Identifying differential regulatory control of APOEɛ4 on African versus European haplotypes as potential therapeutic targets

Abstract

We previously demonstrated that in Alzheimer’s disease (AD) patients, European apolipoprotein E (APOE) ε4 carriers express significantly more APOEε4 in their brains than African AD carriers. We examined single nucleotide polymorphisms near APOE with significant frequency differences between African and European/Japanese APOE ε4 haplotypes that could contribute to this difference in expression through regulation. Two enhancer massively parallel reporter assay (MPRA) approaches were performed, supplemented with single fragment reporter assays. We used Capture C analyses to support interactions with the APOE promoter. Introns within TOMM40 showed increased enhancer activity in the European/Japanese versus African haplotypes in astrocytes and microglia. This region overlaps with APOE promoter interactions as assessed by Capture C analysis. Single variant analyses pinpoints rs2075650/rs157581, and rs59007384 as functionally different on these haplotypes. Identification of the mechanisms for differential regulatory function for APOE expression between African and European/Japanese haplotypes could lead to therapeutic targets for APOE ε4 carriers.

1 |. NARRATIVE

1.1 |. Contextual background

The most significant genetic risk factor for Alzheimer’s disease (AD) is apolipoprotein E (APOE), with the APOE ε4 allele inferring both an increased risk for disease and a decrease in age at onset.^1,2 APOE ε4 is thought to contribute almost 40% of the genetic risk for AD in European populations. Yet, there is no therapeutic intervention to reduce this risk for APOE ε4 carriers, nor is there full understanding of how APOE ε4 leads to AD.

However, the risk of developing AD from the APOE ε4 allele varies greatly among populations³ (Figure 1). Despite having the same APOE ε4 gene, nature has found a way to reduce the risk for APOE ε4 carriers in those individuals with African ancestry. This finding suggests that if we could identify the mechanism that protects (or reduces risk in) APOE ε4 carriers of African ancestry from developing AD from APOE ε4, we could use this information to develop therapeutic treatments to slow or prevent AD in APOE ε4 carriers of all ancestries. In this paper, we identify potential areas of the genome that may begin to fulfill this goal.

FIGURE 1.

Apolipoprotein E (APOE) ɛ4 risk effects in different populations. Odds ratios for the risk of developing Alzheimer’s disease from APOE ɛ4 compared to APOE ɛ3 for homozygous carriers

Using populations with recent ancestral admixture (African Americans, e.g., African and European ancestries, and Latino/Hispanic populations, e.g., African, European, and Amerindian ancestries), Rajabli et al.⁴ and later Blue et al.⁵ demonstrated that this protective mechanism for AD in African APOE ε4 carriers was associated with the ancestral origin of the DNA surrounding APOE ε4, and not elsewhere on the chromosomes. That is, if you inherited the area surrounding the APOE ε4 gene (called the “local ancestry,” or LA) from African ancestors, you have the African risk for AD from APOE ε4, and if you inherited APOE ε4 and its surrounding genetic region from European ancestors, you have the European risk for AD from APOE ε4. However, what the genetic changes are that cause this risk difference for local APOE ε4 ancestry is not known. One major problem is that most genomic studies have been done only in European genomes, thus there is a real gap in knowledge about how differences in genetic sequence among ancestries can affect genomic function and risk.

As the apoE4 protein is similar, if not identical among these populations, other genetic factors in the region other than those producing protein must be driving this protection. This led us to explore and recently publish that the amounts of APOE ε4 expressed in AD brains with European LA are significantly higher (P < 10^–318) than those in which the APOE ε4 gene was surrounded by African ancestry. In this study, we sought to identify what genetic changes in regulation could lead to this difference in APOE ε4 expression among local ancestries.

1.2 |. Study approach and findings

To determine those base pairs in the LA region that are significantly different in allele frequency between African and European APOE ε4 genomes, we used sequence data from the 1000 Genomes Project database⁴ (Figure 2). Next, we focused on the region where these significant variants seem to accumulate, which was located within a predicted regulatory region called a topologically associated domain (TAD) surrounding APOE. TADs are genomic regions with increased cross regulation within its boundaries (for more information on TADs, see Pombo and Dillon⁶). To determine whether these sequence differences led to differences in regulatory function, we used three different methods of “reporter assays.” In these assays, pieces of DNA, each representing one of the identified sequence differences in regions with potential enhancer activity, are placed in a vector with a “reporter” gene, encompassing a barcode or coding for a fluorescence-producing protein. These vectors are then inserted into a living cell and use the cell’s inherent regulatory transcription components to activate expression of the reporter. How the sequence differences enhance the amount of barcode expressed from the vector or fluorescence emitted by the coded protein can be measured. All fragment assays were tested in three different cell types, Neuron (SH-SY5Y), Microglia (HMC3), and Astrocyte (U-118) lines. The first two reporter assays are based on the technique of massively parallel reporter assay (MPRA, described in more detail below), which allows many reporter assays to be done at once through measurement of the barcodes. Within the MPRA approach, we used two different forms of DNA fragments. The haplotype or polymerase chain reaction (PCR)-based assay used larger DNA fragments (≈850 bp) surrounding the variants we were testing. This is important as often not just one variant, but a cluster of variants can affect regulatory function. We also used MPRA with synthetically made DNA fragments (probe-based), which were much smaller and focused on only a small region surrounding a single variant. This allowed us to determine, in those regions where multiple sequence differences between the ancestries were close together, which variant of the clustered variants were likely driving the functional difference. For the variants that had differential function between the ancestries or failed in the MPRA test, we completed a third approach with single reporter assays as well.

