Epigenetic dysregulation in Alzheimer’s disease peripheral immunity

Abstract

The peripheral immune system in Alzheimer’s disease (AD) has not been thoroughly studied with modern sequencing methods. To investigate epigenetic and transcriptional alterations to the AD peripheral immune system, we used single cell sequencing strategies, including assay for transposase-accessible chromatin and RNA sequencing. We reveal a striking amount of open chromatin in peripheral immune cells in AD. In CD8 T cells, we uncover a cis-regulatory DNA element co-accessible with the CXC motif chemokine receptor 3 gene promoter. In monocytes, we identify a novel AD-specific RELA transcription factor binding site adjacent to an open chromatin region in the nuclear factor kappa B subunit 2 gene. We also demonstrate apolipoprotein E genotype-dependent epigenetic changes in monocytes. Surprisingly, we also identify differentially accessible chromatin regions in genes associated with sporadic AD risk. Our findings provide novel insights into the complex relationship between epigenetics and genetic risk factors in AD peripheral immunity.

Graphical Abstract

eTOC Blurb

Ramakrishnan et al. provide a single-cell genomics resource for exploring the peripheral immune system in Alzheimer’s disease. Their extrapolated findings provide novel insights into the complex relationship between epigenetics and genetic risk factors in Alzheimer’s disease peripheral immunity.

Introduction

Alzheimer’s disease (AD) is the most common cause of dementia in older adults and is a major public health concern globally. The development of AD is complex and involves both genetic and environmental factors¹. In recent years, genome-wide association studies (GWAS) and whole genome sequencing analyses have uncovered common and rare genetic variants that are associated with sporadic AD^2–6. These findings implicate brain microglia as important players in the development and progression of AD⁷. However, despite many AD risk factor genes also being expressed by peripheral immune cells⁸, the exact contribution of peripheral immunity to AD is not well understood. Further, while several studies have detected peripheral immune alterations in AD^9–12, this important immune reservoir has not been thoroughly studied with modern, multi-omic sequencing modalities.

The apolipoprotein E (APOE) gene is the main genetic risk factor for age-related sporadic AD¹³. ApoE is involved in the transport and metabolism of lipids in the body¹⁴. It exists in three common allelic forms: APOE ε2, APOE ε3, and APOE ε4. The APOE ε4 allele is the most common risk factor for AD, with carriers of one APOE ε4 allele having a threefold increased risk of developing AD compared to non-carriers, and carriers of two APOE ε4 alleles having a 12-fold increased risk¹⁵. The precise mechanism by which APOE ε4 increases the risk of AD is not fully understood, but it is thought to be related to its effects on amyloid-β (Aβ) metabolism and neuroinflammation¹⁶.

In addition to its role in lipid metabolism, ApoE has also been shown to be involved in immune function. ApoE is expressed by immune cells and regulates their activation and differentiation by modulating the production of cytokines and other immune mediators^17–21. These findings suggest that ApoE may play a role in the immune response in AD and that it may be a potential target for immune-based therapies.

To study the influence of genetic risk factors and APOE genotypes on the peripheral immune system in AD, we analyzed a cohort of AD patients with various APOE genotypes and age-matched healthy controls (HCs). We isolated immune cells from peripheral blood samples and used single cell sequencing strategies to analyze the immune landscape in each group. These strategies included single cell assay for transposase-accessible chromatin (ATAC) sequencing (scATACseq), single cell RNA sequencing (scRNAseq) and T cell receptor (TCR) sequencing (scTCRseq). This multi-omic approach allowed us to measure the accessibility of the genome at the single cell level, providing insights into gene regulatory processes and expression changes associated with genetic risk factors and APOE genotypes in AD. We uncovered epigenetic and transcriptional changes associated with peripheral monocytes and memory CD8⁺ T cells in AD. We also identified differentially accessible chromatin regions with concordant gene expression alterations in inflammatory cytokine genes that were specific to APOE ε4 carriers.

Finally, we analyzed familial and sporadic AD genetic risk factors for chromatin changes in peripheral immune cells. We reveal a surprising number of AD risk genes with chromatin accessibility alterations in various immune populations.

Through these approaches, we provide a comprehensive understanding of the influence of epigenetics and AD genetic risk factors on the peripheral immune response in AD. Our results also provide new insights into the relationship between APOE genotypes and the immune system in AD. As a resource to the AD and immunology fields, our full scRNAseq dataset can be explored online using a data portal located at https://gatelabnu.shinyapps.io/ad_apoe_rna. Altogether, this resource and our extrapolated results have the potential to inform the development of novel immune-based therapies for this devastating neurodegenerative disease.

Results

scATACseq uncovers epigenetic open chromatin regions in AD peripheral immunity

We first established a cohort of 29 AD patients and 26 age and sex-matched HCs with different APOE genotypes (Table 1 and Figure S1A). The HC cohort consisted of 9 APOE ε3/ε3 (4 female, 5 male), 10 APOE ε3/ε4 (4 female, 6 male), and 7 APOE ε4/ε4 (5 female, 2 male) subjects. The AD cohort consisted of 10 APOE ε3/ε3 (4 female, 6 male), 11 APOE ε3/ε4 (6 female, 5 male), and 9 APOE ε4/ε4 (4 female, 5 male) patients. Age distributions of each cohort per assay are displayed in Figure S1A. We measured cerebrospinal fluid (CSF) and plasma biomarkers, which indicated increased levels of phosphorylated Tau at residue 181 (pTau181) (Figure S1B) and total Tau as well as their altered ratios to Aβ species (Table 1). Differences in pTau181 were also evident when stratifying HCs and AD patients by APOE genotype, but were not significantly different by APOE genotype within AD subjects (Table S1). We then generated multi-omic single cell immune profiles of each subject using a variety of bioinformatics approaches (Figure 1A). To generate immune profiles, we isolated peripheral blood mononuclear cells (PBMCs) and performed scATACseq and scRNAseq on the same aliquot of cells. Altogether, we analyzed 476,029 cells by scATACseq and 270,884 cells by scRNAseq (Figure 1B). Our full scRNAseq dataset can be explored online using a data portal accessible via a quick response code linking to https://gatelabnu.shinyapps.io/ad_apoe_rna (Figure 1C).

Table 1.

Demographic table containing blood and CSF biomarker measurements of all study subjects.

