Omics-based biomarkers discovery for Alzheimer's disease

Abstract

Alzheimer’s disease (AD) is the most common neurodegenerative disorders presenting with the pathological hallmarks of amyloid plaques and tau tangles. Over the past few years, great efforts have been made to explore reliable biomarkers of AD. High-throughput omics are a technology driven by multiple levels of unbiased data to detect the complex etiology of AD, and it provides us with new opportunities to better understand the pathophysiology of AD and thereby identify potential biomarkers. Through revealing the interaction networks between different molecular levels, the ultimate goal of multi-omics is to improve the diagnosis and treatment of AD. In this review, based on the current AD pathology and the current status of AD diagnostic biomarkers, we summarize how genomics, transcriptomics, proteomics and metabolomics are all conducing to the discovery of reliable AD biomarkers that could be developed and used in clinical AD management.

Introduction

It is estimated that 50 million people worldwide currently have dementia, 50–70% of whom are AD, and the incidence of AD is expected to triple by 2050. [1]. Although studies on amyloid beta (Aβ) plaques and tau neurofibrillary tangles (NFTs) have provided an understanding of the molecular mechanisms, the pathogenesis of AD is still not fully understood [2]. In addition to the classic pathological features, many emerging new causes/risk factors of AD are discovered. These risks include, but not limited to neurovascular abnormalities, neuroinflammation and immune response disorders, impaired energy metabolism, mitochondrial dysfunction, oxidative stress, compromised autophagy/mitophagy, etc. [3–6]. A comprehensive understanding of the molecular mechanisms under the pathophysiology of AD will help to find out reliable biomarkers to diagnose and track the progression of the disease and provide support for the prevention, treatment and prognosis of the disease [7].

Omics sciences is a holistic, systems-level means to fully expose AD pathophysiology, including genomics, transcriptomics, proteomics and metabolomics [8] The technological innovations of various omics have become the most promising tool to investigate AD (Fig. 1). Genomics main purpose is to collectively characterize and quantify all genes and mutations in an organism, identifying new loci that affect the risk of AD is crucial for us to understand the underlying causes of AD [9]. Transcriptomics is a powerful way to study the potential gene regulation mechanisms of complex trait associations to identify groups of co-expressed genes representing cellular processes and transcriptional programs associated with AD phenotypes [10]. Proteomics studies protein expression levels, posttranslational modifications (PTM), protein–protein interactions, etc., so as to identify the proteins and biological processes that play roles in AD [11]. Metabolomics studies the collection of all metabolites in a cell at a certain moment, reflecting the environment in which the cell is located. It reveals ongoing pathological processes of AD [12]. Thus, omics research can describe the entire biological continuum of AD and identify networks during AD progression.

Fig. 1.

Multi-omics approaches in AD. We list the methods currently available in each omics. GWAS Genome-wide association studies, WGS Whole-genome sequencing, WES Whole-exome sequencing, TS Targeted Sequencing, BS-sequencing Oxi- and Bisulfite sequencing, BS-array Bisulfite-modified DNA-based arrays, MSRE-PCR Methylation-sensitive restriction enzyme-PCR, ESI Soft ionization techniques, MS–MS Tandem-Mass Spectrometry, LC–MS Liquid chromatography-mass spectrometry, MALDI Matrix-assisted laser desorption/ionization, IMS MALDI-imaging mass spectrometry, MALDI-TOF–MS Matrix-Assisted Laser Desorption Ionization Time-of-Flight, iTRAQ Isobaric Tag for Relative and Absolute Quantification, 2-DE Two-dimensional gel electrophoresis, 2D DIGE Differential Gel Electrophoresis, DML Dimethyl labels, MRM Multiple-reaction monitoring, SRM Selected reaction monitoring, QQQ Triple quadrupole mass spectrometer

In this review, we outline the multi-omics interrelated with AD in order to identify its potential biomarkers, and these new biomarkers may potentially facilitate the headway of early diagnosis and therapeutic interventions for the disease.

Available biomarkers for AD diagnosis

The current biomarkers used to diagnose AD in clinical practice are cerebrospinal fluid (CSF) protein markers and brain neuroimaging markers. The core CSF biomarker of AD is Aβ42, which is used to show cortical amyloid deposition. Besides, total tau (t-tau) represents the intensity of neurodegeneration, and phosphorylated tau (p-tau) is related to pathological changes of neurofibrillars. Potential imaging biomarkers that have been approved can be used to observe structural and functional changes, as well as synapses and cell degeneration through positron emission tomography (PET) or functional magnetic resonance imaging (fMRI) [13]. F-fluorodeoxyglucose (FDG) PET can measure the glucose uptake of neurons and glial cells and is sensitive to synaptic function. Amyloid PET has a very high preciseness for cortical amyloidosis. Studies on plasma Aβ, tau and neurofilament light chain (NfL) also have some good results, but none of them has been clinically verified yet [14]. In addition, tau-PET has only been used for a few years. Although it is being actively included in many clinical studies, longitudinal data on accumulated changes in tau are still limited [15].

The combination of proven multiple markers has good predictive value for distinguishing between normal aging in patients with mild cognitive impairment (MCI) and AD [16]. Currently, the diagnosis of AD is determined by experienced clinicians using a series of cognitive tests, coupled with various structural and functional imaging and CSF biomarkers, although these markers for early diagnosis of AD has high sensitivity and specificity, but more biomarkers need to be determined in order to make a diagnosis at an earlier stage of the disease, so that it can monitor disease modification therapies and disease-related processes, and provide novel perceptions in disease mechanisms.

The features of genomics in AD

Genome-wide association study (GWAS) usually examines a large number of single nucleotide polymorphisms (SNPs) in the entire genome to identify disease-related genetic variation, so that thousands of genetic variations (including non-coding regions) can be evaluated simultaneously without the need to make prior assumptions about biological pathways. Next-generation genome sequencing (NGS) technologies, such as whole-exome sequencing (WES) and whole-genome sequencing (WGS), can identify rare SNPs associated with AD, and these genes usually cannot be detected by GWAS [17]. Rare variants are more likely to increase the risk of AD than common variants.

Late-onset Alzheimer's disease

AD that develops after the age of 65 is often referred to as late onset AD (LOAD). LOAD is considered to be a polygenic disease and is sporadic. The APOEɛ4 allele is considered to be the strongest genetic risk factor for LOAD, and its genetic variation greatly increases the risk of AD [18–20].To date, more than 50 common low-risk genetic loci in LOAD have been revealed [21–27] (Fig. 2, Table 1), and their individual contribution to the total heritability of AD is relatively small. Jansen et al. and Kunkle et al. identified new risk loci for LOAD using a "proxy phenotype" approach, which the use of genotypes of unaffected individuals and their affected relatives to identify disease risk loci has the advantage of increasing sample size, although it is less sensitive and less specific for the diagnosis of AD [22, 28].

Fig. 2.

AD-related genes and their SNPs. The outer side of the circle is the genetic factors associated with AD in alphabetical order, and the inner side is the SNP sites of each gene associated with AD. Risk factors are marked in red, protective factors are marked in green, and blue stands for both. The circlize package of the R software (http://www.r-project.org/) was used to generate the diagram

Table 1.

