Hierarchical Markov Random Field model captures spatial dependency in gene expression, demonstrating regulation via the 3D genome

Abstract

AO_SCPLOWBSTRACTC_SCPLOWHiC technology has revealed many details about the eukaryotic genomes complex 3D architecture. It has been shown that the genome is separated into organizational structures which are associated with gene expression. However, to the best of our knowledge, no studies have quantitatively measured the level of gene expression in the context of the 3D genome. Here we present a novel model that integrates data from RNA-seq and HiC experiments, and determines how much of the variation in gene expression can be accounted for by the genes spatial locations. We used Poisson hierarchical Markov Random Field (PhiMRF), to estimate the level of spatial dependency among protein-coding genes in two different human cell lines. The inference of PhiMRF follows a Bayesian framework, and we introduce the Spatial Interaction Estimate (SIE) to measure the strength of spatial dependency in gene expression. We find that the quantitative expression of genes in some chromosomes show meaningful positive intra-chromosomal spatial dependency. Interestingly, the spatial dependency is much stronger than the dependency based on linear gene neighborhoods, suggesting that 3D chromosome structures such as chromatin loops and Topologically Associating Domains (TADs) are strongly associated with gene expression levels. In some chromosomes the spatial dependency in gene expression is only detectable when the spatial neighborhoods are confined within TADs, suggesting TAD boundaries serve as insulating barriers for spatial gene regulation in the genome. We also report high inter-chromosomal spatial correlations in the majority of chromosome pairs, as well as the whole genome. Some functional groups of genes show strong spatial dependency in gene expression as well, providing new insights into the regulation mechanisms of these molecular functions. This study both confirms and quantifies widespread spatial correlation in gene expression. We propose that, with the growing influx of HiC data complementing gene expression data, the use of spatial dependence should be an integral part of the toolkit in the computational analysis of the relationship between chromosome structure and gene expression.