Phosphorylation and pocket-binding ligands rewire the coupled networks in multidomain protein condensate
Curation statements for this article:-
Curated by eLife
eLife Assessment
This important study uses coarse-grained molecular simulations to investigate how phosphorylation and small-molecule ligands reshape the interactions underlying condensation and multidomain assembly of HDAC6. The findings provide mechanistic insights into how modifications of intrinsically disordered regions and ligand binding may regulate protein condensation. The evidence is solid: although the simulations broadly agree with previous experimental observations, concerns about model robustness and sampling in full-length assemblies limit support for the proposed mechanism and its pharmacological implications.
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (eLife)
Abstract
Biomolecular condensates formed by multidomain proteins are increasingly linked to disease and are emerging targets for chemical modulation. Yet most molecular models emphasize intrinsically disordered regions (IDRs), whereas many available ligands bind folded-domain pockets. How IDR modifications and folded-domain ligands are integrated to control condensate organization remains unclear. Here, we address this problem using histone deacetylase 6 (HDAC6), a disease-associated deacetylase whose S22 phosphorylation and catalytic-pocket ligands differentially regulate condensation. Using atomistic-informed Martini 3 coarse-grained simulations, we show that S22 phosphorylation enhances phase separation of the N-terminal IDR1 without compacting isolated chains. Instead, phosphorylation creates a phospho-S22/D26-centered anionic contact node that strengthens dense-phase interactions with Arg/Lys-rich patches. The HDAC6 ligands Nexturastat A and HPOB remodel this IDR network through distinct charged-contact mechanisms: Nexturastat A acts as a focused competitor for cationic patches, whereas HPOB permits partial compensation through alternative charged contacts. In full-length phospho-HDAC6, the assembled network is not IDR1-dominated but is organized by FD3, the C-terminal ZnF-UBP domain. Pocket-bound ligands redirect this folded-domain network from FD3–FD3 association toward pocket–FD3 contacts, with stronger and more persistent remodeling by Nexturastat A. These results reveal a two-layer regulatory mechanism in which phosphorylation rewires IDR electrostatics while pocket-binding ligands redirect folded-domain interactions, suggesting a strategy for modulating multidomain condensates through ligand-induced contact-network remodeling.
Article activity feed
-
eLife Assessment
This important study uses coarse-grained molecular simulations to investigate how phosphorylation and small-molecule ligands reshape the interactions underlying condensation and multidomain assembly of HDAC6. The findings provide mechanistic insights into how modifications of intrinsically disordered regions and ligand binding may regulate protein condensation. The evidence is solid: although the simulations broadly agree with previous experimental observations, concerns about model robustness and sampling in full-length assemblies limit support for the proposed mechanism and its pharmacological implications.
-
Reviewer #1 (Public review):
Summary:
The manuscript authors have used Martini 3-based CG simulations to study how S22 phosphorylation and two catalytic-pocket inhibitors, Nexturastat A and HPOB, regulate HDAC6 condensation.
In isolated IDR1 slabs, phosphorylation lowers the apparent saturation concentration without compacting single chains, and the manuscript attributes this to new contacts between the phospho-S22/D26 region and Arg/Lys-rich patches. The two ligands perturb these contacts differently. In full-length multichain systems, the C-terminal FD3 domain is the most frequent contact partner, and the ligands shift contacts from FD3-FD3 toward pocket-FD3. The manuscript proposes a two-layer mechanism by which pocket-binding ligands regulate multidomain condensates.
Strengths:
The question is well motivated by published …
Reviewer #1 (Public review):
Summary:
The manuscript authors have used Martini 3-based CG simulations to study how S22 phosphorylation and two catalytic-pocket inhibitors, Nexturastat A and HPOB, regulate HDAC6 condensation.
In isolated IDR1 slabs, phosphorylation lowers the apparent saturation concentration without compacting single chains, and the manuscript attributes this to new contacts between the phospho-S22/D26 region and Arg/Lys-rich patches. The two ligands perturb these contacts differently. In full-length multichain systems, the C-terminal FD3 domain is the most frequent contact partner, and the ligands shift contacts from FD3-FD3 toward pocket-FD3. The manuscript proposes a two-layer mechanism by which pocket-binding ligands regulate multidomain condensates.
Strengths:
The question is well motivated by published experiments, the simulated IDR1 construct matches the experimental one, the phosphorylation contact pattern recurs across independent salt conditions, and the Discussion offers testable predictions. My main reservations are whether the full-length results are robust to the model and to sampling, and whether the evidence reaches the mechanistic and pharmacological conclusions drawn from it.
