{"id":"8671aee4-23a4-44df-a12a-c949ab5a49df","arxiv_id":"2508.19017","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"In epithelial monolayers, cell velocity is positively correlated with local density at small coarse-graining scales and negatively at large scales; the crossover aligns with pressure segregation.","lead":"This paper shows that in living sheets of epithelial cells, whether dense regions move faster or slower depends on the size of the viewing window: local dense spots move slightly faster, while globally crowded regions are slower. The work offers a way to reconcile previous contradictory reports by identifying a mechanical pressure length scale that sets the crossover.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Percentile normalization may be responsible for small-scale positive correlation; raw and null tests are missing.","rationale":"The reader's weakest assumption correctly identifies the percentile normalization as the fragile step: the small-scale positive correlation, the novel part of the central claim, is only reported after a per-frame rank transform. My independent reading confirms that the paper provides no raw correlation curves, no validation of the normalization, and no null-model tests. The absence of these controls means the central experimental observation could be an artifact of the analysis pipeline. This concern is load-bearing because the existence of the sign change is the paper's main message; without the positive small-scale branch, the remaining 'large-scale negative' result is already known from contact inhibition and jamming literature. The model's reproduction of the crossover is suggestive but not decisive, since the model does not involve the experimental measurement pipeline and could exhibit the same crossover for different reasons. Therefore the verdict remains CONDITIONAL pending the proposed raw and null analyses. I do not see a reason to change the reader's verdict; the same missing control is the key gap.","tokens_in":9814,"tokens_out":6281,"duration_ms":63315,"concrete_test":"Reanalyze the experimental data to compute the density–velocity correlation versus window size d in three ways: (1) raw fields, Pearson and Spearman, pooled across frames and also per-frame then averaged; (2) with the authors' per-frame percentile normalization (reproduce Fig. 2E); (3) with a null model that shuffles the density field across positions within each frame (or, better, applies random permutation to density ranks) while keeping the velocity field fixed, then recomputes the correlation–window curve. If the positive correlation at d < crossover is absent in (1) or present in (3) at comparable magnitude, the normalization or intrinsic spatial autocorrelation is producing the sign, and the paper's central claim is not experimentally established. Also report the crossover scale from raw fields and compare to the pressure-segregation length (~7 cell diameters).","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central novel claim—positive density–velocity correlation at small scales—is supported experimentally only by Fig. 2E, where both density and velocity fields are percentile-normalized per frame before correlation. This rank transform removes between-frame (global) density changes, but it also changes the marginal distribution of each field to uniform, and because density counts in small windows are discrete with many ties, the rank transform can substantially alter the measured association relative to the raw fields. The paper does not show: (i) the same correlation–window curve for raw, un-normalized fields (Pearson or Spearman), (ii) scatter plots or binned averages at small d, or (iii) any null model in which, e.g., density ranks are shuffled across positions within each frame while velocity is held fixed. Without these controls, the positive small-scale branch could be an artifact of the normalization interacting with the discrete density counts, rather than a biological property of MDCK monolayers. This is load-bearing because if the small-scale positive branch disappears under raw analysis or survives under a null shuffle, the claim that the density–velocity sign is scale-dependent loses its central experimental support. The model alone does not rescue it, since the model does not use the same measurement pipeline.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports that in confluent MDCK epithelial monolayers the correlation between local cell density and local cell speed changes sign with the size of the coarse-graining window: positive at small windows (a few cell diameters) and negative at large windows. The authors further report, using traction force microscopy and Bayesian Inversion Stress Microscopy, that this crossover coincides with a pressure-segregation length defined from the cross-correlation of high- and low-pressure cells. A multi-phase-field model with extensile active shape deformations is used to reproduce the crossover and to argue that the mechanism is a competition between local active force generation and large-scale mechanical confinement. The paper concludes that density-velocity relations in epithelia are inherently scale dependent, resolving conflicting reports in the literature.","tokens_in":10108,"tokens_out":5675,"duration_ms":53155,"significance":"If the central claim holds, the paper would be an important contribution: it would reconcile apparently contradictory observations of positive and negative density-velocity correlations and introduce a mechanically meaningful length scale into the interpretation of collective cell migration. The experimental design has real strengths: four independent experiments, direct PIV and cell-segmentation measurements, independent traction/stress inference, and a forward phase-field model with parameters taken from prior literature rather than fitted to the density-velocity curve. The model also gives a falsifiable perturbation