The non-equilibrium genome - where Physics meets Biology

Kazuhiro Maeshima1,2 ¹National Institute of Genetics, ROIS, Mishima, Japan 2Genetics Course, Graduate Institute for Advanced Studies (SOKENDAI), Mishima, Linker histone H1, the most abundant chromatin protein, condenses chromatin, modulates DNA transactions such as transcription and DNA replication/repair, and participates in differentiation, development and tumorigenesis. While recent studies indicate that nucleosomes are clustered as condensed chromatin domains in higher eukaryotic cells, how histone H1 mechanically condenses chromatin remains unclear. Here, using a combination of direct visualization of single-H1 molecules in living human cells and multiscale molecular dynamics simulations, we demonstrate that the majority of H1 behaves like a liquid inside chromatin domains, rather than binding stably to nucleosomes as suggested by the traditional model. H1 functions as a liquid-like “glue”, mediating dynamic multivalent interactions between nucleosomes within chromatin domains. Consistently, rapid-depletion of H1.2 leads to decondensed chromatin domains both in cells and in silico. Our findings suggest that the H1 “glue” condenses chromatin domains while keeping them fluid and accessible, thereby supporting essential DNA transactions. Reference: M.A. Shimazoe et al. bioRxiv (2025) doi: https://doi.org/10.1101/2025.03.05.641622

Chromatin is the substrate for all DNA-associated processes in eukaryotic cells, and understanding its structural and dynamical properties is crucial for our understanding of gene regulation and other nuclear processes. It was once thought that chromatin would adopt very regular structures, for example, the 30-nm diameter solenoid-like fibres which were often presented in textbooks. But there is now much evidence which suggests that chromatin actually adopts a broad range of different structures within cells, and that these different structures play an important role in function. However, we do not have a clear understanding of what controls fibre structure and properties. I will present results from recent work which combines experiments and computer simulations to study the properties of artificial chromatin fibre constructs which mimic different types of chromatin (heterochromatin-like regularly spaced nucleosomes, or euchromatin-like irregular spacings). Our results reveal a striking and sharp order-to-disorder transition in fibre structure as small levels of irregularity in linker lengths. This suggests that irregularity in nucleosome positioning can lead to broad and dynamic variation of many features of the resulting fibres, which, for example, could be the source of noise, cell-to-cell variability and plasticity in gene expression.

Gene regulation is inherently a dynamic, non-equilibrium process that unfolds in space and time within the crowded nuclear environment. While the one-dimensional genome encodes regulatory information through sequence and epigenetic modifications, its deployment is constrained by the three-dimensional organization and dynamics of chromatin. In this talk, I will discuss how transcriptional activity can be viewed as a four-dimensional process, shaped by the interplay between chromatin mechanics, stochastic molecular interactions, and regulatory network dynamics. Drawing on quantitative live-cell measurements and physical modeling, I will illustrate how dynamic chromatin constraints influence transcriptional bursting, information transmission, and the emergence of reproducible gene expression patterns. This perspective highlights how non-equilibrium chromatin dynamics integrate 1D and 3D genomic processes to enable robust biological function.

Polymers with active segments constitute prospective future materials and are used as a model for some biological systems such as chromatin. The directions of the active forces are typically introduced with temporal or spatial correlations to establish directional motion of the chain and corresponding active dynamics. Instead, here we consider an active-passive copolymer, where the two segments differ only by the magnitude of their fluctuations and feature no artificial correlations. Here we show that although the model itself does not possess directional dynamics, if the chains are concentrated, directional persistent motion spontaneously arises as a consequence of the broken translational symmetry owing to the topological constraints. Using scaling arguments and simulations, we explain the phenomenon and describe the ensuing dynamics. I will discuss the consequences for the chromatin and compare with available data.

We present a new model that enables the purely analytical calculation of the base-pair sequence-dependent free energy of nucleosomes. We showcase various applications of this approach. Furthermore, we discuss the role of biomolecular condensation in large-scale chromatin organization and information transfer.

