Golgi α-mannosidase: opposing structures of Drosophila melanogaster and novel human model using molecular dynamics simulations and docking at different pHs

Abstract The search for Golgi α-mannosidase II (GMII) potent and specific inhibitors has been a focus of many studies for the past three decades since this enzyme is a key target for cancer treatment. α-Mannosidases, such as those from Drosophila melanogaster or Jack bean, have been used as functional models of the human Golgi α-mannosidase II (hGMII) because mammalian mannosidases are difficult to purify and characterize experimentally. Meanwhile, computational studies have been seen as privileged tools able to explore assertive solutions to specific enzymes, providing molecular details of these macromolecules, their protonation states and their interactions. Thus, modelling techniques can successfully predict hGMII 3D structure with high confidence, speeding up the development of new hits. In this study, Drosophila melanogaster Golgi mannosidase II (dGMII) and a novel human model, developed in silico and equilibrated via molecular dynamics simulations, were both opposed for docking. Our findings highlight that the design of novel inhibitors should be carried out considering the human model’s characteristics and the enzyme operating pH. A reliable model is evidenced, showing a good correlation between Ki/IC50 experimental data and theoretical ΔGbinding estimations in GMII, opening the possibility of optimizing the rational drug design of new derivatives. Communicated by Ramaswamy H. Sarma


Introduction
Golgi a-mannosidase II (GMII) is a glycoside hydrolase (GH) present in the Golgi apparatus of eukaryotic cells, playing a crucial role in the processing of carbohydrates, specifically involved in the N-linked glycosylation of proteins (McCarter & Withers, 1994).This enzyme is very important because the abnormal function of glycosylation pathways has been identified as a hallmark in several cancer types, namely, in their development, progression and metastasis (Zhang, 2021).
GMII is involved in the creation of glycoproteins that carry complex carbohydrates, which are, in turn, implicated in metastasis (Lee et al., 2021;Li et al., 2020;van den Elsen et al., 2001).Thus, GMII inhibition represents a powerful strategy in cancer treatment and has been the focus of many studies in the past three decades (Dube & Bertozzi, 2005;Goss et al., 1995;Lee et al., 2021;Rose, 2012).In fact, clinical tests using swainsonine (SWA), a plantderived indolizidine alkaloid, as a GMII inhibitor, have revealed nanomolar inhibition potency, suggesting an anticancer profile for this small molecule (Goss et al., 1997).However, this same inhibitor demonstrates high activity in the structurally related lysosomal a-mannosidase (LMan), another glycosidase from the 38GH family where GMII is included (Lombard et al., 2014), bringing relevant adverse drug reactions (ADRs) and limiting SWA feasibility.Thereby, the search for selective inhibitors has been pursued since SWA's appearance as a potential chemotherapy drug.The SWA structure and of some known inhibitors are shown in Figure 1, together with the natural substrate of GMII, GnMan 5 Gn 2 (Klunda et al., 2021;Koyama et al., 2020;Lee et al., 2021;Li et al., 2004;Shah et al., 2008).
When searching the literature for inhibition values of some potent GMII inhibitors, we face a discrepancy between published tests, either in the pH medium used or in the enzymatic source employed (Armstrong et al. 2020;Kaln� ık et al., 2023;Klunda et al., 2019;2021;K� o� na et al., 2022).Considering the fact that GMII and LMan perform their functions at different pHs within the cell (GMII, pH 6.6; LMan, pH 4.5) (Ashida et al., 2007;Deschamps et al., 2020;Paciotti et al., 2017), the operating pH cannot be an unimportant subject.In our view, the pH question can even open a new route for the rational design of potent or even selective inhibitors, by unravelling the enzymes' structural differences at each pH, claiming for a critical analysis of how the pH modulates the Golgi structure-activity relationship.
In the current investigation, conformational characterization by molecular dynamics (MD) simulations, together with the determination of several GMII:inhibitor complexes assessed by docking experiments on a novel model of hGMII and of the well-known dGMII, are presented.These studies will enable us to perceive whether the usual eukaryotic models, such as drosophila (dGMII) and Jack bean mannosidase sources, are suitable to pursue the discovery of new hGMII inhibitors.We expect this protocol can serve as a sieve for the design of novel anti-cancer compounds targeting Golgi a-mannosidase II.
In fact, molecular simulations and docking studies have progressed in accuracy over the years (Ciccotti et al., 2022;Tripathi & Misra, 2017), especially in the drug design field, consisting of a powerful strategy to obtain molecular details of macromolecules, their protonation states, interactions and surrounding environment, with application in several diseases by targeting their related macromolecules (Navyashree et al., 2021;Salimi et al., 2022).