Summary of genomic loci and genes associated with AD

Locus	Chr	Genes	Potential pathogenesis	Cell type
ABCA7	19	ABCA7	Aβ degradation/clearance, lipid metabolism and phagocytosis, immune response and inflammation	Ubiquitous
ABI3	17	ABI3	Cell growth	Microglia
AC074212.3	19	AC074212.3	–	–
AC099552.4	7	AC099552.4	–	–
ACE	17	ACE, CYB561, TANC2	Aβ degradation/clearance	Neurons
ADAMTS4	1	ADAMTS4	APP processing to Aβ, tau pathology	Microglia
ADAMTS1	21	ADAMTS1	Synapse regulation	Microglia
ADAM10	15	ADAM10	APP processing to Aβ	Ubiquitous
AKAP9	7	AKAP9	Kinase signaling	Ubiquitous
ALPK2	18	ALPK2	–	Microglia
APH1B	15	APH1B	APP processing to Aβ	Ubiquitous
APOE	19	APOE	Aβ degradation/clearance, lipid metabolism and phagocytosis	Ubiquitous
B1N1	2	B1N1	Endocytosis, tau pathology	Ubiquitous
BHMG1	19	BHMG1, FBX046	–	–
BZRAP1-AS1	17	BZRAP1-AS1, RNF43, SUPT4H1, TSPOAP1	–	–
CASS4	20	CASS4, AURKA, CSTF1, FAM209A, FAM209B, GCNT7, RTFDC1	Cell adhesion/actin cytoskeleton, tau pathology	Microglia
CD2AP	6	CD2AP, ADGRF2, ADGRF4, GPR111, GPR115, OPN5, TNFRSF21	APP processing to Aβ, Aβ degradation/clearance, cell adhesion/actin cytoskeleton, endocytosis, immune response, tau pathology	Ubiquitous
CD33	19	CD33, SIGLEC7, SIGLEC9, SIGLECL1	Aβ degradation/clearance, immune response and phagocytosis	Microglia
CELF1/SPI1	11	CELF1, ACP2, AGBL2, C1QTNF4, DDB2, FAM180B, FNBP4, KBTBD4, MADD, MTCH2, MYBPC3, NDUFS3, NR1H3, NUP160, PACSIN3, PSMC3, PTPMT1, RAPSN, SLC39A13, SPI1, LOC101928943	Cell adhesion/actin cytoskeleton, myeloid lineage determination, tau pathology	Myeloid cells
CLNK	4	CLNK	–	–
CLU	8	CLU, EPHX2, SCARA3	Aβ degradation/clearance, lipid metabolism, immune response and inflammation	Astrocytes
CNTNAP2	7	CNTNAP2	Cell adhesion/actin cytoskeleton	Neurons
CR1	1	CR1	Aβ degradation/clearance, immune response and phagocytosis	Microglia
DSG2	18	DSG2, DSG2-AS1, DSG3	–	–
ECHDC3	10	ECHDC3, USP6NL, PROSER2	–	Microglia
EPHA1	7	EPHA1, CASP2, CLCN1, EPHA-AS1, FAM131B, LOC100507507, TAS2R41, TAS2R60, TAS2R62P, ZYX	Synapse regulation, cell adhesion/actin cytoskeleton, immune response and inflammation	–
FERMT2	14	FERMT2, ERO1A, PSMC6, STYX, GNPNAT1	Cell adhesion/actin cytoskeleton, tau pathology	Microglia
FRMD4A	10	FRMD4A	Tau pathology	Ubiquitous
HESX1	3	HESX1, IL17RD, APPL1	Synapse regulation	Microglia
HLA-DRB1/HLA-DRB5	6	HLA-DRB1, HLA-DRB5, C6orf10, BTNL2, HLA-DRA, HLA- DRB6, HLA- DQA1, HLA-DQB1, HLA- DQA2, HLA-DQB2, HLA- DOB	Immune response and inflammation	Microglia
HS3ST1	4	HS3ST1	–	Ubiquitous
INPP5D	2	INPP5D, NEU2, NGEF	Immune response and inflammation	Microglia
IQCK	16	IQCK, KNOP1, C16orf62	Endocytosis and sorting	Microglia
KAT8	16	KAT8	–	Ubiquitous
MAPT	17	MAPT	Synapse regulation, Aβ toxicity	Ubiquitous
MEF2C	5	MEF2C, TMEM161B, MIR9-2, LINC00461, MEF2C-AS1	Immune response and inflammation, tau pathology	–
MS4A	11	MS4A1, MS4A2, MS4A3, MS4A4, MS4A4E, MS4A5, MS4A6A, MS4A6E, MS4A7, MS4A14, OOSP2	Chemosensory receptors, immune response and inflammation	Microglia
NME8	7	NME8, EPDR1, GPR141, SFRP4	Cell adhesion/actin cytoskeleton	–
PFDN1/HBEGF	5	PFDN1, HBEGF	Immune response, cell adhesion/actin cytoskeleton	Ubiquitous
PICALM	11	PICALM, CCDC83, EED	APP processing to Aβ, Aβ degradation/clearance, endocytosis, tau pathology	Ubiquitous
PLCG2	16	PLCG2	Phospholipase signaling, tau pathology	Microglia
PLD3	19	PLD3	–	Ubiquitous
PTK2B	8	PTK2B, CHRNA2, STMN4, TRIM35	Synapse regulation, tau pathology	Neurons
RBFOX1	16	RBFOX1	Aβ degradation	–
SCIMP	17	SCIMP, C1QBP, DHX33, NUP88, RABEP1, RPAIN, USP6, ZFP3, ZNF232, ZNF594	Immune response	Microglia
SLC24A4/RIN3	14	SLC24A4, RIN3, LGMN	Endocytosis and sorting	Endothelial cell
SORL1	11	SORL1	APP processing to Aβ, endocytosis and sorting, lipid metabolism	Ubiquitous
SPPL2A	15	SPPL2A, USP8, USP50, TRPM7	Immune response	Ubiquitous
TM2D3	15	TM2D3	Aβ degradation/clearance	Ubiquitous
TREM2	6	TREM2	Aβ degradation/clearance, immune response and inflammation, tau pathology	Microglia
TREML2	6	TREML2	Immune response	Microglia
TRIP4	15	TRIP4, CSNK1G1, KIAA0101, ZNF609	–	–
TSPOAP1	17	TSPOAP1	Lipid metabolism	–
UNC5C	4	UNC5C	Response to neurotoxic stimuli, cell death	Neurons
WWOX/MAF	16	WWOX, MAF	Immune response, tau pathology	Ubiquitous
ZCWPW1/NYAP1	7	ZCWPW1, ACTL6B, AGFG2, AP4M1, AZGP1, C7orf43, C7orf61, CNPY4, COPS6, CYP3A43, FBXO24, GAL3ST4, GATS, GASTOR3, GIGYF1, GJC3, GNB2, GPC2, LAMTPR4, LOC100128334, LRCH4, MBLAC1, MCM7, MEPCE, MOSPD3, NYAP, OR2AE1, PCOLCE, PCOLCE-AS1, PILRA, PILRB, PPP1R35, PVRIG, SAP25, SPDYE3, STAG3, TAF6, TFR2, TR1M4, TSC22D4, ZKSCAN1, ZNF3, ZSCAN21	Immune response and inflammation, tau pathology	Microglia
ZNF655	7	ZNF655	–	Ubiquitous

We summarized the genomic loci in chromosome order, genes contained in gene locus, potential pathogenesis to which these genes have been functionally linked and cell type-specific expression of these genes [24]. Copyright^© 2019, Springer Nature)

“–”The information has not been obtained from the primary publication or has not been found yet

Rare variants are more likely to increase the risk of LOAD than common variants [29] and can affect different pathological processes [30–33]. The first genome-wide copy number variations (CNVs) study on LOAD identified 3,012 AD-specific CNVs, suggesting that accumulation of CNVs may lead to AD-related pathological changes [34].