FIGURE 2.

Overview of experimental set-up. A, All variants identified to have differential frequency between European (1000 Genomes CEU) versus African (1000 Genomes YRI) apolipoprotein E (APOE) ɛ4 haplotypes; color-coded for significance; red; P < 1*10–5, orange P < 1*10–4 (corresponding to the top 25% results), rest P < .05. B, All variants identified to have differential frequency between Japanese (1000 Genomes JPT) versus African (1000 Genomes YRI) APOE ɛ4 haplotypes; color-coded for significance; red; P < 1*10–5, orange P < 1*10–4, rest P < .05. C; All 39 designed amplicons for inclusion in the polymerase chain reaction (PCR)-based massively parallel reporter assay (MPRA)

Finally, the question remains if the identified functional areas affect the expression of the APOE gene or some other locus? To determine this, we used a three-dimensional interaction method called “promoter Capture C” (for chromatin conformation capture techniques, see Davies et al.⁷). In this chromatin conformation capture technique, physical connections between genomic locations and gene promoters are detected, allowing for identification of enhancer-to-promoter interactions. Here we are specifically looking to see if our identified functional regions are physically interacting with the APOE promoter in the cells.

Through these analyses, we have identified regions in the introns 2–3 of TOMM40 (fragments 10 and 13; Figure 3A–C) that are strong candidates to contribute to the difference in APOE ε4 expression between ancestries. The single-variant assays pinpointed the difference in activity in those fragments to be driven by variants rs2075650/rs157581 (European/Japanese allele G-C; fragment 10) and rs59007384 (European allele T; fragment 13; Figure 3D). Rs2075650 and rs59007384 are predicted to significantly affect binding sites for transcription factors SP1 and PLAG, respectively, based on in silico MotifBreakR analysis. The promoter Capture C analyses showed that these functional regions had strong chromatin interaction with the APOE promoter in both glial cell types, but not in neurons (Figure 3E).

FIGURE 3.

Overview of results and regulatory evidence in the region of interest. A, All 39 designed amplicons for inclusion in the polymerase chain reaction (PCR)-based massively parallel reporter assay (MPRA). B, Representation of results of successful fragments in the PCR-based MPRA. C, Representation of results of successful fragments in the luciferase experiments. D, Representation of results of successful variant analyses in the probe-based MPRA. Blue; astrocytes. Green; microglia. Pink; neurons. E, Significant interactions of apolipoprotein E (APOE) promoter within the region of interest based on the promoter Capture C data at 4-DpnII fragment resolution; interactions between promoter and candidate regulatory element (green), interaction between promoters (red). F, Available in silico annotation data; including previously reported enhancers, enhancers and H3K27ac marks identified in prefrontal cortex from the PsychENCODE project, chromHMM tracks from the RoadMap Epigenome Project (primary HMM) and Transcription factor ChIP-Seq cluster data from ENCODE freeze 3; using data from 338 factors in 130 cell types

Therefore, several findings support these regions being important for APOE regulation: (1) They demonstrated functional differences in expression using microglia and astrocytes, but not neurons. This fits any expected regulatory region of APOE, as brain APOE is primarily expressed in astrocytes and microglia, but minimally in neurons. (2) Higher enhancer activity (corresponding to higher expression) was found for the European alleles compared to the African alleles, supporting our previous observations in AD brains between African LA and European LA APOE ε4.⁸ (3) The promoter Capture C analyses evaluating chromatin interaction within the region of interest showed that these functional regions had strong chromatin interaction with the APOE promoter in both glial cell types, but not in neurons. (4) These variants/regions have been described to be involved in AD before,^9–12 but never in context with APOE ε4 or ancestral background.

1.3 |. Limitations and disease implications

The expression of APOE is thought to be complex,^13–23 and thus, it is highly likely that there are other mechanisms that control APOE ε4 expression, both at genomic, RNA, and protein levels. We have focused on the sequence differences in the genomic regions closest to APOE, but the ancestral regions found in admixed populations around APOE can be larger in size and may include other genomic regions in its regulation. Alternatively, sequence differences between the two APOE ε4 backgrounds could also influence methylation activity in this region. Although differences in methylation between APOE ε4 and the other APOE statuses have been studied,^21–24 the effect of different ancestral backgrounds has not been fully assessed.²⁵ Preliminary data from our studies did not find significant methylation differences involving variants with differential allele frequencies between ancestries.