Protein structure prediction of human GMII
The hGMII primary sequence of the protein was obtained through UniProt Database (UniProt Consortium, 2018) (entry Q16706).The I-Tasser (Roy et al., 2010;Yang et al., 2015;Zhang, 2008) server was used to perform protein structure prediction (PSP) on the hGMII sequence, generating 5 top models, ranked by a C-score, which varies from À 5 to þ2, where a C-score of a higher value signifies a model with higher confidence.I-Tasser is regularly considered the best server performing template-based modelling (TBM) for protein structure prediction in community-wide CASP (Critical Assessment of protein Structure Prediction) experiments (Pearce & Zhang, 2021;Zhang et al., 2018;Zheng et al., 2021).Very recently, deep machine-learning techniques to generate high-quality conformational predictions are gaining relevance, as in the case of AlphaFold 2 (Pearce & Zhang, 2021).However, I-Tasser remains a robust and efficient choice in PSP, widely used even for complex proteins (Castro et al., 2019;Mahtarin et al., 2022;Zheng et al., 2021).
For the alignment, proteins with the following PDB codes were selected by I-Tasser as threading templates: 3dx4, 5jm0, 2ow6 and 3bvx, all of which are alpha-mannosidases.The best model in confidence, with C-score ¼ À 0.40, was chosen and then subjected to MD simulations at different pHs (6.0, 6.6 and 7.4), previously docked with swainsonine (see details below) in order to equilibrate the model (and activity pocket) in each case.

Drosophila melanogaster GMII models
The model of Klunda et al. (2021), here referred to as "Klunda", was used in situ, as published in their supporting information, with the previous elimination of ligands, and converted to PDBQT format for docking experiments.
The 3BLB (X-ray) (Kuntz and Rose) was used after eliminating water molecules and rotamers in the structure.This structure was used for docking directly after this treatment but also undergoes MD simulations at different pHs, complexed with SWA.

Molecular dynamics simulations and analysis
All simulations were performed using the GROMACS 5.1.4version (Spoel et al., 2005), with the GROMOS 54a7 (Schmid et al., 2011) force field.The amino acid charges were attributed in interactive mode to match APBS (Jurrus et al. 2018) protonation at each pH.Proteins were solvated in water with the simple point charge (SPC) water model in a cubic box with a hydration layer of at least 1.5 nm between the macromolecule and the walls.To achieve system neutrality, counter ions were added to simulation boxes.One stage of energy minimization was performed using a maximum 50,000 steps with the steepest descent algorithm for all models.Two equilibration stages, the first with NVT ensemble (constant number of particles, volume and temperature), followed by the second using NPT ensemble (constant number of particles, pressure and temperature) took place, at 310 K (37 � C) and 1 atm, where position restraints were applied to all protein's heavy atoms.
For the new human model, a production run of 100 ns was performed, and equilibration was quickly reached (RMSD plot, Fig. S1).Thus, all complexes enzyme-SWA, at the different pHs, were submitted to 20 ns of MD simulations, without position restraints, to equilibrate the docked SWA molecule at the active centre and to obtain a perception of cavity availability when a ligand is present.Further information regarding MD protocol and methods is described in SI.
GROMACS tools were used to analyse the overall conformation of the simulated enzymes at the different pHs.Root Mean Square Deviation (RMSD) was used to follow the structural deviations along simulation time and to perceive the conformation equilibration of active centres when complexed with SWA.The cluster analysis was used to obtain the most representative structure of each system for further analysis, such as the visual inspection of activity pockets' size/volume.

Model preparation for docking
From MD simulations at each pH, cluster analysis embedded in GROMACS was used to search for the most representative conformation when SWA is present in the active centre, that is, considering the first 10 ns of simulation time, by using the single-linkage method.In this tool, the RMSD of the active centre's amino acids is calculated, and the structure that minimises the variance compared to the others is chosen to proceed with docking experiments.Only one cluster was generated for all systems.These PDB files, without swainsonine, were converted to pqr (modification of the PDB format including charge and radius parameters) using APBS (Jurrus et al. 2018) server with a PARSE force field, to attribute the partial charges corresponding to each pH of MD simulation.The pqr file was converted to pdbqt through AutoDock Tools software.The zinc coordinates, obtained from the original PDB, were transferred to the pdbqt file using a text editor, and partial charge (2.00) and temperature factor (7.04) were attributed.

