Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
License: CC BY 4.0
arXiv:2401.06921v1 [physics.bio-ph] 12 Jan 2024
thanks: knewhall@unc.edu

Different relative scalings between transient forces and thermal fluctuations
tune regimes of chromatin organization

Anna Colettia, Benjamin L. Walker b, Kerry Bloom c, and Katherine A. Newhalla, aDepartment of Mathematics, University of North Carolina at Chapel Hill, Chapel Hill, NC, 27510 bDepartment of Mathematics, University of California, Irvine, Irvine, CA, 92697 cDepartment of Biology, University of North Carolina at Chapel Hill, Chapel Hill, NC, 27510
(January 12, 2024)
Abstract

Within the nucleus, structural maintenance of chromosome protein complexes, namely condensin and cohesin, create an architecture to facilitate the organization and proper function of the genome. Condensin, in addition to performing loop extrusion, creates localized clusters of chromatin in the nucleolus through transient crosslinks. Large-scale simulations revealed three different dynamic behaviors as a function of timescale: slow crosslinking leads to no clusters, fast crosslinking produces rigid slowly changing clusters, while intermediate timescales are optimal for producing flexible clusters that mediate gene interaction. By mathematically analyzing different relative scalings of the two sources of stochasticity, thermal fluctuations and the force induced by the transient crosslinks, we predict these three distinct regimes of cluster behavior. Standard time-averaging that takes the fluctuations of the transient crosslink force to zero can predict the existence of clusters, but not their timescale-dependent lifetimes. Accounting for the interaction of both fluctuations from the crosslinks and thermal noise with an effective energy landscape does capture the timescale-dependent flexible cluster lifetimes. No clusters are predicted when the fluctuations of the transient crosslink force are taken to be large relative to thermal fluctuations. This mathematical perturbation analysis illuminates the importance of accounting for stochasticity in local incoherent transient forces to predict emergent complex biological behavior.

I Introduction

The genome has an intricate and hierarchical organization that allows cells to fit an extraordinary amount of genetic material and perform important nuclear functions within a nucleus mere micrometers in size. Within the nucleus, DNA is complexed with a variety of proteins to form the fibrous material chromatin that makes up chromosomes. Contributing to chromosomal architecture are structural maintenance of chromosome (SMC) protein complexes, namely condensin and cohesin, which facilitate chromosome segregation and gene regulation [18]. Such processes aid in the compaction of genetic material within the nucleus as well as allow the cell to locate and access specific genes depending on cell cycle, cell type, or environmental cue [20]. In particular, condensin helps regulate the retrieval of genetic information by acting on chromatin to form stochastic gene-gene crosslinks [6] and generate loops to induce gene mixing [9, 11, 21, 10].

The emergent behavior induced by condensin crosslinks has been studied computationally with polymer models of chromatin [8, 23, 24, 12, 26, 27] in order to more directly observe the dynamics of DNA over experimental techniques. By representing 5kbp of DNA as beads, linked through worm-like-chain springs to form a long polymer chromatin chain, the timescale of the random SMC crosslink binding between different beads was shown to control the formation of clusters and gene interactions [12, 26]. These clusters, or gene neighborhoods, reveal an underlying organizational framework that optimizes gene interactions, retains a high level of genomic compaction, and yet is highly flexible. A similar detailed level of information is not yet available experimentally; high throughput population methods such as Chromosome Conformation Capture (Hi-C) methods lack temporal evolution information, while microscopic imaging lacks genome-level resolution [22]. The ability to access both the spatial arrangement and the temporal re-arrangement of beads, which is possible in simulation, is crucial for advancing our understanding of life.

Building and remodeling these high-level genetic neighborhoods has the potential to reveal the underlying mechanics for the configuration of the energy landscape within our nuclei. In this paper, we reveal the mathematical mechanism that describes how fluctuations in the model lead to the formation of bead clusters and the temporal mixing of beads between clusters. We vary the timescale of crosslinking forces between beads to change the size of this fluctuating force relative to thermal fluctuations. These various timescales produce three different dynamic regimes. At fast bead crosslinking timescales, rigid, unchanging clusters form. At slow crosslinking timescales, the beads interact in an amorphic state with no clusters. At optimal mean crosslink lifetimes, clusters can both form and exchange beads, indicative of the required gene interaction for proper function. Tuning the timescales of crosslinking provides a unifying mechanism of genome organization that gives insight into how a large number of configurational states can be rapidly remodeled as cells encounter biological challenges.

Configurations that can be rapidly remodeled were shown to take advantage of the interaction between the crosslink binding force fluctuations and the thermal noise fluctuations to promote faster bead exchange between clusters [27]. We build upon this work by incorporating the timescale of the crosslink binding into the mathematical analysis to show how different relative scalings of the two sources of stochastic fluctuations, thermal noise and the force induced by the transient bonds, predict the three distinct regimes of cluster behavior mentioned above. Standard time-averaging that takes the fluctuations of the transient binding force to zero (fast crosslinking timescale) predicts the existence of long-lived rigid clusters. No clusters are predicted when the fluctuations of the transient binding force are taken to be large (slow crosslinking timescale) relative to the thermal fluctuations. Only when accounting for the interaction of both fluctuations from the binding and thermal noise with an effective energy landscape do we recover the flexible clusters and their timescale-dependent mixing.

This work provides the mathematical mechanism underlying the timescale-dependent effects of active agents in biological systems. The timescale controls the production of fluctuations in the forces that the active agents produce. The relative size of these fluctuations to the thermal fluctuations produces different emergent temporal behavior. This mechanism is not restricted to just the chromatin dynamics studied here but is potentially applicable to a wide range of cell-level biological processes such as homology searches for genetic recombination [4, 7, 2], defensive mucus barriers emerging from polymer chains of mucin [28, 17, 25], and error correction through configurational state changes [3, 13, 16]. Thus to deepen our understanding of how such biological systems function, we have shown it is necessary to properly analyze the interplay of different fluctuations and the timescale on which they are generated.

II Results

II.1 Timescale of crosslink binding force drives cluster formation and dynamics

We establish that it is the timescale of the stochastically-switching SMC protein crosslinking force through random binding and unbinding that affects the organization of the beads and the dynamics of the clusters. To show this, we create an idealized 361-bead model that represents the nucleolus, with each bead representing 5 kilobase pairs of DNA in the chromatin chain. The beads bind and unbind based on a biologically feasible random model for the SMC protein crosslinks that is more likely to choose pairs of nearby beads to bind.

Fig. 1 shows the formation of clusters for fast-enough timescale, β𝛽\betaitalic_β, of the random crosslink binding force. The repulsive excluded volume force and thermal fluctuations acting on the beads are in competition with the attractive stochastically changing pairwise binding forces that coalesce many beads into a cluster. Furthermore, changing the timescale of the crosslink binding force, β>0𝛽0\beta>0italic_β > 0, from slow to fast results in three distinct clustering behaviors: The system passes from an amorphic (dissolution of clusters with many bead interactions) to a flexible (frequent bead exchanges between clusters) and then rigid clustering (minimal bead interaction between clusters) behavior.

