Inhibition of mutant RAS-RAF interaction by mimicking structural and dynamic properties of phosphorylated RAS

  1. Metehan Ilter
  2. Ramazan Kasmer
  3. Farzaneh Jalalypour
  4. Canan Atilgan
  5. Ozan Topcu
  6. Nihal Karakas
  7. Ozge Sensoy

  • Curated by eLife

    eLife logo

    Evaluation Summary:

    This is potentially an interesting paper in which extensive MD simulations are used to probe the effect of phosphorylation of a tyrosine residue on the conformational ensemble of Ras GTPase. The insights form the basis for a screen of small molecule(s) that disrupt interaction with its target Raf kinase, and predictions are tested experimentally. Overall, the integrated approach is of interest to a wide range of biochemist and protein scientists and could potentially be used to modulate the activities of other proteins.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #2 agreed to share their name with the authors.)

Reviewed by

Read the full article See related articles

Listed in

Log in to save this article

Abstract

Undruggability of RAS proteins has necessitated alternative strategies for the development of effective inhibitors. In this respect, phosphorylation has recently come into prominence as this reversible post-translational modification attenuates sensitivity of RAS towards RAF. As such, in this study, we set out to unveil the impact of phosphorylation on dynamics of HRAS WT and aim to invoke similar behavior in HRAS G12D mutant by means of small therapeutic molecules. To this end, we performed molecular dynamics (MD) simulations using phosphorylated HRAS and showed that phosphorylation of Y32 distorted Switch I, hence the RAS/RAF interface. Consequently, we targeted Switch I in HRAS G12D by means of approved therapeutic molecules and showed that the ligands enabled detachment of Switch I from the nucleotide-binding pocket. Moreover, we demonstrated that displacement of Switch I from the nucleotide-binding pocket was energetically more favorable in the presence of the ligand. Importantly, we verified computational findings in vitro where HRAS G12D /RAF interaction was prevented by the ligand in HEK293T cells that expressed HRAS G12D mutant protein. Therefore, these findings suggest that targeting Switch I, hence making Y32 accessible might open up new avenues in future drug discovery strategies that target mutant RAS proteins.

Article activity feed

  1. Version published to 10.7554/elife.79747 on eLife
  2. eLife

    Author Response

    Public Evaluation Summary:

    This is potentially an interesting paper in which extensive MD simulations are used to probe the effect of phosphorylation of a tyrosine residue on the conformational ensemble of Ras GTPase. The insights form the basis for a screen of small molecule(s) that disrupt interaction with its target Raf kinase, and predictions are tested experimentally. Overall, the integrated approach is of interest to a wide range of biochemist and protein scientists and could potentially be used to modulate the activities of other proteins.

    We would like to thank the reviewers for their valuable comments/suggestions. We provided detailed responses to the questions raised by the reviewers and also submit the revised manuscript where the modified parts are highlighted in yellow. We believe that the original manuscript is improved in light of these changes.

    In the revised version, we (i) increased the number of replicates of MD simulations to four per system studied, (ii) extended previous simulations, which were presented in the original submission, up to 1 µs to test the statistical significance of the main results, and (iii) increased the number of SMDs to 70 per system. We provided time-line data for each replicate of the classical MD simulation in the SI and showed the results obtained from these combined trajectories in the main text along with respective statistical error values. We also repeated calculations such as RMSF, PCA, and the number of waters including the new trajectories and provided updated values/distribution plots in the revised version.

    In general, we obtained similar results to those presented in the original submission except the flexibilities of G60 and Q61. They seemed to display similar behavior among the systems studied as presented in Table 1 upon inclusion of the new replicates. On the other hand, the two residues reached relatively higher RMSF values in the phosphorylated RAS when considering the error values calculated. We presented these values in Table 1 and revised the text accordingly.

    Also, we revised a part in the original submission pertaining to the criterion used for describing the opening of the nucleotide binding pocket in HRASG12D. We noticed that Q61 was not considered for describing the wideness of the nucleotide binding pocket in the references provided. It is also important to mention that the opening of the nucleotide binding pocket, which was described by the distance measured between the Cα atoms of D12 and D34, did not change by the distance measured between the side chain of Q61 and γ-phosphate atom of GTP. Therefore, we dropped the respective distribution of Q61 in the revised version.

    In the application of the PSP methodology, we increased the number of SMD simulations for each of the ligand-bound and ligand-free systems to 70. We also made a more detailed analysis of the results, and we can now rely on not just the qualitative features of the PMFs, but also on the quantities obtained. In particular, the large barrier to cavity opening (ca. 30 kcal/mol) in the ligand-bound form is now clearly shown, and the fact that cerubidine binding leads to a barrierless transition that requires about 1/3 of the energy is demonstrated.

