Ancestry-related differences in chromatin accessibility and gene expression of APOE4 are associated with Alzheimer disease risk

Abstract

INTRODUCTION:

European local ancestry (ELA) surrounding APOE4 confers higher risk for Alzheimer Disease (AD) compared to African local ancestry (ALA). We demonstrated significantly higher APOE4 expression in ELA vs ALA in AD brains from APOE4/4 carriers. Chromatin accessibility differences could contribute to these expression changes.

METHODS:

We performed single nuclei Assays for Transposase Accessible Chromatin sequencing from frontal cortex of six ALA and six ELA AD brains, homozygous for local ancestry and APOE4.

RESULTS:

Our results showed an increased chromatin accessibility at the APOE4 promoter area in ELA vs ALA astrocytes. This increased accessibility in ELA astrocytes extended genome wide. Genes with increased accessibility in ELA in astrocytes were enriched for synapsis, cholesterol processing and astrocyte reactivity.

DISCUSSION:

Our results suggest that increased chromatin accessibility of APOE4 in ELA astrocyte contributes to the observed elevated APOE4 expression, corresponding to the increased AD risk in ELA vs ALA APOE4/4 carriers.

Graphical Abstract

BACKGROUND:

Alzheimer disease (AD) is the most common neurodegenerative disorder and one of the leading causes of disability worldwide for individuals aged 75 and older¹. The strongest genetic risk factor for late-onset AD is the apolipoprotein E4 (APOE4) allele located on chromosome 19^2,3. However, the risk associated with APOE4 differs dramatically between individuals of European vs African ancestries^4–7 with APOE4 carriers of European ancestry having substantially increased risk for AD compared to individuals with African ancestry.

This observed difference in risk has recently been shown in three independent studies to be due to differences in the genetic local ancestry (LA) region surrounding APOE rather than overall global European or African ancestry^8–10. Using single nucleus RNA sequencing (snRNA-seq) of frontal cortex we have previously demonstrated that APOE4/4 homozygous carriers with European local ancestry (ELA) expressed significantly higher APOE compared to those with African local ancestry (ALA), especially in astrocytes¹¹, suggesting a potential mechanism underlying the differential AD risk seen between ancestries.

Control of gene expression is orchestrated by the integration of cis-regulatory modules, such as enhancer and promoter elements, along with transcription factors^12–14. The cis-regulatory modules of actively transcribed genes are generally in ‘accessible’ euchromatin with low nucleosome occupancy and few high-order structures^15–17, while ‘closed’ chromatin (heterochromatin) generally associates with transcriptional silencing^16,18,19. Accordingly, ancestry-specific changes in accessibility of cis-regulatory elements that modulate the binding of transcription factors is a potential mechanism responsible for the difference in APOE4 expression seen between ancestries.

The single nuclei Assay for Transposase Accessible Chromatin followed by sequencing (snATAC-seq), facilitates the creation of chromatin accessibility maps with single cellular resolution. For example, Corces et al²⁰ demonstrated that open chromatin regions in promoter or enhancer areas are associated with increased gene expression through snATAC-seq. Investigation of chromatin accessibility profiles specifically in AD have mostly focused on pathological hallmarks²¹. More recently, Morabito et al²² conducted an integrated analysis using snATAC-seq and snRNA-seq from brains of AD and control individuals and reported that cell-type-specific chromatin accessibility changes in regulatory elements may be key to gene expression changes in AD. However, to date, the role of chromatin accessibility in AD between different ancestries, specifically at the APOE4 locus, has not been investigated.

Therefore, we performed snATAC-seq combined with snRNA-seq to investigate the cell specific patterns of chromatin accessibility in ELA and ALA APOE4/4 brains. We observed increased accessibility in APOE ELA relative to ALA in astrocytes. The differentially accessible peaks at the APOE promoter are predicted to bind a subset of transcription factors that exhibit differential gene expression in ELA astrocyte samples. These data support the hypothesis that European ancestry at the APOE promoter and in the area around APOE have a more open chromatin conformation than African ancestry. We speculate that this difference is permissive for transcription factor binding and transcriptional activation of APOE. Increased chromatin accessibility in ELA samples was also observed genome-wide in this astrocytic cell type. Thus, these findings provide novel insights into the molecular mechanisms of ancestry specific differences in AD risk.

METHODS:

Brain Samples

Brain autopsy samples were obtained as part of a multi-center collaboration from four Alzheimer Disease Research Centers (ADRCs; Emory University, Northwestern University, Rush University Medical Center, and the University of Pennsylvania) and from the John P. Hussman Institute for Human Genomics (HIHG) at the University of Miami. All ADRC samples were initially selected from the National Alzheimer Coordinating Center (NACC) and compliant with site-specific approved institutional review board protocols as previously described¹¹.

Local ancestry determination

Genotyping arrays were processed to assess both global and local genetic ancestry (LA) in the APOE region. To determine the LA, we phased independently our datasets with the SHAPEIT tool (version 2) using 1000 Genomes Phase 3 as the reference panel⁸. The APOE region was defined within 1Mb on either side of APOE, using chr19:44–46Mb as the LA boundaries for ancestry assignments, broad enough to include potential enhancers and topological associated domains. Individuals homozygous for the ELA or ALA haplotype in this region were Whole Genome Sequenced (WGS) at either the Center for Genome Technology (CGT) at the HIHG or The American Genome Center at Uniformed Services University of the Health Sciences (USUHS).

snATAC-seq and snRNA-seq data generation and analysis

Isolation of Nuclei:

Nuclei were isolated from frozen frontal cortex brain tissue (Brodmann area 9) as described previously^23,24. A suspension of 65,000 nuclei was prepped according to the 10X Genomics Chromium Next GEM Single Cell ATAC 1.1 protocol. Libraries were sequenced on an Illumina NovaSeq 6000 targeting 75,000 reads per nuclei in paired end 50bp sequencing reactions. FASTQ files were processed using the CellRanger ATAC software package v1.2 (10X Genomics). snRNA-seq was performed as previously described¹¹.

