Quantification of PR2 structural deformation induced by drug resistance mutations
We focused on a set of 30 drug-resistant mutants containing from one to three mutations (Fig. 1a). The 3D structure of each mutant was built using FoldX software [39] with five replications (see Methods), resulting in a set of 150 mutant models. To compare the mutant and wild-type structures, we computed the all-atom RMSD, denoted as RMSDaa, between the mutant structure and the three minimized wild-type structures (3EBZmini). The five structures of each mutant could exhibit large RMSDaa (i.e., different conformations) (Additional file 1: Figure S1). As expected, single mutants (i.e., those with only one mutation) exhibited a smaller average RMSDaa than the other mutants, indicating that introducing a single mutation induces less structural change than introducing several mutations (Additional file 1: Figure S1). In addition, there was a link between the number of mutations and structural diversity in terms of RMSDaa. Indeed, the three types of mutants (single, double and triple) did not exhibit the same variability in terms of RMSDaa (Bartlett-test p-value = 2.10− 11): triple mutants showed more conserved structures than the other types of mutants. Thus, compared to single or double mutants, the insertion of three mutations induced more structural diversity relative to the wild-type structure, but the five modeled mutant structures were more similar to each other.
Detection of structural asymmetry in the wild-type and mutant structures of PR2
To explore the structural effects of the studied drug resistance mutations, we compared the structural asymmetry in the three 3EBZmini structures and the 150 modeled mutant structures using an approach based on a structural alphabet [31].
Characterization of structural asymmetry in the three 3EBZmini structures
We determined the structural symmetric and asymmetric positions in the three 3EBZmini structures and their location in the 13 PR2 regions defined in [32, 37, 38] (Additional file 2: Figure S2). Half of the positions (53% of positions) were detected as symmetric (i.e., showing similar local conformations in the two chains in the three 3EBZmini structures). The entire PR2 structure was sampled, particularly the fulcrum, flap, and cantilever regions (Fig. 2a). Thus, the conserved local conformation in both monomers at these positions is important for PR2, particularly for pocket residues 28, 30, 53, 81 and 82, which could be important for ligand binding. The three 3EBZmini structures contained between 34 and 38 asymmetric positions, with 28% of the positions showing asymmetry in the three 3EBZmini structures. These positions were located throughout the structure, particularly in the α-helix and cantilever regions (Fig. 2a). This asymmetry conservation suggests an important role of these positions, particularly for the elbow and flap positions (positions 40–42, 50, 51, 58), which could be important for PR2 deformation.
A total of 18 positions did not exhibit the same asymmetry status in the three 3EBZmini structures (Fig. 2a). These 18 positions were not more flexible (in terms of B-factor value) or more accessible (in terms of accessible surface area (ASA) values) residues relative to other positions (Kruskal-Wallis test p-values = 0.25 and 0.46, Additional files 3 and 4: Figures S3 and S4). This suggests that the asymmetric variability of these positions does not result from the intra-flexibility of PR2. However, we have previously shown that these 18 positions exhibit different conformations in the 18 available structures of PR2 complexed with different ligands [32]. Thus, these positions could modify their local conformation to adapt to different ligands.
Characterization of structural asymmetry in the PR2 mutant set
On average, a mutant structure contains 31.8 ± 4.4 asymmetric positions, which is close to the number of asymmetric positions in the three 3EBZmini structures. The five structures of a given mutant do not always have the same number of asymmetric positions and can exhibit few common asymmetric positions, such as the G48V mutation (m5); see Additional file 5: Figure S5. The variability of the number of asymmetric positions per mutant does not depend on the number of mutations (Bartlett test p-value = 0.14). As expected, a link was found between variability in terms of RMSDaa and both (i) the variability in terms of the number of asymmetric positions (Pearson correlation coefficient = 0.73) and (ii) the number of common asymmetric positions between the five mutant structures (Pearson coefficient between the standard deviation of the RMSDaa value for the 5 mutant conformations and the number of common asymmetric positions = − 0.65). Thus, the mutants exhibiting greater diversity in terms of RMSDaa corresponded to the mutants showing five structures exhibiting different numbers of asymmetric positions with few common asymmetric positions.