While we have used multiple approaches to measure regulatory function, these approaches are still artificial, and thus, the next steps will be to test the effect of changing the sequence in these potential regions by using CRISPR techniques in astrocytes, microglia, and neurons derived from inducible pluripotent stem cells from AD patients. For example, we could assess whether repression of APOE expression through CRISPR interference assays in cell lines derived from European APOE ε4 carriers mimic AD-relevant phenotypes of African APOE ε4 carriers. If validated, these cell lines could then serve as models to test molecules as potential therapeutic interventions for treating APOE ε4 carriers by manipulating the gene expression of APOE ε4 in a similar fashion as has occurred in the African carriers of the risk allele. This work also highlights the importance of including all ancestries in research in AD, as differences in risk can provide windows of opportunity to understand the mechanisms of disease.

It has long been a matter of debate whether increasing or decreasing APOE ε4 would have therapeutic effects^26,27 citing both gain of function and loss of protective function as possible mechanisms for APOE ε4 effects. Interestingly though, anti-apoE4 immunotherapy in mice has consistently showed improvement of AD phenotypes (i.e., cognition, amyloid beta plaque burden).^28–30 More recent functional studies in mice have shown that removal of APOE ε4 in astrocytes is protective against tau-mediated neurodegeneration³¹ and that the positive effect of repetitive transcranial magnetic stimulation on cognition in milder AD³² is mitigated through reduction of apoE.³³ Taken together with the data presented here, these reports support that the increased risk for AD in APOE ε4 on European LA can be ascribed to increased, pathogenic levels of apoE4. Therefore, this gain of toxicity mechanism of APOE ε4 (on European LA specifically but also overall in all carriers) supports the importance of identifying the controlling regions that contribute to the higher expression of APOE ε4 in European ancestry⁸ and development of therapies targeting the overexpression of APOE ε4.

2 |. CONSOLIDATED RESULTS AND STUDY DESIGN

The LA variants identified having significant differential allele frequencies on the different risk APOE ε4 haplotypes peaked in a small topologically associated domain surrounding APOE (chr19:45375k-45440k)⁴. We selected all variants that met Bonferroni significance (adjusted P < .05, N = 56) and had a greater than 0.01 allele frequency (Table 1). For the MPRA assay using large PCR-based genome fragments we amplified 39 larger fragments (≈850 bp) encompassing the significant LA variant haplotypes using PCR (Figure 2). The protocol for library creation in an enhancer reporter plasmid backbone was adjusted from Trizzino et al.³⁹ In short, candidate enhancer fragments and barcodes are subsequently cloned into the vector in front and in the 3′UTR of the reporter gene, respectively, to create unique haplotype-to-barcode links within the same plasmid. The PCR-based MPRA was supplemented with single luciferase reporter assays using the same amplicon primers as for the MPRA. In the second MPRA assay, we used shorter, synthetically made probes differing just in alleles for one variant at a time. Combined variant allele and barcode fragments are cloned into the pGL4 plasmid backbone, with subsequent insertion of the reporter gene between them, as previously described by Tewhey et al.⁴⁰ Differences in proportion of the two haplotypes/alleles in the expressed material (RNA) versus the expected ratio (plasmid DNA library) was assessed by exact binomial test.³⁴

TABLE 1.