Quantitative gene regulation at the population level can be achieved by two fundamentally different modes of regulation at individual gene copies. A “digital” mode involves binary ON/OFF expression states, with population level variation arising from the proportion of gene copies that are ON or OFF, while an “analogue” mode involves graded expression levels at each gene copy. We will discuss the contexts in which each mode of regulation is used, focusing on the importance of digital states in epigenetic memory systems. We will then examine how nonlinear feedback can stabilise digital memory states, examining particularly the case of histone modification read-write feedback. One system where such histone memory states are critical is for memory of winter in plants, needed to ensure that flowering occurs only after winter has passed. We will highlight recent interlinked theoretical and experimental progress in understanding how this quantitative memory is set up and then stably propagated. We will also discuss the case of DNA methylation memory, where due to the very low noise present in the maintenance mechanism, analogue memory states do appear to be possible.

TBC

Dissecting the 3D-chromatin organization in cell physiology is a key area of investigation. Using quantitative super-resolution nanoscopy, we identified a novel chromatin fiber assembly and its relation with naïve pluripotency. We found that nucleosomes are arranged in groups of various sizes, the nucleosome clutches, which control gene function. Recently, by combining imaging and genomic approaches, we developed MiOS, a powerful integrated strategy to model the folding of key pluripotency genes at nucleosome resolution and identified the chromatin structure surrounding them at early stages of embryo development. We are now using a novel integrative modeling framework that streamline the reconstruction of chromatin folding by integrating chromatin tracing data with genomic approaches. Furthermore, building on our observation that transcriptional-dependent supercoiling controls loop formation and 3D-genome organization, we are currently studying how topoisomerase activity controls supercoiling-dependent chromatin folding. Overall, the combination of super-resolution microscopy with genomic and modeling methods allowed us to dissect the functional role of the genome folding in the regulation of cell function.

At many spatial and temporal scales, genome organization and nuclear architecture are closely linked to a variety of biological processes acting on chromosomes. Our understanding of this organization has greatly progressed in recent decades, thanks to advances in sequencing-based technologies for mapping chromosomal contacts and microscopy-based approaches for visualizing chromosome folding and dynamics. Yet beyond visualization, the ability to physically manipulate chromosomes inside living cells holds great potential to reveal what passive observation cannot. In this context, we have developed an experimental approach for the mechanical manipulation of chromosomes by applying magnetic force to a specific genomic locus, enabling its displacement within the nuclear space. From a physicist’s point of view, this allows us to probe the material nature of interphase chromatin. From a biologist’s perspective, it offers a unique perturbation tool to assess causal relationships between genome structure and function. I will discuss what this approach has revealed about the physical properties of interphase chromosomes, the new questions it raises, and how we are beginning to address them.

Technological advances in chromatin structure characterization have continually refined our understanding of transcriptional regulation in eukaryotic systems. Despite these developments, achieving nucleosome-resolution structural characterization remains a significant challenge. As a result, it remains unclear what structural features distinguish active enhancers and promoters and how these elements are organized. To address this, we developed a simulation framework that leverages high-resolution Region-Capture Micro-C (RCMC) contact maps to infer conformational ensembles of megabase-scale chromatin segments at nucleosome resolution. A key component of this framework is a balancing strategy tailored for Micro-C data, which identifies per-nucleosome variation in contact density, in contrast to existing methods that assume uniform contact density. Our model accurately reproduces contact frequencies observed in RCMC data, pairwise spatial distances measured via chromatin tracing, and local structural motifs observed in imaging studies. The high spatial and genomic resolution of the inferred structures reveal a striking departure from the classical view of euchromatin as uniformly open. Instead, euchromatin generally folds into compact domains, consistent with the ``packing domains" observed in imaging studies. These domains are further organized into disordered, insulated clusters of approximately 50 nucleosomes, resembling nucleosome clutches. Notably, kilobase-scale regions surrounding active promoters and enhancers often protrude from these condensed domains, becoming highly accessible. This spatial arrangement effectively compartmentalizes regulatory elements from the surrounding chromatin, facilitating protein binding and promoting enhancer–promoter communication. The distinct structural features of euchromatin revealed here offer new insights into enhancer regulation and may help explain their enigmatic behavior.