The observed timescale-dependent clustering behavior mirrors that seen in [26] which employed a larger polymer-like chromosome model and a different random model for the crosslinking proteins in the nucleolus. This demonstrates the universality of the clustering dynamics that are driven by the timescale of the binding forces.

Refer to caption
Figure 1: (A-C) Snapshot of 361 bead model for increasing binding timescale. We observe three clustering regimes: (A) amorphic, (B) flexible, and (C) rigid. (D) Mixing coefficient and average nearby number of beads for a range of β𝛽\betaitalic_β-values for 361 bead model. Flexible clustering occurs for the range of β𝛽\betaitalic_β-values within the grey region. In this regime, clusters emerge and beads interact.

II.2 Time-averaged force predicts existence of clusters but not their dynamics

To mathematically determine the effective attractive force generated by the stochastic binding we work with a further reduced 3-bead model. We start by showing that a simple time-averaging of the stochastic binding force can predict the existence of clusters but not their timescale-dependent mixing dynamics.

The chosen random model for the crosslinking forces has the added advantage of fitting into a continuous-time Markov chain (CTMC) framework. For the idealized 3-bead model, the four Markov states are easily enumerated: all beads unbound (s=1𝑠1s=1italic_s = 1), beads 1 and 2 bound (s=2𝑠2s=2italic_s = 2), beads 1 and 3 bound (s=3𝑠3s=3italic_s = 3), and beads 2 and 3 bound (s=4𝑠4s=4italic_s = 4). The time evolution of the states follows a general CTMC process with a transition rate matrix S𝑆Sitalic_S. Included in S𝑆Sitalic_S is the bead binding-rate that is proportional to the bead separation distance given by affinity function a()𝑎a(\cdot)italic_a ( ⋅ ) and a constant bond breaking-rate c𝑐citalic_c.

Now that the binding stochasticity is formulated as a Markov chain, finding the time-averaged force on each bead is straightforward [19]. The percent of time spent in each Markov chain state is given by r(x)𝑟𝑥\vec{r}(\vec{x})over→ start_ARG italic_r end_ARG ( over→ start_ARG italic_x end_ARG ), the normalized null vector of the switching matrix S(x)𝑆𝑥S(\vec{x})italic_S ( over→ start_ARG italic_x end_ARG ), so that the time-averaged force for each coordinate is

dxkdt=s=14vk(x;s)rs(x)=vk(x)fork={1,,6}𝑑subscript𝑥𝑘𝑑𝑡superscriptsubscript𝑠14subscript𝑣𝑘𝑥𝑠subscript𝑟𝑠𝑥delimited-⟨⟩subscript𝑣𝑘𝑥for𝑘16\frac{dx_{k}}{dt}=\sum_{s=1}^{4}v_{k}(\vec{x};s)r_{s}(\vec{x})=\langle v_{k}(% \vec{x})\rangle\;\text{for}\;k=\{1,\dots,6\}divide start_ARG italic_d italic_x start_POSTSUBSCRIPT italic_k end_POSTSUBSCRIPT end_ARG start_ARG italic_d italic_t end_ARG = ∑ start_POSTSUBSCRIPT italic_s = 1 end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 4 end_POSTSUPERSCRIPT italic_v start_POSTSUBSCRIPT italic_k end_POSTSUBSCRIPT ( over→ start_ARG italic_x end_ARG ; italic_s ) italic_r start_POSTSUBSCRIPT italic_s end_POSTSUBSCRIPT ( over→ start_ARG italic_x end_ARG ) = ⟨ italic_v start_POSTSUBSCRIPT italic_k end_POSTSUBSCRIPT ( over→ start_ARG italic_x end_ARG ) ⟩ for italic_k = { 1 , … , 6 } (1)

where vk(x;s)subscript𝑣𝑘𝑥𝑠v_{k}(\vec{x};s)italic_v start_POSTSUBSCRIPT italic_k end_POSTSUBSCRIPT ( over→ start_ARG italic_x end_ARG ; italic_s ) is the force on coordinate k𝑘kitalic_k when the system is in Markov state s𝑠sitalic_s. (Note, we have concatenated the x𝑥xitalic_x and y𝑦yitalic_y coordinates of each bead into a single vector x𝑥\vec{x}over→ start_ARG italic_x end_ARG.) This removes the stochasticity and results in a deterministic effective force for the system, vdelimited-⟨⟩𝑣\langle\vec{v}\rangle⟨ over→ start_ARG italic_v end_ARG ⟩.

The stable fixed points of (1) are the observed clusters. One fixed point is a 3-bead cluster in which the beads are in a small triangle configuration; see Fig. 2C. This is distinguished from the unbound state s=1𝑠1s=1italic_s = 1, in which the beads form a triangle but with larger pairwise distances; see Fig. 2A. The other three fixed points are 2-bead clusters in which two beads are superimposed and the third is farther away and unbound; see Fig. 2B. These are distinguished from states s=2,3,4𝑠234s=2,3,4italic_s = 2 , 3 , 4 as the clusters persist longer than any single bond lifetime.

In addition to predicting the existence of clusters, we can test this time-averaging procedure’s ability to predict the lifetime of a cluster. We do this by considering the effects of small perturbations of noise about the average by studying the equation,

dxk=vk(x)dt+2ϵdBk𝑑subscript𝑥𝑘delimited-⟨⟩subscript𝑣𝑘𝑥𝑑𝑡2italic-ϵ𝑑subscript𝐵𝑘dx_{k}=\langle v_{k}(\vec{x})\rangle dt+\sqrt{2\epsilon}\;dB_{k}italic_d italic_x start_POSTSUBSCRIPT italic_k end_POSTSUBSCRIPT = ⟨ italic_v start_POSTSUBSCRIPT italic_k end_POSTSUBSCRIPT ( over→ start_ARG italic_x end_ARG ) ⟩ italic_d italic_t + square-root start_ARG 2 italic_ϵ end_ARG italic_d italic_B start_POSTSUBSCRIPT italic_k end_POSTSUBSCRIPT (2)

