Neuronal and glial 3D chromatin architecture informs the cellular etiology of brain disorders

Most common variants associated with brain disorders lie in non-coding regions and are thought to influence traits by modulating gene regulation. Deciphering regulatory architecture in the human brain is challenging due to its cellular heterogeneity, as neurons and glia have distinct expression and chromatin accessibility profiles. Prior Hi-C studies in iPSC-derived neurons and astrocytes and promoter capture maps in postnatal human cortex offered initial regulatory insights but were limited by in vitro models or partial cell-type coverage. This study aims to generate cell-type-resolved 3D chromatin architecture from purified human brain neurons and glia, integrate it with enhancer and transcriptomic data, and apply these cell-type-specific regulatory maps to refine the cellular etiology and mechanisms of AD, SCZ, and BD.

The paper highlights that previous integrative functional genomics work mapped regulatory relationships in developing and adult human brain but was confounded by cell-type heterogeneity. Hi-C and promoter capture Hi-C in iPSC-derived neurons and astrocytes provided early interaction maps, and promoter interaction profiles were inferred in neurons, oligodendrocytes, and microglia from postnatal cortex. Single-cell RNA-seq studies established extensive neuronal and glial diversity, and open chromatin atlases underscored cell-type-specific accessibility. However, comprehensive, genome-wide, cell-type-resolved chromatin conformation data from adult human brain was lacking, motivating the present work.

- Human tissue and nuclei: Dorsolateral prefrontal cortex (DLPFC) from neurotypical donors was processed. Nuclei were extracted from frozen tissue and sorted by fluorescence-activated nuclear sorting (FANS) into neurons (NeuN+) and non-neuronal glia (NeuN−) using anti-NeuN and DAPI. - Hi-C library generation: Cross-linked nuclei were HindIII-digested, ends filled with biotin-dCTP, ligated in nucleus, sheared (300–600 bp), biotinylated fragments captured, Illumina adapters ligated, PCR-amplified, and sequenced (Illumina 50 bp paired-end). Reads were processed with hiclib, normalized by ICE, and binned at 10/40/100 kb for loops/TADs/compartments. - Reproducibility and dataset comparison: HiCRep computed SCC to assess replicate similarity and compare NeuN+ and NeuN− to other brain Hi-C datasets (adult/fetal cortex, iPSC neurons/astrocytes). - 3D features: A/B compartments called with HiCExplorer at 100 kb using PCA correlated to gene density. TADs called via directionality index and HMM at 40 kb. FIREs and super-FIREs detected with FIREcaller at 40 kb; differential and common FIREs defined using FIRE score thresholds; FIREs linked to genes via promoter overlap (−2 kb/+1 kb of TSS). - Loop calling: Promoter-anchored interactions identified by fitting distance-stratified interaction frequencies to Weibull distributions; significant interactions FDR < 0.01; Fit-Hi-C2 also applied. Overlaps with TADs assessed. - Epigenomic integration: H3K27ac ChIP-seq from NeuN+/NeuN− and purified Glu and MGE-derived GABA neurons reprocessed (Bowtie2, MACS2, DiffBind) to define differential peaks. ChromHMM states used to annotate interacting regions. TF motif enrichment at distal, loop-connected enhancers assessed with GimmeMotifs. - Gene assignment: Cell-type-specific H3K27ac peaks were linked to genes via NeuN+ or NeuN− loops to build regulatory networks; Glu/GABA peaks were integrated with NeuN+ loops to infer neuronal subtype regulatory links. - Disease analyses: - AD epigenetic dysregulation: AD-associated hyper/hypo H3K27ac peaks from bulk tissue were overlapped with NeuN+/NeuN− differential peaks and linked to genes via corresponding NeuN-specific loops. GO, cellular expression, and module enrichment assessed. - GWAS enrichment: LDSC quantified SNP-heritability enrichment in NeuN+/NeuN− and Glu/GABA enhancers for AD, SCZ, BD. - H-MAGMA: Generated NeuN−, Glu, and GABA SNP-to-gene maps (promoter/exonic direct assignment; distal via loops). Ran H-MAGMA to obtain gene-level associations and pathway/cell-type enrichments. - Transcriptomics and expression analyses: NeuN+/NeuN− nuclear RNA-seq generated and analyzed (FastQC, Cutadapt, HISAT2, StringTie, DESeq2). Single-cell RNA-seq datasets used for expression-weighted cell-type enrichment (EWCE) and subtype expression profiling. Co-expression modules from prior studies used for enrichment tests (Fisher’s exact).

