Multi Layered Omics Approaches Reveal Glia Specific Alterations in Alzheimer's Disease: A Systematic Review and Future Prospects

ABSTRACT

Alzheimer's disease (AD) is the most common neurodegenerative dementia with multi‐layered complexity in its molecular etiology. Multiple omics‐based approaches, such as genomics, epigenomics, transcriptomics, proteomics, metabolomics, and lipidomics are enabling researchers to dissect this molecular complexity, and to uncover a plethora of alterations yielding insights into the pathophysiology of this disease. These approaches reveal multi‐omics alterations essentially in all cell types of the brain, including glia. In this systematic review, we screen the literature for human studies implementing any omics approach within the last 10 years, to discover AD‐associated molecular perturbations in brain glial cells. The findings from over 200 AD‐related studies are reviewed under four different glial cell categories: microglia, oligodendrocytes, astrocytes and brain vascular cells. Under each category, we summarize the shared and unique molecular alterations identified in glial cells through complementary omics approaches. We discuss the implications of these findings for the development, progression and ultimately treatment of this complex disease as well as directions for future omics studies in glia cells.

Main Point

• Graphical overview of the literature summarized in this review, spanning glia cells and levels of omics analysis present on Alzheimer's disease.

Abbreviations

Aβ

amyloid beta

AD

Alzheimer's disease

ADAD

autosomal dominant AD

ADHD

attention‐deficit/hyperactivity disorder

ADNI

Alzheimer disease neuroimaging initiative

ALS

amyotrophic lateral sclerosis

AMP‐AD

accelerating medicines partnership in AD

ARM

amyloid response microglia

AsymAD

asymptomatic AD

ATAC‐seq

assay for transposase‐accessible chromatin sequencing

BBB

blood brain barrier

BD

bipolar disorder

CAA

cerebral amyloid angiopathy

CBD

corticobasal degeneration

CBF

cerebral blood flow

CE

cholesterol esters

CGF

colonized germ free

CNS

central nervous system

5xFAD

an AD mouse model

CSVDL

cerebral small vessel disease

CV

common variant

DAA

disease associated astrocyte

DAM

disease associated microglia

DEG

differentially expressed genes

DHS

DNase hypersensitivity site

DLPFC

dorsolateral prefrontal cortex

DMP

differentially methylated position

DNAm

DNA methylation

DOL

disease associated oligodendrocyte

DSB

double stranded break

E‐P

enhancer‐promoter

ECM

extracellular matrix

eQTL

expression quantitative trait loci

EWAS

epigenome wide association study

FACS

fluorescence activated cell sorting

FANS

fluorescence activated nuclei sorting

FTD

frontotemporal dementia

GF

germ free

GWAS

genome‐wide association studies

hMGLs

human embryonic stem‐cell derived microglia‐like cells

IF

impact factor

IT

inferior temporal region

LD

linkage disequilibrium

LDSR

linkage disequilibrium score regression

LOAD

late‐onset AD

LPS

lipopolysaccharide

LTP

long term potentiation

MCI

mild cognitive impairment

MDD

major depressive disorder

MR

Mendelian randomization

MS

multiple sclerosis

MTG

middle temporal gyrus

MWAS

methylome wide association study

NFT

neurofibrillary tangle

NT

neuropil threads

OCC

occipital cortex

OCR

open chromatin region

OEG

oligodendrocyte enriched glia

OL

oligodendrocyte

OPC

oligodendrocyte precursor cell

PBMC

peripheral blood mononuclear cell

PC

principal component

PD

Parkinson's disease

PET

positron emission tomography

PIGs

plaque induced genes

PLAC‐seq

Proximity Ligation‐Assisted ChIP‐Seq

PSP

primary supranuclear palsy

QTL

quantitative trait loci

ROSMAP

religious orders study and rush memory and aging project

RPS

risk profile score

scRNAseq

single‐cell RNA sequencing

SCZ

schizophrenia

sLDSR

stratified LD score regression

SNP

single nucleotide polymorphism

snRNAseq

single‐nucleus RNA sequencing

SPF

specific pathogen free

sQTL

splicing quantitative trait loci

ST

superior temporal region

STG

superior temporal gyrus

Synaptoneurosomes

structures assembled by the sealed presynaptic bouton and the attached post‐synaptic density

TAG

triacylglycerol

TEER

trans‐endothelial electrical resistance

TF

transcription factor

TMT‐MS

tandem mass tag‐mass spectrometry

TREM2

triggering receptor expressed on myeloid cells 2

TWAS

transcriptome‐wide association study

WGCNA

weighted gene co‐expression network analysis

WT

wild type

1. Methods

For this review, we performed a PubMed search using following keywords and filtering criteria:

(Alzheimer's[Title/Abstract] OR Alzheimer[Title/Abstract]) AND ((omics[Title/Abstract]) OR (multiomics[Title/Abstract]) OR (multi‐omics[Title/Abstract]) OR (transcriptome[Title/Abstract]) OR (transcriptomics[Title/Abstract]) OR (proteome[Title/Abstract]) OR (proteomics[Title/Abstract]) OR (methylome[Title/Abstract]) OR (methylomics[Title/Abstract]) OR (epigenome[Title/Abstract]) OR (epigenomics[Title/Abstract]) OR (epigenetics[Title/Abstract]) OR (metabolome[Title/Abstract]) OR (metabolomics[Title/Abstract]) OR (lipidome[Title/Abstract]) OR (lipidomics[Title/Abstract]) OR (GWAS[Title/Abstract]) OR (EWAS[Title/Abstract]) OR (MWAS[Title/Abstract]) OR (PWAS[Title/Abstract])) AND ((microglia[Title/Abstract]) OR (microglial[Title/Abstract]) OR (astrocyte[Title/Abstract]) OR (astrocytic[Title/Abstract]) OR (astroglia[Title/Abstract]) OR (oligodendrocyte[Title/Abstract]) OR (oligodendroglia[Title/Abstract]) OR (oligodendroglial[Title/Abstract]) OR (OPC[Title/Abstract]) OR (oligodendrocyte progenitor cell [Title/Abstract]) OR (oligodendroglia [Title/Abstract]) OR (glia[Title/Abstract]) OR (glial[Title/Abstract]) OR (vascular[Title/Abstract]) OR (endothelial[Title/Abstract]) OR (endothelia[Title/Abstract]) OR (pericyte[Title/Abstract]) OR (pericytic[Title/Abstract]) OR (celltype[Title/Abstract]) OR (cell‐type[Title/Abstract])) AND ("english"[Language]) AND ("2013/01/01"[Date–Publication]: "3000"[Date–Publication]) AND ("freetext"[Filter]) AND ("exclude preprints"[Filter]) NOT ("review"[Filter]) NOT ("comment"[Filter]) NOT ("personal narrative"[Filter]) NOT ("retracted publication"[Filter]) NOT ("case reports"[Filter]) NOT ("technical report"[Filter]) NOT ("personal narrative"[Filter]) NOT ("editorial"[Filter])

Our PubMed search conducted on June 09, 2023 yielded 439 relevant glial omics peer‐reviewed articles. Exclusion of articles in journals with a 5‐year impact factor (IF) < 5, yielded 382 articles. From these 382 articles, additional manual curation was carried out to keep relevant articles, including those that are (1) primary research articles, (2) articles that involve mainly human data/results, (3) for microglia transcriptomics study, articles need to have IF > 10, and (4) exclusion of preprint articles. Manual annotations were performed to assign articles to corresponding omics and cell types for review. New or relevant publications were added as applicable. Our final curation resulted in 236 AD focused articles.

2. Introduction

Multi omics approaches have accelerated Alzheimer's disease (AD) research exponentially in the last decade. Each layer of omics data and analysis has illuminated, enhanced, and henceforth expanded our understanding of the molecular, epigenetic, cellular, and spatial underpinnings of AD pathology. Further investigation, however, is necessary to elucidate late onset AD cause and progression. This review is an amalgamation of a selection of genomics, epigenomics, transcriptomics, and other omics approaches obtained from 236 AD focused studies. The review will have the following sections for glia cells: microglia, oligodendrocytes, astrocytes, and brain vascular cells. Each glia cell section will introduce the topic, evaluate the current literature per each omics type, followed by a short summary. Over the last decade, genomics and transcriptomics were the earliest omic studies published for all brain glial cell types (Figure 1). There are gaps such as an absence of epigenetics‐based studies on vascular cells as well as metabolomics/lipidomics focused studies on oligodendrocytes (Figure 1). There has been a recent spike in transcriptomics and proteomics‐based studies in all cell types (Figure 1), which is likely in part due to advances in technology. Recently, single cell/single nucleus RNA sequencing has advanced our knowledge of glial cell type specific expression profiles associated with AD. Focus on furthering omics studies in vascular cell types and oligodendrocytes with AD pathology, increasing epigenomic and metabolomic/lipidomic studies in all cell types as well as integrating the different omics data types are needed for a comprehensive understanding of the molecular perturbations of glia in AD. In the future, continued investment in these areas and further growth of spatial transcriptomic studies are necessary to push the boundaries of our current understanding of glial multi‐omics changes in AD including those in relation to regional pathology changes.

FIGURE 1.

Bar chart of how many studies there are per cell type per omics per year. The percentages are not relative to other cell types.

3. Microglia

3.1. Introduction

Microglia play crucial roles in immune pathways and cell proliferation in the brain. Dysregulation of these cells correlate with AD related pathologies such as increased Aβ accumulation and tau related pathologies. It is one of the most extensively studied glial cells in AD research where a plethora of genetic, epigenetic, and transcriptomic studies are present on microglia. There is a necessity for increased proteomic‐, metabolomic‐ and lipidomic‐focused studies, however, in this field (Figures 1 and 2). A void in spatial transcriptomic based studies in microglia at present will most likely be met as this field continues to grow (Figure 1).

FIGURE 2.

Key takeaways from Alzheimer's disease omic studies in glial cells (Created with Biorender.com).

3.2. Genomics

Genome‐wide association studies (GWAS) of AD risk identified many AD‐associated variants with genome‐wide significance (GWS, p < 5 × 10⁻⁸) that are within or in close proximity to genes implicated in microglial function. Initial GWAS or meta‐analyses identified a small number of AD‐associated loci, including APOE, CR1, BIN1, CD2AP, EPHA1, CLU, MS4A6A, PICALM, ABCA7, CD33, TREM2, MS4A4/MS4A6E and SORL1 (Lambert et al. 2009; Harold et al. 2009; Hollingworth et al. 2011; Naj et al. 2011; Guerreiro et al. 2013; Jonsson et al. 2012; Miyashita et al. 2013), among which TREM2, MS4A4, MS4A6A, CD33 and SORL1 were most highly expressed in microglia among major central nervous system (CNS) cell types. In 2013, Lambert et al. reported 19 AD‐associated loci, generated from their stage 1 meta‐analysis and stage 2 genotyping results, of which 11 loci were new (Lambert et al. 2013). Some of these loci were implicated in immune response and inflammation pathways (HLA‐DRB5, HLA‐DRB1, INPP5D and MEF2C) and microglial and myeloid cell function (INPP5D, SPI1/CELF1) (Lambert et al. 2013; Huang et al. 2017). A three‐stage GWAS study by Sims et al. (Sims et al. 2017) later reported novel rare coding protective variant rs72824905 (PLCG2) and risk variants rs616338 (ABI3) and rs143332484 (TREM2), while confirming the GWS association of previously reported risk variant rs75932628 (TREM2) (Guerreiro et al. 2013; Jonsson et al. 2012; Jin et al. 2014). PLCG2, ABI3 and TREM2 are highly expressed in microglia with limited expression in other main CNS cell types (Sims et al. 2017). Kunkle et al.'s large scale meta‐analysis of AD GWAS studies, followed by genotyping and analysis in additional samples, discovered five novel loci (IQCK, ACE, ADAM10, ADAMTS1, and WWOX) (Kunkle et al. 2019). Combining the newly and previously identified GWAS variants, 400 protein‐coding genes were discovered within ±500 kb of the sentinel variants with linkage disequilibrium (LD) r ² ≥ 0.5 (Kunkle et al. 2019). According to brain RNAseq data, the highest percentage (20.3%) of these 400 genes were most highly expressed in microglia/macrophages, if excluding fetal astrocytes (Kunkle et al. 2019). Jansen et al. employed a phase 1 GWAS study consisting of AD and control, a phase 2 study with AD‐by‐proxy and control‐by‐proxy, and a phase 3 study with meta‐analysis of phases 1 and 2 (Jansen et al. 2019). Following phase 3, 2357 GWAS variants were identified, located in 29 distinct loci, among which 9 were novel (Jansen et al. 2019). Gene‐set enrichment analyses of gene‐based association p‐values suggested enrichment in microglia (among other CNS cell types), in one mouse brain dataset and one human brain scRNAseq dataset, which was partially dependent on APOE (Jansen et al. 2019). In 2021, Schwartzentruber et al. reported four novel loci near CCDC6, TSPAN14, NCK2, and SPRED2 from genome wide meta‐analysis study (Schwartzentruber et al. 2021). Recently, Bellenguez et al. identified 42 new risk loci and highlighted microglia's involvement in AD and related dementia (Bellenguez et al. 2022).

AD GWAS‐derived analyses, such as polygenic score, expression quantitative trait loci (eQTL), colocalization analysis and linkage disequilibrium score regression (LDSR), revealed association between AD‐associated variants and microglia or immune cell types. Using brain gene expression from temporal cortex and cerebellum of donors with AD, progressive supranuclear palsy (PSP)—a primary tauopathy, and controls without neurodegenerative pathology, we found significant enrichment of cis‐SNPs that associate with brain expression of nearby genes among human disease‐associated variants (Zou et al. 2012), suggesting gene expression changes as a mechanism for many central nervous system (CNS) diseases, including AD. Our combined assessment of brain gene expression and disease GWAS revealed associations between brain levels of many AD GWAS genes and their cis‐SNPs that also associate with AD risk, including genes enriched in microglia or implicated in immune pathways such as MS4A4A (Allen et al. 2012) and HLA‐DRB1 (Allen et al. 2015). We identified a functional regulatory variant (rs9357347) at the TREM locus associated with AD protective effect and higher brain TREML1 and TREM2 levels which suggest that modest increases in brain levels of these genes may confer protection from AD (Carrasquillo et al. 2017). Using brain gene co‐expression network analysis, we determined that a microglial gene‐enriched co‐expression network that has significantly higher levels in AD temporal cortex also harbors many AD risk associated genes including ABI3, HLA‐DRB1/5, INPP5D, MS4A4A/6A, and TREM2 (Conway et al. 2018). Collectively, these findings support a role for microglial gene expression changes in modifying AD risk.

Subsequent studies that evaluated brain gene expression and AD risk GWAS genes or variants provided further support for our findings. Calderon et al. identified an association between microglia and AD status with or without removing the APOE region, through using polygenic score of AD GWAS (Lambert et al. 2013) from proximity region of each protein‐coding gene and cell type specific expression from scRNAseq of fresh healthy human brain (Darmanis et al. 2015; Calderon et al. 2017). Lopes et al. reported the regulatory role of genetic variants in microglia expression from controls and patients of neurodegenerative disorders including AD cases (Lopes et al. 2022). Focusing on AD‐associated genes (Kunkle et al. 2019; Marioni et al. 2018), a cross‐study eQTL meta‐analysis identified genes with shared effects between microglia and monocytes (MS4A6A, RABEP1, CD33, FCER1G, and ABCA7), and genes unique to microglia (BIN1, PICALM, USP6NL and GNGT2) (Lopes et al. 2022). Intriguingly, CASS4 had a significant effect but in the opposite direction in microglia and monocytes, implicating complex regulatory mechanisms. Further, colocalization analyses of AD‐associated variants (Lambert et al. 2013; Kunkle et al. 2019; Jansen et al. 2019; Marioni et al. 2018) and microglia QTL identified genes BIN1, PICALM and USP6NL (Lopes et al. 2022). Gagliano et al. partitioned the genetic heritability in gene sets marking different cell type/tissues or functional categories through sLDSR (Finucane et al. 2015) using AD GWAS summary statistics (Lambert et al. 2013; Gagliano et al. 2016). They identified significant heritability of genes marking cell types involved in innate (e.g., monocytes and neutrophils) and adaptive (e.g., T cells) immune systems (Gagliano et al. 2016). Rosenthal et al. identified two interactome clusters from AD GWAS (Jansen et al. 2019) genes and their proximal genes that were involved in immune activity and highly expressed in microglia (Rosenthal et al. 2022).

GWAS and/or GWAS‐derived analyses of other AD‐relevant phenotypes, such as hippocampal volume, CSF sTREM2 and brain amyloid burden, imply crucial roles for microglia or immune cell types. Lancaster et al. calculated AD related risk profile score (RPS) using 56 genes from a microglia protein–protein network (Sims et al. 2017), which explained the variance of hippocampal volume (Hibar et al. 2015; Elliott et al. 2018) (p < 0.05) (Lancaster et al. 2019). Among these 56 genes, 32 contributed to the AD‐RPS and hippocampal volume association, with the top five being PLCG2, BLNK, HMHA1, NCF4, and ARHGAP24 (Lancaster et al. 2019). Deming et al. performed CSF sTREM2 GWAS in ADNI samples (Deming et al. 2019). A common variant rs1582763, located in MS4A gene region, was associated with elevated CSF sTREM2 level (p = 1.15 × 10⁻¹⁵) (Deming et al. 2019). Analysis within MS4A region identified rs6591561, independent of rs1582763, being associated with reduced CSF sTREM2 level (p = 1.47 × 10⁻⁹) (Deming et al. 2019). Various Mendelian randomization (MR) models estimated a causal effect from CSF sTREM2 on AD risk (Deming et al. 2019). Furthermore, targeting MS4A4A on human macrophages decreases soluble TREM2 in vitro and MS4A4A colocalizes with TREM2 in human macrophages (Deming et al. 2019). Ramanan et al. performed GWAS of longitudinal change in brain amyloid burden measured by (18)F‐florbetapir PET (Ramanan et al. 2015). Variant rs12053868‐G was associated with higher rate of amyloid accumulation (p < 5 × 10⁻⁸), which is located in intronic region of IL1RAP, a gene most highly expressed in microglia in CNS (Ramanan et al. 2015). Furthermore, IL1RAP rs12053868 is associated with a marker of cortical microglial activation (Ramanan et al. 2015). Using quantitative neuropathology measures as outcomes, we found significant association of the PLCG2 AD‐protective variant (rs72824905‐G) with reduced tau pathology in PSP, supporting a role for this variant in suppressing tau pathology (Strickland, Morel, et al. 2020). In sum, association studies with AD‐related endophenotypes provide further evidence for role and potential mechanism of action of AD‐risk associated microglial genes.

