The opening dynamics of the lateral gate regulates the activity of rhomboid proteases

Abstract

Rhomboid proteases hydrolyze substrate helices within the lipid bilayer to release soluble domains from the membrane. Here, we investigate the mechanism of activity regulation for this unique but wide-spread protein family. In the model rhomboid GlpG, a lateral gate formed by transmembrane helices TM2 and TM5 was previously proposed to allow access of the hydrophobic substrate to the shielded hydrophilic active site. In our study, we modified the gate region and either immobilized the gate by introducing a maleimide-maleimide (M2M) crosslink or weakened the TM2/TM5 interaction network through mutations. We used solid-state nuclear magnetic resonance (NMR), molecular dynamics (MD) simulations, and molecular docking to investigate the resulting effects on structure and dynamics on the atomic level. We find that variants with increased dynamics at TM5 also exhibit enhanced activity, whereas introduction of a crosslink close to the active site strongly reduces activity. Our study therefore establishes a strong link between the opening dynamics of the lateral gate in rhomboid proteases and their enzymatic activity.

Revealing a mechanism of activity regulation in rhomboid proteases.

INTRODUCTION

Intramembrane proteases process transmembrane substrates directly within the lipid bilayer and are involved in numerous physiological and pathological processes. Notably, they play important roles in a wide range of diseases including neurodegenerative disorders, diabetes, and malaria (1–4). The bacterial rhomboid protease GlpG is the de facto model system for this class of intramembrane proteases, and its structure, dynamics, function, and inhibition have been studied in detail (5–9).

Unfortunately, little is known about the functional role of bacterial rhomboids, and until recently, only one bacterial substrate was studied in detail: the twin arginine receptor [twin arginine transporter A (TatA)], the natural substrate of AarA in Providencia stuartii (10). In 2020, the physiological substrate HybA of GlpG in Shigella sonnei was found, which has 99% sequence identity with Escherichia coli (E. coli) GlpG (11). It is involved in membrane protein quality control by specifically targeting components of respiratory complexes (11).

In order for GlpG to hydrolyze transmembrane substrates, they need access to the active site within the membrane bilayer. The active site consists of a catalytic dyad, formed by residues S201 and H254 in transmembrane helices (TM) 4 and 6, respectively. It is hydrophilic and therefore shielded from the hydrophobic acyl chains of the lipids, particularly by TM2 and TM5 (see Fig. 1A) and the L5 loop. The exact mechanism how the substrate gains access to the active site is still not determined. Urban, Shi, and colleagues (5, 12) proposed that TM2 and TM5 form a lateral gate that is able to open and allows the substrate to enter (see Fig. 1C). This mechanism was inspired by crystallographic work where TM5 and L5 can adopt either an “open” or a “closed” conformation (5). The extreme bending of TM5 was however questioned, because it might be an artifact of crystal packing. In contrast, Xue and Ha (13) suggested that the substrate enters from the top rather than laterally. In that case, the substrate does not enter the enzyme but bends over to reach into the active site. Consequently, only a slight displacement of TM5 is needed, while the L5 loop mostly influences substrate processing by acting as a cap (see Fig. 1C). Additional to these theories, substrate processing likely involves exosite binding, where the substrate binds to residues of TM2 and TM5 in order for it to be recognized and being positioned for helix unwinding and active site binding (14).

Fig. 1. Lateral substrate gate of GlpG consisting of transmembrane helices TM2 and TM5.

(A) Substrate gate of GlpG with important amino acids for the gate interaction shown in blue in a stick representation. The amino acid pairs F153 + W236, W157 + F232, and Y160 + L229, respectively, were mutated to cysteines to enable the formation of a crosslink. Right: Zoom into the gate with amino acids W236 and F153 highlighted to show the interaction of TM2 and TM5. The active center (S201 and H254) is shown in yellow. (B) Chemical structure of the 1,2-bis(methylsulfonylsulfanyl)ethane (M2M) crosslinker. (C) Cartoon representation showing the scission mechanism of GlpG (gray) cleaving an intramembrane substrate (blue) in two different variations (L5 cap and lateral gate model).

The two different propositions have been previously probed by engineering disulfide bonds connecting TM2 and TM5, with the aim to understand if lateral gate opening is required for access to the active site, and whether crosslinking of TM2 and TM5 would abolish the activity of GlpG. Specifically, the following three variants were investigated: L229C + Y160C (bottom), F232C + W157C (middle), and W236C + F153C (top). Note that the active site is closest to the top position (Fig. 1A). The interaction between TM5 and TM2 is mainly stabilized by hydrophobic and π-stacking interactions. Consequently, mutating aforementioned residue pairs result in a more open conformation and in an increase of activity (12), while disulfide bonds will reestablish these connections.

