Extensive long-range polycomb interactions and weak compartmentalization are hallmarks of human neuronal 3D genome

Abstract

Chromatin architecture regulates gene expression and shapes cellular identity, particularly in neuronal cells. Specifically, polycomb group (PcG) proteins enable establishment and maintenance of neuronal cell type by reorganizing chromatin into repressive domains that limit the expression of fate-determining genes and sustain distinct gene expression patterns in neurons. Here, we map the 3D genome architecture in neuronal and non-neuronal cells isolated from the Wernicke's area of four human brains and comprehensively analyze neuron-specific aspects of chromatin organization. We find that genome segregation into active and inactive compartments is greatly reduced in neurons compared to other brain cells. Furthermore, neuronal Hi-C maps reveal strong long-range interactions, forming a specific network of PcG-mediated contacts in neurons that is nearly absent in other brain cells. These interacting loci contain developmental transcription factors with repressed expression in neurons and other mature brain cells. But only in neurons, they are rich in bivalent promoters occupied by H3K4me3 histone modification together with H3K27me3, which points to a possible functional role of PcG contacts in neurons. Importantly, other layers of chromatin organization also exhibit a distinct structure in neurons, characterized by an increase in short-range interactions and a decrease in long-range ones.

Graphical Abstract

Figure 1.

Global chromatin organization in NeuN(+) and NeuN(−) cells. (A) Anatomical localization of the analyzed brain region, experimental procedure and the design of this study. Hi-C experiments were performed in NeuN(+) and NeuN(−) cells isolated using FANS from the Wernicke's speech area of four human individuals. The brain image was adapted from https://commons.wikimedia.org/wiki/File:BA22_-_lateral_view.png, attribution: Polygon data were generated by Database Center for Life Science (DBCLS), CC-BY-SA-2.1-jp (https://creativecommons.org/licenses/by-sa/2.1/jp/deed.en). (B) Principal component analysis plot based on the Insulation Score (IS) variation among all produced Hi-C maps. Colors represent NeuN(+) and NeuN(−) cells, here and in panels C, E, F, I. (C) Average interactions within all chromosomes (cis contacts, left panel) and between all pairs of chromosomes (trans contacts, right panel) calculated separately for NeuN(+) and NeuN(−) cells. Asterisks indicate Wilcoxon test P-values: ^****P < 0.00001, ns: P > 0.05. (D) Ratio of interactions within and between all chromosomes (NeuN(+)/NeuN(−)). (E) Polymer scaling plot showing average interaction frequencies at various genomic distances, as well as the first derivative of this plot demonstrating the slope. (F) Distal (>3 Mb) to local (<3 Mb) contact ratio (DLR) calculated for every 100-kb genomic region, demonstrating that chromosomes in neurons are more compact at a local scale and less compact at a large scale. Asterisks represent Wilcoxon test P-value <0.00001. (G) A fragment of the Hi-C map featuring the compartment eigenvector (PC1) for NeuN(+) (left) and NeuN(−) (right) cells. Positive PC1 values represent the A compartment, while negative values correspond to the B compartment. (H) NeuN(+) / NeuN(−) ratio of average Hi-C contact probability (observed over expected) of genomic regions arranged by the corresponding PC1 rank (saddle plot). Both saddle plots were created using NeuN(−) PC1. (I) Compartment interaction strength calculated as the highest intra-compartment interactions divided by the lowest inter-compartment interactions.

Figure 2.

Cell-type-specific differences of TADs in NeuN(+) and NeuN(−) cells. (A) Examples of differential TAD profiles at neuronal DE genes. RBFOX3, GALNT17, ATP2B1 gene locations are shown. RBFOX3 produces the neuronal nuclei (NeuN) antigen that is used as a neuronal marker in the study. NeuN is an RNA binding protein involved in the regulation of alternative splicing of pre-mRNA. GALNT17 encodes a glycosyltransferase enzyme involved in the post-translational modification of neuronal proteins. This gene has been shown to be essential for the proper development and function of the nervous system. ATP2B1 is a gene that is involved in neuronal signal transmission encoding a calcium pump that is important for the regulation of calcium levels in neurons. (B) Venn diagram of the intersection of neuronal and non-neuronal TAD borders identified based on the insulation profiles. (C) Box plot of TAD densities for neurons and non-neuronal cells. Asterisks indicate Wilcoxon test P-values: ^**** - P < 0.0001. (D) Average TAD. (E) Grouped stacked bar plots of the chromatin states coverage within TAD borders. The ratio of each state is normalized to the total coverage within the genome in the corresponding cell type. Chromatin state annotation includes six distinct chromatin states, specifically EnhA (active enhancer), TssA (active promoters), TssBiv (bivalent promoters), TssFlnk (promoter flanking region), ReprPC (Polycomb repression region), and Quies (other regions). (F) Expression profiles around the cell-type-specific and common borders. (G) Confusion tables with Fisher test P-values for the enrichment of genes upregulated in NeuN(−) and NeuN(+) cells within corresponding TAD borders. DE genes were determined as genes upregulated in the selected cell type with FC > 1.5. (H) GO terms enrichment (shown with dots sizes) among upregulated genes placed in neuronal-specific or common TAD borders. (I) Average large TADs located within lamina-associated domains (LADs).

