A Computational Approach for the Discovery of Novel DNA Methyltransferase Inhibitors

Round 1

Reviewer 1 Report

A brief summary
The study reported a valuable insight into the discovery of novel DNMT inhibitors sourced from natural compounds. A pharmacophore-based virtual screening was performed, followed by molecular docking and molecular dynamics simulations. 5 natural-derived compounds, from FDA approved “Epi-drugs” have been tested for the inhibitory activity. Their total DNMT inhibitory activity was evaluated revealing promising results for the derived hits with an inhibitory activity ranging within 30–45% at 100 uM of the tested compounds.

General concept comments
The manuscript is relevant for the field, and it is very well written and structure. The design is appropriated. The authors used high level software and protocols. The introduction is clear and complete, the results are reproducible based on the details given in the methods section. The discussion can be improved. The references are correctly reported, and they are mostly recent. No self-citation detected. Conclusions are ok. Ethic statements and data availability statements Not applicable.

I only have some suggestion to improve the study. Please, improve the figures resolution. Then, it could be useful to see the RMSF plot of the ligands. Moreover, Figure S1 of RMSD plots should be better discussed.

Author Response

Please see the attached.

Author Response File: Author Response.pdf

Reviewer 2 Report

The authors of the manuscript cimb-2902138 reports in silico screening to identify potential inhibitors of DNA methyltransferase. I indicate reconsideration of this manuscript after revision.

1) Please, predict the ADMET parameters of the potential inhibitors and incorporated into the manuscript.

2) In the manuscript it is not clear the reason that the authors selected Sinefungin isoform (PDB: 3SWR) to conduct the molecular docking calculations.

3) What is the experimental evidence that the docked compounds are competitive and not non-competitive or allosteric inhibitors of the assayed enzyme? This is extremely important to validate the molecular docking approach since the potential inhibitors are not structurally similar to the enzyme’s substrate to be considered as competitive inhibitors.

4) The authors did not consider the docked compounds under physiological pH, e.g., the carboxylic acid group of compound 5 is negatively charged under physiological pH instead of neutral. Please, take care and recalculate the molecular docking and dynamic simulations.

5) For the pharmacophore-based virtual screening and physicochemical filters do the authors considered the chemical structure under physiological conditions? It is not clear. Please, clarify it in the manuscript.

6) Molecular dynamics simulations must be carried out in triplicate to validate the in silico approach. Please do it and superpose the RMSD, RMSF, and Rg plots of the triplicate as Supplementary Material.

7) What is the positive control for the in silico calculations?

Author Response

Please see the attached

Author Response File: Author Response.pdf

Reviewer 3 Report

The manuscript (MS) entitled ‘A Computational Approach for the Discovery of Novel DNA Methyltransferase Inhibitors' by Kritsi et al report virtual screening protocol and applying it to identify natural chemicals targeting DNMT enzyme.

The manuscript needs revision before publication, and the authors are kindly asked to address the following comments.

1. Please include experimental activity for structures of training set in Table S1.

2. 20 ns MD simulation is below minimum (100 ns) accepted in literature.

3. The Authors stated in Figure 1 that criteria for the final compounds selection included docking score, presence of crucial interactions and interaction pattern similarity across different docking protocols. How did the Authors overcome well known docking problems discussed in 10.1016/j.tips.2014.12.001? Why did the authors then perform MD simulations? Why they did not estimate free energy of binding from existing MD simulations? It is also far from being perfect, but is much better than just docking score.

4. Can Authors explain why they first performed pharmacophore based virtual screening, and only then physicochemical filters? Which step require more computational resources?

5. The Authors performed docking for 10000 compounds? And from those 10000 compounds they selected only 5 of them based on docking score and interaction patterns? How did they analyze 30000 interaction patterns? Manually or some software was used?

6. Please include in SI results of docking, with docking score.

7. Which criteria had higher priority – docking score or interaction pattern?

8. Could the authors, please, measure the distance between two the most distant carbon atoms of compound 5 in docked pose (Figure 5). Is it above 10 A? If yes, I believe that 10*10*10 A^3 grid box is too small!

9. Why did the Authors run DNMT activity assay at compounds’ concentration 100 um?

10. Can Authors comment the toxicity of 5 proposed compounds and their ADME properties?

11. I guess p-p interactions are pi-pi interactions?

12. More detailed protocol for MD simulation is needed. Which ensemble was used, at which temperature simulation was running, were PBC applied, how long range electrostatic interactions were treated… Please indicate all details, so independent person could repeat your simulations.