Urban and colleagues (12) reported that the middle and bottom crosslinks could be formed by regular disulfide bridges, resulting in a loss of GlpG activity. However, they were not able to establish the crosslinking at the top adjacent to the active site. This limitation may have been a result of the improper formation of the crosslink due to geometric constraints and therefore did not have the anticipated effect. Ha and colleagues (13) succeeded to properly establish the top crosslink with the help of the linker 1,2-bis(methylsulfonylsulfanyl)ethane (M2M) (see Fig. 1B). However, they found that GlpG still maintains its activity in the crosslinked variant. This was presented as an argument against the “lateral gate” theory and in favor of the “top-down” theory of substrate access because it is difficult to imagine how the substrate could enter laterally with the top crosslink established.

Since previous structural studies on GlpG were all based on detergent-solubilized proteins, we recently set out to study GlpG in liposomes by solid-state nuclear magnetic resonance (NMR) (15). Solid-state NMR is an ideal structural biology tool for this aim, as it has the unique ability to study functional dynamics of membrane proteins in native-like lipid bilayers at physiological conditions and temperatures (16–20). In support of the lateral gate model, we observed irregular features of TM5 such as a kink at W236. Relaxation dispersion data further indicated the presence of a major and a minor state of TM5 that are in conformational exchange on a time scale of ~40 μs. The importance of investigations in the lipid bilayer is further highlighted by a very recent solid-state NMR study on GlpG, which shows that hydrophobic membrane thickness influences GlpG activity (21). Possibly, the thickness of the membrane could influence the conformational equilibrium of TM5 through lateral pressure and thereby the opening dynamics of the lateral gate.

To investigate the role of the lateral gate in the regulation of the activity of GlpG, the current work focuses on variants of GlpG that are either crosslinked, so that the motion of TM5 relative to TM2 is restricted or alanine mutants with a weakened interaction between TM5 and TM2. The latter GlpG variants are expected to increase the dynamics of the gate, as it was previously reported that these variants entail an enhancement of activity (12, 22). We systematically analyze these variants in a lipid bilayer environment using an integrated biophysical approach comprising solid-state NMR, molecular dynamics (MD) simulations, and a range of functional assays. Our data reveal a clear relationship between the opening dynamics of the TM5 gate and protease activity, reassuring the lateral gating mechanism for substrate binding in the rhomboid proteases.

RESULTS

Correlation between gate opening dynamics and protease activity

According to the proposed lateral gate cleavage mechanism, the interactions between TM2 and TM5 are essential for the activity of GlpG. Therefore, by mutating residues involved in these interactions, altered activity should be achieved as described before (12). We evaluated this on individual F153A and W236A single mutants, as well as on the F153A + W236A double mutant. These variants potentially weaken the interaction between the TM2 and TM5 helices at the top and maybe with the surrounding lipids. Note that here and for all following results we studied a truncated version of GlpG, missing the cytosolic domain: GlpGΔN, i.e., GlpG_87–276 [see the Supplementary Materials; for simplicity, GlpG_87–276 is henceforth referred to as GlpG or wild-type (WT) GlpG].

A fluorescence-based proteolysis activity assay (23, 24) on all alanine mutants showed an increase in activity compared to WT GlpG (fig. S1). The single mutants F153A and W236A both showed approximately a 2.5 times activity enhancement, while the double mutant (F153A + W236A) showed a synergetic effect with fivefold enhancement compared to WT GlpG. This is in line with previous results showing similar enhancement of the enzymatic activity in this double mutant (12). Our results indicate that enzymatic activity is quantitatively correlated with the opening dynamics of the lateral gate.

With the aim to characterize the opening dynamics of these mutants at the atomistic level, we performed solid-state NMR experiments. Intriguingly, the resulting spectra of F153A + W236A showed increased conformational heterogeneity compared to WT GlpG. Furthermore, the weakened interactions between TM5 and TM2 did not only affect the mutated region but the entire protein. The increased conformational heterogeneity also resulted in a marked decrease in sensitivity, preventing unambiguous assignments of the alanine mutants [see fig. S2 for a comparison of ¹H-¹⁵N two-dimensional (2D) spectra].