Overview included variants in analyses and summarized results

Fragment	Centered LA SNPs	Location	Allele most common in ɛ4		Enhancer activity (highest activity, cell type)					GTEx
			AFR	EUR/JPT	PCR-based MPRA	Luciferase	Probe-based MPRA	Capture C APOE promoter interaction	RoadMap Epigenome	brain	non-brain	MotifbreakR
1	rs34278513	intron PVRL2	C	T	Negative	NA	JPT astrocytes				BCAM, PVRL2
2	rs3865427	intron PVRL2	C	A	NA	JPT neurons	JPT astrocytes				BCAM, PVRL2
3	rs71352237*	intron PVRL2	T	C	NA	JPT neurons	EUR astrocytes		Enhancer hepatocytes, adrenal gland		BCAM, PVRL2	RARB & RARG
	rs71171301*	intron PVRL2	–	C			NA				BCAM, PVRL2	NA
	rs34224078*	intron PVRL2	A	G			Negative				BCAM, PVRL2
	rs35879138*	intron PVRL2	T	A			Negative				BCAM, PVRL2
4	rs142042446	intron PVRL2	–	TAA	Negative	NA	NA				BCAM
5	rs283808	intron PVRL2	C	A	NA	negative	AFR astrocytes			PVRL2	TOMM40
	rs283809	intron PVRL2	G	A			Negative			PVRL2	TOMM40
6	rs12972156	intron PVRL2	C	G	Negative	NA	Negative		Enhancer hepatocytes		BCAM	TFAIP2A
	rs12972970	intron PVRL2	G	A			AFR astrocytes				BCAM	FOXD1
7	rs34342646	intron PVRL2	G	A	NA	EUR astrocytes	AFR microglia, astrocytes				BCAM, TOMM40
8	rs6857	3′UTR PVRL2	C	T	NA	AFR microglia	–				BCAM, APOC1
9	rs71352238	Up TOMM40	T	C	NA	NA	AFR astrocytes		Promoter TOMM40		BCAM, TOMM40
10	rs2075650*	intron TOMM40	A	G	EUR microglia	NA	Negative	NPC, astrocytes, microglia	Enhancer brain		BCAM	SP1
11	rs34095326	intron TOMM40	G	A	NA	AFR neurons	Negative	NPC, astrocytes, microglia			BCAM	HMGA2 & EHF
	rs34404554	intron TOMM40	C	G			AFR astrocytes		Enhancer brain		BCAM
12	rs11556505	exonic TOMM40	C	T	NA	Negative	Negative	NPC, astrocytes, microglia	Enhancer brain, hepatocytes, adrenal gland
13	rs59007384*	intron TOMM40	G	T	EUR microglia, astrocytes	EUR microglia	EURmicroglia	NPC, astrocytes, microglia	Enhancer brain, hepatocytes, adrenal gland		APOC1	PLAG
	rs157583	intron TOMM40	T	G			Negative
14	rs157584	intron TOMM40	C	T	Negative	NA	Negative	NPC, astrocytes, microglia	Enhancer adrenal gland		APOC1P1, APOE, APOC1
15	rs157585	intron TOMM40	C	A	NA	Negative	Negative	NPC, astrocytes, microglia	Enhancer adrenal gland			MAFK
16	rs157588	intron TOMM40	T	C	NA	AFR astrocytes	Negative		Enhancer adrenal gland		APOC1P1, APOE
17	rs157590	intron TOMM40	C	A	AFR astrocytes	Negative	Negative		Enhancer adrenal gland		APOC1P1, APOE	FEV & EOMES
18	rs205909	intron TOMM40	G	T	Negative	NA	Negative	NPC		APOE	CBLC
19	rs542555887	intron TOMM40	T	–	NA	NA	NA			APOE	CBLC
	rs491153	intron TOMM40	T	C			Negative				CBLC
	rs490243	intron TOMM40	T	C			Negative			APOE	CBLC
20	rs417357	intron TOMM40	T	C	Negative	NA	NA	Astrocytes, microglia		APOE	CBLC
21	rs394819	intron TOMM40	T	G	Negative	NA	Negative	Astrocytes, microglia		APOE	CBLC
22	rs435380	up APOE	A	G	Negative	NA	Negative	NPC, astrocytes, neurons	Enhancer brain, adrenal gland; promoter APOE hepatocytes	APOE	CBLC	ZFX
	rs446037	up APOE	T	G			Negative			APOE	CBLC
	rs434132	up APOE	G	C			Negative			APOE	CBLC
23	rs449647	up APOE	T	A	NA	Negative	AFR astrocytes	NPC, astrocytes, neurons	Promoter APOE brain, hepatocytes		PVRL2
24	rs769449	intron APOE	G	A	Negative	NA	Negative		Promoter APOE brain, hepatocytes		APOC1
25	rs75627662	intergenic	C	T	NA	Negative	EUR astrocytes		Enhancer hepatocytes, adrenal gland		APOC1, APOE	many
	rs148353395*		–	del	NA			APOE, CLASRP	CBLC
26	rs445925	up APOC1	A	G	NA	EUR microglia	Negative		Enhancer all		APOE	PPARG
	rs10414043*	up APOC1	G	A			Negative				APOC1
	rs7256200*	up APOC1	G	T			Negative				APOC1
27	rs390082	up APOC1	G	T	Negative	NA	Negative		Enhancer all		APOE
28	rs34954997 = rs11568822*	5′ UTR APOC1	–	CGTT	AFR microglia	Negative	NA		Promoter APOC1 all		APOC1, APOE
29	rs5117*	intron APOC1	T	C	NA	NA	Negative				APOC1, APOE
30	rs389261	intron APOC1	A	G	NA	NA	Negative
31	rs12721046	intron APOC1	G	A	NA	NA	EUR astrocytes
32	rs12721051	intron APOC1	C	G	NA	NA	Negative				APOE
33	rs56131196*	down APOC1	G	A	Negative	Negative	Negative				APOC1, APOE
	rs4420638*	down APOC1	A	G			Negative				APOC1, APOE
34	rs157591	intergenic	A	G	NA	Negative	Negative		Enhancer hepatocytes, adrenal gland	CLASRP	CBLC
35	rs157596	intergenic	T	C	Negative	NA	Negative				CBLC
36	rs157597	intergenic	C	A	NA	negative	NA		Enhancer hepatocytes, adrenal gland		CBLC
37	rs111789331	intergenic	T	A	NA	AFR neurons	Negative		Enhancer hepatocytes, adrenal gland
38	rs157599	intergenic	G	A	NA	Negative	Negative		Enhancer all
	rs149345	up APOC1P1	G	T			Negative
	rs66626994	up APOC1P1	G	A			Negative
39	rs10424663	up APOC1P1	A	G	Negative	NA	Negative			APOE, CLASRP	CBLC

Abbreviations: APOE, apolipoprotein E; Down, downstream; EUR: European, AFR: African, JPT: Japanese; LA, local ancestry; MPRA, massively parallel reporter assay; NA, not assessed/low coverage/failed; NPC, neuroprogenitor cells; SNP, single nucleotide polymorphism; Up, upstream.

^*= frequency different general population versus APOE ɛ4 haplotypes.

