Free energy landscapes of KcsA inactivation

Reviewer #1 (Public Review):

This manuscript employs a string method with swarms of trajectories to extract a free energy map of KcsA channel inactivation and its model dependence. The approach connects X-ray structures for closed, partially and fully open, and inactivated KcsA through optimisation of a string defined in a collective variable space consisting of distances involving gate size, cavity-filter and filter pinching (as defined in the proposed X-ray structure for an inactivated state). The final trajectory includes pore opening and filter collapse with water penetration behind the filter, via different intermediates depending on the force field. The authors propose a role for residue L81 in controlling water entry in the final stage of this process. The results suggest that KcsA more easily inactivates with the Charmm force field, with lower barrier and direct passage from a partially open state, whereas the pathway for Amber involves transition first to a fully open state with higher barrier, despite not being the dominant open state seen experimentally under activating conditions. The results also suggest that PG lipids help activate the channel within the Amber force field, consistent with experimental evidence. The work represents large-scale advanced MD simulation. Some questions remain, however, such as if the CV space chosen is sufficient to capture all possible slow coordinates in the inactivation process, and how the resultant free energy surfaces may potentially depend on the end structures and initial pulling procedure.

Collective variable choice:

The explanation for the choice of CVs on page 5 is not sufficient to understand the process and its likely success. How were the most important and unimportant CVs identified exactly? Table 2 on page 19 shows only gate distances, cavity-filter distances and a single variable related to filter structure itself (77 CA - 77 CA) representing a pinch. Is that pinching really the only slow variable associated with inactivation changes in the filter? Why are there no variables, say for carbonyl flipping, E71 or D80 movements or even for ion and water occupancy (although water may be sampled with control of other interactions, such as involving L81)? I understand that the X-ray structure is the one source of information used to define an inactivated structure and is one with just a pinch and no complete carbonyl flipping away from the pore, as has been identified in past studies and discussed as being involved by the authors on page 14. Key changes like carbonyl flipping surely are part of the story and may be slow variables. At the very least, if not part of the CV space, could be analysed.

On page 10 the authors discuss possible differences in Amber and Charmm involving the extent to which the 4 subunits change in respect to the L81-W67 water pathway and W67-D80 hydrogen bond, arguing the different results for force field could be to do with different numbers of subunits doing different things. If I understand, the chosen CVs are all tetramer-based distances (including across subunits) and not subunit-based CVs, so that random and incomplete changes may occur to subunits for a given point in CV space. There is thus potential for the string to converge on a local minimum pathway with partial changes to its interactions within and between subunits, and may not be a unique global solution. Can the authors please explain whether or not this is possible and what analysis has been done to check it?

X-ray endpoints and initial pathway:

The string was created from a pulling/steered MD between existing X-ray structures for the closed (5VKH), partially open (3FB5), fully open (5VK6) and finally inactivated (5VKE) states. The authors write on page 12 that "The block of conduction during inactivation appears to result from pinching at the selectivity filter...", but given the end point was forced to be the X-ray structure with pinching, wasn't this outcome predetermined? This raises a significant point of how much has choice of endpoints predetermined the final states of the string? i.e. How much is an end state actually allowed to draft away from the initial Xray structure. Was a bead placed at the very endpoint and allowed to update via swarms, or was it fixed and all beads just interpolate between those fixed end states? The reason this is important is that it is plausible the inactivated crystal structure with pinching but not other changes (such as complete V76 carbonyl flipping or outer filter splaying), may not be the actual free energy minimum structure for that state and that force field.

Another obvious concern is the possible reliance on the initial pulling procedure used before string optimisation began. Fig.2 Supp 1 shows generally that the Amber path stayed pretty close to the initial steered MD path, whereas Charmm drifted downward away from that path. One could justifiably ask, if a very different initial path was chosen, might different local minimum pathways result, including Amber sampling a path like Charmm? How does one test whether or not the final path has not been trapped in some local trough of free energy? e.g. Imagine starting the Amber string using an initial path like the more diagonal Charmm-like path, or even a more extreme unphysiological one, such as a steered trajectory that initially inactivates before opening the gate. Would the final results be the same? I appreciate the simulations are very expensive and such trials may not be possible, but what evidence is there that the final path has not been trapped away from the global minimum?

