Structural description of the receptors
We obtained the structure of selected proteins of T. gondii to evaluate their inhibition by thiazolidinones. The proteins TgPDI, TgRNR2, TgROP18, and TgCDPK1 were selected due to their relevance in T. gondii and because previous in vitro studies point out they could be inhibited by thiazolidinone derivatives. To obtain the structures of TgPDI and TgRNR2 we used homology modelling; the structures of TgROP18 and TgCDPK1 were obtained from the Protein Data Bank.
The TgPDI model presented a structure similarity with the template and conserved physicochemical properties (Fig. 2a). TgPDI sequence had 38% of identity with the template (human PDI) and a coverage score of 0.96. Additional file 1: Figure S1 shows the sequential alignment between TgPDI and its template. The RMSD between the model and the template was 0.27 Å, which indicated that the tertiary structure was preserved. The Ramachandran plot of the model showed that 88% of residues were located in the most-favored regions and 9.9% are placed into the additional regions permitted. QMEAN value is − 3.77. Altogether, results suggest a good quality model (Fig. 2a). The TgPDI model presents semiconserved residues, such as Ile310 and Ser313 (Leu337 and Tyr340 in the template, respectively). These residues were placed at the highly preserved x-linker structure from PDI family, located between b’ and a’ domains. Also, Phe283 and Phe301 (Phe283 and Phe304 in the template, respectively) residues shape a hydrophobic pocket in the a’ domain, where x-linker residues interact with them. Although Val330 (Tyr332 in the template) is not a conserved residue, it does have semiconservative physicochemical properties [40].
The TgRNR2 model conserves at least eight relevant residues in the catalytic site (Fig. 2b). TgRNR2 sequence has 65% of identity with its template (Plasmodium vivax RNR2) and coverage of 0.75. Additional file 2: Figure S2 shows the sequential alignment between TgRNR2 and its template. The RMSD between the model and the template was 0.11 Å, which indicates the tertiary structure was not altered (Fig. 2b). The Ramachandran plot showed that more than 90% of the residues were located in the most-favored regions and none in the generously and disallowed regions. The QMEAN value is − 2.08. These data suggest a good quality model. It can be observed from the model that many residues surrounding the di-nuclear iron center of RNR2 are conserved. There residues are: Glu172, His175, Ser176, Glu234, Phe238, Glu268, His271, and Tyr179. The residues are located in the helices αB, αC, αE, and αF.
The structure of TgROP18 has several kinase residues conserved (Fig. 2c). The catalytic residues from kinases (known as “the catalytic triad”) are present: Lys281, Asp409, and Asp427. We find residues located in the Glycine rich loop (G-loop) that plays an important role to hold ATP (the substrate) in the active site: Gly259, Gly261, Gly262, Phe263, and Val266. There are conserved residues from the A-loop, Phe428 and Gly429 that play a major role in the regulatory mechanism that induces a conformational change that activates the protein for subsequent γ-phosphate transference to the substrate phosphorylation site. Lys411 and Asn414 are located in the catalytic loop. Ala359 is located in the hinge region, which interacts with the adenine ring. Met356, known as gatekeeper, covers a hydrophobic pocket posterior to the ATP location. Finally, Glu300 is a key residue that intervenes in the salt-bridge formation along with Lys281, allowing the breath-like movement of opening and closing the two lobes to let the ATP substrate in and out and enable phospho-transference.
The active site of TgCDPK1 presents kinases catalytic triad: Lys80, Asp174 and Asp195 (Fig. 2d). In addition, the gatekeeper residue of TgCDPK1 is a glycine (Gly128). In general, gatekeeper residues in human kinases have higher volumes; this difference has been exploited for the selective design of drugs against apicomplexan parasites such as T. gondii and N. caninum [25]. The Ramachandran plot for the TgCDPK1 protein was obtained to have a reference from a crystal structure representation, showing 94.7% of residues located in the most-favored regions.
Molecular docking
Thiazolidinone derivatives were docked into the protein targets with Autodock Vina (Table 1). The results of the molecular docking suggest that the 30C derivative presents the best binding affinity in three of the four protein models evaluated (TgROP18, TgCDPK1, and TgPDI), and it is among the five best derivatives evaluated in the active site of the TgRNR2 model. Moreover, the results of the re-scoring tests performed with DSX-score and X-Score (Additional file 3) present a trend similar to the results delivered by AutoDock Vina, suggesting that 30C is one of the best docked compounds in the active sites of the evaluated proteins. Therefore, the 30C molecule was chosen as the thiazolidinone model to evaluate target-ligand interactions.