The KnockoffScreen-AL approach, recently found to be validated in the AD cohort, is particularly effective in identifying rare variants [35]. NGS approach will identify additional low frequency and rare variants associated with AD, which is important for identifying key pathways in disease pathogenesis and discovering potential targets for drug therapy.

Currently, most GWAS are conducted in European populations, in the largest gene-based cross-ethnic meta-analysis in LOAD to date, the final analysis results are different from the published results [36, 37]. Although the main pathways of AD etiology in other ethnic groups are similar to those in European populations, many disease-associated loci are different [38–41]. The reason for these inconsistencies may be the result of cross-ethnic transfer of risk genotypes described by different populations. However, it has also been argued that the possibility of combining individuals from different populations to obtain a larger sample cannot be considered completely certain [42].

Early-onset Alzheimer's disease

Patients with early onset of AD, usually before age 65, are referred to as early-onset AD (EOAD). EOAD is an almost entirely genetic disorder with heritability ranging from 92 to 100%. Through genetic studies on single genes, high-risk mutations were found in three genes encoding APP, Presenilin 1 and Presenilin 2 (PSEN1 and PSEN2) [43], which mainly regulate Aβ production and degradation [9].

However, most pathogenic variants have not been evaluated by in vitro functional assays. In the case of insufficient genetic evidence, the definite pathogenicity of a variant may therefore still be uncertain. A PSEN1 mutation carrier who did not develop MCI until age 70 was found to have two rare Christchurch (APOEch) mutations in APOE3 after WES, considering that the APOEch homozygous gene was her most likely genetic modifier, which has a strong protective effect [44]. Therefore, can it be considered that the safe and effective editing of APOE may have a profound impact on the treatment and prevention of AD?

Studies on the role of rare variants in EOAD are scarce. In a WGS, FERMT2 was found to regulate AD risk by regulating APP metabolism and Aβ peptide production [45]. A novel missense variant in ACAA1 impairs lysosomal function and promotes Aβ pathology and cognitive decline [46]. It can be seen that the genetic determinants of EOAD involve rare mutations in Aβ network genes. The main advantage of these NGS technologies is that they can identify putative disease genes by directly analyzing the data of related family members belonging to small families.

Genomics related to pathological processes

Aβ is produced by sequentially dividing APPs by β- and γ-secretase enzymes. PSEN1 and PSEN2 form the active core of the γ-secretase complex, which affects the activity of endopeptidases or carboxypeptidases, allowing the production of Aβ40 and Aβ42 to be transferred to longer, more neurotoxic species [9]. Whole-brain amyloid deposition using Pittsburgh compound-B (PiB)-PET imaging confirmed the interrelationship between the APOE locus and brain amyloidosis in vivo [47].

Mutations of tau have been majorly associated with/contribute to other tau proteinopathies; however, there is no evidence that the moment that mutation of tau happens in AD [48]. Thus, we will be focusing on the mutations of other genes that could activate/inhibit tau pathologies in AD. Lots of known autosomal-dominant mutations in MAPT highlight the importance of tau protein in AD, and the gene encodes for tau [49]. Identifying other pathogenic genes that perturb tau accumulation, such as those involved in tau clearance and protein quality control, has the latent to disclose previously unknown mechanisms of AD [49].

A subset of genes has been identified to be associated with classical AD pathology or other pathways (Table 1), and subsequent genetic and functional studies are needed to confirm the role of candidate genes in the relevant pathological processes. Many studies currently focus on one variant or some variants, neglecting the complex polygenic disease context. Further understanding of polygenic risk factors still requires additional efforts and validating the discovery of these genes and elaborating them into targetable mechanisms remains an important issue.

Alzheimer's disease epigenomics

Epigenomics does not alter nucleotide sequences, and DNA methylation and histone modifications are highly variable and play an important role in AD [50].

The large-scale epigenome-wide association study (EWAS) found that although the different brain areas cells are different in composition and have different sensitivities to AD pathology, AD has similar associations with DNA methylation across different brain regions [51]. Through the combination of GWAS and EWAS, not only the contribution of DNA methylation in the AD risk loci to the disease was studied, but also the SNPs related to changes in DNA methylation were identified, which is helpful to determine the functional results of proven risk genes, provides a complement to causal variation at risk loci for AD [52, 53]. A recent epigenomic study found that an epigenomic factor associated with a reduced proportion of activated microglia appears to attenuate the risk of APOEε4 for AD [54].

There are relatively few studies on histone modification. Based on previous studies that determined the correlation between SNPs and gene expression, DNA methylation and histone modification levels, it was found that SNPs have an effect on RNA expression. The effect was exclusively mediated by epigenetic features in 9% of these loci, indicating that genetic variation affects these three factors together [55, 56], as well as a close interaction between epigenetic changes and RNA.

Challenges and opportunities

There is an urgent need to collect and process a large amount of data to discover the regulatory mechanism of AD risk-associated variants, and the consequent number and complexity of new treatment strategies. Linking genetic variation to the intra-AD phenotype that characterizes disease progression and is associated with pathology can help prioritize AD risk genes [9] to simplify the follow-up research on treatment. However, it remains unclear how genetic factors are related to the specific progress of the entire AD neuron network pathology. The relationship between the different risk genes and the molecular mechanisms involved in different pathologies still needs further understanding, so our main bottleneck at present is still insufficient understanding of the mechanism of genetic variation.

Sample size in NGS is generally small, and its efficacy is very limited. Therefore, we still need a large number of data sets to find strong evidence for a single gene or variant identified in the entire WGS and WES. Moreover, analysis of postmortem brain tissue has limited exposure to brain tissue in the early stages of the disease. Therefore, mouse models carrying causal mutations in AD have many advantages, such as precise environmental control, early exposure to brain tissue and a clear high-risk genotype and being an important tool [57].

The features of transcriptomics in AD

The traditional methods of transcriptomics research are pooling cell microarrays and single-cell RNA sequencing (scRNA-seq). Transcriptome-wide association analysis (TWAS) is a powerful way to study the potential gene regulation mechanisms of complex trait associations to identify co-expressed genomes representing cellular processes and transcriptional programs associated with disease phenotypes [10].

AD-related genome-wide RNA profiles include protein-coding RNA (cRNA) and non-protein-coding RNA (ncRNA) transcripts. The proportion of conserved transcripts with intact splice sites is smaller than that of general non-coding genes [58], suggesting that non-coding genes associated with AD have faster functional adaptations and are likely to constitute the key to gene regulation.

The alterations of mRNA in AD

Transcriptional analysis of AD-related regions in the human brain reflects the relevant processes of AD pathology [59]. Co-expression analysis of the brain transcriptome effectively links RNA to molecular pathways, organelles and cell types that influence AD, and forms a network, revealing the transcriptional profile of cell types in AD and cell types that are most likely to express candidate AD genes [60]. Using integrative network methods to analyze different brain regions in different cohorts can determine different combinations of multiple dysregulated pathways in specific molecular subtypes of AD [61].

Studies of gene expression profiles in AD brain tissue have also identified transcriptional changes in AD-specific cellular processes. Studies of the transcriptomes of six major brain cell types have identified myelination, inflammation and regulation of neuronal survival as major perturbations [49].

An approach linking genotype and gene expression helps to provide a clearer picture of the pathology of the whole brain [62]. Recently, splicing disruptions have been found to be a common feature of AD. The proteins encoded by splice introns play a role in autophagy-lysosome-related pathways [63] and have also been identified to be associated with neuroinflammatory plaques, Aβ-load and NFTs [55], highlighting a potential role in AD pathology.

The alterations of long non-coding RNA in AD

Transcripts longer than 200 nucleotides are referred to as long non-coding RNAs (lncRNAs) and can influence the pathogenesis of AD through various regulatory mechanisms, including transcriptional, posttranscriptional and translational regulation (Fig. 3, Table 2).