prediction. However, the experimental evidence for the novel positive small-scale branch rests on a percentile normalization step whose effect is not validated with raw correlations or null-model controls. That gap is load-bearing because the positive branch is the central new observation, and the model does not use the same measurement pipeline.","major_comments":[{"comment":"The positive small-scale density-velocity correlation is shown only after percentile normalization of both fields per frame. The text does not report raw correlations, scatter plots, binned averages, or any null-model test. Because local density counts in small windows are discrete and have many ties, a rank transform can change the sign or magnitude of the association relative to the raw fields. This is not a cosmetic issue: the claim that the sign is scale dependent depends on the small-scale positive branch. Please provide: (i) raw Pearson and Spearman correlation vs. window size, (ii) within-frame z-scored analysis, and (iii) a permutation null in which density ranks are shuffled across positions within each frame while velocity is held fixed. Without these controls the crossover could be an artifact of the normalization interacting with discrete density counts.","section":"Section III, Fig. 2E; Methods (percentile normalization)"},{"comment":"The pressure-segregation length is defined inconsistently. The text states the characteristic length is the distance at which g_cross(r) first rises above unity, giving ~7 cell diameters; the Fig. 3B caption says 'the spacing length scale is given 7 cells as point cross g(r) reaches maximum.' These two definitions are not equivalent and can differ by a factor of two. Since the central quantitative claim is that the density-velocity crossover coincides with the pressure-segregation length, a single definition must be used, the uncertainty in the measured length should be reported, and the crossover from Fig. 2E should be overlaid with the segregation length from Fig. 3B. In the simulations the structure-factor wavelength ~7 is described as twice the segregation length (~3.5); the same convention must be applied to the experimental g_cross(r).","section":"Section III, Fig. 3B and text"},{"comment":"The simulation is said to reproduce the experimental density-velocity relation, but the text does not state whether the simulated density and velocity fields are processed with the same percentile normalization used for the experimental data. If the experimental curve is rank-based while the simulation curve is a raw correlation, the comparison is not meaningful. Please specify the analysis pipeline for the simulation explicitly and present both raw and normalized curves (or state that the same normalization was applied). Also report the number of independent simulation runs used for the error bars in Fig. 4D and how the crossover was extracted.","section":"Section III, Fig. 4D and model methods"}],"minor_comments":[{"comment":"The reduced-activity perturbation in Fig. 5A lists ζ = 1.2×10^-2, identical to the base-case value in Table I. Either the listed value or the label is wrong. This should be corrected, since Fig. 5 is used to support the mechanism.","section":"Appendix Table II, Fig. 5A row"},{"comment":"Typo: 'colar bar' should be 'color bar', and the unit 'µm/10mins' should be 'µm/10 min'.","section":"Fig. 1B caption"},{"comment":"The correlation measure is not specified in the text: is Fig. 2E Pearson or Spearman correlation, and what are the numerical values and error bars? The y-axis label and range are not given in the main text.","section":"Fig. 2E and text"},{"comment":"Reference [31] duplicates reference [29]; both are Nier et al., Biophysical Journal 110, 1625 (2016).","section":"References"},{"comment":"'as point cross g(r) reaches maximum' is unclear. If the intended definition is the first crossing of g_cross(r) above unity, the caption should say so explicitly.","section":"Fig. 3B caption"}],"recommendation":"major_revision","confidential_remarks":"The percentile-normalization concern is substantive and, in my view, blocks acceptance in the current form. The authors should be asked for raw correlations and a null model, not just a textual clarification. If those controls support the positive small-scale branch, the paper would likely merit publication in a high-profile soft-matter venue. The inconsistent definition of the pressure-segregation length also needs to be fixed before the quantitative 'coincidence' claim can be assessed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nThis paper is worth engaging with, but read the central quantitative claim as suggestive rather than established. The genuinely new thing is a systematic window-size sweep of the density-velocity correlation in MDCK monolayers, showing a crossover from positive correlation at small windows to negative at large windows. That observation, if it holds up, would reconcile conflicting reports in the literature, and I don't know of a prior systematic test. The multiphase-field model is a fair forward simulation with literature-based parameters, and the perturbation experiments (reduced activity, reduced packing) give a plausible mechanistic story.\n\nThe experimental observation itself is credible in outline: four independent experiments, standard PIV and segmentation, and the global negative correlation in Fig. 1 is consistent with jamming. The problem is the small-scale positive branch, which rests entirely on the percentile-normalized correlation in Fig. 2E. Because the normalization ranks density and velocity per frame, it changes the marginal distributions and can interact with the discrete cell counts in small windows to produce a sign change that is not present in the raw fields. The paper does not show raw correlations, scatter plots, or any null-model shuffle, so the crossover could be a processing artifact. That is the load-bearing point, and it needs controls.