Author response:
Response to the eLife Assessment
We thank the Editors and the Reviewers for their helpful suggestions, which will help us strengthen and test the key conclusions of this study of condensate dynamics at atomic resolution. In response to the Editors, we will make clearer in the Results and Discussion how the present work advances beyond our initial study of MUT-16 condensates, the scaffold of Mutator foci (Gaurav K et al., Biophys. J. 2025; 124:3987–4004). That study used a multiscale approach — residue-level (CALVADOS2) and near-atomic (Martini3) coarse-grained simulations together with in vitro experiments — to establish that the foci-forming region (FFR) phase separates whereas the adjacent MUT-8-binding region (M8BR) does not, and used atomistic simulations of that non-phase-separating region to dissect client–scaffold recognition. In this way the multi-scale simulations helped to provide a molecular basis for previous in vivo observations by Uebel et al. (PLOS Genet. 2018; 14(7):e1007542). That study did not, however, resolve with atomic resolution the interactions within the phase-separated FFR condensate itself. The present study addresses precisely this gap: from 10 µs of atomistic molecular dynamics of the FFR condensate, we characterise the sub-µs contact dynamics and the protein–ion and protein–water interactions that govern the condensed phase at atomistic resolution — observables inaccessible to the coarse-grained models used previously, but key to understanding the properties of Mutator foci and ultimately how they underpin biological function in small RNA biology.
Reviewer 1:
(1) I have several questions regarding the system preparation that require clarification. The authors state that "65 copies of the coarse-grained MUT-16 FFR were embedded in a slab-shaped simulation," but it is not clear how this initial configuration was generated. Were the molecules randomly distributed in the simulation box, or were they initially arranged in a preformed condensate? Alternatively, were they randomly inserted and allowed to self-assemble into a condensate during NpT simulations? In Figure 1, the atomistic snapshot appears to show a well-defined condensate at the center of the simulation box. It would be important to clarify how this configuration was obtained: Was it generated from coarse-grained simulations starting from random initial conditions? Or was a preassembled condensate used as input? Related to this, how do the authors ensure that the simulations are equilibrated? While 20 μs appears to be a reasonably long simulation time for coarse-grained simulations, it would be useful to demonstrate equilibration explicitly. For example, the authors could plot the center-of-mass positions (in the long axis of the simulation box) of individual proteins over time to show that all molecules reach a steady state and remain within the condensate without systematic drift.
We thank the reviewer for these important clarifying questions regarding system preparation and equilibration.
The initial structure for the atomistic simulation was generated by randomly inserting 65 copies of the coarse-grained MUT-16 FFR into a slab-shaped simulation box using the gmx insert-molecules tool. The molecules were therefore not pre-arranged in a condensate; instead, they were allowed to spontaneously self-assemble from this random configuration during NpT simulations using the Martini3-IDP force field over 20 μs. The well-defined condensate visible in Figure 1 is thus the product of this unbiased self-assembly process.
To make this workflow transparent to the reader, we will revise Figure 1 to include a two-panel illustration of the Martini3 simulation: a snapshot at t = 0 ns showing the randomly distributed chains, and a snapshot at t = 20 μs showing the assembled condensate, connected by an arrow indicating the subsequent backmapping step to the atomistic representation. We believe this will clearly communicate the sequential nature of the pipeline (random insertion → coarse-grained self-assembly → atomistic backmapping).
We appreciate the concrete suggestion for demonstrating equilibration. We will add a supplementary figure showing the center-of-mass positions of individual protein chains along the long axis of the simulation box as a function of simulation time. This will allow readers to verify that molecules converge into the condensate phase and reach a steady state without systematic drift, providing explicit evidence that 20 μs coarse-grained simulation time is sufficient for equilibration under these conditions.
(2) The authors experimentally observe UCST behavior for these condensates. Do the coarse-grained or atomistic simulations reproduce this behavior?
While atomistic simulations may be too computationally demanding to systematically explore temperature dependence, coarse-grained simulations could be used to test whether condensates are stable at lower temperatures and dissolve at higher temperatures. Such an analysis would provide valuable support for the experimental observations.
