The hallmarks of Alzheimer’s disease (AD) are amyloid aggregation, tauopathy and neuronal injury. Along with the microglia response, these are considered as the “core” AD pathology (Scheltens et al., 2021). Indeed, cerebrospinal fluid (CSF) levels of Aβ1-42, total-tau and tau phosphorylated at threonine 181 (p-tau181) levels, mirroring cerebral amyloid, neuronal injury and tau pathology, are currently the best validated biofluid markers of AD pathology. However, these pathology aspects do not fully explain the vast heterogeneity of the observed disease phenotypes. A large body of evidence indicates that other biological pathways and molecular alterations happening at both cerebral and systemic levels are involved in the pathophysiological processes underlying AD. These processes may substantially contribute to the development and acceleration of amyloid pathology and neurodegeneration, precipitate the manifestation of cognitive and neuropsychiatric symptoms (NPS) and may therefore represent targets for both interfering with the developing AD pathology and slowing down or even reverse clinical disease progression. For example, neuroinflammation was proposed to play a key role at the interface between both tau and amyloid pathologies (Calsolaro and Edison, 2016; Luca et al., 2018; Marttinen et al., 2018). Alterations of the lipid metabolism including oxysterols (Gamba et al., 2019; Jahn et al., 2021), cholesterol and non-cholesterol sterols (Popp et al., 2012, 2013), eicosanoids (Biringer, 2019) and other lipid groups (Kao et al., 2020), have been associated with AD pathology (Barbash et al., 2017). Other metabolic pathways, such as one-carbon metabolism (Oikonomidi et al., 2016; Troesch et al., 2016; Dayon et al., 2017), glucose (Arnold et al., 2018) and amino acid metabolism (De Leeuw et al., 2017), amongst others (van der Velpen et al., 2019; Xu et al., 2021), are also implicated in AD. Vascular factors are also part of the AD pathology (Bowman et al., 2018; Nation et al., 2019). Yet it is difficult to paint a complete figure of the endophenotypes of AD as (i) there are large inter-individual variations, (ii) most previous studies have focused on single pathway alterations, (iii) biological systems are interdependent and interconnected and (iv) their relevance for the clinical manifestation and disease progression remains unclear. From a clinical point of view, a better understanding of the AD endophenotype is needed not only for a better diagnosis and prediction of clinical disease progression, but also with regard to targeted interventions. To achieve this, systemic biological approaches are required. These approaches should simultaneously consider multiple biological levels, their interactions and their relationships with known features of AD. This is especially relevant as so far, the traditional “one gene, one drug, one disease” hypothesis has mostly failed to produce clinically relevant drugs, likely because of the complex disease pathophysiology of AD (Fang et al., 2020). Drug development hypotheses should therefore take into account the great complexity of AD spanning multiple biological levels, and their clinical relevance.

Such in-depth analysis of a disease phenotype can hardly be performed using only hypothesis-driven targeted approaches that typically focus on one or a few groups of molecules of interest. Applying more comprehensive untargeted, data-driven assessments is better suited to address the extent and complexity of disease-related alterations. The recent development of various omics technologies allows, for instance, to resolve the complexity of the metabolome and proteome (Sancesario and Bernardini, 2018). It is now possible to measure a large number of molecules simultaneously with relatively high throughput resulting in sensitive read-outs of biological status and for exploration of molecular fingerprints of diseases (Jove et al., 2014; Shevchenko et al., 2015; Hasin et al., 2017). As powerful phenotyping technologies, omics significantly accelerate the understanding of mechanisms of pathophysiological alterations that underlie complex diseases such as AD.

