Cochaperones convey the energy of ATP hydrolysis for directional action of Hsp90

The molecular chaperone and heat shock protein Hsp90 is part of many protein complexes in eukaryotic cells. Together with its cochaperones, Hsp90 is responsible for the maturation of hundreds of clients. Although having been investigated for decades, it still is largely unknown which components are necessary for a functional complex and how the energy of ATP hydrolysis is used to enable cyclic operation. Here we use single-molecule FRET to show how cochaperones introduce directionality into Hsp90’s conformational changes during its interaction with the client kinase Ste11. Three cochaperones are needed to couple ATP turnover to these conformational changes. All three are therefore essential for a functional cyclic operation, which requires coupling to an energy source. Finally, our findings show how the formation of sub-complexes in equilibrium followed by a directed selection of the functional complex can be the most energy efficient pathway for kinase maturation.

The molecular chaperon Heat shock protein-90 (Hsp90) is a dynamic protein and it occupies at least four states, two (using FRET efficiency) open states (0 and 1) and two closed states (2 and 3) which differ kinefically.The authors state that these four states are not coupled to the ATP hydrolysis (based on their previous published work).
In this paper the authors Vollmar et al have used single-molecule FRET and added several cochaperones (Aha1, Sba1, Cdc37) and the client kinase Ste11 to Hsp90.They then showed when the energy of ATP hydrolysis is coupled to Hsp90 conformafional changes and consequently Hsp90 direcfionality.The authors suggest that the presence of Cdc37, Ste11 and ATP lead to changes of Hsp90 kinefic rates, but they do not introduce direcfionality.However, addifion of Aha1 and Sba1 cause direcfionality.Therefore, all three cochaperones are necessary to convey the energy of ATP hydrolysis.This is an interesfing story and provides evidence on how co-chaperones, a kinase client and ATP control Hsp90 dynamic in vitro.There are few issues that the authors need to address in order to strengthen their claims Surely, some Hsp90 mutants need to be tested in order to confirm these states.
Figure 3-Hsp90+Ste11+ATP control data is missing.In the cellular context Cdc37 is requires to for kinase binding to Hsp90.However, in vitro, these two proteins (Hsp90:kinase) are capable of interacfing.
Figure 3-CDC37 phosphomimefic has been used this experiment, however I am sure that the authors know very well that PP5 has to dephosphorylate these sites in order for the client to be released.Therefore, it is unclear how addifion of Aha1 and P23 provides any relevance to understand the "direcfionality".Reviewer #2 (Remarks to the Author): Observafional direcfional cycling between conformafional states in situafions where there is no linear of angular movement is a difficult problem in biophysics.Here the authors aftempt to observe direcfionality in conformafional cyclzin in Hsp90, in presence of various molecular cochaperones.The authors first test their analysis on an arfificial smFRET system where direcfionality of transifion between different FRET states is established by specific combinafions of red and green lasers for specific fime periods.Then, they use their pipeline to analyze smFRET data between FRET pairs placed on each monomer of Hsp90, in presence of various cochaperones.In presence of Cdc37, Ste11, and ATP, the authors observe no direcfionality.This is contrary to current models.However introducfion of two addifional cochaperones -Aha1 and Sba1 -lead to the overall DeltaG of the cycle to be ~-2.1 kT, signifying direcfionality.Overall, this is a neat paper on direcfionality in conformafional cyclin in an important ATPase.The findings may indeed be generally true of GHL NTPases.
The authors perform an arfificial cycling of states by periodically cycling between different lasers for different fime durafions.Although they show in one case that their direcfional cycling (ground truth) matches what the model predicts, the authors really need to establish what the limit are on parameters such as the rafio of rate constants, fime durafions, etc, over which the model works.When would the model fail to predict direcfional cycling even though it is the ground truth?Conversely, how does the model perform when the ground truth is just a stochasfic transifion between 4 states with no direcfional bias?
