{"id":"fcedf99e-b0f0-482f-9664-a1f1ca0dd8db","arxiv_id":"2506.02516","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"Sequential multi-step protein binding produces power-law interaction lifetimes that, even at the same mean as single-step binding, alter condensate exchange dynamics, aging, and size distribution in Brownian dynamics simulations.","lead":"This paper's model shows that proteins can bind and unbind either in one step or through a series of small conformational changes, and these two modes produce very different interaction lifetime patterns even when the average lifetime is the same. Using computer simulations of protein droplets, the authors find that the multi-step mode makes droplets age and exchange molecules differently, which may help explain how biological condensates become gel-like over time.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The N=1 'single-step' lifetime distribution derived from Eq. S6 is not exponential, so the central exponential-vs-power-law contrast and the N=1 condensate results rest on a mis-specified baseline.","rationale":"The paper's central claim is that the shape of the interaction lifetime distribution, not just its mean, controls condensate exchange dynamics and size distributions, with single-step binding giving exponential lifetimes and sequential multi-step binding giving truncated power laws. The reader's verdict (CONDITIONAL) identified the uniform energy landscape as the weakest assumption. I find a more concrete and internal issue: the N=1 case of the paper's own model (Eq. S6) does not produce an exponential lifetime distribution. The solution is the first-passage time of diffusion from a reflecting boundary to an absorbing boundary, whose survival probability is an alternating eigenfunction series. At short times the survival probability stays near 1 (for tau_m=15 s, S(1.2 s) ~ 0.999 vs 0.92 for an exponential), and the PDF has a finite-time peak. The SI acknowledges the N=1 continuum limit 'may be inappropriate' but asserts, without testing, that using it or an exponential gives the same outcome. This is not correct, and it matters because the N=1 simulations draw lifetimes from this peaked distribution. The reported Fex ~ t scaling and lack of aging for N=1 are properties of this non-exponential waiting-time distribution, not of genuine first-order single-step binding. The central exponential-vs-power-law contrast is thus not demonstrated as written. The fix is straightforward: use a true exponential distribution for N=1 and re-run the BD simulations; if the qualitative conclusions survive, the paper's message is strengthened; if not, the reported contrast is an artifact. Until this correction is made, the current evidence for the central claim is unreliable. This is an internal consistency concern, not a disagreement with consensus, and it is directly testable with the code the authors would need to provide. The verdict should remain CONDITIONAL, but with the explicit condition that the N=1 baseline be corrected and the simulations re-verified.","tokens_in":14226,"tokens_out":17604,"duration_ms":166009,"concrete_test":"Replace the N=1 lifetime sampler in the Brownian dynamics simulations with a true exponential distribution of the same mean tau_m (inverse transform of F(t)=1-exp(-t/tau_m)), keeping all other parameters fixed. Recompute Fex(t) in Fig. 3 and the size-distribution CCDF in Fig. 4 for all reported (af, Rb, tau_m). If Fex(t) still scales linearly from early times and the N=1 vs N>1 contrast is unchanged, the central claim survives; if the early-time Fex increases faster, shows a lag, or if N=1 exhibits aging-like saturation, the reported contrast is an artifact of the mis-specified baseline. Additionally, plot the hazard rate -d ln S/dt for Eq. S6 with N=1 and the C values from Table S1: a nonconstant hazard confirms non-exponentiality.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"For N=1, Eq. S6 gives S(t) = (4/pi) * sum_{l=1..inf} (-1)^{l-1}/(2l-1) * exp[-C*((2l-1)pi/2)^2 * t]. This is the first-passage survival probability of a diffusing particle starting at a reflecting boundary at n=1 and absorbing at n=0, not the survival of a first-order single-step process. The leading long-time term 1.273*exp(-a*t) with a=C*pi^2/4 is canceled at short and intermediate times by the higher alternating terms. For tau_m = 15 s (C=0.034), S(1.2 s) is about 0.999 whereas exp(-t/tau_m) is about 0.92; the PDF is near zero at t=0, peaks at a finite time of a few seconds, and only then decays exponentially. The SI explicitly states that the continuum limit for N=1 'may be inappropriate,' but then asserts without verification that using Eq. S6 for N=1 or an exponential distribution gives 'the same' outcome. This is incorrect: the waiting-time distribution is not exponential, the renewal process is not Poisson, and the reported N=1 scalings Fex ~ t and the absence of aging are properties of this peaked distribution, not of true single-step exponential binding. The abstract's claim that single-step binding leads to exponential lifetime distributions is therefore not supported by the model as written.