Weaknesses:
(1) The full-length conclusions rest on FD3-FD3 being the dominant domain contact. The water-interaction correction applies only to the disordered regions, so the folded domains retain standard Martini 3 behavior, which is known to overestimate protein association (e.g., Thomasen et al., J. Chem. Theory Comput. 2022). The contact maps are also normalized per bead, which favors a small domain such as FD3. Please test whether FD3-FD3 dominance survives a folded-domain correction, and show raw contact counts next to the normalized maps. Without this test, and without experimental evidence that this domain self-associates, frame the FD3 result as a model prediction.
(2) The ligand-treated IDR1 slabs are single trajectories with no statistical uncertainty on the saturation concentrations or on the contact-difference maps. The full-length systems have three replicates; contact numbers are still rising at the end of several trajectories (Figure S9), and most differences between Nexturastat A and HPOB overlap within error. The wild-type controls also reverse the ligand ordering (Figs. S12, S14), which suggests that trajectory-to-trajectory variation may drive the domain-level differences. Please add independent replicas for the ligand comparisons, show convergence over consecutive time blocks, and report uncertainties on the difference maps. Claims should then be limited to the differences that are resolved.
(3) The Discussion presents the simulations as explaining why Nexturastat A suppresses HDAC6 condensation more strongly than HPOB. However, every full-length system assembles at a high protein concentration, and total contacts and morphologies are similar across conditions, so the simulations do not show suppression. The IDR-level and domain-level mechanisms are studied in separate systems, and ligand distribution and IDR1 contacts are not analyzed in the full-length assemblies, so the coupling of the two layers named in the title is not tested. Please add an analysis that bears on condensation propensity and on this coupling, or recast "explains" as "is consistent with" and revise the title and abstract accordingly.
(4) For phosphorylation, the IDR1 fragment is strongly basic, so a simple reduction of net charge could lower the saturation concentration without a position-specific contact node. The phosphoserine charge state used is also lower than expected at neutral pH. Moving the same charge to a distal serine, and testing the fully charged phosphoserine, would distinguish these readings. For the ligands, the coarse-grained models show only weak pocket engagement, lack a non-pocket control, have unvalidated nonbonded parameters, and are simulated at millimolar concentrations. The ligand-IDR1 mechanism could therefore reflect nonspecific association. Please add a non-pocket surface control, validate the ligand parameters, and relate the simulated concentrations to those of the motivating experiments.
-
Reviewer #2 (Public review):
Summary:
The authors investigated the phosphorylation- and ligand-dependent regulation of HDAC6 condensates using explicit-solvent coarse-grained simulations. They found these regulations can be attributed to the reorganization of the interaction networks between multiple domains within the condensate.
Strengths:
Large-scale condensate simulations with multiple domains using explicit-solvent models are nontrivial, impressive, and provide a useful example of what can currently be studied using such simulation models.
Comparison between simulations with/without ligand and with/without phosphorylation at different salt concentrations provides a comprehensive framework for understanding conditional residue- and domain-level condensate organization.
The simulation results are mostly consistent with previous …
Reviewer #2 (Public review):
Summary:
The authors investigated the phosphorylation- and ligand-dependent regulation of HDAC6 condensates using explicit-solvent coarse-grained simulations. They found these regulations can be attributed to the reorganization of the interaction networks between multiple domains within the condensate.
Strengths:
Large-scale condensate simulations with multiple domains using explicit-solvent models are nontrivial, impressive, and provide a useful example of what can currently be studied using such simulation models.
Comparison between simulations with/without ligand and with/without phosphorylation at different salt concentrations provides a comprehensive framework for understanding conditional residue- and domain-level condensate organization.
The simulation results are mostly consistent with previous experimental observations and offer potential mechanistic understandings that extend those findings.
The proposed FD3-centered interaction network for phosphorylation- and ligand-dependent regulations is a novel hypothesis.
The study generates a few experimentally testable predictions regarding the IDR- and FD3-mediated interactions.
Weaknesses:
When going through the manuscript, especially the result section, I sometimes found it challenging to distinguish what has already been reported experimentally (mostly from Lu et al., Nature Cancer) from what is newly learned from the simulations. I do appreciate that the authors mentioned which results reproduce experimental trends; however, this is often after all simulations. It may help if each subsection first briefly states the known experimental observation and then explains the specific question addressed by or new mechanisms learned by the simulation. For instance, experiments have shown that phosphorylation enhanced and ligand suppressed phase separation.
