{"id":"785f2c9d-62b2-4aee-b318-8c504469f693","arxiv_id":"1908.05851","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"In living cells, nucleosomes split into fast and slow movers that form transient dynamic domains, revealed by statistical analysis of single-nucleosome trajectories and a polymer model.","lead":"This paper reanalyzes single-nucleosome tracking data from living human cells and reports that nucleosome motion is bimodal: some nucleosomes move fast, others slow, forming dynamic fluid-like domains. It introduces a statistical framework for separating these populations and a minimal polymer model to explain how local constraints create them.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The bimodal MSD distribution that defines fast/slow nucleosomes rests on an uncharacterized Richardson-Lucy inversion; if its valley M* is not stable, every domain-size estimate is conditional.","rationale":"The reader's weakest assumption is the right one: the chain of inference from bimodal P(M) to dynamic domains is only as strong as the deconvolution step. I agree with the reader's CONDITIONAL verdict. I would not move to REJECT because the paper provides a transparent description, reports reproducibility across 10 cells, gives consistency checks with perturbation experiments, and includes an illustrative polymer model; those are real though indirect supports. The missing piece is a direct validation of the RL inversion on a known ground truth matching the experimental noise and sampling regime. The proposed synthetic test would settle whether the bimodality and M* are stable. If the inversion survives the test, the dynamic-domain picture is substantially strengthened. If not, the classification threshold and all downstream domain sizes are in question. Hence no change to the conditional verdict is needed.","tokens_in":11383,"tokens_out":5491,"duration_ms":58618,"concrete_test":"Create synthetic vHCs with a known ground-truth P(M) at t=0.5 s using the experimental sampling density, localization precision, trajectory lifetime, and field of view, for (i) a single broad unimodal P(M) and (ii) a bimodal P(M) with the reported peak positions and valley. Apply the exact RL protocol from Methods with at least three initial guesses and iteration counts spanning any reasonable stopping range. Require that case (ii) recovers M* and the fast fraction within 10% and that case (i) never develops a second peak at or near the 0.5-s valley. If the protocol fails either test, the M* threshold is not robust and the domain-size statistics must be recomputed from a calibrated inversion.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's dynamic-domain conclusion depends on labeling each nucleosome as fast or slow using the threshold M*, defined as the valley of P(M,0.5 s). But P(M,t) is not directly observed; it is the solution of the Fredholm integral equation Eq. 1, Gs(r,t)=∫ dM P(M,t)(1/πM)e^{-r²/M}, obtained by the Richardson-Lucy iteration described in Methods. The main text specifies the initial exponential distribution and the constraints P≥0 and ∫P=1, but not the iteration count, convergence criterion, regularization, or M grid; no ground-truth biological data calibrate the inversion. The validation referenced in Figs. S3–S5 uses simulated polymer systems, whose sampling density, localization error, trajectory length, and finite-nucleus boundary conditions may not reproduce the live-cell experiment. Because M* is the valley of this inverted distribution, any RL-induced peak splitting or valley shift propagates into the fast/slow labels, and then into β exponents, RDF peak positions Dff/Dss, correlation radii Rc, and perturbation comparisons. Hence the load-bearing assumption is that the deconvolution is faithful enough to preserve the true bimodality at t=0.5 s.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper analyzes single-nucleosome trajectories in living human cells to infer the distribution of single-nucleosome mean squared displacement (MSD), P(M,t), using a Richardson-Lucy deconvolution of the self-part of the van Hove correlation function. At t=0.5 s the inferred P(M) is bimodal, and the authors classify nucleosomes as fast or slow according to the threshold M* at the minimum between the two peaks. They then compute separate MSD curves, displacement auto-correlations, density of vibrational modes, and pair-correlation functions for fast and slow nucleosomes, and interpret the resulting spatial oscillations as evidence for fast dynamic domains (f-domains) and slow dynamic domains (s-domains) with diameters of roughly 380 nm and 600 nm, respectively. The paper further examines how these features change under cohesin knockdown, histone hyper-acetylation, formaldehyde crosslinking, and peripheral heterochromatin sampling, and introduces a minimal bead-spring polymer model of two loop domains whose simulated P(M) can show bimodality. The central claim is that chromatin in living cells is organized as a fluid mosaic of fast and slow dynamic domains whose sizes overlap loop-domain and TAD scales.","tokens_in":11673,"tokens_out":4038,"duration_ms":41961,"significance":"If the central claim survives scrutiny, the paper