We thank the reviewer for this valuable suggestion. In previous coarse-grained simulations we have used a coarse-grained force field that does not capture UCST vs LCST behavior (Gaurav K et al. Biophys. J. 2025; 124:3987–4004). It will be very interesting to revisit these coarse-grained simulations with a coarse-grained simulation force field that can capture UCST and LCST behavior such as the Mpipi-T (Chakravarti & Joseph, Protein Sci 2025;34(10):e70284) and HPS-T models (Dignon GL et al. ACS Cent. Sci. 2019; 5(5):821–830). We plan to perform additional coarse-grained simulations at multiple temperatures using the HPS-T force field. The HPS-T model has been shown to capture UCST versus LCST behavior (Changiarath A et al. bioRxiv 2024) in accordance with previous in vitro experiments. These simulations will allow us to test whether the MUT-16 FFR condensates remain stable at lower temperatures and dissolve at higher temperatures, providing direct computational support for the experimentally observed UCST behavior. We will include this analysis in the revised manuscript.
(3) Regarding the analysis of ions, several points could be clarified and extended:
a) It would be helpful to report the total number of ions and quantify how many are located inside vs. outside the condensate. While qualitative trends can be inferred from density profiles, quantitative analysis would strengthen the conclusions.
b) It would also be interesting to analyze the number of contact ion pairs (e.g., Na⁺-Cl⁻ pairs), as described in J. Chem. Phys. 156, 044505 (2022). It is known that some ion models tend to overestimate ion pairing and underestimate solubility (e.g., J. Chem. Phys. 153, 010903 (2020)).
c) In this context, the use of scaled-charge models has been shown to improve the description of ionic solutions and biomolecular systems (e.g., J. Phys. Chem. Lett. 2019, 10, 23, 7531-7536). I would suggest that, at least for one trajectory, the authors perform a test simulation using scaled charges (e.g., scaling by ~0.8) to evaluate whether ion distributions and protein-ion interactions are significantly affected.
We thank the reviewer for these insightful suggestions regarding the ion analysis. We agree that a more quantitative treatment of ion behavior would strengthen the manuscript. To address all three points collectively, we will expand the existing Figure S7 with additional panels. These will include quantitative counts of Na+ and Cl- ions partitioning inside versus outside the condensate complementing the existing density profiles, the Na+–Cl- radial distribution functions to estimate contact ion pair populations following J. Chem. Phys. 156, 044505 (2022).
Following the Reviewer suggestion we will run a simulation with scaled charges (~0.8 scaling factor, J. Phys. Chem. Lett. 2019, 10(23):7531–7536) to evaluate the sensitivity of our results to the choice of ion model. We will compare ion distributions obtained with standard versus scaled charges . We will discuss the contact ion pair results in the context of known force field limitations regarding ion pairing (J. Chem. Phys. 153, 010903 (2020)) and assess whether the scaled-charge treatment leads to any qualitatively different conclusions.
(4) Finally, while the selected water model is known to be accurate, it would be useful to assess its performance for concentrated salt solutions. For example, the authors could estimate the density of a 6 m salt solution and compare it with experimental data or validated models (e.g., J. Chem. Phys. 151, 134504 (2019)). This would help clarify to what extent the conclusions depend on the chosen force field.
We thank the reviewer for this important suggestion. We agree that while the chosen water model is well established for biomolecular simulations, its performance under concentrated salt conditions is a legitimate concern that is worth explicitly validating in the context of this work. We will perform a short bulk simulation of a 6 m NaCl solution and compute the solution density, comparing it to experimental data (J. Chem. Phys. 151, 134504 (2019)). This straightforward validation will allow us to quantify how well our water and ion force field combination reproduces the thermodynamic properties of concentrated salt solutions, and to transparently discuss any deviations and their potential implications for the ion partitioning and protein–ion interaction results presented in the manuscript. The results will be added to the supplementary information alongside the expanded ion analysis in Figure S7.
(5) In the Introduction, it would be helpful to elaborate further on the possible driving forces of LLPS in this region. Are there prior hypotheses or evidence pointing to specific interactions (e.g., cation-π, π-π, electrostatic interactions)? While this work addresses these questions, a brief discussion of previous experimental or theoretical insights would provide useful context.