Docking options
Each enzyme had its target file (.trg) generated through AutoGridFR v. 1.2 (AGFR) (Zhang et al., 2019).The ligandbinding pocket was calculated through AutoSite 1.1 algorithm with peptide scrolling function option ON (size: 500), and all points in the active site were selected.GridBox was defined as involving the active site pocket (dimension of the box in SI) with a spacing of 0.375 Å and smoothing of 0.5.Options such as the calculation of all affinity maps and the creation gradient inside the receptor were also included.Docking was performed with AutoDockFR (ADFR) with improved genetic algorithm (GA) (Ravindranath et al., 2015), with the following options: number of GA evolution (-nbRuns) 150, population size (-popSize) 150, the maximum number of evaluations of the scoring function per GA (-maxEval) 2,500,000, stop GA evaluation after 15 generations with no improvement in best energy in all clusters (-noImproveStop).

dGMII pH-dependent behavior
Our first approach was to follow the effect of pH on a series of dGMII-inhibitor complexes regarding the binding of experimentally tested ligands/inhibitors.In a preliminary step, the ADFR docking software was validated by comparing the experimental inhibition trend reported by Klunda, and expressed in IC 50 , for a set of molecules to their in silico binding energies calculated in ADFR.For this validation, the drosophila 3D structure made available by Klunda et al. was used (Klunda et al., 2019), as well as Klunda's tested compounds (see Table 1).For analysis, the ligand-receptor binding energy (LR) was considered, from the ADFR pdbqt output file.LR value differs from global score energy since it does not consider the intramolecular contributions of the ligand; it only considers Ligand (L) and Receptor (R) interactions.
Importantly, in this study, the focus is not on the molecules' structure and pharmacophore, but on understanding how these known active molecules change their inhibitory profile consonant with the enzymes' structural profile at different pHs, and further when comparing dGMII and hGMII active centres.Thus, docking screening was run for both targets at different protonation conditions, as described in Table 1.
For the validation set (Klunda's dGMII), the resulting binding energies (LR) (Table 1) were correlated with Log (IC 50 ) by a linear function (Figure 2), LR values (kcal/mol) versus Log (IC 50 ) with R 2 ¼ 0.77.This value represents a good correlation and reproducibility of the reported data (Klunda et al., 2019;2021), even by using a different method/docking software: ADFR in our case, GLIDE program in Klunda's work.The authors do not report their docking binding energies, only the binding mode, which is discussed ahead.The direct comparison between estimated binding energies through docking and experimental activity (K i or IC 50 ) must be read carefully, as in docking, important terms for LR energy such as solvation and entropy are only predicted implicitly (Sethi et al., 2019;Yuriev et al., 2011).Thus, one can only expect to observe a similar trend between theoretical and experimental measurements.Nevertheless, the ADFR software was considered accurate, in the prediction of LR energy trends for the class of iminosugars under study and thus validated.
By testing protonated ligands in Klundas's dGMII model, a correlation of 0.6 was found (SI, page 13), which is overall inferior to their neutral counterparts; this can be explained by the presence of activity site aspartate which acts as an acid-base entity in the catalytic cleavage of mannose.This fact generates uncertainty regarding the protonation state of molecules in the active centre, as the Asp residues could also act in some charge transfer towards the ligands.As so, we took the decision to pursue this study with ligands in the neutral form.
To elaborate further on docking experiments using an Xray structure versus a relaxed structure via MD simulations at different pHs, other structures were considered as targets for docking.Direct use of dGMII X-ray (3BLB) increased the range of binding energies by 1.5 À 2.0 kcal/mol, and the R 2 decreased to 0.66.This structure was crystalized at pH 7.4 and the compounds were tested by using an enzymatic assay at pH 6.0.With this result, it can be concluded that using X-ray structures directly in docking studies can induce bias in the results.
To support the previous conclusion, the amino acids of 3BLB dGMII were protonated or deprotonated according to APBS web service (PDB2PQR) (Jurrus et al. 2018) at pH 6.0, then complexed with swainsonine and submitted to MD simulations for 20 ns.From RMSD and cluster analysis, a middle structure for the complex of 3BLB:swainsonine was generated, on which docking protocol was applied to the same compounds.The LR energies were overall more negative when compared with the X-ray structure and R 2 raised to 0.80.It was observed that the active site of 3BLB after MD is slightly different from the Klunda homology structure regarding the amino acids positioning in the activity pocket, that is, Tyr727 differs in location as well as some interacting amino acids such as Arg228 and Asp341 versus Tyr269 and Trp95 (Figure 3), but globally, the binding interaction between ligand-receptor is the same; the terminal hydroxyl and 3-hydroxyl group complexed with Zn 2þ and Asp92 via a hydrogen bond.The main difference is the hydrogen bonding with the guanidine group as "anchor"; in the case of the Klunda model, a hydrogen bond towards Asp409 was formed; in the case of 3BLB at pH 6.0, the hydrogen bond occurs with Arg876.
As said before, the pH used in many enzymatic dGMII inhibition studies varies a lot from study to study.In recent studies (Deschamps et al., 2020), it was concluded that the pH in the cis-and medial-Golgi apparatus, where Golgi a-mannosidase is localized (Futerman et al., 1990), is comprised of a narrow gap between 6.6 and 6.7.Considering this fact, the 3BLB was resubmitted to molecular dynamics for 20 ns at pH 6.6 complexed with a swainsonine molecule, which was demonstrated to be very stable by RMSD analysis  (Figure 4).Structurally, this enzyme presents a cavity which is not as available as in the human model, but not as restricted as the X-ray structure (Figure 5).That is, the empty space, not hindered by the amino acids' side chains, varies from enzyme to enzyme.Docking studies were pursued with the compounds in Table 1.It was observed that the R 2 of a traced regressing line dropped to 0.60.This result drives our attention to pH, which can greatly influence enzyme inhibition.Experimental enzymatic inhibition studies of the same compounds should be conducted at different pHs to confirm this.Another piece of evidence is the binding trend observed at pH 7.4, which points to less negative LR energies, thus weaker inhibition.
In conclusion, by applying the ADFR docking protocol, a 77% correlation between docking binding energy and IC 50 was observed, which in turn provides confidence in predicting the behavior of compounds for the Klunda model, by using Equation (1) for pH 6.0 in the IC 50 ¼ 7-160 lM range.Similarly, for 3BLB after MD modelling at pH 6.0, the correlation of 80% results in a high assurance of using Equation (2), obtaining IC 50 ¼ 7-160 lM range.Consequently, activity values can be predicted by simply applying the correlation functions obtained for the log (IC 50 ) versus LR, as demonstrated in Equations ( 1) and (2).Full statistics analysis regarding the confidence interval of each regression slope at 95% confidence can be found as supporting information.
IC 50 lM ð Þ ¼ 10 0:243�LRþ4:293 (1) Unfortunately, it is not yet possible to predict IC 50 values in the order of nM, since SWA is the only example, and is itself, too high (less negative) in comparison with its own inhibitory power.In order to define the complete curve and equation (probably polynomial) and to accurately predict the inhibitory profile of GMII, inhibitory data for IC 50 values ranging from 20 nM to 7 lM would be necessary.