Integrated analysis of the snATAC-seq and snRNA-seq data:

This was performed using the ArchR software package v1.0.1²⁵. Clustering of cells was performed with five iterations of Iterative Latent Semantic Indexing (LSI)^26,27 followed by batch correction with Harmony²⁸. ATAC clusters were identified using FindClusters in Seurat v4.0 software package²⁹ and refined the cell type definitions by integration with snRNA-seq from the same samples using Seurat’s FindTransferAnchors.

SnATAC-seq peak calling and differential accessibility

We created pseudo-bulk replicates combining all ELA and all ALA samples separately and called peaks for each ancestry within each cluster using the addReproduciblePeakSet function in ArchR utilizing the default parameters of the MACS2 callpeak command v2.2.7.1³⁰. Differential accessibility for each peak within each cluster was calculated using ArchR getMarkerFeatures.

ATAC peak annotations

HOMER³¹ (version 4.11³²), was used to annotate peaks closest to transcription start sites (TSS) and to indicate exon, intron, intergenic, or promoter-TSS regions. A peak was assigned to a gene promoter-TSS when the peak location was ±2kb from the TSS. GREAT³³ (version 4.0.4, http://great.stanford.edu/public/html/) was used to refine annotations where the peak was near multiple genes or intergenic. Distal enhancers were identified with the ELITE GeneHancer database from UCSC³⁴ of enhancers and promoters. We considered as distal enhancer peaks those annotated by GREAT or HOMER farther than ±2kb from a TSS and with coordinates overlapping one of these ELITE enhancers.

Transcription Factor Analysis

Bedtools getfasta³⁵ was used to extract the sequence of each peak (refdata-cellranger-atac-GRCh38–1.2.0 version). The algorithm FIMO³⁶ within MEME Suite version 5.4.1³⁷ was used to identify known transcription factor (TF) binding motifs in the two APOE promoter differentially accessible peaks (DAP). We surveyed defined transcription factor binding site databases JASPAR CORE 2022 (vertebrates non redundant)³⁸, Jolma 2013 (human and mouse)³⁹, and Swiss regulon (human and mouse)⁴⁰, and selected those with false discovery rate (FDR) adjusted p-value <0.05. In addition, we used all the human specific JASPAR 2022 defined TFs of the UCSC HG38 table browser⁴¹ filtering by a JASPAR score ≥ 300.

Functional Enrichment Analysis

Functional analysis was performed using enrichR⁴². using as reference the human specific gene-set libraries KEGG 2021⁴³, the Gene Ontology Biological Process 2021⁴⁴, Reactome 2021⁴⁵ and ENCODE Histone post-translational modifications⁴⁶. Enrichment for cell type specific markers was performed using the cell type gene-set atlas lists from Azimuth 2021⁴⁷ and PanglaoDB 2021⁴⁸. Disease gene enrichment analysis used GeneSigDb⁴⁹, HDSigDb⁵⁰ and MsigDB⁵¹. Enrichment in chromosomal location was performed using a Fischer exact test.

RESULTS:

Brain Sample Characteristics

Twelve brain samples were used for the study (Table 1). Six ELA samples (European global ancestry >96%) were obtained from the HIHG Brain Bank (3), Rush University Medical Center (2) and Emory University (1). Six ALA samples (African global ancestry 68% to 88%) were obtained from Emory University (2), Northwestern University (1) and Rush University Medical Center (3). Samples included seven females (four ALA and three ELA) and five males (two ALA and three ELA). All samples had BRAAK staging scores ranging from IV to VI and had a mean age-of-death of 79 years. Whole genome sequencing confirmed the homozygous APOE4/4 genotype and absence of mutations in known Mendelian genes for AD (PSEN1, PSEN2, APP, MAPT) as well as in known rare AD risk variants in ABCA7, TREM2 and SORL1.

Table 1.

Demographic characteristics of the samples

Sample	Center	Sex	AOD	APOE genotype	Local Ancestry (Chr19: 44–46Mb)	% Of Global Ancestry	BRAAK staging score
1	Emory	Female	82	4,4	AF/AF	87% AF	V
2	NW	Female	85	4,4	AF/AF	86% AF	V
3	Emory	Female	80	4,4	AF/AF	88% AF	VI
4	Rush	Male	93	4,4	AF/AF	68% AF	V
5	Rush	Male	83	4,4	AF/AF	79% AF	VI
6	Rush	Female	77	4,4	AF/AF	85% AF	VI
7	UM/Duke	Male	75	4,4	EU/EU	95% EU	IV
8	Rush	Male	71	4,4	EU/EU	98% EU	VI
9	UM/Duke	Female	70	4,4	EU/EU	95% EU	VI
10	Rush	Male	76	4,4	EU/EU	97% EU	IV
11	UM/Duke	Female	76	4,4	EU/EU	96% EU	V
12	Rush	Female	86	4,4	EU/EU	98% EU	VI

AOD: Age of Death; UM: University of Miami; NW: Northwestern University; RUSH: RUSH University Medical Center; AF: African; EU: European