However, given that biological systems are interdependent and interconnected, it is necessary to consider multiple single-omics datasets together. These approaches are dubbed multi-omics as they integrate data stemming from multiple biological levels, from genes to transcripts, to proteins and to metabolites. Such approaches afford the opportunity to explore in depth the pathway alterations and molecular mechanisms associated with the disease. In the context of AD, distinct endophenotypes and alterations could be identified and associated with distinct disease manifestations, helping us better understand disease heterogeneity and related clinical relevance. Indeed, it is conceivable that distinct endophenotypes underlie cognitive and non-cognitive clinical manifestations, for example. These approaches also have the potential of identifying altered biofluid molecule profiles that could be used as biomarkers for the diagnosis and prognosis in preclinical or early clinical AD stages. Such biomarkers signatures could also be used for monitoring and following up the effects of clinical interventions on relevant pathologies over time.

We conducted a review of the existing literature on multi-omic approaches in Alzheimer’s disease, with a focus on hypothesis-free untargeted approaches that integrated different omics modalities and data types originating from different molecular biology levels in order to capture the dynamic pathomechanisms underlying Alzheimer’s disease. We searched PubMed, MEDLINE and Google Scholar for original articles in English language referenced in these databases between 30th September 2017 and 30th September 2022 using the following search terms: “multi-omics,” “multi omics,” “omics,” “Alzheimer’s disease,” “Alzheimer’s,” “AD” and “neurodegenerative disease.” The titles and abstracts of all references were screened for eligibility. Full texts of potentially eligible references were then retrieved and assessed for inclusion.

To be considered, articles had to include systems biology approaches that considered together at least two different biological modalities at different molecular levels from the following: genomics, transcriptomics, proteomics, lipidomics, metabolomics or ionomics, obtained from human participants with confirmed AD pathology (through validated biomarkers of post-mortem neuropathology analysis). We excluded from our analysis review articles, commentaries or errata, unreviewed pre-prints, articles not using human data or those to which we did not have access.

Relevant information was extracted from the included articles, including omics modalities used, use of an integrative method, as defined according to Ritchie et al. (2015), i.e. the process by which different types of omics data are combined as predictor variables to allow more thorough and comprehensive modelling of complex traits or phenotypes, and their main findings.

After exclusion of duplicates, we obtained 116 different studies. We excluded 32 review articles, 29 studies that were not multi-omic studies, 13 that did not consider AD patients, 11 that were not performed using human samples or were only in silico simulations, 3 unreviewed preprints and 3 to which we did not have access. We therefore considered 33 studies for this review shown in Table 1. Before discussing them, we will address the tools and challenges involved in performing multi-omics studies.

While multi-omics have been used for over a decade to investigate alterations in other diseases, in particular in the field of oncology (Akhoundova and Rubin, 2022), multi-omic investigations of AD are still in their infancy. Nonetheless, in the past 5 years, several multi-omics studies have been performed on human cohorts and data in the context of AD (Table 1). However, few of these studies are truly integrative. While they consider multiple biological levels, they only investigated these alterations independently of each other. For example, Cohn et al. (2021) analysed proteomics, transcriptomics and lipidomics data derived from microglial extracellular vesicles obtained from cryopreserved brain tissue with neuropathologically confirmed AD. They showed that alteration of microglial function in the late stages of AD leads to the release of neuronal debris, including tau in extracellular vesicles. This suggests that extracellular vesicles could be used as not only a source of future omics data but also as a potential readout of disease state. More recently, also using brain tissue from AD patients, Chung et al. (2022) explored genomics and transcriptomics data and found an association of AD with genetic variants in MGMT (O6-methylguanine-DNA methyltransferase, which is involved in DNA damage repair function) amongst women lacking the APOE ε4 allele, the most important genetic risk factor in sporadic (non-familial) AD. Furthermore, the epigenetic regulation of MGMT was associated with tau-related pathology. The APOE ε4 allele has also been associated with proteins within the extracellular matrix and glycosaminoglycan-binding proteins in a study considering genomics, proteomics and transcriptomics data obtained from brain tissue derived from the general population (Johnson et al., 2022). Also, considering genomics and transcriptomics obtained from AD confirmed brain tissue, Min et al. (2021) have shown that single-nucleotide variants of AD-candidate genes are not a causal factor of sporadic AD. Non-integrative multi-omics analyses were also applied to investigate alterations in blood-based miRNA levels in relation to genomics data, Shigemizu et al. (2020) revealing that single-nucleotide polymorphism-microRNA pairs can be used to improve the prediction of conversion from mild cognitive impairment (MCI) to AD dementia.