\n\nSecond, the claimed coincidence between the crossover and the pressure-segregation length does not hold numerically. The experimental text puts the onset of negative correlation above ~14 cell diameters, while the pressure segregation scale is ~7 cell diameters, and the model crossover is ~3.5. The abstract says these coincide, but the paper never defines what 'coincide' means or gives a statistical comparison. This is a real inconsistency, not a cosmetic one.\n\nMinor issues: error bars are SEM across four experiments, but the 70 frames per experiment are temporally correlated; treating them as independent inflates confidence. No code or data are shipped, so the analysis cannot be checked directly.\n\nWho should read this: anyone working on collective cell migration or active matter density-velocity couplings. It is a hypothesis paper with a testable claim. I would send it to peer review, but with a request for raw analyses, null tests, and a proper length-scale comparison.","headline":"Novel and plausible scale-dependence observation in MDCK monolayers, but the quantitative match to pressure-segregation length and the small-scale positive branch both need stronger controls before the central claim is accepted.","tokens_in":10524,"tokens_out":2623,"would_cite":false,"duration_ms":25032,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["87.17.Jj"],"model":"deepseek-v4-flash","headline":"The density–velocity relation in epithelial monolayers is not a fixed sign but a function of the observation window: positive at small scales, negative at large scales, with the crossover set by mechanical pressure segregation.","keywords":["epithelial monolayers","density–velocity correlation","scale dependence","collective cell migration","pressure segregation","phase-field model","active matter","traction force microscopy"],"falsifier":"Recompute the density–velocity correlation from the raw, unranked coarse-grained fields in the same monolayer movies, sweeping the window size from one to thirty cell diameters; if the small-scale correlation is not positive before percentile normalization, or flips sign under a different detrending scheme, the crossover is not established. In the simulations, the same check can be done directly on raw fields.","tokens_in":9704,"feed_emoji":"🧫","tokens_out":7093,"duration_ms":60981,"temperature":0.7,"pith_summary":"This paper claims that whether dense cell packs move fast or slow depends on the ruler used to measure them. In confluent epithelial monolayers, coarse-graining cell motion over small windows of a few cell diameters gives a positive density–velocity correlation: denser local neighborhoods move faster. Large windows of tens of cell diameters recover the familiar negative correlation in which crowding suppresses motion. The crossover occurs at roughly seven cell diameters in experiments and coincides with the spatial segregation of high- and low-pressure regions measured by traction-force microscopy and Bayesian Inversion Stress Microscopy. A minimal phase-field model whose only active ingredient is activity-induced shape deformation reproduces the crossover and the pressure-segregation length, identifying the mechanism as competition between local active force generation and global mechanical confinement. If correct, the result reconciles conflicting reports of positive and negative density–velocity coupling in the literature and makes the observation window a required parameter for interpreting collective cell dynamics.","feed_headline":"Density-speed link flips sign with the measuring window","feed_subtitle":"In dense cell layers, crowding speeds motion locally but slows it globally; pressure segregation sets the crossover.","key_machinery":"The central object is the crossover length scale in the density–velocity correlation, defined by sweeping a square coarse-graining window of size d over the monolayer and computing at each location the local density (cell count inside the window) and local velocity (mean cell speed in the same window). A percentile ranking step removes global density fluctuations before the correlation sign is read off as d varies. On the mechanical side, the paper couples this to the pressure-segregation length: the distance at which the cross-type radial distribution function g(r) of high- and low-pressure cells first rises above unity, measured by Bayesian Inversion Stress Microscopy from traction-force d","core_discovery":"On the paper's own terms, the discovery is that the density–velocity relation in epithelial monolayers is inherently scale-dependent and cannot be captured by any local functional v = v(ρ). Using coarse-graining windows from a few to tens of cell diameters on MDCK monolayers, the authors measure a positive correlation between local density and local speed at small scales and a negative correlation at large scales, with a smooth crossover between them. Traction-force measurements show that this crossover sits at the same length scale as the spatial segregation of high- and low-pressure regions, about seven cell diameters, and a minimal phase-field model with extensile shape deformations repro","pith_inferences":["The small-scale positive correlation is the novel claim, and it emerges after percentile normalization; a natural test is whether the sign survives raw, unnormalized correlations or alternative detrending of the same movies.","If the mechanism is generic, the same sign flip with window size should appear in other confluent tissues, bacterial colonies, and particle-based models of motility-induced phase separation, and could be checked by re-analyzing existing datasets.","The crossover scale may be tunable by substrate stiffness or geometric confinement, since both should alter the mechanical confinement length without changing cell-level activity.","The diverging pressure-segregation length at low activity suggests a connection between this density–velocity