Microglial chromatin interactions are enriched for genetic risk variants associated with neurodegenerative disorders

To determine whether disease risk variants for neurodegenerative disorders are associated with genes linked to distal gene regulatory regions, we used proximity ligation-assisted chromatin immunoprecipitation-seq (PLAC-seq) data generated from human cortical neurons, microglia and oligodendrocytes [40]. Histone H3 lysine 27 acetylation (H3K27ac) is a histone modification enriched at active gene promoters and distal enhancers, whereas histone H3 lysine 4 trimethylation (H3K4me3) is enriched at promoters only [54,55]. By integrating H3K4me3-anchored PLAC-seq-defined chromatin loops with ATAC-seq, H3K27ac chromatin immunoprecipitation (ChIP)-seq and H3K4me3 ChIP-seq from the same cell types [40], we classified chromatin interactions as either: i) promoter-to-enhancer; ii) promoter-to-promoter; iii) promoter-to-ATAC; iv) promoter-to-promoter/enhancer; v) promoter-to-other; vi) H3K4me3-to-H3K4me3; vii) H3K4me3-to-other; and viii) other interactions. Promoter-to-enhancer loops were the most common classification of chromatin interactions for each cell type, representing 29.4%, 39.1% and 38.2% of interactions in microglia, neurons, and oligodendrocytes, respectively (Fig 1a). The next most abundant classifications were chromatin interactions that occurred at promoters-to-other or H3K4me3-to-other (Fig 1a). Promoters are known to interact with more than one enhancer. For microglia, neurons and oligodendrocytes, most promoters interacted with more than one enhancer and for enhancer-to-promoter interactions, most enhancers interacted with a single promoter (Fig 1b). The average distance of these enhancer-to-promoter interactions was 200 kb for microglia (range: 3,443 – 1,003,377 bp; mean: 206,304 bp), 250 kb for neurons (2,657 – 1,008,329 bp; mean: 254,016 bp) and 150 kb for oligodendrocytes (range: 2,974 – 999,211 bp; mean: 159,230 bp) (Fig 1c). Overall, H3K4me3-anchored PLAC-seq chromatin loops in microglia, neurons and oligodendrocytes predominantly identified promoters that were linked to multiple distal enhancers.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Microglia enhancer-to-promoter interactions were enriched for disease-risk variants across multiple neurodegenerative conditions.

A) Doughnut plots of classifications of PLAC-seq interactions identified in human microglia, neurons and oligodendrocytes [40] with the total number of interactions shown in the center. ‘Promoters’, PLAC-seq bins that overlap a H3K4me3 and H3K27ac peak within 2,000 bp of a transcriptional start site (TSS). ‘Enhancers’, PLAC-seq bins that overlap H3K27ac peaks distal to the TSS. ‘H3K4me3’, PLAC-seq bins that overlap H3K4me3 peaks distal to TSS. ‘ATAC’, PLAC-seq bins that overlap chromatin accessible regions devoid of H3K4me3 and H3K27ac. B) Percent distribution of the number of enhancers interacting with individual promoters (top plot) and the number of promoters interacting with individual enhancers (bottom plot). C) Distribution plot of the proportion of distances between midpoints of promoters and midpoints of enhancers that interact based on chromatin interaction PLAC-seq data. D) Heatmap of partitioned heritability sLDSC regression analysis of PLAC-seq-derived functional annotations: (i) total PLAC-seq bins, (ii) promoter & enhancer PLAC-seq bins, (iii) promoter PLAC-seq bins and (iv) enhancer PLAC-seq bins for microglia, neurons and oligodendrocytes in AD [29,30], PD [27] (excluding 23andMe), MS [33], ALS [31], and schizophrenia [32]. Shown are LDSC enrichment p-values with Benjamini–Hochberg FDR correction for the number of diseases and cell types (-log10(q)). Disease enrichment was considered insignificant if the coefficient z-score was negative and assigned a 0.0 -log10(p) score. OLs, oligodendrocytes; SCZ, schizophrenia.