- Global 3D architecture and reproducibility: - Hi-C maps from NeuN+ and NeuN− nuclei were highly reproducible (SCC NeuN+ 0.95–0.97; NeuN− 0.86–0.91) and clustered by cell type. - NeuN− chromatin structure was more similar to adult brain than fetal brain and distinct from iPSC astrocytes, consistent with oligodendrocyte enrichment; NeuN+ resembled adult, fetal, and iPSC-derived neurons. - Extensive compartment switching between cell types corresponded with cell-type-specific gene expression. - FIREs and super-FIREs: - Identified 3966 NeuN+ and 3967 NeuN− FIREs; 1248 common FIREs. - 287 differential FIREs (145 NeuN+, 142 NeuN−) showed strong overlap with corresponding differential H3K27ac peaks and aligned with neuronal functions (e.g., GRIN2B) or oligodendrocyte programs (e.g., OLIG1/OLIG2). - Super-FIREs were highly cell-type-specific: only 9 shared; 253 NeuN+ and 157 NeuN−. >95% overlapped differential H3K27ac; all overlapped promoters; associated with neuron synaptic/ion channel genes (NeuN+) or myelination/cell adhesion (NeuN−). - Promoter-anchored loops and enhancer-promoter wiring: - Detected 187,674 promoter-based interactions in NeuN+ and 167,551 in NeuN−; ~75% within TADs; ~37% enhancer–promoter, ~23% promoter–promoter. - ~50% of interactions span ≥320 kb; promoters often contacted multiple enhancers; gene expression increased with the number of interacting enhancers (up to ≥10). - ChromHMM showed promoters interact with active and bivalent states, indicating poised regulation. - TF motif enrichment: NeuN+ enhancers enriched for neuronal TFs (ZBTB18, SMARCC1, TBR1, NEUROD2); NeuN− enhancers enriched for SOX2/3/4/6/9 and IRFs (IRF4/7/8/9) related to glial immune pathways. - Linking cell-type H3K27ac to loops assigned 10,167 NeuN+ peaks to 7828 genes and 11,242 NeuN− peaks to 8851 genes. Examples: HOMER1 (NeuN+ loops/peaks, higher NeuN+ expression); SOX10 (NeuN− loops/peaks, higher NeuN− expression). NeuN+ assignments enriched for synaptic/axonal functions and developmental synaptic modules; NeuN− for actin/motility; expression matched neuron vs glia. - Neuronal subtype mapping: - Glu-specific peaks (45,911) linked to 6234 genes; GABA-specific peaks (32,169) to 4342 genes via NeuN+ loops. Assignments showed subtype specificity (e.g., Glu genes in Ex1/Ex7; GABA genes in parvalbumin In6). Locus examples: GRIK4 (Glu); GAD1 (GABA). - AD epigenetic dysregulation is cell-type-specific: - AD hyperacetylated peaks are largely active in NeuN−; AD hypoacetylated peaks active in NeuN+ in controls. - Using loops, linked AD hypoacetylation (NeuN+) to 460 genes and hyperacetylation (NeuN−) to 676 genes. Examples: CACNG3 linked to NeuN+ hypo peak; EHD1 linked to NeuN− hyper peak. - NeuN+ hypo genes enriched for synaptic functions and expressed in neurons; NeuN− hyper genes enriched for catalytic/glycoprotein binding and expressed in oligodendrocytes/astrocytes/endothelium. - Co-expression module alignment: NeuN− hyper genes enriched in astrocytic modules upregulated in AD (T-M8, T-M14); NeuN+ hypo genes enriched in neuronal modules downregulated in AD (T-M1, T-M16). - AD genetic risk converges on microglia: - LDSC showed AD SNP-heritability enriched in NeuN− enhancers (p = 8.41×10⁻⁹; NeuN+ p = 0.205). - NeuN− H-MAGMA identified 181 AD risk genes (FDR < 0.05), including BIN1 whose promoter contacts AD GWS loci in NeuN− only. - AD risk genes enriched for Aβ pathways, lipoprotein assembly, and immune processes; higher expression postnatally and in microglia; enriched among AD microglial DEGs and a microglial module upregulated in AD (T-M3). - Refined cellular etiology of SCZ and BD: - LDSC: SCZ and BD heritability enriched in NeuN+ enhancers; both also enriched in Glu and GABA enhancers. - Glu/GABA H-MAGMA yielded 753 Glu-SCZ and 624 GABA-SCZ genes; 143 Glu-BD and 101 GABA-BD genes. Glu-SCZ implicated synaptic organization/ion channels/projections; GABA-SCZ dendrite/axon/hormonal response. Glu-BD enriched for neurogenesis/cell adhesion/synapses; GABA-BD transcriptional regulation/NMDA activity. - Subtype expression revealed shared enrichment in parvalbumin interneurons (In6) and corticothalamic projection neurons (Ex7) across disorders, with divergence in upper-layer Ex1 neurons for BD and Ex2 for SCZ.