On similar lines, I feel an important control to perform to verify if indeed the author's claim of direcfionality is true is to use an ATP analogue that cannot be hydrolyzed, or that hydrolyzes very slowly, as cyclical movement in a direcfional fashion between conformafional states should only be possible in presence of ATP hydrolysis.This addifional experiment would also fie in well to the author's discussion on how the conformafional transifions may be linked to substeps in ATP hydrolysis.The author's findings that direcfionality and ATPase rates are only weekly linked is quite interesfing and would benefit for an addifional discussion on week coupling between structural transifions and chemical substeps in the cycles of GHK ATPases in general, as opposed to more well-studied P-loop NTPases.For example, see PMID 22484318.
Reviewer #3 (Remarks to the Author): My review is uploaded as a 'review aftachment'.
This paper delineates an interesting and quite original approach to identify directionality in protein functional cycles.The authors study the conformational transitions of Hsp90 in the presence of various co-chaperones at the single-molecule level.They use FRET signals to study whether detailed balance is maintained or a net flux is created in the set of transitions they observe.They conclude that the addition of three co-chaperones is necessary in order to push the ATP-driven cycle of Hsp90 out of equilibrium.
This work has the potential to become important and influential.However, in its current form it is not ready for publication and requires significant reworking, for the reasons delineated below.
1.The paper is written in a rather sloppy, perhaps hasty manner.Many experimental details are missing.These include details like the way data is treated, including aspects like the leak between channels, direct acceptor excitation, photobleaching etc.The authors use ALEX, but we are not shown how the acceptor excitation channel looks like, so we cannot tell whether, for example, a signal like the one in Figure 3c starting at 20 seconds is not due to a bleached acceptor.We are also not told how exactly the HMM analysis is done.For example, is it performed on each trace separately or on all traces together?Are the reported rates obtained directly from the HMM analysis, or from dwell-time analysis, like some people are doing?And how are parameter errors obtained from this analysis?2. In general, very little data is shown in the paper.In fact, only two traces are shown.The authors do provide all the data in the form of txt files, but surely they do not expect the reader to plot all these files one by one.Please show a significant number of traces, e.g. in a supporting information file.3. Missing from the paper is the very basic demonstration of the fact that a 4-state model is required to treat the data.We are sent to a 2016 paper, but that is not at all sufficient when this point is so central to the current paper.This issue should be directly shown in the paper and the reader should be convinced that indeed four states with two degeneracies are required and/or are optimal.What about other models?What happens if diagonal transitions (0-2 and 1-3) are included in the analysis?4. The major control used by authors to show that they can detect directionality seems to be FLAWED.They conduct an experimental simulation of a directional cycle by alternating laser excitation on a FRET-labeled sample.However, they use a DETERMINISTIC set of times between switches of the excitation, and then analyze the data using HMM analysis.Yet in general, HMM analysis is derived for stochastic (kinetic) models, where there is a certain probability for switching between states and no deterministic switches are expected.I therefore do not see how their control data can be analyzed with HMM.The authors should re-perform the analysis using stochastic switching.Also, they should cover in their control the range of parameters that they get from the Hsp90 analysis, i.e. down to 2 KbT, rather than just showing analysis with larger values (i.e. 3 and 10 KbT). 5.The quantity shown in equation 2 is NOT entropy production.It is a quantity related to the thermodynamic force in a cycle (sometime also called Affinity), and indeed will be different than 0 when equilibrium is breached, but it should be referred to properly.See for example the book Free Energy Transduction and Biochemical Cycle Kinetics by Terrell Hill, starting on page 12.The correct expression for entropy production in a cycle involving discrete states is: Here the Ps are of course the populations of the states.
The affinity CAN be used to study directionality in a cycle, but the correct name should be given.As a matter of fact, it would be interesting to perform not only a calculation of the affinity but also of the proper entropy production and observe that both of them are showing deviation from 0. 6. Panels b and c in Figure 3 show numbers of transitions (the legend refers to 'nomalized amount'-what is this?)-it is not clear how these were calculated (from Viterbi?), but in any case they should be given for all data sets, and errors on these numbers should also be estimated (from repeats of the experiment).This is very important for analysis of the robustness of the results.7. Additional minor issues: a. What is actually shown in Fig. 2b?The sum of the two channels?A 'FRET signal'?If the latter, why is FRET efficiency not shown with the traces of Figure 3? b.Please pay attention to the signs of energies, they are sometimes given as positive and sometimes as negative.c.The first paragraph of the discussion is a bit difficult to understand.For example, it is not clear what the statement "This could for example be provided as binding free energy" means, or for that matter the term 'upstream equilibrium'.Please re-write in a more friendly manner.d.Line 268-"our data clearly shows that there is no linear succession of binding events"-how is this inferred from the data?