https://doi.org/10.1371/journal.pgen.1011407.g001

To examine whether disease heritability was enriched at brain cell type chromatin interactions, we used stratified linkage disequilibrium score (sLDSC) regression analysis. Cell type disease enrichment by sLDSC regression was assessed using chromatin interactions defined as (i) all PLAC-seq bins irrespective of functional genomic annotations (total PLAC-seq bins), (ii) PLAC-seq bins subset to both active gene promoters (H3K4me3 + H3K27ac) and distal enhancers (H3K27ac only) (promoter & enhancer PLAC-seq bins), (iii) PLAC-seq bins subset to active gene promoters (promoter PLAC-seq bins) and (iv) PLAC-seq bins subset to distal enhancers (enhancer PLAC-seq bins). Cell type disease enrichment was assessed using summary statistics from two complementary AD GWAS; one study was based exclusively on clinical diagnosis [30], while the second included by-proxy cases [29]. GWAS summary statistics were analyzed for three additional neurodegenerative conditions, PD, MS and ALS [27,31,33] and for a neurodevelopmental condition, schizophrenia [32].

Microglia PLAC-seq bins showed enrichment for AD risk variants over other cell types, with a greater enrichment of AD risk variants found at enhancer PLAC-seq bins compared to promoter PLAC-seq bins (Figs 1d and S1). These findings corroborate observations that AD GWAS variants are enriched at microglia enhancers compared to microglia promoters defined using histone modifications [40,56]. The microglia enhancer PLAC-seq bins are likely physically linked to gene promoters and therefore functionally relevant. An enrichment of disease risk variants at microglia PLAC-seq bins was also observed for PD and MS, although there was no clear preference for either promoter or enhancer interacting regions for these disorders (Figs 1d and S1). No enrichment for ALS risk variants was identified at PLAC-seq bins for microglia, neurons or oligodendrocytes (Figs 1d and S1). In contrast, for schizophrenia, heritability showed a strong enrichment of disease risk at PLAC-seq bins identified in neurons and oligodendrocytes (Figs 1d and S1). The heritability enrichment for schizophrenia was stronger in neurons than oligodendrocytes (promoter & enhancer PLAC-seq bins; sLDSC; neurons -log10(q)=15; oligodendrocytes, -log10(q)=4.0) (Fig 1d). This supports previous findings showing that schizophrenia GWAS variants were enriched at neuronal promoters and enhancers using annotations defined by histone modifications [40,44]. Overall, chromatin-interacting regions in microglia show a broad enrichment for disease heritability across multiple neurodegenerative disorders.

Microglial chromatin interactions identify disease risk genes across multiple neurodegenerative conditions

Promoter-to-enhancer interactions link distal gene regulatory regions, such as enhancers, to active gene promoters and can be used to infer disease-risk genes for noncoding GWAS risk variants. H-MAGMA was used to identify disease-risk genes in microglia, neurons and oligodendrocytes for AD, PD, MS, ALS and schizophrenia by incorporating PLAC-seq interactomes for the corresponding cell types (S2 Fig). In all the neurodegenerative GWAS that we assessed, the highest number of risk genes were identified in microglia compared to neurons and oligodendrocytes (Fig 2a and S1 Table). In contrast, for schizophrenia, the highest number of risk genes were identified in neurons (Fig 2a and S1 Table). Schizophrenia had the highest number of H-MAGMA risk genes across all cell types compared to neurodegenerative disorders. The number of identified risk genes is partially influenced by the number of associated risk variants reported in the corresponding GWAS. Schizophrenia had the most variants with a GWAS p-value below 5×10⁻⁸ (19,898 SNPs), more than five times the number reported for the second-highest GWAS analysed in this study (3,555 SNPs in MS).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Microglial disease risk genes were identified for distal GWAS variants using chromatin loops.