3.3. Epigenomics

AD GWAS variants reside within transcription factor (TF) binding motifs and/or enhancer‐promoter (E‐P) linked open chromatin regions (OCRs) in microglia. Further, microglia OCRs account for at least part of AD heritability. Tansey et al. discovered that AD risk variants (Lambert et al. 2013) were enriched in human microglia at assay for transposase‐accessible chromatin sequencing (ATAC‐seq) peaks (Gosselin et al. 2017), and enriched in peaks containing binding motifs of RUNX, SPI1 and SPDEF (Tansey, Cameron, and Hill 2018). In macrophages, these variants are enriched in DNase hypersensitivity sites (DHSs), and DHS containing binding motifs of SPI1, EGR1, MEF2A, and CEBPA (Tansey, Cameron, and Hill 2018). They confirmed that in both microglia and macrophages, AD heritability was enriched in SPI1 motif containing OCRs (Tansey, Cameron, and Hill 2018). Nott et al. found that AD SNP heritability was highly enriched in microglia enhancers and promoters, while psychiatric disorders such as autism, major depressive disorder (MDD), attention deficit hyperactivity disorder (ADHD), schizophrenia (SCZ), and bipolar disease (BD) showed enrichment in neurons (Nott et al. 2019). Other neurological disorders, however, including amyotrophic lateral sclerosis (ALS), multiple sclerosis (MS), Parkinson's disease (PD) and epilepsy did not show enrichment in specific brain cell types (Nott et al. 2019). Importantly, Proximity Ligation‐Assisted ChIP‐Seq (PLAC‐seq) was utilized to identify enhancer‐promoter (E‐P) interactions: 25 PLAC‐linked AD‐risk genes were identified in microglia, of which 14 were unique (such as PICALM and BIN1) and not observed in neurons or oligodendrocytes (Nott et al. 2019). More recently, Kosoy et al. confirmed the enrichment of AD GWAS variants in microglia OCR (Kosoy et al. 2022). In addition, with Hi‐C, ATAC‐seq and H3K27ac data, they predicted E‐P linked OCRs and genes under regulation in microglia, neuronal and non‐neuronal populations in AD and controls (Kosoy et al. 2022). The enrichment of AD risk variants was only observed in microglia (Kosoy et al. 2022). Further, microglia OCRs explained more AD heritability compared to neuronal and non‐neuronal cells, whereas this value was even higher in E‐P linked microglia OCRs (Kosoy et al. 2022). Ramamurthy et al. profiled H3K27ac peaks, reflecting activated transcription, in neuronal, microglial and oligodendrocyte‐enriched glial cells, isolated from AD and non‐AD postmortem brains (Ramamurthy et al. 2022). It was reported that AD risk loci (Kunkle et al. 2019; Jansen et al. 2019) were enriched in microglia‐specific H3K27ac peaks, and microglial‐specific peaks had a statistically significant preference for colocalization with AD associated SNPs (Ramamurthy et al. 2022). A majority (10/17) of genes that colocalized in AD included SNPs overlapping with microglial enhancers, such as USP6NL (Lopes et al. 2022).

AD GWAS variants were discovered to affect TF‐binding in microglia OCRs and/or their target genes. Kosoy et al. constructed TF‐to‐TF regulatory networks, and by weighing the nodes with AD risk variants (Jansen et al. 2019), they predicted 23 TFs in perturbed regulatory hubs, 16 of which were expressed in microglia (Kosoy et al. 2022). Further analyses highlighted SPI1 and TAL1 (Kosoy et al. 2022). Lutz et al. used monocyte data (Garnier et al. 2013) as a surrogate for microglia, and applied a bioinformatics framework to predict the impact of enhancer SNPs within AD risk regions (Kunkle et al. 2019) on TF binding (Lutz and Chiba‐Falek 2022). Genes SPI1, SP1, RUNX3, and FOXO1 were identified as harboring candidate AD risk SNPs in strong enhancer regions that interrupt TF binding (Lutz and Chiba‐Falek 2022). Another study developed scEpiLock, a prediction model for cis‐regulatory element location and variant impact estimation (Gong et al. 2022). This model was trained using single cell ATAC‐seq peaks for various cell types of peripheral blood mononuclear cells (PBMCs) and brain tissue, including microglia (Gong et al. 2022). It predicted that SNP rs10769263, located ~17 kb upstream of SPI1, reduced the regional accessibility in microglia (Gong et al. 2022). Jin et al. developed scGRNom that can construct networks linking TFs, regulatory elements (e.g., enhancers and promoters) and target genes (Jin et al. 2021). When applied to microglia‐specific OCR, it identified gene regulatory networks with MEF2A and RFX3 discovered as the predicted hubs (Jin et al. 2021). Combining microglia‐specific network with AD GWAS SNPs, they identified AD disease genes that were enriched in amyloid β formation and clearance, MAPK signaling, and neuron death pathways (Jin et al. 2021).

DNA methylation is another epigenetic factor that is important in regulating microglia response where methylation level associations with tau pathology are mainly seen in microglia‐astrocyte‐enriched cell populations. Carrillo‐Jimenez et al. found that TET2, an enzyme that can remove methylation from cystine, could regulate the proinflammatory response in mice microglia cells injected with lipopolysaccharide (LPS), and that TET2 was expressed in microglia close to amyloid β (Aβ) plaques in both human AD and 5xFAD mice brains (Carrillo‐Jimenez et al. 2019). Ma et al. (2020) conducted a staged methylome wide association study (MWAS) of AD risk and pathology. They reported four CpG sites that might attenuate the deleterious effect of APOE ε4, three of which were correlated with one another and were subsequently reduced to a single measure, PC1 (principal component 1) (Ma et al. 2020). Both PC1 and the other CpG sites were associated with the proportion of histologically‐defined activated microglia, whereas PC1 was associated with mRNA expression of innate immune genes (Ma et al. 2020). Shireby et al. obtained DNA methylation profiles of neuronal‐enriched, oligodendrocyte‐enriched, and microglia‐astrocyte‐enriched cells from human donors with low (≤ II) and high (≥ V) Braak neurofibrillary tangle (NFT) stages (Shireby et al. 2022). When compared to the neuronal population or total population, the microglia‐astrocyte‐enriched populations showed significantly greater between‐group (low vs. high Braak NFT stages) effect size in a set of 334 differentially methylated positions (DMPs) that were identified from meta‐analyzing bulk cortex tissue MWAS studies of Braak NFT stages (Shireby et al. 2022). Interestingly, one of the “bulk” DMPs was annotated to SPI1, a TF implicated in regulating AD‐associated genes in microglia (Shireby et al. 2022). Further, genes annotated to these bulk DMPs were enriched in immune and inflammatory processes (Shireby et al. 2022). Collectively, these studies suggest that some AD GWAS variants may confer risk via their effects on epigenomics mechanisms in microglia and that association studies of different epigenomics measures with AD‐related outcomes has pinpointed microglial genes as potential culprits.

3.4. Transcriptomics

Single‐cell/single‐nucleus RNA sequencing (sc/snRNAseq) studies have identified microglia subpopulations in human and/or AD mouse models that are reactive to AD pathologies, displaying similarities and discrepancies across species. Keren‐Shaul et al. uncovered disease‐associated microglia (DAM) by scRNAseq that predominantly existed in 5XFAD mouse model of amyloidosis but not wild‐type (WT) mice (Keren‐Shaul et al. 2017). Compared with homeostatic microglia, DAM express higher levels of genes including Apoe, Trem2, Lpl, Cst7 and so on, and less P2ry12, Tmem119 and Cx3cr1 (Keren‐Shaul et al. 2017). Using crosses involving mice lacking Trem2 (Trem2−/−), DAM activation was determined to be a two‐stage process, where stage I was Trem2‐independent and associated with the upregulation of ApoeE and Tyrobp, followed by stage II that was Trem2‐dependent and associated with upregulation of Trem2, Lpl and Cst7 (Keren‐Shaul et al. 2017). Further, DAM, as marked by Lpl and/or Csf1, were mainly localized near plaques in both 5XFAD mice and human AD brains (Keren‐Shaul et al. 2017). Zhou et al. confirmed the upregulation of DAM genes in 5XFAD mice vs. WT mice (Zhou et al. 2020). However, in human AD brains, gene signatures in homeostatic microglia and DAM were found to be different from those in mice (Zhou et al. 2020). In human brain AD vs. control comparison, TMEM119, P2RY12, and CX3CR1, markers of mouse homeostatic microglia, were significantly up in human AD brain donors. Many mouse DAM genes were unchanged in human AD brains or even downregulated, with the exception of those upregulated in AD including APOE, TREM2, CD68 and MHCII (Zhou et al. 2020). The discrepancy between mouse AD model and human microglia was again reported by Srinivasan et al., such that most DAM genes from mice did not show a significant coherent trend between human AD microglia (HAM) and control microglia (Srinivasan et al. 2020). Krasemann et al. identified molecular signature of disease‐associated microglia (MGnD) in multiple disease models including APP‐PS1 mice, which represented lower expression levels of homeostatic genes such as P2ry12, Tmem119, Cx3cr1, Tgfb1, among others and higher levels of inflammatory transcripts such as Clec7a, Spp1, Itgax, and Apoe (Krasemann et al. 2017). Further, in APP‐PS1 mice microglia, Apoe regulated MGnD phenotype and function, while Trem2−/− suppressed the expression of inflammatory molecules including Apoe and restored the homeostatic gene expression (Krasemann et al. 2017). Nguyen et al. identified microglia sub‐populations through snRNAseq of human AD and control brains of variable amyloid and tau (AT) pathologies, namely homeostatic, dystrophic, motile and amyloid responsive microglia (ARM) (Nguyen et al. 2020). ARM were mostly abundant in donor brains with amyloid and no tau and highly expressed CD163, whereas motile microglia were mostly enriched in amyloid and tau positive donor brains (Nguyen et al. 2020). Trajectory analysis in microglia cells implied a different response pattern from that in mice (Keren‐Shaul et al. 2017). Chen et al. identified a set of 57 plaque‐induced genes (PIGs) in App ^NL‐G‐F and C57BL/6 mice through spatial transcriptomic analysis, which were co‐expressed, highly reactive to Aβ accumulation and genotype, enriched in various pathways and in DAM or ARM genes including Apoe, Trem2, and Tyrobp (Chen et al. 2020). A vast majority (51/57) of PIGs were most highly expressed in proximity of Aβ‐plaques, influenced largely by microglia within Aβ‐plaque niches expressing Apoe and Cstl (Chen et al. 2020). In human brains, in situ sequencing confirmed that 18 of the 45 detectable PIGs were enriched in Aβ‐plaque niches, including TYROBP, C1QA, C1QB and C1QC (Chen et al. 2020). More recently, Xu et al. (2023) identified in striata two microglia populations, namely homeostatic Micr‐0 and activated Micr‐1 in AD, PD and control brains (Xu et al. 2023). Micr‐1 marker genes significantly overlapped with DAM genes in mice (Keren‐Shaul et al. 2017). But unlike DAM in mice (Keren‐Shaul et al. 2017), each diagnosis group contributed similar proportion of cells (Xu et al. 2023).

Studies of transcriptomes from carriers of mutations in AD risk genes APOE, TREM, and MS4Aautosomal dominant AD genes (ADAD), uncovered different microglial expression profiles that may play regulatory roles associated with AD. Zhou et al. reported that TREM2 R47H AD carriers expressed lower IRF8, HLA‐DRA, and AIF1 than non‐carriers, suggesting a TREM2‐dependent microglia response to AD pathology in humans (Zhou et al. 2020). Krasemann et al. reported in human AD brains, microglia from TREM2 R47H and R62H carriers did not cluster around Aβ plaques as in TREM2 ^WT, and homeostatic gene TMEM119 intensity was higher when compared to TREM2 ^WT, suggesting a TREM2‐regulated microglia phenotype switch (Krasemann et al. 2017). Nguyen et al. identified ARM as previously discussed, which was likely regulated by TREM2 and APOE, as evidenced by a reduced ARM: amyloid ratio with the increase of amyloid pathology in TREM2 R47H versus TREM2 ^WT and APOE4 versus APOE3 AD carriers (Nguyen et al. 2020). Sepulveda‐Falla et al. performed snRNAseq in different brain regions of an autosomal dominant AD PSEN1 E280A carrier, who also had protective homozygous APOE3 Christchurch variant and developed AD symptoms almost three decades past the expected age of onset (Sepulveda‐Falla et al. 2022). This donor brain displayed an unusual pattern of tau pathology with the frontal cortex being spared and occipital cortex being severely affected (Sepulveda‐Falla et al. 2022). Correspondingly, the microglial APOE expression was higher in the frontal cortex than in the occipital cortex. The set of genes that positively correlated with APOE in microglia in the frontal cortex showed a signature of acute immune response, whereas in the occipital cortex, genes related with active inflammatory processes were seen (Sepulveda‐Falla et al. 2022). Brase et al. profiled the transcriptomics of autosomal dominant AD (ADAD), APOE, TREM2, and MS4A variant carriers, versus non‐carriers (Brase et al. 2023). ADAD brains were enriched in microglia cluster with stress signature, TREM2 risk variant (p.R47H, p.R62H, or p.H157Y) carriers were enriched in a microglia cluster with reduced cellular activity, and MS4A protective variant (rs1582763‐A) carriers were enriched in a microglia cluster with proinflammatory signature (Brase et al. 2023).

Transcriptome studies of human cohorts and/or model systems uncovered additional regulatory genes, such as TYROBP and BIN1. Zhang et al. constructed co‐expression networks of transcriptome from brains of AD and control donors separately (Zhang et al. 2013). An immune network (i.e., module) was enriched in microglia markers, which also showed gain of gene–gene connectivity in AD compared to control samples and was correlated with most AD traits among all modules (Zhang et al. 2013). For this microglia module, a causal Bayesian network was constructed incorporating SNP information. TYROBP, known to interact with TREM2, was highlighted as the top causal regulator with MS4A4A, MS4A6A, and CD33 in the same group (Zhang et al. 2013). DEGs from genetically modified mice overexpressing full or truncated Tyrobp were enriched in this microglia module (Zhang et al. 2013). In Li and De Muynck (2021) TYROBP, TREM2, and OLR1 were reported to be upregulated in human AD superior temporal gyrus (STG) region. Sekiya et al. studied Drosophila expressing human TYROBP and TREM2^{wild‐type(WT)} or TREM2 ^R47H in glial and Aβ42 or tau in neuronal cells. They found that TREM2/TYROBP expression influenced both Aβ42 and tau‐related pathways (Sekiya et al. 2018; Bennett et al. 2012).

Galatro et al. reported a signature of aging from sorted microglia of postmortem human brains of donors who had intact cognition (Galatro et al. 2017). Subsequently, Srinivasan et al. identified human AD microglia (HAM) gene expression profiles from sorted microglia of brains with AD pathologies (Srinivasan et al. 2020). The DEGs between HAM and control microglia overlapped with aging DEGs in Galatro et al. Importantly, the aging DE score in HAM was significantly higher than that in microglia from control samples (Galatro et al. 2017). Using bulk, sorted bulk microglia and single cell RNAseq approaches in fresh neurosurgical brain tissue, we discovered microglial co‐expression network modules associated with age, sex and APOE (Patel et al. 2022), which are enriched for lipid and carbohydrate metabolism genes. We determined that aging‐associated microglial modules have significant overlap with pro‐inflammatory and DAM microglial clusters and harbor known AD‐risk genes including APOE, PLCG2, and BIN1. These findings delineate a link between aging and neurodegeneration involving microglial immunometabolism perturbations as a pathomechanism.

Wan et al. performed co‐expression analyses in over 2000 samples from multiple brain regions of AD and controls (Wan et al. 2020). Co‐expression modules enriched in DEGs were broadly clustered into five clusters, with cluster B enriched in microglia markers and significant overlap with signature genes from AD mice models (Wan et al. 2020). Further, two modules in this cluster were enriched in gene signatures seen in aged, wild‐type mice (Wan et al. 2020). Lopes et al. found in isolated microglia from control, AD and other neurological disorders that the genes associated with chronological aging showed overrepresentation in AD‐derived microglia signature (Srinivasan et al. 2020) and TWAS‐prioritized genes (Raj et al. 2018) for AD (MS4A6A, FCER1G, and CR1) (Lopes et al. 2022). Further, colocalization analyses revealed that AD‐associated variants (Lambert et al. 2013; Kunkle et al. 2019; Jansen et al. 2019; Marioni et al. 2018) had more colocalizing loci in each QTL dataset when compared to BD, SCZ and MS, with microglia‐specific genes including BIN1 and PYCR2 (Lopes et al. 2022).

Interplay between microglia and other cell types, tissues or organisms may play a role in AD. Moutinho et al. showed in mice that soluble TREM2 (sTREM2) species, primarily expressed in microglia, inhibited long term potentiation (LTP) induction in neurons (Moutinho et al. 2023). They discovered three TREM2 transcripts, namely TREM2³⁰⁰ (encoding full‐length protein), TREM2 (Kakkar and Lee 2008) and TREM2 (Mishra et al. 2022) (encoding sTREM2), were increased in postmortem AD brains (Moutinho et al. 2023). Further, both TREM2 (Kakkar and Lee 2008) and TREM2 (Mishra et al. 2022) were translated into proteins in human brains (Moutinho et al. 2023). Welch et al. discovered DNA double‐strand break (DSB)‐bearing neurons in the CK‐p25 mouse model of neurodegeneration and confirmed again that these neurons stimulated microglia activation (Welch et al. 2022). Importantly, in postmortem brains of AD and control donors, neurons with DSB‐bearing expression signature were identified and AD donors displayed a more prominent pattern than controls (Welch et al. 2022). Huang et al. performed snRNAseq of two brain regions from germ‐free (GF), specific pathogen free (SPF) and colonized‐GF (CGF) mice (Huang et al. 2023). The microglial transcriptome was the most modulated of all cell types in GF and SPF mice brains within the prefrontal lobe cortex and hippocampus regions. Further, most of the microglia DEGs between GF and SPF could be rescued by microbial colonization (Huang et al. 2023). These rescued genes were enriched in AD, many being DAM or PIG genes including APOE, TREM2, and C1QA (Huang et al. 2023). Cross‐species analysis of snRNAseq data revealed that microglia populations influenced by microbiota were associated with AD (Huang et al. 2023). These bulk, snRNAseq and scRNAseq studies in human and model systems revealed vast perturbations in microglial transcriptome in Alzheimer's disease and aging, demonstrated the complexity of microglial subtypes. Microglial co‐expression networks and single‐cell/nuclei clusters perturbed in AD harbor AD risk genes, further solidifying the interplay between AD genetic risk variants in microglial genes and their transcriptional perturbations as a disease mechanism. Additionally, the recapitulation of microglial transcriptional changes in some rodent models of amyloidosis suggest the potential effects of AD‐related proteostasis as one of the culprits of microglial transcriptional perturbations. It is likely that the etiology of microglial transcriptome changes in AD are multifactorial including both genetic variants in microglial genes and a response to the AD‐related proteinopathy, with contributions from aging and other factors, such as APOE genotype and sex.

3.5. Proteomics/Metabolomics/Lipidomics

Proteins/metabolites/lipids exert their function downstream of genetics, epigenetics and transcriptomics. Further, their levels could be significantly different from the corresponding mRNAs (de Sousa Abreu et al. 2009). Therefore, proteomics, metabolomics or lipidomics studies can uncover complementary information and elucidate the underlying mechanisms of AD etiology and progression. In recognition of the importance of microglia from other omics studies, many proteomics studies utilized model systems to identify potential AD‐relevant proteins and/or the underlying mechanisms in microglia, and subsequently sought supporting evidence from human tissue.