We performed high resolution (i.e., using the 4 bp-cutter DpnII, as opposed to the more commonly used 6 bp-cutter HindIII) genome-wide promoter-focused Capture C analyses on neuroprogenitor cells (NPCs), astrocyte, microglia, and neuronal cell lines, following the previously described protocol.³⁵ Significant interactions at 4-DpnII fragment resolution were called using CHiCAGO.³⁶

Overall results on enhancer activity—assayed in representatives of microglia, astrocytes, and neurons—across all methods can be found in the overview of Table 1, and are visualized in Figure 3.

Specifically, we identified a significant regulatory region in microglia and astrocytes supported by several levels of evidence. We identified differential activity in microglia and astrocytes between significant LA variant haplotypes for fragments 10 and 13, both located in the first introns of TOMM40 (Table 2). Higher enhancer activity is observed for the European/Japanese APOE ε4 haplotype in fragment 10 (containing two significant LA variant alleles: rs2075650-G and rs34404554G) and the European APOE ε4 haplotype in fragment 13 (containing two significant LA variant alleles: rs59007384-T and rs157583-G; Table 2). Haplotype fragment 13 showed consistently higher activity driven by the significant LA variant rs59007384 European allele T in the PCR-based MPRA and luciferase assays (both glia cell lines), and the probe-based analyses (in microglia), supporting the T allele variant as the driving signal in fragment 13. Besides the two significant LA variants (rs2075650 and rs34404554), European/Japanese and African haplotypes in the PCR-based fragment 10 also contained two additional variants that did not meet Bonferroni significance in the local ancestry comparison (European/Japanese haplotype; rs157580-G and rs157581-T; Table 2). Luciferase analyses of overlapping fragment 11, centering on significant LA variants rs34095326 and rs34404554, did not support differential activity driven by these variants in microglia and astrocytes, suggesting that rs2075650 is the driving factor for the regulatory activity seen with fragment 10. Although two out of three replicates showed increased enhancer activity for the European/Japanese significant LA variant rs2075650-G allele, this effect could not be replicated in the probe-based analyses (Table S1 in supporting information). Rs2075650 and rs157581 are both predicted by the JASPAR transcription factor binding database to be located in transcription factor binding sites and are located within 100 bp of each other, potentially indicating a synergistic effect of these two TAD variants that was missed in the probe-based single variant MPRA. Additionally, according to MotifbreakR, both rs2075650 and rs59007384 are predicted to significantly affect binding sites for SP1 and PLAG, respectively.

TABLE 2.

Results TOMM40 intronic potential enhancer region

SNPs	PCR-based MPRA (AFR - EUR/JPT haplotype)			PCR-based results	Luciferase (AFR EUR/JPT haplotype)		Luciferase results	Probe-based results	ChIPs-Equation (Haploreg)	RegulomDB		MotifBreakR	GTEx
	10	11	12		11	12				Score	JASPAR
rs157580	A-G	–	–	Fragment 10; EURmicroglia	–	–	Fragment 10; NA	NA	–	1f	–	–	APOE skin, APOC1P1 cultured fibroblasts and skin, APOC1 skin
rs2075650	A-G	A-G	–		A-G	–		Negative	–	1b	RREB1	SP1	BCAM tibial nerve
rs157581	T-C	C-C	–		C-C	–		NA	–	2b	ZEB1	–	APOC1mucosa esophagus
rs34095326	G-G	G-A	G-A	Fragment 11; NA	G-A	A-A	Fragment 11; AFR neurons	Negative	POL2 (K562)	4	–	HMGA2 & EHF	BCAM tibial nerve
rs34404554	C-G	C-G	C-G		C-G	G-G		AFR astrocytes	POL2 (K562)	4	–	–	BCAM tibial nerve
rs11556505	–	C-T	C-T	Fragment 12; NA	C-T	C-T	Fragment 12; negative	Negative	POL2 (K562)	4	–	–	–
	13	14			13
rs59007384	G-T	G-T	–	Fragment 13; EUR astrocytes, microglia	G-T	–	Fragment 13; EUR microglia	EUR microglia	GR (A549)	4	–	PLAG	APOC1mucosa esophagus and adrenal gland
rs157583	T-G	T-G	–		G-G	–		Negative	GR (A549)	4	–	–	–
rs157584	C-T	C-T	–	Fragment 14; negative	C-C	–	Fragment 14; NA	Negative	GR (A549)	4	–	–	APOE skin and atrial appendage heart, APOC1 atrial appendage heart

Bold SNPs = significant LA variants, Italic/underline; variants in center of fragment.

Abbreviations: APOE, apolipoprotein E;MPRA, massively parallel reporter assay; PCR, polymerase chain reaction.

Promoter Capture C data in microglia, astrocytes, and NPCs consistently show physical interaction of the region encompassing fragments 10–14 with the APOE promoter. All interactions within the small TAD are shown in Figure 3. Additional available in silico annotation data (Figure 3F) support a regulatory function of this region, with RoadMap Epigenome chromatin mark analyses indicating enhancer activity in the region overlapping with fragments 10–12 (TOMM40 introns 2–3) in AD-relevant brain regions (hippocampus, frontal cortex), and fragment 13 (TOMM40 intron 4) in other brain regions (e.g., cingulate gyrus, substantia nigra). Additionally, H3K27ac peaks from prefrontal cortex tissue in PsychENCODE support active regulatory activity in the TOMM40 introns 1–2 region (overlapping fragments 10–11). No evidence for effect on expression of APOE (or any other gene) was observed in the Genotype-Tissue Expression (GTEx) project for these variants in the brain.

