A map of open chromatin in human pancreatic islets

Abstract

Tissue-specific transcriptional regulation is central to human disease. To identify regulatory DNA active in human pancreatic islets, we profiled chromatin by formaldehyde-assisted isolation of regulatory elements coupled with high-throughput sequencing (FAIRE-seq). We identified approximately 80,000 open chromatin sites. Comparison of FAIRE-seq data from islets to that from five non-islet cell lines revealed approximately 3,300 physically linked clusters of islet-selective open chromatin sites, which typically encompassed single genes that have islet-specific expression. We mapped sequence variants to open chromatin sites and found that rs7903146, a TCF7L2 intronic variant strongly associated with type 2 diabetes, is located in islet-selective open chromatin. We found that human islet samples heterozygous for rs7903146 showed allelic imbalance in islet FAIRE signals and that the variant alters enhancer activity, indicating that genetic variation at this locus acts in cis with local chromatin and regulatory changes. These findings illuminate the tissue-specific organization of cis-regulatory elements and show that FAIRE-seq can guide the identification of regulatory variants underlying disease susceptibility.

Figure 1. FAIRE-seq in human pancreatic islets

(a) Chromatin is cross-linked using formaldehyde, sonicated, and subjected to phenol-chloroform extraction. DNA fragments recovered in the aqueous phase are then sequenced. (b) Reads obtained from sequencing were highly concordant with FAIRE signal obtained from tiling microarrays covering the ENCODE pilot project regions. Arrows indicate the direction of gene transcription.

Figure 2. Both proximal and distant FAIRE sites harbor functional regulatory elements

(a) Genes with high expression in islets (top 20%; red) have more FAIRE enrichment at promoters than genes with moderate (middle 20%; green) or low (bottom 20%; blue) expression. (b) Promoters (−750/+250 bp) bound by RNA Pol II, HNF4A or HNF1A in human islets are significantly over-represented among islet FAIRE sites (red dash indicates expected value; all bars: P<0.001). (c) Intergenic islet-selective and ubiquitous FAIRE sites that are located >2 kb from a TSS are enriched for evolutionary conserved sequences (P<0.001), predicted regulatory modules (PreMod, P<0.001), and transcription factor binding sites (conserved TFBS and MotifMap, both P<0.001). CTCF binding, however, is enriched in ubiquitous FAIRE sites only. Over half of intergenic open chromatin sites are coincident with an experimentally or computationally determined functional annotation (expected value for random sites: 27%). (d) Open chromatin is most enriched directly at sites of experimentally determined CTCF binding. (e) In contrast to ubiquitous FAIRE sites, islet-selective FAIRE sites are rarely located within 2 kb upstream of a TSS or in exon 1, and are instead located predominantly in more distal regions. Shown is the percentage of bases covered by each annotation category in islet-selective FAIRE sites (blue), ubiquitous FAIRE sites (red), and the mappable genome (gray).

Figure 3. Islet-selective FAIRE sites form Clusters of Open Regulatory Elements (COREs)

(a) FAIRE sites are highly clustered. We divided the genome in windows of varying size (x axis), and calculated a χ2 statistic to determine if the number of windows with 0, 1 or >1 FAIRE sites differed from randomly distributed sites. The highest significance was observed in ~20 kb windows. (b) Same as (a) but for islet-selective sites. (c) We defined islet-selective clusters of open chromatin regulatory elements (COREs) as three or more islet-selective FAIRE sites separated from each other by <20 kb. (d) We identified 3,348 islet-selective COREs (blue points). Fewer COREs were generated using randomized FAIRE sites (orange points), and they were smaller than in vivo COREs (e) Most islet-selective COREs were associated with a single gene. (f) RefSeq genes associated with islet-selective COREs were on average inactive in non-islet human tissues, except for brain. Asterisks indicate P<1×10⁻⁵ (one-way ANOVA). (g) Chromatin landscape of the PDX1 locus showing an extended cluster of islet-selective FAIRE sites, contrasting with a closed conformation of the adjacent gut-specific homeodomain gene CDX2. The top panel depicts the density of FAIRE-Seq reads centered on the genomic average density value, the location of moderate stringency FAIRE sites in islets (blue) or in any of the 5 non-islet cells (red), and the binding sites of the CTCF insulator protein in K562 cells. CTCF sites demarcate regions that show broadly consistent FAIRE-Seq enrichment patterns. The bottom panel shows a closer view of a portion of the PDX1 islet-selective CORE, with islet-selective open chromatin sites at previously characterized regulatory elements (Area I–III, Area IV) and in an evolutionarily conserved putative enhancer.

Figure 4. Allele-specific open chromatin and enhancer activity at the TCF7L2 locus

(a) Schematic representation of how FAIRE-seq enables the identification of human sequence variants located in islet open chromatin. From ~4 million SNPs present in dbSNP with average heterozygosity >1%, 38 SNPs associated with T2D or fasting glycemia mapped to islet open chromatin sites. The analysis was carried out with all SNPs in strong linkage disequilibrium (r²>0.8) with an FG- or T2D-associated variant, which are labeled as FG or T2D SNPs, and FAIRE-seq sites identified with a liberal threshold. (b) Among TCF7L2 variants in linkage disequilibrium with rs7903146 (r²>0.2, top panel), only rs7903146 maps to an islet-selective FAIRE site. (c) In all 9 human islet samples that were heterozygous for rs7903146, the risk allele T was more abundant than the non-risk C allele in the open chromatin fraction, in contrast to input DNA or gDNA from unrelated heterozygous individuals. (d) Allelic imbalance for open chromatin at rs7903146 was verified in independent assays using quantitative Sanger sequencing (see also Supplementary Fig. 4b). (e) The risk allele T of rs7903146 exhibits greater enhancer activity than the non-risk allele C in MIN6 cells and (f) 832/13 cells. Standard deviations represent four independent clones for each allele. Results for inserts in the reverse direction are provided in Supplementary Fig. 4. P-values were calculated by two-sided t-test.