in the limit as ϵ0italic-ϵ0\epsilon\to 0italic_ϵ → 0. Even though the vector of forces is the gradient of a potential function for each state s𝑠sitalic_s, i.e. v(x;s)=xU(x;s)𝑣𝑥𝑠subscript𝑥𝑈𝑥𝑠\vec{v}(\vec{x};s)=-\nabla_{x}U(\vec{x};s)over→ start_ARG italic_v end_ARG ( over→ start_ARG italic_x end_ARG ; italic_s ) = - ∇ start_POSTSUBSCRIPT italic_x end_POSTSUBSCRIPT italic_U ( over→ start_ARG italic_x end_ARG ; italic_s ), it is not true that v(x)=xU(x)delimited-⟨⟩𝑣𝑥subscript𝑥𝑈𝑥\langle\vec{v}(\vec{x})\rangle=-\nabla_{x}\langle U(\vec{x})\rangle⟨ over→ start_ARG italic_v end_ARG ( over→ start_ARG italic_x end_ARG ) ⟩ = - ∇ start_POSTSUBSCRIPT italic_x end_POSTSUBSCRIPT ⟨ italic_U ( over→ start_ARG italic_x end_ARG ) ⟩ since the CTMC null vector r𝑟\vec{r}over→ start_ARG italic_r end_ARG depends on the positions x𝑥\vec{x}over→ start_ARG italic_x end_ARG. However, we numerically find an effective potential Ueffsubscript𝑈effU_{\text{eff}}italic_U start_POSTSUBSCRIPT eff end_POSTSUBSCRIPT such that xUeff(x)v(x)subscript𝑥subscript𝑈eff𝑥delimited-⟨⟩𝑣𝑥-\nabla_{x}U_{\text{eff}}(\vec{x})\approx\langle\vec{v}(\vec{x})\rangle- ∇ start_POSTSUBSCRIPT italic_x end_POSTSUBSCRIPT italic_U start_POSTSUBSCRIPT eff end_POSTSUBSCRIPT ( over→ start_ARG italic_x end_ARG ) ≈ ⟨ over→ start_ARG italic_v end_ARG ( over→ start_ARG italic_x end_ARG ) ⟩ along the most probable transition path connecting a minimum of Ueffsubscript𝑈effU_{\text{eff}}italic_U start_POSTSUBSCRIPT eff end_POSTSUBSCRIPT (recall these are the stable fixed points of (1)) to a saddle point using the String Method [5]. We are therefore justified in asymptotically approximating the escape times from a cluster using the well-known Arrhenius law,

log(𝔼[τ])ΔUeffϵsimilar-to𝔼delimited-[]𝜏Δsubscript𝑈effitalic-ϵ\log(\mathbb{E}[\tau])\sim\frac{\Delta U_{\text{eff}}}{\epsilon}roman_log ( blackboard_E [ italic_τ ] ) ∼ divide start_ARG roman_Δ italic_U start_POSTSUBSCRIPT eff end_POSTSUBSCRIPT end_ARG start_ARG italic_ϵ end_ARG (3)

in the limit as ϵ0italic-ϵ0\epsilon\to 0italic_ϵ → 0, where ΔUeffΔsubscript𝑈eff\Delta U_{\text{eff}}roman_Δ italic_U start_POSTSUBSCRIPT eff end_POSTSUBSCRIPT is the change in the effective potential generated by the time-averaged force along the most probable transition path [29].

Refer to caption
Figure 2: (A-C) plots of pairwise distances of three beads over time and a snapshot of the three bead configuration: (A) all beads unbound: balance between excluded volume and confinement forces keep beads in a large triangle configuration, (B) 2-bead cluster state with beads 1-2 bound, (C) 3-bead cluster state: all beads remain close, rapid switching between pairwise bonds keeps beads in a small triangle configuration.

Despite predicting the existence of clusters, the naive time-averaged force does not produce an effective potential Ueffsubscript𝑈effU_{\text{eff}}italic_U start_POSTSUBSCRIPT eff end_POSTSUBSCRIPT that accounts for the observed differences in the cluster lifetime as the crosslink binding timescale β𝛽\betaitalic_β varies in the 361-bead model. The time-averaged force, and thus its effective potential, is invariant to the overall timescale of the transition rate matrix. Scaling S𝑆Sitalic_S by some constant α𝛼\alphaitalic_α does not affect the fraction of time spent in each state because r𝑟\vec{r}over→ start_ARG italic_r end_ARG is also in the null space of αS𝛼𝑆\alpha Sitalic_α italic_S. Thus, Eqs. (1) and (2) are unchanged when scaling the transition rate matrix S𝑆Sitalic_S by α𝛼\alphaitalic_α and therefore cannot predict subsequent changes to the dynamic cluster lifetimes.

II.3 Relative scalings of fluctuations predicts three different clustering regimes

The key to understanding clustering dynamics is accounting for fluctuations from both the stochastic binding and the small perturbations of thermal noise. To analyze the dynamics mathematically we perform an asymptotic expansion as the size of the fluctuations goes to zero. We consider taking both fluctuations to zero at the same rate and at different rates. The latter allows one noise source to dominate. Rigid clusters arise when the binding fluctuations are small and only thermal noise drives transitions between clusters. Amorphic arrangements arise when the binding fluctuations are large and take place on a longer timescale than the fast thermal fluctuations. This binding drives transitions between clusters. Taking fluctuations to zero at the same rate allows for interaction between the noises. We find that this interaction is required to predict the flexible cluster dynamics that depend on the timescale of the binding. Together, these three asymptotic regimes explain the three regimes seen in [26] and Fig. 1.

We include the effects of fluctuations about the mean force so that the position of each bead follows the overdamped Langevin-like equation given by

dxk=vk(x;s)dt+2ϵdBk.𝑑subscript𝑥𝑘subscript𝑣𝑘𝑥𝑠𝑑𝑡2italic-ϵ𝑑subscript𝐵𝑘dx_{k}=v_{k}(\vec{x};s)dt+\sqrt{2\epsilon}dB_{k}.italic_d italic_x start_POSTSUBSCRIPT italic_k end_POSTSUBSCRIPT = italic_v start_POSTSUBSCRIPT italic_k end_POSTSUBSCRIPT ( over→ start_ARG italic_x end_ARG ; italic_s ) italic_d italic_t + square-root start_ARG 2 italic_ϵ end_ARG italic_d italic_B start_POSTSUBSCRIPT italic_k end_POSTSUBSCRIPT . (4)

where s𝑠sitalic_s is the state of the CTMC. How we choose to scale the transition rate matrix S𝑆Sitalic_S determines how the relative rates of the two different fluctuations go to zero.

We start with the scaling 1ϵ2S1superscriptitalic-ϵ2𝑆\frac{1}{\epsilon^{2}}Sdivide start_ARG 1 end_ARG start_ARG italic_ϵ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG italic_S which takes the binding fluctuations to zero faster than the perturbations from thermal noise. We seek an effective potential W𝑊Witalic_W to replace Ueffsubscript𝑈effU_{\text{eff}}italic_U start_POSTSUBSCRIPT eff end_POSTSUBSCRIPT in Eq. (3). Therefore we assume a steady-state distribution

ps(x)=rs(x)exp(1ϵW(x))subscript𝑝𝑠𝑥subscript𝑟𝑠𝑥1italic-ϵ𝑊𝑥p_{s}(\vec{x})=r_{s}(\vec{x})\exp\bigg{(}-\frac{1}{\epsilon}W(\vec{x})\bigg{)}italic_p start_POSTSUBSCRIPT italic_s end_POSTSUBSCRIPT ( over→ start_ARG italic_x end_ARG ) = italic_r start_POSTSUBSCRIPT italic_s end_POSTSUBSCRIPT ( over→ start_ARG italic_x end_ARG ) roman_exp ( - divide start_ARG 1 end_ARG start_ARG italic_ϵ end_ARG italic_W ( over→ start_ARG italic_x end_ARG ) ) (5)

as the asymptotic solution for the system of steady-state Fokker-Planck equations,

