Figures and data
Partitioned Linkage Disequilibrium Score Regression analysis for open chromatin regions of all cell types.
A. The schematic shows the different types of regions defined in our study and 3 different ways overlapping chromatin contact regions – OCRs – gene promoters define cREs.
B. Heritability enrichment by LDSC analysis for each cell type
a) Bar-plot shows the total number of OCRs identified by ATAC-seq for each cell type on bulk cells - blue, or on single cells – red; and the portion of OCRs that fall within cREs identified by incorporating Hi-C (green) or by Capture-C (orange).
b-e) 4 panels of dot-plots show heritability enrichment by LDSC analysis for each cell type, with standard error whiskers. Dots’ colors correspond to -log10(p-values), dots with white asterisks are significant p-values <0.05, and dots’ sizes corresponding to the proportion of SNP contribute to heritability. Dash line at 1, i.e., no enrichment. b)Analysis done on whole OCRs set of each cell type (whiskers colors match with bulk/single cell from bar-plot a);c)On only OCRs that overlapped with promoters (whiskers’ colors match with bulk/single cell from bar-plot a);d)On the putative cREs of each cell type (whiskers’ colors match with Hi-C/Capture-C from bar-plot a);e)On the same cREs as (c) panel with their genomic positions expanded ±500 bases on both sides (whiskers’ colors match with Hi-C/Capture-C from bar-plot a)
b) Analysis done on whole OCRs set of each cell type (whiskers colors match with bulk/single cell from bar-plot a);
c) On only OCRs that overlapped with promoters (whiskers’ colors match with bulk/single cell from bar-plot a);
d) On the putative cREs of each cell type (whiskers’ colors match with Hi-C/Capture-C from bar-plot a);
e) On the same cREs as (c) panel with their genomic positions expanded ±500 bases on both sides (whiskers’ colors match with Hi-C/Capture-C from bar-plot a)
Mapping 771 proxies to the open chromatin regions of each cell type
A. Venn diagram shows how 771 proxies mapped to the OCRs:
Blue area: 758 proxies were located within contact regions of at least one cell type regardless of chromatin state;
Red area: 417 proxies were located within contact regions marked as open by overlapping with OCR;
Yellow area: If we only considered open chromatin regions, 178 proxies were included;
Dotted green bordered area: To focus on just those variants residing within open chromatin and contacting promoter regions in any cell type, we overlapped the genomic positions of these proxies with each cell type’s cRE set, yielding 90 variants (3 from the 99% credible set) directly contacting open gene promoters (eTable 3), with 10 of which located within a promoter of one gene but contacting another different gene promoter. There were an additional 4 variants located within gene promoters but in chromatin contact with promoter(s) of nearby transcript(s) of the same gene (correspond to 3 cREs illustrations in Figure 1A).
White area: proxies that fall into neither defined region of interest. B. Bar-plot shows number of proxies, cell types and target genes mapped at each locus. C. The upSet plot shows the degree of overlap across cell types of the variants; ranked from the most common variant (red) – rs61888800 from BDNF locus, a well-known 5’ untranslated region variant of this gene that is associated with anti-depression and therapeutic response81,82 – appeared in 39 cell types, to the group of variants (grey) which appeared in only one cell type.
Profiles of 111 implicated genes by 94 proxies through cREs of each cell type
Main panel: Bubble plot show corresponding expression level (size) and number of variants (color) target each implicated gene of each cell type. Squares represent genes with variants at their promoters. Circles represent genes with variants contacted through chromatin loops. Some genes were implicated by both types, these “double implications” are represented as diamond shapes, and were identified across several cell types: two cell types (plasmacytoid dendritic cells and pre-differentiated adipocytes) for ADCY3 gene, and five for BDNF (human embryonic stem cells - hESC, differentiated human fetal osteoblast cells - hFOB_Diff, neural progenitor cells derived from induced pluripotent stem cells - NPC_iPSC, PANC-1, and NCIH716 cell lines) Genes with expression undetected in our arrays are shown as triangles.
Top panel: bar-plot shows numbers of cell types each gene was implicated within, color-coded by which systems the cell types belong to.
Right panel: bar-plot shows numbers of genes implicated by the variants with each cell type.
Colocalization of target effector genes with eQTLs
A. Venn diagram shows the overlaps between sets of genes yielded by ColocQuiaL and the variant-to-gene mapping process.
B. Circos plot of the 10 loci demonstrates the differences in the ranges of associations between the two approaches, with long-ranged chromatin contacts between obesity variants and target genes displayed as orange links and short-range eQTLs colocalizations as green links.
Two SNPs – rs35796073, and rs35142762 within the TMEM18 locus, in linkage disequilibrium with rs7579427 – were estimated with high probability (cond.PP.H4=0.78) of colocalizing with the expression of ALKAL2 gene in subcutaneous adipose tissue. These pairs of SNP-gene were also identified by our variant-to-gene mapping approach in natural killer cells, plasmacytoid dendritic cells, unstimulated PBMC naïve CD4 T cells and astrocytes.
The rs7132908 variant at the FAIM2 locus colocalized with the expression of AQP6 in thyroid tissue and with ASIC1 in prostate tissue, not only with high cond.PP.H4 but also with high individual SNP causal probability (SNP.PP.H4 > 0.95). rs7132908 was the second most consistent observation in our variant-to-gene mapping, namely across 25 different cell types (Figure 2B) and all three systems plus the other independent cell lines. The pair of rs7132908-contacting-AQP6 was observed in 15 different cell types - 8 metabolic and 4 neural cell types, and 3 independent cell lines. The pair of rs7132908-contacting-ASIC1 was observed in 11 different cell types - 8 metabolic and 2 neural cell types, and plasmacytoid dendritic cells.
The other eQTL signals that overlapped with our variant-to-gene mapping results were: BDNF at the METTL15 locus with its promoter physically contacted by rs11030197 in 4 cell types and its expression significantly colocalized (cond.PP.H4=0.82) in tibial artery; ADCY9 at its locus with its promoter physically contacted by rs2531995 in natural killer cells and its expression significantly colocalized in skin tissue (“Skin_Not_Sun_Exposed_Suprapubic”, cond.PP.H4=0.97). And ADCY3 in the C panel.
C. ColocQuiaL estimated that these SNPs highly colocalize with the expression of ADCY3 in 11 different tissues, where the overlapping with the 16 cell types is represented, color-coded by the proxies rs numbers.