Figure 3.

Loops in NeuN(+) and NeuN(−) cells. (A) Fragments of 5-kb resolution Hi-C maps for regions of neuronal-specific gene interactions with enhancers. Enhancer tracks are shown at the top of Hi-C maps. (B) Venn diagram of the intersection of neuronal and non-neuronal loop positions. (C) Box plot of loop length distributions. Asterisks indicate Wilcoxon test P-values: ^****P < 0.00001. (D) Average loops within four groups defined based on the increase of NeuN(+)/NeuN(−) intensity ratio. (E) EnhA and ReprPC chromatin states in anchors of four loop groups calculated relative to the average across the genome as log₂(ratio). (F) The number of upregulated genes placed within four loop groups. DE genes were determined as genes upregulated in the selected cell type with FC > 1.5. (G) The number of enhancer - promoter loops within four loop groups. (H) Box plot of gene expression grouped based on the number of interactions with enhancers (‘0’ – no interactions, ‘1–5’ and ‘over 5’ – 1–5 or more than 5 interactions). Asterisks indicate Wilcoxon test P-values: ^****P < 0.00001. Groups ‘0’, ‘1–5’ and ‘over 5’ consist of 3038, 461, 398 and 3227, 496, 295 genes in NeuN(+) and NeuN(−), respectively. (I) Ratio of genes separated by three types of interactions with enhancers within four loop groups. (J) Box plots of loop lengths mediating interactions between a gene and at least one enhancer. Asterisks indicate Wilcoxon test P-values: ^****P < 0.00001. Gr.1–2, 4 consist of 1227 loops, and Gr.3 consists of 1226 loops. (K) Linkage disequilibrium (LD) score regression P-values calculated based on GWAS studies (91–101) for NeuN(−) and NeuN(+) loops. UKBB – UK Biobank (92).

Figure 4.

Features of neuronal dots. (A) Fragment of the Hi-C map with neuronal dots present in NeuN(+) (top), but absent in NeuN(−) (bottom). (B) The average Hi-C signal (observed over expected) of neuronal dots (left) and corresponding pairs of neuronal dot loci in NeuN(−) (right). The value in the corner corresponds to the central pixel. (C) Left: contact scaling of neuronal dots (purple), approximated by a linear regression (black line). Right: contact scaling of pairs of neuronal dot loci in NeuN(−) (orange), approximated by a linear regression (black line). The gray line represents the contact scaling of the whole genome, which is the same as in Figure 1E. (D) Significant interchromosomal interactions of neuronal dot loci in NeuN(+) and NeuN(−). Significance is defined by FitHiC2, P < 0.05. (E) Statistics for the size (order) of cliques (networks) made by neuronal dots (purple bars) or neuronal dot loci in NeuN(−) (orange bars). (F) Average lamin B1 ChIP-seq read coverage profile between pairs of neuronal dot anchors. Horizontal axis is scaled so that separation between each pair of anchors is constant. Anchor locations are marked by gray dotted lines. Pairs of anchors are separated into four quantiles based on corresponding Hi-C interaction intensity (normalized by expected at given loci separation). Areas around the coloured lines - standard deviations. (G) Average H3K9me3 ChIP-seq read coverage profile at neuronal dot anchors. Areas around lines - 95% confidence intervals. (H) Expression of protein-coding genes located within anchors of neuronal dots compared with expression of housekeeping genes. ^*****P-value < 10⁻¹¹, two-sided Wilcoxon test, N = 252. (I) Expression of protein-coding genes located within anchors of neuronal dots. Genes are divided into TFs (magenta) and non-TFs (black). Grey dots – expression of all human protein-coding genes.

Figure 5.

Neuronal dots in excitatory (EN) and inhibitory (IN) neurons. (A) Average snm3C-seq signal (observed over expected) of neuronal dots in excitatory (top) and inhibitory (bottom) neuronal dot loci. Value in the corner corresponds to the central pixel. Dot annotation was taken from NeuN(+) Hi-C data. (B) Snippets of aggregated snm3C-seq contact matrices showing differential neuronal dots between excitatory and inhibitory neurons. A gene name at the bottom of each plot indicates the gene closest to the dot locus. (C) Expression of six genes that form differential neuronal dots. (D) H3K27me3 ChIP-seq signal centered at the transcription start sites (TSS) of six genes that form differential neuronal dots. Tracks are oriented so that each gene body is located to the right of the TSS. Vertical axis scale is equal for GABAergic and glutamatergic neurons.