Extracting multi-way chromatin contacts from Hi-C data

Abstract

There is a growing realization that multi-way chromatin contacts formed in chromosome structures are fundamental units of gene regulation. However, due to the paucity and complexity of such contacts, it is challenging to detect and identify them using experiments. Based on an assumption that chromosome structures can be mapped onto a network of Gaussian polymer, here we derive analytic expressions for n-body contact probabilities (n > 2) among chromatin loci based on pairwise genomic contact frequencies available in Hi-C, and show that multi-way contact probability maps can in principle be extracted from Hi-C. The three-body (triplet) contact probabilities, calculated from our theory, are in good correlation with those from measurements including Tri-C, MC-4C and SPRITE. Maps of multi-way chromatin contacts calculated from our analytic expressions can not only complement experimental measurements, but also can offer better understanding of the related issues, such as cell-line dependent assemblies of multiple genes and enhancers to chromatin hubs, competition between long-range and short-range multi-way contacts, and condensates of multiple CTCF anchors.Author summaryThe importance of DNA looping is often mentioned as the initiation step of gene expression. However, there are growing evidences that ‘chromatin hubs’ comprised of multiple genes and enhancers play vital roles in gene expressions and regulations. Currently a number of experimental techniques to detect and identify multi-way chromosome interactions are available; yet detection of such multi-body interactions is statistically challenging. This study proposes a method to predict multi-way chromatin contacts from pair-wise contact frequencies available in Hi-C dataset. Since chromosomes are made of polymer chains, the pairwise contact probabilities are not entirely independent from each other, but certain types of correlations are present reflecting the underlying chromosome structure. We extract these correlations hidden in Hi-C dataset by leveraging theoretical argument based on polymer physics.

Physical models for chromosome organization to predict multi-contact statistics

“…We repeated our analysis for these experimental data using the three-point contact frequency prediction recently proposed by Liu et al [19]. Their method is based on inferring a set of spring constants from Hi-C data, and is hence computationally more complex than the independent link or independent loop approximations, which calculate a set of three-point contact frequencies with optimal N 3 scaling.…”

“…chromosome conformation changes during the cell cycle). Recent studies have proposed a range of statistical approaches to predict contact conformation frequencies based on pairwise contact frequencies [14, 18, 8, 19]. However, these prediction schemes still lack theoretical motivation, and their performance has not been thoroughly compared or tested.…”

Dissecting the co-segregation probability from genome architecture mapping

“…Given that clinical samples are often limited in number and given in a sectioned form, GAM can be more practical than other methods when studying disease-related genome reorganization. While not easily accessible in Hi-C, a number of important properties of 3D genome, such as higher-order chromatin contact (16)(17)(18)(19)(20)(21), lamin and nuclear body association (22)(23)(24)(25)(26), can be measured by leveraging cryoFISH-combined GAM. From an ensemble of cells, Hi-C measures the cross-linking frequencies between two genomic loci; GAM measures the co-segregation frequencies in a nuclear profile; FISH measures the inter-locus spatial distances.…”

Deciphering Multi-way Interactions in the Human Genome

“…Building a more complete set of principles requires an understanding of multi-way, rather than only pairwise interactions. Recent methods for analyzing Hi-C data have inferred multi-way interactions from pairwise interactions [4, 5, 6, 7, 8, 9], though with the limitation of some ambiguity inherent to artificial assembly of pairwise ligation junctions or machine learning-based predictions. A recent extension of Hi-C preserves multi-way interactions and uses sequencing of long reads (e.g.…”

“…Multi-way contacts have been identified using targeted 3C-based methods [4, 5, 6], through inference from pairwise contacts [7], and on occasion using classical Hi-C [8]. Ligation-free approaches, such as GAM and SPRITE, have recently enabled large scale capture of multi-way interactions [9, 10, 11], though comparisons of different methods find under- and over-representation of higher order contacts in the absence of proximity ligation [12, 13].…”