The human Golgi mannosidase model
Due to the inexistence of experimental solved human Golgi a-mannosidase structure, protein structure prediction was used to determine the 3D structure of hGMII.This model was generated by using UniProt Database (for the human sequence) and the I-TASSER server (for structure prediction).This model was submitted to MD simulations for 100 ns and quickly reaches structural stability/equilibration (RMSD; Fig. S1).
This same 3D structure was complexed with SWA and submitted to MD equilibration for 20 ns at pH 6.0, 6.6 and 7.4.Regarding the active site interactions with SWA, Figure 5 helps to identify the differences in complexation mode when comparing the human model and drosophila Golgi a-mannosidase, as well as, when looking at a crystallographic structure or MD sampled states.
Although very similar in the overall 3D structure, and very conserved across eukaryotic-type cells (Shah et al., 2008), when looking at the active site pocket and interaction patterns, there are important differences between human and drosophila enzymes (Figures 4 and 5), as well as between Xray structures and MD relaxed models.
In fact, these small differences in the amino acids at the active centre can generate important modifications in the binding pocket, and in the interaction pattern with ligands.In this sense, a BLAST search (Altschul et al., 1997) from 3BLB X-ray sequence was done to compare between human Golgi a-mannosidase II (hGMII) and dGMII: 43% identity was found.Looking specifically at the binding pocket, Figure 4 highlights the discrete differences in residues in both organisms.These amino acids were identified using PyMOL and selecting the residues up to 10 Å distance from the Zn 2þ ion. Figure 4b shows that hGMII and dGMII are not completely superimposable, having some distinction near the zinc ion.Also, we highlight the presence or absence of some amino acids in the vicinity of this ion, in each case: hGMII presents Trp286, Thr575, Ala576 and Tyr 808 facing the activity In the left: Swainsonine-binding site interaction diagrams generated using LIGPLOTþ (Laskowski & Swindells, 2011); black spheres are for C, blue for N, red for O and green for Zn 2þ .In the right: activity pockets with cavities highlighted in blue surface and Zn 2þ as a grey sphere, for (a) hGMII relaxed at pH 6.6, (b) dGMII model from Klunda et al. (2021), (c) dGMII 3BLB X-ray and (d) 3BLB relaxed at pH 6.6.pocket, which are not visible in dGMII, while the latter presents a Val166, Met167, Pro168 and Asp169 sequence near the zinc ion not visible in the hGMII conformation.
The stability of activity centres of modelled hGMII model and dGMII 3BLB at pH 6.6 was followed by RMSD analysis (Figure 4a), which indicates stable enzymes and rapid equilibration throughout the simulation time.
By looking at the activity pockets, the surface cavity perspective helps to distinguish the available space in each case, where for hGMII model at pH 6.6 (Figure 5a) it is observed a spacious room for SWA molecule, whilst, for drosophila 3BLB X-ray (Figure 5c) and the homology model derived from it (Figure 5b), less accessible space is perceived.Yet, in the case of 3BLB relaxed via MD simulations, unfilled space is again observed.These available volumes are responsible for an active centre more or less constrained, which leads to the interaction pattern detected in Figure 5 (left side).This is, the more restrictive and rigid the pocket (b and c) the better SWA is positioned, and the more hydrogen bonds are established.In the other two cases, the available space and dynamics allow a looser interaction, sampling some variations between states, with SWA showing fewer interactions.Importantly, a higher number of interactions does not necessarily reflect more negative binding energies, as the steric hindrance in smaller cavities can negatively counterbalance the favourable interactions.
The comprehension of activity pocket size/volume, giving atomistic details about interactions and the overall availability, is very important to design potent and selective inhibitors and, more than that, suggests that inhibitors should be designed considering a realistic target, in this case, a human GMII model.
As for the docking of Klunda's compounds to the hGMII at different pHs, no linear tendencies were found (Figure 6a).The R 2 for linear regression functions found were 0.32-0.39.The ligand binding position and orientation were similar as on dGMII, and also interacted with Asp and His residues; one hydroxy group of the pyrrolidines complexed with Zn 2þ , and hydrogen bonds were formed with Asp570, Asp177 and His175, for compound 1 (Figure 6b), guanidine group as "anchor" hydrogen bonded with Asp425 and Asp426.But now, this set of molecules is inserted in a looser active centre, which can impair the activity, by not restraining the compounds to a certain pocket area and more stable hydrogen bonds.This finding is a strong indication that it can be erroneous to design and develop inhibitors for hGMII looking at dGMII characteristics, as the small differences in amino acid sequences, overall 3D structure and pocket volume can result in a different outcome in binding energies and affinity.
Looking now at the compounds with enzymatic inhibition with consistent pH, in human target, the molecules of Armstrong and colleagues were used to validate the designed human model and compare them towards Drosophila (Armstrong et al. 2020).The docking protocol developed was applied, generating the results in Table 2.The computational models presented a protonation compatible with the pH used for the enzymatic tests (pH 6.5 enzymatic tests, pH 6.6 model used).
The criteria for binding mode energy selection were as follows: two hydroxy groups complexed with Zn 2þ and the closest ligand orientation to the X-ray structure published in Armstrong's supporting information.
Interestingly, the compounds 14 and 15 consistently demonstrated binding energies very different and not correlated with enzymatic results.That is, in any of our models (hGMII, dGMII and 3BLB) and at different pH and X-ray structures, binding energies were more favourable for compound 15 than 14, which is the opposite of that stated in Armstrong's study (Table 2).Still, these compounds also show some inconsistent inhibitory activity towards the glucosylceramidases GBA1 and GBA2 (van den Berg et al., 2011), which in turn, points to a challenging determination of their experimental and in silico properties.
Considering the aforementioned, and due to their structural similarity and the consistency of docking results in several targets, a possible error in the assignment of these molecules' properties, in reported data, was also pondered.Thus, two correlation plots were generated, the first one neglecting the K i values for 14 and 15 (Figure 7   with the fact that molecule 15 has a more negative binding energy (LR) in all our models.
The exchange of the enzyme inhibition K i and IC 50 values from compound 14 to compound 15 and vice versa generate a linear regression LR vs. Log (Ki) with R 2 ¼ 0.80 to hGMII at pH 6.6.
If docking the same set of molecules, in dGMII relaxed at pH 6.6, a correlation is also found between inhibition (IC 50 and K i ) values and docking energies (LR).The R 2 for the correlation with K i is 0.74 and is 0.71 for IC 50 values.That is, these compounds are possibly active against the Drosophila target also.However, and most importantly, the contrary is not always true: was observed before that when a series of compounds tested in dGMII were docked in the hGMII, the obtained binding energies were not satisfactory (Figure 6).
Again, by applying this docking protocol, the linear regression and R 2 values of 80% for the hGMII model and 74% for the 3BLB after MD equilibration at pH 6.6, indicate that one can extrapolate to predict the behaviour of compounds with a high degree of confidence, using Equations ( 3) and ( 4), respectively, obtaining a K i ranging from 10 to 2200 lM and K i ranging 7-250 lM.Calculated docking binding energies for the same molecules in the hGMII model and 3BLB (dGMII), both equilibrated via MD simulations at pH 6.6.

