DISCOVERY AND VALIDATION OF MULTIMODAL BIOMARKER SIGNATURES RELATING TO ALZHEIMER'S DISEASE PATHOLOGY AND PROGRESSION: 2017 Abstract Supplement

Background Biomarkers of Alzheimer’s disease (AD) pathology and progression have now been identified across various modalities. The aims of the two studies presented (ANM-MMB and EMIF-AD biomarker discovery) are to discover and replicate previously identified biomarkers of disease pathology and progression, and moreover to determine whether a multimodal biomarker signature may add value in comparison to a biomarker of a single modality. Methods The ANM-MMB cohort is comprised of 718 AD, MCI converters and non-converters, and control subjects selected from the AddNeuroMed, Alzheimer’s research trust and Dementia case register cohorts. Cognitive measures, serum and urine metabolomics, structural MRI, genomics, whole blood transcriptomics and plasma proteomics data was available. The EMIF-AD biomarker discovery cohort consists of 1221 AD, MCI and control subjects, selected from the EMIF catalogue. All subjects had existing amyloid measures (CSF Aβ or amyloid-PET), structural MRI and clinical data, and furthermore plasma proteomics (targeted and untargeted), CSF proteomics (targeted), metabolomics, genomics and epigenetics data were generated. For both studies univariate and multivariate statistics were utilised to identify candidate biomarkers of AD pathology (neurodegeneration and/or brain amyloid burden), rates of cognitive decline, and MCI progression to dementia. pQTL-eQTL-mQTL analyses, network/pathway analysis, and multimodal classifiers were employed to detect multimodal signatures. Results Initial analyses indicate that in the ANM-MMB study a serum and urine derived 15 metabolite classifier predicts MCI progression to AD with 72% accuracy, and the biological significance of the metabolites included in the biomarker panel was identified. Further analyses will examine whether a multimodal classifier is able to predict with even greater accuracy. We will then seek to replicate this in the EMIF-AD biomarker discovery study. Further analyses will also examine single and multimodal biomarker classifiers of other endophenotypes. Conclusions These two studies could be used to identify novel and replicate previously identified single modality biomarker findings. Furthermore the impact of combining the additional modalities with these findings will be discussed. Computational and technical challenges encountered and the bioinformatics pipeline devised in the multimodal analysis of the ANM-MMB cohort will be used to inform the analysis pipeline of the EMIF-AD biomarker discovery study as a replication.