We thank the reviewer for this helpful suggestion. We will expand the Introduction to briefly discuss the known molecular driving forces of LLPS in IDR-containing proteins. Specifically, we will discuss the role of π–π interactions between aromatic residues (Vernon et al. eLife 2018; 7:e31486), cation–π interactions between aromatic and positively charged residues such as tyrosine–arginine pairs, which have been experimentally demonstrated to drive condensate formation in proteins such as FUS (Qamar et al. Cell 2018; 173:720–734), and the broader sequence-encoded molecular grammar governing these interactions in prion-like RNA-binding proteins (Wang et al. Cell 2018; 174:688–699, Rekhi et al. Nat Chem 2024 16:1113–1124 ). We will discuss previous findings on how ions shape interactions in condensates (MacAinsh et al. eLife 2024; 13:RP100282). We will also note the contribution of electrostatic interactions arising from charge patterning within the IDR, and contextualize how these general principles apply to the specific sequence composition of MUT-16 FFR, motivating the simulation-based investigation presented in this work.
(6) On page 18, the authors state: "MUT-16 FFR satisfies the length (172 residues), aromatic content (20.35%), and Arg enrichment (85.71%) criteria. Its charge content (10.47%) and charge balance (38.89% positive charge fraction) are slightly below the nominal thresholds." It would be very helpful to include a schematic representation of the protein sequence highlighting these features (aromatic residues, charge distribution, etc.) in the corresponding figure, to provide a more intuitive understanding.
We thank the reviewer for this helpful suggestion. We will include a figure showing a schematic representation of the MUT-16 FFR sequence, with aromatic residues, charged residues (positive and negative), and arginine content highlighted.
(7) A question regarding ion hydration: What is the coordination environment of the ions that bridge proteins? Are they still hydrated by water molecules, or does the reduced water content inside the condensate significantly affect their solvation. Typically, Na+ and Cl- ions have coordination numbers around 5-6 in aqueous solution. Do protein interactions and reduced solvent conditions within the condensate alter this coordination? A brief analysis or discussion would be valuable.
We will calculate the coordination numbers of Na⁺ and Cl⁻ ions that mediate residue–residue bridging interactions inside the condensate and compare them against ions in the bulk dilute phase. This will directly reveal the degree to which bridging ions retain or lose their hydration shell when engaging with protein residues, and whether the condensate environment meaningfully perturbs ion solvation. The results will be presented as an additional figure in the Supplementary Information.
Reviewer 2:
(1) The large amount of detail in the results section sometimes makes it difficult to identify the central take-home messages. I encourage the authors to more clearly highlight the principal findings and the physical insights that may generalize to other condensate-forming systems. The authors may also consider streamlining parts of the Results section to improve focus and readability.
We thank the reviewer for this constructive feedback. We will revise the Results section by adding brief concluding remarks at the end of each subsection that explicitly state the key physical insight emerging from that analysis. We will consider which secondary findings can be moved to the Supplementary Information. We will also strengthen the Conclusion section to more clearly distil the principal findings of the study as a whole and highlight the broader insights that may generalize to other condensate-forming systems, ensuring the central take-home messages are clearly communicated to the reader.
Reviewer 3:
(1) In its current form, several technical issues need to be addressed before the main conclusions can be considered robust. Most importantly, the simulated sequence is 172 residues long, while the atomistic slab has box dimensions of only 12 nm in two directions. This length scale is comparable to the expected end-to-end distances of a disordered 172-residue chain. It is therefore not clear whether individual protein chains interact with their own periodic images, which could substantially affect overall chain dynamics and subsequently bias contact lifetimes, residue-residue interaction statistics, and the inferred condensate dynamics. The authors should check, for each chain, histograms of end-to-end distances. For chains for which more than ~2-3% of the end-to-end distances exceed ~11 nm, the authors should explicitly check for self-image interactions (for example, using "gmx mindist -pi") and report whether such interactions occur and for what fraction of the trajectory. Without this control, at least in the Supporting Information, I do not think the simulation-derived contact dynamics are sufficiently trustworthy.