Cell type identification

We obtained data from a total of 60,306 nuclei from snATAC-seq and 94,411 nuclei from snRNA-seq. The snATAC-seq libraries had an average of ~13,000 fragments per nucleus and the snRNA-seq libraries had a median depth of ~116,000 reads per nucleus with on average ~1,900 genes/nucleus (Supplementary Table 1). The clusters with more than 500 nuclei in ELA and ALA showed no statistical differences in the number of nuclei per cluster between the ancestry groups (p-value= 0.497). Integration of the snATAC-seq and snRNA-seq resulted in the identification of 11 cell type clusters (Figure 1A) with no sample specific bias (Supplementary Figure 1A and 1B, Supplementary Table 2). The cell type identity of each cluster was determined using top 50 snATAC-seq predicted gene scores for marker genes of known cell types in the frontal cortex (Figure 1B). We used SLC17A7 for excitatory neurons; SLC32A1 for inhibitory neurons; MAG for oligodendrocytes; AIF1 for microglia; GFAP for astrocytes, CSPG4 for oligodendrocyte precursor cell (OPCs) and COL1A2 for vascular leptomeningeal cell (VLMC) (Supplementary Figure 2). We identified one oligodendrocyte cluster (55.6% of total cells), four excitatory neuron clusters (20.8% of the total cells), one inhibitory neuron cluster (8.3% of total cells), one microglia cluster (7.4% of total cells), one OPC cluster (3.9% of total cells), two astrocyte clusters (2.5% of total cells), and one vascular leptomeningeal cell (VLMC) cluster (1.5% of total cells) (Supplementary Table 2).

Figure 1. Visualization of the integrated single nuclei ATAC and RNA sequencing clusters and cell identification across the samples.

A) UMAP reduction plot using resolution of 0.4 resulting in 11 cell type clusters from the integration of snATAC-seq and snRNA-seq data. B) Heatmap showing cell cluster identification by chromatin accessibility patterns using the top 50 snATAC-Seq predicted gene scores for markers genes of the known cortex cell types.

To further characterize the two astrocyte cell clusters, we determined the genes that identified astrocyte cluster 1 and astrocyte cluster 2 and investigated their expression in the single cell atlas of the Entorhinal Cortex in Human Alzheimer’s Disease (ECHAD)⁵² and astrocyte transcriptomic data from frontal cortex⁵³. We observed that the genes characterizing astrocyte cluster 1 were primarily expressed in astrocyte subclusters a4 to a8, while the genes from our astrocyte cluster 2 were primarily expressed in astrocyte subclusters a2 and a3 from ECHAD, which correspond to reactive astrocytes subpopulations (Supplementary Figure 3)⁵³.

Accessibility analysis at the APOE locus and LA

We first investigated whether the ancestry-related differential expression of APOE4 previously described¹¹ was recapitulated when additional samples from Rush University Medical Center were included in our analyses. Indeed, APOE4 was differentially expressed when considering expression over all cell types between ancestries with greater expression in ELA (FC= 1.31; p-value=9.66E⁻²¹⁹), except one excitatory neurons cluster with higher expression in ALA (FC= 1.28; p-value=5.87E⁶). This pattern of increased expression in ELA samples was true for four cell types (excitatory and inhibitory neurons, astrocytes and microglia) (Supplementary Table 3). Astrocytes had the highest fold change difference in ELA (FC = 1.56) and most significant p-value (p-value=1.24E⁻¹²⁹).

We next determined whether the APOE expression difference in astrocytes is associated with differences in chromatin accessibility. The snATAC-seq and snRNA-seq integrated UMAP showed that APOE expression as well as accessibility was highest in the astrocyte cluster 1 (Figure 2A). Comparison of accessibility between ancestries revealed two peaks with significantly increased accessibility in ELA at the APOE promoter in astrocyte cluster 1 (Figure 2B). These peaks were located at −19bp and −1990bp upstream of the APOE transcription start site (TSS) (FC: 2.57 and 4.52; FDR: 0.001 and 0.02, respectively). Notably, although APOE showed significantly increased expression in several cell types, significant differences in accessibility were only detected in astrocyte cluster 1 (Supplementary Figure 4).

Figure 2. Differential chromatin accessibility and expression of APOE in Astrocyte cluster 1.

A) APOE expression represented in clusters generated by the integrated snATAC-seq and snRNA-seq data; B) Visualization of chromatin differential accessible peaks in the APOE locus between ancestries from Astrocyte cluster 1. *Represents significantly differentially accessible peaks between ancestries (*FDR=0.02; ***FDR=0.001). C) Venn diagram showing transcription factors biding to APOE that are differentially expressed (DEG) and differentially accessible (DAG).

To gain insight into the molecular mechanisms involved in the increased accessibility and increased expression of APOE, we determined which known transcription factor binding sites were present in the differentially accessible peaks and analyzed the accessibility and gene expression of those transcription factors. The two differentially accessible peaks at the APOE promoter overlap predicted binding sites for a set of 15 transcription factors (GLI2, NPAS2, KLF15, HIF1A, LHX2, RXRA, MXI1, FOS, JUNB, KLF2, PURA, SREBF1, TEAD1, KLF6, ZBTB7C) that are also differentially expressed and have differentially accessible genomic region between ancestries (Supplemental Table 4). From these 15 putative APOE binding transcription factors, PURA and SREBF1 have increased accessibility and expression in ELA samples, KLF6 has increased accessibility and expression in ALA samples and the other 12 transcription factors have increased accessibility in ELA and increased expression in ALA.