The authors directly used Automartini to generate the ligand parameters and then checked the contact fraction between the ligand and the pocket. Such analysis only tells whether the ligand visits the pocket but does not test either specificity or binding affinity, and therefore does not support the claim of ligand pocket recognition. The simulation concentration is also in the millimolar regime, probably much higher than the experimental range, so it's not surprising to see ~10% to 20% contact fraction. Considering that no native interactions are used to restrain the ligand-pocket interactions, I suspect the ligand might also have similar contacts to the other regions of HDAC. The authors might want to be careful about their discussions.
The claim regarding the IDR dimension might be problematic. It's still an open question whether a coarse-grained model, or even current all-atom force fields, can reliably resolve the effect of a single phosphorylation on IDR dimensions. I agree that all current results, including the error estimate, support that Rg do not change that much, but the authors might still want to be careful about such claim considering such small perturbation in the sequence.
The proposed electrostatic rewiring is based on the difference between the contact maps of WT and phosphorylation variant. It looks like the difference is relatively small to the contact probabilities. There is also no error estimation or pattern validation from multiple simulations. The authors might want to quantify the uncertainties of such patch-level relative changes by either checking some independent simulations or time-block variations.
A similar comment applies to the claims on the charged-contact reorganization mechanism of two small inhibitors on LLLPS. Error estimates should be provided for these contacts when comparing different variant. In cases when errors are large (e.g. contact number with errors reported in Fig. 4K to N), statistical analysis (p-values) is needed to show the significance of the results.
Regarding the proposed "phospho-S22-centered" contact network, I had some difficulty recognizing the specific regions discussed in Figures 2 and 3. It might be helpful to mark these regions directly on the contact maps. The authors may also wish to consider whether "S22-centered" is the most accurate description, or whether the observed changes are better described as involving a broader segment spanning several residues around S22. And since phosphorylation is introduced at S22, observing changes near S22 is not unexpected and may not be sufficient to establish phospho-S22-centered network.
Within the last result section, the authors argued that FD3 plays an important role in both phosphorylation- and ligand-regulated HDAC assembly. However, only contact-based analysis instead of saturation concentration was presented. I understand the substantial challenge of obtaining converged saturation concentration for such system with multiple domains, if so, the authors need to discuss such caveats.
Though direct experimental evidence of such FD3-centered interaction network does not exist, the experimental paper did provide measurements for a couple domain-deletion constructs, including the deletion of the FD3 (BUZ) region. This might be worth discussing, considering the manuscript focuses on the role of FD3 region. In particular, deletion of the BUZ region did not appear to affect LLPS. This does not necessarily exclude a role for FD3 in phosphorylation- or ligand-dependent regulation. However, it may complicate the description of FD3 as the center of the condensate network. The authors might want to discuss how their model can be reconciled with these domain-deletion results and consider their corresponding conclusion.
Finally, the authors describe their mechanism as coupled networks. They might want to clarify the coupling they meant. If this refers to the coupling between different regions (e.g. S22 and FD3) or domains, they might want to consider more quantitive analysis, for instance by conditional/correlated contacts between domains. Changes in contact patterns at different conditions do not necessarily show the coupling between domains.
-
Reviewer #3 (Public review):
Summary:
In this work, the authors probe the combined role of phosphorylation of an intrinsically disordered region (IDR) and domain pocket interactions with small molecules on the assembly of a multidomain protein HDAC6 using molecular simulations. Their find that IDR phosphorylation and ligand interactions with domain pockets can together reshaped in the interaction networks; thereby, expanding mechanistic understanding of multidomain protein assembly in the context of phase separation beyond an IDR centric grammar. The work is supported by extensive simulations featuring several controls and analysis of the underlying interaction networks.
Strengths:
(1) They find an important role of phosphorylation and ligand binding in shaping multidomain protein assembly.
(2) They demonstrate that even if IDR:IDR …
Reviewer #3 (Public review):
Summary:
In this work, the authors probe the combined role of phosphorylation of an intrinsically disordered region (IDR) and domain pocket interactions with small molecules on the assembly of a multidomain protein HDAC6 using molecular simulations. Their find that IDR phosphorylation and ligand interactions with domain pockets can together reshaped in the interaction networks; thereby, expanding mechanistic understanding of multidomain protein assembly in the context of phase separation beyond an IDR centric grammar. The work is supported by extensive simulations featuring several controls and analysis of the underlying interaction networks.