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a heuristic model in which protein binding-unbinding proceeds via sequential conformational states, described as diffusion on a one-dimensional coordinate with absorbing and reflecting boundaries (Eq. 1 and Eq. S6). The authors derive lifetime distributions from this model and combine them with Brownian dynamics simulations of condensing proteins. They report that single-step (N=1) and multi-step (N>1) interactions can share the same mean lifetime yet produce different exchange dynamics: N>1 yields faster early exchange but aging with F_ex ~ t^{1/2}, while N=1 shows F_ex ~ t and no aging. They also report different cluster size distributions. The central claim is that the shape of the interaction lifetime distribution, not just the mean, controls condensate behavior.","tokens_in":14527,"tokens_out":7310,"duration_ms":64711,"significance":"If the central claim were established, it would offer a mechanistic route to condensate aging and size polydispersity that does not require time-dependent strengthening of interactions, a topic of active current interest. Strengths of the manuscript include an analytic solution for the survival probability, a clearly described hybrid Brownian-dynamics framework, and careful accounting of several simulation parameters (sticking probability, binding regions, area fraction). The paper explicitly acknowledges the absence of condensate-level experimental validation and the need to calibrate N and C for specific proteins. However, the current implementation of the N=1 baseline is mathematically incorrect: Eq. (S6) with N=1 is not an exponential distribution. This error directly affects the abstract's claim of 'exponential vs. truncated power-law' lifetime distributions and the interpretation of all N=1 simulation results. The qualitative finding that distribution shape matters may still survive after correction, but the manuscript as written does not support its headline claim.","major_comments":[{"comment":"For N=1, Eq. (S6) gives S(t) = (4/π) Σ_{l=1}^∞ (-1)^{l-1}/(2l-1) exp[-C ((2l-1)π/2)^2 t], which is the survival probability of a diffusing particle on [0,1] with absorbing and reflecting boundaries, not the exponential distribution of a first-order single-step process. For C=0.034 (τm=15 s), S(1.2 s) ≈ 0.999 whereas exp(-t/τm) ≈ 0.92; the corresponding PDF is zero at t=0 and peaks at a finite time. Therefore the SI's assertion that 'the form S(t) in Eq. S6 predicts an exponential trend for N=1' is incorrect, and the subsequent claim that using Eq. S6 or an exponential distribution gives the same outcome is unverified. Since all N=1 simulation results in Figs. 3 and 4 use lifetimes drawn from this peaked distribution, the single-step baseline is mis-specified. The authors should re-run the N=1 simulations using a true exponential lifetime distribution (or a discrete one-state Markov model) and revise the abstract and discussion accordingly.","section":"SI, Eq. (S6) and main text, Fig. 1c"},{"comment":"The simulation framework feeds the model's lifetime PDFs directly into the Brownian dynamics, so the observed differences in F_ex, aging, and size distributions are consequences of the assumed lifetime distributions rather than independent tests of the model. This is a legitimate heuristic approach, but the phrase 'the lifetime distributions of individual interactions may singularly control condensate dynamics' overstates the evidence. Only three values of N (1, 5, 25) are compared, and parameters p_s, R_b, and a_f are also varied. Please temper the claim and state explicitly that the simulations illustrate possible consequences of the assumed distributions, not validation of the mechanism.","section":"Model and Results, Figs. 3 and 4"}],"minor_comments":[{"comment":"The symbol 'Ã' appears in place of π in Eq. (2) and throughout the SI; this encoding issue should be corrected.","section":"Eq. (2) and SI"},{"comment":"The notation 'Äm' is used for the mean lifetime in several places; it should be typeset as 'τm' consistently.","section":"Main text"},{"comment":"The SI contains a direct contradiction: it states that using Eq. (S6) for N=1 'may be inappropriate' yet then asserts that it nevertheless yields an exponential trend. Given the major comment above, this section needs to be rewritten to provide the correct exponential baseline or to justify a different baseline.","section":"SI1"},{"comment":"The axes of Fig. S1 are not fully described; please define the x- and y-axes and the physical units for lifetime and probability density.","section":"Fig. S1"},{"comment":"The counting of neighbor exchanges is described verbally; adding a pseudocode or a flowchart would improve reproducibility.","section":"SI3"}],"recommendation":"major_revision","confidential_remarks":"The