Molecular dynamics simulations of the bacterial OST PglB from C. lari with bound optimal peptide (GDQNATFG) were started from the crystal structure (PDB code 3RCE). The missing residues of external loop EL5 (residues 283–306) and missing loop residues in the periplasmic domain (residues 605–607) were built with modeler (Fiser et al., 2000) using a protocol that optimizes the positions of all non-hydrogen atoms of the loop without altering the coordinates of the existing residues. The optimization protocol used conjugate gradient minimization followed by molecular dynamics with simulated annealing (Fiser et al., 2000). The missing residues of the EL5 loop (residues 283–306) remain mostly disordered during the simulations, but they sample α-helical conformations with a probability of ∼5%. Residues 284–289 are known to become structured and form a α-helix upon binding LLO (Napiórkowska et al., 2017). Using VMD (Humphrey et al., 1996), hydrogens were added to the crystal structure, the protein and peptide were inserted into a lipid bilayer, and the system was solvated with TIP3P (Jorgensen et al., 1983; MacKerell et al., 1998) water (the PglB/peptide system was electrostatically neutral). To evaluate the effect of the lipid content of the bilayer on the dynamics of PglB, we used either POPC or POPE lipid bilayers.

The simulations were performed with NAMD 2.13 (Phillips et al., 2005) using the CHARMM27 force field for proteins and lipids (MacKerell et al., 1998). In the first phase of the equilibration, all atoms except the hydrocarbon tails of the POPC or POPE lipids were fixed. Conjugate gradient minimization followed by equilibration with constant temperature and volume (NVT ensemble) allowed the lipid tails to melt while the water, lipid head groups, Mg²⁺ ions, PglB, and peptide positions remained fixed. The following steps were then performed thirteen times to generate independent trajectories (ten with POPC lipids, three with POPE lipids). In the second phase of equilibration with constant temperature and pressure (NPT ensemble), restraints were removed and a force was applied to any water molecules that entered the lipid bilayer to prevent infiltration and disruption of the bilayer. At the beginning of the second phase, conjugate gradient minimization was performed on all atoms, followed by velocity reinitialization and temperature and pressure equilibration. The final phase of equilibration was performed with no constraints or applied forces using constant temperature and pressure to allow for final density equilibration of the system. Multiple restraint conditions were tried in the final phase of equilibration and the unrestrained equilibration protocol was determined to be the most appropriate to reproduce the coordination of the Mg²⁺ ion given the uncertainty in the Mg²⁺ coordination due to resolution of the crystal structure (3.4 Å) (Lizak et al., 2011). The coordination obtained by unrestrained equilibration is consistent with previous molecular dynamics simulations of PglB (Pedebos et al., 2015; Lee and Im, 2017).

The production phase of the data generation used constant temperature, pressure, and constant area in the plane of the bilayer (NP_xyP_zT ensemble). Pressure and temperature were maintained at 1 atm and 310 K, respectively, using Langevin dynamics and the Nosé-Hoover Langevin piston method. The SHAKE constraint algorithm (Ryckaert et al., 1997) was used to allow a 2 fs time step and the particle mesh Ewald method (Essmann et al., 1995) was used for electrostatic interactions with periodic boundary conditions. Thirteen independent trajectories (ten with POPC lipids, three with POPE lipids) of 500 ns each were collected for PglB with the GDQNATFG peptide (PglB-NAT). The consensus sequence variant (T→S) was created with the Mutator plugin of VMD (Humphrey et al., 1996) and thirteen independent trajectories of up to 500 ns each (ten with POPC lipids, three with POPE lipids) of PglB with the GDQNASFG peptide (PglB-NAS) were collected using the same simulation protocol as above. Three of the trajectories (1, 5, and 12) were terminated after observing peptide unbinding (at 400 ns, 100 and 300 ns, respectively, see Definition of Peptide Binding).