In the TgPDI model, the 30C derivative interacts with residues along the x-linker structure and a α-helix from the a’ domain (Fig. 3a). A single H-bond is formed between Arg312 and the oxygen from the carbonyl in the thiazolidinone ring; Asp311, Leu314, and Lys333 make π-ion/π-π interactions with the second N-3 aromatic ring substituted; and residues Phe335, Glu336, Ile340, Thr390, Pro391, Leu392, Glu393, and Glu394 interact by Van der Waals forces with the ferrocene group.
In the TgRNR2 model, the compound with lowest score is 34A, which shows H-bonds and π-anion interactions for important and conserved residues surrounding the di-nuclear iron center (Fig. 3b). Asp141 and Ala138 stabilize the 34A through H-bonds aimed at the amine in the thiazolidinone ring; Glu146 and Glu268 show π-anion interactions with the naphthalene moiety of 34A; Tyr179, Phe137, A139, Asn147, Glu172, His175, Ser176, Val233, and Phe238 interacts through Van der Waals forces with naphthalene moiety and thiazolidinone core of the 34A derivative. It is interesting to compare the pose of 34A and 30C. In the case of the 30C derivative, the hydrophobic and Van der Waals interactions between the naphthalene moiety and the residues Glu146, Glu172 and Asn147 are conserved; while amino acids such as Asp141, Ala138, Ala139 and Ser176 interact with the ferrocene moiety instead of thiazolidinone core.
In the active site of the TgROP18, 30C interacts with residues that stabilize ATP (Fig. 3c). Lys281 and Asp427 present π-ion interactions with the naphthalene substituent; Lys411 forms an H-bond with the sulfur atom in the thiazolidinone core; Gly259, Ser260, Gly261, Gly262, Phe263, Ala264, Val266, Phe283, Glu300, Asp362, Ala413, Asn414, Leu416, Gly429, and Thr430 present Van der Waals interactions with the ferrocene, naphthalene and thiazolidinone moieties.
In the active site of the TgCDPK1, 30C interacts with the residues Leu57, Val130, Tyr131, Leu181, Ala78, Glu128 and others with its naphthalene moiety, through hydrophobic interactions. Also, it interacts with Gly60, Lys80, and Glu178 through its thiazolidinone core that presents strong interactions as H-bond. Lastly, 30C interacts with Lys338, Gly58, Glu135, Gly134 and Lys59 with its ferrocene moiety mainly presenting Van der Waals interactions (Fig. 3d).
The general trend of scores suggests that the targets with highest affinity with thiazolidinones are, in decreasing order: TgCDPK1, TgROP18, TgPDI, and TgRNR2. The target chosen for further evaluation was TgROP18 due to the low docking scores and its importance in T. gondii pathogenesis. We chose the protein TgROP18 because it is a key virulence factor for T. gondii [41, 42]. It controls the intracellular proliferation of the parasite [43, 44], it manipulates the host’s immunity and cell apoptosis [45,46,47], and it had been studied in our research group in previous works [48,49,50,51]. Altogether, the complex selected for further evaluation with molecular dynamics is TgROP18/30C.
Molecular dynamics simulations
We performed MD simulations of the complex TgROP18/ATP and TgROP18/30C to characterize their interactions. The system TgROP18/ATP was used as a reference of the natural biological interaction of the protein. We simulated four replicas of each complex. In the MDs of TgROP18/ATP; ATP keeps the conformation of AMP-PNP found in the crystal structure. In the MDs of TgROP18/30C; 30C buries the naphthalene moiety and the thiazolidinone scaffold, improving the contact with the protein and reaching diverse stable structures. One of the final conformations of the TgROP18/30C complex is compact and the substrate has low exposure to the solvent; in the other one TgROP18 is open and 30C is more exposed to the solvent.
In all the MD runs, the systems presented considerable relaxation times before reaching the final structure. For each MD run, we compared snapshots of the trajectory with the initial structure though RMSD in two ways: using the protein backbone (Fig. 4a and b) and using the heavy atoms of the ligand (Fig. 4c and d). All RMSD curves reach a plateau after 100 ns. Therefore, we decided to use the last 200 ns of all MDs in the analyses. It should be mentioned that MD1 of TgROP18/30C complex presents a clear change in the ligand RMSD plot at 167 ns. This jump corresponds to a rotation of a single dihedral angle, which flips the ferrocene group and leaves the interaction unchanged.