One test offered by the authors is a set of unbiased MD simulations launched from points on the string. The authors ran 200ns simulations and write on page 5 that "These simulations have the expected stability based on their starting values. This is a good quality test to check the correct estimation of the general features of the free energy surface". While this sounds reasonable, 200ns MD may only be sufficient to begin to explore locally within the solved free energy trough, much like the swarms in the iterations were able to do. My own examination of Fig2 Supp 5 is that some of these simulations linger around the expected states and some drift away within the general trough of sampling, which is a good sign. What those 200ns simulations may not be able to do is escape that trough and see evidence of other possible solutions, beyond what was sampled with the string that was tied to Xray endpoints and trapped in the solution pathway that was already formed after 100-300 iterations. Overall, the string involved 800 iterations of 10ps swarms (80ns around each bead; albeit 32 trajectories in parallel), allowing good local sampling around the beads in the free energy trough, but in terms of ability to diffuse away from that point, only being comparable in contiguous trajectory time to the unbiased MD tests. It therefore would have been interesting to see if longer simulations remain in this trough; though I understand the challenges in running so much MD. Such simulations may, however, lead to exploration beyond what was seen in the string solutions.

Force field effects and origin:

Regarding the effect of the chosen force field, the authors state that "Given that our simulations were conducted under activating conditions, we had expected the open states to be more populated than the closed ones. Simulations carried out at higher pH may be able to resolve this inconsistency". Also running at high pH would be a nice thing to do to prove the method is in fact sensitive to conditions to see a shift in the distribution of states. But the question is why were open states not more occupied under low pH and 50mM K+? From my analysis of the figures, the results show that the Charmm force field tends to allow for opening of the channel somewhat (at least with similar free energy for partially and fully open to closed) whereas Amber tends to close the channel more (with more uphill energy as the channel opens than Charmm; Fig 2). i.e. at low pH and 50 K+, isn't the Amber model incorrectly reporting fairly strong bias against opening? Moreover, regarding the free energy of the inactivated state itself, why should we not expect equilibrated channels under activating conditions to eventually fall into an inactivated state, in which case we should expect low free energy of that state (as found with Charmm and not Amber in Fig2), but with a slow rate. While much discussion in the manuscript appears to discuss limitations in Charmm (although on page 12 discussion leans either way), these factors may seem to favour Charmm over Amber.

On page 12 the authors explain the possible causes for force field dependence, although this seems limited to ion interactions, glutamate charges and dihedrals. But it would be nice to get a bit more insight into what terms may have influenced the pathway, in particular involving interactions between TM2 and the base of the selectivity filter and hydration behind the filter. Regarding ion interactions, is there a good reason to believe ions are key to the difference seen? i.e. How were ions involved differently in the state transitions involving Amber and Charmm? The authors have noted a role for ion-carbonyl interactions. It is important that the authors explain which is the two competing models has been used and why. i.e. Off-the-shelf Charmm36 force field includes strong K+-backbone carbonyl interaction, previously seen to promote high ion occupancy, similar to Amber, whereas Lennard-Jones parameters modified to match N-methyl-acetamide and water partitioning (such as early Berneche, Noskov and Roux work) reduce ion occupancy and increase water content inside the filter.

Reviewer #2 (Public Review):

The authors describe a computational study into the energetics of KcsA inactivation. Using enhanced sampling, a converged free energy landscape of the inactivation process is achieved in two modern molecular mechanics force fields. The obtained profiles confirm the literature finding of too rapid inactivation, in particular in simulations using the CHARMM force field. Interestingly, it is found that selectivity filter collapse does not gradually follow opening of the inner gate, but proceeds rather switch-like. A key role for residue L81 is proposed as opening gateway in this process.

The study is impressive and interesting. However, I have a number of concerns that the authors may wish to address in a revised version of the manuscript.