A) The number of disease risk genes identified in microglia, neurons and oligodendrocytes using H-MAGMA and GWAS for AD, PD (excluding 23andMe), MS, ALS, and schizophrenia. Gene-to-SNP associations were assigned for SNPs that were located within the promoter or exon of a gene, and within enhancers that were linked to genes through PLAC-seq interactions. B) The number of disease risk genes identified using the sampled down loops for microglia, neurons and oligodendrocytes with H-MAGMA for AD, PD (excluding 23andMe), MS, ALS, and schizophrenia. To account for differences in chromatin interactions between cell types, the number of enhancer-to-promoter interactions was randomly sampled down to 60,000 loops 10 times. Dunn’s test (non-parametric) between cell types within each group: AD (Jansen 2019): microglia-neurons (**), microglia-oligodendrocytes (****), neurons-oligodendrocytes (ns); AD (Kunkle 2019): microglia-neurons (*), microglia-oligodendrocytes (****), neurons-oligodendrocytes (*); PD: microglia-neurons (ns), microglia-oligodendrocytes (*****), neurons-oligodendrocytes (**); MS: microglia-neurons (****), microglia-oligodendrocytes (**), neurons-oligodendrocytes (ns); ALS: microglia-neurons (****), microglia-oligodendrocytes (**), neurons-oligodendrocytes (ns); schizophrenia: microglia-neurons (*), microglia-oligodendrocytes (*), neurons-oligodendrocytes (****). C) UCSC browser visualization of the AD INPP5D locus and the PD MED12L locus. Top track, GWAS SNPs (line indicates GWAS p-value –log10(5e-8) =7.301); middle tracks H3K27ac and H3K4me3 ChIP-seq; bottom track, PLAC-seq interactomes [40]. Yellow, promoters of risk genes identified only by H-MAGMA; pink, promoters of risk genes identified by H-MAGMA and MAGMA. D) EWCE analysis identified cell type enrichment of H-MAGMA disease risk genes from Fig 2A using single-cell RNA-seq from human brain cortical tissue from Tsartsalis et al. (2024) [58]. E) EWCE analysis identified microglia subtype enrichment of microglia H-MAGMA disease risk genes using single-cell RNA-seq of human microglia subtypes from Sun et al. (2023) [59]. EWCE heatmap colour scale represents the number of standard deviations from the mean level of expression found in the target gene lists relative to the bootstrapped mean. SCZ, schizophrenia; OLs, oligodendrocytes.; EWCE p-values *p <0.05, **p<5e-8, ***p=0.

https://doi.org/10.1371/journal.pgen.1011407.g002

The number of PLAC-seq chromatin interactions were higher in microglia compared to other cell types (microglia, 108802; neurons, 93290; oligodendrocytes, 61895; Fig 1a), which may partially explain the increased number of microglia disease risk genes identified across neurodegenerative conditions. To account for the differing number of chromatin interactions identified between the three cell types, the PLAC-seq data was randomly downsampled to 60,000 chromatin interactions per cell type. This was followed by H-MAGMA analysis, which was repeated for 10 iterations (Fig 2b). H-MAGMA analysis using the 60,000 downsampled PLAC-seq chromatin interactions maintained a similar distribution of disease-risk genes across the three cell types (Fig 2b). Importantly, when the number of chromatin interactions was the same for each cell type, the number of disease-risk genes identified remained highest in microglia for AD, PD, MS and ALS (Fig 2b).