crossover and the jamming or glass transition, where the balance point between the two mechanisms moves outside the observable window."],"forward_implications":["Reports of density-regulated migration should state the coarse-graining window; a single-scale measurement of the correlation sign is incomplete.","Contact inhibition of locomotion and activity-enhanced migration are reconciled: negative coupling dominates beyond the pressure-segregation length, positive coupling below it.","The pressure-segregation length becomes a measurable, cell-type- and environment-dependent parameter that organizes collective dynamics and can be compared across tissues.","Local constitutive relations v = v(ρ) are insufficient; any faithful theory of collective cell migration must treat density dependence as nonlocal.","Model perturbations give testable predictions: weakening activity erases the small-scale positive correlation, and weakening crowding by lowering packing fraction erases the large-scale negative one."],"supporting_citations":[{"why":"Supplies the PIVlab implementation used to compute the velocity fields from the time-lapse images.","marker":"[28]"},{"why":"Bayesian Inversion Stress Microscopy, the method that yields the isotropic pressure maps from which the pressure-segregation length is measured.","marker":"[29]"},{"why":"Supplies the radial distribution function framework used to quantify the depletion zone and define the pressure-segregation length scale.","marker":"[42]"},{"why":"Source of the extensile shape-deformation activity term: shows that monolayers of isotropic cells develop active nematic behavior through shape changes.","marker":"[25]"},{"why":"Glass-like dynamics of collective cell migration; the jamming result that motivates the expectation of a global negative density–velocity correlation.","marker":"[26]"},{"why":"Multi-phase-field model of biological tissues; supplies the physiological parameter ranges used in the simulations.","marker":"[33]"},{"why":"Provides the multiphase-field stress derivation used to compute passive stresses and pressure in the model.","marker":"[34]"},{"why":"Frames density–velocity coupling as a functional v([ρ]) in motility-induced phase separation, the background that the observed crossover directly contradicts.","marker":"[21]"},{"why":"The local-approximation limit v = v(ρ) that the paper's scale dependence disproves.","marker":"[43]"}],"fun_headline_variants":["Cell density–speed relation flips sign at a new length scale","Crowding speeds cells locally but slows them globally","Scale-dependent density–velocity link emerges from pressure segregation","Why dense cells move fast or slow depends on the zoom level"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The new small-scale result — that dense neighborhoods move faster — appears only after rearranging density and velocity values by percentile ranking, and the paper does not show the raw correlations, so the sign crossover could in principle be a product of that rearrangement rather than a biological effect.","fun_headline_variants_meta":{"raw":{"variants":["Cell density–speed relation flips sign at a new length scale","Crowding speeds cells locally but slows them globally","Scale-dependent density–velocity link emerges from pressure segregation","Why dense cells move fast or slow depends on the zoom level"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00048,"raw_usage":{"total_tokens":2179,"prompt_tokens":680,"completion_tokens":1499,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":424,"completion_tokens_details":{"reasoning_tokens":1443}},"tokens_in":424,"tokens_out":1499,"duration_ms":10076,"temperature":1.0,"reasoning_tokens":1443,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T16:01:10.630038+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Recompute the density–velocity correlation from the raw, unranked coarse-grained fields in the same monolayer movies, sweeping the window size from one to thirty cell diameters; if the small-scale correlation is not positive before percentile normalization, or flips sign under a different detrending scheme, the crossover is not established. In the simulations, the same check can be done directly on raw fields.","supporting_citations":[{"cited_title":"Thielicke, Pivlab –towards user friendly, affordable and accurate digital particle image velocimetry in mat- lab, Journal of Open Research Software 2, 30 (2014)","cited_arxiv_id":null,"evidence_quote":"Supplies the PIVlab implementation used to compute the velocity fields from the time-lapse images."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Bayesian Inversion Stress Microscopy, the method that yields the isotropic pressure maps from which the pressure-segregation length is measured."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the radial distribution function framework used to quantify the depletion zone and define the pressure-segregation length scale."},{"cited_title":"Mueller, J","cited_arxiv_id":null,"evidence_quote":"Source of the extensile shape-deformation activity term: shows that monolayers of isotropic cells develop active nematic behavior through shape changes."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Glass-like dynamics of collective cell migration; the jamming result that motivates the expectation of a global negative density–velocity correlation."},{"cited_title":"Multi-phase-field Models of Biological Tissues","cited_arxiv_id":"2503.05053","evidence_quote":"Multi-phase-field model of biological tissues; supplies the physiological parameter ranges used in the simulations."},{"cited_title":"Chiang, A","cited_arxiv_id":null,"evidence_quote":"Provides the multiphase-field stress derivation used to compute passive stresses and pressure in the model."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"The local-approximation limit v = v(ρ) that the paper's scale dependence disproves."}],"review_version":1}