First, concerning a set of unbiased simulations spawned at different regions of the investigated free energy landscapes, the authors write: "These simulations have the expected stability based on their starting values".
Fig 2.c shows a rather smooth downhill slope in the free energy curve towards the closed state for AMBER , so wouldn't the expected behavior in that case be that all unbiased trajectories end up in the closed state, or at least travel a substantial amount in that direction? However, that is not observed. This should be further investigated.

Second, "This suggests that stabilization of the partially open state by the removal of bound lipids can explain the increase in open probability" is an odd statement, as "stabilization of the partially open state" means almost the same as "increase in open probability".

The statement "both force fields yield inactivation barriers that are orders of magnitude lower than what is expected from electrophysiology experiments" seems inaccurate. Perhaps the authors mean "inactivation rates that are orders of magnitude lower" rather than barriers?

In addition, the assertion "The CHARMM force field, on the other hand, results in landscapes in agreement with the fact that one of the dominant states in activating conditions is the partially open state, as revealed by a combination of ssNMR+MD." seems to hold for the AMBER force field without PG lipids rather than for CHARMM?

Together with the higher barrier towards the inactivated state as well as covering most known x-ray structures along the inactivation pathway, this would seem to point all in the direction that the studied AMBER force field provides a more faithful picture of the inactivation pathway than CHARMM. I, therefore, find the somewhat inconclusive summary as presented in Fig. 5 a bit uninformative, as it suggests that both mechanisms might be equally likely.

Overall, the study would benefit from a follow-up step to become more conclusive. This could be either in the form of the suggested L81 mutation or changing the simulation conditions to inactivating conditions such as low salt, in which case the inactivated state would be expected to become a minimum, which would provide an additional reference point for validation. Either of these would narrow down the spectrum of possible mechanisms.

Reviewer #3 (Public Review):

The computational study reported in the manuscript "Free energy landscapes of KcsA inactivation" by Pérez-Conesa and Delemotte is quite interesting and insightful.

The computations provide the first complete analysis of how the opening of the activation gate and the constriction of the selectivity filter are coupled in the KcsA channel.

The analysis is careful and is state-of-the-art. The results reveal remarkable differences between the CHARMM and AMBER force fields.

Unfortunately, the "elephant in the room" with regards to K+ channel inactivation is the significance of the dilated structures more recently obtained by Xray and EM. While it is worthwhile doing our best to really understand the constriction mechanism of KcsA, and the present manuscript does an excellent job at that, the ground has shifted and understanding finer points about KcsA constriction has become, unfortunately, not the most prominent issue in the field at the present time.

Let's discuss the current situation about the inactivation of K+ channels. The situation is fairly unsettled. The KcsA channel was the first for which some atomic structure and mechanism, centered on a constriction of the selectivity filter, were proposed. The constricted conformation really does not conduct because the filter is too narrow. More recently a few structures (Xray and EM) for channel mutants known to have more propensity to inactivate have revealed a different conformation of the filter which appears to be dilated toward the extracellular side. This is a conformation that had never been seen previously. Different "camps" co-exist in the K+ channel community about inactivation. Those who were very skeptical about the constricted conformation claim that the new dilated structures is the final truth. While the dilated structures are certainly part of the body of information that we have now, but their significance remains somewhat unclear if anything because of the fact that they are not perfectly occluded and they allow ion conduction! While it is worthwhile doing our best to really understand the constriction mechanism of KcsA, and the present manuscript does an excellent job at that, the ground has shifted and understanding finer points about KcsA constriction has become, unfortunately, not the most prominent issue in the field at the present time.

Author Response:

Reviewer #1 (Public Review):

[…] Collective variable choice:

The explanation for the choice of CVs on page 5 is not sufficient to understand the process and its likely success. How were the most important and unimportant CVs identified exactly? Table 2 on page 19 shows only gate distances, cavity-filter distances and a single variable related to filter structure itself (77 CA - 77 CA) representing a pinch. Is that pinching really the only slow variable associated with inactivation changes in the filter? Why are there no variables, say for carbonyl flipping, E71 or D80 movements or even for ion and water occupancy (although water may be sampled with control of other interactions, such as involving L81)?