Standard gene mapping approaches such as MAGMA rely on the proximity of GWAS risk variants to genes [53]. The number of genes identified by assigning variants located within 10 kb of a gene using MAGMA was 2–3-fold higher than the number of H-MAGMA microglia risk genes identified across the neurodegenerative conditions (S3 Fig and S3 Table). The higher number of MAGMA risk genes likely reflects the lack of cell type specificity that is provided by incorporating microglia enhancer-to-promoter interactomes. However, many H-MAGMA risk genes were not identified by MAGMA (AD Jansen, 62; AD Kunkle, 32; PD, 30; MS, 75; ALS 68 genes) (S3 Fig and S4 and S5 Tables), including the autophagy-related gene ATG16L1 at the INPP5D AD locus and purinergic receptor gene P2RY13 located at the PD MED12L locus (Fig 2c). These more distal H-MAGMA risk genes were not identified using conventional proximity-based strategies for annotating variants to genes. In summary, H-MAGMA disease risk genes were overrepresented in microglia for neurodegenerative disorders compared to neurons for schizophrenia, which is consistent with the cell type distribution of disease heritability identified using sLDSC regression analysis (Fig 1d).

GWAS risk variants may be differentially enriched at chromatin interaction contact sites at enhancers (PLAC-seq bins at intergenic and intronic regions) compared to promoters (PLAC-seq bins at promoters and exonic regions). To determine GWAS risk enrichment across these gene regulatory classifications, H-MAGMA was repeated using PLAC-seq bins subset to enhancers or promoters. For AD, PD and MS, the number of disease-risk genes identified using microglial enhancer contact sites (100, 64, 179 genes respectively) was comparable to the number of disease-risk genes identified using microglial promoter contact sites (111, 62, 175 genes respectively) (S4 Fig). In ALS, more disease-risk genes were identified using microglia promoter contact sites (124 genes) compared to enhancer contact sites (47 genes) (S4 Fig). This suggests that promoters may play a more crucial role in the genetic risk associated with ALS, in contrast to the significance of enhancers for GWAS risk in other neurodegenerative conditions. Lastly, for schizophrenia, the highest number of disease-risk genes were identified at neuronal promoter contact sites compared to enhancers (S4 Fig).

Disease-risk variants often colocalise with gene regulatory regions that are highly cell-type specific, thereby conferring cell-type-associated genetic susceptibility [36,57]. However, the downstream genes associated with these regulatory regions may be expressed exclusively in the disease-associated cell type or across multiple cell types. Expression Weighted Celltype Enrichment (EWCE) analysis was used to determine the cell type expression of the GWAS risk genes identified by H-MAGMA by incorporating single-cell gene expression data from human cortical brain tissue [58]. EWCE analysis revealed that the expression of microglia H-MAGMA-risk genes for all diseases was enriched predominantly in microglia, with some enrichment in lymphocytes for MS and schizophrenia (Fig 2d). In contrast, H-MAGMA-risk genes identified in neurons and oligodendrocytes across the neurodegenerative conditions generally did not exhibit a cell type enrichment in gene expression. Only oligodendrocyte H-MAGMA risk genes identified for PD and MS showed an enrichment in oligodendrocyte signature genes (Fig 2d). Disease risk genes identified by H-MAGMA across all three cell types for schizophrenia were characterized by matching cell type-specific gene expression (Fig 2d). Human microglia exhibit heterogeneous transcriptional profiles, a subset of which have been associated with disease conditions [59]. EWCE analysis indicated that microglial AD H-MAGMA risk genes were associated with inflammatory, phagocytic and cycling transcriptional profiles (Fig 2e). In contrast, MS microglial H-MAGMA risk genes were associated with ribosome biogenesis and phagocytic transcriptional profiles and ALS with phagocytic only (Fig 2e).