The analysis presented in this paper focuses on differences in the bound states of the two peptides. The results of the simulations that used POPC and POPE lipids were found to be the same within the error of the measured observables (interaction energies, H-bond probabilities, distances between PglB and key peptide residues, etc.), indicating that the dynamics of PglB is not influenced by the lipid composition of the bilayer. For example, the difference in the peptide-PglB interaction energy for the two lipid types was less than the standard deviation: PglB-NAT POPC = −228.7 ± 24.8 kcal/mol, PglB-NAT POPE = −240.0 ± 26.7 kcal/mol, Δ(NAT POPC-POPE) = 11.3 kcal/mol. Therefore, we combined all the trajectories for better sampling. PglB-NAT spent 5,063 ns in the bound state and PglB-NAS spent 2,780 ns in the bound state and all histograms were normalized to account for the difference in the number of data points. To ensure that we have collected enough of the relevant states to be able to make meaningful comparisons between the two peptides, we assessed the quality of sampling using jack-knife analysis (Lohr, 1999) of the peptide-PglB interaction energy. The standard deviation of the jack-knife means was calculated and found to be approximately 1/10 of the standard deviation of the interaction energy, indicating that we have good sampling (Supplementary Table S1). Furthermore, a jack-knife style analysis of the first two principal components (where each trajectory is excluded in turn) of the bound frames shows that the trajectories sample similar regions of conformational space independent of the lipid type (POPC: trajectories 1–10, POPE: trajectories 11–13), another indicator of adequate sampling (see Supplementary Figures S2, S3).

Except where otherwise specified, trajectory analysis was performed with VMD (Humphrey et al., 1996) on bound frames only (see Definition of Peptide Binding) excluding the first 50 ns to allow for structural equilibration, figures were prepared with Tableau Software v10.4 (Seattle, WA), and structures were visualized with VMD using the STRIDE algorithm for secondary structure identification (Frishman and Argos, 1995).

Molecular dynamics simulations of C. lari PglB with the optimal acceptor sequence, GDQNATFG (PglB-NAT), and the peptide variant GDQNASFG (PglB-NAS) were used to discover the detailed differences between threonine and serine at the +2 position in the substrate-bound state of the catalytic cycle. Thirteen replicas with a cumulative simulation time of 6.5 and 5.8 μs were collected for PglB-NAT and PglB-NAS, respectively. We observed a striking difference in the behavior of NAT and NAS peptides: out of the thirteen independent trajectories collected for each system, four trajectories of PglB-NAT exhibit partial peptide unbinding while ten trajectories of PglB-NAS exhibit partial peptide unbinding. Since a stably bound substrate is prerequisite to forming a catalytically competent state, we wanted to understand the critical interactions in the bound state that lead to this behavior. To do so, we needed to quantitatively determine when the peptide was in the bound state versus a partially bound state or fully unbound state in our simulations. We then analyzed the frames where the peptide is bound to identify differences in the bound states of the two peptides.

We identified six non-catalytic H-bonds whose absence corresponds to loss of the catalytically relevant H-bonds and disassociation from the binding pocket. These H-bonds were used to define the bound, partially bound, and fully unbound states (see Table 2). The H-bonds p+1 O-M318 N, p+2 N-D465 Oδ, and p+2 Oγ-W463 Nε (see Supplementary Figures S4, S5 for donor-acceptor distances) provide a good indicator of partial unbinding because once all three are lost, with no specific order, the peptide no longer simultaneously forms the catalytically relevant H-bonds with D56 and E319 (the definition of the partially bound state as used in this work) and stable rebinding is never observed.

We analyzed the behavior of the peptides after becoming partially bound and observed that there is no clear pathway that leads to the full unbinding and dissociation from the binding pocket. After the loss of the H-bonds p+1 O-M318 N (Figures 2A,B, Supplementary Figure S8: dark pink/light pink), p+2 N-D465 Oδ (Figures 2A,B, Supplementary Figure S8: blue/green), and p+2 Oγ-W463 Nε (Figures 2A,B, Supplementary Figure S8: blue/lime), the peptide is still in the binding pocket but is more loosely associated with the enzyme (Figure 2C, light green). The peptide can remain in this state for the remainder of the simulation time or it can become fully disassociated (for example, Supplementary Figure S5: trajectory 3 vs. trajectory 1). The presence of three additional H-bonds keeps the peptide in this partially bound state: p+2 Oγ-W464 Nε (Figures 2A,B, Supplementary Figure S8: blue/cyan), p+2 Oγ-D465 Oδ (Figures 2A,B, Supplementary Figure S8: blue/green), and p+3 N-T316 O (Figures 2A,B, Supplementary Figure S8: violet/lavender). Upon losing these H-bonds (which are never observed to reform once all three are lost), the peptide moves further away and becomes fully dissociated from the binding pocket (Figure 2C, dark green and blue). Thus, our results suggest that multiple interactions are important for the stability of the bound substrate: three key H-bonds are essential to keep the peptide tightly bound in the binding pocket of the enzyme (p+1 O-M318 N, p+2 N-D465 Oδ, and p+2 Oγ-W463 Nε) and three additional H-bonds keep the peptide in a partially bound state and prevent the full dissociation from the binding pocket (p+2 Oγ-W464 Nε, p+2 Oγ-D465 Oδ, and p+3 N-T316 O).