Author Response

Please see the attached.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

After a careful reanalysis of the manuscript cimb-2902138, the authors replied to the Reviewer’s questions and clarified some statements of the work. Therefore, I recommend the publication.

Author Response

Response to Reviewer’s 2 comments

Point 1: After a careful reanalysis of the manuscript cimb-2902138, the authors replied to the Reviewer’s questions and clarified some statements of the work. Therefore, I recommend the publication.

Response: We thank the Reviewer for his/her recommendation for publication.

Also, longer Molecular Dynamics Simulations (simulation time = 50 ns) were performed and the results were re-analyzed. Also, additional information for MD simulations were assed in the Materials and Methods section of the revised manuscript.

Reviewer 3 Report

1. I’m aware that there are a lot of publication with 20ns MD simulations. Nevertheless, if the authors check their RMSD analysis, they would see in three cases jump in RMSD, suggesting significant conformational rearrangement. Longer dynamics is needed, or alternatively shorter triplicates.

2. The successful identification of hits among the libraries members proves the validity of the applied screening pipeline. - Maybe, but MD simulations Authors performed does not prove validity of the pipeline.

3. We believe that longer MD simulations constitute a great concept for another publication, targeting more mechanistic data on binding modes in view of hit-to-lead optimization. - I agree, but if the Authors decide to include MD simulations in this manuscript, it should do it properly. The Authors have short MD, and still did not fully analyzed their results.

4. The Authors used docking as one of criteria for prioritizing hit molecules. Docking score has unit of energy, and is in specific sense measure of binding. How the Authors claim that free energy of binding is out of scope of the present study?

5. For how many compounds did the Authors perform docking? The protocol workflow is confusing.

6. How was long term interactions treated? Did the Authors use SHAKE or RATTLE algorithm? Which thermostat was used? Periodic boundary conditions?

7. In experimental part, control (one known DNMT inthibitor) experiment is missing.

Author Response

Response to Reviewer’s 3 comments

Point 1: I’m aware that there are a lot of publication with 20ns MD simulations. Nevertheless, if the authors check their RMSD analysis, they would see in three cases jump in RMSD, suggesting significant conformational rearrangement. Longer dynamics is needed, or alternatively shorter triplicates.

Response: Longer molecular dynamics simulations (simulation time = 50 ns) were performed, as Reviewer 3 suggested. The obtained results are analyzed in the corresponding section (§ 3.2.3). The Supplementary Materials have also been revised accordingly.

Point 2: The successful identification of hits among the libraries members proves the validity of the applied screening pipeline. - Maybe, but MD simulations Authors performed does not prove validity of the pipeline.

Response: Longer molecular dynamics simulations (simulation time = 50 ns) were performed, giving a putative explanation. Hints for molecular dynamics were also added in the Discussion section of the revised manuscript.

Point 3: We believe that longer MD simulations constitute a great concept for another publication, targeting more mechanistic data on binding modes in view of hit-to-lead optimization. - I agree, but if the Authors decide to include MD simulations in this manuscript, it should do it properly. The Authors have short MD, and still did not fully analyzed their results.

Response: Longer molecular dynamics simulations (simulation time = 50 ns) were performed and the results were re-analyzed. Also, additional information for MD simulations were assed in the Materials and Methods section of the revised manuscript.

Point 4: The Authors used docking as one of criteria for prioritizing hit molecules. Docking score has unit of energy, and is in specific sense measure of binding. How the Authors claim that free energy of binding is out of scope of the present study?

Response: The free energy of the binding (ΔG bind) for the proposed molecules were calculated by employing Prime MM-GBSA algorithm using the docked pose retrieved from Glide-XP algorithm. Therefore, a table with binding energy (kcal·mol^-1) and individual energy terms of calculated hDNMT1-selected compounds complexes was added in Supplementary Materials section (Table S6).

Point 5: For how many compounds did the Authors perform docking? The protocol workflow is confusing.

Response: Molecular docking studies was performed to 10,000 compounds, derived from pharmacophore-base virtual screening and physicochemical filtering, as presented in the protocol workflow (Figure 1). 

Point 6: How was long term interactions treated? Did the Authors use SHAKE or RATTLE algorithm? Which thermostat was used? Periodic boundary conditions?