manuscript addresses a timely question and the simulation framework is clear. However, the N=1 baseline error is load-bearing: the abstract and the central contrast between exponential and power-law lifetime distributions are not supported by the current model. The error is fixable by rerunning simulations with a true exponential distribution, but this is a substantial revision rather than a local edit. I would ask the authors to perform that rerun and to report whether the qualitative conclusions (F_ex scaling, aging, size distributions) survive the correction. If they do not survive, the paper's central claim would need to be substantially reframed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe stress-test note is right. The N=1 baseline in this paper is not exponential. Eq. S6 for N=1 is the survival probability of a diffusing particle starting at the reflecting boundary n=1 and absorbed at n=0. For the parameters used (C=0.034, τ_m≈15 s), the lifetime density is zero at t=0, peaks around 7 s, and only then decays. That is a peaked first-passage distribution, not a first-order exponential. The SI states the continuum limit for N=1 'may be inappropriate' but then asserts the outcome would be the same as a true exponential. It would not be. The reported N=1 scalings (Fex ∝ t, no aging) come from this peaked distribution, so the paper's central exponential-vs-power-law contrast is built on a mis-specified baseline.\n\nStill, the paper has real value. Coupling sequential binding-unbinding lifetime distributions to Brownian dynamics is a new move, and the qualitative message—shape of the lifetime distribution, not just the mean, controls condensate aging and size distributions—is plausible and testable. The model fits single-molecule survival data from Armstrong et al. and Garcia et al. reasonably well for N>1, and the simulation methods are described clearly.\n\nSoft spots beyond the N=1 error: the abstract's 'singularly control' overstates a study that uses its own distributions as inputs. There is no condensate-level experimental validation, and no code or data are shipped. The simulations show consequences of the assumed PDFs rather than independently testing them. These are limitations, not fatal flaws, but they should temper the claims.\n\nI would send this to peer review. It deserves a serious referee, but the authors need to fix the N=1 baseline—use a genuine exponential for single-step binding—and then re-check whether the contrast persists. I suspect the qualitative story will survive, but the current N=1 results don't demonstrate it.\n\nNot a citation for me until fixed, but worth bringing to reading group to discuss the subtlety.","headline":"Good idea, but the N=1 exponential baseline is actually a peaked first-passage distribution; the paper's central contrast needs re-doing.","tokens_in":15051,"tokens_out":9928,"would_cite":false,"duration_ms":88843,"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":"The paper claims that the shape of an interaction's lifetime distribution, not just its mean, controls how protein condensates exchange molecules, age, and distribute sizes.","keywords":["biomolecular condensates","interaction lifetime distribution","sequential binding-unbinding","truncated power law","condensate aging","Brownian dynamics simulation","neighbor exchange dynamics","cluster size distribution"],"falsifier":"Measure, for one condensate-forming protein at fixed mean interaction lifetime, both the single-molecule unbinding time distribution and the cumulative neighbor-exchange curve $F_{\\mathrm{ex}}(t)$; observing an exponential unbinding distribution together with $F_{\\mathrm{ex}}\\propto t^{1/2}$, or a truncated power-law unbinding distribution together with $F_{\\mathrm{ex}}\\propto t$, would show that lifetime distribution shape is not the controlling variable the paper claims.","tokens_in":13968,"feed_emoji":"💧","tokens_out":11915,"duration_ms":98348,"temperature":0.7,"pith_summary":"This paper asks how biomolecular condensates can be at once compositionally specific and dynamically fluid, and proposes that the answer lies in the distribution of interaction lifetimes, not just their average. It models two binding mechanisms: single-step binding-unbinding, which gives exponentially distributed interaction lifetimes, and sequential multi-step binding-unbinding through conformational changes, which gives truncated power-law lifetimes. Holding the mean lifetime equal, the two mechanisms produce different condensate behavior in Brownian-dynamics simulations: the multi-step mechanism exchanges neighbors faster at first but then ages, with cumulative exchanges growing as $t^{1/2}$, whereas the single-step mechanism exchanges steadily and linearly in time. The final cluster size distributions also differ. The paper concludes that the shape of the lifetime distribution may by itself control condensate dynamics.","feed_headline":"Sequential protein binding ages condensates and reshapes their sizes","feed_subtitle":"Same mean lifetime, different fate: multi-step interactions age condensates; single-step ones stay fluid.","key_machinery":"The machinery is a one-dimensional diffusion model of binding and unbinding on a conformational coordinate. The probability $P(n,t)$ of being in conformational bound state $n$ evolves by $\\partial P/\\partial t = C\\,\\partial^2 P/\\partial n^2$ on $0\\le n\\le N$, with an absorbing boundary at the unbound state $n=0$ and a reflecting boundary at $n=N$, starting from $n=1$. The survival probability $S(t)=\\int_0^N P(n,t)\\,dn$ has the closed form $S(t)=\\frac{4}{\\pi}\\sum_{\\ell=1}^{\\infty}\\frac{1}{2\\ell-1}\\sin\\!