Systematic identification of protein combinations mediating chromatin looping

Abstract

Chromatin looping plays a pivotal role in gene expression and other biological processes through bringing distal regulatory elements into spatial proximity. The formation of chromatin loops is mainly mediated by DNA-binding proteins (DBPs) that bind to the interacting sites and form complexes in three-dimensional (3D) space. Previously, identification of DBP cooperation has been limited to those binding to neighbouring regions in the proximal linear genome (1D cooperation). Here we present the first study that integrates protein ChIP-seq and Hi-C data to systematically identify both the 1D- and 3D-cooperation between DBPs. We develop a new network model that allows identification of cooperation between multiple DBPs and reveals cell-type-specific and -independent regulations. Using this framework, we retrieve many known and previously unknown 3D-cooperations between DBPs in chromosomal loops that may be a key factor in influencing the 3D organization of chromatin.

Figure 1. The performance of the GGM is consistently better than ARACNE.

Each plot shows the average curve from 10 independent simulations. (a) ROC curve for samples generated from random networks. For each simulation 500 (left) or 1,000 (right) samples were generated from a network of 50 (left) or 100 (right) nodes. (b) ROC curve for samples generated from yeast sub-networks. For each simulation 500 (left) or 1,000 (right) samples were generated from a network of 50 (left) or 100 (right) nodes. (c) Workflow of the DBPnet pipeline.

Figure 2. Constructing the DBP cooperation network in GM12878.

(a) DBP cooperation network in GM12878, with network hubs (EP300, EBF1, CREB1) being highlighted. (b) A significant portion of DBP cooperation is supported by evidence of direct protein–protein interactions. (c) The majority of DBP cooperation is a mixture of 1D and 3D cooperation. (d) An example of 1D-cooperation. (e) An example of 3D-cooperation. (f) An example of mixed cooperation. (g) Disease-associated genotype variations are enriched in 1D-dominant (n=67), 3D-dominant (n=71) and 1D–3D cooperative (n=346) sites.

Figure 3. Network analysis reveals functions of DBP modules in GM12878 and K562.

(a) Communities in the DBP cooperation network and their functions. (b) Top DBP cliques. (c) An example of DBP cliques. (d) An example of K562-specific DBP cliques and the enriched GO terms of their binding sites. (e) An example of GM12878-specific DBP cliques and enriched GO terms of their binding sites. (f) Top conserved DBP modules in K562 and GM12878.

Figure 4. CEBPB–PML–STAT5A cooperates with different DBPs in GM12878 and K562.

(a) DBP-binding profile (left) and enriched de novo motifs (right) in 2468 CEBPB–PML–STAT5A-binding sites in GM12878. (b) DBP-binding profile (left) and enriched de novo motifs (right) in 2620 CEBPB–PML–STAT5A-binding sites in K562. (c) Enriched spacing between de novo motifs found in CEBPB–PML–STAT5A sites in K562 and GM12878.