3 |. DETAILED METHODS AND RESULTS

3.1 |. Methods

See supporting information for full details on the library creation, transfection, and RNA extraction experiments.

3.1.1 |. Reporter assays design

APOE lies within a larger well-defined TAD, based on the 3D genome browser data from hippocampus³⁷ and Hi-C data from the Jin lab³⁸ (Figure S1 in supporting information). The significant LA variants peaked in a small region (≈subTAD) surrounding APOE (chr19:45375k-45440k);⁴ which we selected for further analyses. We selected all variants identified through the 1000 Genomes Project in this subTAD region directly surrounding APOE that met Bonferroni significance in differential frequency on the African versus European or African versus Japanese APOE ε4 haplotype (Bonferroni-adjusted P < .05) and had a greater than 0.01 allele frequency (Table S2 in supporting information). To assess the functional potential of these variants, we designed two different MPRA assays, one using large PCR-based genome fragments and the other using a probe-based design assessing single variants.

In the first assay, we amplified larger fragments (≈850 bp) encompassing the significant LA haplotypes using PCR (PCR-based MPRA; Figures 2, 3 and Table S2). These larger amplicons allow for full inclusion of larger regulatory elements in the fragment, as well as inclusion of multiple variants closely located to each other on the same haplotype permitting inclusion of synergistic signals. This allows the fragments to contain additional LA located on African or European haplotypes that could participate in regulation but did not survive the significance threshold determined by Rajabli et al.⁴ Fragments 26–27 and 39 overlap with reported enhancer regions Multi-enhancer E1 and 2 (ME1–2^18,19); B control region (BCR)²⁰ is not included, as it was located outside the region with significant LA variant convergence. Variant or variant haplotype fragments were amplified in African American heterozygous individuals carrying the variants of interest located in the middle of the fragments (primer sequences in Table S3 in supporting information) and cloned into the enhancer reporter vector pGL4.24 (E8421, Promega). Subsequently, a library of 20 bp barcodes was cloned into the 3′UTR of the vector library. This final library was used for the transfection experiments. The variant (haplotype) was linked to a unique barcode using the methodology published in Trizzino et al.³⁹

The PCR-based MPRA results were supplemented by single variant/haplotype reporter assays using the Dual-Glo Luciferase System (Promega Corporation) for confirmation or when MPRA assays failed to generate sufficient data for analyses. For each region, fragments of both haplotypes were amplified using the same primers as the PCR-based MPRA design (Tables S2 and S3) and cloned separately into the pGL4.24 plasmid to test for enhancer activity.

In the second assay, we used shorter, synthetically made probes (probe-based MPRA). This allows for high-throughput inclusion of variants and single variant assessment but is more limited in the inclusion of adjacent variants because the probe size is restricted to ≈200 bp. This approach was used to identify regulatory variants as part of a larger design, including several AD loci, totaling ≈24,000 variants. For each significant LA variant, 180 bp of reference sequence (90 bp on either side of variant) was extracted for the “ref” allele fragment. An alternate (“alt”) allele fragment was created by replacing just the position of interest with the alternate allele. The same variants as used for the PCR-based MPRA design were included in this approach, except for the five indel variants, which were excluded for technical reasons in the synthesis. Using these sequences, DNA probes were synthesized (Twist Biosciences) after which 20 bp barcodes were attached to these fragments by PCR using a barcoded primer library. Combined fragments were cloned into the pGL4 plasmid backbone and sequenced for variant-to-barcode link determination, as previously described by Tewhey et al.⁴⁰ Once validated, the reporter gene with a minimal promoter was cloned in between the fragments and barcodes.

We performed these reporter assays in immortalized AD-relevant cell lines; neuronal (SH-SY5Y), microglia (HMC3), and astrocyte (U-118) cell lines. All were obtained from ATCC (CRL-2266, CRL-3304, HTB-15) and cultured according to manufacturer’s instructions.

3.1.2 |. MPRA sequencing protocols

For the PCR-based MPRA, we performed 2 × 100 bp paired-end sequencing on the Illumina HiSeq3000- for the determination of barcode to fragment link in the tagmented library. Sequencing was performed at the Center for Genome Technology (CGT) at the John P. Hussman Institute for Human Genomics (HIHG). Using the subassembly method described in Trizzino et al.,³⁹ paired reads were grouped per barcode in read 1 after which sequences of read 2 were used to reassemble the larger fragment and determine variant allele presence. We generated >143 million paired-end reads for both the HindIII or BglII induced libraries of fragment-to-barcode ligated vectors, which totaled >147,000 unique barcodes in either sample. Only barcodes unique across both HindIII and BglII samples and unique for European or African haplotype were used in the further analyses; these amounted to ≈51,000 unique barcodes. We generated on average 19 million, 56 million, and 28 million single-end reads for both replicates of RNA expressed barcodes in microglia, astrocytes, and neurons, respectively.