\\big(\\frac{(2\\ell-1)\\pi}{2N}\\big)\\exp\\!\\big[-C\\big(\\frac{(2\\ell-1)\\pi}{2N}\\big)^2 t\\big]$, whose derivative gives the lifetime distribution: exponential for $N=1$ and truncated-power-law-like for $N>1$. These lifetime distributions feed a Brownian-dynamics simulation in which discs bind with a sticking probability and stay bound for durations drawn from the distribution; the cumulative number of neighbor exchanges and the cluster size distribution are the outputs that carry the argument.","core_discovery":"The central claim is that the lifetime distribution of individual protein interactions, not the mean lifetime alone, may singularly control condensate dynamics. With the same mean lifetime $\\tau_m$, a single-step interaction ($N=1$) yields an exponential lifetime distribution, while a sequential multi-step interaction ($N>1$) yields a truncated power-law distribution. In simulations, these distributions translate into distinct observable behavior: cumulative neighbor-exchange events grow as $F_{\\mathrm{ex}}\\propto t$ for $N=1$ but as $F_{\\mathrm{ex}}\\propto t^{1/2}$ for $N>1$, so the multi-step system exchanges molecules faster initially yet ages, slowing its exchange rate over time. The two mechanisms also produce different complementary cumulative distributions of cluster sizes at the same mean lifetime. The authors interpret this as evidence that condensate aging and size selection can emerge from static binding-unbinding kinetics, without requiring interactions to strengthen over time.","pith_inferences":["If the mechanism generalizes, measuring the scaling exponent of $F_{\\mathrm{ex}}(t)$ in live condensates could serve as a label-free probe of whether the underlying interactions are single-step or multi-step; the paper itself does not propose this diagnostic.","The same lifetime-distribution logic should apply to other protein-driven processes whose output depends on dwell times, such as enzyme cascades or molecular communication, but the paper only gestures at these settings and does not test them.","The uniform, drift-free energy landscape is the simplest possible case; introducing rugged barriers or parallel unbinding routes would likely modify the truncated power-law tail, so single-molecule measurements on engineered proteins with tunable numbers of conformational steps would be a direct test of whether the predicted $N$-dependent scalings survive in real systems."],"forward_implications":["Condensate aging can be driven by the shape of the interaction lifetime distribution alone: a static, sequential multi-step binding mechanism produces a gradual slowdown in molecular exchange without any change in interaction strength over time.","Identical mean interaction lifetimes do not imply identical condensate behavior, so measurements that report only average binding times miss a controlling variable.","The exchange curve provides a fingerprint of the binding mechanism: cumulative exchanges growing as $t^{1/2}$ versus $t$ distinguish multi-step from single-step interactions.","Observed curvature changes in condensate cluster size distributions can be traced to the protein interaction pathway, giving a way to connect structural interaction changes to condensate-level outcome.","Truncated power-law interaction lifetime distributions observed in experiments do not require a mixture of exponentially decaying populations; a single population following sequential multi-step binding can produce them."],"supporting_citations":[{"why":"Supplies the experimental protein–surface desorption data with power-law kinetics that the N>1 survival model is fitted against in Fig. 1b.","marker":"[37]"},{"why":"Supplies the protein–protein transcription-factor data with an intrinsically disordered region that the model also fits in Fig. 1b.","marker":"[40]"},{"why":"Reports non-exponential binding in DNA-coated colloids, evidence that sequential multistep binding-unbinding is not limited to one system.","marker":"[34]"},{"why":"Shows kinetics and non-exponential binding of DNA-coated colloids, further experimental support for truncated power-law lifetime distributions.","marker":"[35]"},{"why":"Provides single-molecule transcription-factor dynamics with power-law behavior, used to argue these distributions arise within