Response: For MD simulations, the thermostat method was Nose-Hoover-chain and the barostat method was Martyna-Tobias-Klein. Also, the relaxation time was defined equal to 1.0 ps and 2.0 ps, respectively. Moreover, the periodic boundary conditions are described in Material and Methods section.  

Finally, during the course of a Desmond simulation, positions and momenta are updated according to the velocity Verlet algorithm. Therefore, the M-SHAKE algorithm was used to constrain bonds containing hydrogen atoms.

Point 7: In experimental part, control (one known DNMT inhibitor) experiment is missing.

Response: At this pilot preliminary phase of our study, we did not consider the use of known inhibitor as compulsory. Our future plans towards hit optimization against DNMT1 will comprise in vitro testing of known inhibitors as well.

Round 3

Reviewer 3 Report

Why did the authors run MD simulations at 300 K, and not at 310?

At which temperature they performed DNMT inhibition assay?

Author Response

ANSWER TO REVIEWER’S COMMENTS

Comment 1: Why did the authors run MD simulations at 300 K, and not at 310?

Response: The focus of the manuscript is the discovery of natural-derived hits which can serve as starting points for further hit-to-lead optimization process seeking for novel Epi-drugs devoid of toxicity. This was accomplished through a solid Virtual Screening protocol workflow (Fig.1).

We chose to run the MD simulations at 300 K, a common temperature for simulating TIP4P solvent model.

A literature survey points to the application of either temperature when performing MD simulations with the DNA methyltransferases (DNMTs). The Reviewer is kindly asked to consider the following list of relevant publications.

Publications heating the system from 0 to 300 K and run the production stage at a constant temperature (300 K)

1. Biochemical Studies and Molecular Dynamic Simulations Reveal the Molecular Basis of Conformational Changes in DNA Methyltransferase‑1.

Fei Ye et al, ACS Chem. Biol. 2018, 13, 772−781; DOI: 10.1021/acschembio.7b00890

2. Insight into the selective binding mechanism of DNMT1 and DNMT3A inhibitors: a molecular simulation study.

Tianli Xie et al, Phys. Chem. Chem. Phys. 2019, 21, 12931; DOI: 10.1039/c9cp02024a

3. Insights into the Inhibitory Mechanisms of the Covalent Drugs for DNMT3A
   Wei Yang et al, Int. J. Mol. Sci. 2023, 24, 12652. DOI: https://doi.org/10.3390/ijms241612652
4. Understanding the R882H mutation effects of DNA methyltransferase DNMT3A: a combination of molecular dynamics simulations and QM/MM calculations. Lanxuan Liu et al, RSC Adv. 2019, 9, 31425–31434 DOI: https://doi.org/10.1039/c9ra06791d

Publications heating the system from 0 to 310 K and run the production stage at a constant temperature (310 K)

1. Molecular dynamics simulations of human DNA methyltransferase 3B with

selective inhibitor nanaomycin A.

Thomas Caulfield and José L. Medina-Franco. Journal of Structural Biology 2011, Volume 176, Issue 2, Pages 185-191. DOI: https://doi.org/10.1016/j.jsb.2011.07.015

2. In silico design of the first DNA-independent mechanism-based inhibitor of mammalian DNA methyltransferase Dnmt1 Vedran Miletić et al, PLoS ONE 2017 12(4): e0174410. DOI: https://doi.org/10.1371/journal.pone.0174410

Comment 2: At which temperature they performed DNMT inhibition assay?

Answer: The DNMT inhibition activity was implemented through Abcam DNMT activity assay kit (https://www.abcam.com/en-gr/products/assay-kits/dnmt-activity-assay-kit-colorimetric-ab113467) according to manufacturer’s instructions.

According to the protocol booklet which can be downloaded from: https://www.abcam.com/en-gr/products/assay-kits/dnmt-activity-assay-kit colorimetric-ab113467#tab=support

the procedure for the enzymatic reaction (page 14) includes the incubation of the strip-well microplate at 37°C for 90-120 min.

In general, 90 min incubation is suitable for active purified DNMT enzymes (which was our case) and 120 min incubation is required for nuclear extracts.

The corresponding Materials & Methods Section (§2.3.1) has been revised to include the sentence: 5 uL of the tested compounds in a final concentration of 100 uM were then incubated in duplicates at 37 °C for 90 minutes.