Strengths:
(1) They find an important role of phosphorylation and ligand binding in shaping multidomain protein assembly.
(2) They demonstrate that even if IDR:IDR interactions are important in isolated IDR systems, folded domain-folded domain interactions can dominate in the full-length protein. Furthermore, ligands can significantly perturb the interaction landscape, e.g., by redirecting folded domain-folded domain interactions to domain pocket ones.
(3) They also show that different ligands can yield differing impact on the interaction networks by both competing with and compensating interactions to varying degrees. This is demonstrated for two ligands (Nexturastat A and HPOB).
(4) They probe both the IDRs in isolation (including both the phosphorylated and unphosphorylated versions) and condensates formed via these IDRs. They also characterize assemblies of the full length proteins (with and without phosphorylation) in presence and absence of the ligands. The combined simulations, in general, support the key conclusions of the paper.
Weaknesses:
(1) There are areas where statistical tests can improve the quantitative presentation of the work. Specifically, the difference in the IDR ensembles for the isolated chains should be assessed for statistical significance.
(2) There appears to be considerable finite-size effects in some of the slab simulations that may affect the conclusions. In particular, the chains along the long axis appear to be interacting across the periodic boundary. This can happen if the slab long axis is not long enough/if the ratio of dilute phase to dense phase is not large enough.
(3) Some decisions in the simulation conditions require further justification/explanation. More specifically, the concentrations of small molecules used, as well as the decision to conduct simulations at 295 K, should be adequately justified.
(4) Given the size of the full-length protein, it is understandable that the full dense phase is not probed for the full-length protein. However, it would be useful to comment on how the conclusions may change in the context of the full length. For instance, there is no real interface that the small molecules need to diffuse across and so discussions on the role of partitioning in the paper should be better contextualized in light of this.
-
Author response:
Reviewer #1 (Public review):
(1) The full-length conclusions rest on FD3-FD3 being the dominant domain contact. The water-interaction correction applies only to the disordered regions, so the folded domains retain standard Martini 3 behavior, which is known to overestimate protein association (e.g., Thomasen et al., J. Chem. Theory Comput. 2022). The contact maps are also normalized per bead, which favors a small domain such as FD3. Please test whether FD3-FD3 dominance survives a folded-domain correction, and show raw contact counts next to the normalized maps. Without this test, and without experimental evidence that this domain self-associates, frame the FD3 result as a model prediction.
We thank the reviewer for highlighting this important concern. We agree that the apparent prominence of FD3–FD3 interactions in …
Author response:
Reviewer #1 (Public review):
(1) The full-length conclusions rest on FD3-FD3 being the dominant domain contact. The water-interaction correction applies only to the disordered regions, so the folded domains retain standard Martini 3 behavior, which is known to overestimate protein association (e.g., Thomasen et al., J. Chem. Theory Comput. 2022). The contact maps are also normalized per bead, which favors a small domain such as FD3. Please test whether FD3-FD3 dominance survives a folded-domain correction, and show raw contact counts next to the normalized maps. Without this test, and without experimental evidence that this domain self-associates, frame the FD3 result as a model prediction.
We thank the reviewer for highlighting this important concern. We agree that the apparent prominence of FD3–FD3 interactions in the full-length simulations requires careful examination, particularly because the folded domains retain the standard Martini 3 protein–protein interaction strength whereas the water-interaction correction used in this study was applied to the disordered regions. We also agree that per-bead normalization may influence the apparent relative contributions of domains of different sizes.
In the revised manuscript, we will examine the robustness of the FD3-centered interaction pattern to the treatment of folded-domain interactions. We will also report raw contact counts alongside the normalized contact maps and explicitly evaluate how domain size and normalization affect the relative contact frequencies.
More generally, we agree that the simulations alone do not establish FD3 self-association as an experimentally validated property of HDAC6 condensates. We will therefore distinguish more clearly between simulation-derived predictions and experimentally established conclusions and will frame the proposed FD3-centered interaction network as a model prediction where appropriate.
(2) The ligand-treated IDR1 slabs are single trajectories with no statistical uncertainty on the saturation concentrations or on the contact-difference maps. The full-length systems have three replicates; contact numbers are still rising at the end of several trajectories (Figure S9), and most differences between Nexturastat A and HPOB overlap within error. The wild-type controls also reverse the ligand ordering (Figs. S12, S14), which suggests that trajectory-to-trajectory variation may drive the domain-level differences. Please add independent replicas for the ligand comparisons, show convergence over consecutive time blocks, and report uncertainties on the difference maps. Claims should then be limited to the differences that are resolved.
