Conformational Landscapes of Halohydrin Dehalogenases and Their Accessible Active Site Tunnels

Round 1

Reviewer 1 Report

The presentation of the results are good, but difficult to evaluate. As a reviewer one must trust these results. As a chemist I would prefer also to see some experimental results form these different enzymes used in reactions with suggested substrates. The manuascript is well written, however there should not be necessary to abbreviate "solvent-accessible 77 surface area to SASA". 

Author Response

Referee 1:

• The presentation of the results are good, but difficult to evaluate. As a reviewer one must trust these results. As a chemist I would prefer also to see some experimental results form these different enzymes used in reactions with suggested substrates

Following referee 1 and 3 recommendations, we have added a new scheme in the paper (scheme 1) that shows the two-step reaction mechanism catalyzed by HHDH (panel a), and a few representative bulky epoxide substrates that the HheG and HheC variant are known to accept experimentally (panel b). We have better detailed the most relevant experimental papers investigating the substrate scope and catalytic activities exhibited by the different HHDH classes in the scheme caption, which reads:

“Scheme 1. A. Reaction scheme of the two-step HHDH catalyzed enzymatic reaction: (1, left) dehalogenation for epoxide formation, followed by the promiscuous (2, right) enantioselective epoxide-ring opening reaction by a nucleophile. Numbering of the residues is based on HheC. B. Representative bulky epoxide substrates accepted by the HheG variant: cyclohexene oxide, limonene epoxide[22], and racemic di-substituted trans-epoxides[24]. Examples of epoxide substrates accepted by HheC are also displayed: epichlorohydrin and epibromohydrin[11].” 

• The manuscript is well written, however there should not be necessary to abbreviate "solvent-accessible 77 surface area to SASA". 

Following referee 1 recommendation, we have removed the SASA abbreviation in the paper.

Reviewer 2 Report

The manuscript “Conformational landscapes of Halohydrin dehalogenases and their accessible active site tunnels” by Miquel Estévez-Gay, Javier Iglesias-Fernández , and Sílvia Osuna

reports a molecular dynamics simulation study, followed by conformational analysis, employing recent dimensionality reduction techniques.

The authors investigate several members of the halohydrin dehalogenase enzyme family aiming for a dynamics-activity relation ship that might explain the promiscuity of those enyzmes w.r.t epoxide-ring opening reactions. They find a “breathing motion” as a common dynamic feature of this enzyme family, but also conformational differences between the analysed systems. Characterisation of the active site accessibility by identification of so-called tunnels reveals their presence to be independent of the “breathing motion” and the width similar across the different systems. Substantial differences are, however, observed for a second, and, in some cases, in the presence of a third, tunnel, which the authors associate with the varying preferences for bulkier substrates.

This work nicely shows how the (conformational dynamics) analysis of a complex data set (here from MD simulations) can be focussed and tailored in a data-driven fashion so as to provide inside in the relationship between structure, function, and dynamics. It also shows, as manifested by the results about the active site tunnels, that unexpected behaviour can be revealed. On the other hand, it becomes also clear, that previous knowledge about the system under study, is important for users choices on initial data/features, thresholds, etc. and last but not least the interpretation that leads to an actual rationalisation.

The manuscript is well written and should be published subject to a few things addressed:

Interestingly, the manuscript does not have a conclusion section. The authors may want to add one, so as to inform the reader about their conclusions drawn from their work.

There must be an error in the last sentence of the MD simulations methods description: “After equilibration in the isothermal-isobaric ensemble (NPT), 4 replicas of 2 ns were run for each system in the canonical ensemble (NVT) for a total simulation time of 6,25 μs.”

The authors should explain what is meant by “ X, Y Z, W Z extracted features” (line 372).

How did the authors decide to use 20 “most informative” TICs? And relatedly, why and how were 4 TICS chosen for the reprentation of the 10 most populated conformations? Ow much (and which) information is lost in such a projection?

Author Response

Referee 2:

• Interestingly, the manuscript does not have a conclusion section. The authors may want to add one, so as to inform the reader about their conclusions drawn from their work.

We thank referee 2 for this suggestion. The last paragraph of the discussion section was actually a nice summary of the main findings of the study. In the new revised version of the manuscript, we have therefore added in the new conclusions section the following paragraph:

“4. Conclusions

The exploration of the conformational landscape of the different HHDH subclasses coupled to the active site tunnel calculations has indicated that the superior activity of HheG towards bulky epoxide substrates is due to the presence of some additional structural elements adjacent to the active site pocket, its higher conformational heterogeneity, and the presence of highly prevalent active site tunnels that present bottleneck radius of ca. 1.9 Å. This is unique to the G subclass, as the rest of the analyzed HHDH are conformationally more restricted and present a reduced number of narrower active site tunnels. Altogether, our study has shown how the HHDH structural dissimilarities influence their conformational landscape, thus impacting their associated active site tunnels, and in turn, their catalytic promiscuity. By means of extensive MD simulations and CAVER analysis, this work has provided key information for rationalizing HHDH promiscuity and for further engineering.”