Fig. 3.

Transcriptomic biomarkers associated with AD. The diagram shows (from outside to inside): (1) related pathogenesis; (2) potential transcriptomic biomarkers; and (3) connection of different pathogenesis related to the same protein. The circlize package of the R software (http://www.r-project.org/) was used to generate the diagram

Table 2.

RNA transcriptomic biomarkers from different origins

Source	RNA transcript	Refs.
PBMC	lnc-AL445989.1–2, TTC39C-AS1, LINC01420, lnc-CSTB-1, LOC401557	[64]
Blood	miR-29c, miR-125b, miR-155, miR-Let-7b, miR-206	[65, 66]
CSF	MALAT1, miR-27a-3p	[67, 68]
Brain tissue	BACE1-AS,51 A, 17 A, NDM29, BC200, GDNFOS, BDNF-AS, Sox2OT, EBF3-AS, NAT-Rad18, CDKN2B-AS, lncRNA n336934, lncRNA n341006, LRP1-AS, NEAT1, SORL1-AS, NEAT1, SNHG1, miR-15, miR-16, miR-29a/b-1 cluster, miR-34a, miR-101, miR-106b, miR-184, miR-188-3p, miR-195, miR-338-5p, miR-346, miR-485-5p	[69–71]
Blood/CSF	mir-567	[72]
Blood/Brain tissue	miR-9, miR-103, miR-107, miR-206	[73, 74]
Blood/CSF/Brain tissue	miR-132	[75]

AD-related lncRNAs in the AD brain may affect coding RNAs at the same site, and some lncRNAs have been identified to be co-expressed with known AD-related genes [76], and studies have also been conducted on its regulatory mechanisms in response to AD, with its corresponding transcripts mainly involved in Aβ metabolism, neuroinflammation, synaptic plasticity, neurotrophic depletion, mitochondrial dysfunction, AD stress response, and other processes [77, 78]. Among the few functionally identified lncRNAs, the lncRNA BACE1-AS is highly expressed in both brain and blood and contributes to the pathogenesis of AD by increasing the levels of Aβ and APP [74]. Recently, the spectrum of lncRNAs in peripheral blood mononuclear cells in AD has also been studied for the first time [64], allowing one to go beyond postmortem brain tissue studies and to open up findings in different blood cells.

The alterations of MicroRNA in AD

MicroRNAs (miRNAs) are non-coding RNAs containing 20–24 nucleotides, which mediates the stability of mRNA and regulates translation inhibition, leading to down-regulation of gene expression [79]. Dysregulation of the transcriptome expressed by miRNAs can be delivered via exosomes and act synergistically with proteins to cause changes in mRNA, providing an early signal to suggest impending neuropathy in the brain.

The expression of miRNAs in blood, CSF and brain has been studied (Fig. 3, Table 2). Inconsistent results have been observed for the same miRNA in different studies analyzing the same brain region, and inconsistent findings regarding the changes in miRNAs in brain, CSF and blood. MiR-567 was currently found to be differentially expressed in all three biological samples from MCI-AD patients [72], but no independently validated miRNA has been found to vary uniformly among the three.

Future directions of transcriptomics

Studies have shown that transcription disorders can be spotted much earlier than the appearance of pathological signs and can be determined as the first sign of brain phenotypic changes [80]. Studies from the whole brain may mask local transcriptome overlap due to the different degree and rate of neurodegeneration in different brain regions. Accurate and consistent pathways in the brain, blood and cerebrospinal fluid from different samples and different disease stages have also not been established. Applying spatial transcriptomics and in situ sequencing techniques to study the cellular stages of AD could complement each other and provide a more comprehensive understanding to unravel the molecular network of AD [81].

In addition, ncRNAs are more readily available and relatively stable. The confirmation of abnormally expressed miRNAs and lncRNAs in AD patients will provide effective support for the early detection, diagnosis and treatment of AD in the future. Future studies on circulating RNAs (circRNAs), PIWI-interacting RNAs (piRNAs) and natural antisense transcripts (NAT) will also help to identify molecular events in the development of AD.

The features of proteomics in AD

Proteomics is the analysis of proteins in different tissues and fluids of the body. It can identify thousands of proteins in complex biological samples and can be used to study the various pathways that may be affected in the body of early stage of AD patients (Fig. 4, Table 3).

Fig. 4.

Potential proteomic biomarkers associated with AD. The diagram shows (from outside to inside): (1) related pathogenesis; (2) potential proteomic biomarkers; and (3) connection of different pathogenesis related to the same protein. The circlize package of the R software (http://www.r-project.org/) was used to generate the diagram

Table 3.

Protein biomarkers from different origins

Source	Protein	Refs.
Blood	α-1-antitrypsin, Complement C3, ACE, AFM, BNP, CDH5, FCN2, fibrinogen-γ-chain, Homocysteine, LGALS3BP, NfL, POSTN, PP, α-2-macroglobulin	[82–85]
CSF	MMP-10, tau368, TREM2, VILIP-1, YKL-40	[78, 86–88]
Brain tissue	GLUT3, SCRN1	[89, 90]
Blood/CSF	α-antichymotrypsin, Isoprostanes (8,12-isoiPF2α-VI), BACE1, CASP8, JAM-B, NRGN, tPA, 8-hydroxy-deoxyguanine, Complement C1q, CRP, Interleukin-6-receptor complex, Aβ1-17	[83, 86, 87, 91–95]
Blood/Brain tissue	ApoE, SMOC1, VGF, ApoA-1	[85, 96, 97]
CSF/Brain tissue	GAP43	[92]
Blood/CSF/Brain tissue	IGFBP-2	[98]

ACE blood angiotensin I-converting enzyme, AFM Afamin, APOA1apolipoprotein A-1, ApoE apolipoprotein E, BACE1Beta-Secretase 1, BNP type B natriuretic peptide, CASP8Caspase 8, Apoptosis-related cysteine peptidase, CDH5Cadherin 5, CRPC‐reactive protein, FCN2Ficolin 2, GAP43 Growth associated protein-43, GLUT3 Glucose transporter 3, IGFBP-2 Insulin-like growth factor binding protein-2, JAM-B Junctional adhesion molecule B, LGALS3BP Galectin 3 binding protein, MMP-10 Matrix metalloproteinase-10, NfL neurofilament light chain, NRGN neurogranin, POSTN Periostin, PP pancreatic polypeptide, SCRN1 Secernin 1, SMOC1 Secreted modular calcium-binding protein 1, tPA tissue plasminogen activator, TREM2 Triggering receptor expressed on myeloid cells 2, VGF Neurosecretory protein VGF, VILIP-1 Visinin-Like protein 1, YKL-40 Chitinase-3-Like Protein 1

Blood-based proteomics

The destruction of the blood–brain barrier (BBB) allows small molecules associated with AD to penetrate into the surrounding area, which makes it possible to discover AD biomarkers in blood.

The identification of a set of blood biomarkers capable of identifying AD at an early stage is pursued, and both plasma and serum samples are currently considered to have sufficient quantitative proteomic properties [99]. Plasma proteins that predict the progression of AD or MCI to AD have been identified based on different phenotypes, such as the degree of hippocampal atrophy, baseline cortical thickness and cognitive abilities; non-amyloid/non-tau proteins markers that can predict brain amyloid deposition as well as other AD phenotypes, or that are associated with cognitive decline and dementia, have also been identified based on different identification and analysis methods [100–102].