pst=k=16xk[vk(x;s)ps]+ϵk=162xk2[ps]+αϵ2j=14Ssjpj.subscript𝑝𝑠𝑡superscriptsubscript𝑘16subscript𝑥𝑘delimited-[]subscript𝑣𝑘𝑥𝑠subscript𝑝𝑠italic-ϵsuperscriptsubscript𝑘16superscript2superscriptsubscript𝑥𝑘2delimited-[]subscript𝑝𝑠𝛼superscriptitalic-ϵ2superscriptsubscript𝑗14subscript𝑆𝑠𝑗subscript𝑝𝑗\frac{\partial p_{s}}{\partial t}=-\sum_{k=1}^{6}\frac{\partial}{\partial x_{k% }}[v_{k}(\vec{x};s)p_{s}]+\epsilon\sum_{k=1}^{6}\frac{\partial^{2}}{\partial x% _{k}^{2}}[p_{s}]+\frac{\alpha}{\epsilon^{2}}\sum_{j=1}^{4}S_{sj}p_{j}.divide start_ARG ∂ italic_p start_POSTSUBSCRIPT italic_s end_POSTSUBSCRIPT end_ARG start_ARG ∂ italic_t end_ARG = - ∑ start_POSTSUBSCRIPT italic_k = 1 end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 6 end_POSTSUPERSCRIPT divide start_ARG ∂ end_ARG start_ARG ∂ italic_x start_POSTSUBSCRIPT italic_k end_POSTSUBSCRIPT end_ARG [ italic_v start_POSTSUBSCRIPT italic_k end_POSTSUBSCRIPT ( over→ start_ARG italic_x end_ARG ; italic_s ) italic_p start_POSTSUBSCRIPT italic_s end_POSTSUBSCRIPT ] + italic_ϵ ∑ start_POSTSUBSCRIPT italic_k = 1 end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 6 end_POSTSUPERSCRIPT divide start_ARG ∂ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG start_ARG ∂ italic_x start_POSTSUBSCRIPT italic_k end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG [ italic_p start_POSTSUBSCRIPT italic_s end_POSTSUBSCRIPT ] + divide start_ARG italic_α end_ARG start_ARG italic_ϵ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG ∑ start_POSTSUBSCRIPT italic_j = 1 end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 4 end_POSTSUPERSCRIPT italic_S start_POSTSUBSCRIPT italic_s italic_j end_POSTSUBSCRIPT italic_p start_POSTSUBSCRIPT italic_j end_POSTSUBSCRIPT . (6)

The element of the CTMC transition matrix Ssjsubscript𝑆𝑠𝑗S_{sj}italic_S start_POSTSUBSCRIPT italic_s italic_j end_POSTSUBSCRIPT gives the transition rates from state j𝑗jitalic_j into state s𝑠sitalic_s. Thus, the last term couples the process between different states. Setting (6) to zero and plugging in (5) yields the leading order equation

Sr=0𝑆𝑟0S\vec{r}=0italic_S over→ start_ARG italic_r end_ARG = 0 (7)

with next order equation

rsvxW+rsxWxW=0.subscript𝑟𝑠𝑣subscript𝑥𝑊subscript𝑟𝑠subscript𝑥𝑊subscript𝑥𝑊0r_{s}\vec{v}\cdot\nabla_{x}W+r_{s}\nabla_{x}W\cdot\nabla_{x}W=0.italic_r start_POSTSUBSCRIPT italic_s end_POSTSUBSCRIPT over→ start_ARG italic_v end_ARG ⋅ ∇ start_POSTSUBSCRIPT italic_x end_POSTSUBSCRIPT italic_W + italic_r start_POSTSUBSCRIPT italic_s end_POSTSUBSCRIPT ∇ start_POSTSUBSCRIPT italic_x end_POSTSUBSCRIPT italic_W ⋅ ∇ start_POSTSUBSCRIPT italic_x end_POSTSUBSCRIPT italic_W = 0 . (8)

Thus, r𝑟\vec{r}over→ start_ARG italic_r end_ARG is the steady-state distribution of S𝑆Sitalic_S and summing Eq. (8) over s𝑠sitalic_s reveals that xW=sv(x;s)rs(x)subscript𝑥𝑊subscript𝑠𝑣𝑥𝑠subscript𝑟𝑠𝑥-\nabla_{x}W=\sum_{s}\vec{v}(\vec{x};s)r_{s}(\vec{x})- ∇ start_POSTSUBSCRIPT italic_x end_POSTSUBSCRIPT italic_W = ∑ start_POSTSUBSCRIPT italic_s end_POSTSUBSCRIPT over→ start_ARG italic_v end_ARG ( over→ start_ARG italic_x end_ARG ; italic_s ) italic_r start_POSTSUBSCRIPT italic_s end_POSTSUBSCRIPT ( over→ start_ARG italic_x end_ARG ) is exactly the naive time-average from above that does not depend on α𝛼\alphaitalic_α.

If instead we take the scaling 1ϵS1italic-ϵ𝑆\frac{1}{\epsilon}Sdivide start_ARG 1 end_ARG start_ARG italic_ϵ end_ARG italic_S, both sources of noise show up in the leading order equation given by

[A(x,xW)+D(xW)+αS(x)]r=0delimited-[]𝐴𝑥subscript𝑥𝑊𝐷subscript𝑥𝑊𝛼𝑆𝑥𝑟0[A(\vec{x},\nabla_{x}W)+D(\nabla_{x}W)+\alpha S(\vec{x})]\vec{r}=0[ italic_A ( over→ start_ARG italic_x end_ARG , ∇ start_POSTSUBSCRIPT italic_x end_POSTSUBSCRIPT italic_W ) + italic_D ( ∇ start_POSTSUBSCRIPT italic_x end_POSTSUBSCRIPT italic_W ) + italic_α italic_S ( over→ start_ARG italic_x end_ARG ) ] over→ start_ARG italic_r end_ARG = 0 (9)

for advection (drift) matrix A𝐴Aitalic_A, diffusion matrix D𝐷Ditalic_D, and switching rate matrix S𝑆Sitalic_S; see [27] for details. In this regime, both sources of noise are important for flexible clustering. We see the effects of varying the binding timescale parameter α𝛼\alphaitalic_α, which controls how quickly the simulation switches between the Markov chain states. This role mirrors that of the kinetic timescale parameter μ𝜇\muitalic_μ in the large-scale simulations of [12, 26] in which μ𝜇\muitalic_μ is a “tuning knob” used to set the kinetic timescale on which the crosslinks bind and unbind.

To further illustrate the changing dynamics with α𝛼\alphaitalic_α, we compare the predicted lifetime of the cluster governed by Eq. (3) with Monte-Carlo simulations. The asymptotic escape times from one 2-bead cluster to another 2-bead cluster are shown in Fig. 3. The slope value is given by the quasipotential barrier height and well expresses the linear relationship of the mean escape times. As α𝛼\alphaitalic_α increases, the effective energy barrier approaches the barrier predicted by the naive time average discussed previously.

Refer to caption
Figure 3: Transition: 2-bead cluster to 2-bead cluster. (a) Quasipotential (solid) for different α𝛼\alphaitalic_α values and for the time-averaged binding force (dotted) along the transition path, normalized to unit length. (b) Comparison of asymptotic escape times computed via Monte Carlo simulation to slope taken from quasipotential barrier height. (c) Quasipotential energy barrier height vs. switching timescale α𝛼\alphaitalic_α.