		Healthy Control	Alzheimer’s disease	P
Demographics		n = 26	n = 30
Sex, n (%)	female	13 (50.0)	14 (46.7)	1.000
	male	13 (50.0)	16 (53.3)
Age	Median (IQR)	72.0 (68.2 to 78.8)	73.5 (66.0 to 81.0)	0.882
Race, n (%)	Asian	1 (3.8)	2 (6.7)	0.506
	Black or African	1 (3.8)
	American
	White	24 (92.3)	28 (93.3)
APOE Genotype		n = 26	n = 30
	ε3/ε3	9 (34.6)	10 (33.3)	0.968
	ε3/ε4	10 (38.5)	11 (36.7)
	ε4/ε4	7 (26.9)	9 (30.0)
CSF Biomarkers (pg/mL)		n = 12	n = 10
CSF pTau181	Median (IQR)	48.0 (32.4 to 63.1)	105.4 (83.9 to 163.4)	0.005
CSF Total Tau	Median (IQR)	302.1 (216.2 to 450.0)	720.2 (560.6 to 1117.8)	0.001
CSF Aß42	Median (IQR)	695.7 (567.8 to 883.2)	556.0 (497.5 to 634.8)	0.166
CSF Aß40	Median (IQR)	9286.5 (8625.5 to 11675.5)	9156.0 (7913.2 to 9776.8)	0.692
CSF Aß42/Aß40	Median (IQR)	0.1 (0.1 to 0.1)	0.1 (0.1 to 0.1)	0.187
Aß42/Total Tau	Median (IQR)	2.1 (1.8 to 3.3)	0.6 (0.4 to 1.0)	0.002
Aß40/Total Tau	Median (IQR)	30.8 (25.8 to 36.5)	11.6 (8.0 to 15.2)	<0.001
Aß42+Aß40/Total Tau	Median (IQR)	33.0 (27.8 to 39.9)	12.3 (8.4 to 16.2)	<0.001
Aß42/pTau	Median (IQR)	13.4 (11.6 to 23.1)	4.3 (3.2 to 6.6)	0.004
Aß42/pTau	Median (IQR)	200.8 (178.3 to 248.8)	75.5 (56.5 to 101.5)	0.001
Aß42+Aß40/pTau	Median (IQR)	215.0 (191.9 to 268.8)	79.9 (59.7 to 108.0)	0.002
Plasma Biomarkers (pg/mL)		n = 23	n = 27
Plasma pTau181	Median (IQR)	1.9 (1.7 to 2.3)	3.5 (2.9 to 4.2)	<0.001

Nonparametric Wilcoxon rank sum and Chi square tests comparing measurements between HC and AD groups stratified by APOE genotype; median and IQR. Values shown for biomarkers are in pg/ml.

Figure 1. Study design and resource generation.

(A) Schematic depicting experimental design and bioinformatics pipeline of the study. (B) Uniform manifold approximation and projection of scATAC cells (left) labeled using matched scRNA cells (right) labeled using a CITE-Seq atlas. (C) Scan QR code to view interactive ShinyCell application hosting RNA and TCR datasets. (D) Schematic portraying study approach; methods include analysis of disease status and the effects on APOE genotype-associated epigenomes and transcriptomes (top) and analysis between APOE genotypes and effects on disease-associated epigenomes and transcriptomes (bottom).

We performed initial analyses of scATACseq data using a software suite for single-cell analysis of regulatory chromatin in R (ArchR)²². Our quality-control results indicated consistent numbers of median DNA fragments and transcription start site enrichment per sample (Figure S1C). Imputed gene scores for canonical cell type markers validated results of unsupervised clustering, with improved cell type identification following integration of the scRNAseq dataset (Figure S1D–E). Our gene score matrix is presented in Figure S1F. Motif enrichment performed on differentially accessible peaks between cell types demonstrated expression of expected cell-type markers (Figure S1G). Finally, we observed even distributions of cells by disease status, APOE genotype, and sex (Figure S1H–J).

To annotate cell clusters, we used weighted-nearest neighbor analysis using a cellular indexing of transcriptomes and epitopes by sequencing (CITE-seq) dataset (Figure S2A)²³. We then performed quality control measures of our scRNAseq dataset using Seurat²³. We obtained consistent numbers of counts (Figure S2B) and features per sample (Figure S2C). Further, numbers of features and percent mitochondrial reads per diagnosis were indistinguishable (Figure S2D). We removed small amounts of ambient RNA contamination with SoupX²⁴ (Figure S2E). Annotated scRNAseq clusters demonstrated robust expression of key marker genes (Figure S2F). Additionally, we observed even distributions of cells by sex, disease status, and APOE genotype (Figure S2G). Lastly, we observed similar proportions of cell types by scATACseq and scRNAseq (Figure S2H). Cell counts for each assay were also not significantly different between groups at the level of individuals (Figure S2I). We utilized this dataset to measure the influence of AD on the APOE genotype-associated transcriptome as well as the influence of APOE genotype on the AD-associated transcriptome (Figure 1D).

Next, we used ArchRtoSignac²⁵ to convert our ArchR project to a Signac²⁶ object and identified differentially accessible peaks between AD and control PBMC cell types. We quantified differentially accessible regions (DARs) of chromatin between AD and HCs by cell type using single cell logistic regression (LR) and pseudobulk (DESeq2)²⁷ methods. We then analyzed the overlapping regions from these two statistical measures (Figure 2A). Importantly, since age²⁸ and ethnicity²⁹ are known to influence immunity, we controlled for these covariates in all analyses. Notably, CD8 T cells had the highest number of DARs (Figure 2A and Table S2). Associating each DAR with its location within the genome revealed most peaks were located in promoter and intronic regions and were more accessible in AD than HC for all cell types (Figure 2B). Pathway analysis on CD8 T cells (the most altered cell type by chromatin accessibility) revealed increased activation and cytokine signaling and decreased cellular senescence (Figure 2C). Although CD8 T cell dysfunction and activation can be transcriptionally intertwined³⁰, these results demonstrate an epigenetic state of CD8 T cells indicative of increased activation and reduced senescence.

Figure 2. Epigenetic dysregulation in AD peripheral immunity and concordance with differential gene expression.

(A) DARs of chromatin between AD and HCs by cell type using single cell LR and pseudobulk DESeq2 methods. The overlapping genes from these two statistical measures are used in all downstream analyses. UpSet plot shows unique and shared DARs by cell type with a corresponding heat map showing the ratios of DARs to regions tested. (B) Proportion of AD versus HC DARs by peak type and direction of accessibility. (C) Pathway analysis of AD vs HC DARs of CD8 T cells [HC cells = 57326, median interquartile range (IQR) = 1969 (963.25 – 2771.25), AD cells = 53974, median (IQR) = 1304 (668.75 – 2682)]. indicating increased activation and cytokine signaling and reduced cellular senescence in AD. (D) DEGs between AD and HCs by cell type using MAST and pseudobulk edgeR methods. The overlapping genes from these two statistical measures are used in all downstream analyses. UpSet plot shows unique and shared DEGs by cell type with a corresponding heat map showing the ratios of DEGs to genes tested. (E) Overlapping significant genes by scATACseq and scRNAseq by cell type indicating CD14 and CD16 monocytes and CD8 T_EM cells as having the most overlapping genes with altered chromatin accessibility and altered gene expression in AD.