We thank the reviewer for raising this important point. Indeed the box size in x and y dimensions is only marginal, which may influence the dynamics in our simulations and could affect our conclusions. In response, we will perform a control simulation with a larger box, increasing the x and y dimensions to ~16 nm. We will compare the contact dynamics of the resulting trajectory with our original results. This control simulation is initiated from an independently assembled coarse-grained condensate (see our response to Question 6) and therefore also addresses the replica-independence concern raised there.
(2) A second major concern is the treatment of ions. The manuscript makes important conclusions about Na+ association and Na+-mediated bridging, but the atomistic ion model is not explicitly stated. This is a reproducibility problem and also affects interpretation - for example, standard Amber ions are known to bind too strongly to the oppositely charged residues. In their results, one acidic residue appears to interact on average with roughly two Na⁺ ions, which is not obviously expected from charge balance alone. The authors should state the exact Na+/Cl- parameters used, justify their compatibility with TIP4P-D and the protein force field, and explicitly interpret why such a strong Na+ association with acidic residues is observed.
We thank the reviewer for raising this important point. We will explicitly state in the Methods section how the Na+ and Cl- ions, including the force field parameters of the ions, were modelled in our setup, and discuss its compatibility with TIP4P-D and the protein force field. In the presented simulations we have used the Joung and Cheatham parameters (Joung et al, J. Phys. Chem. B 2008, 112 (30), 9020–9041) with σ = 0.243934 nm and ε = 0.365846 (kJ mol-1) for Na+ and σ = 0.447766 nm and ε = 0.148913 (kJ mol-1) for Cl-. While similar setups have been used, these ion parameters have not been optimized for TIP4P-D (originally developed for TIP3P water) and thus a lack of compatibility of the parameters could affect our conclusions.
In response to the Reviewer and also in response to Reviewer 1 (Question 3), we will perform a sensitivity check by running an additional molecular dynamics simulation with scaled ion parameters as suggested by Reviewer 1 ( J. Phys. Chem. Lett. 2019, 10, 23, 7531-7536). In this way we will assess to what extent the degree of Na+ association with acidic residues is sensitive to the choice of ion parameters and discuss the implications for our conclusions regarding Na⁺-mediated bridging interactions.
(3) More generally, because the manuscript is centered on contact lifetimes, the choice of the atomistic force field needs stronger justification. Salt bridges, cation-pi contacts, pi-pi stacking, ion coordination, and water-mediated interactions are all force-field-sensitive. Since there is no direct experimental observable used here to validate the simulations, the authors should discuss the expected limitations of the chosen force field (while I do acknowledge that testing different force fields would be computationally too demanding).
We thank the reviewer for this fair comment. We will add a short discussion justifying the choice of both TIP4P-D and Amber99sb-star-ILDN-q force field, discussing their performance for disordered proteins. We will explicitly acknowledge that absolute contact lifetime values should be interpreted with caution given the inherent force field sensitivities of salt bridges, cation-π, and π-π interactions, while relative trends and qualitative insights are expected to be more robust. We believe this transparent discussion will strengthen the manuscript and place our findings in the appropriate context for the reader.
(4) I also find the sequence-comparison section somewhat confusing. The authors compare one specific IDR, MUT-16 FFR, with the average properties of human IDRs and then frame it as more representative than FUS LCD. It is not clear how informative this is because IDR behavior depends strongly on sequence-specific patterning, molecular connectivity, and the particular interaction network of each protein. Averages over human IDRs may provide a broad context, but they do not necessarily define what is physically or biologically representative for phase separation. In addition, FUS LCD is not intended to be a representative human IDR; it is an unusually low-complexity, phase-separating domain. Therefore, the "more representative than FUS" framing should be toned down. At most, this analysis shows that MUT-16 FFR is compositionally less extreme than FUS LCD.
We thank the reviewer for this valid criticism. We agree that the framing of MUT-16 FFR as "more representative than FUS LCD" is an overstatement, and we will revise the text accordingly. The comparison against human IDR averages was intended to provide broad compositional context rather than make claims about functional or dynamical representativeness, and we will make this distinction explicit. We will reframe the statement to simply note that MUT-16 FFR is compositionally less extreme than FUS LCD, without implying broader representativeness, which as the reviewer correctly points out cannot be inferred from sequence composition alone given the strong dependence of IDR behavior on sequence-specific patterning and interaction networks.