The human genome contains multiple layers of information that extend beyond the genetic sequence, where diverse epigenetic patterns and structural variations operate in concert to regulate cellular function. Capturing these layers simultaneously on individual DNA molecules is essential for understanding the great variation in structure and function displayed by cells with identical genetic backgrounds. Here, we present a versatile "write-and-read" framework that leverages chemo-enzymatic DNA labeling with synthetic tags to record multiomic information, such as the locations of DNA-binding proteins and histone modifications, directly onto the sequenced DNA molecule. This framework enables the simultaneous recording of genetic sequences, naturally occurring modifications such as methylation and hydroxymethylation, and chemically encoded protein-binding sites. This bio-orthogonal labeling strategy significantly extends the detectable molecular alphabet, enabling the simultaneous analysis of multiple omic features on single molecules. To provide a long-range structural context, we utilize single-molecule Optical Genome Mapping, a high-throughput method where long chromosomal fragments are stretched in nanochannel arrays. This technology facilitates the de-novo construction of complex genomes by imaging long DNA molecules over large genomic distances, uncovering structural insights inaccessible by standard sequencing. We use these high-resolution optical maps as a physical scaffold for the precise alignment of multiomic nanopore data. By integrating the structural scaffolding of optical mapping with the nucleotide-specific multiomic resolution of nanopore sequencing, this platform enables the characterization of multiple genomic observables on individual DNA molecules, advancing our ability to decode the complex layers of information within the human genome.

Three-dimensional chromatin organization plays a crucial role in gene regulation, yet understanding how chromatin architecture arises from genomic features remains challenging. Existing computational approaches face a fundamental trade-off: deep learning models predict contact maps at scale but lack mechanistic interpretability, while polymer physics simulations provide mechanistic insight but are not scalable genome-wide. We present the differentiable loop extrusion model (dLEM), a scalable framework that reformulates cohesin-mediated loop extrusion, a central mechanism organizing genomes, as a mechanistic yet trainable process. dLEM represents extrusion through position-specific velocity profiles for leftward and rightward cohesin movement and a cohesin detachment rate, deconvolving complex 2D contact maps into interpretable 1D parameters that capture extrusion dynamics and align naturally with genomic and epigenomic features. We demonstrate that dLEM accurately reconstructs chromatin architecture across multiple resolutions and cell types, with median correlations of 0.68 in observed-over-expected contact frequency. The fitted velocity parameters show strong directional correspondence with CTCF binding: leftward velocity is reduced at forward-oriented CTCF sites while rightward velocity is reduced at reverse-oriented sites, consistent with the known asymmetric blocking mechanism. Analysis across three human cell lines (H1hESC, IMR90, K562) reveals that while CTCF remains the dominant regulator of cohesin dynamics, transcriptional machinery (indicated through markers such as H3K4me3, H3K36me3, and chromatin accessibility) makes reproducible contributions to extrusion rates. The model's physically-grounded parameters enable quantitative perturbation analysis: we successfully predict structural outcomes of depletion of loop extrusion components. For WAPL depletion, we show that modifying only the detachment rate parameter successfully recapitulates the emergence of new chromatin loops observed experimentally. For CTCF depletion, we demonstrate that residual CTCF in depleted cells retains functional barrier activity at precisely the level predicted by the fraction of remaining ChIP-seq peaks (0.33). These results demonstrate that dLEM enables principled hypothesis testing that is both interpretable, unlike black-box deep learning models, and scalable genome-wide, unlike polymer physics simulations. Extending this framework, deep dLEM demonstrates how biophysical processes can serve as interpretable layers in deep learning. By embedding dLEM as a fixed mechanistic layer within a neural network trained on DNA sequence and chromatin accessibility, we achieve competitive performance with state-of-the-art models (comparable to the state-of-the-art models Orca and C.Origami) while using up to 700-times fewer parameters. This design uses extrusion dynamics through dLEM parameters as a biophysically grounded latent space, preserving interpretability while enabling genome-wide prediction of contact maps from widely available data and removing the need for expensive experimental contact maps or CTCF ChIP-seq measurements. Together, dLEM and deep dLEM provide a unified framework for understanding how chromatin architecture arises from genomic features, achieving both mechanistic interpretability for perturbation prediction and flexible learning for sequence-to-structure mapping.