Our multi-omic strategy was designed to allow us to jointly profile chromatin accessibility and gene expression changes associated with the peripheral immune system in AD. We thus quantified differentially expressed genes (DEGs) between AD and HCs by cell type using Model-based Analysis of Single-cell Transcriptomics (MAST)³¹ and pseudobulk (edgeR)³² methods. We then analyzed the overlapping genes from these two statistical measures (Figure 2D). Differential expression analysis of the full transcriptome revealed innate immune CD14 and CD16 monocytes as the most dysregulated cells type in AD (Figure 2D and Table S3). We next quantified the overlapping significant genes by scATACseq (LR and DESeq2) and scRNAseq (MAST and edgeR) to detect cell types with genes containing altered chromatin accessibility and altered expression in AD, respectively. Here, we noted CD14 and CD16 monocytes and CD8 T effector memory (T_EM) cells as having the most overlapping genes with altered chromatin accessibility and altered gene expression in AD (Figure 2E and Table S3).

Epigenetic changes to NF-κB signaling molecules in peripheral monocytes of AD patients

We next focused our attention on the genes that showed chromatin accessibility changes and gene expression changes in CD14 and CD16 monocytes. Notably, we detected an open chromatin region in the nuclear factor kappa B subunit 2 (NFκB2) gene that was associated with increased expression in CD14 and CD16 monocytes (Figure 3A). Intriguingly, transcription factor motif scanning of monocyte DARs revealed enrichment of the major transactivating NF-κB subunits, REL and RELA, in accessible chromatin regions in AD (Figure 3B and Table S4). Moreover, transcription factor footprinting of monocytes revealed increased signals centered around motif binding sites for each transcription factor in AD (Figure S3A). To further interrogate transcription factor binding in monocytes, we utilized a deep neural network model for genome-wide transcription factor binding site (TFBS) prediction called maxATAC³³. We used maxATAC to predict high-resolution TFBSs in AD monocytes resolved to 32 base pairs. We first focused on identifying TFBSs in the NFκB2 gene that were unique to AD monocytes (Table S4). We identified a RELA binding site within NFκB2 that was unique to AD monocytes (Figure 3C). We then confirmed this AD-specific RELA binding site was directly adjacent and slightly overlapping with the previously identified DAR in NFκB2 (Figure 3D). Further, pathway analysis of upregulated CD14 and CD16 monocyte genes indicated enrichment of NFκB signaling (Figure 3E).

Figure 3. Epigenetic changes to NF-κB signaling molecules in peripheral AD monocytes.

(A) Scatterplots showing AD vs. HC DARs intersecting DEGs by fold change in CD14 and CD16 monocytes. (B) Transcription factor motif scanning analysis showing enrichment of REL and RELA binding sites in AD monocytes, where transcription factors are ranked by product of log(p-value) and log2(fold-change). (C) maxATAC analysis of TFBS in the NFκB2 gene indicating an AD-specific RELA binding site. (D) Chromatin track of the NFκB2 gene demonstrating the location of the AD-specific RELA binding site. The RELA binding site (red) is adjacent to an NFκB2 DAR (light blue). (E) Pathway analyses of upregulated DEGs of AD versus HC CD14 and CD16 monocytes indicating enrichment of NFκB signaling. ATAC-seq assay: HC monocytes = 9763, median (IQR) = 311 (163.25 – 451.75), AD monocytes = 7162, median (IQR) = 242.5 (112.5 – 397.75). RNA-seq assay: HC CD14 monocytes = 4414, median (IQR) = 105 (55 – 254.25), AD CD14 monocytes = 3455, median (IQR) = 61.5 (11 – 148), HC CD16 monocytes = 1837, median (IQR) = 42 (5 – 108), AD CD16 monocytes = 899, median (IQR) = 13 (4 – 41.25).

Notably, by applying maxATAC TFBS prediction across the entire genome in monocytes, we found a high number of SET Domain Bifurcated Histone Lysine Methyltransferase 1 (SETDB1) binding sites that were unique to AD in monocytes (Figure S3B). SETDB1 is a histone methyltransferase that serves as an epigenetic checkpoint³⁴, suggesting SETDB1 involvement in epigenetic modification to the peripheral AD immune system. Altogether, our multi-omic analysis uncovered epigenetic alterations to NF-κB signaling in AD and an epigenetic modifier in peripheral monocytes of AD patients.

Cis-regulatory sequence accessibility correlates with quantity of transcription of select CD8 T cell and monocyte genes

We next aimed to identify co-accessible pairs of DNA elements using our scATACseq data to connect regulatory elements to their putative target genes. We thus applied Cicero³⁵ to investigate how dynamically accessible elements might orchestrate gene regulation in monocytes and CD8 T cells in AD. We assessed chromatin regions co-accessible with gene promoters using a 0.01 co-accessibility threshold³⁶ (Table S5). Next, we investigated which regions’ accessibility was correlated to expression of the co-accessible promoter region. By this analysis, in the top five cell types by number of AD correlations, CD14 monocytes and CD8 T_EM cells showed the largest increases in significant correlations between AD and HC (Figure 4A). Since CD8 T_EM cells had a large number of chromatin alterations connected to gene promoters, we next assessed which cre-linked genes were also DEGs between AD and control CD8 T_EM cells (Figure 4B). Intriguingly, we detected an open chromatin region distal to the CXC motif chemokine receptor 3 (CXCR3) gene that was co-accessible with the CXCR3 promoter (Figure 4C). Next, we used flow cytometry to measure CXCR3 protein expression among various CD8 T cell phenotypes in a separate cohort of patients diagnosed with mild cognitive impairment or AD. We noted protein expression was associated with CD8⁺ T_EM CD45RA⁺ (T_EMRA) cells (Figure S4A–B). Notably, we previously identified CD8 T_EMRA cells as clonally expanded in the CSF of AD patients⁹. Here, we also identified CXCR3⁺CD3⁺CD8⁺ T cells in post-mortem AD hippocampus (Figure 4D) and leptomeninges (Figure 4E).

Figure 4. Cis-regulatory sequence correlates with quantity of transcription of CXCR3 in CD8 T cells, which home to the AD brain.

(A) Number of gene promoter to region connections filtered for only significant correlations with corresponding gene expression [> 95th % of Pearson correlation coefficient (PCC) per cell type; P-adjusted<0.05]. (B) Overlap of significant promoter-region correlations with AD vs HC DEGs in CD8 T_EM cells. (C) Representative Cicero CXCR3 promoter-region connections (bottom; HC in gray, AD in red). CXCR3 distal region accessibility (CD8 T cells) to gene expression (CD8 T_EM cells) correlation in AD samples (top). (D) CXCR3-expressing CD8 T cells in post-mortem AD hippocampus. Arrowheads indicate CXCR3⁺CD3⁺CD8⁺ T cells. Scale bar=15 μm. (E) CXCR3⁺CD3⁺CD8⁺ T cells in post-mortem AD leptomeninges. Scale bar=20 μm. (F) A CD3⁺ T cell interacting with an Aβ plaque-associated Iba1⁺ microglial cell. Scale bar=10 μm.