We agree that additional statistical characterization is required for the ligand-dependent comparisons. In particular, the current ligand-treated IDR1 slab simulations do not provide sufficient information about trajectory-to-trajectory variability, and differences between Nexturastat A and HPOB in the full-length simulations should not be interpreted without considering the uncertainty across independent trajectories.
In the revised manuscript, we will rigorously assess the convergence and statistical robustness of the current simulations, including time-block analyses of the saturation concentrations and contact patterns. We will additionally perform independent replicate slab simulations to evaluate the robustness of the estimated saturation concentrations and interaction patterns. We will also increase the independent sampling of the full-length multichain systems to better characterize trajectory-to-trajectory variability and the robustness of the domain-level interaction patterns. Together, these analyses will allow us to provide uncertainty estimates and determine which ligand-dependent differences are reproducibly resolved.
We will also reassess the differences attributed to Nexturastat A and HPOB in light of these uncertainties. Where the differences are not statistically resolved, we will refrain from assigning ligand-specific mechanistic interpretations and will revise the corresponding conclusions accordingly.
(3) The Discussion presents the simulations as explaining why Nexturastat A suppresses HDAC6 condensation more strongly than HPOB. However, every full-length system assembles at a high protein concentration, and total contacts and morphologies are similar across conditions, so the simulations do not show suppression. The IDR-level and domain-level mechanisms are studied in separate systems, and ligand distribution and IDR1 contacts are not analyzed in the full-length assemblies, so the coupling of the two layers named in the title is not tested. Please add an analysis that bears on condensation propensity and on this coupling, or recast "explains" as "is consistent with" and revise the title and abstract accordingly.
We agree that the current full-length simulations do not directly demonstrate suppression of condensation by either ligand. All full-length systems were simulated under conditions in which assemblies form, and similarities in the overall contact numbers and morphologies do not provide a direct measure of differences in saturation concentration or phase-separation propensity.
We will therefore revise the title, Abstract, Results, and Discussion where necessary to avoid implying that the full-length simulations directly explain the experimentally observed suppression of HDAC6 condensation. Instead, we will describe the simulated interaction-network changes as being consistent with, or as providing a possible molecular basis for, the experimentally observed effects.
We also agree that the coupling between the IDR-level and domain-level regulatory mechanisms needs to be more clearly established. We will extend the analysis of the full-length simulations to examine ligand distributions and interactions involving IDR1 and the folded domains and will assess whether conditional or correlated contact patterns provide quantitative evidence linking these interaction networks. If such analyses do not provide sufficient evidence for direct coupling, we will revise the terminology accordingly and describe the results as condition-dependent reorganization of multiple interaction networks rather than as a demonstrated coupled mechanism.
(4) For phosphorylation, the IDR1 fragment is strongly basic, so a simple reduction of net charge could lower the saturation concentration without a position-specific contact node. The phosphoserine charge state used is also lower than expected at neutral pH. Moving the same charge to a distal serine, and testing the fully charged phosphoserine, would distinguish these readings. For the ligands, the coarse-grained models show only weak pocket engagement, lack a non-pocket control, have unvalidated nonbonded parameters, and are simulated at millimolar concentrations. The ligand-IDR1 mechanism could therefore reflect nonspecific association. Please add a non-pocket surface control, validate the ligand parameters, and relate the simulated concentrations to those of the motivating experiments.
We thank the reviewer for raising these important points regarding the interpretation of both the phosphorylation- and ligand-dependent effects.
For the phosphorylation-dependent mechanism, we would like to clarify that our simulations were motivated by prior experimental observations showing that phosphorylation at different serine residues of HDAC6 does not produce equivalent effects on phase separation. In particular, phosphorylation at S22 was found to have a pronounced effect on HDAC6 phase separation, whereas phosphorylation at the other tested serine sites did not show a comparable effect. These experimental observations argue against a simple model in which the effect of phosphorylation is determined solely by an increase in the overall negative charge of the protein, and instead indicate that S22 represents a site-specific regulatory position.
Our simulations were therefore designed to explore the molecular basis of this experimentally established site specificity, rather than to establish S22 specificity from the simulations alone. In this context, the enhanced contacts involving the phosphorylated S22-containing region and positively charged segments of IDR1 provide a possible molecular explanation for why phosphorylation at this particular site can have a distinct effect on phase separation. We will clarify this experimental motivation and the scope of our interpretation in the revised manuscript, and will avoid implying that the simulations alone demonstrate the site specificity of S22 phosphorylation.