Nevertheless, most experiments failed to reproduce the results with similar accuracy during the validation phase. Different stages of AD may be related to the temporal evolution of the pathophysiology of AD in the brain and cerebrospinal fluid, so that different events occur at different time points of disease progression. Recent efforts have also been made in this regard, identifying some proteins that can distinguish the different stages of AD [103].

CSF-based proteomics

It is CSF that is in direct contact with the extracellular space of the brain and has brain tissue-specific proteins whose concentrations can reflect physiological or pathological changes in the AD brain.

Studies have shown that CSF proteomic changes are more representative of tau pathology than amyloid pathology [88, 104]. CSF tau-368 is considered likely to detect pathological changes before the onset of symptoms in cognitively normal individuals [88], and that protein levels involved in neuronal plasticity and blood–brain and blood-CSF barrier dysfunction are associated with CSF t-tau [105]. CSF proteomics can also respond to changes in AD patients such as neuronal proliferation, innate immune response and BBB dysfunction, synaptic, vascular, myelin and metabolic pathway dysfunction [87, 106]; however, no single marker is reflective of the complete pathology and the selection of a set of CSF biomarkers to characterize may be crucial for diagnosis and prognosis. In particular, changes in CSF NPTX2 have been suggested as possible biomarker candidates for AD prognosis [107].

Brain tissue-based proteomics

Proteomics studies of the human brain can only be based on autopsy and are therefore primarily aimed at discovering the biomolecular mechanisms involved in AD, rather than at diagnosis.

Brain degeneration is progressive and dynamic, with very little possibility of continuous observation, and little is known about how pathological changes at different stages of the disease affect the overall expression of brain proteins. However, it has also been found that different disease stages have different AD-related specific proteins [96], and a comprehensive overview of the interaction groups of t-tau and p-tau in the brain at different stages has also been identified [108–110]. Functionally distinct human brain regions show different and region-specific alterations in protein expression, and analysis of the spatial proteome may help to identify specific proteins that represent early states of the disease [111].

A consensus AD brain protein co-expression molecular network was established to look at modules associated with the disease, define different protein pathways, and analyze representative proteins for each pathway, finding that the most closely related to AD is astrocyte/microglia metabolism, and that proteins involved in vesicular endocytosis and secretion pathways begin to change in the early stages of AD [96, 112–116], and in this process gradually led to the derivation of redox proteomics, phosphorylation proteomics, synaptic proteomics, mitochondrial proteomics and microglia proteomics, which together complement the entire proteome.

Limitations and future directions

Proteins are a powerful intermediary for evaluating novel biomarkers for diagnosis and treatment of AD. The proteomics research process is complex and the possible problems and possible solutions for each step have been described in detail [1]. A comprehensive analysis of the ultra-depth proteomes of cerebral cortex, cerebrospinal fluid and serum revealed that the majority of proteins co-present in the three tissues were mitochondrial proteins, but the number of differentially expressed proteins studied and the way they are related in most practical studies lack consistency. The combination of quantitative proteomics, differential expression analysis and co-expression network analysis provides a useful approach to understand the pathogenesis of AD at a systemic level [112]. The combination of nanotechnology and existing analytical tools can enrich the identification of biomarkers of AD [117].

There are currently some studies on urine, saliva, buccal mucosal cells and other samples that also have great potential in the future to provide utterly non-invasive alternative to present CSF and blood sampling procedures. More importantly, the lower molecular parts of blood or CSF proteomics including peptides and protein fragments should also receive more attention and may have more disease-specific information than the full-length proteins themselves.

The features of metabolomics in AD

Metabolomics is the mensuration of small molecules called metabolites in biological samples, such as blood. Understanding how the metabolic changes in various pathways in AD will help us develop effective disease remission therapies (Table 4).

Table 4.

Metabolites associated with AD

Source	Metabolites	Potential pathogenesis	Refs.
Serum	Lactic acid, α-ketoglutarate, isocitric acid, glucose, oleic acid, adenosine and cholesterol urea, valine, aspartic acid, pyroglutamate, glutamine, phenylalanine, asparagine, ornithine, pipecolic acid, histidine, tyrosine, palmitic and uric acid, tryptophan, stearic acid and cysteine	Energy deficiencies, oxidative stress, dysfunction in mitochondrial activities, hyperammonemia	[118]
Serum	Sphingolipids and glycerophospholipids	Tau phosphorylation, Aβ metabolism, calcium homeostasis, acetylcholine biosynthesis, apoptosis	[119]
Serum	C12, C14:1, C16:1, C18, PC aeC36:2, PC ae C40:3, PC ae C42:4, PC ae C44:4, SM(OH)C14:1, SM C16:0, SM C20:2, α-aminoadipic acid, valine	Tau phosphorylation, Aβ metabolism, oxidative stress, lipid metabolism	[120]
Plasma	Phosphatidylcholines, (PC diacyl (aa) C36:6, PC aa C38:0 PC aa C38:6, PC aa C40:1, PC aaC40:2, PC aa C40:6, PC acyl-alkyl (ae) C40:6), Lysophosphatidylcholine (lysoPC a C18:2), and acylcarnitines (ACs) (Propionyl AC (C3) and C16:1-OH)	Synaptic dysfunction, dysfunction in mitochondrial activities, lipid metabolism	[16]
Plasma	Long chain cholesteryl esters (ChEs) (ChE 32:0, ChE 34:0, ChE 34:6, ChE 32:4, ChE 33:6,)	Lipid metabolism, Aβ metabolism,	[121]
Plasma	Cholic acid, chenodeoxycholic acid, allocholic acid, indolelactic acid, and tryptophan	Tau phosphorylation, Aβ metabolism, dysfunction in mitochondrial activities, lipid metabolism	[122]
Plasma	Branched-chain amino acids (isoleucine, leucine, and valine), creatinine and very low density lipoprotein (VLDL)-specific lipoprotein lipid subclasses	Energy deficiencies, lipid metabolism	[123]
Plasma	Anthranilic acid, glutamic acid, taurine, hypoxanthine	Oxidative stress, BBB damage, synapse damage, apoptosis	[124]
Plasma	Thymine, arachidonic acid, 2‐aminoadipic acid, N,N‐dimethylglycine, and 5,8‐tetradecadienoic acid	Energy deficiencies, lipid metabolism, DNA damage	[125]
Plasma	16-a-hydroxypregnenolone, stearic acid and PC 16:0/22:5(4Z,7Z,10Z,13Z,16Z)	Inflammation, Aβ metabolism, Tau phosphorylation, oxidative stress	[126]
CSF	Uracil, xanthine, uridine, methyl‐salsolinol, nonanoylglycine, dopamine–quinone, caproic acid, vanylglycol, histidine, pipecolic acid, hydroxyphenyl‐pyruvate, creatinine, taurine, sphingosine‐1‐phosphate, tryptophan, and 5′‐methylthioadenosine	Dysfunction in mitochondrial activities, BBB damage, energy deficiencies, oxidative stress	[127]
CSF	Total cysteine, kininogen-1 and cholesteryl ester 27:1 16:0	Energy deficiencies, Aβ metabolism	[128]
Brain	Unsaturated fatty acids (UFAs) (linoleic acid, linolenic acid, docosahexaenoic acid, eicosapentaenoic acid, oleic acid and arachidonic acid)	Tau phosphorylation, Aβ metabolism, energy deficiencies, inflammation	[129]
Brain	Diacylglycerol (14:0/14:0), triacylglycerol (58:10/FA20:5, 48:4/FA18:3), phosphatidylethanolamine (p-18:0/18:1), phosphatidylserine (PS) (18:1/18:2, 14:0/22:6)	Aβ metabolism, inflammation, apoptosis	[130]
Brain	Lauric acid, stearic acid, myristic acid, palmitic acid, palmitoleic acid, and four unidentified mass spectral features	Energy deficiencies	[131]
Plasma, CSF	Lysine, cholesterol, and sphingolipids	Energy deficiencies, DNA damage	[132]
Urine	Glycine, free cortisol/creatinine, prostaglandinsE2, β-alanyl-l-histidine	Oxidative stress, Aβ metabolism, inflammation	[133–135]
Urine, Serum	Serotonin, tryptophan, xanthurenic acid	Inflammation, neurotransmitter regulation	[136]
Fecal	Tryptophan metabolites, short-chain fatty acids (SCFAs), and lithocholic acid	Inflammation, neurotransmitter regulation	[137]

We summarized the metabolites related to the development of AD found in different metabolomic studies from different sources, and each study analyzed the pathogenesis in which these metabolites may be involved

Metabolomics research is mainly predicated on nuclear magnetic resonance (NMR) spectroscopy and mass spectrometry (MS). NMR has high repeatability but low sensitivity, whereas MS provides high throughput and sensitivity with good metabolite coverage [138].