TREM2 plays an important role in the metabolic fitness and phagocytosis of AD microglia. Ulland et al. reported roles of TREM2 in maintaining metabolic fitness of microglia in AD (Ulland et al. 2017). TREM2^−/− 5XFAD mice microglia presented more LC3⁺‐denoted autophagy, which might reflect cells' failure to meet energy demand (Ulland et al. 2017). Importantly, the metabolites and mRNA levels of metabolic enzymes were different in TREM2−/− bone marrow derived macrophages compared to wildtype, highlighting the various metabolomic pathways. Further investigation suggested that TREM2 deficiency causes an impaired mTOR pathway (Ulland et al. 2017). In post‐mortem human AD brains, LC3+ microglia were significantly more abundant in TREM2 R47H or R62H AD‐risk variant carriers than in non‐carriers (Ulland et al. 2017). Nugent et al. identified important roles of TREM2 in regulating cholesterol transportation and myelin phagocytosis in mice, which were reproduced in human iPSC‐derived microglia (Nugent et al. 2020). Under chronic demyelination condition, when compared to TREM2^+/+ or TREM2^+/− mice, the transcriptomics of microglia from TREM2^−/− mice brain failed to transit to DAM‐like pattern with significant expression change of genes in cholesterol metabolism pathway (Nugent et al. 2020). Further, the lipidomics analysis revealed that TREM2 ^−/− mice accumulated cholesterol esters (CE) in bulk brain and in sorted microglia (Nugent et al. 2020). Importantly, CRISPR‐edited TREM2^−/− iPSC‐derived microglia had reduced myelin phagocytosis and increased CE accumulation (Nugent et al. 2020). Liu et al. introduced isogenic mutations to CD33, INPP5D, TREM2, and SORL1 in human embryonic stem‐cell derived microglia‐like cells (hMGLs), and performed ATAC‐seq, ChIP‐seq, RNA‐seq and proteomics (Liu et al. 2020). Compared with wild‐type (WT), the integrated‐omics networks of TREM2 ^{knock‐out(KO)} , TREM ^2R47H , SORL1 ^KO , and SORL1^A528T hMGLs all highlighted an APOE‐centered sub‐network (Liu et al. 2020). Regarding expression, APOE and several other DAM genes (SPP1, LPL) were down‐regulated in TREM2 ^KO and up‐regulated in TREM2 ^R47H , SORL1 ^KO , and SORL1 ^A528T hMGLs compared to WT, indicating that the latter hMGLs might have introduced an AD‐primed state through APOE (Liu et al. 2020). Further proteomic analysis supported that SORL1 ^R744X hMGLs increased TREM2 expression to enhance APOE expression, suggesting an interplay between SORL1 and TREM2 (Liu et al. 2020).

Several other proteins have been implicated in microglia with relevance to AD, including Moesin (MSN), ERK1, ERK2, C1Q, PROS1, and P2X4. Rayaprolu et al. identified that the MSN protein was highly abundant in human AD brains and microglia surrounding Aβ plaques of 5xFAD mouse brains (Rayaprolu et al. 2020). Further, in primary mouse microglia with Aβ or LPS treatment, silencing the Msn gene resulted in decreased Aβ phagocytosis and increased the production of pro‐inflammatory cytokines (Rayaprolu et al. 2020). According to a co‐expression study of proteins in AD patient brains, MSN was a hub protein of an immune module that was significantly associated with neuropathological burden and cognitive outcomes (Johnson et al. 2020). Chen et al. observed that levels of phosphorylated ERK1 and ERK2 increased in sorted microglia of 5xFAD mouse brain compared to WT (Chen et al. 2021). It was confirmed in a proteomics dataset of human brains that ERK1 and ERK2 were more abundant in AD samples compared to controls; and study of brain phosphoproteome in the human brain showed that ERK2 had higher phosphorylated sites in AD when compared to controls (Chen et al. 2021). Further, in primary microglia population, inhibition of ERK signaling affected expression of a group of genes enriched in DAM genes, including CCL3, SPP1, IGF1, and ITGAX (Chen et al. 2021). Inhibition of ERK signaling has also been shown to reduce phagocytic function of microglia in primary mouse cultures including a reduction of fibrillar Aβ42 phagocytosis and neuronal phagocytosis of N2a cells, a mouse neuroblastoma line (Chen et al. 2021). Dejanovic et al. showed that the deletion of complement C1q, which is highly expressed in microglia, has neuroprotective effects in Tau^P301S mice (Dejanovic et al. 2022). Subsequently, it was confirmed in human brain synaptoneurosome that complement factors C1q and C4 protein levels were much more abundant in AD samples (Dejanovic et al. 2022). Kim et al. discovered a panel of extracellularly secreted proteins in hippocampi of 5XFAD mice that were correlated with Aβ accumulation (Kim et al. 2019). Several such proteins, including PROS1, were specifically or highly expressed in microglia (Kim et al. 2019). Microgliosis stimulated by Aβ accumulation led to the increased secretion of PROS1 from microglia (Kim et al. 2019). Importantly, in human blood serum, AD group had more abundant PROS1 than mild cognitive impairment (MCI) and, control groups via western blot analysis (Kim et al. 2019). Further, serum PROS1 levels were correlated with two AD neuroimaging biomarkers: adjusted hippocampal volume and global amyloid retention (Kim et al. 2019). Thygesen et al. found that Ctsz, Clu, App, and Apoe protein levels were elevated in CNS myeloid cells of APP_SWE/PS1_ΔE9 mouse model of amyloidosis. These proteins were also elevated in hippocampi of APP_SWE/PS1_ΔE9 mice injected with LPS or vehicle (PBS) (Thygesen et al. 2018). In the neocortex of AD cases, these proteins appeared to be associated with Aβ plaques (Thygesen et al. 2018). Hua et al. discovered that in myeloid cells, the purinergic receptor P2X4 interacted with ApoE, regulating ApoE in both myeloid cells and microglia cells of APP/PS1 mice, and affecting both memory and soluble Aβ in this model (Hua et al. 2023). Further, it was coexpressed with ApoE in Aβ‐plaque‐associated microglia in cortices of human AD brains (Hua et al. 2023).

Bulk tissue proteomics of human brain and/or CSF uncovered co‐expression modules related with microglia and AD traits. Johnson et al. applied tandem mass tag (TMT) mass spectrometry method to profile proteome with deep coverage within post‐mortem brains of patients with AD, asymptomatic AD (AsymAD) and control donors (Johnson et al. 2018). Protein co‐expression module M27 was identified, which was correlated with tau burden, and was enriched in GWAS AD‐risk loci including PICALM, FERMT2, and TMEM106B, and also in glial genes especially microglial marker genes (Johnson et al. 2018). Subsequently, alternative exon‐exon junction (alt‐EEjxn) peptides were identified: these correlated with M27 eigenproteins, most of which were involved in membrane scaffolding, endosomal transport, autophagy, and protein translation (Johnson et al. 2018). Later, Johnson et al. profiled proteomics from large‐scale brain samples of control, AsymAD and AD (Johnson et al. 2020). Protein co‐expression module M4, which was strongly upregulated in AD, was enriched in astrocyte/microglia markers, in sugar metabolism pathways, in proteins encoded by AD GWAS risk genes, and in microglial anti‐inflammatory markers of mouse models (Johnson et al. 2020). In cerebrospinal fluid (CSF) analysis, M4 proteins including CD44, LDHB and PKM were elevated in AD or AsymAD, and correlated with cognitive function (Johnson et al. 2020). More recently, Johnson et al. profiled proteomics in deep coverage in large‐scale brain samples followed by co‐expression analysis (Johnson et al. 2022). They identified modules M11 cell‐extracellular matrix (ECM) interaction and M7 MAPK/metabolism to correspond to the aforementioned module M4 from their prior work (Johnson, Carter, et al. 2022).

Spatial proteomic imaging at single cell resolution has revealed microglia populations that are associated with AD pathologies in human brains. Vijayaragavan et al. developed a framework to simultaneously image 36 proteins quantitatively on FFPE samples of hippocampus region of AD and control donors (Vijayaragavan et al. 2022). They evaluated different cell types including neurons, astrocytes, vasculature and microglia, and different proteopathies including Aβ plaques and neurofibrillary tangles‐neuropil threads (NFT‐NTs) (Vijayaragavan et al. 2022). The ratio of NFT‐NTs associated microglia over proteopathy‐free microglia were increased from close to zero in control samples to around two in AD samples in the CA1 region. These microglia also expressed more markers associated with reactivity including APOE, IBA1, CD33, and CD45 (Vijayaragavan et al. 2022). A similar, but a lesser degree trend of these reactive marker expression was observed in microglia surrounding Aβ plaques when compared to microglia independent of Aβ plaques (Vijayaragavan et al. 2022). Shahidehpour et al. developed QUIVER method, applied it with nine antibody panels to FFPE samples of superior mid‐temporal gyrus to profile the spatial heterogeneity of microglia in the presence of amyloid and PHF‐1⁺ marked tau pathologies (Shahidehpour et al. 2023). Using IBA1 as a pan‐microglia/macrophage marker, five populations were discovered, with P2Y12⁺ homeostatic population being the most abundant, followed by an IBA1⁺ only population, followed by Ferritin⁺ and/or CD68⁺ reactive microglia/macrophage (Shahidehpour et al. 2023). Near PHF‐1⁺ cells or amyloid‐plaques, most microglia/macrophage cells were Ferritin⁺ and/or CD68⁺⁸⁰. Muñoz‐Castro et al. developed a protocol to phenotype astrocytes and microglia in FFPE cortex samples from human AD and control donors (Muñoz‐Castro et al. 2022). More than 6000 IBA1⁺ microglia cells were identified, with CD68, FTL, MHC2, TMEM119, and TSPO used as markers for microglia phenotyping (Muñoz‐Castro et al. 2022). Three distinct microglia phenotypes were revealed—homeostatic, intermediate that had more abundant FTL, TMEM119 and CD68 when compared to homeostatic microglia, and reactive, that showed increased levels of all markers (Muñoz‐Castro et al. 2022). In AD samples, reactive type was the largest microglia population followed by intermediate and then by homeostatic, whereas the trend was reversed in control samples (Muñoz‐Castro et al. 2022). Reactive microglia were relatively close in proximity to Aβ‐plaques or PHF1⁺ NFTs, followed by intermediate type and then by homeostatic (Muñoz‐Castro et al. 2022).

The change of microglia proteome in AD pathologies is also investigated in non‐brain tissue, for example, retina, and extracellular vesicles (EVs) in the following studies. Koronyo‐Hamaoui et al. investigated the pathological features of AD, MCI and cognitively unimpaired controls in postmortem retina and brains (Koronyo et al. 2023). Histopathological and microscopic analyses identified that Aβ₄₂ burden significantly increased in MCI and AD in superior temporal (ST) and inferior temporal (IT) peripheral retinal regions, along IBA1⁺‐microgliosis and GFAP⁺‐macrogliosis (Koronyo et al. 2023). Further, novel intraneuronal Aβ oligomers were found to increase in MCI and AD in retina (Koronyo et al. 2023). Intriguingly, although AD retina demonstrated microgliosis, the capability of microglia phagocytosis of Aβ in retina was reduced (Koronyo et al. 2023). Finally, a proteomics study performed in the retina and human temporal cortex tissue of AD and controls, elucidated shared differentially expressed proteins (cell death‐, inflammatory‐, mitochondria‐ or signaling‐ related) in brain and retina (Koronyo et al. 2023). Cohn et al. performed proteomics, lipidomics and miRNA studies on small microglia EVs isolated from cryopreserved human parietal cortex of AD and controls with normal/low pathology (NL) (Cohn et al. 2021). In AD, P2RY12, and TMEM119, which mark homeostatic microglia, were significantly reduced, whereas FTH1 and TREM2, marking DAM, were increased based on either proteomics data or immunoblotting (Cohn et al. 2021). Regarding lipids, free cholesterol was increased in AD, whereas levels of docosahexaenoic acid (DHA)‐containing polyunsaturated lipids were decreased in AD EVs from microglia (Cohn et al. 2021). Mallach et al. investigated the proteome profiles of exosomes secreted from human iPSC‐derived microglia (iPS‐Mg) with TREM2^R47H,het or the common variant (CV), and demonstrated microglia‐secreted exosomes responded to different stimuli and influenced the downstream functions of neurons and microglia (Mallach, Gobom, Zetterber, et al. 2021; Mallach et al. 2021b). Compared to CV, exosomes of iPS‐Mg with TREM2^R47H,het exhibited proteome profiles more similar to those stimulated with LPS which induced an inflammatory signal. Further, the latter contained more abundant DAM signature proteins and were less able to promote the outgrowth of neuronal processes and increase mitochondrial metabolism in neurons (Mallach, Gobom, Zetterber, et al. 2021; Mallach et al. 2021b). Taken together, proteome/lipidome/metabolome studies in bulk or isolated microglia from humans and complementary studies in mouse and induced microglia‐like cells demonstrate perturbed proteome networks, lipids and metabolites; highlight complex interactions with other brain cell types and enrichment in biological processes beyond immunity, such as cell signaling and extracellular matrix.

3.6. Summary

Genetic studies have revealed a number of AD risk variants in genes with elevated expression in microglia compared to other brain cell types, such as TREM2, CD33, and SORL1. Many of these risk variants are implicated in immunity and inflammation‐based pathways, and associated with AD‐related outcomes such as hippocampal volume and brain amyloid or tau burden. eQTL studies have identified gene expression associations with AD associated microglia‐enriched genes. Epigenetic studies have shown enhancer, promoter and open chromatin regions in microglia to be associated with AD heritability, and TET2, an enzyme that can remove methylation, regulated pro‐inflammatory response in microglia of mice with LPS stimulus. Transcriptomic bulk and sc/snRNAseq studies demonstrated microglial subpopulations in humans and AD mouse models with varying associations with AD pathology. Microglia‐specific profiles of AD risk genes such as APOE and TREM have been identified. Proteomic studies demonstrated AD associated proteins and protein networks enriched in microglial markers and implicated in multiple pathways including extracellular matrix and cell signaling. Lipidome/metabolome studies, while less abundant (Figure 1), also reveal perturbations (Figure 3). Spatial proteomic imaging have shown amyloid plaque associated microglial proliferation. Hence, there is abundant multi‐omics evidence of microglial perturbations in AD that may both be causal (e.g., driven by AD risk genes) or a consequence (e.g., driven by pathology) in the AD etiopathogenesis. Given the heterogeneity of microglial subtypes, their interactions with other brain cells, their varying omics profiles with age, genetics, external stimuli, studies that focus on refining these profiles longitudinally in peripheral human tissue and model systems, that contextualize these omics changes vis‐à‐vis pathology and other factors and that delineate their interactions with those from other cell types are necessary. Such studies should aim to characterize the complex molecular landscape of microglia temporally and spatially in both AD and healthy aging, while seeking validations and prioritization of key molecules in appropriate model systems.

FIGURE 3.

Alluvial plot of shared and distinct biodomains nominated through the different omic lenses for each reviewed glia cell type.

4. Oligodendrocytes

4.1. Introduction

Oligodendrocytes are involved in myelination and synaptic maintenance. Omics based studies on oligodendrocytes in the AD field are relatively limited in comparison to those of microglia and astrocytes (Figures 1 and 2). Genomic studies have identified variants within genes specific to oligodendrocytes, while epigenetic studies have found associations between AD risk variants and open chromatin regions in this cell type. The most frequently published omics types for oligodendrocytes are transcriptome followed by proteome studies, while metabolome/lipidome studies are lacking and therefore an area of opportunity, especially given the abundant energy demand and lipid content of myelin.

4.2. Genomics and Epigenomics

Although many studies assess the genetic and epigenetic architectures of AD, relatively less is known about the multi‐omics signatures of oligodendrocytes compared to the other glia types (Figure 3). To contextualize the leading SNPs found to be associated with age of AD onset, He et al. (2021) performed cell‐type‐specific eQTL analysis identifying significant associations with expression of LRRC37A2, LRRC37A3, and KANSL1 in oligodendrocytes but not in microglia. Interestingly, LRRC37A genes and KANSL1 reside on or near a large polymorphic inversion on chromosome 17 encompassing the microtubule‐associated protein tau (MAPT) gene (Boettger et al. 2012), haplotypes of which are associated with AD and primary tauopathies such as progressive supranuclear palsy (PSP) (Allen, Burgess, et al. 2016).

To interrogate the epigenetic changes in AD for a specific cell population, fluorescence‐activated nuclear sorting (FANS) with cell markers that are specific to or absent from the cell type of interest have been used. For example, Shireby et al. used SOX10 as a selection marker to obtain an oligodendrocyte‐enriched nuclei population from dorsolateral prefrontal cortex (DLPFC) tissue of 21–28 donors (Shireby et al. 2022). Comparison of the 334 differentially methylated positions (DMPs) identified from their bulk brain tissue analyses of 2013 donors associated with AD pathology to those from the NeuN+ (neuronal‐enriched), SOX10+ (oligodendrocyte‐enriched) and NeuN−/SOX10− (microglia‐ and astrocyte‐enriched) populations revealed that most of the bulk DMPs reflect non‐neuronal rather than neuronal DNA methylation (DNAm) changes (Shireby et al. 2022). These DMPs were often hypermethylated with increased pathology, including Braak, Thal, and CERAD. They also showed that the effect sizes of the DMP associations are smaller in the SOX10+ cell population than those in the NeuN−/SOX10− population suggesting that although many of the identified bulk DMPs may be driven by microglia/astrocyte changes, there are likely also oligodendrocyte specific DNAm changes related to AD pathology (Shireby et al. 2022). Conversely, cell‐type specific markers have also been employed to exclude certain cell populations and enrich for others. Ramamurthy et al. sorted 16 hippocampal and 10 DLPFC tissue samples from 19 brain donors using NeuN and Pu.1 as neuronal and microglial markers, respectively, for H3K27ac ChIP‐seq (Ramamurthy et al. 2022). The double negative population was entitled oligodendrocyte‐enriched glia (OEG), as they were found to be hyperacetylated (suggesting transcriptional activation) in the promoter regions of oligodendrocyte‐specific genes, including OLIG2. This study, albeit limited in size, determined that OEGs had the largest H3K27ac changes associated with brain Aβ load, although this was both sex‐ and brain‐region dependent. In 8 hippocampal samples from female donors, there was hypoacetylation of AD risk and myelination genes in the presence of Aβ, consistent with downregulation of myelination transcriptional networks in AD that we (Allen et al. 2018) and others (McKenzie et al. 2017) previously reported from bulk brain tissue from both sexes. In contrast, Ramamurthy et al. (2022) found hyperacetylation of myelination and AD risk genes in 10 DLPFC samples from both sexes, which may either represent region‐ or sex‐specific differences or simply false‐positive findings due to lack of power. Nonetheless, integration with AD risk GWAS data revealed co‐localization of AD risk SNPs for some of the loci, including BIN1 and PSEN1 to overlap with H3K27ac peaks enriched in OEGs. This is consistent with BIN1 and PSEN1 being members of myelination transcriptional networks discovered in AD (Allen et al. 2018) and other neurodegenerative disease affected patient brains.

Lastly, Morabito et al. (2021) performed snATACseq and snRNAseq using prefrontal cortex tissues from 12 AD and 8 control brains. They were able to identify the oligodendrocyte population based on gene expression and chromatin accessibility of known cell‐type marker genes. Subclustering of the oligodendrocytes highlighted an immune oligodendrocyte population with an increased proportion in AD compared to control. Importantly, co‐accessibility analysis showed that open chromatin regions overlap with cell type markers and AD DEGs, indicating that the cis‐regulatory elements are acting in a cell‐specific manner. This study highlighted NRF1 and SREBF1, two transcription factors with differential snATACseq and snRNAseq findings in AD compared to controls and their target genes as potential oligodendrocyte‐specific disease‐related molecules and pathways in AD. Their integrative multi‐omics analyses including pseudotime revealed SREBF1 to be an oligodendrocyte transcriptional activator, which is itself downregulated in AD, as well as its targets. This study is consistent with others demonstrating downregulation of myelination genes in AD (Allen et al. 2018; McKenzie et al. 2017; Wang, Allen, et al. 2020), and adds NRF1, SREBF1, and their targets to the growing list of potential AD risk genes involved in myelination pathways.