Contact frequency analysis
We calculated the frequency of contact of each residue with the ligand, ATP or 30C. It was considered that a contact exists when an atom of a residue was closer than 2.5 Å to any atom of the ligand. In the interest of clarity, residues were grouped according to their spatial localization: disordered linker (residues 211–218), G-loop (residues 258–266), β3-β5 sheet (residues 276–284 and 353–358), hinge region (residues 359–365), gatekeeper residue (Met323), and catalytic loop (residues 411–416). This classification follows the specifications of the crystal structure and is depicted in Fig. 5a.
The most important contacts with ATP and 30C are shown in Fig. 5b, displayed by group and separated by MD run. The residues that contact ATP interact with different moieties of ATP. The residues that contact the adenine portion are: Leu258 and Val266 (G-loop); Met356 and Met357 (β3-β5 sheet); Ala359 (hinge region); and Leu416 (catalytic loop). In addition, residues that interact with the ribose moiety are: Gly259 (G-loop); Asp362 (hinge region); and in some MDs Ala413 (catalytic loop), Gln216 and Ala217 (disordered linker). Lastly, the triphosphate portion interacts with: Gly261 and Gly262 (disordered linker); Lys281 (β3-β5 sheet); and less frequently with Asn414 and Lys411 (catalytic loop).
On the other hand, with respect to 30C (Fig. 5b), the residues that contact thiazolidinone portion are: Ser260, Gly261, and Ala264 (G-loop); and Lys281 (β3-β5 sheet), that interacts with the carbonyl group of thiazolidinone scaffold and carboxyl group of Glu300. In addition, residues that interact with naphthalene moiety are: Val266 (disordered linker); Met356 (β3-β5 sheet); Met323 (gatekeeper); Ala359 (hinge region); and Leu416 (catalytic loop). Lastly, the ferrocene portion interacts with: Leu258 and Gly259 (G-loop); Ala361 (hinge region); and four residues Gln214, Gln216 (disordered linker) and Asp362, Lys365 (hinge region) that eventually get close and form H-bonds with each other. As can be noted, these interactions differ from the ones obtained with molecular docking. This is because the moieties have changed their location from the initial pose, maximizing their interactions and minimizing the solvent exposure of hydrophobic moieties.
We observe that, in general, the contacts of ATP or 30C do not change along the MD runs. To this aim, we divided each simulation into four parts of 50 ns repeated the contact analysis independently in each 50 ns part. In Additional file 4: Figure S3, we show residues that had more than 50% of contact in at least one of the 50 ns portions of the MDs. In general, the contact frequencies are preserved and show no trend with the simulation time. Nevertheless, there are some exceptions that should be mentioned. In the system TgROP18/ATP, there are changes in the residues: Gln216 and Ala217 (disordered linker); Gly262 (G-loop); Lys281 (β3-β5 sheet); Met323 (gatekeeper); and Lys411 and Asn414 (catalytic loop). In the system TgROP18/30C, there are changes in: Gln214 (disordered linker) and Ala361 (hinge region). In addition, some contacts present variance along the MDs without a trend; thus, indicating more labile interactions.
Contact frequencies can vary between MD runs of the same complex. Some contacts are strong and exist with high frequency in three or four MDs of a complex; other contacts are weaker and present high frequency in only one or two MDs. In the case of TgROP18/ATP, the strong contacts are Leu258, Gly259, Met356, Met357, and Ala359; while moderate contacts are Gln216, Ala217, Gly261, Gly262, Val266, Asp362, Lys411, Ala413, Asn414, and Leu416. In the case of TgROP18/30C, the strong contacts are Leu258, Val266, Lys281, Ala359, and Leu416; while moderate contacts are Gln214, Gln216, Leu218, Gly259, Ser260, Gly261, Ala264, Met323, Met356, Ala361, Asp362, Lys365, and Ala413.
Some contacts are unique to a certain ligand; others are common to both. The residues that exclusively contact with ATP are: Ala217 (disordered linker); Gly262 (G-loop); and Lys411 and Asn414 (catalytic loop). The residues that contact with 30C are: Gln214 and Leu218 (disordered linker); Ala264 (G-loop); and Ala361 and Lys365 (hinge region). On the other hand, the residues that present contact with both residues are: Gln216 (disordered linker); Leu258, Gly259 and Val266 (G-loop); Lys281 and Met356 (β3-β5 sheet); Met323 (gatekeeper); Ala359 and Asp362 (hinge region); and Ala413 and Leu416 (catalytic loop). These contacts indicate that the overall binding site is conserved but presents some differences. In general, 30C is more buried allowing stronger interactions the G-loop (Ala264 and Val266) and the disordered linker (Leu218). In addition, this enables the interaction between Gln214 and Lys365 that surround 30C ferrocene moiety when they bind. Conversely, 30C can’t form interactions with polar residues of the catalytic loop (Lys411 and Asn414).