Notably, CXCR3 signaling mediates development of AD-like pathology in mouse models³⁷. Mice lacking CXCR3 have reduced concentrations of proinflammatory cytokines in their brains and attenuated behavioral deficits³⁷. Recently, CXCR3 was found to mediate T cell homing to human AD three-dimensional cultures³⁸. Our results reveal a potential contribution of epigenetic modification to CXCR3 that influences its expression in CD8 T_EM cells in humans with AD. While the function of T cells in the AD brain remains unclear, we observed them interacting with Aβ plaque-associated microglia in AD post-mortem tissue (Figure 4F). Cumulatively, these results suggest that the upregulated expression of CXCR3 via epigenetic alteration might promote T cell homing to the AD brain where they modulate microglial function.

We next assessed which cre-linked genes were also DEGs between AD and control CD14 monocytes (Figure S5A). Notably, ATP binding cassette subfamily A member 1 (ABCA1) was among the cre-linked genes that was also an upregulated DEG. We detected an intronic region of ABCA1 that was co-accessible with the ABCA1 promoter in AD monocytes (Figure S5B). ABCA1 plays an important role in the lipidation of ApoE^39,40. Moreover, some have postulated that variants in ABCA1 alter sporadic AD risk^41–46, while others have argued that variants do not influence risk⁴⁷. Interestingly, due to its role in the efflux of phospholipids and cholesterol from cells, some have argued that ABCA1 variants play a role in AD development in combination with APOE ε4 genotype⁴³.

APOE genotype-dependent innate immune dysregulation in AD

To interrogate the relationship between APOE and the peripheral immune system, we compared AD and HC subjects of each genotype. We noted stark differences in the numbers of DARs by APOE genotype comparison. Specifically, there were more unique DARs when comparing AD APOE ε4/ε4 carriers to HC APOE ε4/ε4 carriers versus other APOE genotype comparisons (Figure 5A and Table S6). In fact, we detected a step-wise, APOE ε4 gene dose-dependent increase in the number of DARs for nearly all peripheral immune cell types (Figure 5A). We next quantified the overlapping significant genes by scATACseq (LR and DESeq2) and scRNAseq (MAST and edgeR) to detect cell types with genes containing altered chromatin accessibility and altered gene expression by APOE genotype (Table S7). Here, we noted a similar step-wise, APOE ε4 gene dose-dependent increase in the number of genes in CD14 monocytes (Figure 5B). Pathway analysis of CD14 monocyte DEGs in AD versus HC APOE ε4/ε4 carriers signaled enrichment of NFκB signaling and inflammatory response (Figure S6). Thus, our scATACseq analysis uncovered APOE genotype-specific epigenetic chromatin accessibility and transcriptomic differences in peripheral immune cells of AD patients.

Figure 5. APOE genotype-dependent innate immune dysregulation in AD.

(A) UpSet plot of AD vs. HC APOE genotype comparisons indicating increasing number of unique DARs by number of APOE ε4 alleles. Corresponding heatmap shows DARs by cell type for each APOE genotype comparison between AD and HC subjects. (B) Total number of overlapping DARs and DEGs by cell type for each APOE genotype comparison between AD and HC subjects indicating increasing number of genes in AD monocytes by number of APOE ε4 alleles. (C) Scatterplots showing DARs intersecting DEGs by fold change in CD14 monocytes for each AD vs. HC APOE genotype comparison. (D) Representative chromatin tracks of DARs intersecting DEGs in AD vs. HC APOE ε4/ε4 carriers. For each gene, significant DARs are shown in grey. (E) Transcription factor motif analysis in monocytes indicating enrichment of inflammatory transcription factors in AD APOE ε4/ε4 carriers, where transcription factors are ranked by product of log(p-value) and log2(fold-change). Group sizes for all comparisons: n = 9 HC APOE ε3/ε3, 10 HC APOE ε3/ε4, 7 HC APOE ε4/ε4, 10 AD APOE ε3/ε3, 11 AD APOE ε3/ε4, 8 AD APOE ε4/ε4. ATAC-seq assay: HC APOE ε3/ε3 monocytes = 2647, median (IQR) = 227.5 (135.75 – 542.5), HC APOE ε3/ε4 monocytes = 3436, median (IQR) = 275 (250 – 410), HC APOE ε4/ε4 monocytes = 3680, median (IQR) = 357 (296 – 657), AD APOE ε3/ε3 monocytes = 1968, median (IQR) = 217.5 (163 – 280), AD APOE ε3/ε4 monocytes = 3734, median (IQR) = 299 (114 – 470), AD APOE ε4/ε4 monocytes = 1460, median (IQR) = 214 (58 – 291). RNA-seq assay: HC APOE ε3/ε3 CD14 monocytes = 833, median (IQR) = 75 (61.5 – 127), HC APOE ε3/ε4 CD14 monocytes = 2255, median (IQR) = 127 (11 – 262), HC APOE ε4/ε4 CD14 monocytes = 1326, median (IQR) = 198.5 (85.75 – 281.25), AD APOE ε3/ε3 CD14 monocytes = 1607, median (IQR) = 62 (14 – 166), AD APOE ε3/ε4 CD14 monocytes = 1334, median (IQR) = 115 (59.5 – 168), AD APOE ε4/ε4 CD14 monocytes = 514, median (IQR) = 11 (3.5 – 40).

We next assessed the overlap of DARs and DEGs for CD14 monocytes to determine if chromatin changes were associated with gene expression alterations. Among the more accessible DARs associated with increased gene expression in CD14 monocytes of APOE ε4/ε4 carriers were inflammatory cytokines C-C motif chemokine ligand 4 like 2 (CCL4L2), CC motif chemokine ligand 3 like 1 (CCL3L1) and CXC motif chemokine ligand 2 (CXCL2) (Figure 5C). Notably, these cytokines have been implicated in microglial inflammation in AD⁴⁸. We also uncovered an open chromatin region associated with increased transcription of the aforementioned ABCA1 gene in APOE ε4/ε4 carriers (Figure 5C). We then confirmed increased accessibility of each of these genes by generating chromatin tracks of their location in the genome (Figure 5D). Finally, transcription factor enrichment analysis of CD14 monocytes across APOE genotypes revealed several transcription factors involved in cytokine gene transcription enriched in APOE ε4/ε4 carriers. These transcription factors included FOS like 1 (FOSL1), FOS like 2 (FOSL2), JUN and JUNB (Figure 5E). Cumulatively, these data indicate epigenetic innate immune dysregulation related to a pro-inflammatory response in AD APOE ε4/ε4 carriers.

Altered chromatin accessibility of AD risk genes in the peripheral immune system

In addition to APOE, GWAS have identified many AD risk genes involved in inflammation and immunity^2–6. Much attention has been paid towards the role of these genes in the immune response to Aβ and its clearance by microglia⁷. Yet, we identified numerous DARs in established AD risk genes in several peripheral immune cell types of AD subjects (Figure 6A). Notably, some DARs were present in multiple cell types, like those within inositol polyphosphate-5-phosphatase D (INPP5D) in B and NK cells, while others were present in only one cell type, like clusterin (CLU) in CD8 T cells (Figure S7A). We next assessed DARs in AD risk genes that distinguished APOE ε4/ε4 AD carriers from HC APOE ε4/ε4 carriers. We again noted numerous DARs in several peripheral immune cell types of AD APOE ε4/ε4 subjects (Figure 6B). Interestingly, among the DARs that distinguished AD APOE ε4/ε4 carriers from HC APOE ε4/ε4 carriers was ABCA1 (Figure 6C), suggesting a convergence of APOE ε4/ε4 genotype and ABCA1 epigenetics in AD.