The probabilities of all six of the hydrogen bonds that stabilize the bound and partially bound states are decreased in PglB-NAS compared to PglB-NAT (Table 1). To evaluate the stability of the bound substrate peptides, we calculated the probability of maintaining or transiently losing the first three and all six of these bonds simultaneously. We find that the NAS peptide is much less stably bound than the NAT peptide. All three of the H-bonds whose presence defines the bound state are simultaneously maintained in 68% of PglB-NAT bound states while only 24% of PglB-NAS bound states meet this criterion. Similarly, the probability of maintaining all six of these H-bonds is decreased and the probability of transiently losing all three or six of these H-bonds is increased by factors of 5–10 for PglB-NAS (Supplementary Table S5). This difference is seen over all trajectories and when comparing only those trajectories where the peptide eventually becomes partially bound (Supplementary Table S5, starred probabilities). Surprisingly, when comparing the peptides in the bound state, they have the same probability of forming the catalytically relevant H-bonds p0 Nδ-D56 Oδ (Figures 3A,B, Supplementary Figure S8: red/yellow) and p0 Nδ-E319 Oε (Figures 3A,B, Supplementary Figure S8: red/orange) (Table 1), which are formed transiently in the absence of LLO [as seen previously (Pedebos et al., 2015)]. However, when considering the full trajectories (bound, partially bound, and unbound frames) of PglB-NAT and PglB-NAS, the population of catalytically competent states (where both catalytically relevant H-bonds are present) is lower for the NAS peptides by half compared to NAT peptides (Supplementary Table S6) because the NAS peptides populate the bound state with lower probability (Supplementary Table S6). From this we conclude that the differences between NAT and NAS peptides that lead to increased unbinding in NAS are crucial to understanding the difference in catalytic competency.

None of the six H-bonds identified as important for stable binding should be impacted by the T/S variation in the consensus sequence at position p+2 and two out of six (p+1-M318 and p+3-T316) do not involve p+2 at all. To understand the difference in stability of the bound NAT and NAS peptides, we calculated the interaction energy between the peptide residues and PglB. The largest change is observed for p+2, however, the interaction energy while bound is affected beyond this residue and shows a large degree of compensation at peptide residue p-2 (Table 3).

The interaction energy between PglB and the NAS peptide is overall less favorable (Table 3), consistent with increased disassociation. To understand the changes in conformation underlying the altered interaction energy, we compared the average bound structures of PglB-NAT and PglB-NAS and found differences in both the conformations of the peptide and of the periplasmic domain. The backbone RMSD between average structures (see Materials and methods, Eq. 1) is 1.46 Å for the peptide and 2.48 Å for the periplasmic domain. In contrast, the transmembrane domain shows minimal change between the two structures (backbone RMSD of 0.73 Å).

The compensation observed in the interactions with p-2 can be explained by the changes observed in the NAS peptide. The aspartic acid at position p-2 shifts relative to its transmembrane domain hydrogen-bonding partners R147 and R331, decreasing the favorable interaction energy with R147 (PglB-NAT interaction more favorable by −4.1 kcal/mol) and increasing the favorable interaction energy with R331 (PglB-NAS interaction more favorable by −12.6 kcal/mol), accounting for 80% of the difference in interaction energy at p-2. The remaining difference comes from numerous long-range electrostatic interactions of the charged aspartic acid side chain.

For p+2, the key interacting residues of the WWD motif and I572 from the crystal structure (Lizak et al., 2011) show deviations in average position between PglB-NAS and PglB-NAT. The backbone RMSD between average structures is 0.87–1.18 Å for the WWD motif, 1.67 Å for I572, and 0.93 Å for p+2, indicating that conformational change of both the peptide and the periplasmic domain contribute to the differences in the interaction energy.