Conclusions
For many years, human Golgi a-mannosidase inhibition has been studied by using drosophila or Jack bean models, either experimentally or in silico, or opposing Golgi and lysosomal mannosidases when evaluating selectivity (Pol� akov� a et al., 2016).The present study highlights the importance of the in silico use of human models for a more rational design of cancer-fighting inhibitors.
The in silico conformational characterization of a novel model of human GMII as opposed to the well-studied drosophila GMII for docking energy comparison was carried out.Both hGMII and dGMII enzymes together with one dGMII 3D structure (X-ray source) were studied at different pHs.These structures, although highly conserved, have important differences in their activity pocket, especially in spatial availability.Docking protocols developed for hGMII and dGMII 3D structures confirm experimental inhibitory trends, considering the medium pH.Docking energies were found to be strongly related to pH, in all cases.The direct use of X-ray structures can lead to biased results because the pH of enzyme inhibitory studies can be different from the pH of enzyme crystallization, which can change the active centre arrangement and the protonation state of the amino acids.
The docking studies performed in two sets of reported inhibitors against human and Drosophila targets reinforce that the design of novel inhibitors should be carried out considering the particularities of the human model.The proposed 3D structure of hGMII, generated by homology modelling and relaxed at pH 6.6, showed good agreement with enzymatic results and can now be used as a platform to predict the behaviour of other molecules for a more rational design of novel anti-cancer molecules.
Alves, and Castro contributed to the overall design and conceptualization of the project; Castro also acts as supervisor of modelling work.

