Transcriptional activation during cell reprogramming correlates with the formation of 3D open chromatin hubs

ARTICLE https://doi.org/10.1038/s41467-020-16396-1 OPEN Transcriptional activation during cell reprogramming correlates with the formation of 3D open chromatin hubs 1234567890():,; Marco Di Stefano 1,2 ✉, Ralph Stadhouders Thomas Graf 2 ✉ & Marc A. Marti-Renom 2,5, Irene Farabella 1,2,3,4 ✉ 1,2, David Castillo 1,2, François Serra 1,2,6, Chromosome structure is a crucial regulatory factor for a wide range of nuclear processes. Chromosome conformation capture (3C)-based experiments combined with computational modelling are pivotal for unveiling 3D chromosome structure. Here, we introduce TADdyn, a tool that integrates time-course 3C data, restraint-based modelling, and molecular dynamics to simulate the structural rearrangements of genomic loci in a completely data-driven way. We apply TADdyn on in situ Hi-C time-course experiments studying the reprogramming of murine B cells to pluripotent cells, and characterize the structural rearrangements that take place upon changes in the transcriptional state of 21 genomic loci of diverse expression dynamics. By measuring various structural and dynamical properties, we find that during gene activation, the transcription starting site contacts with open and active regions in 3D chromatin domains. We propose that these 3D hubs of open and active chromatin may constitute a general feature to trigger and maintain gene transcription. 1 CNAG-CRG, Centre for Genomic Regulation (CRG), Barcelona Institute of Science and Technology (BIST), Baldiri i Reixac 4, 08028 Barcelona, Spain. 2 Centre for Genomic Regulation (CRG), Barcelona Institute of Science and Technology (BIST), Dr. Aiguader 88, 08003 Barcelona, Spain. 3 Universitat Pompeu Fabra (UPF), 08002 Barcelona, Spain. 4 ICREA, Pg. Lluís Companys 23, 08010 Barcelona, Spain. 5Present address: Department of Pulmonary Medicine and Department of Cell Biology, Erasmus MC, Rotterdam, the Netherlands. 6Present address: Computational Biology Group—Barcelona Supercomputing Center (BSC), 08034 Barcelona, Spain. ✉email: ; ; NATURE COMMUNICATIONS | (2020)11:2564 | https://doi.org/10.1038/s41467-020-16396-1 | www.nature.com/naturecommunications 1 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-16396-1 T he three-dimensional (3D) structure of the genome has been shown to modulate transcriptional regulation1–3 and to play a role in cancer and developmental abnormalities4. In the effort of characterizing 3D genome structures, chromosome conformation capture (3C)-based experiments5 allow to capture a single snapshot of the genome conformation at a given time. A plethora of theoretical approaches have been developed to take advantage of 3C-based experimental data and model genome spatial organization. Restraint-based modelling approaches6 take 3C-based contact frequencies as input and employ ad hoc conversions to spatial distances for determining 3D genome structure7–12. This approach has provided valuable insights into the structural organization of chromosomal regions in various organisms13. Complementary, thermodynamics-based approaches14–22 use physics-based principles to test specific interactions or interaction mechanisms to explain the molecular origins of the contact patterns obtained in 3C-based experiments. Together, these theoretical strategies provide insights into chromatin conformation16,17,23,24 and the possible mechanisms that form chromosome territories18, compartments19 and topologically associating domains (TADs)20,22,25,26. Decreased sequencing costs, together with more refined experimental protocols, has permitted performing 3C-based timeresolved experiments to monitor genome conformation dynamics of biological processes at high resolution. For example, Highthroughput chromosome conformation capture (Hi–C) experiments have been applied to study the dynamics of nuclear organization during mitosis27,28 or meiosis29–31, during hormone treatment32 and during induced neural or adipose cells differentiations33,34 or cell reprogramming35. However, none of the computational strategies developed so far can take full advantage of these time-series datasets. Hence, approaches specifically designed for the simulation of time-dependent conformational changes (4D) are urgently needed. To fill this gap, we introduce TADdyn, a computational method allowing to model 3D structural transitions of chromatin using time-resolved Hi–C datasets. We combine in TADdyn a physics-based model of chromatin fiber18,36 with dynamic restraint-based modelling. For any genomic locus, this integrated strategy allows for simulating a plausible 4D trajectory that is data-driven and at the same time satisfies basic physical properties of the chromatin fiber. The potential of TADdyn to provide insights beyond the Hi–C datasets is highlighted by the simulation of 21 loci of the mouse genome during cell reprogramming of pre-B lymphocytes into pluripotent stem cells (PSCs)35. By measuring structural and dynamical properties from the simulations, we characterize the interplay between 3D structure and gene transcription at an extent unreachable from the experimental datasets alone. Interestingly, we find that transcription starting sites (TSS) of simulated loci embed into in a cage-like structure that favors contacts with open and active regions located (even) several kilo-bases (kb) away from the gene promoter. Hence, TADdyn simulations are compatible with the formation of 3D hubs37 as a general mechanism to modulate gene transcription. Results The TADdyn modelling strategy. TADdyn is based on the following methodological steps (“Methods” and Fig. 1): (i) collection of experimental data, (ii) representation of selected chromatin regions using a bead-spring polymer model, (iii) conversion of experimental data into time-dependent restraints, (iv) application of steered molecular dynamics to simulate the adaptation of chromatin models to satisfy the imposed restraints, and (v) analysis of the conformation dynamics. As discussed below, each 2 of these steps constitutes per se an extension of all the existing restraint-based strategies for chromosome modelling and, in particular, of TADbit38, a modelling tools previously developed in our lab. We applied TADdyn to a previously published in situ Hi–C interaction time-series dataset (GEO accession number GSE96611). We could use at once restraint-based modelling for seven distinct time-points of in situ Hi–C experiments during C/EBPα priming followed by Oct4, Sox2, Klf4 and Myc (OSKM)-induced reprogramming of B cells to PSCs35. To collect statistics on distinct expression dynamics, we focused on 20 different ~2 mega-bases (Mb) regions of the mouse genome encompassing a total of 21 different loci (Supplementary Data 1). The selected genes are representative of different time-dependent patterns of transcriptional activity (Supplementary Fig. 1 and “Methods”), which allowed us to study how various different transcription dynamics interplay with changes in the 3 (...truncated)