We nevertheless agree that the representation of the phosphoserine charge is an important modeling consideration. We will examine the sensitivity of the phosphorylation-dependent results to the phosphoserine charge state and clarify the corresponding limitations of the coarse-grained representation in the revised manuscript.
We also agree that the ligand simulations require a clearer assessment of specificity. In the revised manuscript, we will compare ligand contacts with the catalytic pocket to contacts with non-pocket regions of HDAC6 and further evaluate the extent of nonspecific protein association. We will also validate the physical reasonableness of the coarse-grained ligand parameterization, with particular attention to the nonbonded parameters relevant to ligand-protein interactions, and report the validation in the revised manuscript. We will explicitly state the limitations of using these models to infer binding specificity or affinity
Finally, we will clarify the relationship between the ligand concentrations used in the simulations and those used in the motivating experiments. The high ligand concentrations used in the simulations were chosen to obtain sufficient sampling of ligand-protein encounters on accessible simulation timescales, and we will make clear that these simulations should not be interpreted as reproducing experimental binding equilibria at the corresponding concentrations.
Reviewer #2 (Public review):
When going through the manuscript, especially the result section, I sometimes found it challenging to distinguish what has already been reported experimentally (mostly from Lu et al., Nature Cancer) from what is newly learned from the simulations. I do appreciate that the authors mentioned which results reproduce experimental trends; however, this is often after all simulations. It may help if each subsection first briefly states the known experimental observation and then explains the specific question addressed by or new mechanisms learned by the simulation. For instance, experiments have shown that phosphorylation enhanced and ligand suppressed phase separation.
We agree that the manuscript should more clearly separate the experimental observations reported previously from the questions addressed by the present simulations.
In the revised manuscript, we will reorganize the relevant Results subsections so that each section first briefly states the pertinent experimental observation and then defines the specific mechanistic question addressed by the simulations. We will also distinguish more explicitly between trends reproduced by the simulations and hypotheses that emerge uniquely from the simulation analysis. This restructuring should make it clearer which conclusions are supported directly by prior experiments and which represent new mechanistic predictions of the present work.
The authors directly used Automartini to generate the ligand parameters and then checked the contact fraction between the ligand and the pocket. Such analysis only tells whether the ligand visits the pocket but does not test either specificity or binding affinity, and therefore does not support the claim of ligand pocket recognition. The simulation concentration is also in the millimolar regime, probably much higher than the experimental range, so it's not surprising to see ~10% to 20% contact fraction. Considering that no native interactions are used to restrain the ligand-pocket interactions, I suspect the ligand might also have similar contacts to the other regions of HDAC. The authors might want to be careful about their discussions.
We agree that observing contacts between the ligands and the catalytic pocket does not by itself establish specific recognition or binding affinity. The current coarse-grained simulations were designed primarily to examine how the presence of the inhibitors perturbs the protein interaction network rather than to calculate ligand-binding affinities.
In response to this concern, we will analyze ligand interactions with non-pocket regions and compare these with the catalytic-pocket contacts, allowing us to better quantify the degree of preferential localization. We will also validate the physical reasonableness of the coarse-grained ligand parameterization, including the nonbonded parameters relevant to ligand-protein interactions, and report the validation and its limitations more explicitly in the revised manuscript.
Throughout the manuscript, we will revise language implying that the coarse-grained simulations demonstrate specific ligand recognition or reproduce experimental binding affinities. We will also clarify the significance of the relatively high ligand concentrations used in the simulations and relate these conditions to the experimental concentrations.
The claim regarding the IDR dimension might be problematic. It's still an open question whether a coarse-grained model, or even current all-atom force fields, can reliably resolve the effect of a single phosphorylation on IDR dimensions. I agree that all current results, including the error estimate, support that Rg do not change that much, but the authors might still want to be careful about such claim considering such small perturbation in the sequence.
We appreciate this point and agree that the absence of a detectable change in radius of gyration in the present simulations should not be interpreted as evidence that a single phosphorylation cannot alter IDR dimensions more generally.
We will provide appropriate uncertainty estimates and statistical comparisons for the single-chain conformational ensembles and revise the wording to state more narrowly that, within the resolution of the present Martini 3 simulations, we do not detect a substantial phosphorylation-dependent change in IDR1 dimensions. We will also acknowledge the limitations of current coarse-grained models in resolving subtle conformational effects associated with a single post-translational modification.