Finally, if we keep S𝑆Sitalic_S independent of ϵitalic-ϵ\epsilonitalic_ϵ, retaining all the fluctuations, the system equilibrates within each Markov chain state and there are no clusters predicted. In this regime, the effects of the binding noise are pushed into a higher order so that the leading order does not contain any contributions from the CTMC. We observe the beads are often unbound, settling into a “large triangle” configuration; see Fig. 2A. This long-observed state of the system is a balance between the excluded volume and the confinement forces. We do not consider this a cluster as it would dissolve in the absence of the confinement forces. Fig. 4 shows the formation of clusters as α𝛼\alphaitalic_α increases. The system is able to remain in a cluster state despite all the beads being unbound as the system does not have time between binding events to reach the large triangle equilibrium.

Refer to caption
Figure 4: Average proportion of simulation spent in s=1𝑠1s=1italic_s = 1 unbound state of the Markov chain and average proportion of simulation not spent in a cluster state vs. α𝛼\alphaitalic_α.

II.4 Effects of crosslink binding fluctuations on different transitions

Scaling S𝑆Sitalic_S by 1ϵ1italic-ϵ\frac{1}{\epsilon}divide start_ARG 1 end_ARG start_ARG italic_ϵ end_ARG revealed the dependence of cluster lifetimes on the crosslink binding timescale, α𝛼\alphaitalic_α. By accounting for both sources of randomness in the system we computed an accurate quasipotential that predicted a higher energy barrier height between minimum states as α𝛼\alphaitalic_α increased, and thus longer cluster lifetimes. However, this increase is not uniform across all transitions as we show in Fig. 5.

Of the ‘uphill’ most probable transition paths the system takes from one minimum state to the saddle point shown in Fig. 5A-D, only the 3-bead on its way to a 2-bead cluster has noticeable dependence on α𝛼\alphaitalic_α (Fig. 5C). This transition also has the largest change in quasipotential energy barrier with α𝛼\alphaitalic_α (Fig. 5F) indicating removing a bead from a cluster is more sensitive to the binding timescale that rearrangements within the cluster (pathway in Fig. 5D). The transition shown in Fig. 5D to the “collinear” saddle point configuration also has the lowest energy barrier height amongst the transitions, with limited dependence on α𝛼\alphaitalic_α for either the pathway or the height. This indicates that the binding force is not important for this transition, as the stochasticity of the binding does not help the system make the transition.

Starting in the 2-bead cluster state, our system can transition into a different 2-bead cluster (Fig. 5A) or into the “small triangle” 3-bead cluster (Fig. 5B). These transitions have similar pathways up to each saddle point that do not significantly depend on α𝛼\alphaitalic_α, and have similar quasipotential energy barrier heights for the range of α𝛼\alphaitalic_α values from 0 to 20. This suggests that the system will take either transition with similar probability and do so in approximately the same amount of time.

Refer to caption
Figure 5: (A-D) Transition pathways to a saddle point: (A) 2-bead cluster on the way to a different 2-bead cluster (B) 2-bead cluster on the way to a 3-bead cluster, (C) 3-bead cluster on the way to 2-bead cluster for α=1𝛼1\alpha=1italic_α = 1, α=10𝛼10\alpha=10italic_α = 10, and time-averaged binding force (α𝛼\alpha\to\inftyitalic_α → ∞), (D) 3-bead cluster to the collinear saddle point configuration that continues to a rearrangement of the 3-bead cluster, or to a different saddle point on the way to a 2-bead cluster. (E,F) Energy Barrier heights vs. α𝛼\alphaitalic_α for these transition pathways

III Discussion

The underlying mathematical mechanism behind the observed chromatin clustering behaviors is the competition between thermal fluctuations and the timescale of the model condensin crosslinking force. This timescale controls the production of random fluctuations in the crosslink binding force and in turn the dominating terms in the asymptotic perturbation when searching for an effective thermal equilibrium. This analysis emphasizes the importance of accounting for stochasticity in local incoherent transient forces to predict emergent complex biological behavior.

The key to understanding flexible clustering dynamics is accounting for fluctuations from both the stochastic binding force and the thermal noise, taking both fluctuations to zero at the same rate. To produce this interaction mathematically, we scale the binding rate matrix S𝑆Sitalic_S by 1ϵ1italic-ϵ\frac{1}{\epsilon}divide start_ARG 1 end_ARG start_ARG italic_ϵ end_ARG. Taking ϵ0italic-ϵ0\epsilon\to 0italic_ϵ → 0 with this scaling, an effective binding force remains that allows for stable clusters. This limit also accounts for the fact that the fluctuations of the binding force aid in transitions between clusters, as it creates finite periods of time when the thermal noise has to overcome smaller forces to push the system into a new cluster state. Since our added binding timescale parameter α𝛼\alphaitalic_α controls the size of the fluctuations, it too controls the size of the effective energy barrier for predicting transition times between stable cluster states given by the Arrhenius law.

If we take the fluctuations to zero at different rates, one source of noise “dominates” over the other in the limit as ϵ0italic-ϵ0\epsilon\to 0italic_ϵ → 0. Choosing to scale S𝑆Sitalic_S by 1ϵ21superscriptitalic-ϵ2\frac{1}{\epsilon^{2}}divide start_ARG 1 end_ARG start_ARG italic_ϵ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG, the fluctuations of the binding force go to zero faster than the thermal fluctuations. An effective binding force remains that allows for stable clusters, but the fluctuations in the binding force are now much smaller and do not interact with the thermal noise to aid in transitions between clusters. Indeed, we recover the naive time average to describe the effective energy barrier, which is also the limiting barrier as α𝛼\alpha\to\inftyitalic_α → ∞ of the above-mentioned fluctuation-interaction case.

If we flip which fluctuation source dominates in the limit as ϵ0italic-ϵ0\epsilon\to 0italic_ϵ → 0 by keeping S𝑆Sitalic_S independent of ϵitalic-ϵ\epsilonitalic_ϵ, switching between Markov chain states occurs so infrequently that the system equilibriates while in this one state. No effective force is generated by the superposition of different states, thus no clusters are predicted. In the 3-bead model, two cluster-like states appear but their generating mechanism would not produce clusters in the 361-bead model. The first is a 3-bead triangle configuration, but it is a result of the balance between the repulsive excluded volume force and confinement force. In the 361-bead model, this would correspond to all beads in an amorphic arrangement. The second is a 2-bead-like cluster formed by the fact that a crosslink binds two beads; this cluster returns to the 3-bead triangle configuration once the bond is broken. In the 361-bead model, this would correspond to another amorphic arrangement with many pair-wise bound beads.

The arrangement of beads within clusters provides a physical model for the dynamics of gene clusters within the nucleus. Gene clustering is a mechanism that facilitates the co-regulation of un-linked genes or of long linear arrays of repeated genes such as rDNA. The ability to reconfigure the network is central to the cell’s ability to re-wire its transcriptional circuitry. Using a relatively small number of beads to reveal root mechanisms, we account for both state changes and dynamics between states with thermal fluctuations and cross-linking forces. The energy barrier for rearrangements within a cluster is lower than that required for exchanges between clusters (Fig. 5E, F). The system is remarkably plastic, with small bursts of energy able to potentiate new configurations depending on the biological demand.