Microglial genetic-susceptibility genes are associated with disease-specific pathways

Genetic heritability estimates using sLDSC and the identification of putative GWAS risk genes identified using H-MAGMA highlight the importance of microglia in the genetic susceptibility of neurodegenerative conditions. This may suggest shared dysregulated microglial processes across these disorders. However, an intersection of GWAS-risk genes identified using H-MAGMA for microglia showed a minimal overlap between the different diseases (Fig 3a), including for a subset of the top significant H-MAGMA risk genes (filtered on H-MAGMA p-value; AD, PD, MS, ALS p<5e-8 and schizophrenia p<5e-12) (Fig 3b). Similarly, there was a minimal overlap across diseases for GWAS risk genes identified for neurons and oligodendrocytes (Fig 3a and 3b). While most risk genes were unique to each disorder, some genes were shared across two or more conditions. For example, the major histocompatibility complex (MHC) was identified as a disease-risk locus in MS and schizophrenia (Fig 3b). Disease-risk genes that overlapped across PD, ALS and schizophrenia were KANSL1-AS1 (microglia and oligodendrocytes) and KANSL1, ARHGAP27, and PLEKHM1 (microglia). Interestingly, KANSL1 and ARHGAP27 were identified as comorbid genes for PD and ALS [60]. The microglial H-MAGMA-risk genes BAG6, NEU1, PRRC2A, PSMB8, PSMB8-AS1 and PSMB9 were associated with MS, ALS and schizophrenia. PSMB8-AS1 was also identified as a microglial risk gene for AD. These findings indicate that microglia are an important cell type associated with genetic susceptibility across multiple neurodegenerative disorders. However, the microglial genes that are impacted by genetic risk are mostly disease specific.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Microglia disease-risk genes impacted disease-specific pathways.

A) UpSet visualization of unique and intersecting H-MAGMA disease-risk gene numbers between AD, PD (excluding 23andMe), MS, ALS and schizophrenia for each cell type. B) Heatmaps of H-MAGMA identified risk genes based on enhancer-to-promoter interactions from PLAC-seq data for AD, PD (excluding 23andMe), MS, ALS (H-MAGMA p<5e-8) and schizophrenia (H-MAGMA p<5e-12) for microglia, neurons and oligodendrocytes. Shown are H-MAGMA FDR corrected p-values (-log10(q)). C) Gene ontology pathway analysis of microglial risk genes identified by H-MAGMA for AD, PD (excluding 23andMe), MS, ALS, and schizophrenia; shown are top 20 pathways. SCZ, schizophrenia.

https://doi.org/10.1371/journal.pgen.1011407.g003

We next assessed specific cellular and biological pathways associated with microglia H-MAGMA-risk genes for each disorder using gene ontology (GO) analysis. GO pathways linked to H-MAGMA-risk genes were mostly unique for each neurodegenerative condition (Fig 3c). This is consistent with the observation that most disease-risk genes were unique to each GWAS (Fig 3a and 3b). The top GO pathways associated with microglial AD-risk genes included lipoproteins, amyloid processing and endocytosis (Fig 3c and S2 Table) compared to neuronal and oligodendrocyte AD-risk genes which were associated with amyloid and tau protein catabolic processes (S5 and S6 Figs). PD microglial-risk genes were associated with the endolysosomal/autolysosomal pathways, synaptic vesicles and epigenetic signaling (Figs 3c and S2). Whereas substantia nigra gliosis, epigenetic signaling and synaptic vesicle pathways were evident in neuronal PD-risk genes, reinforcing the vulnerability of the midbrain in PD (S5 Fig). Both microglia and oligodendrocyte MS risk-genes were associated with MHC protein complexes, autoimmunity, and antigen presentation and processing (Figs 3c and S6 and S2 Table). Risk genes assigned to the MHC Class II complex were also associated with AD and PD, as well as MS (Fig 3b). ALS exhibited associations with vacuoles and kinases, while also sharing pathways with PD related to lysosomes and autophagosomes (Fig 3c and S2 Table). Microglial-associated GO pathways for schizophrenia H-MAGMA-risk genes were distinct from the neurodegenerative disorders and primarily included epigenetic and gene regulatory pathways (Fig 3c and S2 Table). Neuronal GWAS schizophrenia risk genes were primarily implicated in synaptic processes (S5 Fig). The only common microglial-associated GO pathways between neurodegenerative disorders were autolysosome, cytoplasmic vesicle, secondary lysosome, vacuole and intracellular vesicle (S7 Fig). However, the combination of genes contributing to these pathways varied between diseases, with only HLA−DRB1/HLA−DRA locus shared for the autolysosomal and secondary lysosome pathways in ALS and AD (S7 Fig). Collectively, pathway analysis confirmed the observation from gene set overlaps, indicating that microglial risk genes and associated biological pathways are mostly disease specific.

