{"id":"22a7257c-bdb0-49ec-a5dc-8ee48753312a","arxiv_id":"2411.13279","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"high","formal_verification":"none","parameter_count":6,"one_line_summary":"A coarse-grained simulation with a fitted three-body protein attraction reproduces multilayer transferrin adsorption on polystyrene nanoparticles and predicts glassy dynamics in the inner soft-corona layer.","lead":"A computer simulation shows that proteins around a polystyrene nanoparticle form a soft outer layer that moves slowly and in a glassy way, rather than exchanging freely with the fluid. The authors build a simplified open-source computer model, tune it with transferrin adhesion data, and check the results against centrifuge measurements.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The glassy-corona claim is not yet empirically or statistically secured: it rests on an autocorrelation curve without error bars, from a model calibrated to the same adsorption data it explains.","rationale":"The reader identifies the three-body potential as the weakest assumption; I accept that concern but see an even more immediate evidentiary gap. The central claim is not simply that a multilayer forms, but that the inner soft-corona layer exhibits glassy dynamics. The only support is a simulated autocorrelation curve without error bars, from a model whose parameters are fit to the same fraction-bound data it is used to explain. The DCS validation is external and useful for NAds, but it does not measure dynamics; moreover, the paper's own numbers (-33% and +18%) contradict the 10% error claim in the Discussion. I credit the open-source code and the DCS data as real evidence for the multilayer adsorption aspect, so the paper should not be rejected. However, the glassy claim should be conditional on an explicit robustness test: ensemble-autocorrelation statistics and a perturbation of the interaction shape that preserves the adsorbed-count fit. If those tests do not stabilize the plateau, the claim should be removed or downgraded to a model prediction rather than a demonstrated property of the soft corona.","tokens_in":25931,"tokens_out":5451,"duration_ms":60020,"concrete_test":"Re-run the [Tf]/[NP]=1500, CNP=1 mg/mL, kappa=35 nm case with at least 10 independent Langevin trajectories and report C1(t) and C2(t) from Equation 13 as ensemble averages over trajectories and initial times, with SEM at each lag, over an observation window at least 10 times longer than the 0.1 s shown. As a control, repeat with the Gaussian-well width omega doubled (and kappa re-fitted to the same fraction-bound curve) to test whether the plateau is specific to the calibrated interaction. If the SC1 plateau collapses or shifts by more than the SEM under either change, the 'glassy evolution' claim is an artifact; if it persists, the claim is credible.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Central claim: soft corona is dominated by glassy evolution (Abstract; Section 3.3.2). The evidence is the SC1 density-autocorrelation plateau and SC2 power law in Figure 9. This is the only dynamic observable in the paper, and it is presented without ensemble error bars or multiple independent runs. Equation 13 averages over initial times t0 in what appears to be a single trajectory of one NP; for the small occupancy fluctuations of a single NP layer, the long-lag estimator is biased by finite observation, and the plateau may be an averaging artifact. Nothing outside the model tests the dynamics: DCS validates only NAds values, and those values contradict the \"within 10%\" summary (deviations of about -33% at [Tf]/[NP]=400 and +18% at 1500; Table 1 and Figure 11). The model is explicitly non-transferable (Section 2.2) and kappa is re-fit for each CNP (Section 3.1), so the CNP=1 mg/mL trajectory used for Figure 9 is not extrapolated to the experimental 0.1 mg/mL condition. The glassy claim therefore rests on an untested, parameter-dependent simulation observable.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces a coarse-grained computational model (BUBBLES) with a three-body protein–protein–nanoparticle interaction to simulate multilayer adsorption of transferrin (Tf) on polystyrene nanoparticles. The model parameters are calibrated against the fraction-bound data of Milani et al. (2012), extrapolated to the experimental nanoparticle concentration, and used to predict two soft-corona layers (SC1 and SC2) on top of the hard corona. The authors analyze the density autocorrelation of these layers and claim that the inner soft-corona layer exhibits glassy dynamics related to crowding. They attempt to validate the model with new DCS experiments measuring the size distribution and adsorbed protein counts. The central claim is that the soft corona is dominated by a glassy evolution, with consequences for the biological identity of nanoparticles.","tokens_in":26216,"tokens_out":7607,"duration_ms":75859,"significance":"If the glassy-corona claim were robust, it would be a valuable advance in understanding protein corona dynamics, and the open-source BUBBLES package with its tutorial is a useful community resource. However, the central claim is not empirically secured: the interaction parameters are fitted to the same fraction-bound data the model is said to reproduce, the glassy autocorrelation is computed from an apparently single trajectory with no error bars, and the DCS validation is semi-quantitative at best and internally inconsistent with the 'within 10%' summary. The