An Integrative Multi-omics Approach Reveals New Central Nervous System Pathway Alterations in Alzheimer’s Disease

Abstract

Background: Multiple pathophysiological processes have been described in Alzheimer’s disease (AD). Their inter-individual variations, complex interrelations, and relevance for clinical manifestation and disease progression remain poorly understood, however. We tested the hypothesis that cerebrospinal fluid (CSF) integrative multi-omics analysis highlights novel interacting pathway alterations in AD.Methods: We performed multi-level CSF omics in a well-characterized cohort of older adults including subjects with normal cognition, mild cognitive impairment, and mild dementia. Proteomics, metabolomics, lipidomics, one-carbon metabolism, and neuroinflammation related molecules were analysed applying Elastic-net regression and Multi-Omics Factor Analysis followed by pathway enrichment. Multivariate analysis was used to select best predictive models of AD pathology and cognitive decline.Results: Multi-omics integration identified five major dimensions of heterogenicity explaining the variance within the cohort and differentially associated with AD . Further analysis exposed multiple interactions between single ‘omics modalities and distinct multi-omics molecular signatures differentially related to amyloid pathology, neuronal injury, and tau hyperphosphorylation. Enrichment pathway analysis revealed overrepresentation of the hemostasis, immune response and extracellular matrix signalling pathways in association with AD. Further, combinations of four selected molecules significantly improved prediction of both AD (protein 14-3-3 zeta/delta, clusterin, interleukin-15, and transgelin-2) and cognitive decline (protein 14-3-3 zeta/delta, clusterin, cholesteryl ester 27:1 16:0 and monocyte chemoattractant protein-1). Conclusions: Applying an integrative multi-omics approach we confirmed previously reported associations with AD pathology and report new molecular and pathways alterations. These findings are relevant for the development of personalized diagnosis and treatment approaches in AD.

Predicting Alzheimer’s Disease with Multi-Omic Data: A Systematic Review

“…Some, such as Yu et al [17], Shigemizu et al [20], Gupta et al [21], and Binder et al [22], aimed to identify biosignatures, a specific combination of biomarkers, which together would predict biomarkers. Others were more general in their biomarker identification, highlighting hundreds of biomarkers which are associated with AD, such as in Maddalenda et al [23], Song et al [24][23], Clark et al [25], Darst et al[26] [33], Khullar and Wang [27], and Corce et al [28]. A third category emerged which did not aim to identify specific biomarkers, but instead focused on creating a model which would predict AD based on a continuous feature representation in a machine learning model, such as Abbas et al [29] and Venugopalan et al [30].…”

“…However, age and cognitive tests are only considered important after patients have already become symptomatic. APOE 4 status, which can be measured at any time, is also often cited a most important risk factor and together with age and clinical test data can predict AD with 88% AUC [25]. Additional meta-analysis on this topic should be conducted in order to categorize which new biomarkers are the most effective [18].…”

Multi‐Omics‐Based Autophagy‐Related Untypical Subtypes in Patients with Cerebral Amyloid Pathology

“…Neff et al [ 8 ] only used RNA sequencing data from two different public cohorts (ROSMAP and MSBB), whereas we used four different multi‐layers (targeted‐sequencing, miRNA transcriptome, proteomics, blood‐based biomarkers) that were generated autonomously. Furthermore, Clark et al [ 19 ] utilized MOFA software in R to combine their multi‐modal datasets in AD but were restricted to the MOFA software, whereas i) we adopted systems‐biological analyses with HENA AD network, KDA, and MCA methods to narrow‐down the key‐drivers, ii) performed a longitudinal analysis to elucidate the informative clusters, and iii) utilized different types of biological samples to validate our key‐drivers such as PBMCs, iPSC‐derived cerebral organoids, and human post‐mortem brain samples. Furthermore, our systems‐biological approaches identified that there are significant associations between the top‐rated targets and the enriched pathways or meanings of each cluster in both models.…”

Multiomics Approach to Understanding Olaparib Resistance and Predicting Drug Response

“…To find features that distinguish responders from nonresponders using minimal information, we selected the features with the highest weights from MOFA. We tested weight cutoffs from 0.1 to 0.8 and used features with weights greater than 0.8 for further analysis (27).…”

“…We used 3,296, 2,361, 636, and 289 features for gene expression, level of DNA methylation, deleterious mutation, and protein activity, respectively (Figure 2b). Each set of features was then matrix-factorized into nine factors (27). Among the four domains, deleterious mutations had the highest variance for the most dominant factor, while gene expression had the lowest variance.…”