PLAC-seq datasets

PLAC-seq data for human microglia, neurons and oligodendrocytes was pre-processed by Nott et al., 2019 [40]. PLAC-seq data was generated using epilepsy resections of the frontal, parietal and temporal cortex of seven individuals aged 5 months to 17 years of Caucasian, Hispanic and Pacific Islander ethnicities (6 males/1 female). PLAC-seq chromatin interactions were anchored to active gene promoters by immunoprecipitation of histone H3 lysine 4 trimethylation (H3K4me3), which is a histone modification enriched at active gene promoters [54,55]. Chromatin interactions were 5 kb resolution and were anchored to promoters using chromatin immunoprecipitation of the histone modification H3K4me3 [40].

Classification of PLAC-seq interactions

PLAC-seq chromatin interactions were classified as i) promoter-to-enhancer; ii) promoter-to-promoter; iii) promoter-to-ATAC; iv) promoter-to-promoter/enhancer; v) promoter-to-other; vi) H3K4me3-to-H3K4me3; vii) H3K4me3-to-other; and viii) other interactions for microglia, neurons and oligodendrocytes. ‘Promoter’ were classified as PLAC-seq bins that overlapped with H3K4me3 and H3K27ac regions within 2,000 bp of the nearest TSS and ‘enhancer’ were classified as PLAC-seq bins that overlapped H3K27ac regions distal to TSS as defined by Nott 2019 [40]; promoter/enhancer were classified as PLAC-seq bins that overlapped both promoter and enhancer regions; ‘H3K4me3’ were PLAC-seq bins that overlapped H3K4me3 regions distal from TSS; ‘ATAC’ were PLAC-seq bins that overlapped chromatin accessibility regions that were devoid of H3K4me3 and H3K27ac; ‘other’ were PLAC-seq bins that did not overlap with H3K4me3, H3K27ac or chromatin accessibility regions [40]. To identify the number of enhancers interacting with each promoter and number of promoters interacting with each enhancer, cell type PLAC-seq bins were overlapped with active promoter and active enhancer regions.

GWAS datasets

The following GWAS summary statistics were used in this study were downloaded from EBI’s GWAS catalogue (https://www.ebi.ac.uk/gwas/) and were of European ancestry:

AD (Jansen 2019) (GCST007320): n= 71,880 cases and 383,378 controls [29];

AD (Kunkle 2019) (GCST007511): n = 21,982 cases and 41,944 controls, Stage 1 [30];

PD (Nalls 2019) (GCST009325): n = 33,674 cases and 449,056 controls (excluding 23andMe) [27];

MS (Andlauer 2016) (GCST003566): n = 4,888 cases and 10,395 controls [33];

ALS (van Rheenen 2021) (GCST90027164): n = 27,205 cases and 110,881 controls [31];

schizophrenia (Trubetskoy 2022) (GCST90128471): n = 53,386 cases and 77,258 controls [32].

Quality control of GWAS summary statistics

GWAS summary statistics were standardised and underwent quality control steps before running H-MAGMA. GWAS summary statistics were filtered using format_sumstats function in “MungeSumstats” package (version 1.6.0, available on Bioconductor) in R (version 4.2.1) [103]. Quality control steps included SE > 0, SNPs that have N<5 standard deviation above mean, indels were kept, chromosomes X, Y and mitochondrial chromosome were removed and multiallelic SNPs were kept. GWAS summary statistics had the following imputation quality scores: AD (Jansen 2019) >0.91; AD (Kunkle 2019) >0.4; PD >0.8; MS ≥0.8; ALS >0.95; schizophrenia >0.9. The pre-filtered imputation quality scores provided by the original GWAS were used for AD, PD, MS and ALS. For schizophrenia, we used an imputation quality score that matched the AD GWAS (Jansen 2019).