manuscript is honest about the model's non-transferability, but this limits the generality of the conclusions. The structural layer predictions and the glassy dynamics are outputs of a calibrated potential rather than independent predictions, so the significance for the physical system remains conditional.","major_comments":[{"comment":"The parameters of the three-body interaction, ε3b and κ, are fitted to the same Milani fraction-bound data that the model is then said to reproduce; the structural layers (Figure 6) and the glassy autocorrelation (Figure 9) are outputs of this fitted potential, not independent tests. Section 2.2 explicitly states that the model is not transferable and requires preliminary adsorption isotherms for each protein–NP pair and thermodynamic condition, so the claim that the soft corona is 'dominated by a glassy evolution' cannot be read as an experimentally validated property of the physical system unless out-of-sample dynamic data are provided.","section":"§3.1, Eq. (5), Figure 4"},{"comment":"The glassy autocorrelation is computed from what appears to be a single trajectory of one nanoparticle: no error bars, no multiple independent runs, and the long-lag estimator for a finite observation window is biased, so the distinction between a plateau (SC1) and a power law (SC2) may be an averaging artifact. With an observation window of only 0.1 s (about two decades), the nonexponential decay and plateau need statistical support before the glassy claim can be accepted.","section":"§3.3.2, Eq. (13), Figure 9"},{"comment":"The DCS validation is not 'within a 10% error' as claimed in the Discussion: the same section reports deviations of about −33% at [Tf]/[NP]=400 and +18% at [Tf]/[NP]=1500. In addition, the DCS-derived corona thickness (2.9–4.7 nm) is comparable to a single Tf radius (3.72 nm), not the multilayer structure extending to ~20 nm predicted by the model; thus the DCS data provide, at best, a test of adsorbed mass in an effective monolayer, not of the multilayer soft corona or its dynamics.","section":"§3.4, Table 1, Figure 11, §4"},{"comment":"The extrapolation of κ to the experimental concentration CNP=0.1 mg/mL is based on an ad hoc function (Eq. 11) fitted to only four simulated concentrations, and the dynamic simulations used for the glassy analysis are all performed at CNP=1 mg/mL with κ=35 nm. The connection between the simulated glassy behavior and the actual experimental condition is therefore not established; either simulations at the extrapolated κ value or a sensitivity analysis over κ is needed.","section":"§3.1, Eq. (11), §2.3"}],"minor_comments":[{"comment":"The numerical values of the reduced surface potentials γ_Tf and γ_NP (or the zeta potentials from which they derive) are not reported; please provide them.","section":"§2.2"},{"comment":"The displayed formula contains garbled text ('radicaltpext/radicaltpext'); the equation should be typeset correctly.","section":"Eq. (11)"},{"comment":"The notation C1(t) and C2(t) is used in the text without explicitly tying it to Ci(t) from Eq. (13); please define the index i in the text.","section":"§3.3.2"},{"comment":"It is unclear whether ε3b=3.75 kBT was also optimized or fixed a priori; the sentence 'we adjust the model's parameter and find that, by fixing ε3b=3.75 kBT and varying κ' suggests both, so clarify the fitting protocol.","section":"§3.1"},{"comment":"The preliminary calculation for RNP=41 nm is cited to an unpublished reference ('Jareño and Delia, 2015'); please mark it as a personal communication or remove it.","section":"Footnote 1"},{"comment":"The DCS experiments use ~110 nm diameter NPs while the simulation uses RNP=35 nm; the authors state they adjusted protein concentration to maintain the same [Tf]/[NP] ratios, but this means the surface area per NP differs, so the comparison of NAds should be justified more explicitly.","section":"§3.4"},{"comment":"The axes of the autocorrelation plot are not legible in the manuscript text; ensure the time axis with units is clearly visible.","section":"Figure 9"}],"recommendation":"major_revision","confidential_remarks":"The manuscript would benefit from a careful internal consistency check: the 'within 10%' claim is contradicted by Table 1, and the DCS-derived thicknesses are inconsistent with a multilayer picture. The central glassy claim is a model prediction; the authors should either temper the abstract or provide direct support. The open-source code and tutorial are a positive contribution."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know about this one. The paper adds a three-body protein-protein-NP attraction to the authors' existing BUBBLES coarse-grained model, and that interaction produces a three-layer corona with an inner soft layer that shows a plateau in the density autocorrelation function. That is a genuine extension beyond their 2016 hard-corona model. They also ship open-source code, a tutorial, and new DCS measurements of adsorbed protein counts, which is real work.\n\nThe structural part holds up. The radial density profile shows three distinct layers that saturate in sequence, and single-protein tracking shows the inner layer is irreversible while the outer two exchange. That is consistent with the hard/soft corona picture and a useful demonstration that a simple three-body term can generate multilayer adsorption.