An Atlas of Promoter Chromatin Modifications and HiChIP Regulatory Interactions in Human Subcutaneous Adipose-Derived Stem Cells

Abstract

The genome of human adipose-derived stem cells (ADSCs) from abdominal and gluteofemoral adipose tissue depots are maintained in depot-specific stable epigenetic conformations that influence cell-autonomous gene expression patterns and drive unique depot-specific functions. The traditional approach to explore tissue-specific transcriptional regulation has been to correlate differential gene expression to the nearest-neighbor linear-distance regulatory region defined by associated chromatin features including open chromatin status, histone modifications, and DNA methylation. This has provided important information; nonetheless, the approach is limited because of the known organization of eukaryotic chromatin into a topologically constrained three-dimensional network. This network positions distal regulatory elements in spatial proximity with gene promoters which are not predictable based on linear genomic distance. In this work, we capture long-range chromatin interactions using HiChIP to identify remote genomic regions that influence the differential regulation of depot-specific genes in ADSCs isolated from different adipose depots. By integrating these data with RNA-seq results and histone modifications identified by ChIP-seq, we uncovered distal regulatory elements that influence depot-specific gene expression in ADSCs. Interestingly, a subset of the HiChIP-defined chromatin loops also provide previously unknown connections between waist-to-hip ratio GWAS variants with genes that are known to significantly influence ADSC differentiation and adipocyte function.

Keywords: 3D organization; adipose tissue; adipose-derived stem cell; chromatin; epigenome; transcriptome.

Figure 1

Sample acquisition and study design. (A) Principal component analysis (PCA) plot of the ten subjects used to isolate the ABD and GF-ADSCs based on clinical parameters. The first two principal components (PC) are plotted and colored according to body shape. PCA was performed using all clinical data collected during the clinical study. (B) Overview of the experimental workflow. Subcutaneous adipose tissue biopsies were performed on five apple-shaped and five pear-shaped subjects. From each biopsy, the stroma vascular fraction was isolated and the ADSCs were cultured in media supplemented with serum and growth factors. Cells were harvested, their chromatin was isolated and used for RNA isolation, ChIP, ATAC, and HiChIP assays, followed by sequencing.

Figure 2

Depot-enriched gene expression and chromatin modification analysis of human ADSCs according to body shape. (A) Heat map showing the differentially expressed genes between ABD and GF-ADSCs in apple and pear-shaped subjects based on RNA-seq. The DEGs were grouped into clusters according to their level of expression in apple and pear samples. DESeq2 analysis, FDR < 0.01, FC > 0.75 Genes potentially involved in adipose tissue expansion are cited. (B) Dot plot showing the significant pathways of the DEGs in each cluster. Only the pathways with p < 0.05 are represented. (C) Legend showing the ChromHMM annotated states, with their emission values for individual chromatin marks. (D) Visualization of selected chromatin states (2, 3, 5, 6, 9, 10) around the TSS (±5 kbp) of DEGs per depot and body shape groups (rows) within gene clusters (columns). Arrows highlight when the states are visually different between ABD and GF. (E) Visualization of Genic enhancer chromatin state (state#1) around the TSS (±5 kbp) of DEGs per groups (colored) within gene clusters. Arrows highlight when the state is visually different between the ABD and GF samples.

Figure 3

Association between depot-enriched expression and depot-enriched chromatin marks at the TSS (±2 kbp) in apple samples. Volcano plots show for each gene and each histone mark studied the average fold change of the ChIP-seq signal between ABD and GF-ADSCs at the TSS. Data are represented by cluster of DEGs (rows). Negative fold changes (green) indicate the ChIP-seq signal is significantly enriched in the GF samples, while positive fold changes (orange) indicate the ChIP-seq signal is significantly enriched in the ABD samples.

Figure 4

Mapping epigenomic landscapes in ABD and GF-ADSCs. (A) Principal component analysis (PCA) plot of normalized in-loop H3K27ac HiChIP read counts. The first two principal components (PC) are plotted and colored according to body shape. (B) Dot plot showing the correlation of read densities between ABD and GF-ADSCs in apple subjects. Differential loops are colored in yellow (ABD-enriched) and green (GF-enriched). The non-significant loops are represented in gray. p-value < 0.05 logFC > 1.75. (C) Density plot showing the correlation between differential looping (x-axis) and differential H3K27ac (y-axis) at loop anchors. The H3K27ac signal was binned into 12 groups based on the magnitude of difference in H3K27ac. Data were plotted for the apple subjects. Similar observations were made for the pear subjects. (D) Genome browser visualization of SKAP2-HOX locus (left) and TBX15 locus (right) in ABD (yellow) and in GF (green) samples. Data were derived from apple subjects. Similar observations were made with data derived from pear subjects. From top to bottom: H3K27ac HiChIP interaction matrices, domainogram of insulation score, CTCF, ATAC-seq, H3K4me3, H3K27ac, H3K4me2, RNAPII, H3K9me3, H3K27me3, ChromHMM states, H3K27ac loops, and gene annotation. Color coding for ChromHMM plots is the same as Figure 2C.

Figure 5

Integration of loop anchors and GWAS-SNPs associated with WHR. (A) Permutation test showing the overlap between loop anchors and SNPs. Green lane shows the observed overlap (n = 417) and the gray histogram shows the expected distribution of overlaps by shuffling the SNP positions 2500 times. Dotted line indicates the mean expected overlap, which was used to calculate significance at p-value < 0.05 (red lane). (B) Venn diagram showing number of overlapping SNPs with loop anchors, grouped by enriched loops in the ABD (yellow) or GF (green) samples and common loops between the ABD and GF samples (grey). (C) Boxplots showing the level of expression of genes annotated to loop anchors overlapping with WHR-SNP in ABD and GF-ADSCs. Paired Wilcoxon test * p < 0.05 ** p < 0.01 (D,E) Genome browser visualization of SKAP2-HOXA locus (D) and PPARG locus (E) in ABD (yellow) and in GF (green) samples. From top to bottom: H3K27ac loops, ChromHMM states, gene annotation. The zoom in windows of HOXA locus (D) and PPARG last exons (E) show H3K27ac in the ABD (yellow) and GF (green) samples, the loop anchors (colored in yellow when belonging to the ABD-enriched loop), and WHR-SNPs. Color coding for ChromHMM plots is the same as Figure 2C.