The emergence of a wide range of clustering regimes (amorphic, flexible, rigid)(Fig. 1) and exchange within and between clusters with a pared-down 3-bead model (Fig. 5) provides a framework for understanding the governing principles of genome organization: The continuous chain of each chromosome biases their individualization. The crosslinking forces provide a mechanism to build gene (bead) clusters and generate informational circuits that are agnostic to the position of a gene in a given chromosome. This work shows that by controlling the timescale of the crosslinking forces, the plasticity of the gene clusters can be tuned based on biological needs. The predictions of cluster dynamics for larger systems will depend greatly on thermodynamics and tuning the crosslinking forces. This model and analysis provide critical insight into the diversity of network configurations with a minimal set of parameters.

Computationally, our analysis is limited to relatively small numbers of beads due to the need to enumerate all possible pairwise bound states for the Markov chain. With the growing power of machine learning, it is likely possible to build a physics-informed neural network [14, 15] to learn the effective potential. This would allow predictions of cluster dynamics for larger systems but would need to be re-applied for each timescale or set of model parameters. The effective potential alone gives little insight into the mechanism underlying the emergent behavior that our analysis of the 3-bead model has provided. This new outlook on the importance of noise at the proper timescale aids in deepening our understanding of life at the cellular level.

IV Methods

The idealized chromatin model is based on the polymer-like chain of beads model in [12, 26]. The position of bead i𝑖iitalic_i is given by the overdamped Langevin equation

d𝐱i=(𝐟ci+𝐟EVi+𝐟bondi)dt+2ϵd𝐁.𝑑subscript𝐱𝑖superscriptsubscript𝐟𝑐𝑖superscriptsubscript𝐟EV𝑖superscriptsubscript𝐟bond𝑖𝑑𝑡2italic-ϵ𝑑𝐁d\mathbf{x}_{i}=(\mathbf{f}_{c}^{i}+\mathbf{f}_{\text{EV}}^{i}+\mathbf{f}_{% \text{bond}}^{i})dt+\sqrt{2\epsilon}d\mathbf{B}.italic_d bold_x start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT = ( bold_f start_POSTSUBSCRIPT italic_c end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_i end_POSTSUPERSCRIPT + bold_f start_POSTSUBSCRIPT EV end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_i end_POSTSUPERSCRIPT + bold_f start_POSTSUBSCRIPT bond end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_i end_POSTSUPERSCRIPT ) italic_d italic_t + square-root start_ARG 2 italic_ϵ end_ARG italic_d bold_B . (10)

The parameter ϵitalic-ϵ\epsilonitalic_ϵ scales the vector of Brownian increments d𝐁𝑑𝐁d\mathbf{B}italic_d bold_B preparing for asymptotic analysis to take the fluctuations to zero. The deterministic confinement force,

𝐟ci=η𝐱i,superscriptsubscript𝐟𝑐𝑖𝜂subscript𝐱𝑖\mathbf{f}_{c}^{i}=-\eta\mathbf{x}_{i},bold_f start_POSTSUBSCRIPT italic_c end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_i end_POSTSUPERSCRIPT = - italic_η bold_x start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT ,

replaces the hard-wall constraint of the nucleus membrane while the excluded volume force,

𝐟EVi=jiaev(𝐱i𝐱j)exp(|𝐱i𝐱j|2cev),superscriptsubscript𝐟EV𝑖subscript𝑗𝑖subscript𝑎𝑒𝑣subscript𝐱𝑖subscript𝐱𝑗superscriptsubscript𝐱𝑖subscript𝐱𝑗2subscript𝑐𝑒𝑣\mathbf{f}_{\text{EV}}^{i}=\sum_{j\neq i}a_{ev}(\mathbf{x}_{i}-\mathbf{x}_{j})% \exp\bigg{(}-\frac{|\mathbf{x}_{i}-\mathbf{x}_{j}|^{2}}{c_{ev}}\bigg{)},bold_f start_POSTSUBSCRIPT EV end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_i end_POSTSUPERSCRIPT = ∑ start_POSTSUBSCRIPT italic_j ≠ italic_i end_POSTSUBSCRIPT italic_a start_POSTSUBSCRIPT italic_e italic_v end_POSTSUBSCRIPT ( bold_x start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT - bold_x start_POSTSUBSCRIPT italic_j end_POSTSUBSCRIPT ) roman_exp ( - divide start_ARG | bold_x start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT - bold_x start_POSTSUBSCRIPT italic_j end_POSTSUBSCRIPT | start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG start_ARG italic_c start_POSTSUBSCRIPT italic_e italic_v end_POSTSUBSCRIPT end_ARG ) ,

remains similar to its form in [12, 26]. Parameters η𝜂\etaitalic_η, aevsubscript𝑎𝑒𝑣a_{ev}italic_a start_POSTSUBSCRIPT italic_e italic_v end_POSTSUBSCRIPT, and cevsubscript𝑐𝑒𝑣c_{ev}italic_c start_POSTSUBSCRIPT italic_e italic_v end_POSTSUBSCRIPT are given in Table 1 for both the 361-bead (𝐱i3,i=1361)formulae-sequencesubscript𝐱𝑖superscript3𝑖1361(\mathbf{x}_{i}\in\mathbb{R}^{3},i=1\dots 361)( bold_x start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT ∈ blackboard_R start_POSTSUPERSCRIPT 3 end_POSTSUPERSCRIPT , italic_i = 1 … 361 ) and the 3-bead (𝐱i2,i=13)formulae-sequencesubscript𝐱𝑖superscript2𝑖13(\mathbf{x}_{i}\in\mathbb{R}^{2},i=1\dots 3)( bold_x start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT ∈ blackboard_R start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT , italic_i = 1 … 3 ) models. Note we have neglected the spring force linking the beads together to form a chain, as we show it is secondary to the clustering dynamics.