\n\nThe soft spots are all in the glassy claim. The three-body parameters, especially kappa, are fit to Milani's fraction-bound data and then the model is said to reproduce that data. That is calibration, not prediction. Kappa is then extrapolated to the experimental NP concentration with an ad hoc formula, so the autocorrelation in Figure 9 is computed at 1 mg/mL, not at the 0.1 mg/mL used in the experiments. That curve has no error bars and no statement of how many independent runs went into it; for the small occupancy fluctuations of one layer, the long-lag estimator is biased, and the plateau may be an averaging artifact. And the paper says 'within 10%' agreement, but its own Table 1 and Figure 11 show -33% and +18% deviations at the two ends. That is an overstatement.\n\nNone of this kills the multilayer modeling, but it does mean the headline claim — soft corona is dominated by glassy evolution — is not empirically or statistically secured. It is a model output from a fitted potential. The authors are upfront that the model is non-transferable and needs recalibration per condition, which is honest, but it also means the glassy behavior is a property of the model, not yet of the corona.\n\nThis paper is for people doing coarse-grained corona kinetics, and those who might reuse BUBBLES. It deserves a serious referee, not a desk reject, but the referee should ask for multiple independent runs with error bars on the autocorrelation, a corrected statement about validation accuracy, and a clear separation between calibrated reproduction and independent prediction. I would not cite the glassy result, but I might cite the multilayer extension and the code.","headline":"A useful multilayer extension of the BUBBLES corona model, but the glassy soft-corona claim rests on a single unvalidated autocorrelation curve from a calibrated potential.","tokens_in":26766,"tokens_out":3554,"would_cite":false,"duration_ms":34941,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper claims that the soft corona of loosely bound proteins around a nanoparticle does not settle into equilibrium, but evolves with glassy, multi-timescale dynamics, and that a coarse-grained simulation with a surface-induced…","keywords":["protein corona","soft corona","hard corona","coarse-grained simulation","glassy dynamics","transferrin","nanoparticle","multilayer adsorption"],"falsifier":"Measure the population autocorrelation of the inner soft-corona layer directly, for example by fluorescence correlation spectroscopy or single-molecule tracking over the 0.01 to 0.1 second window: if it decays exponentially with a single characteristic time rather than showing a plateau, the claimed glassy dynamics are not present.","tokens_in":25633,"feed_emoji":"🧪","tokens_out":6030,"duration_ms":58593,"temperature":0.7,"pith_summary":"This paper claims that the soft corona, the loosely attached outer layers of proteins around a nanoparticle, does not settle into a simple equilibrium but evolves in a glassy way, reorganizing over many timescales at once. Using a coarse-grained simulation of transferrin on polystyrene nanoparticles, the authors reproduce the experimentally observed multilayer adsorption and the numbers of adsorbed proteins. They find two soft-corona layers with distinct dynamics: the inner layer shows a density autocorrelation with a plateau, a signature of glassy relaxation, while the outer layer decays roughly as a power law. If this is right, the protein shell that cells actually encounter keeps changing slowly, so the biological identity of a nanoparticle depends on how long it has been in a biological fluid.","feed_headline":"Nanoparticle soft corona moves like a glass","feed_subtitle":"Inner layer of loosely bound proteins relaxes on many timescales, so corona identity keeps changing.","key_machinery":"The load-bearing machinery is a coarse-grained Langevin-dynamics model in which proteins are soft spheres, the nanoparticle-protein interaction is a DLVO-type potential with a deep contact minimum, and a new three-body potential U3b = -epsilon_3b exp(-di dj / $kappa^{2}$) exp(-(rij-delta)^2/(2 $omega^{2}$)) adds an attraction between two proteins only when at least one of them is near the nanoparticle surface. This three-body term stands in for the hypothesis that surface adsorption partially unfolds transferrin and exposes sticky residues; with epsilon_3b = 3.75 kBT fixed and the decay length kappa tuned for each nanoparticle concentration, it produces the second and third corona layers and the glassy relaxation. A buffer region surrounding the reaction volume maintains the protein concentration without insertion or deletion events, letting the simulations reach timescales of seconds at experimental protein-to-nanoparticle ratios.","core_discovery":"The central discovery is that the soft corona is dominated by a glassy evolution related to crowding. In simulations of transferrin adsorbing onto carboxylated polystyrene nanoparticles, the corona organizes into three layers: an irreversibly bound hard corona, an inner soft-corona layer SC1 whose density autocorrelation function decays nonexponentially and develops a plateau, and an outer soft-corona layer SC2 whose autocorrelation decays approximately as a power law. The plateau in SC1 is interpreted as dynamical arrest caused by crowding from the outer layer, meaning the inner soft corona is trapped in local free-energy minima rather than relaxing to equilibrium. The same coarse-grained model, calibrated on the fraction of proteins bound from earlier experiments, predicts the number of adsorbed proteins per nanoparticle measured here by differential centrifugal