Blood-derived metabolites of AD

Lipid metabolism has been found to be the focus of detection of novel potential AD biomarkers, mainly long-chain triglycerides (LCTs) and cholesteryl esters (ChoEs) [139]. Changes in etheric lipid metabolism pathways are a novel finding that may affect downstream mechanisms of AD [140]. Alteration in the levels of some amino acids were thought to be associated with decreased language ability, visuospatial ability, cognitive performance and changes in brain atrophy in AD patients [123, 141]. Gender-specific systemic metabolic changes have been reported in preclinical and clinical AD serum [142–144], and some metabolic effects are specific to female APOE ε4 carriers [145].

Metabolomics and MR analysis can help to identify causal mediators of AD, which is a more feasible approach to explore metabolic risk factors for AD [146, 147], and integrating metabolomic analysis can be of great help in identifying different subgroups of AD patients for future targeted and precise treatment [148, 149]. In addition, different populations have different specificities of blood metabolomes [150, 151], suggesting that identification of relevant metabolic biomarkers for diagnostic purposes requires diversity of sample populations.

CSF-derived metabolites of AD

The most variable in metabolomics in CSF is amino acid. In CSF, the concentration of several aromatic, branched and urea cycle amino acids were significantly correlated with CSF Aβ1-42, tau and p-tau-181 [152]. Metabolic-pathways of CSF metabolites, including caffeine, as well as nicotinic acid and nicotinamide, have also recently been found to be associated with AD-related core pathology [153]. The first study to fully integrate and analyze untargeted CSF metabolomics with structural MRI brain imaging data found that lower levels of gray matter in the hippocampus, thalamus, and parietal cortex were associated with dysregulated urea cycling and amino acid metabolic pathways in the CSF, which is a novel study that could continue to be developed [154], but the association of most blood metabolites with CSF metabolites is not obvious.

Brain tissue-derived metabolites of AD

In brain metabolomics, most of the differential metabolites are sphingolipids and glycerophospholipids. Analysis of various brain regions susceptible to AD lesions identified many different metabolites that are associated with amino acids, lipids, nucleotide-related bases, vitamins, neurotransmitters, energy metabolism, and oxidative stress [89, 143, 155, 156]. Recently, study found differences in gray and white matter metabolomics in the AD brain and showed differences across APOE carriers and in different stages of the disease, suggesting the need for possible personalized treatment in the future [157].

Future development of metabolomics

Current metabolomics studies all have relatively small sample sizes, which will limit our ability to study correlation and replication. It is important to determine the common and different metabolic pathways in different tissue types, which requires us to conduct follow-up studies in larger cohorts and larger data.

In recent years, urine and saliva metabolomics have gradually become very promising and convenient biological samples [158, 159]. Diversity of study specimens can contribute to the versatility of global AD biomarker studies. The comprehensive analysis of the isomers of chiral metabolites is also becoming more and more important. Comprehensive analysis of chiral metabolite isoforms is also becoming increasingly important. In addition, AD has been shown to be associated with gut flora and small molecule metabolites may play a key role in mediating microbial effects on neurotransmission and disease development. The erythrocyte lipid phenotype may be a marker for the risk of AD development [160], which we believe is the direction of strong future development.

Multi-omics science on neuroinflammation

Neuroinflammation is an immune-related response driven by activation of neuroglia. Microglia are the major immune cells in the central nervous system and are thought to be a key factor in the progression of neuroinflammation [161].

As we have previously described, AD GWAS have identified many risk genes that are highly or selectively expressed in microglia, such as ABI3, BIN1, CD3, CLU, CR1, MS4A6A, PLCG2, SPI1 and TREM2, confirming the importance of microglia-mediated neuroinflammation in neurodegeneration.

We have also stated that many AD risk variants exist in noncoding regions of the human genome and introducing genome-wide analyses of chromatin-accessible regions and histone modifications, as well as single-cell analyses, could contribute to the understanding of the microglia regulatoryome and how it is affected by AD risk variants [162], and many efforts have been made to accurately characterize microglia primarily through transcriptomic approaches. The disease-associated microglia (DAM) profile observed in mouse models of AD reflects activation states that can regulate AD risk or progression. Pro-inflammatory DAM appeared earlier in AD mouse models and is characterized by pro-inflammatory genes, surface markers CD44, potassium channel Kv1.3 and regulatory factors (NF-κB, STAT1, RELA), whereas anti-inflammatory DAM expresses phagocytic genes, surface markers with different regulatory factors CXCR4 (LXRα/β, AFTL) [163]. Deficiency of COP1 in microglia promotes changes associated with chronic microglia activation and the development of neuroinflammation [164]. Since TREM2 is the most studied star gene, more microglia-related studies have focused on this gene, and it was further found that increasing TREM2 gene doses altered microglia transcriptional programs as well as morphological and functional responses in the brains of AD mouse models [165]; GAL3 acts as a TREM2 ligand driving AD microglia activation and is thought to be an AD an upstream regulator of microglia immune responses [166]; TREM2 regulates microglia pathology in AD downstream of CD33 [167]; TREM2 deficiency has differential effects on the brain transcriptome of mice expressing different types of human APOE and significantly reduces microglia barrier action around plaques [168]. In an epigenomic study, a key role of DNA methylation in regulating the inflammatory state of microglia was recognized and TET2 was found to drive pro-inflammatory activation of microglia [169]. Deletion of HDAC1 and HDAC2 in microglia reduces amyloid load by enhancing microglial amyloid phagocytosis [170].

Although there are more studies based on transcriptomic and single-cell data, comparison of AD microglia profiles with different mouse microglia profiles showed little similarity between characterized human Alzheimer's microglia (HAM) profiles and the DAM profiles observed in mouse microglia activation profiles [162, 171]. Most of the current histological studies have been performed in mouse models for observational validation, followed by comparison with published human microglia expression profiles, and further by combining the findings with existing GWAS and human postmortem brain proteomes to determine the relevance of DAM modules in human AD, which is a limitation of the current studies, but already in maximizing the opportunity to identify cellular heterogeneity within cell populations. The use of induced pluripotent stem cell (IPSC) technology from humans has been considered as an approach to overcome this problem [161].

