Pattern recognition of topologically associating domains using deep learning

BACKGROUND

Recent increasing evidence indicates that three-dimensional chromosome structure plays an important role in genomic function. Topologically associating domains (TADs) are self-interacting regions that have been shown to be a chromosomal structural unit. During evolution, these are conserved based on checking synteny block cross species. Are there common TAD patterns across species or cell lines?

RESULTS

To address the above question, we propose a novel task-TAD recognition-as opposed to traditional TAD identification. Specifically, we treat Hi-C maps as images, thus re-casting TAD recognition as image pattern recognition, for which we use a convolutional neural network and a residual neural network. In addition, we propose an elegant way to generate non-TAD data for binary classification. We demonstrate deep learning performance which is quite promising, AUC > 0.80, through cross-species and cell-type validation.

CONCLUSIONS

TADs have been shown to be conserved during evolution. Interestingly, our results confirm that the TAD recognition model is practical across species, which indicates that TADs between human and mouse show common patterns from an image classification point of view. Our approach could be a new way to identify TAD variations or patterns among Hi-C maps. For example, TADs of two Hi-C maps are conserved if the two classification models are exchangeable.

Mysteries of gene regulation: Promoters are not the sole triggers of gene expression

•

TF binding peaks were widely distributed in nonpromoters, especially downstream regions of transcription termination sites.

•

Exons of non-coding regions were the prominent regions of TF binding.

•

TAD boundaries were colocalized with activating histone marks and TF binding.

•

Genes with distinct functions demonstrated substantially different behaviors in cis-regulation and epigenetic signatures.

Cis-regulatory elements of promoters are essential for gene regulation by transcription factors (TFs). However, the regulatory roles of nonpromoter regions, TFs, and epigenetic marks remain poorly understood in plants. In this study, we characterized the cis-regulatory regions of 53 TFs and 19 histone marks in 328 chromatin immunoprecipitation (ChIP-seq) datasets from Arabidopsis. The genome-wide maps indicated that both promoters and regions around the transcription termination sites of protein-coding genes recruit the most TFs. The maps also revealed a diverse of histone combinations. The analysis suggested that exons play critical roles in the regulation of non-coding genes. Additionally, comparative analysis between heat-stress-responsive and nonresponsive genes indicated that the genes with distinct functions also exhibited substantial differences in cis-regulatory regions, histone regulation, and topologically associating domain (TAD) boundary organization. By integrating multiple high-throughput sequencing datasets, this study generated regulatory models for protein-coding genes, non-coding genes, and TAD boundaries to explain the complexity of transcriptional regulation.

Comparison of the somatic TADs and lampbrush chromomere-loop complexes in transcriptionally active prophase I oocytes

In diplotene oocyte nuclei of all vertebrate species, except mammals, chromosomes lack interchromosomal contacts and chromatin is linearly compartmentalized into distinct chromomere-loop complexes forming lampbrush chromosomes. However, the mechanisms underlying the formation of chromomere-loop complexes remain unexplored. Here we aimed to compare somatic topologically associating domains (TADs), recently identified in chicken embryonic fibroblasts, with chromomere-loop complexes in lampbrush meiotic chromosomes. By measuring 3D-distances and colocalization between linear equidistantly located genomic loci, positioned within one TAD or separated by a TAD border, we confirmed the presence of predicted TADs in chicken embryonic fibroblast nuclei. Using three-colored FISH with BAC probes, we mapped equidistant genomic regions included in several sequential somatic TADs on isolated chicken lampbrush chromosomes. Eight genomic regions, each comprising two or three somatic TADs, were mapped to non-overlapping neighboring lampbrush chromatin domains - lateral loops, chromomeres, or chromomere-loop complexes. Genomic loci from the neighboring somatic TADs could localize in one lampbrush chromomere-loop complex, while genomic loci belonging to the same somatic TAD could be localized in neighboring lampbrush chromomere-loop domains. In addition, FISH-mapping of BAC probes to the nascent transcripts on the lateral loops indicates transcription of at least 17 protein-coding genes and 2 non-coding RNA genes during the lampbrush stage of chicken oogenesis, including genes involved in oocyte maturation and early embryo development.

The generation of PD-L1 and PD-L2 in cancer cells: From nuclear chromatin reorganization to extracellular presentation

The immune checkpoint blockade (ICB) targeting on PD-1/PD-L1 has shown remarkable promise in treating cancers. However, the low response rate and frequently observed severe side effects limit its broad benefits. It is partially due to less understanding of the biological regulation of PD-L1. Here, we systematically and comprehensively summarized the regulation of PD-L1 from nuclear chromatin reorganization to extracellular presentation. In PD-L1 and PD-L2 highly expressed cancer cells, a new TAD (topologically associating domain) (chr9: 5,400,000-5,600,000) around CD274 and CD273 was discovered, which includes a reported super-enhancer to drive synchronous transcription of PD-L1 and PD-L2. The re-shaped TAD allows transcription factors such as STAT3 and IRF1 recruit to PD-L1 locus in order to guide the expression of PD-L1. After transcription, the PD-L1 is tightly regulated by miRNAs and RNA-binding proteins via the long 3'UTR. At translational level, PD-L1 protein and its membrane presentation are tightly regulated by post-translational modification such as glycosylation and ubiquitination. In addition, PD-L1 can be secreted via exosome to systematically inhibit immune response. Therefore, fully dissecting the regulation of PD-L1/PD-L2 and thoroughly detecting PD-L1/PD-L2 as well as their regulatory networks will bring more insights in ICB and ICB-based combinational therapy.