Comparison of the average structures suggests a difference in the conformation of PglB when bound to NAT versus NAS peptides. The difference between PglB Cα-Cα distances was calculated for PglB-NAT and PglB-NAS to identify regions of PglB that differ in conformation when bound to the two peptides. We found that the distance between the periplasmic domain and the transmembrane domain is increased in PglB-NAS compared to PglB-NAT (Figure 3). The average Cα-Cα distance between the periplasmic domain and the transmembrane domain (excluding the large external loops EL1 and EL5) is increased by 0.8 Å per residue pair in PglB-NAS compared to PglB-NAT, indicating that on average, the region between the two domains (containing the catalytic pocket) is more open in PglB-NAS (designated the open state) than in PglB-NAT (designated the closed state).

In contrast, the intradomain Cα-Cα distances of the periplasmic domain and the transmembrane domain differ remarkably little between PglB-NAT and PglB-NAS, with average changes of 0.01 Å and 0.04 Å per residue pair, respectively. The minimal change in Cα-Cα distances within the periplasmic and transmembrane domains indicates that the overall structure of both the periplasmic and transmembrane domains do not change between PglB-NAS and PglB-NAT. These results suggest that while the structure of the two domains is conserved, their relative positions change and that there should be a “hinge” that allows for rigid domain motion. If true, we would expect the hinge to meet the following criteria: 1) the hinge should have low backbone RMSD between average structures, 2) the hinge should be located in an interfacial region between transmembrane and periplasmic domains, and 3) the magnitude of the displacement observed between the open and closed positions should correlate with the distance from the hinge. In our system, this means we expect that the further a residue is from the hinge, the more different its position will be between the open state and the closed state for the mobile portion of the structure (the periplasmic domain). Two pairs of structures were used to identify the differences between PglB-NAS and PglB-NAT: the average structure and as an additional validation, the most populated structure (calculated as described in Analysis of Trajectories). The results obtained using either the average or the most populated structure are the same.

From the RMSD between of PglB-NAT and PglB-NAS, we chose residues with low RMSD in regions of the structure appropriate for the hinge. We then calculated the correlation of the backbone RMSD and the distance from each of the chosen residues to see which of these residues agreed best with the criteria for the hinge residue (Supplementary Table S7). From this, we found that N448 and Y473 best meet the criteria for the hinge between the closed state (PglB-NAT) and the open state (PglB-NAS). Both residues are located in regions of low RMSD (Figure 4A) and have the highest correlation coefficient (0.95) between distance and backbone RMSD of the periplasmic domain residues (Supplementary Table S7). These two residues are adjacent in the structure of PglB at the interface between the two domains (Figure 4B): N448 (red asterisk) is located on a helix in between the transmembrane domain and the periplasmic domain with Y473 directly below N448 at the terminus of the helix containing the WWD motif (black asterisk), indicating that the hinge is located at the junction of these two helices. Additionally, the backbone RMSD between the periplasmic domain residues (the mobile portion of the structure) increases linearly with distance from N448 and Y473 but the correlation is weak for other regions of PglB (see Figure 4C for example). Lastly, we chose other residues distant from the hinge region with low RMSD across the sequence of PglB and these residues show low correlation between distance and RMSD (Supplementary Table S8), validating that the strong linear relationship is unique to the hinge region.

The interaction energy between p+2 and the residues of PglB was calculated for both peptides (Table 4) to understand how the T/S difference results in widespread changes in the peptide-PglB interaction (Table 3) and how it affects the orientation of the periplasmic domain (Figure 3). The location of residues that differ in interaction energy between PglB-NAT and PglB-NAS by ≥ |±0.5| kcal/mol (absolute value of the difference ≥0.5) are shown in Figure 5A. For each of these residues, the interaction energy is less favorable in PglB-NAS (Table 4). As expected, the loss of the methyl group in p+2:S decreases the interaction energy with I572 but also its neighbor R571. The WWD motif (463–465) shows the largest differences in interaction energy and surprisingly, residues in the structured C-terminal portion of EL5 (315–316) show significant differences as well.