provides a valuable and nontrivial advance: it turns single-nucleosome trajectory data into a spatial picture of dynamically distinct chromatin regions, connecting single-molecule dynamics to domain-scale chromatin organization. The use of the Richardson-Lucy deconvolution to extract MSD distributions from limited trajectories is creative, the collapse of P(M) across 10 cells is a genuine strength, and the perturbation experiments (cohesin KD, TSA, FA, periphery) give a useful, falsifiable handle on the underlying physical constraints. The polymer model, while qualitative, illustrates how tethering and compaction can produce fast/slow differences. However, the main conclusions rest on the stability of the deconvolved P(M) and on the statistical significance of pair-correlation peaks, and these load-bearing elements are not yet established in the manuscript. With additional validation and quantified uncertainty, this could be an important contribution to the chromatin-dynamics literature.","major_comments":[{"comment":"The Richardson-Lucy deconvolution is not fully characterized. The Methods state the iteration formula and the constraints P>=0 and integral P=1, but they do not specify the number of iterations, the convergence criterion, the M-grid discretization, any regularization, or the sensitivity of the result to the initial exponential guess P1. Since M* is defined as the minimum of the deconvolved P(M,0.5 s), and since every fast/slow label and all downstream domain-size estimates inherit M*, the bimodality must be shown to be stable under reasonable variation of these RL parameters. The validation in Figs. S3-S5 is performed on simulated polymer systems; please add synthetic tests that mimic the experimental localization error, trajectory length, finite-nucleus boundary, and sampling density, and report explicitly how M* and the two peak positions vary across these conditions.","section":"Methods, Eq. 1 and Fig. 1B"},{"comment":"The demonstration that fast and slow nucleosomes have different average MSD is partly by construction, because the same threshold M* extracted from P(M,0.5 s) is used to define the two classes and then to compute their respective MSD curves in Fig. 2B. The authors should quantify the construction effect, for example by classifying nucleosomes using P(M) at one time and testing the classification at another time, by applying the threshold to independent trajectory subsets, or by comparing against a permutation/null-label control. Without such a check, the different exponents beta_f and beta_s do not independently establish that the two populations obey different physical mechanisms.","section":"Results, Fig. 2"},{"comment":"The domain sizes D_ss=600 nm, D_ff=380 nm, and D_fs=(D_ff+D_ss)/2 are inferred from peaks in the radial distribution functions g_ab(r), but the text acknowledges that g_ab(r) is small for r<200 nm and that pair sampling is sparse. The manuscript gives no uncertainty estimates for the peak positions, no specified peak-detection rule, and no statistical test against a null model of uniformly distributed or independently classified nucleosomes. Because these peaks are the main quantitative evidence for the f-domain/s-domain mosaic picture, the authors should provide bootstrap confidence intervals for D_aa and D_fs, and a test that the oscillatory pattern is not an artifact of sparse sampling.","section":"Eq. 3 and Fig. 4"},{"comment":"The polymer model's bimodal P(M) is demonstrated for specific hand-picked interaction energies (e.g., epsilon_I=1.2 k_BT and epsilon_II=0.9 k_BT in Fig. 6D), and the model relies on a fixed reference point to represent tethering. The paper claims only qualitative consistency, but even this is weakly evidenced because no parameter scan, no comparison of the simulated P(M) to the experimental P(M), and no prediction of the perturbation responses (cohesin KD or TSA) are given. Please either add quantitative criteria for consistency or explicitly limit the claim to 'illustrative plausibility', which would remove the risk that the model is seen as a fit to the conclusion.","section":"Fig. 6 and polymer model"}],"minor_comments":[{"comment":"The heading 'Significan Statement' contains a typo; it should read 'Significance Statement'.","section":"Abstract/Significance Statement"},{"comment":"The Gaussian basis is variously written as q(r,M), q(M,t), and q(r,M); please use a consistent notation throughout.","section":"Eq. 1 and Methods"},{"comment":"The definition of xi^{ab}(r) in Eq. 5 would be easier to follow if the numerator xi^{ab}_{vv}(r) were defined before Eq. 5 rather than in Eq. 6, and if the relation between xi^{ab}_{rr} and g^{ab}(r) were stated explicitly.","section":"Eqs. 2-6"},{"comment":"The statement that 'the pair correlation functions of position, i.e., the radial distribution functions' are small for r<200 nm conflicts with the later discussion of peaks at 380-600 nm; please clarify whether the plotted g_ab(r) is normalized so that the small-r deficit is meaningful or is a sampling artifact.","section":"Results, pair correlations"},{"comment":"The sentence 'nucleosomes are driven primarily by thermal fluctuating motion' goes beyond