    Reviewer #3 (Public Review):

    In their manuscript "Inhibition of mutant RAS-RAF interaction by mimicking structural and dynamic properties of phosphorylated RAS", Ilter, Kasmer, et al. search for druggable sites in the RAS mutant G12D in computer calculations, and verify their results by experiments. RAS is a major oncogene for various types of cancer and is notoriously hard to target with drugs. Any significant insight into how to find drugs targeting RAS mutants is therefore of high interest. The present manuscript tries to provide such insight, and the connection between simulation and theory appears sound, as the identified compound cerubidine apparently indeed blocks mutant RAS activity.

    As I am an expert in simulations, but not in experiments, I will only focus on the presented computational part. In this function, however, I see some significant problems with the results: The data basis that the authors base their analysis on is quite small (only two simulations of 2.5 µs total simulation time), and from the presented data set I do not see any information on if the results on Y32 dynamics are anecdotal or reproducible. All presented distance distribution plots miss error bars/error ranges, as well as some time course plots that the simulations have indeed converged. So I cannot confirm whether the presented results are valid or if the authors were just lucky in their small data set.

    We would like to thank the reviewer for sharing his comments pertaining to inadequacy of the data used. During the revision period, we performed additional simulations to have four replicates, each of which is about 1 µs, per system. For ligand-bound RAS systems, we ran the simulations until Switch I was displaced from the nucleotide-binding pocket and extended it for an additional ca. 200-300 ns to check if it comes back to its original position. Respective time-line plots of replicates of both ligand-bound and non-liganded systems were provided in Figures S4 – S6 and S11-S14 the SI of the revised MS. We also provided error values in the caption of corresponding figures in the main text. The updated simulation times were provided in the methods section. We presented the total simulation times of each ligand-bound RAS system in the SI.

    To show the convergence of the systems, we provided RMSD profiles for each replicate of the system studied in panel A of Figures S4–S6 and S11-S14. For HRASWT, HRASG12D, and HRASPY32, RMSDs reached a plateau after some time while those of ligand-bound systems did not, as Switch I was highly fluctuating. Importantly, we observed similar behavior in each replicate of the systems so it can be said that the results presented in the original MS are reproducible.

    Interestingly, Switch I was displaced in one of the four replicates of HRASG12D which might lead to the release of the nucleotide from the pocket, thus triggering transitioning towards the apo state. In fact, this observation does not contradict with the findings in the literature.

    It has been shown that mutant RAS can also adopt the apo state albeit with low probability due to its low intrinsic GTPase activity. Therefore, except for KRASG12C, which has a relatively higher intrinsic GTPase activity, either the GDP or GTP-bound state of RAS mutants have been targeted for therapeutic purposes. This information is now included in the manuscript on page 2 of the current version.

    Furthermore, it might be that I have overlooked this information, but this work is not the first finding of druggable sites in RAS (see e.g. review of Moor et al., Nat. Rev. Drug Discov. 2020). The authors should include such a comparison in their manuscript.

    We would like to thank the reviewer for suggesting this comprehensive review. We included it along with the sentence below in the revised version of the manuscript (page 2):

    ‘In these studies, the mutant RAS was targeted directly or in combination with other proteins including SOS, tyrosine kinase, SHP2, and RAF. Also, except for the KRASG12C mutant, the GTP-bound state has been targeted, as RAS mutants either lose their intrinsic or GAP-mediated GTPase activity. However, the intrinsic GTPase activity of KRASG12C is relatively higher than the other mutants which enables targeting the GDP-bound state of KRAS (Moor et al., Nat. Rev. Drug Discov. 2020).’

    We would also like to clarify that we do not claim our study is the first in the field presenting druggable sites in RAS but rather we claim that the study provides a perspective for mimicking the impact of phosphorylation in targeting undruggable mutant RAS.

    Especially the PMF presented in Figure 9 is erroneous, and all arguments based on this plot need to be discarded from the manuscript. From the Methods and Eq. (9), I assume the authors indeed use only the first two cumulants to calculate the PMF. The artificially low PMF with a difference of up to ~800 kcal/mol is a well-understood artefact (see Jäger et al., J. Chem. Mol. Model. 2022) that indicates the breakdown of the second-order approximation in Eq. (9) due to the presence of different pathways in the steered MD data set. This artefact overlays the PMF and obfuscates any information on the true free energy profile.