• There must be an error in the last sentence of the MD simulations methods description: “After equilibration in the isothermal-isobaric ensemble (NPT), 4 replicas of 2 ns were run for each system in the canonical ensemble (NVT) for a total simulation time of 6,25 μs.”

We thank referee 2 for noticing this typo. We have now corrected this sentence, which now reads: “After equilibration in the isothermal-isobaric ensemble (NPT), 5 replicas of 250 ns were run for each system (i.e. 1,25 μs per HHDH subclass) in the canonical ensemble (NVT) yielding a total MD simulation time for all systems of 6,25 μs.”

• The authors should explain what is meant by “ X, Y Z, W Z extracted features” (line 372).

We would also like to thank referee 2 for noticing this additional typo. We have now corrected this sentence, which now reads: “C-alpha coordinates of the aligned protein subclasses at each nanosecond of MD simulation were used as initial features, resulting in 182250000, 168000000, 189000000, 168000000, 192750000 extracted values (features*frames*replicas) for the A2, B, C, D2, and G HHDH subclasses making the statistical analysis unfeasible.”

• How did the authors decide to use 20 “most informative” TICs?

We selected a variance threshold of 25% to select the number of TICs used in the subsequent t-SNE transformations. By selecting the 20 most informative TICs we reached the minimum amount of dimensions in which the 25% of variance threshold was fulfilled. This has now been better detailed in the methods sections: “After applying t-ICA, we further reduced the dimensionality of the data by applying the t-SNE method to the 20 most informative t-ICA dimensions. These 20 most informative t-ICA dimensions describe the 25% of the total variance.”

•  And relatedly, why and how were 4 TICS chosen for the representation of the 10 most populated conformations?

We thank the referee for the comment. In the paper, the fourth slowest TICs were chosen to represent the most populated clusters as a balance between accuracy and clarity of the presented results. These TICS represent the slowest transitions within each HHDH variant and, therefore, the most informative conformational movements. In particular, the 4th slowest TICs describe the 5.3%, 6.0%, 6.5%, 6.3% and 6.5% of the variance in HheA2, HheB, HheC, HheD2, and HheG, respectively. It should be noted that additional less-informative TICS were also analyzed, which led to a loss of clarity of the presented results and a less insightful discussion.  

• How much (and which) information is lost in such a projection?

As mentioned above, the 20th most relevant TICs that describe a 25% of the variance are used for the subsequent t-SNE transformation. By applying the t-SNE transformation to the t-ICA space, less than 75% of the variance described by the TICs is lost in the transformation. Using this two-step protocol, we ensure that the most informative conformational changes are maintained, whereas the least relevant conformational changes are discarded. If more dimensions are included in the t-SNE transformation, the t-SNE data becomes extremely scattered being similar conformations grouped in different clusters, thus adding noise to the analysis. We have better detailed this point in the methods section, which now reads: “The resulting 2D t-SNE space was clustered with the HDBSCAN algorithm[37], with a minimum cluster size of 200 and other default parameters, resulting in 133, 126, 134, 124, 119 clusters for the A2, B, C, D2, and G variants, respectively. By applying the t-SNE dimensionality reduction less than 75% of the variance was lost.   ”

Reviewer 3 Report

This manuscript is well written with a clear and polite style. I have noted, however, that along the manuscript the Authors have placed the references just after the full stop, as if the citations were referred to the next sentence, instead of to the former. My advice is to place citation numbers before the full stops of the sentence they refer to. In the Supplementary Materials , Figure S3  requires a significant page enlargement  to become easily readable. Finally, I would suggest the insertion of a reaction scheme for the enzyme catalytic action, clearly showing the two-step path (dehalogenation-epoxide formation, and nucleophilic attack to the arising epoxide), to help readership, not so familiar with such an enzyme family, better understanding enzyme mechanism and potential technological applications.

Author Response

Referee 3:

• I have noted, however, that along the manuscript the Authors have placed the references just after the full stop, as if the citations were referred to the next sentence, instead of to the former. My advice is to place citation numbers before the full stops of the sentence they refer to.

We have followed referee 3 recommendation and added the citations before the full stops of the sentence in this new revised version of the paper. 

•  In the Supplementary Materials , Figure S3  requires a significant page enlargement  to become easily readable.

We thank referee 3 for this comment. We have splitted Figure S3 into 5 separate Figures that only display the random forest classifier for one of the analyzed HHDH subclasses. In this way, the labels are substantially bigger and easier to read without the need of page enlargement.