Although the average differs by ≥ |±0.5| kcal/mol for the residues listed in Table 4, a histogram of the interaction energy shows that the two peptides sample states with similar p+2 interaction energies (Supplementary Figure S9). The measured difference in the average interaction energy of p+2 is due to the higher population of states with near-zero interaction energies by the NAS peptide (Supplementary Figure S9), decreasing the average interaction energy of the residues highlighted in Figure 5A. This suggests that p+2:S interacts in a similar way with residues of PglB where the interaction energy differs by ≥ |±0.5| kcal/mol, but that the interactions are less stable, consistent with the decreased stability of H-bonds identified as important for peptide binding involving many of these same residues (Table 1 and Supplementary Table S5).

The total effect of sampling higher energy conformations with higher probability in PglB-NAS can be seen in the total interaction energy between p+2 and PglB (Figure 5B). Both PglB-NAT and PglB-NAS have bimodal distributions indicating the sampling of two states, one more favorable than the other. However, the more favorable (primary) mode and the less favorable (secondary) mode are more favorable in PglB-NAT by approximately 6 and 7 kcal/mol, respectively. Additionally, the more favorable state is more populated in both PglB-NAT and PglB-NAS but the difference in the population of the two states is significantly larger for PglB-NAT (Figure 5B).

These data indicate that the loss of the methyl group in p+2:S results in fewer interactions between the peptide and PglB. The result of these changes is less favorable total interaction energy in PglB-NAS (Figure 5B) due to fewer simultaneous interaction partners. The interactions of the residue at p+2 with the periplasmic domain and the EL5 loop provide a mechanism for the T/S variation at the +2 position to affect the periplasmic domain orientation: threonine may act as a pin to close a latch, preventing opening of the periplasmic domain hinge (near N448/Y473). We define residues of PglB with an absolute difference in the p+2 interaction energy ≥1.0 kcal/mol as the latch residues (W463 and D465 of the WWD motif, I572, and EL5 residue T316). Our hypothesis is that p+2 interacting with these latch residues acts as a pin to close the latch which limits the hinge motion when p+2 is threonine, but that this ability is impaired when p+2 is serine. The two sides of the latch are formed by the interacting residues of the periplasmic and EL5 domains (Figure 5A) and by linking the two regions which contain the latch, threonine at p+2 closes the latch and restricts the motion of the hinge. The ability of serine at p+2 to pin the latch closed is impaired because serine is incapable of forming stable simultaneous interactions with the periplasmic domain and the EL5 loop due to the loss of the methyl group.

As shown in Figure 4C, distance from the hinge (N448/Y473) is strongly correlated with backbone displacement of the periplasmic domain, consistent with a hinge-like motion. We propose that the difference between the two peptides when bound to PglB is consistent with a simple hinge-latch model (Figure 6A). In this model, the pin connects the two sides of the latch when closed. An arm connects the hinge and latch and mediates the restriction of the domain motion when the pin closes the latch. In PglB, N448/Y473 is the hinge, residues W463, D465, I572, and T316 form the two sides of the latch, p+2 acts as the pin to hold the latch closed when threonine, and the helix connecting the hinge and latch is the arm (residues 463–472) (Figure 6B).

If the hinge-latch model is accurate for PglB, the motion should be different when latched (PglB-NAT) versus unlatched (PglB-NAS). The motion should be more restrained in PglB-NAT and the motion should be freer in PglB-NAS. Principal component analysis of PglB (excluding the unstructured EL5 N-terminus) shows that the motion of PglB-NAT and PglB-NAS is similar (Supplementary Figures S2, S3). However, the frequency of sampling different regions of the conformational space is different (Supplementary Figure S6). PglB-NAT is more concentrated in the center of the conformational space, which is consistent with restrained motion of the hinge when the latch is pinned (Supplementary Figure S6A). PglB-NAS is more broadly distributed with higher sampling in the tails, which is indicative of increased motion consistent with an unlatched hinge (Supplementary Figure S6B).