To characterize the structural asymmetry in the mutant set, we then computed the asymmetry occurrence (AO) for each position (i.e., the number of mutant structures exhibiting asymmetry for a considered position). A total of 26% of the positions that were symmetric in all mutant structures were also symmetric in the three 3EBZmini structures. This indicates that the resistance mutations do not affect the structural symmetry of these positions, particularly the mutations occurring at some of these symmetric positions (K7R, V10I, V71I, I82F, and I82L). These positions were frequent in the fulcrum, flap, and cantilever regions and absent in the Nter, Cter, elbow, R3, and R4 regions (Fig. 2b). Five of them (28, 30, 53, 81, 82) were located in the binding site, confirming the important role of these positions in ligand binding.
In the mutant set, 36 positions correspond to overrepresented asymmetric positions and were denoted as ORasym positions. Their asymmetry does not arise at random. These ORasym positions are located throughout the structure except in the R1 and catalytic regions (Fig. 2b). These positions did not consist of more flexible (in terms of B-factor) or exposed (in terms of ASA) residues on average than other asymmetric positions (T-test p-values = 0.40 and 0.53, respectively, additional files 3 and 4: Figures S3 and S4). Seventy percent of the ORasym positions were also asymmetric in the three 3EBZmini structures. These positions were particularly common in the α-helix, flap, cantilever, and fulcrum regions (Fig. 2b). Thus, the studied drug resistance mutations do not modify the structural asymmetry of these positions, reinforcing the important role of the structural asymmetry of these positions. These overrepresented asymmetric positions could be important for PR2 structure or activity, particularly the four residues belonging to the dimerization region (4, 5, 97, and 98) and the five pocket residues (23, 32, 47, 50, 80).
The remaining 11 ORasym positions were symmetric in the three 3EBZmini structures. In addition, two positions (40 and 41) were asymmetric in the three 3EBZmini structures and were not overrepresented in the mutant set. Among these 13 positions, four were close (44 and 80) or corresponded (47 and 90) to mutated positions. Thus, these drug resistance mutations could be responsible for the changes in asymmetry at these mutated positions and their nearby residues, and they could modify the asymmetry of more distant residues.
Link between drug resistance mutations and changes in asymmetry occurring in mutants
For each mutant, we determined how many of its five structures exhibited a change in asymmetry at each position relative to the three wild-type structures. The number of changes in asymmetry observed for each mutant varied from 6 (mutant m2) to 36 (mutant m5) and did not depend on the number of mutations (P-value of the Kruskal-Wallis test = 0.55). Figure 3 presents the network connecting a mutant with its asymmetric positions. We observed that some changes in asymmetry occurred at positions connected with many mutations (which correspond to central nodes in the network), while others were connected to few mutations (which correspond to external nodes in the network, Fig. 3). For example, changes in asymmetry at positions 40, 41, 33, 18, and 98 were observed in more than 20 mutants (Fig. 3). These changes in asymmetry were not specific to certain mutations, suggesting that they were not induced by mutations. In contrast, structural backbone asymmetry at positions 6 and 78 was observed only in mutants I54M/I84V (m17) and K7R/I46V/L99F (m26), respectively, while such asymmetry at position 62 was observed in mutants I84V (m12), G48V (m5), and I84V/L90M (m23). The loss of structural asymmetry at positions 12, 64, and 75 was only observed in mutant G48V (m5), but the five structures of these mutants did not exhibit this loss.