4.3. Transcriptomics

4.3.1. Bulk Transcriptome

To understand the oligodendrocytic gene expression changes in AD from bulk brain transcriptome data, several studies have applied cell‐type annotation, enrichment and deconvolution approaches. In 60 temporal lobe samples, Barbash et al. (2017) reported a decreased expression of transcripts known to have proteins exclusively expressed in oligodendrocytes in brain donors who were non‐demented but had early AD pathology compared to controls. However, no oligodendrocyte‐specific expression differences were detected between control and AD brains from demented donors with advanced pathology. Gene expression‐network‐based methods have also been used to understand the involvement of oligodendrocytes in the pathophysiology of AD. We (Allen et al. 2018) and others McKenzie et al. (2017) identified down‐regulation of eigengene expression in AD within co‐expression network modules that are enriched with oligodendrocyte genes involved in myelination, including PLP1, PLLP, CNP, and MOBP. Collectively, these two studies comprise 637 AD and 249 non‐AD donors spanning 4 different brain regions of temporal cortex (TCX), cerebellum, DLPFC and visual cortex highlighting their power as well as independent reproducibility of these findings. Applying various gene expression module approaches to the three Accelerating Medicines Partnership in Alzheimer's Disease (AMP AD) cohorts (Allen, Carrasquillo, et al. 2016; De Jager et al. 2018; Wang et al. 2018) (Mayo Clinic, Mount Sinai, Rush ROS‐MAP) comprising > 1100 brains, modules enriched in oligodendrocyte and/or oligodendrocyte progenitor cell (OPC) genes have been identified (Wan et al. 2020; Morabito et al. 2020; Wang, Chen, et al. 2022), although their direction of association with AD were not consistent with prior studies from the individual cohorts (Allen et al. 2018; McKenzie et al. 2017). This may be due to differences in the analytic approaches, the datasets or a combination thereof that impact module compositions, enrichments and/or the direction of associations with AD.

Using analytic deconvolution of bulk transcriptome data to brain cell‐type specific expression measures, we (Wang, Allen, et al. 2020) and others (Bordone and Barbosa‐Morais 2020; McKenzie et al. 2018; Patrick et al. 2020) estimated proportions of specific brain cell types, with some studies also subsequently performing cell‐specific or cell‐intrinsic differential gene expression analyses. These studies use available brain single cell/single nucleus RNA sequencing (sc/snRNAseq) data as reference for cell‐type annotations; digital sorting algorithms such as DSA (Zhong et al. 2013) or CIBERSORTx (Newman et al. 2019) for proportion estimations and deconvolution methods including CellCODE (Chikina, Zaslavsky, and Sealfon 2015), PSEA (Kuhn et al. 2011) or WLC (Wang, Allen, et al. 2020). Oligodendrocyte proportions were estimated to be higher in AD compared to control brains in the AMP AD bulk transcriptome data from Rush DLPFC and Mount Sinai TCX data (Wang, Allen, et al. 2020) but not significantly different in the Mayo Clinic TCX data (Wang, Allen, et al. 2020; Bordone and Barbosa‐Morais 2020), underscoring dataset‐specific differences. Differential gene expression analyses for oligodendrocytes revealed both downregulated (Wang, Allen, et al. 2020) and upregulated (Wang, Allen, et al. 2020; Bordone and Barbosa‐Morais 2020) genes in AD. Importantly, enrichment of downregulated oligodendrocyte genes with myelination (Allen et al. 2018; McKenzie et al. 2017) and upregulated ones with ceramide/sphingomyelin (Filippov et al. 2012; Czubowicz et al. 2019; Toledo et al. 2017) biological terms are consistent with known myelin biology and reveal the complex transcriptomic changes that occur in oligodendrocytes in AD. Unraveling this complexity requires multiple molecular profiling and analytic approaches such as deconvolution, sn/scRNAseq and spatial transcriptomics approaches discussed below.

4.3.2. Single‐Nucleus RNAseq

A powerful approach facilitating our understanding of oligodendrocyte‐specific gene expression changes at a single‐cell resolution is snRNAseq. Sadick et al. profiled the gene expression landscape in AD using snRNAseq finding robust oligodendrocyte populations (Sadick et al. 2022). Although demyelination in AD is well‐established, Sadick et al. reported distinct gene expression changes in specific clusters of oligodendrocytes in AD, involving either up‐regulation or down‐regulation of genes related to synaptic maintenance (Sadick et al. 2022). This dichotomy is also supported by other snRNAseq studies (Zhou et al. 2020; Lau et al. 2020; Belonwu et al. 2022). It is possible that some oligodendrocytes are compensating for the loss of myelin by increasing their metabolism to accelerate remyelination and axon ensheathment. At the same time, the down‐regulation of metabolism could be a reflection of damaged oligodendrocytes or of myelination no longer being needed as neuronal axons degenerate. Alternatively, cellular senescence may be playing a role as related senescence and aging genes have been found upregulated in AD oligodendrocytes compared to healthy controls (Zhao, Xie, and Liu 2022). Looking at oligodendrocytes as a whole, there are also data showing changes in oligodendrocyte gene expression as AD progresses (Morabito et al. 2021; Wang, Chen, et al. 2022).

Brase et al. generated oligodendrocyte expression profiles of ADAD and TREM2 risk variant carriers showing both up‐regulated and down‐regulated DEGs consistent with other snRNAseq studies (Brase et al. 2023). Spliceosome‐related genes also associated with ADAD, indicating potential involvement of splicing dysfunction (Brase et al. 2023). Oligodendrocytes enriched with TFEB expression, a gene known to play roles in autophagy and lysosomal pathways and is involved in myelination, were implicated in TREM2 risk variant carriers (Brase et al. 2023) suggesting possible immune‐oligodendrocyte crosstalk. Another study from Zhou et al. investigating snRNAseq data from 5xFAD mouse amyloidosis models with TREM2 risk variants identified a unique AD‐reactive oligodendrocyte population (Zhou et al. 2020). Although the expression signature of disease‐associated oligodendrocytes (DOL) in mouse models of AD is characterized by Serpina3 and C4b (Kenigsbuch et al. 2022; Pandey et al. 2022; Park et al. 2023), their human orthologs are primarily expressed in astrocytes.

4.3.3. Spatial Transcriptomics

In contrast to bulk and single cell transcriptomics, limited work has been conducted to assess the spatial expression patterns of oligodendrocytes. Using APP transgenic mice, Chen et al. identified region‐specific correlations between oligodendrocyte genes and Aβ deposition in both the early and late stages of AD as defined by amyloid burden in this rodent model (Chen et al. 2020). Following validation in human tissue, they found that the cellular niches around amyloid plaques are depleted of oligodendrocytes, suggesting a toxic effect of Aβ (Chen et al. 2020). Chen et al. used 10× Visium platform to profile gene expression in the middle temporal gyrus (MTG) of AD donor brains (Chen et al. 2022). They identified brain layer and cell type‐specific changes including oligodendrocyte gene dysregulation proximal to Aβ plaques and neurofibrillary tangles (Chen et al. 2022). Together, the results of these transcriptomics studies highlight the complex gene expression perturbations of oligodendrocytes in AD.

4.4. Proteomics

Recently, several proteomic datasets have been generated at bulk tissue resolution from AD brains (Figure 1). Similar to bulk RNAseq analyses, deconvolution and cell proportion estimations were performed to understand AD‐related changes in cellular composition. Congruent to some of the aforementioned RNAseq data (Wang, Allen, et al. 2020; Bordone and Barbosa‐Morais 2020), there were no oligodendrocyte proportion differences between AD and control regardless of which algorithm was used (Johnson et al. 2018; Miedema et al. 2022). Application of WGCNA identified a module enriched with oligodendrocyte proteins, however, no association between this module eigenprotein and AD pathology was detected despite the module being enriched with AD GWAS risk genes (Seyfried et al. 2017). The lack of oligodendrocyte protein module associations with AD this group and others have reported (Johnson et al. 2018; Swarup et al. 2020; Higginbotham et al. 2019) is inconsistent with the previously discussed robust oligodendrocyte transcriptional network associations with AD (Allen et al. 2018; McKenzie et al. 2017). This suggests that different omics modalities can capture complementary cellular perturbations. In the case of oligodendrocyte proteome, disease‐specific perturbations may not be detectable potentially owing to higher stability of this proteome, ongoing remyelination despite damaged oligodendrocytes or a combination of these factors. Nevertheless, a brain region‐specific proteome analysis identified a module enriched with oligodendrocyte genes and validated their downregulation in AD in the entorhinal cortex (Gao et al. 2022), suggesting that the earliest affected regions in AD may involve oligodendrocyte protein changes. Further, a CSF proteome study identified elevation of myelination network genes in AD in both the CSF and brain (Higginbotham et al. 2020), highlighting oligodendrocytic protein perturbations in AD brains that are also reflected in CSF.

Oligodendrocyte specific expression changes were also assessed in different APOE genotype carriers and also in relation to aging. Although there are no differences in oligodendrocyte proportions between different APOE allele carriers, the expression patterns of the oligodendrocyte module genes are different between E2, E3, and E4 AD cases and E3 controls (Dai et al. 2018). Interestingly, in healthy participants, the levels of oligodendrocyte total proteins and phosphorylated proteins increase over age (Andres‐Benito et al. 2023), which must be taken into account when evaluating proteome studies for this cell type.

4.5. Summary

Relatively fewer omics‐based studies have been undertaken with a focus on oligodendrocytes (Figures 1 and 2). Downregulation of transcripts from oligodendrocyte specific genes involved in myelination is observed. Age of onset associated AD SNPs were found in oligodendrocyte‐specific/enriched genes. SnRNAseq has revealed perturbation of gene regulation in oligodendrocytes where cellular subtypes are either up‐ or down‐regulated in AD. FANS sorting has been used to enrich for oligodendrocytes and DNA methylation profiling has discovered most differentially methylated positions to be hypermethylated with high AD pathology. Further studies are required to establish understanding regarding sex and brain region associated histone acetylation patterns related to myelination in AD brains. Spatial transcriptomic studies have observed regions with amyloid aggregates lacking oligodendrocytes and dysregulation of related genes in these regions. While some proteomic based studies have not observed oligodendrocyte protein network expression changes in AD brains, others demonstrate such changes in CSF and/or brain though the direction may differ from those observed for transcripts. In summary, there is genetic, epigenetic and transcriptome evidence of oligodendrocyte involvement in AD pathogenesis, with emerging evidence also observed in proteome studies, though this requires further confirmation and may be tissue‐type, brain region or disease stage related. Importantly, there is a critical knowledge gap regarding metabolome/lipidome changes in oligodendrocytes as these cells regulate the high‐energy and lipid demanding myelination processes essential for neuronal function.

5. Astrocytes

5.1. Introduction

It is well documented in AD literature that the relative abundance of astrocyte populations increase and are correlated with amyloid and tau burden from measured neuropathology (Johnson et al. 2018). This and other lines of evidence suggest that astrocytes play an important role in both AD pathophysiology and aging, including maintenance of the blood brain barrier (BBB) (Is et al. 2024), cell death (Guttenplan et al. 2021) and immunity (Sekar et al. 2015; Lian et al. 2016; Liddelow et al. 2017) especially through its interactions with other CNS cell types such as neurons, pericytes and microglia. To date, omics studies have revealed astrocytic substates and multiple roles of astrocytes in AD including in cell activation, inflammation, and APOE and amyloid related pathways (Figure 3).

5.2. Genomics

Two relatively recent studies employing a GWAS approach of AD related endophenotypes identified significant genetic variants near astrocytic genes that may be important in AD through their impact on the endophenotype. The first study by He et al. investigated genetic variants that associated with age of AD onset finding two genome‐wide significant variants, rs56201815 in the ERN1 gene and rs12373123 in the SPPL2C gene at the MAPT locus, a known AD risk locus (He et al. 2021; Allen et al. 2014). Interestingly, when integrating with cell‐type specific gene expression data from the Religious Orders Study and Rush Memory and Aging Project (ROSMAP) cohort, they found rs12373123 associated with increased levels of MAPT, LRRC37A2, LRRC37A3, and KANSL1 expression in astrocytes as well as other cell types while rs56201815 associated with expression of ERN1 in astrocytes and other cell types (He et al. 2021). In their gene‐set enrichment analysis, they also found high enrichment in pathways related to regulation of astrocyte activation suggesting that regulating astrocyte activation is likely important to AD age of onset. In the second AD related endophenotype study, we reported (Oatman et al. 2023) a genetic variant, rs9890231 within the ITGB4 gene, that significantly associated with levels of soluble Aβ40 in the temporal cortex of AD patients. Using single cell RNAseq data from Mathys et al. (2019), we observed upregulation of ITGB4 expression in astrocytes of AD patients compared to controls (Oatman et al. 2023).

Although GWAS approaches derive associations, they do not identify causation and so additional experiments are required to determine causality or functionality of associated genetic variants and/or their nearby genes. Lee et al. (2022), through a genome‐wide, gene interaction analysis identified a significant interaction between genetic variation at the FMNL2 locus with cerebrovascular risk factors that modifies AD risk. They further found FMNL2 expression was upregulated in patient brains with infarcts and AD pathology as well as increased FMNL2 protein in astrocytes among individuals with cerebrovascular pathology (Lee et al. 2022). Subsequently, Lee et al. investigated functional relevance of fmnl2 in zebrafish models treated with amyloid as well as APP/PS1dE9 AD mouse model of amyloidosis finding that the cerebrovascular pathology (aggravated by cardiovascular risk factors), likely interacts with FMNL2 to alter normal astroglial‐vascular mechanisms, impairing the ability of the brain to clear excess amyloid and tau through the BBB (Lee et al. 2022). Another study investigated the potential gene expression changes of TMEM106B in the AD brain (Satoh et al. 2014) as this gene was previously identified by AD GWAS and is known to be important in the development of FTLD‐TDP (Debaisieux and Schiavo 2014). They identified TMEM106B expression in reactive astrocytes as well as other cell types in both AD and non‐AD brains (Satoh et al. 2014). Similarly, Pan et al. investigated the potential causal role of circulating blood levels of S100B, an astrocytic marker, in AD as well as other neuropsychiatric and neurological diseases (Pan et al. 2023). Through Mendelian randomization, Pan et al. found that there was no evidence of circulating S100B levels having a causal effect on AD (Pan et al. 2023). In summary, genetic studies discovered associations between AD‐related endophenotypes and variants near genes that have altered expression in astrocytes.

5.3. Epigenomics

There is emerging evidence of DNAm variability linked to changes in AD‐related phenotypes including those that are potentially specific to astrocytes (Tulloch et al. 2018; Phipps et al. 2016; Coppieters et al. 2014; Neal and Richardson 2018). Nonetheless, the application of genome‐wide methylomic approaches to identify astrocyte specific DNAm changes or sub‐state specific changes have been extremely limited or non‐existent. Most studies investigate DNAm levels in bulk tissue with various attempts to correct for cell type composition, but the lack of large, well‐defined reference panels for cell type specific DNAm makes this challenging. These studies often rely on statistically adjusting for surrogate cell type specific variables like gene expression levels of GFAP for astrocytes, leaving open questions about the driving factors behind significant DNAm associations (De Jager et al. 2014; Silva et al. 2022). There are more recent efforts to advance this field by sorting cells into neuronal (NeuN+) or glial (NeuN−) (Gasparoni et al. 2018; Guintivano, Aryee, and Kaminsky 2013) cell types, such as oligodendrocyte (SOX10+), or microglia/astrocyte (NeuN−/SOX10−) (Shireby et al. 2022) enriched populations. These have then been used to investigate enriched cell type specific methylation changes in relation to AD, Braak stage or have used a cell type specific reference (Li, Sun, and Wang 2020). Interestingly, these studies suggest that DNAm variation occurs in large part within glial cells when compared to neuronal populations within the context of AD, but specific astrocytic variations often remain elusive with some exceptions. Gasparoni et al. (2018) identified differentially methylated CpG sites associated with Braak stage at the S100B locus which is an astrocytic marker and at the GNG7 locus of which the expression has been found to be dysregulated in mouse astrocyte cultures upon inflammatory stimuli. Beyond cell type specific methylation, there have been studies functionally characterizing bulk epigenome‐wide association study (EWAS) differentially methylated hits in cell type specific gene expression data. For example, Mastroeni et al. (2017) took the AD EWAS differentially methylated positions identified by De Jager et al. (2014) and looked at differential expression between AD cases and matched controls of nearby genes in laser capture microdissections of brain tissue including that of GFAP‐positive astrocytes. Interestingly, although they did not find astrocyte specific differential expression of ANK1, a top differential methylated locus in AD EWAS (De Jager et al. 2014), they did find differential gene expression of ANK1‐like genes in astrocytes including ANKRD36, ANKRD52, and ANKRD18CP. These results suggested that genes containing an ankyrin repeat domain may be important for brain astrocytes in AD, but there remains a significant gap in knowledge regarding specific functions of DNAm in astrocytes.

There are even fewer human omics studies investigating other epigenetic modifications in astrocytes within the context of AD. Some of the first studies to investigate cell type specific chromatin changes mainly relied on pre‐sorting cells or nuclei into specific cell type populations then performing sequencing on these cell type specific populations. One commonly used marker for neuronal cell types is NeuN, which has been used to sort cell populations into neuronal or glial cells. Barrera et al. (2021) used FANS with NeuN to sort nuclei and glial populations to then perform ATAC‐seq and snRNAseq in each population separately. They identified numerous AD specific chromatin changes in the neuronal population but did not identify significant changes in the glial cell population until they stratified by sex where they then identified multiple female specific AD related chromatin changes including those at known AD risk loci like APOE (Barrera et al. 2021). Although their sample size was relatively small, the authors took advantage of a tiered replication approach and compared the known AD risk loci to their snRNAseq data finding that, although not significant, many of the genes in these loci showed gene expression patterns predicted by the chromatin data, including the glial chromatin data to gene expression in the astrocyte annotated clusters. Interestingly, Barrera et al. (2021) also found decreased chromatin accessibility at the APOE locus in the female AD glial populations. These results suggest that there are AD and sex specific changes in glial cell chromatin accessibility translating into gene expression changes, including at the APOE locus. As a result, since astrocytes are the main producers of APOE in the brain, additional work needs to be done to determine if these are astrocyte specific results, and if this difference in APOE accessibility may explain some of the differences in AD risk between males and females. Morabito et al. (2021) published one of the first studies to integrate snATACseq and snRNAseq data generated from the same late‐stage AD and age‐matched cognitively healthy control post‐mortem prefrontal cortex tissue samples. Although the sample size was small (12 AD and 8 control individuals), they identified distinct cell type specific clusters including astrocytic clusters marked by higher GFAP expression and enrichment of open chromatin enrichment around the GFAP promoter. Interestingly, Morabito et al. found that proportions of GFAP ^high/CHI3L ⁺ astrocyte clusters increased in AD compared to controls while GFAP ^low/WIF1 ⁺/ADAMTS17 ⁺ astrocyte cluster proportions decreased (Morabito et al. 2021), which was consistent with previous AD mouse models (Habib et al. 2020). Using this disease associated astrocyte gene signature and pseudotime analysis, their data also suggest GFAP ⁺ increases with AD and that the astrocytic state may be regulated towards an activated state by expression of the FOSL2 transcription factor vs. a more homeostatic state by CTCF (Morabito et al. 2021). In sum, while bulk tissue or sorted brain cell studies revealed epigenetic changes in astrocytes in AD, where some of these alterations also have corresponding gene expression correlates, there is a relative paucity of well‐powered epigenetic studies exploring astrocyte‐specific perturbations.