We identified 32 additional differentially accessible peaks with greater accessibility in ELA than ALA astrocytes in gene promoters (within 2kb from the TSS) in the LA region surrounding APOE (chr19:44–46MB), which corresponds to 19 additional genes, including the APOE proximal genes TOMM40 and APOC1 (Table 2). No genes in the defined LA region were more accessible in the ALA. In addition, by overlapping the differentially accessible peaks in the APOE LA region with previously classified enhancers (ELITE GeneHancer from USCS) we identified 32 additional peaks among 23 LA genes, all with increased accessibility in ELA brains (Supplementary Table 5).

Table 2.

Differentially accessible peaks in promoter regions of Local Ancestry genes in Astrocyte cluster 1.

Gene	Chromosome	Peak location	Distance to TSS	FC	FDR	Region
CTB-171A8.1	chr19	44718456	−21	3.66	0.003	promoter-TSS
BCL3	chr19	44748309	12	2.77	0.007	promoter-TSS
BCAM	chr19	44808802	−7	5.11	0.043	promoter-TSS
BCAM	chr19	44808802	−51	5.11	0.043	promoter-TSS
CTB-129P6.4	chr19	44890205	421	2.70	0.002	promoter-TSS
TOMM40	chr19	44890205	−781	2.70	0.002	promoter-TSS
APOE*	chr19	44903514	−1990	4.52	0.022	promoter-TSS
APOE*	chr19	44905485	−19	2.57	0.001	promoter-TSS
APOC1	chr19	44914897	278	2.05	0.000	promoter-TSS
APOC1	chr19	44914897	900	2.05	0.000	promoter-TSS
APOC1P1	chr19	44926992	266	2.23	0.025	promoter-TSS
CLPTM1	chr19	44955080	84	3.61	0.007	promoter-TSS
CLPTM1	chr19	44955080	84	3.61	0.007	promoter-TSS
CTB-179K24.3	chr19	45091377	−1236	2.19	0.027	promoter-TSS
PPP1R37	chr19	45091377	−1335	2.19	0.027	promoter-TSS
MARK4*	chr19	45294609	504	3.65	0.011	promoter-TSS
ERCC2	chr19	45370425	−88	2.40	0.002	promoter-TSS
ERCC2	chr19	45370425	−57	2.40	0.002	promoter-TSS
PPP1R13L	chr19	45405279	−491	3.68	0.004	promoter-TSS
CD3EAP	chr19	45405279	−1162	3.68	0.004	promoter-TSS
PPP1R13L	chr19	45405279	820	3.68	0.004	promoter-TSS
FOSB	chr19	45467194	−551	2.61	0.024	promoter-TSS
VASP	chr19	45506348	−818	3.58	0.020	promoter-TSS
OPA3	chr19	45584758	−166	2.16	0.000	promoter-TSS
EML2*	chr19	45644874	505	7.63	0.032	promoter-TSS
EML2*	chr19	45645399	−20	3.28	0.002	promoter-TSS
FBXO46	chr19	45730087	567	4.28	0.025	promoter-TSS
FBXO46	chr19	45730692	−38	2.01	0.028	promoter-TSS
FBXO46	chr19	45730692	−38	2.01	0.028	promoter-TSS
RSPH6A	chr19	45815742	−673	3.10	0.033	promoter-TSS
IRF2BP1	chr19	45885449	471	4.72	0.025	promoter-TSS
NOVA2	chr19	45973825	−529	5.11	0.004	promoter-TSS
CCDC61	chr19	45995053	−158	2.38	0.000	promoter-TSS
CCDC61	chr19	45995053	−164	2.38	0.000	promoter-TSS

FC: Fold change; TSS: Transcription Starting Site; Negative value in Distance to TSS reflects peaks falling upstream of TSS

Genome-wide differential accessibility

Since the global ancestry of the ELA samples were predominantly European and the European local ancestry blocks are uniformly distributed among ALA samples (Supplementary Figure 5), we performed genome-wide differential accessibility analysis. In total we identified 5,154 significant (FDR ≤ 0.05 and FC ≥ 2) differentially accessible peaks between the ancestries in three cell types (astrocytes, excitatory neurons and microglia), representing less than 1% of all called ATAC peaks. 99% of these differentially accessible peaks were seen in the astrocyte cluster 1, with an overall increased chromatin accessibility in the European ancestry blocks compared to African ancestry blocks (more accessible in ELA: 4,546 peaks; more accessible in ALA: 107 peaks) (Figure 3A). This astrocyte specific accessibility difference was widely spread across all chromosomes (Figure 3B). Among all differentially accessible peaks, 25.8% (2,248 peaks) are in promoters and TSS of genes, 33.5 % (2,920 peaks) are intragenic, 34.1% (2,970 peaks) are intergenic and 6.6% (574 peaks) are in distal ELITE enhancers (Figure 3C), altogether corresponding to 6,067 genes.

Figure 3. Chromatin accessibility differences between ancestries genome-wide in Astrocyte cluster 1.

A) Volcano plot representation of global chromatin accessibility difference between ancestries in Astrocyte cluster 1; B) Chromatin accessibility differences across the genome displayed by chromosomes. Blue is increased accessibility in Europeans, Red is increased accessibility in African ancestry; C) Pie chart showing the differentially accessible peak region distribution in Astrocyte cluster 1.

Sixteen percent of the differentially accessible peaks were found on chromosome 19, a significant enrichment as compared to a random chromosomal distribution of peaks (Fischer exact test adjusted p-value=2.501E⁻³¹) with 36.4% of these differentially accessible peaks in the promoter regions of chromosome 19 genes (Supplementary Figure 6).

