Investigation of α-Glucosidase Inhibitory Metabolites from Tetracera scandens Leaves by GC–MS Metabolite Profiling and Docking Studies

Authors have brought out some useful information on their studies. But still the rationale behind the study needs to be simplified and explained in understandable way (as a figure or flow chart will enhance reader understanding).

We added a flow chart in the introduction to simplify the concept

Why the overall yield of the extract is low?

The extraction yield in our study (11.34% to 14.60%) is comparable to the yield reported from the leaves of the same plant in previous studies: 10% to 13.69% (Umar et al 2010, doi:10.1016/j.jep.2010.06.016), 10% (Thanh et al, doi:10.1016/S2221-1691(15)30009-5).   

The introduction needs to highlight various other similar studies carried out and their observations.

Some similar previous studies were already highlighted in brief in the introduction (2^nd paragraph). But we added more about their findings as per your recommendation

Explain in detail about the OPLS model

We explained more about the concept behind the PLS model and its recent extension; the OPLS model, in the introduction (2^nd paragraph). We tried to simplify this concept to make it easier to understand for the readers who are not familiar with multivariate statistics

What is the significance of the error in figure 6?

The OPLS model was developed based on 2689 X-variables. Before the determination of the ions (i.e. X variables) contributing to the activity (i.e. Y variable), this huge number needed to be shortlisted. Initially, the variable importance for the projection (VIP) plot was developed. This plot summarizes the importance of the variables both to explain X and to correlate to Y. the VIP plot is sorted from high to low, and shows confidence intervals for the VIP values at the 95% level. VIP values larger than 1 indicate important X-variables. So, we excluded all X variables with VIP values less than 1. Then the loadings column plot was developed. Thereafter, the columns (representing the X variables with VIP values higher than 1) were observed one by one to investigate the error bars. All columns with too large error bars (i.e. the error bar is extending out of the column in both up and down directions) were further excluded (indicate non-significant difference). Now we ensured that the remaining columns (i.e. X variables) were considered statistically significant.

Add details in the figure caption.

Done

What is the pH used to generate the PDB2PQR?

We added the pH value (pH = 6.5)

Explain in detail about the parameters used for Molecular docking experiments.

We explained in more detail the procedure followed for molecular docking.

How many times were the AutoDock experiments repeated?

The docking job was done once, but after your comment we repeated it more 2 times on different occasions and the results of the 3 times are summarized in the table below

Add a figure of the docking profile with protein and the overlap image of all possible docked metabolites

Done (figure 7)

Give the significance of similar and dissimilar residues from the binding studies in table 2 by highlighting … like text in bold

Common interacting residues within the same metabolite class (i.e. fatty acids, anthraquinones and phytosterols) are highlighted by bold and/or underline

Generate a figure showing the best-docked pose of each molecule with the protein in a bound form

It would be too large to be published in the article, so we added it as supplementary data (File S2)

Docked metabolite

Predicted binding energy (Kcal/mol)

Trial 1

Trial 2

Trial 3

Maltose (control docking)

-6

-6

-6

Quercetin (positive control)

-8.8

-8.8

-8.8

Palmitic acid

-6.1

-5.9

-5.9

Phytol

-6.8

-6.6

-6.6

Linoleic acid

-6.4

-6.3

-6.6

α-Linolenic acid

-6.5

-6.7

-6.7

1-Monopalmitin

-5.9

-6.1

-5.9

5-Methoxy-8,8-dimethyl-10-

(3-methyl-2-butenyl)

-2H,8H-pyrano[3,2-g] chromen-2-one

-8.2

-8.2

-8.2

Questin

-8.3

-8.3

-8.3

Emodin

-8.2

-8.2

-8.2

Catechin

-8.4

-8.4

-8.4

α-Tocopherol

-8.3

-8.4

-8.3

Stigmasterol

-9.1

-9.1

-9.1

β-Sitosterol

-8.8

-8.4

-8.8

Cycloartenol

-9.8

-9.8

-9.8

24-Methylenecycloartenol acetate

-8.8

-8.8

-8.9

Page 5: There is no title for neither X-axis nor Y-axis in Figure 1A and B.

Performed as per request

Page 6: I suggest Figure 2 and Figure 3 can be merged as one for better visualization and comparison as they are acquired by the same platform.