5.4. Transcriptomics

5.4.1. Substate Signatures

Compared to other omics measures, the most utilized technique to evaluate astrocytic states in AD is transcriptomics (Figure 1). In the last decade, numerous transcriptomics studies have demonstrated the presence of multiple astrocytic states across multiple brain regions, including nominating gene signatures characterizing astrocytic subpopulations and their contributions to AD pathophysiology. In general, there are two broad astrocytic states that have been identified through transcriptomics: a homeostatic state, previously termed as A1, and an activated state, termed A2. In the last few years, however, these two states have been found to most likely be an extremely reductionist categorization of multiple important distinct astrocytic states within the brain during the progression of AD. First, identifying markers of activated astrocytic states has been an active area of research because of their significant population increases in AD, their association with multiple AD related phenotypes, and interactions with microglia and neurons. Bulk RNAseq studies have identified modules of co‐expressed genes enriched in astrocytic markers that associate with increased pathology measures and decreased cognition, suggesting that astrocytes play a role in AD severity and disease progression (Morabito et al. 2020; Wang, Chen, et al. 2022). These modules are often enriched with immune related signaling pathways indicating an early activation or reactivity of astrogliosis in AD (Morabito et al. 2020; Wang, Chen, et al. 2022; Sekar et al. 2015) that may also influence immunity possibly through astrocyte interactions with microglia (Lian et al. 2016; Liddelow et al. 2017) or its role in the BBB (Is et al. 2024). Li et al. nominated the astrocytic gene CDK2AP1 as a key gene strongly associated with cognitive and neuropathology AD measures (Li and De Muynck 2021). A functional gene network analysis built with bulk transcriptomic and epigenomic data to identify top genes with cell‐specific differential expression discovered astrocytic transcript associations with AD pathology and dementia measures for genes including HLA‐C, ITGA9, FOS, CXCR4, ACTB, PLCB1, and SOD1 (Lin et al. 2022). It should be noted, however, that these four studies (Li and De Muynck 2021; Morabito et al. 2020; Wang, Chen, et al. 2022; Lin et al. 2022) all leveraged the same AMP AD datasets (Mayo (Allen, Carrasquillo, et al. 2016), Mount Sinai (Wang et al. 2018), and ROSMAP (De Jager et al. 2018)) in their discovery or replication analyses.

Beyond bulk RNAseq, single cell and single nuclei RNAseq have been vital in identifying transcriptomic signatures for astrocytes in AD. In these single cell/nuclei studies, authors rely on enrichment of cell type specific gene expression markers to annotate clusters to cell types. Astrocytic cluster markers often used in human datasets to initially annotate astrocytic clusters include AQP4 (Zhou et al. 2020; Xu et al. 2023; Sadick et al. 2022; Lau et al. 2020; Mathys et al. 2019; Qian et al. 2023), SLC1A2 (Xu et al. 2023; Lau et al. 2020; Qian et al. 2023), GFAP (Zhou et al. 2020; Sadick et al. 2022; Lau et al. 2020; Qian et al. 2023), GJA1 (Sadick et al. 2022), ALDOC (Sadick et al. 2022), CLU (Sadick et al. 2022), ALDH1L1 (Sadick et al. 2022), SLC1A3 (Sadick et al. 2022), GLUL (Sadick et al. 2022), ADGRV1 (Lau et al. 2020; Qian et al. 2023), GPC5 (Lau et al. 2020), and RYR3 (Lau et al. 2020). The first snRNAseq study of human AD and control brain tissue was performed by Mathys et al. (2019) with 48 samples (24 AD cases and 24 no pathology controls) from the ROSMAP cohort finding GFAP, SYTL4, MT1E, ZFP36L1, MT2A, and GJA1 significantly upregulated in AD vs. control astrocytes and genes APOE, STMN1, CIRBP, TMEM241, CABLES1, PREX2, GRM3, KCNIP4, RLBP1, SMURF2 downregulated. Mathys et al. also found consistency between astrocytic gene expression changes and bulk RNAseq gene expression changes. They then subclustered the astrocytes finding 2 distinct sub‐clusters, Ast0 characterized by an association with lower pathology and better cognition, and Ast1 characterized by an association with higher pathology, worse cognition, and higher expression of GLUL and CLU suggesting a reactive astrocyte cluster (Mathys et al. 2019). Interestingly, there was also a sex association where Ast0 was enriched with male cells while Ast1 was enriched with female cells, although this was not as striking as other cell type specific clusters (Mathys et al. 2019). Subsequent to the Mathys et al. article, many other groups also published single cell/nuclei data from human AD cases (Zhou et al. 2020; Xu et al. 2023; Sadick et al. 2022; Lau et al. 2020). Similarly, they found distinct substates in their astrocytic clusters including the broad homeostatic and activated astrocytic states but often additional distinct subpopulations within the activated astrocytic state after reclustering (Zhou et al. 2020; Xu et al. 2023; Sadick et al. 2022; Lau et al. 2020; Qian et al. 2023). Xu et al. (2023) found 2 activated astrocyte states in the putamen, entorhinal cortex, and prefrontal cortex with Ast‐1 characterized by higher expression of APOE, CKB, CST3, FTL, SPARCL1, ITM2C, CPE, TUBB2B, VIM, ATP1B2, GLUL, PSAP, GJA1, AGT, ENO1, GPR37L1, CLU, and APOC1 while Ast‐2 was characterized by higher expression of DPP10, GFAP, CD44, PLEKHA5, VCAN, KAZN, C3, SYNM, TNC, and MAOB. Zhou et al. (2020) identified a total of 6 astrocytic subclusters, Lau et al. (2020) and Sadick et al. (2022) identified 9, while Qian et al. (2023) identified 8 after integrating data from 15 previous studies. Interestingly, there has been recent evidence suggesting that senescent astrocytes may play a role in AD, however, expression signatures for senescent astrocytes have not been described by current sn/sc‐transcriptome studies in AD. This may be due to these populations being relatively small and therefore hard to detect using current techniques or difficulty in using gene expression signatures to identify senescence, though this is still an active area of research. Taken together, the application of single cell/nuclei techniques to understand astrocytic mechanisms in AD have proven invaluable and future integration of these datasets through interfaces like The Alzheimer's Cell Atlas (Zhou et al. 2022) will be imperative for identification of additional transcriptomic mechanisms/signatures. These studies provide important evidence of the heterogeneity within astrocytic cells indicating that they may have different molecular roles in both the healthy brain and in Alzheimer's disease.

5.4.2. Deep Endophenotyping to Identify Astrocytic Mechanisms

5.4.2.1. Identification of Astrocyte Activation Mechanisms

There has been substantial progress in identifying mechanisms that may lead to activation of astrocytes during neurodegeneration through transcriptomic analyses. Although complement component 1q (C1q) and its subunits (C1QA, C1QB, and C1QC) have been used as markers of activated microglia (Zhou et al. 2020; Xu et al. 2023; Lau et al. 2020; Mathys et al. 2019), there is evidence that C1q, along with IL‐1α and TNF‐α, likely secreted by nearby microglia, may induce activation of astrocytes into a neurotoxic state (Labib et al. 2022) potentially through an NF‐κB signaling cascade (Leng et al. 2022). Once activated, these astrocytes are then able to induce apoptosis in neurons and oligodendrocytes by a signaling mechanism (Liddelow et al. 2017) recently suggested to be long‐chain saturated lipids contained in APOE and APOJ lipoparticles (Guttenplan et al. 2021). Knockout of the ELOVL1 enzyme in astrocytes, which prevents the formation of long‐chain saturated lipids by astrocytes, reduced the astrocyte induced neurotoxicity in vitro and in vivo (Guttenplan et al. 2021). Knockout of C1q in the P301S mouse tauopathy model reduced hippocampal volume loss and improved hyperactive mouse behavior but did not improve phospho‐tau immunoreactivity, microgliosis nor astrogliosis suggesting C1q acts downstream to tau deposition and gliosis (Dejanovic et al. 2022). Dejanovic et al. further reported that many astrocytic protein markers were increased in proteomic data of postsynaptic density fractions of these mice in a C1q‐dependent manner and showed similar trends in human synaptoneurosome proteome data (Hesse et al. 2019) comparing AD cases and controls (Dejanovic et al. 2022). They also found a high level of spatial contact between GFAP⁺ astrocytes and Homer1⁺ excitatory synapses and a large amount of Homer1⁺ puncta in astrocytic lysosomes, both of which was reduced when C1q was knocked out suggesting that under pathological conditions, phagocytosis of synapses by astrocytes are C1q dependent (Dejanovic et al. 2022). Interestingly, downstream activation of the classical complement pathway by C1q initiation is the cleavage of complement factor C3 and recent work has shown upregulation of C3 levels in activated astrocyte clusters in AD (Sadick et al. 2022; Guttenplan et al. 2021; Liddelow et al. 2017; Qian et al. 2023) and AD mouse models (Lian et al. 2016; Wu et al. 2019). Using either C3 or C1q knockout in P301S vs. wild type mice, Dejanovic et al. (2022) found evidence that C3 deposition on synapses is downstream of C1q activation and facilitates synapse elimination by astrocytes. Sadick et al., however, did not identify the same C3 protein differences between AD cases and controls, although this was in a total of 6 samples and was a global comparison between the proportion of C3⁺ and GFAP⁺ cells rather than for enrichment in specific astrocyte subpopulations in proximity to synapses (Sadick et al. 2022). Interestingly, Dejanovic et al. further reported that in the TauPS2APP combined amyloid and tau mouse model, astrocytes and microglia phagocytosed more Homer1⁺ synapses near plaques than in plaque free areas but, when TREM2 was knocked out, microglia reduced their phagocytic activity while astrocytes increased this, suggesting that astrocytes may compensate for some of the Trem2‐dependent microglial synapse engulfment activities near plaques (Dejanovic et al. 2022). Interestingly, La Rocca et al. found that low levels of Aβ along with TNF‐α and IL‐1α can activate astrocytes into a neurotoxic like transcriptional phenotype similar to the state induced by the microglial C1q mechanism and that this is distinct from the high concentrations of Aβ often used in in vitro studies (LaRocca et al. 2021). Moreover, Iguchi et al. found that in an App ^NL‐G‐F transgenic mouse model, Tyrobp deficiency significantly reduced GFAP⁺ astrocyte activation but with concomitant Inpp5d haplodeficiency, this reduction was partly restored (Iguchi et al. 2023). INPP5D encodes a phosphatase which is mostly expressed by microglia and is a downstream inhibitor of the TREM2‐TYROBP‐mediated phosphatidylinositol‐(3,4,5)‐triphosphate signaling pathway suggesting that the cross‐talk between microglia and astrocytes as well as their proximity to amyloid plaques are important contributors to the activation of astrocytes (Iguchi et al. 2023). Similar results were also identified with spatial transcriptomics and in situ sequencing in the App ^NL‐G‐F transgenic mouse model finding co‐expression of 57 genes, including activated astrocyte marker genes H2‐D1, B2m, C4b, Gfap, Serpina3n, in the microenvironment of amyloid plaques termed plaque induced genes (Chen et al. 2020). These 57 plaque induced genes contained important connections between classical complement cascade genes including C1q subunits a, b, and c, microglial genes Trem2 and Tryobp, and astrocytic genes Apoe, B2m, and Ctsd, which showed increased connectivity as Aβ load increased suggesting a complex but coordinated response of microglia to astrocytes as nearby levels of amyloid increase (Chen et al. 2020). Nonetheless, only 5 astroglial genes including GFAP, CLU, CTSH, CST3, and IGFBP5 were replicable in human AD superior frontal gyrus tissue enriched in the amyloid plaque microenvironment, but they did identify expression of NPC2, S100A6, ITGB5, PRDX6, and VSIR in AD astrocytes but not controls (Chen et al. 2020). Moreover, although there seems to be a convergence of microglial activation of astrocytes by proinflammatory cytokines C1q + IL‐1α + TNF‐α through an NF‐κB signaling cascade, recent in vitro data with human iPSC‐derived astrocytes and pooled CRISPRi screening suggest that there are two distinct inflammatory astrocytic states that are induced, one that is IL‐1/IL‐6 responsive marked by C3 and SERPINA3 and another that is TNF/IFN‐responsive marked by VCAM1 (Leng et al. 2022). Interestingly, Leng et al. (2022) found only the IL‐1/IL‐6 responsive astrocyte abundance was different between AD cases and controls. They also found about 18% of their iPSC‐derived astrocytes were C3⁺ and VCAM1⁺ which they grouped with the IL1/IL‐6‐responsive group because of their similar response, but found only a very small proportion of astrocytes (< 0.00002%) that were C3⁺ and VCAM1⁺ in human AD tissue (Leng et al. 2022). Further, Leng et al. found that STAT3 is a key regulator of these distinct inflammatory signatures with increased STAT3 expression promoting the IL‐1/IL‐6‐responsive state and inhibiting the TNF/IFN‐responsive state (Leng et al. 2022). Lastly, work by Lee et al. (2022) have also shown that FMNL2 is an important astrocytic regulator of cardiovascular risk factors for AD by altering the normal astroglial‐vascular mechanisms. Lee et al. (2022) found FMNL2 expression upregulated in human astrocyte clusters enriched with interferon response genes and likely that dysregulation of FMNL2 in astrocytes disrupts the clearance of amyloid and tau from the brain into the vasculature. Together, these data demonstrate that mechanisms related to inflammation, amyloid β and microglial‐crosstalk are important activators of astrocytes in AD, that astrocytes are important contributors to synaptic pruning in the context of neurodegeneration and vascular homeostasis, and there are likely multiple activator‐dependent activated astrocytic states. Despite similar conclusions from different studies, there are also study‐specific differences. For example, Brase et al. found that astrocytic clusters show reduced expression in AD cases suggesting that astrocytes may lose functionality in AD (Brase et al. 2023), but Barbash et al. found that astrocyte specific genes increase expression in AD cases compared to controls (Barbash et al. 2017). These study‐to‐study discrepancies need to be evaluated taking into consideration study cohort, method, analytic differences as well as power.

5.4.2.2. Identification of Astrocytic Mechanisms With a Genetic Underpinning

Astrocytes are the main producers of APOE in the brain and because the largest AD genetic risk factor is an isoform of this gene, APOEε4, there has been interest in investigating if there are astrocyte‐specific effects of APOE genotypes in brain aging and AD. Interestingly, in AD, expression of APOE increases in microglia and decreases in astrocytes (Mathys et al. 2019; Grubman et al. 2019), however, more recent deep endophenotyping studies have also suggested differences dependent on genetic background which may contribute to the known ancestry dependent differences in APOEε4 risk, namely that African Americans and Latin Americans have lower APOEε4‐associated AD risk than non‐Hispanic whites (reviewed in Ertekin‐Taner (2007)). In a small cohort of homozygous APOEε4 AD patients with either European (N = 7) or African (N = 4) local genetic ancestry at the APOE locus, Griswold et al. (2021) found through scRNAseq that those with European ancestry had overall increased APOE expression than those with African ancestry, with the largest APOE expression difference found in a microglia cell cluster, though there was also a similar direction of difference in an astrocyte cluster. There was also a 13‐fold difference in cell proportions of an astrocytic cluster enriched for activated astrocyte marker genes between individuals with European and African local ancestry at the APOE locus. Although this might suggest that ancestry can have an impact on astrocytic substates, this cluster was primarily composed of cells from two European samples indicating that this finding may simply represent heterogeneity in AD (Griswold et al. 2021) or possibly a technical artifact in a low power study. The expression of GFAP, a marker of astrocyte activation, was significantly upregulated in AD cases with APOEε2/3 and APOEε3/3 genotypes but not APOEε3/4 providing further evidence that APOE genotype may impact astrocytic state (Panitch et al. 2021). Additional studies have also investigated broader transcriptional differences between AD cases and controls dependent on APOE genotypes. Brase et al. found pathways relating to COPI coating of Golgi vesicles in astrocytes enriched in APOEε4 carriers (Brase et al. 2023). Integrating the Mathys et al. (2019) and Grubman et al. (2019) snRNAseq datasets and stratifying by APOEε3/3 and APOEε3/4 genotypes, Belonwu et al. (2022) found shared and distinct differential expression between AD cases and controls dependent on APOE genotype in the astrocytic clusters. Shared astrocyte cluster pathways perturbed in AD prefrontal and entorhinal cortices from donors of both APOE genotypes included synaptic signaling and transmembrane transport, while APOEε3/3 donor astrocytes showed distinct enrichment in ion transport and ferritin sequestration pathways and APOEε3/4 had distinct enrichment in neuronal and axon development/myelination related pathways (Belonwu et al. 2022). In comparison to microglia, astrocytes have less transcriptional differences dependent on APOE genotype, but these modest gene expression changes were related to lipid metabolism and extracellular matrix in APOEε2 and APOEε4 carriers compared to APOEε3 (Serrano‐Pozo et al. 2021). These astrocytic APOE genotype‐related gene expression changes were, however, generally not statistically significant in either normal nor AD brain tissue potentially suggesting that transcriptomic changes in astrocytes are less dependent on APOE and rather driven by other factors such as microglia or neuropathology (Serrano‐Pozo et al. 2021). Conversely, although lipid metabolism was not significant in the Serrano‐Pozo et al. (2021) study, Farmer et al. (2021) did identify APOEε4 attributable differences in astrocytic metabolism. It will be important for future studies to better understand these APOE‐dependent effects on astrocytic metabolism as astrocytes play a major role in the brain supporting neurons and synapses through their metabolic activities and cholesterol synthesis.

Although less well studied, additional known genetic variants implicated in AD pathogenesis have been investigated for their potential effects on the astrocyte transcriptome. Variants in TREM2 including the common variant and the R47H hypomorphic variant were investigated in humanized TREM2 5XFAD mouse astrocytes finding similar expression levels of astrocyte reactive genes including PDE4DIP, NCAN, HILPDA, and SERPINA3 across both variants (Zhou et al. 2020). Compared to control, ADAD (Dumitrescu et al. 2020) samples showed transcriptional signatures highly correlated to that of sporadic AD suggesting a similar astrocytic activation in both types of AD (Brase et al. 2023). However, two astrocytic substates were specific to ADAD brains, one of which was enriched for cytoplasmic transport and cytokine signaling pathways suggesting there may be ADAD specific activation of astrocytes (Brase et al. 2023). Lastly, carriers of the MS4A variant rs1582763‐A had a significant decrease in proportions of astrocytes with a resting signature (LUZP2, SLC7A10, and MFGE8) and a trending increase in astrocytes with an activated signature (GFAP, ID3, AQP4, ID1, and FABP7), and because astrocytes do not express MS4A, this suggests a potential MS4A‐dependent cross‐talk mechanism between astrocytes and microglia (Brase et al. 2023).

5.4.2.3. Heterogeneity in Astrocyte Populations Dependent on Brain Region

Studies often use different brain regions to increase their power to identify astrocytic mechanisms and gene signatures but do not comment on the potential differences and effects these different brain regions may have. Previous studies have identified important gene expression signature similarities across the brain (Wang, Allen, et al. 2022), but many have shown there are also distinct brain region specific patterns (Xu et al. 2023; Li and De Muynck 2021; Morabito et al. 2020; Wang, Chen, et al. 2022; Sadick et al. 2022; Belonwu et al. 2022; Qian et al. 2023). Astrocytes are often observed in areas of the brain with decreased cortical thickness and neurodegeneration either through measures of cell‐type specific abundance or a novel method called case–control virtual histology, using RNAseq data and in vivo structural MRI scans to estimate cortical thickness (Kerrebijn et al. 2023). Leveraging spatial transcriptomics of AD donor and control brain tissue, Chen et al. found a higher predicted proportion of astrocytes in layer I compared to other layers of the middle temporal gyrus, a brain region vulnerable to AD, but no significant difference of astrocyte proportions between AD cases and controls in any other cortical layer (Chen et al. 2022). Interestingly, Chen et al. (2022) not only found layer specific differential expression of astrocyte genes between AD cases and controls (upregulation of SLC1A3, KIF5A, CSRP1, and CRYAB in layers II/III and V, SNCG, GLUL and CD63 in layer II/III and decrease of PAQR6 in layer II/III), they also identified differential expression based on proximity to amyloid plaques and neurofibrillary tangles. These differences included upregulation of SLC1A3, KIF5A, CSRP1, GLUL, and CD63 in layer II/III, CRYAB in layer V and downregulation of SLC1A3 in layer V close to amyloid plaques while there was upregulation of CD63 and CRYAB in layers II/III and V, SLC1A3, KIF5A, SNCG, CSRP1, and GLUL in layer II/III, and downregulation of CSRP1 and GLUL in layer V close to tau deposits. Comparisons of astrocytic substates across brain regions revealed conserved expression patterns in pathways such as inflammatory response that may be shared across regions as well as substate‐specific gene expression heterogeneity dependent on brain region (Xu et al. 2023; Sadick et al. 2022; Qian et al. 2023). It is unknown whether this heterogeneity is an intrinsic substate difference based on brain region versus a consequence of regional microenvironment differences (e.g., due to amyloid and tau) resulting in different astrocyte activation signals.

