4D nucleome modeling

Abstract

The intrinsic dynamic nature of chromosomes is emerging as a fundamental component in regulating DNA transcription, replication, and damage-repair among other nuclear functions. With this increased awareness, reinforced over the last ten years, many new experimental techniques, mainly based on microscopy and chromosome conformation capture, have been introduced to study the genome in space and time. Owing to the increasing complexity of these cutting-edge techniques, computational approaches have become of paramount importance to interpret, contextualize, and complement such experiments with new insights. Hence, it is becoming crucial for experimental biologists to have a clear understanding of the diverse theoretical modeling approaches available and the biological information each of them can provide.

Figure 1

Exploration map. Radar chart displaying spanned areas of current experimental (a) and computational (b) approaches used to study the 4D nucleome. All axes are in arbitrary units. Vertical axis (‘resources’) indicates the required resources to execute the experiments/computations. Left axis (‘space’) indicates the coverage and depth of 3D space by either experiments or computation. Right axis (‘time’) indicates the coverage and depth of time by either experiments or computation. Dashed grey lines exemplifies a ‘perfect approach’ that requires very little resources but can provide the maximum insight in both space and time. Both the experimental and computational approaches have extensively charted the space dimension, but yet there is some work to do in unraveling the effects of the local scale on the global ones, and vice versa. In this sense, hybrid modeling has not yet exploited this to the fullest. The time axis has a great potential for further 4D nucleome modeling in parts of the exploration map still inaccessible to experiments. The resource dimension is currently the limiting factor, since both experiments and computation tend to use it at maximum. Experimental resource needs could be limited by, for example, reducing material requirements, as for instance a recently introduced low-input Hi-C technique [60]. As for computational resources, data-driven approaches are generally less demanding than top-down approaches, but the implementation of more efficient software may balance out this difference. For instance, bottom-up computational methods usually rely on few local force-fields and thus ought to be more computationally scalable than data-driven ones, given an efficient software implementation [34].

Figure 2

Illustration of selected 4D modeling studies. On the left, each panel (a)–(f) shows the portion of the exploration map charted by the corresponding approach. These graphs illustrate the space and time dimensions one could explore given the available resources applying the various approaches. (a) Whole-genome models were obtained using single-cell Hi-C data at different cell-cycle phases [21^••]. Models at different timepoints were crucial to unveil structural features only implicit in the Hi-C contact patterns such as the difference in decompaction speed between A (fast) and B (slow) during the G1 progression. (b) Center: Chrom3D whole-genome models in human adipose stem cells [18^••] at the single-TAD level resolution. Each chromosome is indicated with a distinct color. Right: Three groups of TADs (cliques) in a repressive compartment predicted to be interacting with the nuclear lamina. (c) Trajectories of the Sox2 locus dynamics were simulated during the mouse B-to-iPSC reprogramming using the TADdyn tool [27^••]. Upon expression activation in day 6 (D6), several regions (red beads) with enhancer characteristics (Open and Active chromatin) gather around the TSS of the locus (black bead) and form a 3D superenhancer hub. (d) Bead-and-spring polymer models with temporary crosslinking interactions were used in Ref. [41^•] to study the structure and dynamics of the Igh locus in live mouse B-lymphocytes. The authors showed that the observed constrained motion of chromatin is consistent with a network of long-lived loops ensuring that the genomic region is ordered, but maintains enough fluidity. (e) 4D application of the loop extrusion model during mitotic chromosome folding [54^••]. In particular, going from Prophase to Prometaphase, condensin II acts first by forming a helical scaffold of large adjacent loops (400 kbp), then condensin I further compacts these loops by folding them into shorter nested loops (80 kbp). (f) 3D organization of heterochromatin domains (SAHDs) in cycling and senescent human fibroblasts was modeled [23^••] by polymer simulations accounting for the capacity of SAHDs to self-interact and to associate with the nuclear lamina. The slow 3D reorganization of SAHDs into large internal foci is consistent with a substantial weakening of lamina-SAHDs interactions mediated by HMGA-2.