Performed as per request

And it would be better if you can include the special compound name for each assigned peak in the legend but not just ask readers to find from Table 1.

Performed as per request

Page 7: Figure 4 and Figure 6 can be merged since both of them were generated by OPLS analysis.

Performed as per request

Score plot is always paired with loading plot or S-plot. Page 7: Figure 5 is mainly a supplementary validation for the prediction function of OPLS model. But the purpose of the model in your study is to identify the target peaks who contribute to the bioactivity or clustering happened in score plot but not the prediction of IC50. So I suggest to move it to Supplementary Information.

Performed as per request. The observed vs predicted plot is attached as a supplementary figure (Figure S1)

Page 8, Figure 6: The label of x-axis for loading plot should be pair-identifier for the peak (X-variable), it can be represented by mass to charge ratio and Rt(m/z/Rt).

However, I would recommend you including S-plot instead of loading plot. Because S-plot is specific to OPLS model and it can provide a more reliable rankling list by removing unrelated variation in the orthogonal direction.

We used XCMS (R version) for preprocessing of the mass spectral data. This software gives a specific name for each feature, based on the m/z and RT. Unlike XCMS online which mentions the RT in minutes, XCMS R package mentions the RT in seconds (e.g. M146T4450). XCMS gives 2 more columns for the m/z and RT (seconds) data individually. So, it would be confusing to the reader if we used such pair-identifier names to represent the features. Instead, we developed a new column for RT in minutes to represent the features, as we think it is more important to the reader than the m/z data.

We used the UV scaling during the multivariate analysis of this study, so it is not possible to develop the S plot for the established OPLS model (the S plot only developed with Pareto or Ctr scaling; source: SIMCA help). However, we tried to use pareto scaling of the data, just to compare the results of the S plot with the results we got from the loading column plot. We got a new OPLS model (1+2+0). The features representing the metabolites questin, emodin, 1-monopalmitin, catechin and 5-Methoxy-8,8-dimethyl-10-(3-methyl-2-butenyl)-2H,8H-pyrano[3,2-g] chromen-2-one were situated on the wing of the S plot, indicating high model influence and high reliability. Furthermore, as we go toward the vertical diagonal, most of the remaining putative metabolites were found. Unfortunately, the different extracts were not discriminated in a good way in the scores scatter plot of this new OPLS model and the R²Y (cum) and Q²Y (cum) for this model were 0.88 and 0.825, respectively. So, we think that the initial OPLS model was better (best describes the variation in X and Y as indicated by the higher R²Y and Q²Y, nice discrimination of the different extracts in the scores scatter plot, identified the same putative metabolites as the S plot of the new model did). Moreover, as per our knowledge, UV scaling is normally used for FT-IR and mass spectra, while pareto scaling is better for NMR spectra. 

Page 4: In the data preprocessing and statistical section, please indicate which kind of scaling and normalization method you use if have applied. Different scaling and normalization methods may lead totally different results. You may refer to Proteome Res.2016, 15(8): 2595-2606 for more details.

We used only the UV scaling. We mentioned it as per your recommendation.

Most importantly, validation part is missing in this manuscript. No matter multivariate data analysis or docking strategy is just for narrowing down the putative candidates, biological validation is necessary for the robustness of your result if it is going to be published. What’s more, most of the metabolites identified by your study are common and commercially available, at least the top 3 of them with highest scores need to be validated by the α-glucosidase inhibitory assay.

The in-vitro α-glucosidase inhibitory potential has already been investigated previously for almost all the commercially-available standards of the putative active metabolites identified in this study, that is why we skipped this step. Palmitic acid, stearic acid, linoleic acid and linolenic acid standards showed IC₅₀ values of 95.27, 88.22, 2.79 and 4.31 µg/mL, respectively (Indrianingsih & Tachibana, 2017). Moreover, β-sitosterol and stigmasterol were isolated from the leaves of Dillenia indica and showed 52.5 and 34.2% inhibition in α-glucosidase activity (kumar et al., 2013). Furthermore, emodin and questin were isolated from Cassia obtusifolia seeds and showed α-glucosidase inhibitory IC₅₀ values of 1.02 and 136.19 µg/mL, respectively (Jung et al., 2017). Similarly, catechin was isolated from the leaves of Muehlenbeckia tamnifoliaa and showed α-glucosidase inhibitory IC₅₀ of 5.5µM).