index, the unannotated features in the top 10% of weights included components that were not annotated, demonstrating that the annotated analysis missed key features. Conclusion: Our results indicate that considering only annotated metabolites misses features in the full features set in a study of children with CM in the association between a CSF metabolome and behavioral outcomes. The unknown features identified to be linked to CBCL scores and not included in the annotated set provide potentially important future areas of inquiry.