The proposed electrostatic rewiring is based on the difference between the contact maps of WT and phosphorylation variant. It looks like the difference is relatively small to the contact probabilities. There is also no error estimation or pattern validation from multiple simulations. The authors might want to quantify the uncertainties of such patch-level relative changes by either checking some independent simulations or time-block variations.
We agree that the relative changes visible in the phosphorylation-dependent contact maps need quantitative uncertainty estimates.
In the revised manuscript, we will quantify the reproducibility of these contact changes using time-block analyses together with additional independent slab simulations, and provide uncertainty estimates for the difference maps. We will use these analyses to identify which phosphorylation-dependent contact changes are robust and will restrict the mechanistic interpretation to features that are reproducibly resolved.
A similar comment applies to the claims on the charged-contact reorganization mechanism of two small inhibitors on LLLPS. Error estimates should be provided for these contacts when comparing different variant. In cases when errors are large (e.g. contact number with errors reported in Fig. 4K to N), statistical analysis (p-values) is needed to show the significance of the results.
We agree that the same consideration applies to the comparison of the ligand-dependent interaction networks.
We will report uncertainty estimates for the relevant contact observables and strengthen the statistical assessment using additional independent slab simulations together with expanded independent sampling of the full-length multichain systems. Statistical comparisons will be based on independent estimates, and where the variability is comparable to or greater than the apparent differences between conditions, we will explicitly report this and avoid assigning mechanistic significance to those differences.
Regarding the proposed "phospho-S22-centered" contact network, I had some difficulty recognizing the specific regions discussed in Figures 2 and 3. It might be helpful to mark these regions directly on the contact maps. The authors may also wish to consider whether "S22-centered" is the most accurate description, or whether the observed changes are better described as involving a broader segment spanning several residues around S22. And since phosphorylation is introduced at S22, observing changes near S22 is not unexpected and may not be sufficient to establish phospho-S22-centered network.
We appreciate this suggestion. We agree that describing the observed interaction changes as strictly "S22-centered" may be overly specific, because phosphorylation is introduced at S22 and the altered contacts involve a broader sequence region surrounding this residue.
We will mark the relevant sequence regions directly on the contact maps to make these patterns easier to identify. We will also reconsider the terminology throughout the manuscript and, where more appropriate, describe the effect as a phosphorylation-responsive interaction region or network surrounding S22 rather than implying that S22 alone constitutes a discrete interaction node.
Within the last result section, the authors argued that FD3 plays an important role in both phosphorylation- and ligand-regulated HDAC assembly. However, only contact-based analysis instead of saturation concentration was presented. I understand the substantial challenge of obtaining converged saturation concentration for such system with multiple domains, if so, the authors need to discuss such caveats.
We agree that the full-length simulations provide information about the organization of the assembled state but do not directly determine the saturation concentration of full-length HDAC6. Given the size and complexity of the full-length system, obtaining converged coexistence concentrations would require substantially larger simulations and longer sampling than those performed here.
We will make this distinction explicit in the revised manuscript and clarify that the full-length contact analysis addresses changes in the internal interaction network of the assembled state rather than directly measuring changes in phase-separation propensity.
Though direct experimental evidence of such FD3-centered interaction network does not exist, the experimental paper did provide measurements for a couple domain-deletion constructs, including the deletion of the FD3 (BUZ) region. This might be worth discussing, considering the manuscript focuses on the role of FD3 region. In particular, deletion of the BUZ region did not appear to affect LLPS. This does not necessarily exclude a role for FD3 in phosphorylation- or ligand-dependent regulation. However, it may complicate the description of FD3 as the center of the condensate network. The authors might want to discuss how their model can be reconciled with these domain-deletion results and consider their corresponding conclusion.
We appreciate the reviewer drawing attention to the previously reported BUZ/FD3 deletion experiments. We agree that the observation that deletion of the BUZ region does not abolish LLPS indicates that FD3 is not required for baseline phase separation.
In the revised manuscript, we will discuss these experiments explicitly and will avoid describing FD3 as an obligatory center of the HDAC6 condensation network. Instead, we will discuss the possibility suggested by the simulations that FD3 contributes to the organization or regulation of interactions within the full-length assembly without being necessary for assembly itself. We will also clarify that this interpretation remains a model-derived hypothesis requiring experimental validation.