Cell-type-resolved 3D chromatin maps from human NeuN+ neurons and NeuN− glia reveal that compartments, FIREs, and loops are highly cell-type-specific and tightly linked to gene expression programs. Integrating these maps with enhancer landscapes enables accurate assignment of distal regulatory elements to target genes and deconvolution of disease-associated epigenetic changes. In AD, neuronal hypoacetylation and glial hyperacetylation correspond to downregulated neuronal modules and upregulated astrocytic programs, respectively, while genetic risk localizes predominantly to microglial regulatory architecture, underscoring distinct cellular mechanisms (genetic risk in microglia versus epigenetic alterations in astrocytes/oligodendrocytes). For SCZ and BD, neuronal enhancer networks from Glu and GABA neurons refine risk gene cellular substrates, identifying convergence in parvalbumin interneurons and corticothalamic projection neurons, and divergence between upper-layer excitatory neurons in BD and deeper-layer neurons in SCZ. These results support a model in which cell-type-specific regulatory architecture constrains and informs the cellular origins and mechanisms of diverse brain disorders.

This work provides high-resolution, cell-type-specific chromosome conformation maps for adult human neurons and glia, integrated with enhancer and transcriptomic data to build regulatory networks. Applying these maps to AD, SCZ, and BD reveals distinct and shared cellular etiologies: AD genetic risk converges on microglia, whereas AD epigenetic dysregulation highlights neuronal and oligodendrocyte/astrocyte programs; SCZ and BD risks map to both excitatory and inhibitory neuronal circuits with subtype-specific distinctions. These resources improve functional interpretation of non-coding variation and disease-associated epigenetic changes. Future directions include generating neuronal subtype-specific Hi-C maps (e.g., Glu and GABA), dissecting whether AD epigenetic alterations reflect cellular composition versus functional state changes, and causal validation of enhancer–gene links using CRISPR-based perturbations in relevant cell types.

- Lack of neuronal subtype-specific Hi-C (e.g., direct Glu/GABA Hi-C) limits resolution of subtype interactions; subtype analyses rely on integrating subtype H3K27ac with NeuN+ loops. - AD epigenetic patterns from bulk tissue may reflect cellular composition changes (neuronal loss, glial expansion) versus regulatory state changes; disentangling these requires further study. - Findings are derived from postmortem adult DLPFC; developmental and regional differences are not directly assessed. - Assignments of risk variants to genes, while improved by Hi-C, remain correlative and require experimental validation (e.g., CRISPR perturbations).