Figure 6. Epigenetic dysregulation of AD risk genes in the peripheral immune system.

(A) Heatmap depicting number of DARs in AD risk genes by cell type comparing AD vs. HC. (B) Heatmap depicting number of DARs in AD risk genes by cell type comparing AD vs. HC APOE ε4/ε4 carriers (n = 7 HC APOE ε4/ε4, 8 AD APOE ε4/ε4). (C) Chromatin track of ABCA1 showing locations of significant DARs in various peripheral immune cell types of AD vs. HC APOE ε4/ε4 carriers (n = 7 HC APOE ε4/ε4, 8 AD APOE ε4/ε4). (D) Chromatin track of BIN1 showing locations of significant DARs in various peripheral immune cell types of AD vs. HC APOE ε4/ε4 carriers (n = 7 HC APOE ε4/ε4, 8 AD APOE ε4/ε4). (E) Chromatin track of BIN1 showing the significant DARs in CD8 T cells of AD vs. HC APOE ε4/ε4 carriers (n = 7 HC APOE ε4/ε4, 8 AD APOE ε4/ε4). (F) Uniform manifold approximation and projection plot of clonal and non-clonal T cells from scTCRseq analysis of scRNAseq data. (G) Volcano plot showing BIN1 upregulation in clonal CD8 T_EM cells in AD vs. HC APOE ε4/ε4 carriers [n = 7 HC APOE ε4/ε4, 8 AD APOE ε4/ε4. HC APOE ε4/ε4 clonal CD8 TEMs = 4019, median (IQR) = 652.5 (485.75 – 942.25), AD APOE ε4/ε4 clonal CD8 TEMs = 7120, median (IQR) = 526.5 (185.25 – 1264.25)].

We also noted in APOE ε4/ε4 carriers several AD-specific DARs in bridging integrator 1 (BIN1), which is the second-highest genetic rick factor for sporadic AD after APOE⁶ (Figure 6D). We confirmed two DARs in APOE ε4/ε4 CD8 T cells by generating a chromatin track of BIN1 and the corresponding DARs (Figure 6E). The memory T cell pool is a dynamic repository of antigen-experienced T cells that results from the clonal expansion of TCRs⁴⁹. We thus analyzed scTCRseq data to determine whether AD risk factor genes were associated with T cell clonal expansion. We annotated cells as non-clonal or clonal based on whether they shared an identical TCRαβ sequence with another cell (Figure 6F). We then compared clonal T cell transcriptomes between AD and HC for each APOE genotype (Figure S7B and Table S8). Interestingly, we noted that BIN1 was differentially upregulated by clonal CD8 T_EM cells in AD versus HC APOE ε4/ε4 carriers (Figure 6G). Further, the same BIN1 DARs were associated with upregulated expression of BIN1 in AD CD8 T_EM cells when comparing DARs to DEGs (Figure S7C). Relatedly, we noticed that a DAR in an intronic region of the aforementioned CXCR3 was also associated with upregulated CXCR3 expression in CD8 T_EM cells (Figure S7C–D). Altogether, these results signal APOE genotype influence on altered chromatin accessibility in genes historically linked to AD via genetics analysis.

Adaptive immune dysregulation in memory CD8⁺ T cells in AD

We next sought to determine whether the epigenetic dysregulation observed in memory T cells occurred concomitantly with altered gene expression in AD. To do so, we cross-compared all APOE genotypes between AD subjects and HCs to identify cell types and genes that were specifically altered in AD. Comparing the log₂ fold-change of number of DEGs in AD versus number of DEGs in HCs across APOE genotype comparisons revealed CD8⁺ T central memory (T_CM) cells as consistently altered across APOE genotype comparisons in AD (Figure S8A). Notably, the genes upregulated in APOE genotype comparisons within AD subjects included those involved in TCR signaling and T cell migration, such as JUN, JUNB, KLF transcription factor 6 (KLF6), aquaporin 3 (AQP3) and S100 calcium binding protein A11 (S100A11) (Figure S8B). Since we observed APOE allele-specific influence of gene regulation in CD8 T_CM cells, we next studied the transcriptomes of clonally expanded T cells associated with various APOE genotypes within AD and HC groups (Table S9). We then compared DEGs between AD and HC clonally expanded T cell subsets. These results revealed disparate levels of differential expression among clonal CD8 T_CM subsets by APOE genotype comparisons within AD subjects (Figure S8C). Yet, comparisons across APOE genotypes within clonal CD8 T_CM cells of AD subjects again revealed a host of genes involved in TCR signaling and T cell migration, including JUN, JUNB, KLF6, AQP3 and S100A11 (Figure S8D). These results suggest that T cell clonal expansion drives the upregulated expression of genes involved in TCR signaling and T cell migration in AD. Altogether, these results demonstrate transcriptomic changes to the peripheral clonal T cell pool and indicate disruption of peripheral adaptive immunity in AD.

Discussion

Our transcriptomic resource provides novel insight into alterations to the peripheral immune system in AD. Importantly, we validated many of the cell-type-specific DEGs at the protein level using a published dataset comparing AD to HC plasma proteomes⁵⁰ (Figure S9). Thus, our dataset allows for the identification of specific cell-types that produce differentially expressed plasma proteins. We also reveal a surprising increase in open chromatin regions in peripheral immune cells of AD patients. Specifically, we uncover a novel AD-specific RELA binding site in the NFκB2 gene. APOE ε4 has been shown to enhance brain inflammation by modulation of the NFκB signaling cascade^21,51,52. Our results indicate a potential feedback loop in which epigenetic modification to NFκB2 associated with binding of RELA modulates the NFκB signaling cascade in AD peripheral monocytes. Yet, this hypothesis warrants further experimental investigation. We also reveal cis-regulatory elements associated with transcriptome abundance in various peripheral immune cell types in AD. We highlight CD8 T cell CXCR3 and monocyte ABCA1 as cre-linked genes associated with differential expression. Intriguingly, we found an influence of APOE genotype on chromatin accessibility of pro-inflammatory cytokine genes CCL4L2, CCL3L1 and CXCL2 in monocytes. These data suggest an enhanced peripheral immune inflammatory response in AD APOE ε4/ε4 carriers that may influence AD risk or disease severity in these subjects.