Biological function relies on an actively maintained, non-equilibrium genome state: epigenetic marks and 3D chromatin organization must be continuously written, erased, and remodeled to stabilize gene regulation and cell fate. In a warm, fluctuating environment, stochastic drift (thermal fluctuations, replication/repair errors, remodeling noise) continuously perturbs this genome state, implying a nonzero energetic “maintenance floor” required to keep the system in a functional attractor. We model a self-replicator receiving metabolic power P(t) but delivering only a fraction to genome maintenance, and define aging as the positive maintenance shortfall integrated over time, an accumulated aging load that generically yields Gompertz-like failure statistics without assuming Gompertz a priori. This frames epigenetic drift and progressive disorganization of genome architecture as the inevitable consequence of running a driven, leaky information-bearing system below its required maintenance throughput, including irreducible information-theoretic costs for correcting errors. Moreover, we derive a thermodynamic lower bound on rejuvenation, viewed as finite-time genome-state restoration by a molecular controller (endogenous repair/remodeling or engineered reprogramming). Using trajectory-level stochastic thermodynamics, each successful correction must dissipate at least a physical free-energy cost plus a Landauer-type reset cost proportional to the control information processed, implying a throughput constraint on how fast genome order can be restored. In this view, rejuvenation is aging-load clearance under a deadline, requiring a transient increase in correction rate and thus in required power and dissipation. Together, these results yield quantitative trade-offs between speed, scope, and specificity for epigenetic reprogramming and chromatin remodeling, suggesting experimentally testable scaling relations for energy usage and heat dissipation during genome-state maintenance versus rejuvenation.

Biological function relies on the coordinated regulation of genome organization in 1D and 3D, despite recurrent perturbations by the cell cycle, most notably DNA replication and cell division. We investigate the “unreasonable effectiveness” of equilibrium gene regulation in this dynamic setting by quantifying how regulatory systems preserve stable input–output behavior despite genome replication-driven disruptions (Vilar & Saiz, Cell Systems, 2024). Our results show that common regulatory architectures can deploy compensatory kinetic mechanisms that suppress non-equilibrium effects associated with genome duplication, including perturbations induced by replication fork passage, thereby explaining the paradoxical accuracy of equilibrium predictions in inherently non-equilibrium regimes. This robustness is consistent with single-cell evidence that DNA looping effects can persist across many generations through competition between fast repressor rebinding and slow transcriptional initiation (Chang et al., PNAS 2022). By combining statistical physics with single-molecule and single-cell dynamics, we outline practical principles for reconciling equilibrium models with cell-cycle-driven non-equilibrium processes in living cells. References: J. M. G. Vilar and L. Saiz, The unreasonable effectiveness of equilibrium gene regulation through the cell cycle, Cell Systems 15, 639-648 (2024). C. Chang, M. Garcia-Alcala, L. Saiz, J. M. G. Vilar, and P. Cluzel, Robustness of DNA Looping Across Multiple Cell Divisions in Individual Bacteria, Proc. Natl. Acad. Sci. USA 119, e2200061119 (2022).

Cells compact and organize their genomes through DNA-binding proteins that can undergo phase separation with DNA. However, how the underlying DNA sequence modulates this process remains poorly understood. I will present our combined computational and experimental study uncovering the sequence-dependent mechanisms that govern protein–DNA co-condensation and genome compaction. Brownian dynamics simulations show that nonspecific binding yields a single condensate, whereas sequence-dependent interactions generate multiple condensates whose sizes are dictated by the interplay between DNA–protein and protein–protein affinities. Experiments on the Mycobacterium tuberculosis (Mtb) nucleoid-associated protein Lsr2 reveal that it compacts DNA via co-condensation by selectively binding to AT-rich regions. Remarkably, these Lsr2–DNA condensates appear to “sense” the average binding energy landscape rather than discrete sequence motifs. Together, these findings establish a unified physical framework for how DNA sequence encodes the propensity for protein-mediated condensation, providing new insight into the principles of genome organization and regulation.