a single protein population.","marker":"[38]"},{"why":"Provides the trap-model framework for weak ergodicity breaking and aging that the paper invokes to interpret the $t^{1/2}$ exchange scaling.","marker":"[66]"},{"why":"Supplies the continuous-time random-walk trap model whose heavy-tailed waiting times yield the $t^{1/2}$ jump scaling expected for N>1.","marker":"[67]"},{"why":"Reports experimental size distributions of intracellular condensates, the comparison point for the paper's claim that interaction pathway changes cluster size distribution.","marker":"[8]"},{"why":"Establishes the reaction-limited versus diffusion-limited aggregation criterion used to interpret the simulations' growth regime at sticking probability 0.1.","marker":"[59]"}],"fun_headline_variants":["Lifetime distribution, not mean, controls condensate aging","Sequential binding alters droplet aging and size","Protein interaction timing shapes condensate fate","Same average binding, different condensate outcomes","Multi-step binding slows condensate exchange over time"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The predictions rest on modeling sequential binding and unbinding as a drift-free, constant-rate random walk up and down a uniform one-dimensional ladder of conformational states with only one exit; if real interactions have uneven energy barriers, drift, or parallel unbinding routes, the truncated power-law distributions and the downstream condensate scalings may not survive.","fun_headline_variants_meta":{"raw":{"variants":["Lifetime distribution, not mean, controls condensate aging","Sequential binding alters droplet aging and size","Protein interaction timing shapes condensate fate","Same average binding, different condensate outcomes","Multi-step binding slows condensate exchange over time"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00062,"raw_usage":{"total_tokens":2814,"prompt_tokens":825,"completion_tokens":1989,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":441,"completion_tokens_details":{"reasoning_tokens":1920}},"tokens_in":441,"tokens_out":1989,"duration_ms":14137,"temperature":1.0,"reasoning_tokens":1920,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T11:22:24.206837+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure, for one condensate-forming protein at fixed mean interaction lifetime, both the single-molecule unbinding time distribution and the cumulative neighbor-exchange curve $F_{\\mathrm{ex}}(t)$; observing an exponential unbinding distribution together with $F_{\\mathrm{ex}}\\propto t^{1/2}$, or a truncated power-law unbinding distribution together with $F_{\\mathrm{ex}}\\propto t$, would show that lifetime distribution shape is not the controlling variable the paper claims.","supporting_citations":[{"cited_title":"Armstrong , author J","cited_arxiv_id":null,"evidence_quote":"Supplies the experimental protein–surface desorption data with power-law kinetics that the N>1 survival model is fitted against in Fig. 1b."},{"cited_title":"Garcia , author T","cited_arxiv_id":null,"evidence_quote":"Supplies the protein–protein transcription-factor data with an intrinsically disordered region that the model also fits in Fig. 1b."},{"cited_title":"Biancaniello , author A","cited_arxiv_id":null,"evidence_quote":"Reports non-exponential binding in DNA-coated colloids, evidence that sequential multistep binding-unbinding is not limited to one system."},{"cited_title":"Rogers , author T","cited_arxiv_id":null,"evidence_quote":"Shows kinetics and non-exponential binding of DNA-coated colloids, further experimental support for truncated power-law lifetime distributions."},{"cited_title":"Garcia , author G","cited_arxiv_id":null,"evidence_quote":"Provides single-molecule transcription-factor dynamics with power-law behavior, used to argue these distributions arise within a single protein population."},{"cited_title":"Bouchaud ,\\ title title Weak ergodicity breaking and aging in disordered systems ,\\ @noop journal journal Journal de Physique I \\ volume 2 ,\\ pages 1705 ( year 1992 ) NoStop","cited_arxiv_id":null,"evidence_quote":"Provides the trap-model framework for weak ergodicity breaking and aging that the paper invokes to interpret the $t^{1/2}$ exchange scaling."},{"cited_title":"Monthus \\ and\\ author J","cited_arxiv_id":null,"evidence_quote":"Supplies the continuous-time random-walk trap model whose heavy-tailed waiting times yield the $t^{1/2}$ jump scaling expected for N>1."},{"cited_title":"Lee , author C","cited_arxiv_id":null,"evidence_quote":"Reports experimental size distributions of intracellular condensates, the comparison point for the paper's claim that interaction pathway changes cluster size distribution."},{"cited_title":"Family , author P","cited_arxiv_id":null,"evidence_quote":"Establishes the reaction-limited versus diffusion-limited aggregation criterion used to interpret the simulations' growth regime at sticking probability 0.1."}],"review_version":1}