Historically, the majority of research effort surrounding neuroinflammation in AD has centered around brain microglia. Yet, we identify numerous epigenetic open chromatin modifications to AD-associated genes in several peripheral immune subtypes. Perhaps most surprising is that we identify an open chromatin region in BIN1 that corresponds to upregulated gene expression in clonally expanded CD8 T_EM cells. We also identify an influence of APOE genotype on AD risk genes within AD subjects. Specifically, APOE ε4/ε4 carriers harbor numerous DARs in AD risk genes. We highlight INPP5D as a gene with multiple DARs across several peripheral immune cell types, while a DAR in CLU was associated with only CD8 T cells. These findings suggest that AD risk factors are not contributing solely to altered microglial responses in the AD brain but may also contribute to dysregulated peripheral immune function in AD. Notably, BIN1 and ABCA1 DARs appear to be influenced by APOE genotype in AD, suggesting convergence of epigenetic changes to AD risk factors and APOE genotype. We suggest future studies on the peripheral function of AD risk genes highlighted by this resource, including BIN1, CLU, INPP5D and ABCA1, particularly the impact of APOE genotype on their respective functions.

These data also uncover APOE genotype-dependent epigenetic and transcriptomic changes to clonally expanded, CD8 memory T cells in AD. Again, expression of BIN1 was influenced by APOE genotype and associated with clonally expanded CD8 T cells in AD. These results dovetail with our recent findings that clonally expanded CD8 T cells patrol the CSF of AD patients⁹. Whether the concordant epigenetic and transcriptomic changes to peripheral CD8 memory T cells we observe in these subjects might influence AD pathophysiology (e.g. by modulating microglia) warrants further scientific investigation.

Since epigenetic changes do not involve alterations in the DNA sequence, it is enticing to speculate what may be generating chromatin accessibility alterations in AD. The fact that we observe numerous pro-inflammatory cytokine genes involved in the immune defense suggests that immune challenges promote epigenetic alterations in AD subjects. Thus, the AD field would greatly benefit from a longitudinal examination of peripheral immunity that tracks subjects prior to the onset of clinical dementia. Since APOE ε4 is enriched in the AD population, our results suggest that previous studies of peripheral immunity that did not stratify by APOE genotype, or did not include APOE as a covariate, were influenced by this confounding genetic factor. Moreover, the fact that we identify many epigenetic changes associated with APOE ε4/ε4 carriers may partially reflect why these individuals are more susceptible to side effects from conventional anti-Aβ therapies^53,54. Finally, we provide our findings as an explorable online resource to aid therapeutic target identification to restrict peripheral inflammation in AD.

Limitations of the study

This study is comprised entirely of human data and our claims are based primarily on underlying assumptions of bioinformatic algorithms. As such, there are limitations regarding functional or mechanistic evidence regarding the impact of epigenetic changes on immune function. We also regard this study as an initial understanding of epigenetics in the AD peripheral immune system. While our dataset is rich for a modern multi-omic study, some analyses were underpowered. Specifically, we report raw, uncorrected p-values for DARs and DEGs by pseudobulk methods. For this reason, we present the overlap of genes significant by single cell and pseudobulk measures to try to reduce false positive results. As the accessibility of multi-omic studies increases in the future, we suggest further interrogation of epigenetics of the peripheral immune system in AD to replicate our findings using larger sample sizes.

STAR Methods

Lead Contact

Further information and requests for resources should be directed to and will be fulfilled by the lead contact, David Gate (dgate@northwestern.edu).

Materials Availability

No new unique reagents were generated for this study.

Data and Code Availability

• Single-cell ATAC-seq RNA-seq data have been deposited at GEO and are publicly available as of the date of publication. Raw RNA+TCR .fastq files and gene expression matrices can be downloaded from GSE226602. Raw ATAC .fastq files and peak matrices can be downloaded from GSE226267.

• All original code has been deposited at Zenodo and is publicly available as of the date of publication. DOIs are listed in the key resources table.

• Any additional information required to reanalyze the data reported in this work paper is available from the Lead Contact upon request.

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
rat anti-CD3	Abcam	Cat #ab11089
rabbit anti-CD8	Cell Signaling	Cat #85336
mouse anti-CXCR3	Abcam	Cat #ab64714
rabbit anti-Aβ	Cell Signaling	Cat #42284
goat anti-Ibal	Abcam	Cat #ab5076
CD8α-Pacific blue	BioLegend	Cat #344718
CD3-BV650	BD Biosciences	Cat #563916
CD45RA-PE	BioLegend	Cat #304108
CCR7-488	BioLegend	Cat #353206
CD27-PE-Cy7	BioLegend	Cat #302838
CXCR3-647	BioLegend	Cat #353712
Fixable Live/dead dye	Thermo	Cat #L34992
Bacterial and virus strains
Biological samples
Adult PBMCs	Stanford University Alzheimer’s Disease Research Center (ADRC)	n/a
Table 1 Demographic information	This paper	n/a
Chemicals, peptides, and recombinant proteins
Critical commercial assays
10x Genomics Chromium Next GEM Single Cell 5’ v2 with immune profiling kit	10x Genomics	User Guide CG000331, Rev D
10x Genomics Chromium Next GEM Single Cell ATAC v1.1 kit	10x Genomics	User Guide CG000209, Rev F
10x Genomics Nuclei Isolation for Single Cell ATAC Sequencing	10x Genomics	Demonstrated protocol CG000169, Rev D
Deposited data
Raw and processed single cell RNA+TCR sequencing data	This study	GEO: GSE226602
Raw and processed single cell ATAC sequencing data	This study	GEO: GSE226267
Experimental models: Cell lines
Experimental models: Organisms/strains
Oligonucleotides
Recombinant DNA
Software and algorithms
Code for analysis	This study	https://github.com/gatelabnw/ad_apoe_pub and DOI: 10.5281/zenodo.10472192
Cellranger v6.0.0	10x Genomics	https://support.10xgenomics.com/single-cell-gene-expression/software/pipelines/latest/what-is-cell-ranger
SoupX v1.6.2	Young and Behjati24	https://github.com/constantAmateur/SoupX
DoubletFinder v2.0.3	McGinnis et al.59	https://github.com/chris-mcginnis-ucsf/DoubletFinder
Cell Ranger ATAC v2.0.0	10x Genomics	https://support.10xgenomics.com/single-cell-atac/software/pipelines/latest/what-is-cell-ranger-atac
ArchR v1.0.2	Granja et al.22	https://github.com/GreenleafLab/ArchR
ArchR2Signac v1.0.2	Shi et al.25	https://github.com/swaruplabUCI/ArchRtoSignac
Seurat v4.1.0	Hao et al.23	https://satijalab.org/seurat/
edgeR from Delegate v1.0.0	Christoph Hafemeister, Developmental Cancer Genomics group at St. Anna Children’s Cancer Research Institute (CCRI)	https://github.com/cancerbits/DElegate
DESeq2 from Delegate v1.0.0	Christoph Hafemeister, Developmental Cancer Genomics group at St. Anna Children’s Cancer Research Institute (CCRI)	https://github.com/cancerbits/DElegate
Signac v1.8.0	Stuart et al.26	https://github.com/stuart-lab/signac
maxATAC v1.0.0	Cazares et al.33	https://github.com/MiraldiLab/maxATAC
Cicero v1.3.8	Pliner et al.35	https://github.com/cole-trapnell-lab/cicero-release
Other
ShinyCell app for interactive data analysis	This paper	https://gatelabnu.shinyapps.io/ad_apoe_rna/