Pathway enrichment of differentially accessible and expressed genes

To gain mechanistic insights from the integrative snRNA-seq and snATAC-seq approach, we determined the genome-wide overlap between DEG in the astrocyte clusters identified by snRNA-seq and genes associated with differentially accessible peaks identified by snATAC-seq (DAG). From the 6,067 genes with differentially accessible peaks, 239 genes were both differentially expressed and accessible (DEG-DAG) between the ancestries (Supplementary table 6). However only 55 genes had concordant differences in expression and accessibility: 48 with increased accessibility and expression in ELA (including APOE) and 7 with reduced expression and accessibility in ELA. This set of 55 DEG-DAG in astrocytes were used to compute a functional enrichment analysis.

KEGG pathways (KEGG) and Gene Ontology (GO) analysis of gene set enrichment consider different category sets. KEGG pathway analyses identify over-representation of genes participating in the same biological process, while GO analyses aim to reduce complexity and focus on the common functional properties of the gene products. Thus, both analyses were performed, as it is expected that they will highlight different aspects of the data. Analysis of KEGG Molecular functional pathways showed enrichment in signaling pathways including calcium signaling and MAPK signaling pathway. The GO gene-sets highlighted alterations in the biological process of cholesterol metabolism, including regulation of cholesterol metabolic process and biosynthetic process, pathways linked to synapses and axonal transport, spine development and pathways linked to glial cell and astrocyte projection (Supplementary Table 7).

Using Human MSigDB database of disease associated genes from published transcriptomic studies revealed the highest overlap with lipopolysaccharide (LPS) treated astrocytes to induce reactivity (adjusted p-value =8.16E⁻¹⁰, GSE75246)⁵⁴ and with genes showing altered expression in brains (adjusted p-value =2.26E⁻¹⁹, GSE79666)⁵⁵ and astrocytes (adjusted p-value =2.92E⁻¹⁹)⁵⁶ of Huntington Disease (HD) individuals. The overlapping genes between the ancestry DEG-DAG and genes up-regulated in astrocytes of HD patients include a molecular signature of reactive astrocyte markers also observed in the HD striatal astroglia data, but also extends to pathways related to glutamatergic synapse and signaling. In addition, the ENCODE database of brain histone post-translational modifications identified an enrichment of DEG-DAG in those genes having H3K27me3 signals in astrocytes (H3K27me3 astrocytes Hg19, adjusted p-value = 0.02) (Supplementary Table 7).

Chromatin accessibility in AD candidate genes

We also assessed chromatin accessibility differences between the ancestries in AD associated genes, including Mendelian genes (APP, PS1, PS2, and MAPT) and genes suggested by GWAS and rare variant association studies across ancestries^57,58. 32 differentially accessible peaks between the ancestries were identified in seven out of 76 AD-associated loci as defined by the ADSP Gene Verification Committee (https://adsp.niagads.org/gvc-top-hits-list/)⁵⁸. The correlation between accessibility and expression of these seven genes with differential accessibility is shown in Table 3. Aside from APOE, nine differentially accessible peaks were observed to fall in the promoter areas of three genes, SORL1, VRK3 and ABCA7, all with increased accessibility in ELA regions. Ten differentially accessible peaks fall in intergenic regions close to five AD genes, all with increase accessibility in ELA. In addition, twelve peaks are found in the gene body of four AD-associated genes (CLU, PTK2B, SORL1, SPHK1), also with increased accessibility in ELA. Nonetheless, none of these genes containing differentially accessible peaks between local ancestries showed significant differential expression between ancestries (Table 3).

Table 3.

Differentially accessible peaks in Astrocyte cluster 1 in Alzheimer disease GWAS candidate genes.