5.4.3. Astrocytes in Alzheimer's Disease (AD) Versus AD Related Disorders

Studies have also compared astrocytic gene signatures in AD and other related dementias to identify conserved and distinct astrocytic mechanisms. Multiple studies have found shared bulk differential expression patterns between AD and Parkinson's Disease (PD) (Gil et al. 2021) cases (Bordone and Barbosa‐Morais 2020; Sadeghi et al. 2022). Xu et al. (2023) found three shared astrocytic substate cluster signatures in AD and PD cases as well as controls, all with a significant overlap and concordant change in differentially expressed genes and enriched pathways suggesting these distinct astrocytic states are conserved in the brain independent of specific disease context. Sadeghi et al. (2022) also found a module enriched for astrocytic genes and metabolism upregulated in AD and PD, as well as pathologic aging, autism spectrum disorder, schizophrenia, and bipolar disorder which had gene expression patterns most similar to those from the medulla oblongata. Nevertheless, a meta‐analysis of the substantia nigra transcriptome (Phung et al. 2020) for genes nearby known AD risk loci did not reveal shared differential expression patterns between AD and PD. When investigating astrocyte cell abundance from deconvolved bulk RNAseq data, compared to controls, astrocytes were found to increase in AD cases (Bordone and Barbosa‐Morais 2020; Johnson, Webster, and Hales 2022) and Frontotemporal degeneration (FTD) (Johnson, Webster, and Hales 2022), but have no significant proportional changes in PD (Bordone and Barbosa‐Morais 2020). Others have compared bulk AD gene signatures to other disorders including Vascular Dementia and FTD/ FTD‐tau (Johnson, Webster, and Hales 2022; Santiago, Bottero, and Potashkin 2020) finding largely correlated gene expression changes across all three conditions. Santiago, Bottero, and Potashkin (2020) found almost no shared astrocytic pathway enrichment across diseases but did find perturbations in AD‐specific pathways known to be important in astrocytes like PI3K‐AKT and calcium signaling, although these pathways have general importance across multiple brain cell types. Conversely, Johnson et al. found astrocyte differentiation and function related pathways upregulated in AD and FTD‐tau (Johnson, Webster, and Hales 2022). Interestingly, Santiago, Bottero, and Potashkin (2020)) identified a single transcription factor, KLF4, with shared perturbation between AD, Vascular Dementia, and FTD. This transcription factor has been implicated in the activation of astrocytes after ischemic conditions or injury (Wang and Li 2023; Park et al. 2014; Huang et al. 2020).

5.4.4. Astrocytes in Human Versus Model Organisms

The human astrocytic AD gene signatures have also been compared to those from other model organisms as understanding these similarities and differences is imperative to evaluating the applicability of these model organisms in deciphering AD‐related astrocytic changes. Similar to human datasets, astrocytic cluster markers often used in mouse datasets to initially identify astrocyte clusters include Aqp4 (Zhou et al. 2020), Gfap (Chen et al. 2020), Gja1 (Zhou et al. 2020), Mlc1 (Dejanovic et al. 2022), Slc1a4 (Dejanovic et al. 2022), Slc1a3 (Chen et al. 2020), and Clu (Chen et al. 2020). By applying these astrocytic mouse signatures to snRNA‐seq datasets Habib et al. (2020) found three distinct states of astrocytes, Gfap‐high, Gfap‐low, and disease associated astrocytes (DAA) expressing Serpina3n, Ctsb, Apoe and Clu, as well as two states that seem to be transitional. In comparing human astrocyte clusters from sn/scRNAseq data to mouse, Xu et al. (2023) and Zhou et al. (2020) identified a substantial number of differences in expression of astrocyte substate transcriptional signatures between human and mouse, although Xu et al. (2023) described some signature homology between astrocyte clusters. Morabito et al. (2021) investigated the DAA signature from Habib et al. (2020) in their human snRNAseq/ATACseq study and found that these gene signatures correlated with disease trajectory in a pseudotime analysis where GFAP‐high and DAA modules increased while GFAP‐low modules decreased over time. Lastly, although less frequently used for investigation of astrocytic pathways in AD, zebrafish are facile models for AD‐related pathologies where GFAP can be used as an astroglia marker. Cosacak et al. (2022) compared human signatures to ortholog genes in zebrafish finding similar pathways enriched across different cell types but that this similarity was seen least in astrocytes. Notably, zebrafish share more similar markers of proliferative and progenitor cells when compared to fetal human brain tissue than adult human brain tissue which suggests that the neurogenic capacity of zebrafish brain cells may be present in early human brain development but are lost as humans age and in AD (Cosacak et al. 2022). Despite these differences, there were similarities between human and zebrafish astroglia gene expression in the presence of amyloid including upregulation of immune and protein quality control related pathways (Cosacak et al. 2022).

5.5. Proteomics

Beyond transcriptomic signatures of astrocytes in AD, there has also been growing work to identify their proteomic signatures. Proteomic signatures can be powerful tools to identify spatiotemporal organization of astrocytic substates in situ via tissue staining and enable purification of substate populations from tissue or cultures for deep endophenotyping and further characterizations. Three proteomic methods frequently used separately or in combination to investigate astrocytic signatures in AD include: (1) challenging astrocytic cell cultures with a molecule(s) to induce a state change and then measuring differential proteins (Labib et al. 2022; Kleffman et al. 2022; Wang, Cai, et al. 2020), (2) performing bulk proteomics on tissue and computationally identifying enrichment of astrocytic proteins (Johnson et al. 2020; Miedema et al. 2022; Seyfried et al. 2017; Swarup et al. 2020; Gao et al. 2022; Dai et al. 2018; Qian et al. 2023; Ojo et al. 2021a), and (3) performing localized proteomics on tissue through fractionation or microdissection methods (Dejanovic et al. 2022; Drummond et al. 2017). Similar to the previously described transcriptomic studies, proteomic studies also point to an important role of astrocytes in AD.

Through protein co‐expression network and differential protein expression (DPE) analyses, modules and protein profiles enriched with astrocytic marker proteins related to gene ontology terms were linked to sugar metabolism (Johnson et al. 2020), immune/inflammatory response (Seyfried et al. 2017; Swarup et al. 2020), cell to cell adhesion (Swarup et al. 2020), and extracellular matrix (Miedema et al. 2022; Seyfried et al. 2017) pathways. These modules also often associate with amyloid and tau pathology in the brain (Johnson et al. 2020; Johnson et al. 2018; Seyfried et al. 2017; Swarup et al. 2020) and cognition (Johnson et al. 2018; Seyfried et al. 2017). These results are consistent with known astrocytic functions including cholesterol metabolism, ion homeostasis, regulation of inflammation, and formation of the blood brain barrier. This suggests that dysregulation of these pathways at the proteomic level is important in the progression of AD. Interestingly, astrocytic enriched protein co‐expression modules and differentially expressed proteins also show enrichment in genetic risk factors of other neurodegenerative diseases including FTD, PSP/CBD, and PD dementia (Johnson et al. 2020; Miedema et al. 2022; Swarup et al. 2020) but not autism spectrum disorder nor schizophrenia (Johnson et al. 2020, 2018) suggesting that astrocytes likely play a larger role beyond AD in neurodegeneration but possibly not as much in other neurological disorders. Johnson et al. (2020) also found that their astrocytic enriched module associated with normal aging, although to a lesser extent than with AD, suggesting that there may also be a shared mechanism between aging and AD. Gao et al. (2022) found a positive association of S100A10 protein levels with aging and AD, a higher density of S100A10 expressing astrocytes in AD hippocampus and entorhinal cortex tissue compared to healthy donors, and that these S100A10 expressing astrocytes showed indications of phagocytosing apoptotic neurons. This enrichment of neuroprotective astrocytes was also identified by Dejanovic et al. (2022) demonstrating that astrocytes may compensate for pruning of inhibitory synapses in a Cq1‐dependent manner around amyloid plaques when TREM2 deficiency prevents microglia from performing this function. Hence, TREM2, a microglia‐enriched gene, has AD risk mutations which may impact AD pathogenesis by altering specific functions of multiple cell types in the brain. This is even more evident with APOE, which harbors the APOEε4 allele, the largest known genetic risk factor for AD. When investigating the proteins of AD donor brains with different APOE genotypes, Dai et al. (2018) found that predicted astrocyte proportions correlated with APOE genotype, finding the lowest number of astrocytes in APOEε2 carriers and the highest in APOEε4 carriers. These results suggest that there are astrocyte specific functional changes in AD tied to genetic variation, the consequences of which individually and in combination have yet to be investigated fully.

Similar to the transcriptomic studies, proteomic studies suggest that astrocytes change their state during AD when compared to healthy control tissue, although the precise mechanisms for this remain elusive. Studies investigating the role of proinflammatory factors and amyloid β suggest that these are able to induce astrocyte state changes (Labib et al. 2022; Kleffman et al. 2022; Ojo et al. 2021a), although these changed states were not identical. Exposure to inflammatory factors of TNF, IL1α, and C1q induced iPSC derived astrocytes into an activated substate characterized by increased expression of proteomic markers VCAM1, BST2, ICOSL, HLA‐E, PD‐L1, and PDPN (Labib et al. 2022). On the other hand, exposure to amyloid‐β fragments produced from melanoma metastases in the brain suppressed inflammatory polarization of astrocytes nearby as characterized by a reduced level of complement pathway C3 when compared to astrocytes farther from the amyloid β secreting metastases (Kleffman et al. 2022). Expression levels of astrocytic marker proteins including GFAP, AQP4, S100β, SLC4A4, and APOE were also found to vary depending on the level of cerebral amyloid angiopathy (CAA) pathology, that is, amyloid deposits in the cerebrovasculature, from the inferior frontal gyrus tissue fractions enriched for cerebrovessels (Ojo et al. 2021a). Moreover, when comparing the localized proteome around amyloid plaques from patients with rapidly progressive AD and sporadic AD via microdissection, Drummond et al. (2017) found significantly higher levels of astrocytic proteins around amyloid plaques in the sporadic AD cases as a result of increased numbers of plaque associated astrocytes as measured by increased GFAP levels via IHC. The relationship between astrocytes and amyloid β was further investigated by Wang, Cai, et al. (2020) who stressed astrocytoma U251 cell models with amyloid β and treated with verbascoside that has extensive pharmacological properties including antioxidative. They found that this treatment alleviated ER stress, indicating that amyloid β may impact astrocytes through ER stress related pathways. Together, these results suggest that the microenvironment, particularly that of inflammatory molecules and amyloid, plays a crucial role in the proteomic state of astrocytes in AD.

More recently, proteomic approaches have also been used to identify AD biomarkers linked to changes in astrocytes in CSF or peripheral blood. As astrocytes change states likely throughout the progression of AD and in line with pathological and cognitive phenotypes as discussed in the previous section, identifying astrocytic protein biomarkers in peripheral tissue will be useful for staging of AD progression. Levels of LDHB, PKM, GAPDH, PRDX1, DDAH2, and PARK7 identified from brain protein co‐expression modules enriched with astrocytic markers were also found to be differentially expressed in CSF from AD cases and Asymptomatic AD or controls (Johnson et al. 2020). Astrocyte related proteins including A2M, SERPINA3, APOE, CD14, and UBA52 were found to be robustly associated with AD related changes in CSF of AD mouse models and showed replication in comparative human CSF proteome analyses, however, the latter three may also be related to microglial and/or neuronal specific changes (Eninger et al. 2022). Beyond bulk proteomic changes in the CSF, proteomic changes in extracellular vesicles (EV) are of interest due to their key role in inter‐cellular communication and the potential to classify originating and destination cell types. Through label free proteomics of CSF‐derived EVs, Muraoka, Jedrychowski, et al. (2020) found enrichment of astrocyte specific proteins to a greater extent in AD cases compared to MCI. This was similar to EV's isolated from brain tissue where around 30% of the identified proteins were astrocyte specific and differentially expressed proteins between AD cases and controls included CAV1, PBXIP1, CYBRD1, MAOB, SLC16A1, MT3 higher in AD, and HSPA2, FBXO2, JAM2, and PLCE1 lower in AD (Muraoka, DeLeo, et al. 2020). Huang et al. (2022) also found higher astrocyte protein markers isolated from brain EVs in AD cases compared to controls including CD44 and GJA1. However, LRP1 and ITGA6 which were identified as EV markers derived from astrocyte differentiated iPSCs were not differentially expressed in EVs from AD donor brains or CSF (You et al. 2022). Of note, although EVs have a potential to be particularly useful in the context of centrally‐linked peripheral biomarkers (i.e., biomarkers revealing brain molecular perturbations in peripheral tissue like blood or CSF), the studies described above lack consistent observations of specific astrocytic proteins that are differentially expressed in AD. Although these inconsistencies could be a result of cohort, EV isolation, or proteomic technique specific differences, this could also point to a potentially limited ability for astrocytic changes in AD to be reliably detected in EVs and overall generalizable.

There has also been recent experimental advances related to proteomics or generated as a result of proteomics which has and will continue to improve our understanding of astrocytes in AD. A new proteomic method developed by Johnson et al. (2018) and expanded in 2020 (Johnson et al. 2020) uses isobaric tandem mass tag (TMT) mass spectrometry and offline prefractionation method more than doubling the depth of brain proteome coverage when compared to their previous online “single‐shot” label free quantification by liquid chromatography tandem mass spectrometry from the same samples. This increased proteome depth and coverage is critical to identify important astrocytic protein changes across samples which may otherwise be unquantified or dropped due to high missingness. There has also been advances in cyclic multiplexed fluorescent immunohistochemistry to identify spatial relationships of specific cell types and substate types, morphological characterization, and affiliations with pathological features from FFPE brain tissue by Muñoz‐Castro et al. (2022). Using 7 astroglial markers, Muñoz‐Castro et al. (2022) identified 3 distinct clusters of astrocytes including a reactive substate characterized by increases in GFAP, YKL‐40, vimentin TSPO, EAAT1 and reduced GS, a homeostatic cluster with the reverse expression of the previous markers, and an intermediate cluster marked by increased EAAT2, moderate GS increase, and intermediate levels of GFAP and YKL‐40, marking either a potential transitional or resilient state. This provides additional and highly important contextual information to understanding the complex dynamics of astrocytes in brain tissue, their relationship with pathological features, and their likely important connection with microglia.

5.6. Lipidomics and Metabolomics

Astrocytic processing of cholesterol and lipids in the brain is an extremely important process as neuronal cells rely on neighboring astrocytes for this support, particularly around synapses and during hyperactivation. Despite this, the application of lipidomic or metabolomic approaches to understand the crucial role astrocytes play in providing cholesterol and lipid metabolism in the brain is limited. Multiple recent studies have begun exploring the effect of the APOE genotype of lipid metabolism in astrocytes. Astrocytes constitute one of the main APOE producing cells in brain; APOE transports lipids between astrocytes and neurons, and APOE harbors the largest known AD genetic risk variant, but the molecular links between astrocytes, APOE and lipid metabolism are not fully known. Using lipidomics approaches in human derived iPSC astrocytes, APOE‐dependent lipid processing mechanisms were discovered revealing that APOE genotype significantly affects lipid homeostasis in astrocytes and that this may be one of the mechanisms underlying AD risk for APOEε4 carriers and protection for APOEε2 carriers (Sienski et al. 2021; Lindner et al. 2022). These studies demonstrated that human iPSC derived APOEε4 astrocytes store triacylglycerol (TAG) in the form of lipid droplets much more than APOEε3 carriers and that this difference is even more dramatic when treated with excess TAG (Sienski et al. 2021; Lindner et al. 2022). This is important because an increase in lipid droplet accumulation in the brain is associated with aging or stress as well as AD and other neurodegenerative disorders (Sienski et al. 2021; Lindner et al. 2022). Moreover, when induced by oleic acid treatment to store TAG in lipid droplets, Lindner et al. (2022) found that astrocytes reach a saturation point and begin releasing TAG‐rich particles into the extracellular space in an APOE‐dependent manner, a process more active in APOEε4 astrocytes than APOEε2 or APOEε3 astrocytes. APOEe2 astrocytes behaved inversely to APOEε4 and more similarly to APOE‐KO astrocytes suggesting that APOEε2 may provide a protective effect in AD by allowing uptake of excess oleic acid but release of significantly less TAG into the extracellular space (Lindner et al. 2022). These results suggest that in low fat level conditions, APOEe4 astrocytes accumulate TAG more quickly than APOEε2 or APOEε3, but in fatty conditions, astrocytes can begin releasing TAG‐loaded particles into the extracellular space and that APOEε4 astrocytes activate this mechanism much more than APOEε2 or APOEε3 astrocytes. TAGs were also measured in a small cohort of temporal lobe samples, revealing that although TAG levels were similar between AD cases and controls, there was a significant downregulation in individuals with AD neuropathology but without dementia suggesting TAG metabolism may be related to cognitive resilience (Barbash et al. 2017). There is also emerging evidence that APOE genotype impacts astrocytic ability to process cholesterol. In human iPSC derived astrocytes, it was found that cholesterol accumulates more in APOEε4 carriers (Lin et al. 2018). Accumulation of cholesterol was not seen in FACS isolated astrocytes from forebrains of Apoe knockout mouse model compared to those with Apoe suggesting that this accumulation is a result of the APOEε4 genotype itself, although there were Apoe dependent effects on specific cholesteryl esters (CE18:1 and CE22:6) (Nugent et al. 2020). When investigating other lipids like cholesteryl esters (CE), Nugent et al. (2020) also treated the these mice with cuprizone, a demyelinating agent finding an even higher increase in CE species 18:1 and CE22:6 in the Apoe knockout astrocytes isolated via FACS (Nugent et al. 2020). Interestingly, Sienski et al. (2021) treated their iPSC derived astrocyte cultures finding that supplementation with CDP‐choline, a soluble phospholipid precursor, rescued the APOEε4 genotype dependent accumulation of both TAG and cholesterol. Together, these results strongly indicate that there are broad APOE and specific APOE genotype effects on lipid metabolism in astrocytes, particularly for cholesterol and TAGs, and that targeting these mechanisms may alleviate some of the AD risk attributed by APOEε4. Additional studies investigating glycolytic activities in astrocytes of APOEε4 carriers also found an increase in glucose flux through aerobic glycolysis in mouse derived astrocyte cultures (Farmer et al. 2021). APOEε4 effects were also observed beyond the brain in the plasma metabolome of young adult participants where APOEe4 carriers had higher plasma lactate concentrations and an enrichment for pathways related to aerobic glycolysis (Farmer et al. 2021). Taken together, these studies highlight the strong role APOE has in lipidome and metabolome in astrocytes and that future studies investigating these should account for APOE genotype and sex in their analyses.