Molecular mechanics-generalized born surface area
Results suggest 30C could bind to TgROP18 acting as a competitive inhibitor of ATP. The binding-free energy of the ligands was estimated through molecular mechanics-generalized Born surface area (MM-GBSA), using the single MD approach without entropy calculation. We obtained an average binding energy of − 11.5 ± 10.3 kcal/mol for ATP and − 29.0 ± 4.4 kcal/mol for 30C. The value obtained for 30C suggests it can bind TgROP18. This value was compared with the ones obtained by Lyne et al. [52]. They used MM-GBSA to evaluate the binding of ligands to four kinases, which rendered linear fits of the predicted-experimental data. From the reported curves, we calculated the pIC50 value that corresponds to − 29.0 kcal/mol (complex TgROP18/30C). The obtained values range from 4.5 to 8.1. These values, even if they present large variance, are in agreement with the ones reported by Carradori et al., who studied the ability of the 30C ligand to inhibit in vitro tachyzoites of T. gondii and reported a pIC50 of 5.3 (IC50 of 5 μM) [13]. On the other hand, the comparison of MM-GBSA values obtained herein suggests that 30C could be a competitive inhibitor of ATP. However, even if the results support this idea, they are not conclusive due to two reasons. First, because MM-GBSA method is sensitive to the ligand charge model [53] and we employed different charge models for ATP and 30C (RHF/6–31 + G* and B3LYP/6-31G* sets, respectively). Second, because the precision of the method makes the difference in values not very significant [54].
Residues that contribute to the binding free energy were obtained with free-energy decomposition per residue [55]. Figure 6 shows schematic structures of the binding sites of ATP and 30C. The most important residues for ATP are: Leu258, Ser260, Gly261, and Val266 (G-loop); Lys281 and Met357 (β3-β5 sheet); Ala359 (hinge region); Lys411 and Leu416 (catalytic loop); and Asp427. The contributions in ATP binding are mainly due to electrostatic interactions. The most relevant residues for 30C ligand are: Leu258, Gly259, Ala264, Arg265 and Val266 (G-loop); Met356 (β3-β5 sheet); Ala359 and Ala361 (hinge region); and Leu416 (catalytic loop). Their contribution is mainly due to Van der Waals interactions and non-polar solvation. The residue Lys281 (β3-β5 sheet) hinders the binding of 30C, due to polar solvation energy. As expected, the residues that contribute to the binding free energy had also been found by the contact analysis. However, the residue Asp427, important for ATP stabilization through the indirect interaction with Mg+ 2, was not found in the former analysis due to the distance threshold.
Principal components analysis
To characterize the structural differences between TgROP18/ATP and TgROP18/30C, we performed the principal component analysis (PCA) method. All the eight MD runs were concatenated and aligned [56]. The covariance matrix was calculated from the positions of the α-carbons of the enzyme, skipping the first and last five residues to avoid noise of ending regions. This left 351 atoms in the analysis and a coordinate space of dimension 1053. The analysis of the accumulated square fluctuations in the PC-modes indicated that these modes contain 75.7% of the total fluctuations of the system (Fig. 7c). In addition, the eigenvalues of the modes decay rapidly. These data indicate that four eigenvectors are enough to explain the variance of the samples.
The trajectories were projected onto the first four PC modes (Fig. 7a and b). The MDs of TgROP18/ATP sample three clusters of points, indicating three states: MD1 samples around {20, − 5, − 15, 10}; MD2 around {13, − 3, 18, 8}; and MD3/MD4 sample around {15, 15, 0, 0}. These states present high overlap in the 1st and 2nd mode and sample more diversely in the 3rd and 4th mode. In general, each MD run populates a cluster of points and no simulations jump between clusters, except for MD2 that samples two small clusters that we classified into only one state (Fig. 7b). This observation suggests that each MD run samples around single conformation of TgROP18 (or two in the case of MD2). In addition, MD4, which was started from a representative structure of MD3, samples around the same conformation as MD3 and therefore validates the sampling of that state.