CVs for steering simulations were selected based on structural comparisons between the X-ray structures as well as the information about the inactivation available in the literature. These steering CVs were later used as CVs for the string method with the exception of those found to be irrelevant in preliminary string simulations (see methods for details). For example we discarded CVs that would just oscillate freely and thus represent fast equilibrating CVs. We will add additional explanations to the methods section of the manuscript in revisions.

Carbonyl flipping, E71 and D80 movement and SF occupancy were observed in the initial steering simulation to correlate with the 77 CA - 77 CA opening and the opening of the L81-W67 contact. They were not biased but followed the expected path as a consequence of the motion of the imposed selectivity filter constriction. Therefore, they did need not be explicitly biased. The same can be said with respect to water occupancy behind the selectivity filter, which correlates with the opening of the L81-W67 contact.

I understand that the X-ray structure is the one source of information used to define an inactivated structure and is one with just a pinch and no complete carbonyl flipping away from the pore, as has been identified in past studies and discussed as being involved by the authors on page 14. Key changes like carbonyl flipping surely are part of the story and may be slow variables. At the very least, if not part of the CV space, could be analysed.

Indeed, the reviewer is correct in stating that there are molecular motions of interest aside from the ones included in the CV space. Figure 3 and associated supplementary figures indeed extensively investigate the probability distributions of many of those as the system progresses along the inactivation pathway. These results are presented in the section titled “Free energy landscapes offer insights into atomistic-resolution mechanistic details”. Carbonyl flipping seemed to be highly correlated with the 77CA- 77CA distance and this analysis was therefore not presented.

On page 10 the authors discuss possible differences in Amber and Charmm involving the extent to which the 4 subunits change in respect to the L81-W67 water pathway and W67-D80 hydrogen bond, arguing the different results for force field could be to do with different numbers of subunits doing different things. If I understand, the chosen CVs are all tetramer-based distances (including across subunits) and not subunit-based CVs, so that random and incomplete changes may occur to subunits for a given point in CV space.

In fact, some of the CVs represent intrasubunit distances, for example L81-W67 while others represent distance across subunits. This distinction never represented a criterion to select CVs.

There is thus potential for the string to converge on a local minimum pathway with partial changes to its interactions within and between subunits, and may not be a unique global solution. Can the authors please explain whether or not this is possible and what analysis has been done to check it?

This indeed represent a well-recognized shortcoming of all string-based enhanced sampling methods. The string-of-swarms method used herein indeed assumes that there is a dominant minimum free energy path and requires a reasonable starting path. One major advantage of this methodological choice, however, is that the path can be described in high dimension, thus avoiding stark dimensionality reduction as is the case in many collective-variable based methods such as metadynamics.

We do note that though the initial path was the same for the two force fields, the final pathway is different, which tends to indicate that the results do not only depend on the initial path but also on the force field guiding the dynamics of the process.

X-ray endpoints and initial pathway:

The string was created from a pulling/steered MD between existing X-ray structures for the closed (5VKH), partially open (3FB5), fully open (5VK6) and finally inactivated (5VKE) states. The authors write on page 12 that "The block of conduction during inactivation appears to result from pinching at the selectivity filter...", but given the end point was forced to be the X-ray structure with pinching, wasn't this outcome predetermined? This raises a significant point of how much has choice of endpoints predetermined the final states of the string? i.e. How much is an end state actually allowed to draft away from the initial Xray structure. Was a bead placed at the very endpoint and allowed to update via swarms, or was it fixed and all beads just interpolate between those fixed end states? The reason this is important is that it is plausible the inactivated crystal structure with pinching but not other changes (such as complete V76 carbonyl flipping or outer filter splaying), may not be the actual free energy minimum structure for that state and that force field.

The reviewer is right to point out that this observation is most likely a consequence of the choice of the end points of the initial string. The string method assumes that the end points of the string are fairly representative of the initial and final states of the processed studied. In this case, for ease of use, the endpoints of the simulation were fixed. When endpoints are left free to relax, they drift towards the closest minima and make comparisons between force fields, between simulation conditions, etc more difficult.

We do agree that the selection of initial and final states as well as the starting string are important modeling choices. For this reason, we were very mindful and made these choices based on the existing published evidence (available at the time).