the presented evidence, since the perturbations only show that constraints affect mobility; please soften this claim or add a supporting reference for the thermal-driving assumption.","section":"Discussion"}],"recommendation":"major_revision","confidential_remarks":"The manuscript addresses an interesting and timely question, and the experimental perturbations provide a strong proof-of-concept framework. The main risk is statistical: the entire fast/slow taxonomy depends on an under-characterized deconvolution, and the domain sizes are read off from pair-correlation peaks without error bars or null-model controls. These are fixable with additional analysis, so I recommend major revision rather than rejection. I do not see a novelty-disclosure problem; the approach is sufficiently distinct from prior single-nucleosome tracking papers. The authors should also ensure that the validation of the RL procedure is performed on synthetic data that reproduce the experimental noise and sampling conditions, not only on idealized polymer trajectories."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe thing to know: this is a serious statistical reanalysis of existing single-nucleosome tracking data that makes a new, testable claim—nucleosomes in living cells sort into fast and slow dynamic domains with sizes comparable to TADs and loop domains. The bimodal MSD distribution is the load-bearing observation, and the paper is mostly upfront about where it comes from. I read it as a genuine extension of Nozaki et al., not a repackaging.\n\nWhat's new: the fast/slow categorization from P(M) at 0.5 s, the correlation-based domain sizes (D_ff ≈ 380 nm, D_ss ≈ 600 nm), and the perturbation responses (cohesin KD, TSA, FA, periphery). The polymer model is minimal and illustrative, but it does give a mechanistic story consistent with the data. Credit where due: the analysis is reproducible across 10 cells, the RL scheme is tested on simulated polymers, and the authors don't oversell—they flag tethering effects and say quantitative work remains.\n\nThe soft spots are real but not fatal. The main one is the Richardson-Lucy inversion: the main text gives the update rule but not the iteration count, convergence criterion, regularization, or M grid. Since M* is defined as the valley of the deconvolved P(M), any numerical artifact in that valley shifts every fast/slow label and every downstream domain size. The validation on simulated polymers is reassuring but not a substitute for a ground-truth test with known bimodality and experimental sampling density. I'd want to see either a stability analysis of M* under iteration count and initial guess, or a synthetic-data test that mimics the actual trajectory lengths, localization error, and finite-nucleus boundary. Second, the pair-correlation peaks at 380–600 nm rest on sparse pair sampling; there are no error bars on D_ff and D_ss themselves, only on the curves. That's worth stating. Third, the threshold M* being chosen from the same distribution used to define fast/slow is partly by construction—the two populations' average MSDs will differ by design. That doesn't kill the interpretation, but it means the biological content is in the domain sizes and correlation lengths, not in the mere existence of two MSD populations.\n\nThe citation pattern is fine; the prior single-nucleosome work is cited and the model builds on it.\n\nBottom line: the fluid mosaic picture is plausible and worth taking seriously, but the paper needs a revised version that (1) documents the RL inversion's numerical behavior, (2) gives uncertainty estimates for domain sizes, and (3) either constrains the polymer parameters with data or relabels the model as purely illustrative. That is enough for me to send it out for serious peer review. I'd invite it to our reading group, and I'd cite it once the inversion is pinned down.","headline":"A credible, well-scoped reanalysis of single-nucleosome tracking data that makes a new fast/slow domain claim, but the load-bearing bimodal distribution rests on an under-characterized deconvolution that the authors should pin down before publication.","tokens_in":12180,"tokens_out":2154,"would_cite":true,"duration_ms":20335,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Nucleosome motion partitions living chromatin into fast and slow fluid domains.","keywords":["chromatin dynamics","single-nucleosome tracking","mean squared displacement distribution","iterative deconvolution","fast and slow nucleosomes","dynamic domains","polymer model","live-cell imaging"],"falsifier":"Simulate trajectories with a known unimodal distribution of single-particle squared displacements, generate a noisy van Hove function from them, and run the same iterative deconvolution with the same initialization and stopping rule; if the reconstruction returns two peaks, the paper's fast/slow split is an artifact of the method. Alternatively, re-analyze the raw single-nucleosome trajectories with longer observation times, where each nucleosome's own MSD can be estimated without deconvolution, and check whether the bimodality survives.","tokens_in":11155,"feed_emoji":"🧬","tokens_out":6769,"duration_ms":63987,"temperature":0.7,"pith_summary":"The