While all these studies provide important and confirmatory clues for the better understanding of the pathophysiological mechanisms of AD and reveal potential new risk-factor and diagnosis tools, they do not fully leverage the power of multi-omics data integration. Firstly, they do not use integration algorithms, as defined according to Ritchie et al. (2015), that consider the interactions that most likely exist between the different biological levels considered. Indeed, no analysis of the combined data was performed but rather each individual data modalities were investigated separately without integrating the different results were integrated into a single framework. Another difficulty is that the most studies only consider two or three biological levels, such as. genomics and transcriptomics, neglecting the downstream events happening at the proteome and metabolome levels. The next most considered modality is proteomics, but very few studies consider lipidomics, metabolomics or ionomics (a term defining the entire elemental composition of a living organism) data, all of which have previously been associated with AD (Zheng et al., 2016; De Leeuw et al., 2017; Gamba et al., 2019; van der Velpen et al., 2019). Of note, it is also possible to consider in multi-omics studies non-molecular data such as neuroimaging and clinical data. For example, investigating both genomics and proteomics of brain tissues derived from MCI and AD patients, Xie et al. (2021) found 20 SNPs from 20 genes associated with the function and structure of six frontal cortex regions. These SNPs were also associated with increased amyloid deposition in these areas. Integration of such clinical and functional data greatly enhances the real-world application chances of the generated results.

Some recent studies have however pursued a true integrative approach. Importantly, Odom et al. (2021) have used such an approach to analyse genomics and transcriptomics derived from brain tissue from AD patients. They have shown that most pathways involved in AD pathology are not picked up by single-omics approaches. Indeed, combining multiple weak signals from a number of individual molecules is more likely to identify key biological hubs or pathway alterations relevant for disease. For example, Wang M. et al. (2021) used post-mortem brain tissue samples with confirmed AD to identify ATP6V1A (encoding an ATPase component) as a key network regulator involved in multiple protein networks associated with AD. This suggests that this gene is not only a potential risk factor for AD but could also serve as a target for disease modifications, highlighting the power of integrative approaches for discovering potential disease modifying drug targets.

The common findings of all the studies presented in Table 1 could help build a clearer and more accurate picture of the different processes underlying AD pathophysiology. These muti-omics approaches have provided evidence for alteration of molecular and cellular pathways underlying AD pathophysiology such as the immune system response (including complement activation), haemostasis pathways, energy metabolism and mitochondrial function, lipid homeostasis and processing, extracellular matrix regulation, synapse biology and plasticity, neuronal function, proteostasis, oxidative stress, signal transduction pathways and programmed cell death.

There is also a potential for multi-omics approaches to classify AD subtypes by their biological endophenotype in the first place. Such biologically defined subtypes would be devoid of any clinical bias while still allowing for considering the clinical manifestation and disease severity, and progression over time. Defining disease subtypes biologically would also be very helpful for grouping participants in future clinical trials specifically targeting the altered biological pathways. Integration of multi-omics data obtained from cerebral tissue of cerebrospinal fluid will allow to explore and unravel the disease pathomechanisms and could help identify and investigate drug-targets and biological pathways alterations. On the other hand, blood (or systemic)-derived multi-omics data are better suited to identify biomarkers for individual diagnosis, prognosis and monitoring. For example, using readily accessible blood-derived genomics, transcriptomics and proteomics data from patients with cerebral amyloid pathology, Park et al. (2022) have identified possible molecular drivers of AD heterogeneity or AD subtypes. This suggests that specific clinical manifestations of AD could be associated with distinct biological alterations. More generally, the molecular signature of disease could differ from one patient to another, according to cognitive status, age or sex for example (Xu et al., 2021). Therefore, identification of specific molecular signatures is key for (i) better understanding the mechanisms underlying the heterogeneity of AD, (ii) developing specific diagnosis tests based on multi-omics derived biomarkers and (iii) for developing patient-tailored interventions that take into account individual biological specificities. Taken together, such approaches are likely to increase the efficacy of clinical care strategies and interventions.