We will make these details explicit in a revised version of the manuscript.

Another obvious concern is the possible reliance on the initial pulling procedure used before string optimisation began. Fig.2 Supp 1 shows generally that the Amber path stayed pretty close to the initial steered MD path, whereas Charmm drifted downward away from that path. One could justifiably ask, if a very different initial path was chosen, might different local minimum pathways result, including Amber sampling a path like Charmm? How does one test whether or not the final path has not been trapped in some local trough of free energy? e.g. Imagine starting the Amber string using an initial path like the more diagonal Charmm-like path, or even a more extreme unphysiological one, such as a steered trajectory that initially inactivates before opening the gate. Would the final results be the same? I appreciate the simulations are very expensive and such trials may not be possible, but what evidence is there that the final path has not been trapped away from the global minimum?

As stated above, the reviewer is right to point out the weakness of the method of converging to the closest local minimum free energy path. It is unfortunately computationally infeasible to test many possible paths. For this reason, we chose to initiate our calculations with a pathways based on experimental data; in this case based on available X-ray structures. In addition, it is necessary to contrast the results of the simulation with available experimental evidence: the string method with swarms of trajectories, when aptly used, has a history of bringing useful insights to several biological systems (Lev et al. 2017b; Suh et al. 2019, Fleetwood et al 2021, 2019; McComas et al. 2022).

As already noted, the fact that the two force field yield very different energy landscapes is evident since they would otherwise converge to the same final pathway given the same initial pathway guess.

One test offered by the authors is a set of unbiased MD simulations launched from points on the string. The authors ran 200ns simulations and write on page 5 that "These simulations have the expected stability based on their starting values. This is a good quality test to check the correct estimation of the general features of the free energy surface". While this sounds reasonable, 200ns MD may only be sufficient to begin to explore locally within the solved free energy trough, much like the swarms in the iterations were able to do. My own examination of Fig2 Supp 5 is that some of these simulations linger around the expected states and some drift away within the general trough of sampling, which is a good sign. What those 200ns simulations may not be able to do is escape that trough and see evidence of other possible solutions, beyond what was sampled with the string that was tied to Xray endpoints and trapped in the solution pathway that was already formed after 100-300 iterations. Overall, the string involved 800 iterations of 10ps swarms (80ns around each bead; albeit 32 trajectories in parallel), allowing good local sampling around the beads in the free energy trough, but in terms of ability to diffuse away from that point, only being comparable in contiguous trajectory time to the unbiased MD tests. It therefore would have been interesting to see if longer simulations remain in this trough; though I understand the challenges in running so much MD. Such simulations may, however, lead to exploration beyond what was seen in the string solutions.

We agree with the authors that longer simulations would be very interesting to understand the behavior of the string-of-swarms method and how it behaves for this intricate FES. Note however, that 80 ns divided over 32 trajectories yields an overall trajectory length that is ~two orders of magnitude below a single 200 ns-long simulation. We thus still stand by our statement that the fact that these simulations behave as expected from the free energy landscapes is a good quality check of the CVs and of the resulting free energy landscapes.

Force field effects and origin:

Regarding the effect of the chosen force field, the authors state that "Given that our simulations were conducted under activating conditions, we had expected the open states to be more populated than the closed ones. Simulations carried out at higher pH may be able to resolve this inconsistency". Also running at high pH would be a nice thing to do to prove the method is in fact sensitive to conditions to see a shift in the distribution of states.

Indeed this is the logical next step for future work.

But the question is why were open states not more occupied under low pH and 50mM K+? From my analysis of the figures, the results show that the Charmm force field tends to allow for opening of the channel somewhat (at least with similar free energy for partially and fully open to closed) whereas Amber tends to close the channel more (with more uphill energy as the channel opens than Charmm; Fig 2). i.e. at low pH and 50 K+, isn't the Amber model incorrectly reporting fairly strong bias against opening? Moreover, regarding the free energy of the inactivated state itself, why should we not expect equilibrated channels under activating conditions to eventually fall into an inactivated state, in which case we should expect low free energy of that state (as found with Charmm and not Amber in Fig2), but with a slow rate. While much discussion in the manuscript appears to discuss limitations in Charmm (although on page 12 discussion leans either way), these factors may seem to favour Charmm over Amber.