Finally, the authors describe their mechanism as coupled networks. They might want to clarify the coupling they meant. If this refers to the coupling between different regions (e.g. S22 and FD3) or domains, they might want to consider more quantitive analysis, for instance by conditional/correlated contacts between domains. Changes in contact patterns at different conditions do not necessarily show the coupling between domains.
We agree that condition-dependent changes in two interaction networks are not, by themselves, sufficient to demonstrate coupling between them.
We will clarify what we mean by coupling and will examine whether conditional or correlated contact analyses in the full-length simulations provide quantitative evidence that changes involving one region or domain are associated with changes elsewhere in the interaction network. If these analyses do not support such a relationship, we will revise the terminology and describe the results more conservatively as coordinated or condition-dependent reorganization of multiple interaction networks.
Reviewer #3 (Public review):
(1) There are areas where statistical tests can improve the quantitative presentation of the work. Specifically, the difference in the IDR ensembles for the isolated chains should be assessed for statistical significance.
We agree that the comparisons of the isolated-chain ensembles should include quantitative estimates of uncertainty and statistical support.
In the revised manuscript, we will quantify the variability of the relevant conformational observables across independent simulations and perform appropriate statistical comparisons between the unmodified and phosphorylated IDR1 ensembles. We will use these results to refine the statement regarding the effect of phosphorylation on the single-chain dimensions and avoid interpreting differences that are not resolved within the uncertainty of the simulations.
(2) There appears to be considerable finite-size effects in some of the slab simulations that may affect the conclusions. In particular, the chains along the long axis appear to be interacting across the periodic boundary. This can happen if the slab long axis is not long enough/if the ratio of dilute phase to dense phase is not large enough.
We appreciate the reviewer raising this important issue. Interactions between the condensed slab and its periodic images could bias estimates of the dilute-phase concentration and therefore need to be explicitly evaluated.
We will examine the trajectories and density profiles for evidence of interactions across the periodic boundary and quantify the separation between the condensed phase and its periodic images. We will also evaluate the sensitivity of the relevant conclusions to the simulation box dimensions and, where necessary, perform additional checks using enlarged simulation dimensions for representative systems. We will report the outcome of these tests and discuss any remaining finite-size limitations when interpreting the saturation concentrations.
(3) Some decisions in the simulation conditions require further justification/explanation. More specifically, the concentrations of small molecules used, as well as the decision to conduct simulations at 295 K, should be adequately justified.
We agree that these simulation conditions require clearer justification.
The relatively high ligand concentrations were used to increase the number of ligand–protein encounters that can be sampled within the accessible coarse-grained simulation timescales, rather than to quantitatively reproduce the experimental ligand concentrations. In the revised manuscript, we will state this rationale more explicitly, compare the simulated concentrations with the experimental conditions, and discuss the implications of the elevated ligand concentrations for interpreting ligand–protein contact frequencies and possible nonspecific interactions.
The simulations were conducted at 295 K to match the temperature used in the corresponding experimental measurements that motivated this study. We will clarify this point in the revised manuscript and explicitly state that the simulation temperature was chosen to maintain consistency with the experimental conditions.
(4) Given the size of the full-length protein, it is understandable that the full dense phase is not probed for the full-length protein. However, it would be useful to comment on how the conclusions may change in the context of the full length. For instance, there is no real interface that the small molecules need to diffuse across and so discussions on the role of partitioning in the paper should be better contextualized in light of this.
We agree that the full-length simulations should be distinguished more clearly from the slab simulations of IDR1. The full-length systems probe the organization of an assembled high-concentration state and do not contain a sufficiently large coexisting dilute phase and interface from which equilibrium partitioning or a saturation concentration can be determined.
We will revise the relevant discussion of ligand partitioning to reflect this limitation. Where the current wording implies equilibrium partitioning between dilute and dense phases in the full-length simulations, we will instead describe the measured quantity directly—for example, ligand localization or enrichment within the simulated protein assembly.
More generally, we will clarify throughout the manuscript which conclusions concern the internal organization of the full-length assembled state and which conclusions concern phase-separation propensity inferred from the IDR1 slab simulations.
-
-
-
-
Phosphorylation and pocket-binding ligands rewire the coupled networks in multidomain protein condensate https://www.biorxiv.org/content/10.64898/2026.08.07.743494v1
-
Phosphorylation and pocket-binding ligands rewire the coupled networks in multidomain protein condensate https://www.biorxiv.org/content/10.64898/2026.08.07.743494v1