On the other hand, the MDs of TgROP18/30C sample two states: MD1/MD4 sample around {− 50, 0, 0, 0}; and MD2/MD3 sample around {20, − 10, 0, − 5}. These states, contrary to TgROP18/ATP, present high overlap in the 3rd and 4th mode and samples more diversely in the 1st and 2nd mode. As in the case of TgROP18/ATP, each MD samples around a conformation of the protein. In this system, MD4 was initiated with a representative structure of MD1. We note that the regions sampled by MD1 and MD4 have overlap and resemble to some extent, but they can be unequivocally distinguished in the 2nd and 4th PC-mode. The inspection of the structures of both MDs reveals these differences are due to the position of the disordered linker, which is intrinsically highly mobile, and the β4a/β4b sheet [24], which is far from the binding pocket.
Each PC-mode can be described by concerted changes in the sub-structures of TgROP18. To describe the motions related to the PC-modes, we have generated coordinates of structures that sample along the PC-modes (Additional file 5: Figure S4). The TgROP18 structure presents two domains termed N-lobe (residues 209–360) and C-lobe (residues 186–208 and 361–548). The 1st PC-mode represents a hinge motion between the N-lobe and the C-lobe, where domains are near rigid bodies (Additional file 5: Figure S4a). The 2nd PC-mode represents three concerted changes: a lid-like motion of the β4a/β4b sheet (residues 337–342 and 345–350), the rotation of the C-helix (residues 283–310), and the displacement of the disordered linker (Additional file 5: Figure S4b). The 3rd PC-mode also represents concerted changes: a lid-like motion of the C-helix, a rearrangement of the disordered linker, and the opening of the loop formed by residues Val437-Arg451 (Additional file 5: Figure S4c). Finally, 4th PC-mode represents a pseudo-rotation that involves the G-loop, the β4a/β4b sheet, the C-helix, and the disordered linker (Additional file 5: Figure S4d).
The structural representation of the PC-modes allows an intuitive characterization of states of TgROP18/ATP. In this sense, the sampling of TgROP18/ATP along the 2nd, 3rd and 4th modes can be thought as natural motions of the Michaelis complex. These movements involve the displacement of the disordered linker, which is in agreement with the poor definition of electron density in that region of the crystal structure. In addition, the lack of sampling along the 1st PC-mode indicates that that in the MDs of TgROP18/ATP the enzyme does not undergo a wide inter-lobe hinge motion.
Dynamic domains analysis
PCA revealed TgROP18/30C has two states: one that overlaps with the conformations sampled by TgROP18/ATP and another clearly different. In addition, we observed that the PC modes involved rigid-body like rearrangements of the substructures in TgROP18. Therefore, we aimed to identify dynamic domains that could explain the difference between the conformations sampled by TgROP18/30C and those sampled by TgROP18/ATP. To this aim, we obtained 50 structures of MD1/MD4 of TgROP18/30C, 50 from MD2/MD3 of TgROP18/30C, and 50 structures from all the MDs of TgROP18/ATP. These groups of structures were compared with the DynDom software [57], that models the difference between conformations as screw motions.
The comparison of the state sampled by MD1/MD4 of TgROP18/30C with the one sampled by TgROP18/ATP predicted three domains (Fig. 8a): the C-lobe (residues 189–215, 310–326, and 360–539), the C-helix subdomain of the N-lobe (residues 283–302), and the remaining residues of the N-lobe (residues 223–282 and 327–359). Upon the screw motion, the domains presented an RMSD of 1.6 Å, 1.4 Å, and 1.4 Å for the C-lobe, C-helix, and N-lobe, respectively. The low RMSD values indicate the modelling is successful. The screw motion between the C-lobe and the N-lobe (Fig. 8b) presents a closure of 95% and requires a rotation of 34°. The residues identified to bend are from the disordered linker (residues 209–223), the hinge region (residues 359–360), and residues 326–327. On the other hand, the screw motion between the C-lobe and the C-helix (Fig. 8c) has a closure of 97%, requires a rotation of 58° and a translation of 0.9 Å. The residues that bend in this motion are from the C-helix (residues 302–311). Lastly, the motion between the N-lobe and the C-helix (Fig. 8d) has a closure of 87% and requires a rotation of 40° and a translation of 2.4 Å. In this case, the residues that bend are from the β3-β5 sheet (residues 281–282).