To highlight the changes in asymmetry that are putatively induced by resistance mutations, we studied the conservation and location of structural changes in all structures containing at least one of the nine mutations observed in multiple mutants (K7R, I46V, I54M, V62A, V71I, I82F, I84V, L90M, and L99F); see Fig. 4. We noted that some changes in asymmetry occurred in all mutants exhibiting a given mutation. For example, a structural change at position 83 was observed in all mutants exhibiting the I82F or I84V mutation. The location of conserved structural changes in the PR2 structures allowed us to differentiate two types of changes in asymmetry: those occurring far from a mutated residue and those occurring close to a mutated residue, which were putatively induced by mutation.
Asymmetric changes occurring far from mutated residues and, thus, putatively not related to mutations
Figure 4 shows that the nonspecific changes in asymmetry occurred at positions 40 and 41 in most mutants, but they were located far from mutated residues. As previously assumed, the high frequency of these two changes in asymmetry suggests that they are not induced by a mutation. This was confirmed by the fact that they occurred at two exposed residues located among the elbow residues (Additional file 4: Figure S4). Thus, the loss of asymmetry observed at positions 40 and 41 could be induced by residue flexibility. In the mutants with K7R, I46V, I54M, I84V, or L99F mutations, changes in asymmetry at positions 40 and 41 were accompanied by the loss of asymmetry at positions 18, 21, 58, 59, and 60 (Fig. 4). As these positions are located close to flexible residues 40 and 41, we suggest that these changes in asymmetry could be induced by the changes in asymmetry at positions 40 and 41. The nine mutants with the I54M mutation also presented a loss of asymmetry at position 33, which was also observed in all mutants exhibiting either the I84V or L90M mutation. Although this change in asymmetry was specific to certain mutations, it was not located close to the mutated residues (Fig. 4). The seven mutants containing the L90M mutation exhibited asymmetry at position 8, which was located far from the mutated residue 90 (Fig. 4). Thus, it is difficult to reach a conclusion regarding the link between the changes in asymmetry that occurred at position 33 or 8 and mutations I54M and L90M.
Asymmetric changes close to mutated residues putatively related to mutations
Figure 4 shows that the three mutants with the V62A mutation exhibited asymmetry at positions 38 and 39 and loss of asymmetry at positions 40 and 41. Mutated residue 62 was close to residue 38, which was close to residues 39–41 (Fig. 4). Thus, the V62A mutation could be responsible for the changes in asymmetry at positions 38–41. However, the loss of asymmetry at positions 40 and 41 was highly recurrent in the mutant set, whereas the asymmetry at positions 38 and 39 was specific to V62A. Thus, it is difficult to conclude that the V62A mutation induces changes in asymmetry at positions 38 and 39.
Some changes in asymmetry occurred at residues involved in the PI-binding pocket. For example, all mutants containing the I46V (or I84V) mutation exhibited asymmetry at pocket residue 48 (or 31 and 84). As residues 48 and 84 establish hydrophobic contacts with PIs (Additional file 6: Figure S6), we concluded that the I46V and I84V mutations could induce structural backbone changes at positions 31, 48, and 84 that could modify structural asymmetry and PI interactions.
Other changes in asymmetry occurred at residues that were located outside of the binding site but were important for its conformation, such as residues 33, 83, and 89 [10, 32]. Figure 4 shows that the structures of all mutants with I82F, I82L, or I84V mutations exhibited a loss of asymmetry at these important positions. Thus, these changes in asymmetry could be a consequence of structural changes induced by mutations at positions 82 and 84, which could modify the conformation of the binding pocket and, thus, indirectly alter PI binding.
Concerning the L99F mutation, we observed that the four mutants containing this mutation exhibited changes in asymmetry at positions 91, 92, 93, and 98. Residues 91, 92, 93, and 98 are involved in the interface between the two monomers: residues 91 and 92 establish non bonded contacts at the interface, and residue 98 establishes hydrogen bonds with residues 2 and 96 of the second monomer (Additional file 7: Figure S7). Thus, the L99F mutation could induce structural asymmetry at interface residues that could have an impact on the PR2 interface and modify the stability of the dimer.