For microglia proteomics studies, relatively few studies have been conducted because of the generally low protein yields of their isolated cells, contamination of non-microglia proteins using existing enrichment strategies, and the limitations of the technical requirements of mass spectrometry analysis [172], but some exciting results have been obtained, identifying COTL1, MSN as microglia-specific markers with increased expression in human AD [172, 173]. In a CSF proteomics, SIRT2 and HGF were suggested to be chemotactic for many different immune cells involved in the inflammatory response and activate them [174]. In addition, information on microglia alterations at different stages of Aβ deposition in AD has been investigated [175] to provide help in monitoring the efficacy of microglia repair strategies. Although we were able to obtain data on microglia proteomic changes, it was found that transcript levels do not necessarily reflect the levels of proteins that ultimately control cellular function [176]. This also makes it more difficult to further identify the role of microglial cell proteins in AD for studies related to more readily available cerebrospinal fluid and blood specimens compared to brain tissue. Polyunsaturated fatty acid metabolites were identified by metabolomic analysis corresponding to microglia activation and inflammation [177, 178]. An association between blood metabolomics and neuroinflammation did not identify AD-specific biomarker candidates in a study [179], which further emphasizes the complexity of AD etiology and blood biomarker diagnosis.

Interventions against neuroinflammation exacerbated by over-activated microglia represent a promising therapeutic strategy for AD [180]. Current research efforts need to identify specific pathways that can be actually and effectively targeted, and therefore, further efforts are needed to understand and manipulate microglia in AD, to further understand the various mechanisms by which microglia counteract AD pathology, and the exact timing when microglia-related interventions need to be applied.

Multi-omics integration of AD

Comprehensive analysis of data obtained from different omic approaches is essential to gain insight into disease pathological processes. Systems biology (SB) is to take as much information as possible at every level and integrate them to explain the interactions between molecules [181]. A multi-omics approach identified AD autoantibodies with disease diagnostic capabilities and demonstrated that autoantibodies against isovaleryl coenzyme A dehydrogenase (IVD), ADD2 and CYFIP1 were able to distinguish AD patients from healthy controls [182]. Madrid, L et al. found that blood and cortical biomarkers showed opposite profiles in APOE2 and APOE4 carriers [183]. A study by Clark, C et al. found that total cysteine, kininogen-1 and cholesteryl ester 27:1 16:0 were associated with AD pathology [128]. Two important genes (ABCA1 and CPT1A) and two proteins (lipocalin and neutrophil gelatinase-associated lipid transport protein (NGAL)) involved in acylcarnitine and amine regulation were identified by Horgusluoglu, E et al. through an integrated multi-omics approach [184]. These cross-omics pathways suggest that the development of AD may be the result of genetic variation with potential cascading effects on the downstream transcriptome and proteome levels as well as on the metabolome. Multi-omics studies provide a better understanding of the disease, from the original cause of the disease to functional consequences or related interactions.

Many recent advances in statistics have made it possible to integrate information from multiple data modalities to thoroughly explore disease-related endophenotypic networks and biological interactions. In this context, multifaceted and integrated research networks applied to AD have been developed. One idea that unifies these different approaches is to use it to address the relationships between multiple biological pathways and to correlate them. The Genome-wide Positioning Systems Platform for Alzheimer's Drug Discovery (AlzGPS) is used to accelerate the discovery of biologically relevant targets and clinically relevant drug candidates for AD by combining systems pharmacology and integrated web-based analysis of multi-omics data [185]. Another similar platform that uses gene regulatory networks as targets for integrating multiple levels of data to identify key disease pathways and driver genes is superior to the former in that it also includes in vitro and in vivo functional validation within the platform [186]. An open-source comparative analysis of computational pipelines called scGRNom (single-cell Gene Regulatory Network prediction from multi-omics) reveals interactions between genomic function, pathways, cell types, and disease, enabling the ability to predict the clinical phenotype of AD from its cellular type disease genes [187]. Considering that AD is a heterogeneous disease, understanding how molecular interactions evolve over time can help determine the optimal time point for biomarker measurements or the time window of therapeutic action of a specific drug, it has been shown that by a comprehensive analysis of multi-omics data with animal genome-scale metabolic models (GEM), that is, by correlating gene expression changes with subsystem and metabolite levels, AD mice show tissue- and time-dependent changes [188], which will also help to translate into diagnostic or therapeutic approaches. In addition, identifying its association with AD-related changes in brain function and structure through integration of multi-omics studies is also an approach to help guide the interpretation of molecular data by providing a global view of the disease [189, 190].

Potential therapeutic approaches based on multi-omics

The omics research has enriched our knowledge of AD at multiple levels, leading us to develop better and more targeted therapies (Table 5).

Table 5.

Potential therapy targets for AD

Targets	Drugs	Stage	Trial ID	Refs.
Genomics and epigenomics
ABI3	–	In vitro	–	[191]
ABCA7	–	–	–	[192]
ADAM10	CRISPR/Cas9	In vitro	–	[193]
ATP6V1A	NCH-51	In vitro	–	[186]
BACE1	Atabecestat (JNJ-54861911)	IIb	NCT02569398	[194]
	Lanabecestat (AZD3293)	I	NCT02005211	[195]
	Verubecestat (MK-8931)	II	NCT01953601	[196]
Chi3l1/YKL-40	–	In vitro	–	[196]
IL10	HEK293T cells	In vivo	–	[197]
TREM2	PHSA@PF/pTREM2	In vitro	–	[180]
	AL002	I	NCT03635047	[198]
TYROBP / DAP12	–	In vitro	–	[199]
UNC5C	HEK293T cells	In vivo	–	[200]
H3K27ac, H3K9ac	–	In vitro	–	[56]
H3K4me3	WDR5-0103	In vitro	–	[201]
H3K9me2	BIX01294	In vitro	–	[202]
Transcriptomics
miR-346	–	In vivo	–	[203]
lncRNA XIST	–	In vitro	–	[204]
Proteomics
AMPKα1	–	In vitro	–	[205]
BDNF	–	In vitro	–	[206]
CSF1R	PLX3397 JNJ-40346527	In vitro In vitro	– –	[207] [208]
Eotaxin-1	–	–	–	[209]
Fyn	Saracatinib	IIa	NCT02167256	[210]
RIPK1	DNL747	Ib	NCT03757325	[211]
NEP	PPAR-siSOX9	In vitro	–	[212]
NGF	AAV2-NGF2	II	NCT00876863	[213]
p38 MAPKβ	VX-745	In vitro	–	[214]
Metabolomics
BCAAs	–	–	–	[141]
Homocysteine	Vitamin B	III	NCT00056225	[215]
EGCG	–	III	NCT00951834	[216]
LOX	Baicalin	In vitro	–	[217]
Oxidative stress products	Vitamin C	I	NCT00117403	[216]
Sterol derivative	NE3107	III	NCT04669028	[218]

“–”The information has not been obtained from the primary publication or has not been found yet

Targeting a specific component of these pathways may bring new directions for drug discovery and treatment. TREM2 is at the forefront of these therapeutic candidates, and a novel TREM2-based treatment for microglia function is currently being clinically tested [198]. Another potential treatment strategy may be based on a special diet or pharmaceutical preparations to maintain the normal level of these metabolites in the early stages of AD [219]. But at the same time, because it is a characteristic of metabolites to affect multiple targets and signaling pathways, which makes it more and more difficult to prove their pharmacological activity when studying pharmaceutical preparations. The results of two large Phase III clinical trials conducted with Aducanumab showed no clinical benefit for AD [220]. In AD, drug trials designed to eliminate toxic amyloid in the brain have so far failed to achieve their main clinical goals. Therefore, targeting the protein co-expression module most relevant to the pathophysiology of AD is a promising method for drug development. Chaperone-mediated autophagy in neurons can maintain protein homeostasis and have a positive impact on Aβ deposition and tau pathology, which is a promising therapeutic approach [221]. What is more, the first anti-tau antibody, semorinemab, did not improve outcomes in a phase II trial in patients with prodromal or mild AD [222]. There are other anti-tau agents that have not yet obtained efficacy data, and we look forward to an inspiring result.