REVIEWER COMMENTS
Replies in blue, changes and additions to the text of the manuscript in brown.These changes are also highlighted in the main text in green.
Reviewer #1 (Remarks to the Author): The molecular chaperon Heat shock protein-90 (Hsp90) is a dynamic protein and it occupies at least four states, two (using FRET efficiency) open states (0 and 1) and two closed states (2 and 3) which differ kinetically.The authors state that these four states are not coupled to the ATP hydrolysis (based on their previous published work).
In this paper the authors Vollmar et al have used single-molecule FRET and added several cochaperones (Aha1, Sba1, Cdc37) and the client kinase Ste11 to Hsp90.They then showed when the energy of ATP hydrolysis is coupled to Hsp90 conformational changes and consequently Hsp90 directionality.The authors suggest that the presence of Cdc37, Ste11 and ATP lead to changes of Hsp90 kinetic rates, but they do not introduce directionality.However, addition of Aha1 and Sba1 cause directionality.Therefore, all three cochaperones are necessary to convey the energy of ATP hydrolysis.This is an interesting story and provides evidence on how co-chaperones, a kinase client and ATP control Hsp90 dynamic in vitro.There are few issues that the authors need to address in order to strengthen their claims Thank you for appreciating the importance of this study for co-chaperone, client and ATP control of Hsp90's dynamics.In response to your comments, we have strengthened our claims as follows: Point 1: Figure 2-The authors do not explain how data from single-molecule detection and analysis of dsDNA is applied and capable of quantifying directionality of Hsp90.Surely, some Hsp90 mutants need to be tested in order to confirm these states.
Reply1: We apologize that the reason for the dsDNA experiment was not clearly described: The data from dsDNA is a proof of concept that our method of retrieving directionality from single-molecule data works.Therefore, the underlying ground truth has to be known and this was achieved with a dsDNA sample that contained the same fluorophores as Hsp90.Fig. 2 shows that the ground truth (underlying laser trigger pattern) can well be retrieved from our single-molecule (fluorophore) data.We clarified this point in the revised manuscript by including the following sentences: "To address the challenge of observing directionality in conformational changes, we have developed a new testing procedure that uses external laser triggering to create artificial directionality in experimental data.Therefore, we created an artificial smFRET system using labelled lowFRET dsDNA (Hellenkamp et al. 2018) Point2: Figure 3-Hsp90+Ste11+ATP control data is missing.In the cellular context Cdc37 is requires to for kinase binding to Hsp90.However, in vitro, these two proteins (Hsp90:kinase) are capable of interacting.
Reply2: Thank you for pointing this out.We have now also measured the Hsp90+Ste11+ATP control and added this to the Fig. 3g.We find a small ΔG of -0.8±0.3 kBT, which further supports our conclusion that several cochaperones are needed to obtain a directional cycle.
Point3: Figure 3-CDC37 phosphomimetic has been used this experiment, however I am sure that the authors know very well that PP5 has to dephosphorylate these sites in order for the client to be released.Therefore, it is unclear how addition of Aha1 and P23 provides any relevance to understand the "directionality".Reply3: Thank you for pointing out the importance of phosphorylation, which we have now investigated in more detail and discuss on a few points in the revised manuscript.
More interesting was the second experiment ( Hsp90+Cdc37_noMut+Ste11+ATP+Aha1+p23), where we did not have a clear expectation.Again we do not see a directionality, highlighting the importance of phorphorylation for this system.
We have to work with phosphomimetics here because we work with recombinant yeast proteins produced in E. coli.
