Dendrograms generated for each chromosome are internally reordered, without disrupting the clustering structure, to match subcompartments among different chromosomes. Next, compartment domains are clustered using a divisive hierarchical clustering approach exclusively based on their interdomain contacts (3D-proximity) and ignoring their contiguity along the genome sequence (1D-proximity). First, it computes whole-chromosome contact similarities among genomic loci (Fisher’s z-transformed correlations) and identifies compartment domains by segmenting each chromosome into regions having high intraregion similarity and low inter-region similarity.
#NUMBER OF HAND COMPARTMENTS PLUS#
Our approach consists of two main steps plus an optional one. We introduce an algorithmic approach that infers a complete hierarchy of compartment domains using exclusively intrachromosomal interactions, which are more frequent than interchromosomal ones and thus alleviate the requirements on data resolution (Fig. By inferring chromatin subcompartments across multiple cell types and states, we could study repositioning of compartment domains and its association with cell phenotypes. Here, we propose an algorithm able to infer compartment domains and hierarchies of subcompartments in Hi-C contact maps with highly variable total number of reads. As a consequence, the identification of chromatin subcompartments and compartment domains remains unfeasible for the vast majority of available Hi-C datasets. However, as we will show, this approach frequently fails to correctly infer subcompartments when challenged with relatively low-resolution experiments or when patterns of interchromosomal interactions deviate significantly from the training dataset. Machine learning-based imputation of Hi-C contacts has been used to enhance data resolution and allow subcompartment inference in eight additional Hi-C experiments 18. For example, the GHMM approach was exclusively applied to the GM12878 cell line (4.9 billion read pairs). Computational inference of subcompartments has so far relied on interchromosomal contacts and, thus, it has been possible only for high-resolution experiments. Indeed, up to 6 subcompartments have been inferred by clustering interchromosomal interactions using a Gaussian Hidden Markov Model (GHMM) 2, 10 and, recently, an intermediate compartment was characterized using Hi-C and imaging techniques in colorectal tumor samples 17. It was previously proposed that the A and B dichotomy might not be sufficient to explain chromatin compartmentalization 16. However, chromatin activity encompasses multiple states 15, some of which can only be captured through a more refined subcompartmentalization 2. Importantly, A and B compartments have been shown enriched for distinct histone modifications, which are consistent with the observed transcriptional activity. Compartment domains from the same compartment preferentially interact among each other and, in Hi-C contact correlation maps, the alternance of A and B compartment domains generates a chessboard or “plaid” pattern reflecting enrichment or depletion of Hi-C interactions 4.Ĭomputational inference of A and B compartments has been performed across multiple cell types and it showed, for example, that compartments are less conserved across cell types than TADs 5 and phenotypic changes are more frequently associated with compartment repositioning of a given genomic region than with structural disruption of loops or TADs 14. Interestingly, recent experiments based on CTCF and cohesin depletion have shown that although often overlapping and sometime coincident, TADs and compartment domains are in fact distinct SEs 13. Within each chromosome, DNA regions belonging to a given compartment are defined as compartment domains. At a broad scale, chromatin segregates into two major compartments, one preferentially localized at the core of the nucleus and exhibiting high transcriptional activity (A compartment), and another localized closer to the nuclear lamina and enriched for repressed and gene-depleted chromatin (B compartment). Loop extrusion has been associated with the formation of topologically associating domains (TADs) and structural loops 8, 9, 10, whereas chromatin compartmentalization segregates the chromatin into spatial compartments 11, 12 and compartment domains 10.
The formation of chromatin SEs is driven by two major mechanisms: loop extrusion of chromatin fibers mediated by CTCF and cohesin, and chromatin compartmentalization, which segregates chromatin regions with different patterns of histone acetylation and methylation 7. Hi-C technologies have allowed to quantify and computationally model such interactions to unravel chromatin spatial organization 4, 5, 6. In interphase, the chromatin is packaged into a hierarchy of three-dimensional (3D) structural elements (SEs) emerging from interactions and insulation of distinct DNA regions 1, 2, 3.
0 kommentar(er)