sedimentation within about 10% at intermediate concentrations. The authors conclude that the corona's composition and structure can keep evolving over long times, which should matter for how nanoparticles interact with cells.","pith_inferences":["Beyond the paper: if the glassy behavior is generic, many reported corona 'equilibration' times may be underestimates, and the biological identity of a nanoparticle should be reported together with its exposure history rather than as a single final composition.","Beyond the paper: the three-body interaction is a stand-in for explicit unfolding, so a direct extension would add conformational degrees of freedom to the coarse-grained protein and test whether the same plateau and power-law decay emerge without tuning kappa.","Beyond the paper: the crowding mechanism predicts a testable experimental signature, namely that the inner soft-corona layer should show slowed, collective exchange whose plateau height shifts when the outer layer is made more or less crowded.","Beyond the paper: the need to recalibrate kappa for each nanoparticle concentration suggests the model is not yet predictive across conditions without input from adsorption isotherms, so applying it to other protein-nanoparticle pairs still requires per-system experimental calibration."],"forward_implications":["If the soft corona is glassy, it does not reach equilibrium within the seconds-long observation window; its inner layer can continue to reorganize on longer timescales, so a corona composition measured at one time may not represent what a cell encounters later.","The three-layer structure means protein counts above monolayer saturation require multilayer adsorption, with the fraction-bound curve showing slope changes at layer saturation points, around [Tf]/[NP] = 320 and between roughly 700 and 1000.","Layer stabilization times grow outward: approximately 0.2 seconds for the hard corona, about twice that for the inner soft-corona layer, and two to three times longer for the outer soft-corona layer at the highest concentration studied.","At protein concentrations approaching those in blood, the glassy slowdown could become biologically relevant and should be considered when analyzing nanoparticle-cell interactions over time.","The model's adsorbed-protein counts agree with the differential centrifugal sedimentation measurements to within about 10% at intermediate concentrations, supporting the use of this coarse-grained approach in protein-rich environments."],"supporting_citations":[{"why":"Supplies the experimental fraction-bound data, the [Tf]/[NP] thresholds, and the reversible-versus-irreversible binding picture used to calibrate and validate the model.","marker":"Milani et al., 2012"},{"why":"Provides the base coarse-grained model, the DLVO protein-nanoparticle parameters, the buffer/reservoir method, and the non-Langmuir adsorption theory that this work extends.","marker":"Vilanova et al., 2016"},{"why":"Supports the hypothesis that proteins partially unfold upon corona formation, which motivates the three-body interaction.","marker":"Park, 2020"},{"why":"Supplies the theoretical characterization of glassy relaxation and plateau behavior in autocorrelation functions used to interpret the inner soft-corona layer.","marker":"Gotze and Sjogren, 1992"},{"why":"Cited for the relationship between distance from equilibrium and the height of the plateau in the autocorrelation function.","marker":"Kumar et al., 2006"},{"why":"Provides the protein layer density used to convert differential centrifugal sedimentation diameters into adsorbed protein counts.","marker":"Perez-Potti et al., 2021"},{"why":"Cryo-electron microscopy evidence that the corona may be a uniform layer rather than packed globular proteins, acknowledged as a limitation of the model's fixed-sphere picture.","marker":"Sheibani et al., 2021"}],"fun_headline_variants":["Crowding turns soft corona glassy","Soft corona acts like glass due to crowding","Simulation reveals glassy soft corona dynamics","Inner soft corona trapped by crowding, goes glassy"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole multilayer and glassy picture rests on the assumption that a protein adsorbed in the hard corona partially unfolds and thereby attracts other transferrin molecules through a particular three-body force, whose strength is fixed at 3.75 kT but whose range kappa has to be retuned for each nanoparticle concentration.","fun_headline_variants_meta":{"raw":{"variants":["Crowding turns soft corona glassy","Soft corona acts like glass due to crowding","Simulation reveals glassy soft corona dynamics","Inner soft corona trapped by crowding, goes glassy"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000443,"raw_usage":{"total_tokens":2221,"prompt_tokens":899,"completion_tokens":1322,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":515,"completion_tokens_details":{"reasoning_tokens":1265}},"tokens_in":515,"tokens_out":1322,"duration_ms":11907,"temperature":1.0,"reasoning_tokens":1265,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T16:38:11.422256+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the population autocorrelation of the inner soft-corona layer directly, for example by fluorescence correlation spectroscopy or single-molecule tracking over the 0.01 to 0.1 second window: if it decays exponentially with a single characteristic time rather than showing a plateau, the claimed glassy dynamics are not present.","supporting_citations":[],"review_version":1}