5.7. Summary

Following microglia, astrocytes are the next most well‐studied glial cells by omics studies in AD (Figure 1), leading to the discovery of many perturbed pathways (Figure 3) and insights about their roles in AD pathophysiology (Figure 2). An increase in relative abundance of astrocytes correlates with increased neuropathology burden in AD with aggregation of astrocytic proteins around amyloid plaques. Higher number of astrocytes have been correlated with APOEε4 genotype and the lowest with the APOEε2 genotype. The APOE genotype may affect astrocytic production of APOE to clear/maintain lipids and cholesterol at neuronal synapses. Astrocyte specific expression has been linked with some risk variants for AD. Epigenetic factors such as astrocytic DNA methylation may play a role in increased risk for AD. The methylation landscape is more perturbed in glial cells compared to neurons in AD, however astrocyte specific changes are yet to be identified. The approach to investigate this is somewhat limited due to lack of reference panels to infer cell type composition when working with bulk DNA tissue. Cell sorting techniques, however, have provided the means to categorize cells by either neuronal or glial cell state. Differentiation of glial cell types needs further improvement through identifying effective/specific cell type markers. Investigation of chromatin structure in relation to AD in astrocytes using ATAC‐seq and snRNA‐seq following fluorescence activated nuclei sorting have revealed sex‐specific chromatin changes at AD risk loci, especially at the APOE loci in females where chromatin accessibility is limited. In contrast, open chromatin regions at and near the GFAP genes are associated with astrocytic increase in AD. Understanding the role of astrocytes is important to ascertain as they are responsible for significant APOE production in the brain. Bulk RNA‐seq data has revealed that genes enriched alongside astrocytic markers are also associated with AD pathology. Astrocytic cells have a homeostatic state and subpopulations of various activated states, which can vary depending on brain regions and AD versus other neurologic diseases. Astrocytic subclustering can also be found in mouse models with some conserved homology. There is cross‐talk of astrocytes and microglia, which may be dependent on AD risk genes including APOE and TREM2. Astrocytes may play a compensatory role in immune response alongside microglia. Certain markers of microglia may be involved in activating astrocytes through secretion of C1q and other immune factors, initiating apoptosis of neurons and oligodendrocytes or causing phagocytosis of synapses. Although APOE is mainly produced by astrocytes, in the AD brain APOE expression increases in microglia and decreases in astrocytes. This may be dependent on allelic ancestry of APOE. Gene expression signatures or cell abundance patterns are shared between AD and other neurodegenerative diseases including PD and FTD. Proteomic based studies of AD have revealed astrocytic marker proteins to be related to sugar metabolism and inflammation pathways among others. Enriched astrocytic proteins in AD overlap with those in FTD, PD and PSP/CBD. AD related astrocytic proteins in CSF and peripheral blood have been discovered. Although brain derived extracellular vesicles have been found to contain higher astrocytic protein markers, further studies are required in this area. In the future, further investigation is needed in astrocytic subtypes, their heterogeneity and differential methylation profiles, sex‐specific methylation differences, the role of senescent astrocytic cells and allelic or ancestral differences in genotypes associated with AD in astrocytes.

6. Vascular Cells

6.1. Introduction

Omics based studies focused on the vascular contributions to AD pathogenesis requires further investment and expansion (Figure 1). Vascular dysfunction is well known to occur in AD. In addition to amyloid β plaques and neurofibrillary tangles, AD brains often have vascular co‐pathologies such as cerebral amyloid angiopathy (CAA) with microhemorrhages and other cerebrovascular disease, like infarcts. A study performed by Toledo et al. revealed that around 80% of AD cases displayed vascular pathology (Toledo et al. 2013). Additionally, the prevalence of vascular pathology was higher in AD cases compared to other neurodegenerative diseases (Toledo et al. 2013). Cerebrovascular dysfunctions in AD may also manifest as reduced cerebral blood flow (CBF) and loss of integrity in the blood brain barrier (BBB) in AD. New insights from omics technologies have begun to enhance our knowledge on the molecular underpinnings of vascular AD‐related pathological alterations, which will be the primary focus of this section. Although AD related damage manifests itself in all blood vessel types (arteries, capillaries, and venules), we primarily focus on capillaries and microvasculature in the cerebral cortex, as important vascular cell types for the initiation and propagation of AD.

Reduced CBF (hypoperfusion) to the brain is one of the early events in Alzheimer's disease. Factors that help regulation of CBF are thought to be protective towards AD (Bookheimer et al. 2000; Ruitenberg et al. 2005; Smith et al. 1999; de Eulate et al. 2017; Roher et al. 2012; Austin et al. 2011; Iturria‐Medina et al. 2016). The underlying molecular mechanisms and cause of dysregulated CBF in AD have not been clarified. Nonetheless, decreased CBF can have detrimental effects on cognitive function, brain homeostasis and even accelerate the progression of AD by affecting amyloidogenic pathways (Li et al. 2009; Zhang et al. 2007; Sun et al. 2006). In addition, intact BBB is critical to maintain brain functions such as synaptic operation, information processing and neuronal connectivity. Loss of BBB integrity in AD has been documented in several studies, that were compiled as reviews by Kurz et al. (2022) and Barisano et al. (2022). Imaging studies demonstrated increased permeability of BBB, microbleeds, impaired glucose transport and P‐glycoprotein functions, and leukocyte infiltration to the brain. In the light of omics studies, our understanding of molecular factors underlying BBB dysfunction in AD is expanding.

Human brain vasculature comprises four main vascular cell types: Endothelia, mural cells (such as pericytes and smooth muscle cells), fibroblasts and perivascular macrophages (PVM). The walls of blood vessels are made up of endothelial cells, arranged as a monolayer through tight junctions. Endothelial cells regulate the transport of nutrients from blood to brain and toxic wastes from brain to blood (Bosseboeuf and Raimondi 2020). The pericytes of the brain surround and wrap parts of the endothelial cells and display a distinct morphology. In the brain, pericytes cover the endothelial cell lining with a higher ratio of coverage than that in other tissues (Sims 1986). Smooth muscle cells are also endothelium wrapping cells that regulate cerebrovascular resistance and CBF. Perivascular fibroblasts, on the other hand, reside in perivascular spaces, meninges, and choroid plexus of the brain and spinal cord. PVMs modulate the developing vasculature and have multiple functions that are critical for brain homeostasis and pathology. Despite undefined clear‐cut roles of PVM in human brain, their potential roles are to detect and engulf hazardous foreign substances in the blood vessels. Each cell type has a specific role in the brain vasculature and coordination among these cells ensures maintenance and support of human brain homeostasis. This balance is dysregulated in AD, the molecular basis of which can begin to be deciphered through omics studies.

6.2. Genomics

Genome wide association and whole genome sequencing based studies support the role of vascular contribution to AD (Figure 1). These studies revealed genes and loci that are common between AD and cardiovascular diseases and helped nominate vascular molecular targets that are implicated in cholesterol/lipid metabolism, vasculature development and gliovascular interactions. Examples of such novel vascular targets are LINC‐PINT and TRIM47, and centrally linked peripheral molecular target, ST18. The genetic background of individuals with AD‐related vascular burden such as cerebral amyloid angiopathy and cerebral small vessel disease is important to consider both as additional risk for AD and also to uncover the mechanisms behind such cerebrovascular disease in AD.

6.2.1. Cholesterol/Lipid Metabolism‐ and Vascular Cells

Several AD related genes can influence deterioration of cerebrovascular functions by affecting endothelial cells through their effects on cholesterol/lipid metabolism. APOE4, which modulates cholesterol transport, is the biggest risk factor for AD. Recent evidence suggests that APOE4 also directly impacts endothelia cell functions that are relevant to AD‐related phenotypes (Rieker et al. 2019). In addition to APOE4, several other studies have identified pleiotropy between AD and cardiovascular risk genes. Using the results of multiple GWAS, Broce et al. discovered that a subset of genes associated with cardiovascular disease are also strongly associated with AD (Broce et al. 2019). In their study, 90 SNPs on 19 different chromosomes were found to be associated with an increased risk for both AD and cardiovascular disease. These SNPs were associated with plasma lipid levels rather than other cardiovascular risk factors and located within the genes involved in cholesterol/lipid function such as apolipoproteins, ATP‐binding cassette transporters, and phospholipases (Broce et al. 2019).

6.2.2. Vasculature Development

Impact of vascular phenotype is not solely based on the APOE4 genotype. Moreno‐Grau et al. discovered that several additional vascular genes were implicated in AD in addition to cholesterol biosynthesis and transport genes (Moreno‐Grau et al. 2019). In their study, they performed GWAS on 4120 AD patients and 3289 controls as part of the Genome Research at Fundacio ACE (GR@ACE) study and categorized the LOAD genetic variants into three classes based on vascular burden: (a) purest AD, (b) mixed with vascular disease c) effects in all clinical endophenotypes. They detected genes implicated in vasculature development and blood vessel morphogenesis only in the purest AD category (Moreno‐Grau et al. 2019). A significant correlation between vascular genes and vascular burden highlights the role of vascular pathology in AD.

6.2.3. Gliovascular Interactions

A study by Lee et al. identified FMNL2, which encodes formin‐related protein that plays a role in AD vascular pathology by modulating gliovascular interactions (Lee et al. 2022). They performed GWAS for the association of cerebrovascular risk factors (CVRF) to AD on 6568 AD patients and 8101 controls from Washington Heights–Inwood Columbia Aging Project (WHICAP), Estudio Familiar de Influencia Genetica en Alzheimer (EFIGA), the National Alzheimer's Coordinating Center (NACC), and ROSMAP. Through an adaptive gene–environment interaction test, they identified nominally significant FMNL2 interactions in all groups but African Americans. Furthermore, their subsequent experimental validation in ROSMAP, Mount Sinai, and Mayo Clinic Brain Bank cohorts indicated that patients with brain infarcts or AD had higher levels of FMNL2 expression, which correlates with accumulation of amyloid and tau.

6.2.4. Cerebral Amyloid Angiopathy

Cerebral amyloid angiopathy (CAA) is observed in the majority of AD cases to some extent (Ellis et al. 1996; Boyle et al. 2015). It is caused by the deposition of amyloid β around cerebral vessels and impairs blood vessel integrity (Ghiso, Fossati, and Rostagno 2014). In Reddy et al., we performed GWAS in AD donors from the Mayo Clinic Brain Bank (n = 853) for association with the severity of CAA pathology scored across 5 brain regions and discovered a novel locus, LINC‐PINT, the top rare variant of which is associated with lower CAA among APOEε4‐negative individuals (Reddy et al. 2021). This variant was also associated with altered splicing of this gene, suggesting a functional consequence. LINC‐PINT, is a long non‐coding RNA expressed in multiple tissues and across multiple brain regions (GTEx Consortium 2017), and possibly involved in neuroprotection (Simchovitz et al. 2020). Using transcriptome data from 1186 donors from the AMP‐AD consortium, we determined LINC‐PINT to be replicably higher in AD brains compared to controls. Importantly, pathway analyses of other GWAS findings from this study demonstrated an enrichment for genes involved in neuronal/synaptic health and function, implying the potential complex interplay of this AD vascular pathology with neurodegeneration.

6.2.5. Cerebral Small Vessel Disease

Another vascular pathological phenotype, observed frequently in AD is cerebral small vessel disease (CSVD) (Wardlaw, Smith, and Dichgans 2019). Mishra et al. validated the causal association of AD with CSVD through complementary GWAS (n = 41,326) and whole exome association study (n = 15,965) from the UK Biobank and Heart and Aging Research in Genomic Epidemiology (CHARGE) cohorts (Mishra et al. 2022). They identified 11 extreme CSVD loci with potential to be involved in AD (Mishra et al. 2022). They also nominated a novel gene, TRIM47, as the most plausible causal gene at the SVD risk locus (Mishra et al. 2022). The TRIM47 gene encodes a protein ligase that ubiquitinates SMAD4 and CYLD for proteasomal degradation (Ji et al. 2018) and its knockdown increased endothelial cell permeability (Mishra et al. 2022).

6.2.6. Plasma Based AD Vascular Damage Marker

AD related vascular damage can also manifest itself in the plasma. Jiang et al., demonstrated that plasma and CSF soluble ST2 (sST2) levels are higher in AD patients (n = 345) compared to healthy controls (n = 345) (Jiang et al. 2022). sST2, is a secreted form of interleukin‐33 (IL‐33) receptor and can function as decoy receptor of IL‐33 signaling (Kakkar and Lee 2008). Its plasma level is altered not only in inflammatory and cardiac diseases (Villacorta and Maisel 2016; Watanabe et al. 2018; Homsak and Gruson 2020), but also in MCI and AD (Saresella et al. 2020). In a pQTL analysis of ST2 genetic variants with plasma and CSF levels, rs1921622‐A was associated with a lower plasma ST2 level (Jiang et al. 2022). In their previously published snRNAseq data, this group also performed cell‐type‐specific association analyses and found that rs1921622‐A is associated with lower expression of sST2 in brain endothelial cells. According to their Mendelian randomization approach, this variant protects against AD.

Genetic studies have highlighted the vascular component of AD and nominated several genes and loci that are implicated in vascular pathology. However, without complementary omics studies, our knowledge about vascular cell type specific mechanisms would be limited since most of the identified genes and loci are expressed by multiple cell types of the brain. Especially single cell transcriptomics approaches can shed light upon the complexity of vascular dysfunction in AD, which will be discussed in next section.

6.3. Transcriptomics

The abundance of cell types vary greatly in the human brain, both in those unaffected by neurodegeneration, as well as in AD (Wang, Allen, et al. 2020). Bulk tissue transcriptomic approaches preferentially capture more information about more abundant brain cell types, such as oligodendrocytes (Allen et al. 2018; McKenzie et al. 2017) or neurons (Strickland, Reddy, et al. 2020). The transcriptional perturbations of rarer cell types, such as endothelia may be more difficult to detect due to both lower abundance and technical or inter‐person variability of their proportion. The impact of peripheral insults on the endothelial cells of the BBB can be assessed using bulk transcriptomics approaches. APOE, insulin, C reactive protein, and drugs have been examined for their effects on brain vasculature health in these studies, and have provided further insights into vascular dysfunction under different conditions.

6.3.1. Diversity of Human Vascular Cells

Obtaining the transcriptomic profile of vascular cells has been difficult owing to technical challenges. Even by using single nucleus transcriptomic approaches, most of human snRNAseq studies have depleted vascular cells without enrichment strategies (Mathys et al. 2019; Grubman et al. 2019; Velmeshev et al. 2019). The first study that reported single nucleus transcriptomic profiles of brain endothelia in AD was carried out by Lau et al., who analyzed 169,496 nuclei obtained from AD (n = 12) and control patients (n = 9) (Lau et al. 2020). Their results did not yield the profile of other vascular cell types and the number of endothelial cells profiled was not sufficient to further characterize the endothelia cells.

Yang et al., developed a method to extract nuclei from human brain microvessels and set to generate a single cell vasculature atlas of hippocampus and frontal cortex (Yang and Vest 2022). They applied their method to generate snRNAseq profile from isolated nuclei of 25 hippocampus and cortex samples, that comprised of tissue from AD (n = 9) and no cognitive impairment (n = 8) donors. After quality control, they obtained the transcriptomic profile of 143,793 nuclei, including endothelia, mural cells such as smooth muscle cells and pericytes, tip cells, and fibroblasts. Endothelial cells showed distinct transcriptional clustering dependent on their locations as follows: arterial endothelia (VEGFC and ALPL), capillary endothelia (MFSD2A and SLC7A5), and venous endothelia (L1R1 and NR2F2). Similar to endothelial cells, fibroblasts also clustered based on their location. While meningeal fibroblasts expressed SLC influx solute transporters, perivascular fibroblasts exclusively expressed ABC efflux pumps. On the other hand, pericytes clustered under two different populations based on their transcriptional identity: T‐Pericytes (transport) and M‐Pericytes (matrix). T‐pericytes are enriched in small molecule transmembrane transport, while M‐pericytes are enriched in extracellular matrix (ECM) organization. They also identified an unconventional cluster, which they call “Tip” cells, that express markers such as L1R1 and NR2F2.

Garcia et al. developed and optimized dextran‐based density ultracentrifugation method to enrich vascular cells from fresh‐frozen human brain samples (Garcia et al. 2022). They generated scRNAseq data from 17 temporal lobe surgical resections of patients with intractable epilepsy. After QC and annotation, they identified and profiled 4992 vascular cells that included endothelia, pericytes, smooth muscle cells, and perivascular fibroblasts. The transcriptomic segregation of vascular cell subtypes is similar to Yang et al. study (Yang and Vest 2022). Garcia et al., also revealed zonation dependent transcriptomic signature in endothelial cells and annotated endothelia cells based on blood vessel size: arterial (VEGFC, BMX, and EFNB2), capillary (MFSD2A and TFRC), and venule (TSHZ2 and LRRC1) (Garcia et al. 2022). Despite showing two different populations, pericytes showed zonation dependent profile in this dataset. The pericyte 1 markers indicate possible localization on the venular side of capillaries, while the pericyte 2 cells are found on the arteriolar side of blood vessels (Ihara and Yamamoto 2022). In addition, Garcia et al. identified three perivascular fibroblast clusters while Yang et al. found only 1 cluster. The discrepancy of vascular cell profiles between these two studies might stem from their vascular enrichment method or utilization of different brain locations.

Another study performed by Sun et al., tackled the issue of vascular enrichment bias by performing snRNAseq from a large cohort of subjects (Sun et al. 2023). The study analyzed 22,514 vascular cells from 11 cell types across 220 individuals with AD and 208 age‐matched control donors (Sun et al. 2023). These 11 vascular cell types comprised 3 endothelia, 2 pericytes, 2 smooth muscle cells, 3 perivascular fibroblast and 1 ependymal cell clusters. Like the previous two studies, endothelia cells showed zonation dependent transcriptomic signature. However, they have not compared these to the identity of cells in Yang et al. or Garcia et al.

These studies also provided in‐depth information about the profile of smooth muscle cells in the human brain. Yang and Vest (2022) identified two smooth muscle clusters based on their region, arterial (expressing ACTA2 and TAGLN) or arteriolar (SLIT3 and CTNNA3) smooth muscle cells. Sun et al. also annotated two smooth muscle cluster subtypes with blood vessel types that they are associated with, arterial or venule. Neither study reported extensive transcriptomic changes in AD for these cells. In arterial smooth muscle cells, Sun et al. (2023) reported AD‐specific downregulated expression of PFDN1 that has a genetic variant with known association with AD. Similar to smooth muscle cells, there has yet to be extensive research to our knowledge on the transcriptomics profile of perivascular macrophages owing to their rarity in the brain and close progeny to microglial cells. Vascular‐focused studies such as Sun et al. and Yang et al. did not report perivascular macrophage clusters (Yang and Vest 2022; Sun et al. 2023).

According to these studies, there is consensus regarding the subtypes of endothelial cells. However, there is much discrepancy about the pericyte subtypes. More studies are needed to further profile pericyte clusters and also to generate well‐powered datasets to study vascular smooth muscle cells and perivascular macrophages.

6.3.2. Cell Type Specific Vascular Changes in AD

SnRNAseq approaches on human brain have provided insights into vascular cell type specific changes. The first snRNAseq study from Mathys et al., excluded pericytes and endothelia cells from their analysis owing to low number of counted cells (Mathys et al. 2019). Lau et al., one of the first studies that reported AD specific vascular changes, generated snRNAseq profile of 169,496 total nuclei from prefrontal cortex of AD (n = 12) and healthy patients (n = 9) (Lau et al. 2020). Out of these cells, only around ~2400 cells were annotated as endothelia (3.0% ± 0.9% AD vs. 1.2% ± 0.3% Controls). There were seven subclusters of endothelia cells, of which only three showed AD‐specific changes. Pathways related to angiogenesis and antigen presentation were upregulated within these clusters (Lau et al. 2020).

The first comprehensive, AD related cell type specific transcriptomic alteration within vascular cells was reported by Yang and Vest (2022). By developing and optimizing vessel isolation and nuclei extraction for sequencing, they profiled 143,793 blood vessel associated cells from the hippocampus and superior frontal cortex in AD (n = 9) and control (n = 8). Their results did not highlight new disease associated vascular cell subtypes. On the contrary, they detected overall loss of vascular nuclei and selective vulnerability of vascular cells, namely M‐Pericytes, in AD. This notion has also been supported by the very high number of downregulated genes in AD pericytes. Most of the differentially expressed genes were specific to vascular cell type and localization and involved in cerebral blood flow and vasoconstriction. They also detected APOE4 genotype specific transcriptional changes within inflammation signaling in AD endothelia cells. Additionally, they assessed AD risk GWAS genes for their expression within vascular cells. A total of 30 out of 45 AD GWAS genes were enriched in blood vessel associated cells, indicating the importance of genetics in vascular dysfunction in AD. Vascular expression of AD risk genes included both expected findings such as vascular expression of PICALM, as well as surprises such as immune‐related PLCG2 in arteries, endocytic expression of INPP5D and USP6NL genes in capillaries and AGRN in smooth muscle cells and pericytes.