paper tries to establish that nucleosomes in living human cells are not dynamically uniform: their motions fall into two populations, fast and slow, whose correlated movement forms fluid-like domains. The bimodal distribution of single-nucleosome mean squared displacement at 0.5 s provides the classification, and displacement auto- and pair-correlations reveal dynamic domains with diameters of hundreds of nanometers. The inferred domain sizes overlap loop domains and topologically associating domains (TADs), suggesting that the structural units seen by Hi-C have a dynamic counterpart in living cells. Perturbation experiments and a minimal polymer model support the view that tethering and nucleosome-nucleosome interactions, rather than active driving, set which nucleosomes move slowly.","feed_headline":"Fast and slow nucleosomes reveal a fluid mosaic in living chromatin","feed_subtitle":"Two nucleosome populations with correlated motion form domains whose sizes sit near loop domains and TADs.","key_machinery":"The load-bearing object is $P(M,t)$, the distribution of single-nucleosome mean squared displacement at lag time $t$, reconstructed from the self-part of the van Hove correlation function by an iterative deconvolution scheme. At $t=0.5$ s this distribution is bimodal, so the minimum $M^*$ between the peaks partitions nucleosomes into fast and slow classes. Everything downstream—the separate MSD curves, the auto-correlation functions $\\eta_a(t)$, the density of vibrational modes $D_a(\\omega)$, and the pair-correlation functions $g_{ab}(r)$ and $|\\xi_{ab}(r)|$—is computed separately for the two classes, and the domain radii $R_c^{ab}$ are read off from the range of correlated displacement directions. The same $P(M)$ statistic is computed for a bead-spring polymer ring with two interaction regions, which lets the authors test whether compact versus open local geometry plus tethering can produce the observed bimodality.","core_discovery":"On the paper's own terms, the central discovery is that chromatin in living human cells organizes itself into fast dynamic domains (f-domains) and slow dynamic domains (s-domains), regions within which nucleosome displacements are correlated and which alternate in a mosaic-like spatial arrangement. Fast and slow nucleosomes are defined by the two peaks of the distribution $P(M,0.5\\,\\mathrm{s})$, the distribution of per-nucleosome mean squared displacement at 0.5 s, separated by a threshold $M^*$. The radial distribution functions show oscillations at characteristic distances $D_{ff}\\approx 380$ nm and $D_{ss}\\approx 600$ nm, and the displacement correlation functions have finite ranges $R_c^{ff}\\approx R_c^{fs}\\approx 190$ nm and $R_c^{ss}\\approx 300$ nm, which the paper reads as domain radii. A finite value of the vibrational density of states at zero frequency marks the domains as fluid rather than solid. Perturbed cells (cohesin knockdown, histone hyperacetylation, crosslinking, and peripheral heterochromatin) shift the fast/slow balance in ways consistent with constraints from cohesin-mediated chain bundling and nuclear tethering, and a two-region polymer model reproduces the bimodal MSD pattern when one region is compact and the other open.","pith_inferences":[],"forward_implications":["If the fast/slow classification is real, Hi-C domains and TADs are mirrored in living-cell dynamics: the f-domain radius of about 190 nm maps to roughly 50–300 kb (near the 185 kb median loop-domain size) and the s-domain radius of about 300 nm maps to roughly 150–500 kb (near clusters of loop domains or TADs).","Cohesin knockdown should increase the fast fraction while making slow nucleosomes slower, reflecting enhanced A/B compartmentalization; the observed box plots support this, and direct comparison with Hi-C contact maps would test it.","Histone hyperacetylation is predicted to dissolve s-domains and mix fast and slow populations, shortening the correlation length of $g_{fs}(r)$, whereas formaldehyde crosslinking freezes chromatin and sharply reduces the fast population.","A finite $D_a(0)$ predicts that chromatin at the 30-nm scale behaves as a fluid rather than a regular solid fiber, so structural models of chromatin must accommodate liquid-like dynamic domains.","Going beyond the paper, the dynamic domains need not be stable along the DNA sequence; if boundaries fluctuate from cell to cell, the same locus could be fast in one cell and slow in another, which would reconcile dynamic imaging with single-cell Hi-C variability.","A testable extension is to check whether RNA polymerase II clusters act as mobile tethers that locally create s-domains; the paper names transcription machinery as a possible factor but does not test it, so inhibiting transcription should shift the fast/slow balance toward fast if this tethering picture is right.","Another extension: if bimodality is driven by local compaction and tethering rather than sequence, perturbing nuclear tethering (for example by lamin knockdown) should shift $M^*$ and the domain radii in a quantitatively predictable way."],"supporting_citations":[{"why":"Supplies the single-nucleosome trajectories in living human