The 2018 ATN criteria were recommended as a standard framework for AD research, which simply combines the three CSF AD biomarkers—Aβ (A), tau (T) and neurodegeneration (N)—to move AD research toward personalized medicine [223]. We believe that the combination of the ATN framework with omics can help high-quality pre-screening of patients for further selection procedures in clinical trials. The potential role of plasma ATN biomarker profiles in clinical trials predicting AD pathology and clinical progression is currently receiving positive results and is considered to act as a therapeutic response marker [224, 225]. The current maximal plasma proteomics study of the ATN framework, which compared plasma protein profiles in various ATN variants and replicated four proteins (FCN2, PAI-1, CRP, and sVCAM1) in relation to the ATN framework and clinical diagnosis of AD, further suggests that these proteins are involved in AD pathological pathways and may be promising targets for future therapy [226]. An MRI-based proteomics study of the ATN framework showed that the underlying biological factors associated with neurodegeneration are different at different levels of disease progression [227]. We know that candidate biomarkers representing "N" encompass different physiological processes, that neurodegeneration is a dynamic process, and that longitudinal measures of neurodegeneration can better capture the brain changes associated with AD. Whereas the multi-omics characterization of neurodegeneration has tremendous value, the proposed ATN framework itself hopes to achieve future flexibility by adding other biomarkers at the time of discovery and validation, the combination with multi-omics can provide information on clinical changes and progression, ensure reliability of biomarker assays, and identify biomarkers with sufficient dynamic range during short-term treatment. In contrast, A and T are considered intermediate targets whose modification may lead to neuroprotection and disease modification through related mechanisms [228], and biomarkers of these related pathological mechanisms identified through multi-omics may play an important role in finding targets for drug therapy for individuals with or at risk of developing AD. Scholars believe that this may change as data accumulate on the impact of treatment on ATN biomarkers and the impact of treatment on clinical outcomes, and that data from emerging biomarker collections may lead to a better understanding of ATN biomarker trajectories and biomarker-clinical outcome measures [229], then multi-omics studies are a very good way to obtain more emerging biomarkers. We believe that this approach will help design clinical trials that stratify subjects by ATN profile, thus enabling multi-omics data to identify biological stages that improve disease.

Although omics is an exciting research to develop treatments for AD, there are still many limiting factors that need to be overcome, the most important one is how to connect the various omics together. At present, the integration of multi-omics and SB in AD is still at an early stage, and the development of SB and gene networks in various molecular systems is relatively mature [230]. Drug development urgently needs a holistic and systematic approach to deal with molecular data, omics data and clinical information, which can effectively provide accurate biomarker-guided targeting methods [8].

The idea of designing AD multi-target drugs (MTD) has become a research hotspot in recent years. Its central goal is to synthesize drugs that combine two or more pharmacophores to simultaneously modulate multiple factors involved in the development of AD, which is more effective in influencing the disease network [231]. There are many studies on neurotransmitter pathways and classical pathological pathways of AD, however, MTD for AD have not achieved the expected effect [232], and most scientists believe that in addition to developing new drugs, redeveloping old drugs, or traditional herbal medicines may a promising treatment [233]. However, it is not clear whether MTD produces a synergistic therapeutic effect. If we can detect the relevant markers that the drug plays a role in the interaction network between the key factors regulating the occurrence and development of AD, it will be of great help for us to evaluate the efficacy or drug design. It is believed that the knowledge transformation from the combination of omics and SB to MTD is expected to bring a breakthrough in the treatment of AD patients in the future.

Conclusions

In general, biomarkers in vivo are critical not only for early diagnosis of diseases but also for prognostic assessment, classification of disease progression and treatment of disease remission. Omics technologies help to provide a more comprehensive view of the molecular pathways involved in disease progression. Through data mining, lists of disease-associated genes are generated and potential biomarkers are discovered through functional enrichment analysis, protein–protein interactions, and analysis of structure and function. Advances in high-throughput technologies have made AD studies more possible and are no longer limited to a single omic analysis, but the data sets generated can be very large and may require increased computational power. Therefore, it is challenging to obtain a large amounts of multi-omics data in a single study. Appropriate bioinformatics tools play an important role in solving this problem. New technologies and platforms have been generated in the course of studies in various omics, a lot of models for predicting or accurately distinguishing AD and control are generated, different work processes and techniques are used to carry out relevant measurements in different studies, so it is necessary to standardize the study design and define a specification to guide the studies in each omics research, and design protocols that are economically parsimonious and capable of finding the best diagnostic performance.

We believed that in the near future, omics-based studies will help to understand how these molecular interactions evolve over time and could help to determine the optimal time point for biomarker measurement or the time window for therapeutic action of a specific drug. The integrated method has great potential in future diagnostic evaluations. Combine omics technology with clinic, developing novel methods that allow clinical trials to be carried out in a pre-clinical stage, and possibly in smaller-scale cohorts, to determine biomarkers that can be easily run in the clinic.

Abbreviations

AD

Alzheimer's disease

Aβ

Amyloid beta

NFTs

Neurofibrillary tangles

PTM

Posttranslational modification

CSF

Cerebrospinal fluid

t-Tau

Total tau

p-tau

Phosphorylated tau

PET

Positron emission tomography

fMRI

Functional magnetic resonance imaging

FDG

F-fluorodeoxyglucose

NfL

Neurofilament light chain

MCI

Mild cognitive impairment

GWAS

Genome-wide association studies

SNPs

Single nucleotide polymorphisms

LOAD

Late-onset AD

CNVs

Copy number variations

NGS

Next-generation genome sequencing

WES

Whole-exome sequencing

WGS

Whole-genome sequencing

PSEN1

Presenile hormone 1

PSEN2

Presenile hormone 2

EOAD

Early-onset AD

PiB

Pittsburgh compound-B

EWAS

Epigenome-wide association study

scRNA-seq

Single-cell RNA sequencing

cRNA

Protein-coding RNA

ncRNA

Non-protein-coding RNA

TWAS

Transcriptome-wide association studies

lncRNAs

Long non-coding RNAs

miRNAs

MicroRNAs

circRNAs

Circulating RNAs

piRNAs

PIWI-interacting RNAs

NAT

Natural antisense transcripts

BBB

Blood–brain barrier

NMR

Nuclear magnetic resonance

MS

Mass spectrometry

LCTs

Long-chain triglycerides

ChoEs

Cholesteryl esters

DAM

Disease-associated microglia

HAM

Human Alzheimer's microglia

IPSC

Induced pluripotent stem cell

SB

Systems biology

GEM

Genome-scale metabolic models

MTD

Multi-target drugs

Author contributions

JTY proposed the idea and constructed the structure of the review, as well as revising the manuscript. QA performed the literature search and manuscript writing. ZTW, KMW, XYH and QD participated the literature research and made all illustration. All authors read and approved the final.

Funding

This study was supported by grants from the National Natural Science Foundation of China (82071201,81971032), Shanghai Municipal Science and Technology Major Project (No.2018SHZDZX01), Shanghai Talent Development Funding for The Project (2019074), Research Start-up Fund of Huashan Hospital (2022QD002), Excellence 2025 Talent Cultivation Program at Fudan University (3030277001), and ZHANGJIANG LAB, Tianqiao and Chrissy Chen Institute, and the State Key Laboratory of Neurobiology and Frontiers Center for Brain Science of Ministry of Education, Fudan University.

Data availability

Not applicable.

Declarations

Conflict of interest

The authors declare that they have no competing interests.

Ethical approval

Not applicable.

Consent to participate

Not applicable.

Consent for publication

All participants were properly consented.