While substantial advance in the understanding of AD have already been made using multi-omics approaches some important limitations remain. First, the findings need to be related to the “core” aspects of AD pathology to better understand their contribution to disease. Indeed, they may be associated with amyloid pathology, tau-related or neurodegeneration, but they could also occur as independent pathophysiological processes. We therefore propose the research framework shown in Figure 1 for multi-omics investigations of AD going forwards. In this paradigm, both untargeted and hypothesis-driven targeted, single omics and multi-omics approaches are performed in parallel and combined to allow for bench-side discovery and for clinical translation (Wood, 2014). Molecules associated with AD should be identified in both single- and multi-omics datasets. The identified molecules should then be subsequently assigned to biochemical pathways using pathway and metabolic network-embedded analysis as well as ontology approaches. Finally, the identified analytes/molecules and pathway alterations should be related to specific aspects of AD pathology at both clinical and molecular levels. Of note, this framework could easily be applied to other neurodegenerative diseases or other pathologies. Moving towards a more unbiased and heterogeneous framework encompassing genes, proteins and molecules contributing to synaptic function, neuroinflammation, transcriptional and translational regulation as well as brain structure and function will be key in unlocking in-depth understanding of AD. It will also allow for the discovery of as-of-yet unimagined biomarkers at various stages of the disease. Such approaches require not only large cohorts but also great amounts of computing power as well as substantial financial resources and therefore can only be undertaken as collaborative multi-centric endeavours (Ritchie et al., 2016).

In addition to performing single- and multi-omics together, the next promising approach could be the meta-analysis of different cohorts as multi-omics data become more and more available in public repositories. While this would drastically strengthen understanding of AD pathology this approach comes with its own challenges. Indeed, there is a vast amount of heterogeneity between the cohorts used and one would also have to deal with data completeness and labelling issues amongst others.

Despite their relatively small numbers, multi-omics studies in AD have already shed light on the complex evolution of AD pathophysiology, including at individual level. This is especially relevant for a complex disease such as AD where clinical manifestations arise through interactions of multiple biological pathway that result in multifactorial alterations. While many of the identified alterations have been previously investigated in single-omic or hypothesis-driven studies, multi-omics studies have helped heighten our understanding of AD pathophysiology and clinical manifestations and opened perspectives of diagnosis and treatment options beyond the core pathophysiological mechanisms of AD. This will help identify new targets for prevention or treatment for patients at risk of or affected by AD symptoms. In addition, specific alterations detected in peripheral blood plasma or serum could serve as non-invasive and/or more effective biomarker signatures of AD. These have the potential to diagnose AD and its subtypes at early preclinical stages, to predict clinical progression and to monitor the development of relevant aspects of the AD pathology. Indeed, identification of certain metabolic profiles could help clinicians determine whether the patient will experience rapid or rather slow cognitive decline and develop neuropsychiatric symptoms, and therefore help propose better targeted and more efficient clinical care. Importantly, all identified biomarkers or drug targets will need to undergo absolute quantification and validation in independent studies and “real life” cohorts before they can be used in clinical practice.

CC and JP wrote the manuscript and advised on content. MR and LD provided guidance on content and editing. All authors contributed to the article and approved the submitted version.

This work was supported by grants from the Swiss National Research Foundation (SNF 320030_204886) and Synapsis Foundation—Dementia Research Switzerland (grant number 2017-PI01).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.