The stochastic binding force,

𝐟bondi=jiκbij(𝐱j𝐱i),superscriptsubscript𝐟bond𝑖subscript𝑗𝑖𝜅subscript𝑏𝑖𝑗subscript𝐱𝑗subscript𝐱𝑖\mathbf{f}_{\text{bond}}^{i}=\sum_{j\neq i}\kappa b_{ij}(\mathbf{x}_{j}-% \mathbf{x}_{i}),bold_f start_POSTSUBSCRIPT bond end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_i end_POSTSUPERSCRIPT = ∑ start_POSTSUBSCRIPT italic_j ≠ italic_i end_POSTSUBSCRIPT italic_κ italic_b start_POSTSUBSCRIPT italic_i italic_j end_POSTSUBSCRIPT ( bold_x start_POSTSUBSCRIPT italic_j end_POSTSUBSCRIPT - bold_x start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT ) ,

models the binding SMC proteins found in the biological system. This force binds two beads, corresponding to bij=1subscript𝑏𝑖𝑗1b_{ij}=1italic_b start_POSTSUBSCRIPT italic_i italic_j end_POSTSUBSCRIPT = 1 if beads i𝑖iitalic_i and j𝑗jitalic_j are bound and 0 otherwise. Each bead can be bound to only one other bead; these stochastic bonds form and break at exponentially distributed times with a binding-rate proportional to the bead separation distance, a(𝐱i𝐱j)𝑎subscript𝐱𝑖subscript𝐱𝑗a(\mathbf{x}_{i}-\mathbf{x}_{j})italic_a ( bold_x start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT - bold_x start_POSTSUBSCRIPT italic_j end_POSTSUBSCRIPT ), and constant breaking-rate c𝑐citalic_c. The time evolution of the states follows a general CTMC process; for the 3-bead model the transition rate matrix is given by

S=(bccca(𝐱1𝐱2)c00a(𝐱1𝐱3)0c0a(𝐱2𝐱3)00c)𝑆matrix𝑏𝑐𝑐𝑐𝑎subscript𝐱1subscript𝐱2𝑐00𝑎subscript𝐱1subscript𝐱30𝑐0𝑎subscript𝐱2subscript𝐱300𝑐S=\begin{pmatrix}b&c&c&c\\ a(\mathbf{x}_{1}-\mathbf{x}_{2})&-c&0&0\\ a(\mathbf{x}_{1}-\mathbf{x}_{3})&0&-c&0\\ a(\mathbf{x}_{2}-\mathbf{x}_{3})&0&0&-c\end{pmatrix}italic_S = ( start_ARG start_ROW start_CELL italic_b end_CELL start_CELL italic_c end_CELL start_CELL italic_c end_CELL start_CELL italic_c end_CELL end_ROW start_ROW start_CELL italic_a ( bold_x start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT - bold_x start_POSTSUBSCRIPT 2 end_POSTSUBSCRIPT ) end_CELL start_CELL - italic_c end_CELL start_CELL 0 end_CELL start_CELL 0 end_CELL end_ROW start_ROW start_CELL italic_a ( bold_x start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT - bold_x start_POSTSUBSCRIPT 3 end_POSTSUBSCRIPT ) end_CELL start_CELL 0 end_CELL start_CELL - italic_c end_CELL start_CELL 0 end_CELL end_ROW start_ROW start_CELL italic_a ( bold_x start_POSTSUBSCRIPT 2 end_POSTSUBSCRIPT - bold_x start_POSTSUBSCRIPT 3 end_POSTSUBSCRIPT ) end_CELL start_CELL 0 end_CELL start_CELL 0 end_CELL start_CELL - italic_c end_CELL end_ROW end_ARG ) (11)

with b=a(𝐱1𝐱2)a(𝐱1𝐱3)a(𝐱2𝐱3)𝑏𝑎subscript𝐱1subscript𝐱2𝑎subscript𝐱1subscript𝐱3𝑎subscript𝐱2subscript𝐱3b=-a(\mathbf{x}_{1}-\mathbf{x}_{2})-a(\mathbf{x}_{1}-\mathbf{x}_{3})-a(\mathbf% {x}_{2}-\mathbf{x}_{3})italic_b = - italic_a ( bold_x start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT - bold_x start_POSTSUBSCRIPT 2 end_POSTSUBSCRIPT ) - italic_a ( bold_x start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT - bold_x start_POSTSUBSCRIPT 3 end_POSTSUBSCRIPT ) - italic_a ( bold_x start_POSTSUBSCRIPT 2 end_POSTSUBSCRIPT - bold_x start_POSTSUBSCRIPT 3 end_POSTSUBSCRIPT ).

The affinity function a(𝐱)𝑎𝐱a(\mathbf{x})italic_a ( bold_x ) and parameter values are given in Table 1 for both the 361-bead and the 3-bead models. Both a(𝐱)𝑎𝐱a(\mathbf{x})italic_a ( bold_x ) and c𝑐citalic_c are scaled by β𝛽\betaitalic_β for the 361-bead model and α𝛼\alphaitalic_α for the 3-bead model to explore the kinetic timescale on which the crosslinks bind and unbind, mirroring the kinetic timescale parameter μ𝜇\muitalic_μ in the large-scale simulations of [26]. Both a(𝐱)𝑎𝐱a(\mathbf{x})italic_a ( bold_x ) and c𝑐citalic_c are further scaled by different powers of ϵitalic-ϵ\epsilonitalic_ϵ to perform the asymptotic analysis of the fluctuations.

Table 1: Model parameters and formula.
361-Bead Model 3-Bead Model
a(𝐱)𝑎𝐱a(\mathbf{x})italic_a ( bold_x ) 21+exp(20(|𝐱|75))2120𝐱75\displaystyle\frac{2}{1+\exp({20(|\mathbf{x}|-75)})}divide start_ARG 2 end_ARG start_ARG 1 + roman_exp ( 20 ( | bold_x | - 75 ) ) end_ARG 21+exp(20(|𝐱|0.75))2120𝐱0.75\displaystyle\frac{2}{1+\exp({20(|\mathbf{x}|-0.75))}}divide start_ARG 2 end_ARG start_ARG 1 + roman_exp ( 20 ( | bold_x | - 0.75 ) ) end_ARG
η𝜂\etaitalic_η 0.002 1
aevsubscript𝑎𝑒𝑣a_{ev}italic_a start_POSTSUBSCRIPT italic_e italic_v end_POSTSUBSCRIPT 0.0332 2
cevsubscript𝑐𝑒𝑣c_{ev}italic_c start_POSTSUBSCRIPT italic_e italic_v end_POSTSUBSCRIPT 30600 0.5
κ𝜅\kappaitalic_κ 10.9 5
c𝑐citalic_c 0.01 0.5

Monte Carlo Simulations

We perform Monte Carlo simulations of the 3-bead system to compare the scaling of the escape times with ϵitalic-ϵ\epsilonitalic_ϵ to the effective energy barriers predicted by the theoretical calculations. The simulations are started in either the 2-bead or 3-bead cluster, and continued until a stopping condition is met, indicating that the system has left the basin of attraction of the initial state. From the 2-bead cluster, we look for either a different 2-bead cluster or the 3-bead triangle cluster. From the 3-bead triangle cluster, we look for either a 2-bead cluster or a collinear configuration that is the saddle point between different arrangements of the 3-bead triangle cluster. The mean escape time, τ𝜏\tauitalic_τ, is computed by the maximum likelihood estimate that divides the sum of all escape times by the number executing the desired transition.

Code for reproducing results is available on GitHub [1].

Acknowledgements.
AC and KN were partially supported by National Science Foundation (NSF) grants DMS-1816394 and DMS-2307297 and AC was also partially supported by NSF grant DMS-1929298. KB was partially supported by the National Institutes of Health (NIH) grant R01 GM32238.