cells that all analyses and perturbation comparisons are built on.","marker":"[22]"},{"why":"Provides the iterative deconvolution algorithm used to reconstruct the MSD distribution from noisy van Hove data.","marker":"[30]"},{"why":"Co-develops the same iterative deconvolution scheme, cited as the basis for extracting $P(M,t)$.","marker":"[31]"},{"why":"Gives the 185 kb median loop-domain size used as the comparison target for f-domain sizes.","marker":"[6]"},{"why":"Defines topologically associating domains, the structural unit used for s-domain size comparisons.","marker":"[5]"},{"why":"Provides the FISH-based domain radius distribution used to convert measured f- and s-domain radii into genomic-size estimates.","marker":"[41]"},{"why":"Supplies a polymer model exhibiting back-scattering and viscoelastic displacement correlations that motivates the correlation analysis.","marker":"[24]"},{"why":"Establishes the criterion that a finite zero-frequency vibrational density of states indicates fluid behavior, which the paper uses to interpret $D_a(0)\\neq 0$.","marker":"[38]"},{"why":"Shows enhanced A/B compartmentalization upon cohesin depletion, used to interpret the cohesin-knockdown results.","marker":"[48]"},{"why":"Shows that cohesin loss eliminates loop domains, used to explain why cohesin knockdown changes the fast/slow balance.","marker":"[49]"}],"fun_headline_variants":["Fast and slow nucleosome domains shape living chromatin","Living chromatin shows fluid domains of fast and slow nucleosomes","Single-nucleosome motion reveals fast and slow dynamic domains","Nucleosomes split into fast and slow dynamic fluid domains","Chromatin's fast and slow nucleosomes form fluid domains"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument assumes that the two peaks in the reconstructed distribution of single-nucleosome squared displacements are a real feature of nucleosome motion and not an artifact of the iterative deconvolution used to build that distribution from noisy images.","fun_headline_variants_meta":{"raw":{"variants":["Fast and slow nucleosome domains shape living chromatin","Living chromatin shows fluid domains of fast and slow nucleosomes","Single-nucleosome motion reveals fast and slow dynamic domains","Nucleosomes split into fast and slow dynamic fluid domains","Chromatin's fast and slow nucleosomes form fluid domains"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000734,"raw_usage":{"total_tokens":3296,"prompt_tokens":974,"completion_tokens":2322,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":590,"completion_tokens_details":{"reasoning_tokens":2243}},"tokens_in":590,"tokens_out":2322,"duration_ms":16791,"temperature":1.0,"reasoning_tokens":2243,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:03:11.042817+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate trajectories with a known unimodal distribution of single-particle squared displacements, generate a noisy van Hove function from them, and run the same iterative deconvolution with the same initialization and stopping rule; if the reconstruction returns two peaks, the paper's fast/slow split is an artifact of the method. Alternatively, re-analyze the raw single-nucleosome trajectories with longer observation times, where each nucleosome's own MSD can be estimated without deconvolution, and check whether the bimodality survives.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the single-nucleosome trajectories in living human cells that all analyses and perturbation comparisons are built on."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the iterative deconvolution algorithm used to reconstruct the MSD distribution from noisy van Hove data."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Co-develops the same iterative deconvolution scheme, cited as the basis for extracting $P(M,t)$."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the 185 kb median loop-domain size used as the comparison target for f-domain sizes."},{"cited_title":"& Heard, E","cited_arxiv_id":null,"evidence_quote":"Defines topologically associating domains, the structural unit used for s-domain size comparisons."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the FISH-based domain radius distribution used to convert measured f- and s-domain radii into genomic-size estimates."},{"cited_title":"A., Wolynes, P","cited_arxiv_id":null,"evidence_quote":"Supplies a polymer model exhibiting back-scattering and viscoelastic displacement correlations that motivates the correlation analysis."},{"cited_title":"& Goddard, W","cited_arxiv_id":null,"evidence_quote":"Establishes the criterion that a finite zero-frequency vibrational density of states indicates fluid behavior, which the paper uses to interpret $D_a(0)\\neq 0$."},{"cited_title":"Two independent modes of chromatin organization revealed by cohesin removal","cited_arxiv_id":null,"evidence_quote":"Shows enhanced A/B compartmentalization upon cohesin depletion, used to interpret the cohesin-knockdown results."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Shows that cohesin loss eliminates loop domains, used to explain why cohesin knockdown changes the fast/slow balance."}],"review_version":1}