Figure 2 .
Figure 2. Linear regression between docking binding energies (LR) in dGMII and compound activity (log IC 50 ) of the compounds presented in Table 1, with different conditions represented in different colours (X-ray, homology model, MD relaxed and pH variation).

Figure 3 .
Figure 3. Snapshot of docking poses for molecule 1 in (a) Klunda dGMII homology model and (b) 3BLB structure MD relaxed at pH 6.0.

Figure 4 .
Figure 4. (a) Comparison of active centre amino acids RMSD of hGMII and dGMII (3BLB) at pH 6.6, (b) superposition of representative structures from MD simulation of hGMII (green) and 3BLB (salmon) and (c) comparison of amino acid sequences from the active sites of hGMII and 3BLB.

Figure 5 .
Figure 5.In the left: Swainsonine-binding site interaction diagrams generated using LIGPLOTþ(Laskowski & Swindells, 2011); black spheres are for C, blue for N, red for O and green for Zn 2þ .In the right: activity pockets with cavities highlighted in blue surface and Zn 2þ as a grey sphere, for (a) hGMII relaxed at pH 6.6, (b) dGMII model fromKlunda et al. (2021), (c) dGMII 3BLB X-ray and (d) 3BLB relaxed at pH 6.6.
a), and the second, switching the values attributed by Armstrong and colleagues (Figure7 b).Otherwise, the linear functions will point to a correlation of around 30%, which is incompatible

Figure 6 .
Figure 6.(a) Linear regression between docking binding energies (LR) and Klunda compounds activity (log IC 50 ) in hGMII at different pHs.(b) Binding mode of molecule 1 as a representative example.
Calculated docking binding energies for molecules considered by Klunda in dGMII model, hGMII model, 3BLB equilibrated at several pHs and 3BLB X-ray.