We would like to thank the reviewer for raising these points. We can only speculate about what might be the reasons for these discrepancies, and we have tried to be as honest as possible in our manuscript and avoid overinterpretation of our results. It is interesting that Reviewer 2 gathered from our data that the AMBER results were more consistent with expectations while this reviewer thought the opposite. This does reinforce our decision to avoid taking sides and present both options. Our personal opinion is currently that both force fields are imperfect at describing all the aspects of the activation-inactivation gates coupling. We will include more discussion in the revisions of the manuscript.

On page 12 the authors explain the possible causes for force field dependence, although this seems limited to ion interactions, glutamate charges and dihedrals. But it would be nice to get a bit more insight into what terms may have influenced the pathway, in particular involving interactions between TM2 and the base of the selectivity filter and hydration behind the filter. Regarding ion interactions, is there a good reason to believe ions are key to the difference seen? i.e. How were ions involved differently in the state transitions involving Amber and Charmm? The authors have noted a role for ion-carbonyl interactions.

We agree that this would be interesting, but judged that this would be better done in a separate study. We do note that the K-carbonyl interactions have been reported as candidates for these discrepancies, as mentioned and cited in the manuscript. Very recent simulations using ab initio MD support that the overstimation of the K-carbonyl interaction is the reason for the low conductance of potassium channels in classical MD, refer to Hui et al. Biophysical Journal, vol. 122, issue 3, p. 520a. We will add this reference in revisions.

It is important that the authors explain which is the two competing models has been used and why. i.e. Off-the-shelf Charmm36 force field includes strong K+-backbone carbonyl interaction, previously seen to promote high ion occupancy, similar to Amber, whereas Lennard-Jones parameters modified to match N-methyl-acetamide and water partitioning (such as early Berneche, Noskov and Roux work) reduce ion occupancy and increase water content inside the filter.

We have used “off-the-shelf” or conventional CHARMM36 as described in the literature cited.

Reviewer #2 (Public Review):

[…] The study is impressive and interesting. However, I have a number of concerns that the authors may wish to address in a revised version of the manuscript.

First, concerning a set of unbiased simulations spawned at different regions of the investigated free energy landscapes, the authors write: "These simulations have the expected stability based on their starting values".

Fig 2.c shows a rather smooth downhill slope in the free energy curve towards the closed state for AMBER , so wouldn't the expected behavior in that case be that all unbiased trajectories end up in the closed state, or at least travel a substantial amount in that direction? However, that is not observed. This should be further investigated.

It is true that this would be the effect we should observe after a significant simulation time. Resorting to 200ns-long simulations, our goal was to test whether the local free energy basins identified by the string-of-swarms method were indeed metastable. If that were the case, we would expect the trajectories to remain within the basins on medium timescales due to the kinetic barriers that would need to be overcome to transfer to other basins. Of course, if simulations were long enough, all basins would eventually be explored by the trajectory with a probability related to the relative free energy of the basins.

Second, "This suggests that stabilization of the partially open state by the removal of bound lipids can explain the increase in open probability" is an odd statement, as "stabilization of the partially open state" means almost the same as "increase in open probability".

It is true that one appears to necessarily imply the other. An increase in open probability could potentially come from two effects: a stabilization of the open state or a destabilization of the closed one. In a two-state system, the two cases are indistinguishable since only relative difference in free energies matter. However, this is a three state system, if one takes as a reference the energy of the inactivated state, there is an effective difference in the physics of the system if a stabilization of the open state or a destabilization of the closed state occurs.

The statement "both force fields yield inactivation barriers that are orders of magnitude lower than what is expected from electrophysiology experiments" seems inaccurate. Perhaps the authors mean "inactivation rates that are orders of magnitude lower" rather than barriers?

Yes, this was a mistake on our part. We will amend the manuscript.

In addition, the assertion "The CHARMM force field, on the other hand, results in landscapes in agreement with the fact that one of the dominant states in activating conditions is the partially open state, as revealed by a combination of ssNMR+MD." seems to hold for the AMBER force field without PG lipids rather than for CHARMM?