For the probe-based MPRA, we performed 2 × 150 bp sequencing on the Illumina NovaSeq at the CGT allowing for complete readthrough of the total assembled amplicon harboring the fragment with the variant allele of interest and barcode. We generated close to 6 million paired-end reads of the library without reporter gene, in which 3.5 million unique barcodes were detected across ≈24,000 total included variants, including the 51 non-indel APOE subTAD variants. We generated on average 118 million, 74 million, 94 million, and 164 million single-end reads for all three replicates of DNA and RNA expressed barcodes in neurons, microglia, and astrocytes, respectively. Barcodes matching multiple fragments or opposite alleles in the same fragment within either set-up were excluded from further analyses.

Practically, the one-step incorporation of fragment and barcode to the vector in the probe-based MPRA proved to be significantly more efficient (percentage uniquely linked barcodes versus total barcodes) at identifying unique barcodes than the two-step incorporation in the PCR-based MPRA.

3.1.3 |. Reporter assay statistical analyses

For the MPRA analyses, we performed the exact binomial test to find whether the proportion of the two alleles in the expressed material (RNA-Alt, RNA-Ref) is significantly different from the expected ratio (observed in MPRA library; DNA-Alt, DNA-Ref).³⁴ We implemented analysis using in-house R scripts. Raw data can be found in Table S1. For the PCR-based MPRA, we required at least one out of two replicates to survive multiple testing, with the other meeting P < .05 for confirmation. For the probe-based MPRA, we required at least one out of three replicates to survive multiple testing, with a second replicate also surviving multiple testing, or both replicates meeting P < .05, for confirmation. For the single luciferase assays, results from three independent experiments were evaluated by t test. A P-value of < .05 was considered significant in these analyses.

3.1.4 |. Capture C data

We performed high resolution (i.e., using the 4 bp-cutter DpnII, as opposed to the more commonly used 6 bp-cutter HindIII) genome-wide promoter-focused Capture C analyses on HMC3 cells, primary human astrocytes (NHA, Lonza cat#CC-2565), and healthy control induced pluripotent stem cell (iPSC)-derived NPCs and neuronal cells, following the previously described protocol.³⁵ NPC and neurons were derived from two iPSC lines from control individuals (CHOPWT10 and CHOPWT14) as described in Su et al.⁴¹ Libraries were generated in triplicate for each cell type and donor and paired-end sequenced on the Illumina Novaseq 6000 platform (51 bp read length) at the Center of Spatial and Functional Genomics at Children’s Hospital of Philadelphia (CHOP). Reads were preprocessed using the HiCUP pipeline (v0.5.9⁴²), with the bowtie aligner and hg19 as the reference genome. Significant interactions at 4-DpnII fragment resolution were called using CHiCAGO.³⁶

The chromatin interactions observed in the promoter Capture C experiments represent a summary of interaction across a pool of cells at any given time point, allowing for short-term differences within a cell population to still be picked up, though larger temporal effects in cell development may be missed.

In silico annotation

For each variant included in the probe-based MPRA design, we evaluated their presence in ChIP-seq transcription factor binding sites (TFBS) from ENCODE,⁴³ in putative enhancer or promoter regions (based on RoadMap Epigenome data⁴⁴) and in predicted TFBS (using the RegulomeDB JASPAR query⁴⁵ and MotifbreakR⁴⁶). Although ENCODE has ChIP-seq data for many transcription factors in many cell types, data on transcription factors in brain cell types is very limited; neural cell (SMC3, CTCF, EP300, MXI1, RAD21), astrocytes (CTCF), SH-SY5Y (GATA2, GATA3), and NPC (CTCF, EZH2). PrimaryHMM results from RoadMap Epigenome project of brain regions (hippocampus, dorsolateral prefrontal cortex, singular and angulate gyrus, substantia nigra, and anterior caudate), as well as adrenal gland and liver cells were evaluated. We used the MotifbreakR HOmo sapiens COmprehensive MOdel COllection (HOCOMOCO⁴⁷) for transcription factor binding site identification with a P-value significance threshold of P < 5 × 10^–5. Further, we evaluated expressive quantitative trait locus status of the variants in brain or non-brain tissues using GTEx. Last, PsychENCODE H3K27ac peak analysis was used to identify putative enhancer regions based on prefrontal cortex tissue.⁴⁸

3.2 |. Results

Detailed results can be found in Table S1 (MPRA) and Figure S2 (promoter Capture C).

3.2.1 |. Outcomes PCR-based MPRA

We identified 56 LA variants with significant different frequency on European/Japanese versus African APOE ε4 haplotypes, for which we designed 39 amplicons of ≈850 bp in the PCR-based MPRA, which included 11 amplicons with two or more significant LA. Of the 39 projected amplicons in the PCR-based MPRA, three were excluded as we were unable to optimize amplification of an 850 bp fragment in that region centering around the variant(s) of interest due to repetitive elements. After library creation and analyses to identify barcodes uniquely linked the European/Japanese or African haplotypes, we detected 51,000 unique barcodes originally identified in the DNA plasmid library. Of these, we observed ≈17,000, ≈15,800, and ≈14,500 in RNA of microglia, astrocytes, and neurons to be used in the differential expression analyses.