Topologically associating domain boundaries that are stable across diverse cell types are evolutionarily constrained and enriched for heritability

Topologically associating domains (TADs) are fundamental units of three-dimensional (3D) nuclear organization. The regions bordering TADs-TAD boundaries-contribute to the regulation of gene expression by restricting interactions of cis-regulatory sequences to their target genes. TAD and TAD-boundary disruption have been implicated in rare-disease pathogenesis; however, we have a limited framework for integrating TADs and their variation across cell types into the interpretation of common-trait-associated variants. Here, we investigate an attribute of 3D genome architecture-the stability of TAD boundaries across cell types-and demonstrate its relevance to understanding how genetic variation in TADs contributes to complex disease. By synthesizing TAD maps across 37 diverse cell types with 41 genome-wide association studies (GWASs), we investigate the differences in disease association and evolutionary pressure on variation in TADs versus TAD boundaries. We demonstrate that genetic variation in TAD boundaries contributes more to complex-trait heritability, especially for immunologic, hematologic, and metabolic traits. We also show that TAD boundaries are more evolutionarily constrained than TADs. Next, stratifying boundaries by their stability across cell types, we find substantial variation. Compared to boundaries unique to a specific cell type, boundaries stable across cell types are further enriched for complex-trait heritability, evolutionary constraint, CTCF binding, and housekeeping genes. Thus, considering TAD boundary stability across cell types provides valuable context for understanding the genome's functional landscape and enabling variant interpretation that takes 3D structure into account.

SuperTAD: robust detection of hierarchical topologically associated domains with optimized structural information

Topologically associating domains (TADs) are the organizational units of chromosome structures. TADs can contain TADs, thus forming a hierarchy. TAD hierarchies can be inferred from Hi-C data through coding trees. However, the current method for computing coding trees is not optimal. In this paper, we propose optimal algorithms for this computation. In comparison with seven state-of-art methods using two public datasets, from GM12878 and IMR90 cells, SuperTAD shows a significant enrichment of structural proteins around detected boundaries and histone modifications within TADs and displays a high consistency between various resolutions of identical Hi-C matrices.

Reconstructing high-resolution chromosome three-dimensional structures by Hi-C complex networks

BACKGROUND

Hi-C data have been widely used to reconstruct chromosomal three-dimensional (3D) structures. One of the key limitations of Hi-C is the unclear relationship between spatial distance and the number of Hi-C contacts. Many methods used a fixed parameter when converting the number of Hi-C contacts to wish distances. However, a single parameter cannot properly explain the relationship between wish distances and genomic distances or the  locations of topologically associating domains (TADs).

RESULTS

We have addressed one of the key issues of using Hi-C data, that is, the unclear relationship between spatial distances and the number of Hi-C contacts, which is crucial to understand significant biological functions, such as the enhancer-promoter interactions. Specifically, we developed a new method to infer this converting parameter and pairwise Euclidean distances based on the topology of the Hi-C complex network (HiCNet). The inferred distances were modeled by clustering coefficient and multiple other types of constraints. We found that our inferred distances between bead-pairs within the same TAD were apparently smaller than those distances between bead-pairs from different TADs. Our inferred distances had a higher correlation with fluorescence in situ hybridization (FISH) data, fitted the localization patterns of Xist transcripts on DNA, and better matched 156 pairs of protein-enabled long-range chromatin interactions detected by ChIA-PET. Using the inferred distances and another round of optimization, we further reconstructed 40 kb high-resolution 3D chromosomal structures of mouse male ES cells. The high-resolution structures successfully illustrate TADs and DNA loops (peaks in Hi-C contact heatmaps) that usually indicate enhancer-promoter interactions.

CONCLUSIONS

We developed a novel method to infer the wish distances between DNA bead-pairs from Hi-C contacts. High-resolution 3D structures of chromosomes were built based on the newly-inferred wish distances. This whole process has been implemented as a tool named HiCNet, which is publicly available at http://dna.cs.miami.edu/HiCNet/ .

Comparison of computational methods for the identification of topologically associating domains

BACKGROUND

Chromatin folding gives rise to structural elements among which are clusters of densely interacting DNA regions termed topologically associating domains (TADs). TADs have been characterized across multiple species, tissue types, and differentiation stages, sometimes in association with regulation of biological functions. The reliability and reproducibility of these findings are intrinsically related with the correct identification of these domains from high-throughput chromatin conformation capture (Hi-C) experiments.