Gene	Chr.	Peak location	Cell cluster	Distance to TSS	SnATAC FC	SnATAC FDR	SnRNA FC	SnRNA FDR	Region
PTK2B	chr8	27340471	Astrocyte 1	29239	2.327	0.026	−1.05	4.67E-13	intron
CLU	chr8	27596089	Astrocyte 1	18692	4.235	0.019	−1.15	3.39E-73	intergenic
CLU	chr8	27597084	Astrocyte 1	17697	2.556	0.035	−1.15	3.39E-73	exon
CLU	chr8	27597084	Astrocyte 1	17697	2.556	0.035	−1.15	3.39E-73	intron
CLU	chr8	27597917	Astrocyte 1	16864	18.113	0.008	−1.15	3.39E-73	exon
CLU	chr8	27597917	Astrocyte 1	16864	18.113	0.008	−1.15	3.39E-73	intron
CLU	chr8	27607577	Astrocyte 1	7204	6.695	0.032	−1.15	3.39E-73	intron
ACER3	chr11	76799174	Astrocyte 1	−61449	2.41	0.008	1.04	1.12E-30	intergenic
SORL1	chr11	121566449	Astrocyte 1	114496	3.538	0.044	−1.05	4.51E-53	exon
SORL1	chr11	121566449	Astrocyte 1	114496	3.538	0.044	−1.05	4.51E-53	intron
SORL1	chr11	121566449	Astrocyte 1	−26	3.538	0.044	−1.05	4.51E-53	promoter-TSS
SORL1	chr11	121591362	Astrocyte 1	139409	7.332	0.011	−1.05	4.51E-53	intron
SORL1	chr11	121591362	Astrocyte 1	1193	7.332	0.011	−1.05	4.51E-53	promoter-TSS
SORL1	chr11	121596337	Astrocyte 1	6168	6.485	0.046	−1.05	4.51E-53	intron
SORL1	chr11	121596337	Astrocyte 1	144384	6.485	0.046	−1.05	4.51E-53	intron
SORL1	chr11	121695992	Astrocyte 1	105823	3.063	0.039	−1.05	4.51E-53	intergenic
SORL1	chr11	121695992	Astrocyte 1	244039	3.063	0.039	−1.05	4.51E-53	intergenic
SORL1	chr11	121795198	Astrocyte 1	205029	3.099	0.027	−1.05	4.51E-53	intergenic
SORL1	chr11	121795198	Astrocyte 1	343245	3.099	0.027	−1.05	4.51E-53	intergenic
SORL1	chr11	121929505	Astrocyte 1	477552	3.494	0.004	−1.05	4.51E-53	intergenic
SORL1	chr11	121932043	Astrocyte 1	480090	11.926	0.031	−1.05	4.51E-53	intergenic
SPHK1	chr17	76397939	Astrocyte 1	13539	4.176	0.041	−1.01	9.98E-43	intergenic
SPHK1	chr17	76397939	Astrocyte 1	12897	4.176	0.041	−1.01	9.98E-43	intron
SPHK1	chr17	76414406	Astrocyte 1	30006	15.903	0	−1.01	9.98E-43	intergenic
ABCA7	chr19	1039766	Astrocyte 1	−87	2.429	0.001	1.09	2.95E-21	promoter-TSS
ABCA7	chr19	1040877	Astrocyte 1	−119	4.091	0.002	1.09	2.95E-21	promoter-TSS
ABCA7	chr19	1040877	Astrocyte 1	1024	4.091	0.002	1.09	2.95E-21	promoter-TSS
VRK3	chr19	50024717	Astrocyte 1	360	5.864	0.043	1.09	4.88E-24	promoter-TSS
VRK3	chr19	50024717	Astrocyte 1	979	5.864	0.043	1.09	4.88E-24	promoter-TSS
VRK3	chr19	50025268	Astrocyte 1	−116	3.408	0.002	1.09	4.88E-24	promoter-TSS
VRK3	chr19	50025268	Astrocyte 1	428	3.408	0.002	1.09	4.88E-24	promoter-TSS
VRK3	chr19	50050285	Astrocyte 1	−24589	5.267	0.002	1.09	4.88E-24	intergenic

FC: Chr.: Chromosome; FC:Fold change; TSS: Transcription Starting Site; Negative value in Distance to TSS reflects peaks falling upstream of TSS

DISCUSSION:

To understand the complex mechanisms by which APOE4 confers differential risk for AD in the context of genetic ancestry, we evaluated both gene expression and chromatin accessibility profile differences between homozygous APOE4/4 ELA and ALA Alzheimer disease brains at single cell resolution. We confirmed our previous finding¹¹ that APOE is significantly more expressed in astrocytes of ELA compared to ALA. In fact, APOE is the most differentially expressed gene in astrocytes in the 2Mb local ancestry region surrounding APOE. Beyond being the most differentially expressed, the promoter of APOE in astrocytes showed a significant increase in accessibility in ELA compared to ALA. These results suggest that the increased APOE4 expression previously reported in ELA APOE4 carriers¹¹ may be partly due to differences in chromatin remodeling at the promoter of APOE4 between ancestries. We have also shown using Capture Chromatin Conformation analysis and massively parallel reporter assays that specific DNA sequence differences in areas physically interacting with the APOE promoter have functional effects in microglia and astrocytes, with greater expression occurring in ELA sequence variants than ALA⁵⁹. Thus, multiple mechanisms affecting APOE4 gene expression in astrocytes may be activated at different times in life or by stress and could affect the differences in AD risk seen between ancestral homozygous APOE4 carriers. This emphasizes that APOE loci is under a complex regulatory environment that involves cell type-specific processes and strengthens the primary functional role of APOE in astrocytes⁶⁰.

Though our principal research question focused on the underlying mechanisms leading to the higher expression of APOE in ELA brains using snATAC-seq, this study also provides a global view on the chromatin landscape of AD APOE4/4 carriers from different genetic ancestries. One challenge is the relative scarcity of available autopsy material for African Americans. This is made further challenging as we required homozygous local ancestry for African local ancestry and homozygosity for the APOE4 genotype. These requirements limited the available number for final analysis. It highlights the need for an effort to understand the concerns of the African American and Hispanic/Latino populations on participating in autopsy studies and work with these populations to increase the number of autopsies in AD family members. Non-Hispanic Europeans are not admixed and thus almost exclusively European local ancestry, so that the only additional limiting criteria is APOE4 homozygosity.

We observed in astrocyte cluster 1 a significantly higher chromatin accessibility in ELA brains genome wide. A recent study using mesenchymal progenitor cells has reported that APOE accumulation in the nucleus can disrupt and destabilize heterochromatin, which would increase chromatin accessibility, similar to what we observe here in the ELA with higher APOE expression⁶¹. As astrocytes are the primary cells expressing APOE, this could explain the finding of increased global accessibility primarily in this cell type. Notably, only a small percent of significant peaks of higher chromatin accessibility overlap with the significant differentially expressed genes identified by snRNA-seq. Conversely, only 40% of the significant differentially expressed genes showed significant changes in accessibility between the ancestries. This is in accordance with the growing understanding that gene expression depends on multiple regulatory elements⁶², many located far from the gene locus itself²². Some of the DEG-DAG show a positive correlation between expression and accessibility, suggesting a functional relationship. However, we also observed an anti-correlation in accessibility-expression pairs. This anti-correlation could represent binding by repressors leading to a decrease in expression as previously described^63,64. In addition, highly accessible sites not associated with enhanced expression could represent poised enhancers/promoter with no active transcription occurring⁶⁵. Besides this known biological partial correlation of accessibility and expression, in this study the snATAC and snRNA were derived from separate, but adjacent, brain tissue. Differences in tissue composition could compound the discordance between these two outcomes. Future studies with paired snATAC-seq and snRNA-seq from the same nuclei such as the 10X Genomics Single Cell Multiome will minimize tissue composition effects if present. In fact, while we do identify differential expression of APOE between ancestries in several cell types (e.g., microglia), we only see significant differential accessibility of its promoter and immediate region in astrocytes.