Fragments 2 and 3 showed consistent low incorporation in the library (evidenced by low read numbers in the library of fragment-to-barcode ligated vectors), indicating Gibson assembly for these two fragments did not perform optimally. Of the remaining 34 fragments, we were able to obtain data on unique barcodes for 24 fragments, though 7 of these (15, 16, 23, 31, 32, 36, 38) had low total read counts for the barcodes in the DNA library or expressed RNA samples in the final analyses. For those with sufficient data (Figure 3B/Table 1 overview/Table S1), we did not identify a significant allele difference for 13 fragments (1, 4, 6, 14, 18, 20, 21, 22, 24, 27, 33, 35, and 39), but observed (nominally) significant differences in both replicates for fragments 13, 17 in astrocytes, fragments 10, 13, 28 in microglia. For both fragment 10 and 13, the haplotype with alleles more common on the European ε4 background induce higher expression, whereas for fragments 17 and 28 the alleles more common on the African ε4 background display higher expression.

3.2.2 |. Outcomes luciferase reporter gene analyses

Of 19 fragments tested in all three cell types (Figure 3C/Table 1 overview/Table S1), we observed very high enhancer activity values compared to the base vector with minimal promotor for fragments 2, 28 (all cell types); 3, 5, 11, 15, 25, 39 (microglia); 7 (neurons); 23 (astrocytes); and 37 (astrocytes, neurons). We detected significant differences (P < .05) in enhancer activity driven by opposite haplotypes for fragments 16 in astrocytes; fragments 8, 13 in microglia; and fragments 2, 3, 37 in neurons with > 25% activity level difference. Lower fold difference at P < .05 were detected for fragment 7, 13 in astrocytes; 26, 33 in microglia; and 11 in neurons. Higher enhancer activity of alleles more common on European APOE ε4 was observed for fragments 2, 3, 7, 13, 26, and 33, whereas more common alleles on African APOE ε4 drove higher activity for fragments 2, 8, 11, 16, and 37.

Based on ENCODE ChIP-Seq data indicating many transcription factor binding sites, besides POL2- in fragment 9 overlaying TOMM40’s promoter, we evaluated the effect of haplotypes in this fragment in the context of promoter activity using pGL4.10 as base vector. We observed a significant higher expression in microglia compared to neuronal or astrocyte cell lines overall as expected and a two2fold higher expression level of the haplotype more common on European APOE ε4 (driven by rs71352238 C) than the African APOE ε4 background in all three cell types.

3.2.3 |. Outcomes probe-based MPRA analyses

Of the 56 variants in the region of interest, we included all non-indel variants (N = 51) in the probe-based MPRA design. After library creation, barcode linking and detection in all extracted RNA samples, we identified three variants with insufficient coverage in DNA or expression libraries of one or both alleles. Additionally, we observed 2081 barcodes that were uniquely linked to one allele of one variant in the APOE region and were detected in all cell experiments in the differential expression analyses; to be used in the subsequent analyses. We observed significant differences for variants rs34278513 (included in larger fragment 1), rs3865427 (fragment 2), rs71352237 (fragment 3), rs283808 (fragment 5), rs12972970 (fragment 6), rs34342646 (fragment 7), rs71352238 (fragment 9), rs34404554 (fragment 11), rs449647 (fragment 23), rs75627662 (fragment 25) in astrocytes, and rs34342646 (fragment 7), and rs59007384 (fragment 13) in microglia (Figure 3D). Alleles more common on either APOE ε4 haplotype show higher activity; European/Japanese rs34278513 (fragment 1), rs3865427 (fragment 2), rs71352237 (fragment 3), rs59007384 (fragment 13), rs75627662 (fragment 25), and rs12721046 (fragment 31) versus African rs283808 (fragment 5), rs12972970 (fragment 6), rs34342646 (fragment 7), rs71352238 (fragment 9), rs34404554 (fragment 11), and rs449647 (fragment 23).

APOE promoter variants previously reported to influence overall APOE expression were not significantly different in frequency between European or Japanese and African APOE ε4 haplotypes (rs405509/rs440446)⁴ or did not consistently affect enhancer activity in the current analyses (rs449647; PCR-based MPRA fragment 23 or probe-based MPRA).

Research in Context.

1. Systematic review: The mechanisms by which apolipoprotein E (APOE) ε4 risk is decreased in African ancestry carriers are not understood but may be due to variants with different frequency on African versus European/Japanese local ancestry affecting gene regulation. We reviewed literature regarding the regulatory elements influencing APOE expression.

2. Interpretation: In this reporter gene assay and promoter Capture C analysis study, the TOMM40 intron 2–3 region showed higher enhancer activity for European/Japanese versus African APOE ε4 haplotypes as well as chromatin interaction with APOE in microglia and astrocytes, supporting a role of this region in the observed differential APOE ε4 risk.

3. Future directions: Cellular investigation of the relationship between the regulatory elements in the TOMM40 intronic region and APOE ε4 expression in Alzheimer’s disease and ancestry-specific genomic context will be critical to expand upon this work.