Point4: Figure 3d-The error bars for some of the presented data is very high.Has any statistical analysis been used to analyze this data?Reply4: We have revised and extended our statistical analysis.We did two-sample t-tests and corrected for the multiplicity problem by the Holm-Šidák method.We now give selected p-values in Fig. 3g; p values and the testing procedure for all comparisons can be found in Supplementary Data file 3.

Reply5:
The ATPase rate of Hsp90 in presence of Ste11 is shown in Supplementary Fig. 6 in the 4 th bar from the left.The addition of Ste11 does not change the ATPase rate of Hsp90 significantly (One-way ANOVA, p = 0.1).Therefore, Ste11 on its own does not control Hsp90's ATPase activity.In the publication by Sophie Jackson's lab (https://doi.org/10.1006/jmbi.2001.5245)this was shown for human Hsp90 with the GR-LBD as client.In comparison to the yeast homologue, the ATPase activity of human Hsp90 is even much lower.In this paper they also discuss, that partially unfolded clients (like the kinase Ste11 used by us) do not increase human Hsp90's ATPase activity.This already hints to different mechanisms for different types of clients.We added the following sentence to clarify this point: "For human Hsp90, it was previously shown that the sole addition of a client (the ligandbinding domain of the glucocorticoid receptor) is sufficient to control Hsp90's ATPase activity (McLaughlin et al. 2002).However, the presence of equimolar amounts of Ste11 did not increase the ATPase of the Hsp90 yeast homolog used within this paper (Supplementary Fig. 6)." Supplementary Fig. 6: Relative and absolute ATPase rates of yeast Hsp90 (2 µM) in presence and absence of cochaperones Cdc37, Aha1 and Sba1, and the client kinase Ste11.Hsp902 (1 µM, i.e. 2 µM monomers) is a slow ATPase with 1.3 hydrolysed ATP per minute (violet).The addition of free Cdc37 (2 µM, equimolar) further decreases the ATPase rate (dark blue, striped).As the Cdc372-Hsp902fusion shows the same behaviour, the functionality of this construct (dark blue) is proven.Ste11 (2 µM) slightly decreases Hsp90's ATPase activity (green).Additionally, Ste11 hinders Cdc37's ATPase-decreasing effect on Hsp90 of both the freely added protein (middle blue, striped) as well as the Cdc372-Hsp902-fusion (middle blue).Aha1, Cdc37, Sba1 and Ste11 together (1 µM each, lightest blue) strongly increase the ATPase to 2.6 ATP/min.This effect is not achieved by only adding Cdc37 and Aha1 (1 µM each, darkest blue).Statistical significance tested by one-way ANOVA and Tukey post hoc test (described in methods).
Observational directional cycling between conformational states in situations where there is no linear of angular movement is a difficult problem in biophysics.Here the authors attempt to observe directionality in conformational cyclzin in Hsp90, in presence of various molecular cochaperones.The authors first test their analysis on an artificial smFRET system where directionality of transition between different FRET states is established by specific combinations of red and green lasers for specific time periods.Then, they use their pipeline to analyze smFRET data between FRET pairs placed on each monomer of Hsp90, in presence of various cochaperones.In presence of Cdc37, Ste11, and ATP, the authors observe no directionality.This is contrary to current models.However introduction of two additional cochaperones -Aha1 and Sba1 -lead to the overall DeltaG of the cycle to be ~-2.1 kT, signifying directionality.Overall, this is a neat paper on directionality in conformational cyclin in an important ATPase.The findings may indeed be generally true of GHL NTPases.
Thank you for appreciating this study and for pointing out a possible generality for GHL NTPases.