Astrocytes have well established roles in neuronal metabolic support and neuroprotection, synapse formation and function, ion signaling homeostasis, integrity of the blood brain barrier, tissue repair and complex brain functions such as memory and sleep^66–70. Thus, it is not surprising that a relevant role for astrocytes in AD pathology is suggested⁷¹. Our data suggest that astrocytes are involved in the modulation of APOE4-afforded AD risk and pinpoint this cell type as one of the most sensitive to differences in genetic ancestry. The significant enrichment of astrocytic ancestry-associated DEG-DAG in genes upregulated in HD astrocytes suggest that the differences associated with diverse ancestry entail functional properties that are potentially common to other neurodegenerative disorders. In addition to this astrocyte-ancestry association, we also evidence a difference in DEG-DAG between our astrocyte clusters 1 and 2. We observed that our astrocyte cluster 1 was transcriptionally most similar to astrocytes subclusters a4 and a8 described by Grubman et al⁵², clusters where APOE was upregulated in AD⁵²; while our astrocyte subcluster 2 was transcriptionally comparable to astrocytes subcluster a2 where APOE was downregulated.

Interestingly, there was a significant enrichment of differential chromatin accessibility and expression on chromosome 19 relative to the rest of the genome (Supplemental Figure 5) in astrocyte cluster 1. These results correlate with the observation of a comparatively high number of ATAC-seq peaks in chromosome 19⁷², potentially due to the high gene density, GC content and DNA binding proteins in chromosome 19 compared to other chromosomes^73–75. It is possible the high gene density and the structural nature of chromosome 19 could modulate the differential binding of chromatin modifying enzymes, explaining the enrichment in differentially accessible chromatin and gene expression in this specific chromosome observed between ancestries.

Our analysis of transcription factors with binding sites in the differentially accessible peaks at the 5’end of APOE that were both differentially expressed and accessible between ancestries suggests transcriptional modulators of APOE expression differences. Of the 15 transcription factors we identified, only SREBF1 and PURA showed increased accessibility (intergenic and intronic) and expression in ELA samples and KLF6 showed increased accessibility (distal enhancer) and expression in ALA samples. A polymorphism in SREBF1 has been shown to influence the risk of AD specifically in APOE4 carriers⁷⁶, while PURA has been shown to regulate expression of AD and Aβ clearance related genes⁷⁷. On the other hand, KLF6 is found to be differentially expressed in AD brains and implicated in Aβ-induced oxidative stress⁷⁸. Their role with APOE activation has not been previously described, although SREBF1 is a major regulator of lipid metabolism.

Although many genetic risk factors for disease are expected to be shared among populations, genomic diversity among ancestries can provide new opportunities for discoveries regarding disease susceptibility. In AD, understanding why the APOE4 allele confers a lower risk to African ancestry carriers versus European ancestry carriers is instrumental for the identification of potential AD therapeutics. Here we provide chromatin accessibility and gene expression data from AD APOE4 carriers of African and European ancestries at single cell resolution. One of the main challenges in comparing local ancestry effects is the current limitation of available samples from diverse ancestries. The need for both specific genotype and local ancestry criteria further restricted the number of samples available. This highlights the importance of increasing autopsy material from diverse populations. Expansion of this study into other APOE genotypes and more diverse ancestries could provide further insight into the importance of brain astrocytes and chromatin landscape of APOE, as well as genome-wide changes, implicating specific molecular pathways and regulatory proteins in this risk difference. Finally, these findings support the concept of reducing APOE4 expression as a potential therapeutic pathway for AD.

CONCLUSIONS:

Our results provide novel compelling insights to understand the AD risk difference known to exist between European and African APOE4 local ancestry carriers. Here we demonstrate that differences in chromatin accessibility between the African and European APOE locus could explain the expression differences in APOE expression in astrocytes and supports the concept that this contributes to the risk difference between African American and European populations for AD in APOE4 carriers. We found that this increase in accessibility in the region surrounding APOE4 in European astrocytes extended genome-wide, suggesting a wider mechanism of regulatory dysfunction is occurring in these cells. Not only is it important to include diverse ancestries to ensure all individuals are represented in research, but also we have shown here and in previous work that including diversity in research can provide additional windows of opportunity to elucidate disease and biological mechanisms. Understanding regulatory dysfunction in AD has not been well studied and should be and area of future research in AD.

Supplementary Material

Suppl Fig

Supplementary Figure 1. SnATAC-seq data analysis for sample bias.

Supplementary Figure 2. SnATAC-seq cell marker cluster identification by chromatin accessibility.

Supplementary Figure 3. Astrocyte subtype classification.

Supplementary Figure 4. APOE chromatin accessibility in the other cell clusters where APOE is highly expressed but not differentially accessible between the ancestries.

Supplementary Figure 5. Visualization of local ancestry across all chromosomes per individual sample.