    We thank the reviewer for these details. The pulling directions remain the same. We indeed found that the absence of enough number of samples along with the breakdown of the second-order approximation due to the presence of different pathways in the SMD data set led to this behavior. We have also included a more detailed error analysis by implementing block averaging (this information now appears on page 18). We hope that the conclusions we draw from the updated PMF curves support the findings to the satisfaction of the reviewer.

  3. eLife

    Evaluation Summary:

    This is potentially an interesting paper in which extensive MD simulations are used to probe the effect of phosphorylation of a tyrosine residue on the conformational ensemble of Ras GTPase. The insights form the basis for a screen of small molecule(s) that disrupt interaction with its target Raf kinase, and predictions are tested experimentally. Overall, the integrated approach is of interest to a wide range of biochemist and protein scientists and could potentially be used to modulate the activities of other proteins.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #2 agreed to share their name with the authors.)

  4. eLife

    Reviewer #1 (Public Review):

    Using molecular dynamic simulations, the authors first explored the impact of phosphorylation on the structure and dynamics of G12D-Ras, a protein of great interest due to its involvement in various cancers. The results then motivated the authors to screen for small molecules that mimic the effect of phosphorylation, which perturbs the conformation of SwI and therefore the interaction with RAF. The prediction was then tested experimentally and shown to be correct. Therefore, the authors have established an approach that is potentially applicable to the modulation of other proteins for biomedical purposes.

  5. eLife

    Reviewer #2 (Public Review):

    In the manuscript, the authors explored RAS-RAF interactions upon phosphorylation of Y32, mutation of G12D, and the ligand binding using molecular dynamics simulations. Their in-silico findings were validated by in-vivo cell assays successfully. The strength of this paper is that from both simulations and experiments, they could show molecular mechanisms underlying RAS-RAF interactions related to the above three factors, which opens the possibility of new anti-cancer drugs. In contrast, the weakness of this paper comes from the technical aspects of their molecular dynamics simulations. 2.5 microsecond simulation is fine as it is a standard simulation length nowadays. However, a few replicated simulations are required to evaluate the statistical significance of the main results. Actually, no statistical error is shown in all the main-text figures. Also, more quantitative analysis is required, if the authors want to discuss PMF (Figure 9, for instance). Not only SMD but also umbrella sampling is recommended to get more reliable PMF for ligand-bound and unbound states.

    I consider that the insight from their MD simulations has sufficient impact on the HRAS research as well as anti-cancer drug developments if technically sound methods are applied to their computational studies.

  6. eLife

    Reviewer #3 (Public Review):

    In their manuscript "Inhibition of mutant RAS-RAF interaction by mimicking structural and dynamic properties of phosphorylated RAS", Ilter, Kasmer, et al. search for druggable sites in the RAS mutant G12D in computer calculations, and verify their results by experiments. RAS is a major oncogene for various types of cancer and is notoriously hard to target with drugs. Any significant insight into how to find drugs targeting RAS mutants is therefore of high interest. The present manuscript tries to provide such insight, and the connection between simulation and theory appears sound, as the identified compound cerubidine apparently indeed blocks mutant RAS activity.

    As I am an expert in simulations, but not in experiments, I will only focus on the presented computational part. In this function, however, I see some significant problems with the results: The data basis that the authors base their analysis on is quite small (only two simulations of 2.5 µs total simulation time), and from the presented data set I do not see any information on if the results on Y32 dynamics are anecdotal or reproducible. All presented distance distribution plots miss error bars/error ranges, as well as some time course plots that the simulations have indeed converged. So I cannot confirm whether the presented results are valid or if the authors were just lucky in their small data set.
    Furthermore, it might be that I have overlooked this information, but this work is not the first finding of druggable sites in RAS (see e.g. review of Moor et al., Nat. Rev. Drug Discov. 2020). The authors should include such a comparison in their manuscript. Especially the PMF presented in Figure 9 is erroneous, and all arguments based on this plot need to be discarded from the manuscript. From the Methods and Eq. (9), I assume the authors indeed use only the first two cumulants to calculate the PMF. The artificially low PMF with a difference of up to ~800 kcal/mol is a well-understood artefact (see Jäger et al., J. Chem. Mol. Model. 2022) that indicates the breakdown of the second-order approximation in Eq. (9) due to the presence of different pathways in the steered MD data set. This artefact overlays the PMF and obfuscates any information on the true free energy profile.

  7. Version published to 10.1101/2022.04.24.489309v1 on bioRxiv