H-MAGMA

Annotating genetic variants to target genes was performed using H-MAGMA workflow [52,98]. H-MAGMA workflow incorporates 1) generating input files which provide the background profile of gene-SNP associations based on chromatin interaction data and 2) using MAGMA software to run gene-level analysis (S2 Fig). To generate cell type-specific promoter-enhancer profiles, 1) chromatin interaction data from PLAC-seq for microglia, neurons and oligodendrocytes, and 2) reference data for SNPs (22665064 SNPs) from Phase 3 of 1,000 Genomes for European ancestry were used (genome Build 37) (https://cncr.nl/research/magma/). Chromatin interaction data from PLAC-seq with FDR-corrected p-value cut-off of 0.01 were filtered to interactions with promoters in at least one end by overlapping with cell-type promoter regions to only investigate active chromatin interactions [40]. Gene-SNP association profiles were created by assigning exonic and promoter SNPs directly to target genes based on genomic location using a gene model Gencode v41 [104]. Promoters were defined as 1.5kb upstream and 500 bp downstream of the TSS of each gene isoform based on prior publications [105,106]. Intronic and intergenic SNPs were assigned to genes based on cell-type chromatin interactions (see PLAC-seq datasets) with promoters and exons [52]. Intronic and intergenic SNPs were filtered to enhancer SNPs by overlapping with cell-type enhancer regions [40]. To investigate disease enrichment in active chromatin interactions, significant cell-type specific chromatin interactions with FDR-corrected p-value cut-off of 0.01 were filtered to interactions with promoters in at least one end by overlapping cell-type promoter regions [40]. Filtered chromatin interactions were overlapped with Gencode 41 exon and promoter coordinates to identify exon-based and promoter-based interactions [52,98].

To determine whether enhancer or promoter/exon SNPs were driving the disease enrichment of genes, H-MAGMA input files were generated either with promoter/exon SNPs or enhancer SNPs only. Exonic and promoter SNPs were directly assigned to target genes based on genomic location using a gene model Gencode v41 (https://www.gencodegenes.org/human/release_41lift37.html) [104]. Promoters were defined as 1.5kb upstream and 500 bp downstream of the TSS of each gene isoform. Intronic and intergenic SNPs were assigned to ‘ALL’ Gencode v41 genes based on cell-type chromatin interactions (see PLAC-seq datasets) with promoters and exons [52]. Intronic and intergenic SNPs were filtered to enhancer SNPs by overlapping with cell-type enhancer regions [40]. After running gene-level analysis using MAGMA software (see MAGMA section below) outputted genes with an FDR-corrected p-value <0.05 were selected for downstream analysis (https://ctg.cncr.nl/software/magma).

MAGMA

MAGMA analysis pipeline was used to run the cell type-specific gene level association with a disease based on gene-SNP profiles based on chromatin interaction data [53]. The association was established using the default “SNP-wise mean” gene analysis model, which is a test of mean SNP association using the sum of squared SNP Z-statistics as a test statistic. In brief, SNP-level p-values from GWAS summary statistics were aggregated into gene-level p-values and a reference data set (1,000 Genomes European panel) was used to account for linkage disequilibrium between SNPs. Since some of the GWAS summary statistics used in the study are SNP meta-analysis results, individual sample sizes per SNP may have significant variation and may affect the gene test-statistic results. Therefore, if available, individual sample sizes per SNP were used (ncol modifier in –pval parameter in MAGMA). The analysis was run as follows: magma --bfile g1000_eur --pval <GWAS summary statistics> use=SNP,P ncol=NSUM --gene-annot <Input annotation file> --debug set-spar=tmp_snps_used --out <Output file>