The 3D genome structure in living human cells is subjected to perturbations such as physical forces, DNA damaging agents, osmotic and chemical stresses, and nuclear shape changes. These stresses on a cell can induce cell fate changes that are related to changes in nuclear architecture. The 3D genome’s response to such stresses can also reveal properties of structures such as spatial compartments or CTCF-mediated loops. Here, we present integrated observations of genome structure changes across a set of different perturbations. We find that cancer cells which migrate through multiple constrictions experience stable epigenetic, 3D genome structure, and nuclear lamin alterations that associate with a stable change in cell phenotype. In contrast, the chromosome structure changes induced by transient low or high salt (hypo- or hyper-osmotic) stress are largely reversible. We find that human and Drosophila chromosome structures actually respond quite differently to hyperosmotic stress, revealing different underlying folding mechanisms that can lead to similar contact patterns. Meanwhile, the DNA-intercalating cancer drug curaxin disrupts the 3D genome structure in some ways that mimic salt stress (including removing CTCF from the chromosomes). But, unlike salt stress, curaxin treatment results in signatures of increased cohesin loop extrusion. Our observations begin to shed light on the robustness and sensitivities of the 3D genome structure to perturbation and how the network of 3D contacts in the genome can accomplish both gene regulatory functions and contribute to necessary physical properties of the nucleus.

Cell fate decisions emerge as a consequence of a complex set of gene regulatory networks. Models of these networks are known to have more parameters than data can determine. Recent work has instead mathematically formalized the concept of a Waddington landscape. These landscapes are minimally parameterized descriptions of cell fate decisions. We will describe how to construct these landscapes, and demonstrate their utility with several examples. We will show how they can be extended to describe spatial patterns.

Olfactory sensory neurons select a single olfactory receptor (OR) gene from over a thousand possibilities, but the mechanisms underlying this specificity remain unclear. I will present a predictive two-step model in which OR choice emerges from interactions between OR genes and their enhancers. We validated the model’s predictions by integrating large-scale single-cell and spatial transcriptomics data and reconstructing zonally resolved expression maps using optimal transport. This combined modeling–data approach provides a quantitative framework for OR choice.

Mixtures of chromatin proteins and DNA often undergo liquid-liquid phase separation in vitro that can be used as reductionist models of chromatin. We present models describing the phase diagrams of HP1$\alpha$-DNA and H1-DNA mixtures. In both cases, the models require cooperative and non-cooperative binding states that lead to a subtle competition between attractive and non-attractive states. We will conclude with a discussion of these competing states within the full context of chromatin.

Transcription factors (TFs) regulate gene expression by binding to specific DNA sequences within a complex and heterogenous chromatin environment to assemble transcriptional machinery at specific genomic loci. The mechanisms by which TFs bind to their target sequences on dynamic chromatin to regulate gene expression remains elusive. Single-molecule tracking (SMT) has emerged as a powerful approach to explore chromatin and transcription factor dynamics and their interaction in living cells. We used single-molecule tracking to directly measure the interaction dynamics of a broad spectrum of transcription factors, including steroid hormone receptors, chromatin remodelers and architectural proteins in live cells. We found that TFs follow power-law distributed binding times, suggesting that most TFs interact with chromatin with a broad range of binding affinities, rather than the conventional model of specific and non-specific binding affinities. Using machine-learning based analysis of single molecule diffusion, we show that chromatin displays two distinct low-mobility, highly sub-diffusive states. Our experimental observations are consistent with a minimal active copolymer model for interphase chromosomes. Remarkably, we find that a diverse set of transcription factors, co-regulators and remodelers also exhibit two distinct low-mobility states. Our studies further reveal that lowest mobility state requires an intact DNA binding domain and oligomerization domains, while protein-protein interactions between TFs and co-regulators define the higher mobility state. Finally, we present some recent results on how mechanical properties of the microenvironment shape chromatin mobility and TF dynamics. Together, our results elucidate how transcription factor and chromatin mobility regulates transcriptional activation in mammalian cells.