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Humans

For scATACseq and scRNAseq experiments, PBMC samples were acquired through the Stanford University ADRC. Collection of PBMCs was approved by the Institutional Review Board of Stanford University and written consent was obtained from all subjects. PBMCs were isolated by standard Percoll centrifugation then cryopreserved in freezing medium containing 90% fetal bovine serum and 10% dimethyl sulfoxide. If a sample was not >80% viable post-thaw, replacement samples of similar demographics were chosen to the best of our ability to minimize group-related batch effects. Live PBMCs were sorted using a Sytox blue live/dead dye on a Sony SH800 cell sorter. Age, sex and race demographics for all study subjects are presented in Table S1. Study subjects were categorized as HC or AD based on Clinical Dementia Rating (CDR) ratings and standardized neuropsychological assessments to determine cognitive and diagnostic status, including procedures of the National Alzheimer’s Coordinating Center. Participant eligibility included normal or corrected to-normal vision/hearing, native English speaking, no history of neurologic or psychiatric disease, CDR assessment, and performance within the normal range on a standardized neuropsychological test battery^55,56. All healthy control participants had CDR global scores of zero and were deemed cognitively unimpaired during a clinical consensus meeting consisting of neurologists, neuropsychologists, and research coordinators. All AD participants had a CDR score greater than zero. For single cell transcriptomics assays, a total of 56 unique samples were used, including 26 HC and 30 AD samples. Sample size estimation was based on our prior gene expression studies^9,57. Samples were selected to produce a balance of APOE genotypes for each diagnostic group. After potential experimental samples were identified, a mix of ~10 samples were chosen for each experimental batch with the intention of balancing diagnosis, sex, and APOE genotype composition.

Methods details

Protein biomarker measurements

Plasma and CSF biomarker data were generated by the Stanford University ADRC and were analyzed as previously described⁵⁸. Protein levels were measured on a modified version of the Lumipulse G CSF p-Tau181 assay for CSF (Cat. # 231654, Fujirebio Diagnostics, US, Malvern, PA) using the LUMIPULSE G1200 instrument as previously described. The Lumipulse G plasma p-tau181 assay antibody combination is based on the INNOTEST assay targeting tau epitopes proximal to Thr181, including antibody AT270 for capture, with HT7 and BT2 used for detection. Samples were thawed on wet ice, centrifuged for 5 min at 4°C at 500 × g before being loaded on the fully automated LUMIPULSE G1200 instrument. To minimize potential for non-specific binding, plasma samples were mixed with a heterophilic blocking reagent (200 μg/ml, Scantibodies Inc., Santee, CA) prior to measurement. Individual-level variability was assessed on 6 independent plasma aliquots using a different lot of reagents one year later and showed high test-retest reliability (Pearson’s r = 0.98). 100% of plasma samples from the current study fell within the quantifiable range (range 0.46–11.47 pg/ml). Samples were tested by experimenters blind to diagnostic information or APOE genotype. Following quality control measures, profiles for 50 unique samples were kept.

Droplet-based scRNA+TCRseq

The 10x Genomics Chromium Next GEM Single Cell 5’ v2 with immune profiling kit was used for scRNA+TCRseq of PBMC samples. Libraries were prepared according to 10x Genomics protocols. Libraries were sequenced by Novogene on an Illumina Novaseq 6000 instrument. Bases were called using the Illumina RTA3 method. RNA reads were aligned to the hg38 genome build and gene expression matrices were generated using Cell Ranger 6.0.0 software. TCR reads were also aligned to the hg38 genome build and clonotype/contig matrices were generated using Cell Ranger.

scRNA+TCRseq quality control

Empty droplets were removed via Cell Ranger 6.0.0 using the EmptyDrops method per 10x Genomics’ protocol. Gene expression matrices were corrected for background contamination using R package SoupX 1.6.2. Known monocyte/dendritic markers (S100A8 and S100A9) were used to estimate the contamination fraction of each sample. Counts were adjusted using the SoupX subtraction method using the calculated contamination fraction on a per sample basis. Doublets were removed using R package DoubletFinder 2.0.3⁵⁹ using an approximate doublet formation rate of 5.4% which is consistent with the expected multiplet rate according to 10x Genomics Single Cell 5’ v2 kit protocol. Any cells with fewer than 200 mapped features were eliminated, as well as any features present in fewer than three cells. Any cells with at least 10% mitochondrial reads were also eliminated. TCR clonotypes and contigs were also filtered for empty droplets using Cell Ranger 6.0.0. Only TCR sequences associated to cells annotated with a respective T cell identity by RNAseq were retained.

Droplet-based scATACseq

The 10x Genomics Single Cell ATAC v1.1 kit was used for scATACseq of PBMC samples. Nuclei were isolated and libraries were prepared according to 10x Genomics protocols. Libraries were sequenced by Novogene on an Illumina Novaseq 6000 instrument. Bases were called using the Illumina RTA3 method. ATAC fragments were aligned to the hg38 genome build using Cell Ranger ATAC 2.0.0 software.

scATACseq quality control

Empty droplets were removed via Cell Ranger ATAC 2.0.0 using the EmptyDrops method per 10x Genomics’ protocol. Using ArchR 1.0.2, fragments were filtered to retain those between 10 and 2000 base pairs and with TSS enrichment > 10. Cells were filtered for those with between 1e3 and 1e5 fragments. After quality control, 50 unique sample profiles were retained. Constrained integration was used to map each cell in the scATACseq dataset to the closest cell in the scRNAseq dataset. The ArchR project was then converted to Signac using the ArchR2Signac package. Due to the matrix size limitation in R, the project was first split into CD4⁺ T Cells (which comprised ~45% of all cells) and all remaining cells. Seurat objects were then generated separately for these two projects.

scRNAseq cell type annotations

Corrected and filtered gene expression matrices were SCTransformed with Seurat 4.1.0²³ on a per sample basis and then integrated through harmonizing ‘anchors’ as recommended for cell type identification in Seurat documentation. Number of reads, number of features, and percent of mitochondrial reads were regressed out in the data scaling step of SCTransform, and the top 1000 most variable features were used. Principal component analysis was then run on the integrated assay. The first fifteen principal components were then used to generate a shared nearest neighbor graph which was then clustered under the Louvain algorithm with a resolution of 0.3. Uniform manifold approximation and projection was then performed using the first 15 PCs and 30 nearest neighbors. The object was then mapped to a PBMC CITE-seq reference dataset²³ using the Seurat FindTransferAnchors() and MapQuery() functions in the SCT assay.