Sun et al. analyzed 22,514 vascular nuclei from six brain regions in AD (n = 220) and control donors (n = 208). They have observed regional variations in the proportion of vascular cells, but not among diagnosis, sexes, AD pathologies, or age groups (Sun et al. 2023). However, they observed downregulation of capillary endothelia (ABCB1, ATP10A, PTPRB, and TEK) and pericyte (PDGFRB) markers in AD, which might indicate loss of BBB integrity. They identified 2676 AD‐related DEGs, 2142 of which are specific to one vascular cell type (Sun et al. 2023). Capillary endothelia cells had the highest number of AD related DEGs compared to other vascular cell types. In capillary endothelia and pericytes, these DEGs were mostly regionally conserved and were implicated in functions such as cell junctions and adhesion, solute transporters, and sterol transporters. APOD, a component of cholesterol metabolism, was upregulated in AD pericytes. They also identified insulin signaling genes among shared DEGs across vascular cell types.

Additionally, Sun et al. (2023) analyzed the impact of donor genotype to vascular gene expression changes by integrating known AD associated loci from GWAS. They reported 1010 of 2676 AD related variants that are associated with AD through cis, trans, and intercellular regulatory mechanisms to vascular DEGs related to AD. These AD variant associated DEGs were involved in cellular pathways such as cholesterol transport, endothelial migration, protein folding machinery, and cytokine and growth factor stimulus response. Their DEG analysis between subjects harboring vs. lacking APOE ε4 alleles revealed 2482 APOE related DEGs that significantly overlap with AD related DEGs in a cell type specific manner. These genes have most overlap within endothelia and pericyte cells and are involved in the pathways that regulate transport across BBB, cell junction organization, and angiogenesis and are significantly associated with cognitive resilience.

Finally, although transcriptome studies in general provide a molecular landscape for brain cells in AD and unaffected brains, most fall short of uncovering the mechanisms by which these transcriptional changes may affect biological processes in a way that contributes to AD. To address this knowledge gap in the context of BBB dysfunction in AD, we conducted a brain transcriptome study focusing on two key cell types in the gliovascular unit, pericytes and astrocytes, which are critically important in the maintenance of the BBB, followed by validation studies in blood of living human participants as well as model systems (Is et al. 2024). Following snRNAseq of 12 AD and 12 neuropathologically unaffected control brains in the superior temporal gyrus region, we determined that the majority of the vascular changes in AD occurred in pericytes. Using our data and a molecular interaction prediction algorithm, NicheNet (Browaeys, Saelens, and Saeys 2020), we identified predicted interactions between astrocytic ligands and vascular targets. Of these, pericytic SMAD3, up in AD brains, had the highest number of ligands, which included astrocytic VEGFA, which is down in AD. Combining our data with external datasets comprising 4730 pericyte and 150,664 astrocyte nuclei, we validated these findings. In independent antemortem cohorts where blood SMAD3 transcript levels were measured, we identified associations with AD‐related neuroimaging outcomes. Specifically, higher blood SMAD3 was associated with less brain amyloid β and less cortical atrophy. This could either mean that SMAD3 per se is protective in AD, but it may also reflect an association, rather than causation. Regardless, these findings highlight SMAD3 as a potential centrally linked peripheral signature that informs on the integrity of the BBB. Importantly, we demonstrated an inverse relationship and a functional link between SMAD3 and VEGFA using human iPSC‐derived pericytes and a zebrafish model, where VEGFA, a stimulant of blood vessel growth, reduces the signaling molecule SMAD3. Consistently, VEGFA blockage, increases activated form of SMAD3 (pSMAD3) and impairs BBB in the zebrafish model. In addition to nominating perturbed pericytic‐SMAD3 and astrocytic‐VEGFA as a functionally validated mechanism of BBB dysfunction in AD, this study provides a roadmap of combining human brain omics data with blood omics and model systems to validate and prioritize molecular disease pathways in AD.

Although the vascular component of AD has not been studied using single cell transcriptomic methodology to the same extent as other brain cell types (Figure 1), through the existing studies we have gained a deeper understanding of the vascular molecular changes in AD (Figures 1 and 2). Furthermore, these studies added a new level of complexity to AD‐related changes in vascular cells. Different transcriptional changes can be observed even in the same type of vascular cells based on their zonation within brain regions. Numerous genes and pathways were also revealed to be involved in vascular component of AD through their impact on vascular cell specific mechanisms.

6.3.3. Impact of Peripheral Factors on Cerebrovasculature

Bulk transcriptomic findings have provided further information about the conservation of nominated targets. Impact of brain APOEε4 on AD pathogenesis has been long studied. However, we know less about the role of peripheral APOE. A study by Liu et al., addressed this issue by investigating the effects of peripheral APOEε4 on the brain (Liu et al. 2022). In a mouse model lacking Apoe in the brain, they conditionally expressed human APOEε3 or APOEε4 in the liver. As a result of peripheral APOEε4 interfering with cerebrovascular function in endothelial cells, synaptic plasticity and cognition were impaired in this model which also had upregulated immune response genes in endothelia, which can adversely affect the BBB. This group also treated iPSC derived endothelia with plasma from liver APOEε3 or APOEε4 expressing mice, and evaluated the integrity of the BBB by measuring trans‐endothelial electrical resistance (TEER). Plasma from liver APOEε3‐expressing mice increased TEER of iPSC derived endothelia further, whereas plasma from APOEε4 mice did not. RNAseq of iPSC models after each treatment, followed by weighted gene co‐expression network analysis (WGCNA) (Langfelder and Horvath 2008) revealed modules enriched in cell motility, collagen‐containing ECM, and lipid biosynthesis and metabolism biological terms to be perturbed by treatment with plasma from liver APOEε4‐expressing mice. One of the hub module genes was Timp3, an extracellular matrix binding protein. Researchers suggested that APOEε3 provides benefits to vascular health by increasing Timp3. Treatment with Timp3 recovered the loss of TEER caused by APOEε4 and validated the benefits of Timp3 (Liu et al. 2022). This study provided an alternative mechanism of action for APOE via influence of peripheral form of this molecule on cerebrovasculature, BBB and neuronal function.

Roles of peripheral molecules on the integrity of BBB is not limited to APOE. Wang et al., investigated the impact of insulin signaling on BBB dysfunction by treating human cerebral microvascular endothelial cell (hCMEC/D3) monolayers with insulin at different timepoints and analyzed the affected pathways by RNAseq (Wang et al. 2022c). While genes that are implicated in the regulation of vascular development and actin cytoskeleton reorganization are upregulated after insulin treatment, pathways of inflammation are downregulated. These pathways are critical to maintain BBB integrity.

6.4. Proteomics and Other OMICs

Untargeted proteomics and metabolomics studies are beginning to emerge in the context of vascular cells in AD. Two main methodologies have been used in these studies: surgical isolation of cerebral vessels manually (Ojo et al. 2021a) or with microdissection of FFPE fixed brain tissue (Hondius et al. 2018; Handa et al. 2022).

6.4.1. Cerebrovascular Proteomic Changes

Owing to technical difficulties, very few studies to date address the cerebrovascular proteomic changes in aging and AD. To investigate CAA related proteomic changes in AD, Ojo et al., utilized a novel protein extraction approach that is compatible with next generation proteomics methods that enable isolation and separation of cerebrovascular cells from human inferior frontal gyrus in AD (n = 20) and control brains (n = 24) (Ojo et al. 2021a). Young (n = 9) and elderly (n = 15) donors were used as controls, while AD donors were divided into CAA scoring groups (n = 10 per group). Their cerebrovascular fractions mostly comprise endothelia (CLEC14A, VWF, Cav1, and PLEC) and pericyte cells. Comparison of the proteomic signature of age matched controls versus AD donors with low CAA scores highlighted 37 significant proteins out of 1271, that are related to cytoskeletal, cytosolic and extracellular cell membrane proteins associated with mitochondrial bioenergetics. This study identified 230 significant cytoskeletal, cytosolic, and extracellular cell membrane proteins associated with cell adhesion molecules, mitochondrial bioenergetics, and nuclear proteins through proteomic comparison of age matched controls versus AD donors with high CAA scores. Similarly, high CAA versus low CAA analysis identified 84 proteins that had roles in cytoskeletal, cytosolic, and extracellular cell membrane, membrane protein biosynthesis, and mitochondrial bioenergetics.

Two studies investigated CAA‐related proteomic changes in AD using laser microdissection‐assisted proteomics techniques on human brain tissue samples. Using fresh‐frozen brain tissue samples, Hondius et al. examined the protein content of the occipital lobe of AD and control cases (nAD = 14, nControl = 6), and found that AD cases had higher clusterin (CLU), apolipoprotein E (APOE), and serum amyloid P‐component (APCS) levels than controls (Hondius et al. 2018). In addition, collagen alpha‐2 (VI) (COL6A2) and norrin (NDP) were identified as CAA specific markers in AD. Further, Handa et al., analyzed proteomics changes in the occipital lobes of AD patients having AD neuropathologic change (ADNC) with or without severe vascular deposition of Aβ (nADNC+/CAA+ = 6, nADNC+/CAA− = 5, and nADNC−/CAA‐ = 5) through laser microdissection (LMD), processed by pressure cycling technology (PCT), and applied to SWATH (sequential window acquisition of all theoretical fragment ion spectra). In this study, they found 56 proteins with more than a 1.5‐fold difference between ADNC+/CAA+ and ADNC −/CAA− donors and 90% of these proteins were specific to vascular pathology. These proteins are associated with fibrosis, oxidative stress, and apoptosis.

In another study, Ojo et al. investigated the proteomic signature of the APOE genotype in AD (n = 26) and control (n = 29) (Ojo et al. 2021b) by applying a previously used technique (Ojo et al. 2021a). Comparison within control brains revealed 217 master proteins significantly changing with APOEε2/2, 260 with APOEε4/4, and 59 with APOEε3/4 compared to APOEε3/3 genotype (Ojo et al. 2021b). The APOEε4 genotype is associated with proteins within EIF2, EIF4, P70S6k, and mTOR signaling pathways, while the APOEε2 genotype is associated with oxidative phosphorylation, EIF2 signaling, and mitochondrial dysfunction compared to APOEε3 genotype. Using AD versus controls with the same genotypes, they identified 102 master proteins within pathways for spliceosomal cycle, DNA methylation, and repair in APOEε2/Eε3; 41 within pathways for mTOR and EIF2 in APOEε3/E3; 192 in APOEε3/E4, and 189 in APOEε4/E4 genotype‐carriers in pathways for EIF2, EIF4, and p70S6k signaling.

6.4.2. Central Vascular Pathology Linked to Peripheral Proteomic and Metabolomic Signatures

Biomarkers in CSF and plasma can be used to trace vascular brain pathology to the periphery. Shi et al. (2021), followed this approach and analyzed 5032 plasma proteins in cognitively healthy individuals (n = 1061) through Somalogic's Somascan assay for their association with hippocampal volume and white matter hyperintensities (WMH), that are known to be associated with cognitive decline and AD progression (Burke et al. 2019; Prins and Scheltens 2015; Jack et al. 2000). One hundred and seventy‐eight proteins showed nominally significant association with WMH, 2 of which remained significant after FDR correction. Another study by Higginbotham et al. identified CSF biomarkers that are associated with AD pathophysiology by using an integrative proteomics approach (Higginbotham et al. 2020). A network analysis was conducted to identify overlapping modules between the brain and the CSF, and 67% of protein changes in the CSF can be traced back to the brain. These overlapping modules include two vascular modules with endothelial cell markers and vascular ontologies (Higginbotham et al. 2020). In addition to proteomic approaches, metabolomic approaches using serum could inform about vascular brain pathology. In their study, Fleszar et al. examined the metabolic changes in the serum of healthy controls, patients with AD and vascular dementia, focusing in particular on the L‐arginine/nitric oxide (NO) pathway (Fleszar et al. 2019). The levels of certain metabolites varied according to the severity and pathological state of vascular dementia and AD.

6.5. Summary

Cerebral amyloid angiopathy and cerebral vascular dysfunction such as BBB dysfunction and reduced cerebral blood flow are examples of AD related vascular pathology. The vascular cell types that support and maintain homeostasis but are perturbed in AD include endothelia and pericytes. Several genes have been associated with vascular function in AD. FMNL2 facilitates gliovascular interactions and has higher expression in AD patients. LINC‐PINT has been associated with decreased CAA in patients without APOEe4 and possible protection against AD. Genes associated with cerebral small vessel disease have also been linked with AD such as TRIM47. There are pleiotropic genes affecting both AD and cardiovascular disease. Vascular damage can also be measured in plasma in AD through assessment of change in sST2 levels. Enrichment of vascular cells has been difficult due to lack of effective isolation mechanisms. Transcription of vascular cells can vary dependent on location (endothelial and fibroblast cells) or function (pericytes). Transcriptional profile of endothelial cells can be influenced by blood vessel size and location of pericytes. Further studies are required to establish subtyping of endothelial cells and pericytes. Loss of vascular cells such as pericytes and their specific downregulation of genes in AD has been observed. Most genes from an AD GWAS study were observed to be associated with blood vessel related vascular cells, which may point to associated dysfunction of the BBB in AD, including downregulation of AD related genes in pericyte and capillary endothelia cells. Many of the differentially expressed genes show association with functions maintaining the BBB and therefore could also be linked with cognitive resilience. In vascular cells, AD related genes involved in processes such as solute transportation, insulin signaling and cell adhesion were mostly associated with capillary endothelia cells. Mouse studies have shown that peripherally expressed APOE can affect the BBB. Insulin signaling can also affect the integrity of the BBB through downregulation of inflammation genes. Analysis of proteomic signatures of high or low CAA scored AD donors revealed that most proteins were associated with cell membrane, cytoskeletal structures, or mitochondrial bioenergetics pathways. Clusterin, apolipoprotein E and serum amyloid P‐component have been found in higher levels in AD versus controls. APOE genotypes have certain proteomic signatures, such as those involved in the mTOR signaling pathway, mitochondrial dysfunction, and oxidative phosphorylation. Protein changes in the CSF can reflect changes in brain vascular pathology.

Across omics approaches, each analysis method has provided valuable insight on vascular cells. As a result of genomics approaches, new vascular genetic targets have been identified that may contribute to AD‐related vascular alterations such as decreased CBF, loss of BBB integrity, and CAA. By using transcriptomics approaches, particularly snRNAseq, researchers have been able to uncover specific alterations in individual cell types, zonation dependent transcriptomic signatures, and the vulnerability of certain types of vascular cells, that is, pericytes in AD. Proteins and pathways implicated novel biological processes and metabolites with potential to serve as therapeutics or biomarkers, which have been identified through proteomics and metabolomics methods. Although studies which combine omics approaches with functional studies involving model systems are limited, systematic prioritizations of vascular molecules followed by validations are beginning to yield molecular mechanisms that may underlie complex perturbations, such as those involving BBB impairment in AD, as exemplified by pericytic‐SMAD3 and astrocytic‐VEGFA interactions at the gliovascular unit.

In spite of advances in omics approaches regarding vascular cells, there remains questions and gaps for these cells in terms of omics learnings in AD. There have been very few studies investigating the vascular epigenetic alteration in AD, which has been an important component of cardiovascular diseases (Shi et al. 2022). It might be possible to connect genomics and transcriptomics through epigenetic studies that examine the methylation, chromatin access, and histone modification profiles of vascular cells. In addition, single nucleus transcriptomic studies should be conducted to assess transcriptomic changes within different brain regions. In AD, vascular alterations are not only confined to one type of cell, but are coordinated by multiple types of gliovascular unit cells. In light of this, more studies focusing on the interactome of vascular cells are needed. It would be beneficial to use spatial transcriptome approaches to understand cell‐to‐cell communication in AD brains, in the future (Figure 1).

7. Review Summary

Omics studies provide valuable insight into the brain cell type landscape and its related functions and pathology in AD. Each layer of omics helps reveal and lead to further investigations for our understanding of the molecular underpinnings of AD pathology. This review has discussed a plethora of GWAS and transcriptomic studies for each of the glia cell types and the need for further research especially for certain cell types (vascular) and relatively understudied or new approaches such as spatial transcriptomics and epigenetic approaches. Of the cell types discussed in this review, astrocytes and microglia have the most published studies with the vast majority of omics based studies focused on them (Figures 1 and 2). Within especially the proteomics and transcriptomics studies on both these cell types, biodomains such as immune response, amyloid, APOE effects and cell activation/state were most investigated and/or uncovered (Figure 1). The cell types with the least omics data available are oligodendrocytes and vascular cells (Figure 1). Most of the transcriptomics and proteomics studies on oligodendrocytes have focused on myelination and cell function, while in vascular cells the biodomains involve vasculature, amyloid effects and cell activation (Figure 1). Biodomains that require further investigation within each cell type are resilience, epigenetics, biomarkers and tau homeostasis (Figure 1). Many aspects of omics studies require additional examination, such as sex specific changes related to glial cell type, optimization of glial cell sorting, heterogeneity in sub states (especially astrocytes and microglia), vascular cell location and perturbed function, gliovascular interactions and increased omics‐based studies of understudied cell types in AD.

There are a number of key takeaways from this review (Figures 1 and 2). Literature has shown astrocytes and microglia to have homeostatic versus activated states that may play crucial roles in AD pathogenesis. The APOE genotype can influence presence of astrocytic numbers, while APOE expression can influence apoptosis/phagocytosis (astrocytes), function of the BBB (vascular cells) and APOE accumulation (astrocytes). Many high risk AD genetic variants are enriched in microglia and age of AD onset SNPs in oligodendrocyte genes. Differential methylation is present in AD oligodendrocytes and methylation patterns are influenced by the presence of AD tau pathology in microglia. As astrocytic cells increase around amyloid plaques, there is a depletion of oligodendrocytes around these plaques. Current scientific literature has focused heavily on astrocytic and microglia based omics studies and although there is further room for expansion in these areas, there is a need for omics based studies on oligodendrocytes and vascular cells in AD research. Finally, integrative omics studies that combine multiple omics types, that focus on functional multi‐cellular units (such as the gliovascular unit) and that provide a systematic prioritization and functional validation of identified multi‐omics changes are necessary to capture the full and holistic picture of AD‐related glial perturbations. These findings are expected to pave the way for precision therapies, centrally‐linked peripheral biomarkers and better model systems for AD.

Author Contributions

Nilüfer Ertekin‐Taner and Özkan İş conceptualized the study; Özkan İş, Xue Wang, Stephanie R. Oatman performed literature search; Yuhao Min, Xue Wang, Stephanie R. Oatman performed methodology to curate articles; Özkan İş, Yuhao Min, Xue Wang, Stephanie R. Oatman, Ann Abraham Daniel performed article screening, Özkan İş, Yuhao Min, Xue Wang, Stephanie R. Oatman, Ann Abraham Daniel, Nilüfer Ertekin‐Taner wrote, reviewed and edited the original draft of manuscript; Özkan İş, Yuhao Min, Stephanie R. Oatman, Ann Abraham Daniel created visuals; Özkan İş and Nilüfer Ertekin‐Taner supervised the project; Nilüfer Ertekin‐Taner oversaw the study and provided direction, funding, and resources.

Conflicts of Interest

The authors declare no conflicts of interest.

Data Availability Statement

Data sharing is not applicable to this article as no new data were created or analyzed in this study.