References

  • [1] bwalker1/quasi-string-reprod: Quasipotential Strings v1.
  • [2] Jason C. Bell and Stephen C. Kowalczykowski. RecA: Regulation and Mechanism of a Molecular Search Engine. Trends in Biochemical Sciences, 41(6):491–507, June 2016.
  • [3] Kerry Bloom and Elaine Yeh. Tension Management in the Kinetochore. Current Biology, 20(23):R1040–R1048, December 2010.
  • [4] Alyson J. Conover, Claudia Danilowicz, Ruwan Gunaratne, Vincent W. Coljee, Nancy Kleckner, and Mara Prentiss. Changes in the tension in dsDNA alter the conformation of RecA bound to dsDNA–RecA filaments. Nucleic Acids Research, 39(20):8833–8843, November 2011.
  • [5] Weinan E, Weiqing Ren, and Eric Vanden-Eijnden. String method for the study of rare events. Physical Review B, 66(5):052301, August 2002.
  • [6] John F Marko Elnaz Alipour. Self-organization of domain structures by DNA-loop-extruding enzymes. October 2012.
  • [7] E. Feinstein, C. Danilowicz, A. Conover, R. Gunaratne, N. Kleckner, and M. Prentiss. Single-molecule studies of the stringency factors and rates governing the polymerization of RecA on double-stranded DNA. Nucleic Acids Research, 39(9):3781–3791, May 2011.
  • [8] Geoffrey Fudenberg and Leonid A. Mirny. Higher-order chromatin structure: bridging physics and biology. Current Opinion in Genetics & Development, 22(2):115–124, April 2012.
  • [9] Mahipal Ganji, Indra A. Shaltiel, Shveta Bisht, Eugene Kim, Ana Kalichava, Christian H. Haering, and Cees Dekker. Real-time imaging of DNA loop extrusion by condensin. Science, 360(6384):102–105, April 2018.
  • [10] Anton Goloborodko, Maxim V Imakaev, John F Marko, and Leonid Mirny. Compaction and segregation of sister chromatids via active loop extrusion. eLife, 5:e14864, May 2016.
  • [11] Yunyan He, Josh Lawrimore, Diana Cook, Elizabeth Erin Van Gorder, Solenn Claire De Larimat, David Adalsteinsson, M. Gregory Forest, and Kerry Bloom. Statistical mechanics of chromosomes: in vivo and in silico approaches reveal high-level organization and structure arise exclusively through mechanical feedback between loop extruders and chromatin substrate properties. Nucleic Acids Research, 48(20):11284–11303, November 2020.
  • [12] Caitlin Hult, David Adalsteinsson, Paula A. Vasquez, Josh Lawrimore, Maggie Bennett, Alyssa York, Diana Cook, Elaine Yeh, Mark Gregory Forest, and Kerry Bloom. Enrichment of dynamic chromosomal crosslinks drive phase separation of the nucleolus. Nucleic Acids Research, 45(19):11159–11173, November 2017.
  • [13] Alexander A. Kukreja, Sisira Kavuri, and Ajit P. Joglekar. Microtubule Attachment and Centromeric Tension Shape the Protein Architecture of the Human Kinetochore. Current Biology, 30(24):4869–4881.e5, December 2020.
  • [14] Yang Li, Shengyuan Xu, Jinqiao Duan, Xianbin Liu, and Yuming Chu. A machine learning method for computing quasi-potential of stochastic dynamical systems. Nonlinear Dynamics, 109(3):1877–1886, August 2022.
  • [15] Bo Lin, Qianxiao Li, and Weiqing Ren. A Data Driven Method for Computing Quasipotentials. In Proceedings of Machine Learning Research, volume 145, pages 652–670, 2022.
  • [16] Andrew D. McAinsh and Geert J. P. L. Kops. Principles and dynamics of spindle assembly checkpoint signalling. Nature Reviews Molecular Cell Biology, 24(8):543–559, August 2023.
  • [17] Jay Newby, Jennifer L. Schiller, Timothy Wessler, Jasmine Edelstein, M. Gregory Forest, and Samuel K. Lai. A blueprint for robust crosslinking of mobile species in biogels with weakly adhesive molecular anchors. Nature Communications, 8(1):833, October 2017.
  • [18] Matthew Robert Paul, Andreas Hochwagen, and Sevinç Ercan. Condensin action and compaction. Current genetics, 65(2):407–415, April 2019.
  • [19] Grigorios A. Pavliotis and Andrew M. Stuart. Multiscale Methods: Averaging and Homogenization. Springer, New York, NY, 2008.
  • [20] Stephanie Andrea Schalbetter, Anton Goloborodko, Geoffrey Fudenberg, Jon-Matthew Belton, Catrina Miles, Miao Yu, Job Dekker, Leonid Mirny, and Jonathan Baxter. SMC complexes differentially compact mitotic chromosomes according to genomic context. Nature Cell Biology, 19(9):1071–1080, September 2017.
  • [21] Tsuyoshi Terakawa, Shveta Bisht, Jorine M. Eeftens, Cees Dekker, Christian H. Haering, and Eric C. Greene. The condensin complex is a mechanochemical motor that translocates along DNA. Science (New York, N.Y.), 358(6363):672–676, November 2017.
  • [22] Nynke L. van Berkum, Erez Lieberman-Aiden, Louise Williams, Maxim Imakaev, Andreas Gnirke, Leonid A. Mirny, Job Dekker, and Eric S. Lander. Hi-C: A Method to Study the Three-dimensional Architecture of Genomes. Journal of Visualized Experiments : JoVE, (39):1869, May 2010.
  • [23] Paula A Vasquez and Kerry Bloom. Polymer models of interphase chromosomes. Nucleus, 5(5):376–390, September 2014.
  • [24] Paula A. Vasquez, Caitlin Hult, David Adalsteinsson, Josh Lawrimore, Mark G. Forest, and Kerry Bloom. Entropy gives rise to topologically associating domains. Nucleic Acids Research, 44(12):5540–5549, July 2016.
  • [25] Paula A. Vasquez, Ben Walker, Kerry Bloom, Daniel Kolbin, Neall Caughman, Ronit Freeman, Martin Lysy, Caitlin Hult, Katherine A. Newhall, Micah Papanikolas, Christopher Edelmaier, and M. Gregory Forest. The power of weak, transient interactions across biology: A paradigm of emergent behavior. Physica D: Nonlinear Phenomena, 454:133866, November 2023.
  • [26] Benjamin Walker, Dane Taylor, Josh Lawrimore, Caitlin Hult, David Adalsteinsson, Kerry Bloom, and M. Gregory Forest. Transient crosslinking kinetics optimize gene cluster interactions. PLOS Computational Biology, 15(8):e1007124, August 2019.
  • [27] Benjamin L. Walker and Katherine A. Newhall. Numerical computation of effective thermal equilibria in stochastically switching Langevin systems. Physical Review E, 105(6):064113, June 2022.
  • [28] Timothy Wessler, Alex Chen, Scott A. McKinley, Richard Cone, M. Gregory Forest, and Samuel K. Lai. Using Computational Modeling To Optimize the Design of Antibodies That Trap Viruses in Mucus. ACS Infectious Diseases, 2(1):82–92, January 2016.
  • [29] Joseph Xu Zhou, M. D. S. Aliyu, Erik Aurell, and Sui Huang. Quasi-potential landscape in complex multi-stable systems. Journal of the Royal Society Interface, 9(77):3539–3553, December 2012.