Mesoscale genomic spatial organization in vertebrates is driven by two major mechanisms: spatially segregated euchromatin and heterochromatin compartments form by microphase separation and loops dynamically form through the activity of loop-extruding protein motors. Loop extrusion by cohesin has been observed to interfere with compartmentalization. Depletion of the cohesin unloader, WAPL, extends cohesin’s residence time on chromatin, which allows the formation of larger loops and suppresses compartmental contacts in genomic contact maps (Hi-C). Inhibition of compartmentalization was generally thought to be due to indiscriminate bridging of genomic loci from different compartments. However, co-depletion of WAPL with other cohesin regulators, PDS5 proteins, dramatically increases cohesin residence time, but surprisingly, enhances near-cis chromatin compartmentalization compared to WAPL depletion alone. Using molecular dynamics simulations, we identified a physical mechanism through which loop extrusion dynamics govern chromatin compartmentalization. The simulations indicate that the ability of the chromatin fiber to phase separate into compartments depends non-monotonically on cohesin residence times. This effect depends on the presence of active loop extrusion; differences in static loop architecture have minimal effects on chromatin compartmentalization. By measuring chromatin polymer relaxation times, we find an unexpected interplay between polymer and extruder dynamics. Stabilizing loop extruders can promote compartmentalization when loop extruders turnover more slowly than the time for polymer segment equilibration to occur. This relationship affects chromatin compartmentalization differentially across length scales, corresponding to their respective polymer segment relaxation times. Furthermore, compartmentalization is therefore also controlled by other factors altering loop dynamics, such as extrusion velocity, extruder density, and extruder stalling. Our simulations also suggest a molecular role for PDS5 proteins in regulating cohesin’s extrusion velocity, which is supported by a combination of experiments, including single-molecule observations of cohesin, PDS5 proteins, and the loop-extrusion processivity factor NIPBL. Altogether, our study develops a mechanistic framework for understanding how the nonequilibrium dynamics of chromatin loops interacts with the equilibration of chromatin compartments, and consequently, how molecular perturbations to loop extrusion may alter mesoscale genome architecture.

The three-dimensional genome architecture is crucial for regulating gene expression, DNA replication, and other essential processes. While extensively studied in interphase, chromatin organization during mitosis remains poorly understood, with conflicting findings. GPSeq (Genomic Loci Positioning by Sequencing), the first high-resolution method for mapping radial genome organization, reveals a steep GC content gradient from the nuclear periphery to the center. We hypothesize that this feature persists through the cell cycle, meaning mitotic chromosomes retain a “memory” of interphase radiality. To test this, we applied GPSeq to DT40 chicken cells synchronized at the G2/M boundary, ensuring a homogeneous prometaphase population. Our results show a strong correlation between GC content and GPSeq score, a measure of radiality, supporting our hypothesis. Integration of Repli-Seq data further reveals that late-replicating, AT-rich, gene-poor regions are preferentially localized to peripheral chromatin shells. To deepen our understanding, we will incorporate ChIP-seq for condensins—key architectural proteins in mitotic chromosome folding—and design DNA FISH probes to visualize loci spanning core-to-periphery gradients in individual mitotic nuclei. Importantly, by tracking selected loci from interphase through mitosis and into the next G1, we aim to determine whether radial positioning is conserved or re-established, shedding light on the continuity and regulation of nuclear architecture across the cell cycle. Together, these approaches will help us construct a predictive model of mitotic genome organization and confirm whether GC content serves as a universal organizing principle throughout cell division.

How mechanical stress is buffered across scales from individual DNA strands to the whole nucleus remains an open physical problem. Here we develop a quantitative multiscale theory in which interactions between the nuclear envelope protein LEM2, the DNA-binding protein BAF, and DNA give rise to an emergent load-bearing system. When DNA is held at a given force in optical tweezers, addition of LEM2 and BAF causes a force increase that is proportional to the initial force. We show that the observed force-dependent stiffening in protein–DNA assemblies can be captured by an effective spring model arising from multivalent protein–protein and protein–DNA interactions. This molecular description coarse-grains into a mesoscale elastic surface hydrogel localized at the nuclear envelope. In this framework, LEM2–LEM2 interactions contract the hydrogel relative to its relaxed state, generating intrinsic pre-stress. Using parameters obtained from single-molecule measurements, we construct a minimum continuum surface model that yields (i) free-energy-minimizing nuclear shapes and (ii) an effective area stiffness of the nuclear envelope, both in agreement with experiment. Together, these results identify a protein–DNA surface hydrogel as a pre-stressed elastic shell that mechanically reinforces the nucleus and buffers genomic stress across scales.

In this talk, I will present some of my recent work, in close collaboration with colleagues in experiment and theory, on the organization of chromatin by the binding of proteins such as transcription factors. We will focus on transcription factors that induce DNA curvature, with an outlook towards studying DNA twist in future work, and how their lateral and circumferential distribution couples to chromatin geometry.