AMBER simulations with or without bound PG lipids have a fully open state basin within the minimum free energy path (Fig 4a, 4b) which is not the case for CHARMM (Fig 2b). In that sense, the CHRAMM force field seems to be in better agreement with the ssNMR data. The ssNMR+MD study however suggests that the PO open state basin should be the lowest in free energy. In both cases, however, the C basin is lower in free energy than the PO. We can only speculate about why that may be.

Together with the higher barrier towards the inactivated state as well as covering most known x-ray structures along the inactivation pathway, this would seem to point all in the direction that the studied AMBER force field provides a more faithful picture of the inactivation pathway than CHARMM. I, therefore, find the somewhat inconclusive summary as presented in Fig. 5 a bit uninformative, as it suggests that both mechanisms might be equally likely.

Although the X-ray structures do suggest an AMBER-like path, structural information in isolation is not sufficient to fully understand a phenomenon of dynamical nature. The X-ray structures of metastable structures particularly of open states require the use of engineered mutations and other techniques to trap these states. We of course do not question that a lot of very valuable information can be derived from them, but they should be considered in the context of other computational and experimental techniques. We believe we are very explicit in the text in discussing the weakness and strengths of either possibilities. In fact, we find it interesting that Reviewer 1 gathered from our data that the CHARMM results were more consistent with expectations. This does reinforce our decision to avoid taking sides and present both options. Our personal opinion is currently that both force fields are imperfect at describing all the aspects of the activation-inactivation gates coupling.

Overall, the study would benefit from a follow-up step to become more conclusive. This could be either in the form of the suggested L81 mutation or changing the simulation conditions to inactivating conditions such as low salt, in which case the inactivated state would be expected to become a minimum, which would provide an additional reference point for validation. Either of these would narrow down the spectrum of possible mechanisms.

We absolutely agree with this reviewer. These are great suggestions for further investigations that will definitely be considered in future studies.

Reviewer #3 (Public Review):

[…] The analysis is careful and is state-of-the-art. The results reveal remarkable differences between the CHARMM and AMBER force fields.

Unfortunately, the "elephant in the room" with regards to K+ channel inactivation is the significance of the dilated structures more recently obtained by Xray and EM. While it is worthwhile doing our best to really understand the constriction mechanism of KcsA, and the present manuscript does an excellent job at that, the ground has shifted and understanding finer points about KcsA constriction has become, unfortunately, not the most prominent issue in the field at the present time.

Let's discuss the current situation about the inactivation of K+ channels. The situation is fairly unsettled. The KcsA channel was the first for which some atomic structure and mechanism, centered on a constriction of the selectivity filter, were proposed. The constricted conformation really does not conduct because the filter is too narrow. More recently a few structures (Xray and EM) for channel mutants known to have more propensity to inactivate have revealed a different conformation of the filter which appears to be dilated toward the extracellular side. This is a conformation that had never been seen previously. Different "camps" co-exist in the K+ channel community about inactivation. Those who were very skeptical about the constricted conformation claim that the new dilated structures is the final truth. While the dilated structures are certainly part of the body of information that we have now, but their significance remains somewhat unclear if anything because of the fact that they are not perfectly occluded and they allow ion conduction! While it is worthwhile doing our best to really understand the constriction mechanism of KcsA, and the present manuscript does an excellent job at that, the ground has shifted and understanding finer points about KcsA constriction has become, unfortunately, not the most prominent issue in the field at the present time.

We appreciate the reviewer’s comments and we are also grateful for the contextualization of the current state of the literature with respect to KcsA inactivation.

Although we acknowledge the importance of these new findings and look forward to a lively debate in the literature regarding the importance of this alternative mechanism, this information was not available at the time when this study was started. In any case, for an initial study with a novel technology and with methodological choices such as the force field choice, studying the more established path seems still a valid choice. Of course, the techniques used to study this method can be used to study new hypotheses and contrast them with our current work. This will be an important line of work going forward. We will add further literature discussion to the manuscript and better outline how we decided on the scope of this study.