The comparison of the state sampled by MD1/MD4 of TgROP18/30C with the one sampled by MD2/MD3 of TgROP18/30C renders results comparable to the ones shown above. Two domains are found: one corresponding to the C-lobe (residues 190–211, 308–325, and 360–539), and one corresponding to the entire N-lobe with the C-helix included (residues 222–307 and 326–359). In this case, the domains presented an RMSD of 1.6 Å and 2.5 Å for the C-lobe and N-lobe, respectively. The screw motion between the domains has a closure of 95%, requires a rotation of 34°, and a translation of 1.3 Å. The residues that bend are from the disordered linker (residues 211–222), the hinge region (residues 359–360), the C-helix (residues 307–310), and residues 325–326. We find that the domain that corresponds to the C-lobe is analog to the one found in the previous DynDom analysis. In addition, the high RMSD value of the N-lobe domain indicates the model is not fitting that well. Also, the N-lobe domain is analog to the merging of C-helix and N-lobe domain found previously. That being said, the data indicates that the domains identified in the two comparisons are approximately equivalent.
In addition, we checked if domains could be defined with several other combinations of MDs. The comparison of the MDs of TgROP18/ATP with each other does not render domains. The comparison of the state sampled by MD2/MD3 of TgROP18/30C and the one sampled by TgROP18/ATP also does not render domains. This indicates that, in these cases, the intradomain and interdomain displacements are comparable and, thus, no domains can be identified.
Finally, the PC-modes can be related to the DynDom analysis. The 1st PC-mode, which accumulates 58.4% of the sampled fluctuation, represents the C-lobe/N-lobe hinge movement found in both DynDom analyses. The 2nd, 3rd, and 4th PC-modes, that include an additional 17.3% of the fluctuations, include movements of the disordered linker, the C-helix, and the β4a/β4b sheet. This is also in agreement with the DynDom analysis. First, because C-helix was recognized as a domain; and second, because the disordered linker was identified as a bending region, which is consistent with the diverse displacements it can undergo.
Backbone correlation analysis
To further characterize the systems, we performed correlation analysis between α-carbons atoms. For each system and each MD run, we obtained the displacement vectors between frames of the α-carbons. Then, the correlation per residue-pair was computed as the average of the dot product of the displacement vectors. Correlations matrices are displayed as heat maps in Fig. 9a. The correlation maps of TgROP18/ATP and TgROP18/30C are very similar. The correlation values are found between − 0.3 and 0.8. The presence of inverse diagonals with high correlation reflects the multiple antiparallel β-sheets that the structure of TgROP18 contains; on the other hand, the increase of correlation nears the diagonal of the matrix marks α-helices in TgROP18.
To assess the effect of 30C, we calculated the difference between the correlation matrices of TgROP18/ATP and TgROP18/30C (Fig. 9b). The difference matrix shows few non-zero spots that amount to values from − 0.20 to 0.25 (circles in Fig. 9b). This implies there are no major differences in the correlations of single residue pairs.
On a different approach, the difference matrix was divided into two regions (Fig. 9b): the ones that correspond to residues from the same lobe (i.e., regions with intra-lobe correlations) and the ones that correspond to residues from different lobes (i.e., regions with inter-lobe correlations). The two white triangles represent the intra-lobe correlations of N-lobe (residues 209–360) and a part of the C-lobe (361–548); the black square represents inter-lobe correlations. Interestingly, the spots in the matrix show intra-lobe correlations increases while inter-lobe correlation decrease. In the interest of clarity, Fig. 9c displays these changes of correlation in the structure of TgROP18 indicating when the pair-wise correlation increases (orange) or decreases (blue).
This analysis shows some regions are more affected by the change of correlation. These regions can be visually appreciated in Fig. 9c as the nodes with many connections. The regions that are affected the most are, in decreasing order: the C-helix, the activation loop, the catalytic loop, and the β4a/β4b sheet. On the other hand, it is interesting that some changes of correlation are found in residues that are far from the binding site, such as the correlation between residue Cys478 and residues Leu492-Pro498, from the C-lobe. A careful inspection of the correlation matrices of MD1/MD4 and MD2/MD3 TgROP18/30C (data not shown) indicates that the changes in correlations are mainly due to the state samples in MD1/MD4. The conformation sampled in MD2/MD3 only corresponds to the decrease in correlation of the G-loop (small blue spots in Fig. 9b) while the conformation sampled in MD1/MD4 corresponds to the rest of the changes of correlation (large blue spots and orange spots, in Fig. 9b). These are consistent with previous analyses. PCA and DynDom analyses showed that 30C induces a more open conformation in MD1/MD4 of TgROP18, while it does not change the conformation of TgROP18 in MD2/MD3.