The motion of PglB-NAT versus PglB-NAS supports the existence of a latch. To verify that our choice of latch and pin residues is correct, we looked at the distance between these residues in the latched (PglB-NAT) and unlatched (PglB-NAS) conditions. For the latch and pin to restrict the motion of the hinge (as we propose occurs in PglB-NAT), the latch and pin complex must be stable (the distances between these residues should not fluctuate significantly). When the pin is unable to close the latch (as we propose occurs in PglB-NAS), we expect to see greater fluctuations in the distance between the pin and latch residues. The histograms of the distances between the PglB residues on opposite sides of the latch (W463 and I572 of the periplasmic domain with T316 of the EL5 loop) show that the latch residues in PglB-NAT predominantly sample a narrow range of distances (Figures 6C,D, gray). The distances between the pin (p+2) and the latch residues show a similar preference for sampling a restricted range of distances (Figures 6E–H, gray). In PglB-NAS, the distances between the latch and pin residues sample larger distances more often than in PglB-NAT (Figures 6C–H, red). These results show that when p+2 is threonine, the latch/pin complex is stable, as required to restrict the motion of the hinge in the hinge-latch model. The increased sampling of larger distances between the pin and latch residues in PglB-NAS is instead consistent with a pin that is unable to keep the latch closed.

The behavior of the pin and latch residues explains the interplay between peptide binding and periplasmic domain orientation. The distances between the latch residues in PglB-NAS increase because serine lacks the methyl group and is unable to form the stable simultaneous interactions necessary to pin the latch closed (keeping the two sides of the latch in closer proximity). The NAS peptide therefore shifts its position to interact with the latch residues on opposite sides (periplasmic domain and EL5) because the latch residues are further apart. As the NAS peptide moves towards one side of the latch, it loses the interaction with the other side of the latch (reflected in the increased sampling of near-zero interaction energies (Supplementary Figure S9). The increased distance between the latch residues and the subsequent motion of the NAS peptide explain the weakened H-bonds important for substrate binding because these bonds between the enzyme and substrate overlap with the latch/pin complex (Table 1 and Supplementary Table S5). Of the six H-bonds identified as important for a stably bound peptide, four involve the PglB latch residues, two involve neighboring residues, and both sides of the latch are important (four H-bonds with the periplasmic domain and two H-bonds with EL5). Similarly for the substrate, of these six H-bonds, four involve the pin residue (p+2) and two involve adjacent substrate residues (p+1/p+3). For these reasons, the state of the latch (pinned or unpinned) is inextricably linked to the partial unbinding of the substrate.

Network analysis was used to look for allosteric communication mediated by the peptide as suggested by the hinge and latch model. The network of PglB-NAT consists of eight communities while the PglB-NAS network contains nine communities (Figure 7, Supplementary Figure S7). Seven of these communities are analogous in PglB-NAT and PglB-NAS with 60–100% similarity containing 626 residues of PglB-NAT and 603 residues of PglB-NAS (out of 711 PglB residues, 8 peptide residues, and 2 Mg²⁺ ions). The two largest communities contain most of the periplasmic domain residues (Figure 7, Supplementary Figure S7: orange, green) and are 60–70% analogous. The third largest community (transmembrane helices 4–9; Figure 7, Supplementary Figure S7: red) is 98% analogous in the two networks and is consistent with previous results that showed these helices move as a single unit to accommodate the presence of LLO (Lee and Im, 2017). The last four analogous communities contain 30–50 residues each and are comprised of the EL1 residues (Figure 7, Supplementary Figure S7: light blue) and subsets of the remaining transmembrane helices (Figure 7, Supplementary Figure S7: black, purple, pink). Of the residues not found in the analogous communities, 32 PglB-NAT residues are part of a community not connected in the network of PglB-NAS (Figure 7, Supplementary Figure S7: blue, EL5) and 11 PglB-NAS residues are part of a community not connected in the network of PglB-NAT (Figure 7, Supplementary Figure S7: cyan, transmembrane helix). An additional five residues (483–487) containing part of the bacterial DGGK motif (residues 481–484; Supplementary Figure S1C, cyan), which is suggested to be important for promoting proper conformation of the p0 side chain for LLO attachment (Pedebos et al., 2015; Napiórkowska et al., 2017), is part of a separate community (Figure 7, Supplementary Figure S7: yellow) in PglB-NAS instead of the large community containing the WWD motif as in PglB-NAT (Figure 7, Supplementary Figure S7: orange).