Supplementary Figure 6. Enrichment of differentially accessible peak between ancestries in chromosome 19.

Suppl Tab

Supplementary Table 1. Summary of snATAC and snRNA sequencing results.

Supplementary Table 2. Total cell count and proportion per cluster in the integrated sequencing analysis.

Supplementary Table 3. snRNA sequencing clusters with significant APOE4 differential expression between ancestries.

Supplementary Table 4. TFs (DEGs-DAGs) binding to differentially accessible peaks at the promoter of APOE.

Supplementary Table 5. APOE Local ancestry accessibility analysis.

Supplementary Table 6. Differentially accessible and expressed genes (DEG-DAG) in astrocyte cluster 1.

Supplementary Table 7. Subset of the differential functional analysis of the astrocyte cluster 1 DEG-DAG with positive correlation between expression and accessibility using EnrichR software.

LIST OF ABBREVIATIONS:

AD

Alzheimer Disease

ABCA7

ATP Binding Cassette Subfamily A Member 7

ACER3

Alkaline Ceramidase 3

ADRC

Alzheimer Disease Research Centers

AIF1

Allograft inflammatory factor 1

ALA

African Local Ancestry

APOC1

Apolipoprotein C1

APOE

Apolipoprotein E

APOE4

Apolipoprotein E allele 4

APOE4/4

Homozygous for Apolipoprotein E allele 4

APP

Amyloid Beta Precursor Protein

ATAC

Assay for Transposase-Accessible Chromatin

Azimth

App for reference-based single-cell analysis

Aβ

Amyloid beta

BRAAK

Braak staging (pathology score for Alzheimer Disease)

CGT

Center for Genome Technology

CLU

Clusterin

COL1A2

Collagen Type I Alpha 2 Chain

CSPG4

Chondroitin Sulfate Proteoglycan 4

DAG

Differentially Accessible Genes

DAP

Differentially Accessible Peaks

DEG

Differentially Expressed Genes

DEG-DAG

Differentially Expressed and Accessible Genes

ECHAD

Entorhinal Cortex in Human Alzheimer’s Disease

ELA

European Local Ancestry

ELITE

enhancer gene with both a high likelihood enhancer definition and a strong enhancer–gene association

ENCODE

Encyclopedia of DNA Elements

FC

Fold Change

FDR

false discovery rate

FOS

Fos Proto-Oncogene, AP-1 Transcription Factor Subunit

GC

Guanine-Cytosine content

GeneSigDb

Gene Signature Data Base

GFAP

Glial Fibrillary Acidic Protein

GLI2

GLI Family Zinc Finger 2

GO

Gene Ontology

GRCh38

Genome Reference Consortium Human Build 38

GREAT

Genomic Regions Enrichment of Annotations Tool

H3K27me3

tri-methylation of lysine 27 on histone H3 protein

HD

Huntington Disease

HDSigDb

Huntigton Disease Signature Database

HIF1A

Hypoxia Inducible Factor 1 Subunit Alpha

HIHG

John P. Hussman Institute for Human Genomics

HOMER

Hypergeometric Optimization of Motif EnRichment

JASPAR

transcription factor binding profile database

JOLMA

Human specific transcription factor binding

JUNB

JunB Proto-Oncogene, AP-1 Transcription Factor Subunit

KEGG

Kyoto Encyclopedia of Genes and Genomes

KLF15

Kruppel Like Factor 15

KLF2

Kruppel Like Factor 2

KLF6

Kruppel Like Factor 6

LA

Local Ancestry

LHX2

LIM Homeobox 2

LPS

lipopolysaccharide

LSI

Iterative Latent Semantic Indexing

MAG

Myelin Associated Glycoprotein

MAPT

microtubule-associated protein tau

MB

Million Base pair

Mb

Mega base

MsigDB

Molecular Signatures Database

MXI1

MAX interactor-1

NACC

National Alzheimer Coordinating Center

NPAS2

Neuronal PAS Domain Protein 2

OPCs

oligodendrocyte precursor cells

PanglaoDB

Single-cell sequencing mouse and human database

PCR

Polymerase chain reaction

PSEN1

Presenillin-1

PSEN2

Presenillin-2

PURA

Purine Rich Element Binding Protein A

Reactome

pathway database

RSB

Resuspension Buffer

RXRA

Retinoid X Receptor Alpha

SLC17A7

Solute Carrier Family 17 Member 7

SLC32A1

Solute Carrier Family 32 Member 1

snATAC-seq

single nuclei Assay for Transposase-Accessible Chromatin sequencing

snRNA-seq

single nuclei RNA sequencing

SORL1

Sortilin Related Receptor 1

SPHK1

Sphingosine Kinase 1

SREBF1

Sterol Regulatory Element Binding Transcription Factor 1

TEAD1

TEA Domain Transcription Factor 1

TF

Transcription Factor

TOMM40

Translocase Of Outer Mitochondrial Membrane 40

TREM2

Triggering Receptor Expressed On Myeloid Cells 2

TSS

Transcription Starting Site

TTS

transcription terminating site

UCSC

University of California Santa Cruz

USUHS

Uniformed Services University of the Health Sciences

VLMC

vascular leptomeningeal cell

VRK3

VRK Serine/Threonine Kinase 3

ZBTB7C

Zinc Finger And BTB Domain Containing 7C

WGS

Whole Genome Sequencing

Availability of data and material

Data are available through the National Institute on Aging Genetics of Alzheimer’s Disease Data Storage Site (NIAGADS) Data Sharing Service (DSS): https://dss.niagads.org/datasets/ng00067/