(5) The ion- and water-bridging analyses are also potentially overinterpreted. A distance-based simultaneous contact with two residues does not by itself establish functional mediation or regulation of condensate dynamics. The authors should either add appropriate controls, such as local-density-normalized baselines or randomized-contact expectations, or soften the language to describe these as geometrically defined co-contact events rather than mechanistic bridging interactions.
We thank the reviewer for this valid point. We agree that distance-based co-contact events do not by themselves establish mechanistic bridging or functional regulation, and we will revise the manuscript language throughout to describe these observations as geometrically defined co-contact events rather than mechanistic bridging interactions. We will also explore appropriate controls such as local-density normalized baselines or randomized-contact expectations. In this respect we will also consider our results in light of a recent paper that showed that salt-bridges are overestimated in atomistic molecular dynamics simulations (Ivanović et al, JACS Au 2026, 6(3), 1900–1913). We will ensure the interpretation is appropriately cautious and does not overstate the mechanistic implications of these findings.
(6) Finally, the independence of the atomistic replicas is unclear. The manuscript should state whether all ten all-atom simulations were initiated from the same coarse-grained condensate configuration or from distinct CG frames. If the starting structures came from one CG trajectory, the authors should report how far apart those frames were in simulation time and provide evidence that the initial atomistic configurations are structurally independent. If only velocities differ, the simulations should not be described as fully independent structural replicas.
We thank the reviewer for this important clarification request. We confirm that all ten atomistic replicas were initiated from the same coarse-grained condensate configuration following backmapping, but were equilibrated independently using different random velocity seeds. Only the last 800 ns of each trajectory was used for analysis, discarding the initial 200 ns as equilibration. We will add these details explicitly to the Methods section and make clearer that these simulations are not fully independent structural replicas. We will report the overlap of residue–residue contact maps between replicas to provide an indication of how the contact statistics have decorrelated, given the shared starting structure.
In response to this question and also question 1, we are initiating an all-atom simulation from an independently formed CG condensate (16 nm x 16 nm x 60 nm). This will provide a valuable check as to the conclusions from our ten initial simulation trajectories.
References
Blazquez S, Conde MM, Abascal JLF, Vega C. J. Chem. Phys. 2022;156(4):044505.
Chakravarti A, Joseph JA. Protein Sci. 2025;34(10):e70284.
Changiarath A, Flores-Solis D, Michels JJ, Herrera Rodriguez R, Hanson SM, Schmid F, Zweckstetter M, Padeken J, Stelzl LS. bioRxiv. 2024. doi:10.1101/2024.03.16.585180.
Dignon GL, Zheng W, Kim YC, Mittal J. ACS Cent. Sci. 2019;5(5):821–830.
Gaurav K, Busetto V, Páez-Moscoso DJ, Changiarath A, Hanson SM, Falk S, Ketting RF, Stelzl LS. Biophys. J. 2025;124:3987–4004.
Ivanović MT, Holla A, Nüesch MF, von Roten V, Schuler B, Best RB. JACS Au. 2026;6(3):1900–1913.
Joung IS, Cheatham TE III. J. Phys. Chem. B. 2008;112(30):9020–9041.
Kirby BJ, Jungwirth P. J. Phys. Chem. Lett. 2019;10(23):7531–7536.
MacAinsh M, Dey S, Zhou HX. eLife. 2024;13:RP100282.
Panagiotopoulos AZ. J. Chem. Phys. 2020;153(1):010903.
Qamar S, et al. Cell. 2018;173:720–734.
Rekhi S, Garcia CG, Barai M, Rizuan A, Schuster BS, Kiick KL, Mittal J. Nat. Chem. 2024;16:1113–1124.
Uebel CJ, Anderson DC, Mandarino LM, Manage KI, Aynaszyan S, et al. PLOS Genet. 2018;14(7):e1007542.
Vernon RM, Chong PA, Tsang B, Kim TH, Bah A, Farber P, Lin H, Forman-Kay JD. eLife. 2018;7:e31486.
Wang J, et al. Cell. 2018;174:688–699.
Zeron IM, Abascal JLF, Vega C. J. Chem. Phys. 2019;151:134504.