Lastly, 63 PglB-NAT residues and 102 PglB-NAS residues are unconnected (Figure 7, Supplementary Figure S7: white). In PglB-NAT, the residues in unconnected communities are part of the transmembrane domain. These residues include the N-terminal tail and residues around the stable core (see Analysis of Trajectories). Half of the unconnected residues in PglB-NAS are similarly located in the transmembrane domain. Significantly, the remaining unconnected residues in PglB-NAS are the entire EL5 loop (43 residues) and the peptide (8 residues) (Figure 7, Supplementary Figure S7: white). In PglB-NAT, the EL5 loop is part of three communities: blue (residues 283–309, 311–315), orange (residue 310) and green (residues 316–324) (Figure 7, Supplementary Figure S7). The NAT peptide is also part of the green community, which contains most of the residues important for both peptide binding and the pin and latch behavior demonstrated for p+2:T (see Figure 5A and Table 4), excluding only the WWD motif which is located in the adjacent, strongly connected orange community in both PglB-NAT and PglB-NAS (Figure 7, Supplementary Figure S7). The lack of edges connecting the communities containing the peptide and EL5 in PglB-NAS suggests that allosteric communication between the periplasmic domain and EL5, mediated by the peptide, is lost in the PglB-NAS complex.

Suboptimal path analysis can show paths of allosteric communication between different parts of the network (Scarabelli and Grant, 2014; Yao et al., 2016). The periplasmic latch residues (W463, D465, and I572) are connected to the EL5 latch residue (T316) by a diverse network of paths passing through the pin (p+2, 99.6% of paths) (Figure 8). Network attributes such as the number of edges per node (degree) and the number of shortest paths that pass through each node (betweenness) further illustrate the high connectivity of the NAT peptide and that adjacent peptide residues p+1 and p+3 play a supporting role in the network of PglB-NAT (Supplementary Table S9). In contrast, at the threshold used for the network analysis (cross-correlation >0.7), no paths connect the periplasmic domain and EL5 in PglB-NAS (Figure 7). Since a threshold imposes an arbitrary cutoff, we examined the cross-correlation values between the three most important peptide residues in the network of PglB-NAT (p+1, p+2, and p+3) and the latch residues (W463, D465, I572, and T316). Only three residue pairs have cross-correlation values near the threshold in PglB-NAS (cross correlation >0.68): p+1 with W463 and W464 and p+2 with the adjacent I317 (Supplementary Table S10). In PglB-NAT, 12 out of 18 pairs meet the 0.7 threshold and 15 out of 18 pairs meet the 0.68 threshold (Supplementary Table S10); the cross-correlation cutoff would have to be lowered to 0.51 in PglB-NAS to match the level of connectivity in the network of PglB-NAT (Supplementary Table S10). These results support that the allosteric communication between the periplasmic domain and EL5 is disrupted in PglB-NAS because the NAS peptide does not mediate the interaction effectively.

In the cryo-EM structure of the human OST complex with acceptor peptide (OST-B, PDB code 6S7T (Ramírez et al., 2019), the p+2 binding pocket is structurally similar to the binding pocket in PglB (Supplementary Figure S10), thus our findings are likely generalizable to the eukaryotic OSTs. However, a lysine residue (K674) is present in the analogous position of I572 in the structure of PglB (Supplementary Figure S10, orange). The cryo-EM structure suggests that K674 in the human OST may form a hydrogen bond with D606 of the WWD motif (Supplementary Figure S9, blue). This H-bond would stabilize the nearby tryptophan (W677), which is structurally analogous to V575 in the extended bacterial MIV motif (Maita et al., 2010) (Supplementary Figure S10, yellow). The methyl group of p+2 is rotated towards W677 in the human structure (Supplementary Figure S10, red/yellow), suggesting W677 replaces the stabilizing van der Waals interaction formed with I572 in the bacterial structure. W677 is conserved in eukaryotes (Supplementary Figure S1D) as part of an extended hydrophobic motif after the previously identified DXXK motif (Maita et al., 2010). W677 also stacks with S402 (of the SVSE motif, Supplementary Figure S1B) in the structurally analogous position to T316 (Supplementary Figure S10, purple), possibly explaining the conservation of the smaller serine in this position for eukaryotes (Jaffee and Imperiali, 2011). Overall, the structural similarity between the binding pockets suggests that a similar hinge/latch mechanism is present in eukaryotic OSTs with variations that modulate the catalytic efficiency for serine-containing substrates.