RESULTS

Here, we test and compare 22 computational methods to identify TADs across 20 different conditions. We find that TAD sizes and numbers vary significantly among callers and data resolutions, challenging the definition of an average TAD size, but strengthening the hypothesis that TADs are hierarchically organized domains, rather than disjoint structural elements. Performances of these methods differ based on data resolution and normalization strategy, but a core set of TAD callers consistently retrieve reproducible domains, even at low sequencing depths, that are enriched for TAD-associated biological features.

CONCLUSIONS

This study provides a reference for the analysis of chromatin domains from Hi-C experiments and useful guidelines for choosing a suitable approach based on the experimental design, available data, and biological question of interest.

Genome-Wide Chromatin Structure Changes During Adipogenesis and Myogenesis

The recently developed high-throughput chromatin conformation capture (Hi-C) technology enables us to explore the spatial architecture of genomes, which is increasingly considered an important regulator of gene expression. To investigate the changes in three-dimensional (3D) chromatin structure and its mediated gene expression during adipogenesis and myogenesis, we comprehensively mapped 3D chromatin organization for four cell types (3T3-L1 pre-adipocytes, 3T3-L1-D adipocytes, C2C12 myoblasts, and C2C12-D myotubes). We demonstrate that the dynamic spatial genome architecture affected gene expression during cell differentiation. A considerable proportion (~22%) of the mouse genome underwent compartment A/B rearrangement during adipogenic and myogenic differentiation, and most (~80%) upregulated marker genes exhibited an active chromatin state with B to A switch or stable A compartment. More than half (65.4%-73.2%) of the topologically associating domains (TADs) are dynamic. The newly formed TAD and intensified local interactions in the Fabp gene cluster indicated more precise structural regulation of the expression of pro-differentiation genes during adipogenesis. About half (32.39%-59.04%) of the differential chromatin interactions (DCIs) during differentiation are promoter interactions, although these DCIs only account for a small proportion of genome-wide interactions (~9.67% in adipogenesis and ~4.24% in myogenesis). These differential promoter interactions were enriched with promoter-enhancer interactions (PEIs), which were mediated by typical adipogenic and myogenic transcription factors. Differential promoter interactions also included more differentially expressed genes than nonpromoter interactions. Our results provide a global view of dynamic chromatin interactions during adipogenesis and myogenesis and are a resource for studying long-range chromatin interactions mediating the expression of pro-differentiation genes.

RUNX1 contributes to higher-order chromatin organization and gene regulation in breast cancer cells

RUNX1 is a transcription factor functioning both as an oncogene and a tumor suppressor in breast cancer. RUNX1 alters chromatin structure in cooperation with chromatin modifier and remodeling enzymes. In this study, we examined the relationship between RUNX1-mediated transcription and genome organization. We characterized genome-wide RUNX1 localization and performed RNA-seq and Hi-C in RUNX1-depleted and control MCF-7 breast cancer cells. RNA-seq analysis showed that RUNX1 depletion led to up-regulation of genes associated with chromatin structure and down-regulation of genes related to extracellular matrix biology, as well as NEAT1 and MALAT1 lncRNAs. Our ChIP-Seq analysis supports a prominent role for RUNX1 in transcriptional activation. About 30% of all RUNX1 binding sites were intergenic, indicating diverse roles in promoter and enhancer regulation and suggesting additional functions for RUNX1. Hi-C analysis of RUNX1-depleted cells demonstrated that overall three-dimensional genome organization is largely intact, but indicated enhanced association of RUNX1 near Topologically Associating Domain (TAD) boundaries and alterations in long-range interactions. These results suggest an architectural role for RUNX1 in fine-tuning local interactions rather than in global organization. Our results provide novel insight into RUNX1-mediated perturbations of higher-order genome organization that are functionally linked with RUNX1-dependent compromised gene expression in breast cancer cells.

Structural and functional diversity of Topologically Associating Domains

Recent studies have shown that chromosomes in a range of organisms are compartmentalized in different types of chromatin domains. In mammals, chromosomes form compartments that are composed of smaller Topologically Associating Domains (TADs). TADs are thought to represent functional domains of gene regulation but much is still unknown about the mechanisms of their formation and how they exert their regulatory effect on embedded genes. Further, similar domains have been detected in other organisms, including flies, worms, fungi and bacteria. Although in all these cases these domains appear similar as detected by 3C-based methods, their biology appears to be quite distinct with differences in the protein complexes involved in their formation and differences in their internal organization. Here we outline our current understanding of such domains in different organisms and their roles in gene regulation.

Architectural proteins, transcription, and the three-dimensional organization of the genome

Architectural proteins mediate interactions between distant sequences in the genome. Two well-characterized functions of architectural protein interactions include the tethering of enhancers to promoters and bringing together Polycomb-containing sites to facilitate silencing. The nature of which sequences interact genome-wide appears to be determined by the orientation of the architectural protein binding sites as well as the number and identity of architectural proteins present. Ultimately, long range chromatin interactions result in the formation of loops within the chromatin fiber. In this review, we discuss data suggesting that architectural proteins mediate long range chromatin interactions that both facilitate and hinder neighboring interactions, compartmentalizing the genome into regions of highly interacting chromatin domains.