Point1:
The authors perform an artificial cycling of states by periodically cycling between different lasers for different time durations.Although they show in one case that their directional cycling (ground truth) matches what the model predicts, the authors really need to establish what the limit are on parameters such as the ratio of rate constants, time durations, etc, over which the model works.When would the model fail to predict directional cycling even though it is the ground truth?Conversely, how does the model perform when the ground truth is just a stochastic transition between 4 states with no directional bias?Reply1: We have now simulated several more parameters close to what we measure for Hsp90 and also just stochastic transitions (ΔG=0 kBT).For the experiments stochastic switching is not yet possible, but the directional artificial cycling experiment shows that the succession of states can be detected from single fluorophores (see also our reply to point 4a of reviewer#3).Therefore, what remains to show is that ground truth can be recovered, which is well doable by simulations.We now have done five different simulations of singlemolecule dynamics to cover more of the parameter space and to test if directionality can be retrieved in a wider parameter space.As we obtain a ΔG of around -2 kBT from our Hsp90 experiments, we simulated data around this ΔG.In addition, the smallest rate for Hsp90 is 0.0022 Hz, therefore we created a parameter set with -2 kBT and a smallest rate of 0.002 Hz.The ratio of the smallest to the biggest rate is then 1/335.Additionally, we also tested how the analyses behave, when we have equally distributed state populations -their size differing from the populations in the Hsp90-like condition.Finally, we also created a parameter set in detailed balance (stochastic transitions, ΔG = 0 kBT).Altogether, these experiments show that the relevant ΔG and rate constants can well be detected with our experiments and analysis.We have added Supplementary Fig. 1 and the following text to the manuscript: "To quantify the directionality that can be retrieved by our software SMACKS and to compare it to another software (Hidden-Markury (Gebhardt 2021)), we used simulated data with a directionality of 0 kBT, -3 kBT and -10 kBT, respectively (Fig. 2e, f and Supplementary Fig. 1).Additionally, two data sets with the same ΔG value of -2 kBT but with different transition rates (one with rates similar to rates measured for Hsp90, and the other with equally distributed state populations) were simulated (Fig. 2e, f).The data was simulated with MASH-FRET (Börner et al. 2018) and then analysed with SMACKS and Hidden-Markury.Both programmes, which have previously been tested in a comparative study (Götz et al. 2022), were able to retrieve the pre-specified ΔGs.For ΔG = 0 kBT, Hidden Markury obtained a ΔG of 0.505 kBT, while SMACKS resulted in a ΔG of 0.02 kBT.This important control confirmed that for data representing a system in detailed balance, none of the programmes overestimated the ΔG values of the ground truth, i.e. resulted in a falsely positive directionality.For ΔG = -2 kBT (Hsp90 like), Hidden Markury recovered 93 % of the ground truth while SMACKS recovered about 80 %.In all cases with a ΔG ≠ 0, the directionality was underestimated by a maximum of 25% .A certain underestimation is expected, as the result from Hidden Markov Modelling (HMM) is only a lower limit (Godec and Makarov 2023).Altogether, our findings show that we can distinguish systems in detailed balance from those with a directional flux -even if the analysed system does not show strong directionality, i.e. small ΔGs." Our data analysis is mainly limited by the recording rate of our experiment (i.e. 2 Hz).
Point2: On similar lines, I feel an important control to perform to verify if indeed the author's claim of directionality is true is to use an ATP analogue that cannot be hydrolyzed, or that hydrolyzes very slowly, as cyclical movement in a directional fashion between conformational states should only be possible in presence of ATP hydrolysis.This additional experiment would also tie in well to the author's discussion on how the conformational transitions may be linked to substeps in ATP hydrolysis.Reply2: Thank you for suggesting this additional experiment.We added the very slowly hydrolysable ATP analogue ATPγS to Cdc371-Hsp902 + Ste11 + Aha1 + Sba1.As expected, ATPγS induces more closing of Hsp90.A cyclical movement was not observable.In fact, a two state model with only one open and one closed state represented the data best.We show the results of the suggested experiment in the new Supplementary Fig. 5.

Figure 2 -
Figure 2-The authors do not explain how data from single-molecule detecfion and analysis of dsDNA is applied and capable of quanfifying direcfionality of Hsp90.

Figure 3d -
Figure 3d-The error bars for some of the presented data is very high.Has any stafisfical analysis been used to analyze this data?

Point5:
Figure 4-Addition of Ste11 to Hsp90 and measuring the ATPase activity is missing.Previous work by Sophie Jackson's lab has shown that addition of client is sufficient to control Hsp90 ATPase activity.