Standard gene mapping with MAGMA was applied to assign risk genes for each disease without incorporating epigenetic or chromatin interaction datasets, with variants assigned to genes based on proximity within a 10 kb window. The analysis was run as follows: 1) magma --annotate window=10,10 --snp-loc g1000_eur.bed --gene-loc gencode41_gene_coord.txt --out <Output Annotation>; 2) magma --bfile g1000_eur --pval <GWAS summary statistics> use=SNP,P ncol=NSUM --gene-annot <Input annotation file> --debug set-spar=tmp_snps_used --out <Output file>

Partitioned heritability (sLDSC regression)

Partitioned heritability using sLDSC regression analysis was used to identify brain cell type annotations that were enriched for heritability of AD, PD (excluding 23andMe), MS, ALS and schizophrenia (LDSC version 1.0.1) by functional category while controlling for 97 annotation categories of the full baseline model (model version 2.2) [99]. Cell type annotations per functional category were run jointly. Functional categories included cell type 1) total PLAC-seq bins, 2) promoter and enhancer PLAC-seq bins, 3) promoters PLAC-seq bins, and 4) enhancer PLAC-seq for microglia, neurons and oligodendrocytes. Baseline model LD scores, standard regression weights, and allele frequencies that were used were built from 1000 Genomes Phase 3 for European population. The enrichment P-values were FDR multiple testing corrected for the number of GWAS studies and number of cell types using Benjamini-Hochberg correction method. Disease enrichment was considered insignificant if the coefficient z-score was negative. Cell type annotations for all the functional categories were created using plink format.bed/.bim/.fam filesets of 1000 Genomes Phase 3 for European population and LD scores were computed based on a 1 centiMorgan (cM) window. Since the annotations were built on top of the baseline model, 1000 Genomes Phase 3 was used together with the HapMap3 SNPs. A quality control step of GWAS summary statistics was performed before LDSC analysis using munge_sumstats.py where SNPs had INFO <= 0.9, MAF <= 0.01 and N < 32290, were out-of-bounds p-values, strand-ambiguous, with duplicated IDs and alleles did not match Hap-Map SNPs. To prevent bias from variable imputation quality both between and within each GWAS study, all the GWAS SNPs were filtered to HapMap3 SNPs, as these SNPs are well imputed in most studies.

EWCE

Expression weighted cell type enrichment (EWCE) analysis (v1.6.0) was used to identify cell type-specificity of the H-MAGMA outputted risk genes for each disease type [107]. Single-cell RNA-seq data from brain cortical tissue from Tsartsalis et al. (2014) [58] was used to generate probability distribution associated with cell type-specific H-MAGMA outputted risk genes having an average level of expression within a cell type of the brain. Single-cell RNA-seq data for human microglial subtypes from Sun et al. (2023) [59] was used to generate probability distribution associated with microglia H-MAGMA outputted risk genes having an average level of expression within the subtypes. Significant cell type-specificity was determined based on the p-value <0.05.

GO analysis

Gene set enrichment analysis was performed on the list of H-MAGMA outputted significant risk genes identified per cell type to identify biological pathways at risk in each cell type for each disease. The R package “gprofiler2” (v0.2.1) was used for gene set enrichment, which contains data sources including Gene Ontology (GO), KEGG, Reactome, WikiPathways, miRTarBase, TRANSFAC, Human Protein Atlas, protein complexes from CORUM and Human Phenotype Ontology [108]. Risk genes inputted into the analysis were filtered based on the FDR adjusted p-value<0.05 and were ordered based on the Z-score generated by the H-MAGMA. Identified pathways were also FDR corrected using p-value <0.05. For visualization, if pathways contained the same set of genes, the one with the highest FDR corrected p-value was included in the bar plots.