Differential expression

The Seurat function FindMarkers was used to identify DEGs in a cell type-specific manner across diagnosis and APOE genotypes. MAST was chosen to test significance as it employs a hurdle model specifically tailored to bimodal expression distributions often observed in scRNAseq. Only genes expressed in at least 10% of cells were tested. Sex was included as a latent variable in all comparisons and APOE genotype was additionally included as a latent variable in AD versus HC comparisons. P-values were adjusted for multiple comparisons using the Benjamini-Hochberg procedure. Genes with an adjusted p-value less than 0.05 and average log-fold change magnitude greater than 0.125 were considered significantly differentially expressed. To mitigate the effect of pseudoreplication bias, we intersected these significant DEGs with results from the edgeR pseudobulk method. We implemented the DElegate 1.0.0 package, which aggregates counts for each sample and cell type combination, to test with edgeR (edgeR::glmQLFit). Genes with p-values less than 0.05 and average log-fold change magnitude greater than 0.125 were considered significantly differentially expressed. The intersection of results from these two methods were used for all downstream analyses. Age and ethnicity were included as covariates in all analyses.

Differential accessibility

The Seurat function FindMarkers was used to identify DARs in a cell type-specific manner across diagnosis and APOE genotypes. Logistic regression was chosen to test significance as recommended by the Signac documentation. Regions with an adjusted p-value less than 0.05 and average log-fold change magnitude greater than 0.125 were considered significantly differentially accessible. To mitigate the effect of pseudoreplication bias, we intersected these significant DARs with results from a pseudobulk method, DESeq2. We implemented the DElegate 1.0.0 package which aggregates counts for each sample and cell type combination to test with DESeq2 (DESeq2::DESeq(test = ‘Wald’)). Regions with p-values less than 0.05 and average log-fold change magnitude greater than 0.125 were considered significantly differentially expressed. The intersection of results from these two methods were used for all downstream analyses.

Immunohistochemistry and confocal imaging

We stained 5μm paraffin embedded brain tissue sections using antibodies rat anti-CD3 (Abcam ab11089), rabbit anti-CD8 (Cell Signaling 85336), mouse anti-CXCR3 (Abcam ab64714), rabbit anti-Aβ (Cell Signaling 42284), and goat anti-Iba1 (Abcam ab5076). Sections were deparaffinized, then antigen retrieval was performed using citrate buffer pH 6.0 for 30 min at 95°C. Sections were blocked in phosphate buffered saline containing 10% normal donkey serum and 0.1% triton-x. Sections were stained overnight in primary antibodies. The following morning, sections were incubated with highly cross-absorbed, species-appropriate secondary antibodies. Sections were imaged on a Nikon AXR confocal microscope with a 60x objective.

Flow cytometry

PBMCs from 12 patients clinically diagnosed with mild cognitive impairment or AD were processed for flow cytometry as previously described⁹. Flow cytometry was conducted using an LSRFortessa (BD Biosciences). A panel consisting of antibodies conjugated to six different fluorophores and Fc block was used to classify subsets of memory T cells and their expression of CXCR3. Antibodies used were: CD8α-Pacific blue (BioLegend), CD3-BV650 (BD Biosciences), CD45RA-PE (BioLegend), CCR7–488 (BioLegend), CD27-PE-Cy7 (BioLegend) and CXCR3–647 (BioLegend). A fixable live-dead dye fluorescing APC-Cy7 was added to each sample. Cells were fixed for 12 minutes in 4% paraformaldehyde. A compensation matrix was developed using singly stained and unstained controls. All analysis was conducted in Cytobank.

Transcription factor binding

We first investigated transcription factor binding using motif scanning. A motif matrix was first added to the Seurat object using the JASPAR 2020 motif database. The Seurat object was then filtered for the cell type of interest. Up and down accessible regions were identified using thresholds specified above. Enriched transcription factor motifs were identified within these up and down accessible regions separately using the Signac FindMotifs() function. The hypergeometric test was used to determine significance and the Benjamini-Hochberg procedure was applied for multiple comparison correction. Rank was calculated based on the product of log-transformed adjusted p-values and log-fold change, with equal rank given to transcription factors with equal values. We next implemented maxATAC to investigate transcription factor binding without the constraints of predetermined DARs. Fragments from each sample and cell type combination were converted to Tn5 cut sites using the prepare function, then averaged across diagnosis group. Transcription factor binding was then predicted using the 127 available transcription factor models. Peaks were then called on these prediction tracks using the default calculated prediction value cutoff, which were subsequently divided into 32 base pair regions to aid in identifying unique and shared sites.

Transcription factor footprinting

Transcription factor footprints were calculated using Signac’s Footprint() function, specifying to only examine motifs which fell in peaks. The PlotFootprint() function was used to visualize these footprints with the subtract normalization method used to account for Tn5 DNA sequence bias. Motifs were chosen for footprinting based on their enrichment by hypergeometric test in diagnosis or APOE genotype-specific DARs.

DAR and DEG correlation

To identify genes and regions with correlated expression and accessibility, respectively, we first implemented Cicero on each cell type and diagnosis combination separately. We filtered the resulting links for a co-accessibility score of 0.01 and links containing at least one promoter region, as previously described³⁶. For each promoter link, we ran a Pearson correlation between the average expression of the gene and the average accessibility of the linked region across shared samples between RNA and ATAC modalities. We then applied the Benjamini-Hochberg procedure on the resulting p-values. Any link with a correlation coefficient above the 95^th percentile of all correlations for the given cell type and diagnosis combination and adjusted p-value under 0.05 was deemed significant.

Quantification and statistical analysis

R 4.1.1 were used for all statistical analyses. Statistical methods are described in the figure legends, methods, or main text as appropriate.

Additional resources

ShinyCell is an R package developed to quickly generate interactive Shiny-based web applications to visualize the core analysis of scRNAseq data⁶⁰. We have generated a modified ShinyCell app allowing users to view metadata and gene expression on a uniform manifold approximation and projection, compare gene expression between groups via violin/box plots, and other built-in analyses. This site can be found at: https://gatelabnu.shinyapps.io/ad_apoe_rna.

Supplementary Material

1

Table S2. DAR gene lists between AD and HC subjects, related to Figure 2.

2

Table S3. DEG gene lists between AD and HC subjects, related to Figure 2.

3

Table S4. AD vs. HC transcription factor enrichment analysis by motif scanning and maxATAC in monocytes, related to Figure 3.

4

Table S5. Cis-regulatory sequences and DEG significant correlations, related to Figure 4.

5

Table S6. AD vs. HC DARs by APOE genotype comparison, related to Figure 5.

6

Table S7. AD vs. HC DEGs by APOE genotype comparison, related to Figure 5.

7

Table S8. DEG gene lists for clonal T cells between AD and HC subjects by APOE genotype, related to Figure 6.

8

Table S9. DEG gene lists between AD and HC subjects by APOE genotype, related to Figure 6.

Highlights.

• AD peripheral immune cells have more open chromatin

• AD CD8 T cells have a chromatin modification associated with expression of CXCR3

• AD monocytes have APOE genotype-specific chromatin modifications

• Genes associated with sporadic AD are altered at the chromatin level