HUGIn: Hi-C Unifying Genomic Interrogator

Abstract

Motivation

High throughput chromatin conformation capture (3C) technologies, such as Hi-C and ChIA-PET, have the potential to elucidate the functional roles of non-coding variants. However, most of published genome-wide unbiased chromatin organization studies have used cultured cell lines, limiting their generalizability.

Results

We developed a web browser, HUGIn, to visualize Hi-C data generated from 21 human primary tissues and cell lines. HUGIn enables assessment of chromatin contacts both constitutive across and specific to tissue(s) and/or cell line(s) at any genomic loci, including GWAS SNPs, eQTLs and cis-regulatory elements, facilitating the understanding of both GWAS and eQTL results and functional genomics data.

Availability and implementation

HUGIn is available at http://yunliweb.its.unc.edu/HUGIn

Supplementary information

Supplementary data are available at Bioinformatics online.

1 Introduction

Elucidating the functional roles of the vast majority (>80 to 90%) of non-coding variants identified from GWAS, and subsequently finding their target gene(s), are pressing and daunting tasks. Investigators have only recently realized the power and value of high throughput chromatin conformation capture (3C) technologies, such as Hi-C (Lieberman-Aiden et al., 2009), TCC (Kalhor et al., 2011) and ChIA-PET (Fullwood et al., 2009), for potential target gene identification (Dekker et al., 2017; Pombo and Dillon, 2015). Several 3D genome browsers, including WashU Epigenome Browser (Zhou et al., 2013) and Juicebox (Durand et al., 2016), have been developed to visualize 3C-based data. However, the vast majority of genome-wide unbiased chromatin organization studies published to date have used cultured cell lines, limiting their generalizability. Understanding chromatin structure and understanding complex trait genetic architecture are each daunting challenges. Understanding both simultaneously is even more daunting, but of paramount importance to unveil genetic mechanisms underlying complex diseases (Gamazon et al., 2015; Jakobsdottir et al., 2009).

2 Materials and methods

We developed HUGIn (Hi-C Unifying Genomic Interrogator) to host our recently published compendium of Hi-C data across 14 human primary tissues and 7 cell lines (Schmitt et al., 2016) (http://yunliweb.its.unc.edu/HUGIn/). For each tissue or cell line, HUGIn visualizes the observed and expected read counts between two genomic loci, and the corresponding statistical significance, for long range chromatin interactions, in both heatmap and virtual 4C plots (detailed tutorial in Supplementary Material Section S1 and Fig. S1). Compared to existing 3D genome browsers, HUGIn has five major unique features. First, HUGIn simultaneously visualizes a compendium of Hi-C data, which is continually updated as new data become available (Supplementary Material Section S2). Second, it directly enables detection of chromosomal organizations both specific to, and constitutive across, tissue(s) and/or cell line(s). Third, HUGIn can display Hi-C data anchored at genomic locus of any size, suggesting potential target gene(s) for GWAS variants in a tissue or cell line informative manner. This feature is particularly important for evaluating a GWAS locus, which can range from a single nucleotide to a region of linkage disequilibrium extending over tens of thousands of base pairs (Smith et al., 2005). Fourth, HUGIn also hosts gene expression and a rich collection of epigenomic data, including typical and super enhancers, CTCF binding sites, frequently interacting regions (FIREs) (Schmitt et al., 2016), and several core histone modifications (Supplementary Material Section S1). Lastly, HUGIn generates a list of most likely target gene(s) for any genomic loci of interest. Such a gene list can be used for gene set enrichment analysis or prioritization for follow-up functional studies.

3 Results

Figure 1 is a virtual 4C plot in which we display a long range chromatin interaction between the type 2 diabetes (T2D)- and adiponectin-associated SNP rs6450176 with the gene FST (Civelek et al., 2017). Although the SNP is located within the ARL15 gene, it was shown to be an eQTL for FST but not ARL15 in subcutaneous adipose tissue. In addition, expression level of FST but not ARL15 is correlated with adiponectin levels (Civelek et al., 2017). More details and the virtual 4C plot for the rs6450176-FST interaction are shown in Supplementary Figure S2. Two more examples, a long range interaction between BMI-associated SNP rs9930506 and the gene IRX3 (Smemo et al., 2014), and between schizophrenia-associated SNP rs1191551 and the gene FOXG1 (Won et al., 2016), are shown in Supplementary Figures S3 and S4. We further investigated all intergenic SNPs in the NHGRI catalog of published GWAS for seven diseases including schizophrenia, leukemia, Alzheimer’s disease, autism, depression, type 1 diabetes and type 2 diabetes (Supplementary Material Section S3). We found that, across all diseases examined, >85% of HUGIn-annotated genes are not the gene closest to the GWAS SNP (Supplementary Table S1); and that HUGIn-annotated gene lists tend to provide stronger evidence for enrichment with biologically relevant terms or gene sets (Supplementary Tables S2 and S3).

Fig. 1.

Virtual 4C plot view of the interaction between the type 2 diabetes (T2D)- and adiponectin-associated SNP rs6450176 and its potential target gene FST. Virtual 4C plot of the long range chromatin interactions anchored at the 40 Kb bin (chr5: 53 280 001–53 320 000), shown as a gray bar, containing the type 2 diabetes (T2D)- and adiponectin-associated GWAS SNP rs6450176 (chr5: 53 298 025) (Civelek et al., 2017), shown as a vertical black line within the gray bar, in H1 human embryonic cell line (A), in liver tissue (B) and in spleen tissue (C). The observed and expected chromatin contact frequencies (or counts) are represented by the black and red lines, respectively. The left Y axis displays the range of chromatin contact frequency. The statistical significance (i.e., –log10(P-value)) of each long range chromatin interaction reported by Fit-Hi-C (Ay et al., 2014) is represented by the blue line, with its range listed in the right Y axis. The cell line or tissue specific FDR 5% threshold is shown as a cyan horizontal dashed line, and the more stringent Bonferroni 0.05 threshold is shown as a blue horizontal dashed line. Above each virtual 4C plot is a diagram of genes that are within the 1Mb viewing window (chr5: 52 598 001–53 598 000). The highlighted yellow bar is the gene FST (chr5: 52 776 264–52 782 304) (Color version of this figure is available at Bioinformatics online.)

4 Conclusions

With the continuing accumulation of high resolution 3C-based data in disease-relevant human tissues, we expect that HUGIn will become an increasingly valuable tool for many investigators, including biologists interested in fundamental chromosome organization across cell lines, tissue types, or developmental stages, and geneticists seeking to understand the genetic architecture and mechanism underlying complex diseases and traits.