Activity of gate-closing crosslinked variants

Next, we investigated whether preventing the opening of the TM5 helix by adding a crosslink between TM2 and TM5 decreases activity. To crosslink TM2 and TM5 at several positions, we generated three double mutant pairs: F153C + W236C, W157C + F232C, and Y160C + L229C (Fig. 2A). Compared to WT GlpG, the introduced cysteine mutations at the top of TM2 and TM5 helices showed a similar behavior as the alanine mutations at the same position (Fig. 2B and fig. S3), increasing the activity threefold. In contrast, a double mutation at the middle part of TM2/TM5—the W157C + F232C mutation—results in only a twofold increase in activity, while the double mutation at the bottom section of TM2/TM5—Y160C + L229C—remained even only similar activity compared to the WT. These results suggested that weakening the interaction of the top part of TM2 and TM5 is more efficient for enhancing protease activity. Crosslinking all three double cysteine mutants (F153C + W236C, W157C + F232C, and Y160C + L229C) via M2M strongly decreased the activity by five- to eightfold, which, importantly, could be recaptured to almost 100% by reversing the link with dithiothreitol (DTT) (Fig. 2B). Compared to WT GlpG, the activity of the mutants at the top and at the middle part of the TM5 helix decreased about two- to threefold. The activity was not completely abolished in the crosslinked variants, possibly due to the inefficiency of the crosslinking approach.

Fig. 2. Activity assessment of wild-type (WT) GlpG and mutants without crosslink, with crosslink, and with a reversed crosslink.

(A) Cartoon representation of GlpG with the mutant pairs F153C + W236C, W157C + F232C, and Y160C + L229C, respectively, indicated in blue. An exemplary crosslink with the 1,2-bis(methylsulfonylsulfanyl)ethane (M2M) linker is shown between F153C + W236C and depicted in red. (B) Initial rates of GlpG WT and mutants (each without crosslink, crosslinked with M2M, and with broken crosslink via dithiothreitol (DTT) processing the fluorescein isothiocyanate–labeled twin arginine transporter A (TatA) substrate. The intensity of the fluorescent signal is measured in relative fluorescence units (RFU). Each experiment had three technical replicates.

Active site not directly influenced by crosslinking

To probe the influence of the crosslinks on the active site, we conducted fluorophosphonate-based experiments (Fig. 3) (25). Fluorophosphonate probes specifically label the serine of enzymatically active serine hydrolases through coupling with tetramethylrhodamine (TAMRA) and therefore enable in gel fluorescence detection (Fig. 3A). Here, we found that the fluorophosphonate was able to bind to the active site in all of the investigated mutants to a similar degree, including the crosslinked and uncrosslinked variants (Fig. 3B). This indicates that the structure and function of the active site are not perturbed by the presence of the crosslinks, and the observed changes in the rate of TatA processing are solely due to changes in site accessibility through the lateral gate rather than perturbation of the catalytic dyad.

Fig. 3. Analysis of the active site of GlpG and its corresponding mutants.

(A) Cartoon representation of tetramethylrhodamine (TAMRA)–fluorophosphonate (FP) assay. (B) GlpG and its respective mutants, crosslinked and uncrosslinked, were labeled with a fluorophosphonate probe in the active center (S201). The top SDS–polyacrylamide gel electrophoresis (SDS-PAGE) shows the Coomassie staining of all proteins, whereas the bottom SDS-PAGE shows the fluorophosphonate signal under ultraviolet light. All protein samples show a strong signal. As a negative control, GlpG with the inhibitor JLK6 (8) is used, resulting in the absence of the fluorophosphonate signal. Note that the 1,2-bis(methylsulfonylsulfanyl)ethane (M2M) crosslink changes the mobility of the protein, which runs slightly faster compared to uncrosslinked proteins.

We again complemented our functional data by solid-state NMR experiments to obtain structural insights at the atomic level. For this, we focused on the mutant with the most pronounced effect of the M2M crosslink, namely, F153C + W236C. Figure 4 shows a comparison between WT GlpG and the crosslinked GlpG F153C + W236C mutant. While a number of peaks match in the two spectra, there are some considerable differences (see Fig. 4A for an overlay of 2D NCα projections from 3D hCαNH spectra). The most noticeable difference is the complete absence of peaks originating from residues in TM5 in the spectrum of the crosslinked cysteine mutant. In addition to TM5, peaks corresponding to residues in TM1 (113), loop L1 (117 and 118), TM2 (149 and 150), TM3 (189, 191 to 193), loop L3 (196), TM4 (200 and 213), loop L4 (223 to 225), and TM6 (251 to 256) are also missing in the crosslinked double cysteine mutant (blue in Fig. 4B). There are also previously unknown peaks arising in the crosslinked double cysteine mutant (red in Fig. 4B) that could not be observed in WT GlpG. Specifically, several residues in loop L1 (124 to 125, 138 to 140, and 142 to 144), the lower part of TM6 (264 to 267), and a few residues in other parts of the protein (97 in TM1, 165 in TM2, and 187 in TM3). Furthermore, of the peaks that could be observed in both samples (dark gray in Fig. 4B), we found a number of residues with considerable chemical shift differences in loop L1, around the active site (L3/TM4), and in loop L4 (Fig. 4C).

Fig. 4. Comparison between wild-type (WT) GlpG and crosslinked GlpG F153C + W236C based on solid-state nuclear magnetic resonance (NMR) experiments.

(A) 2D ¹⁵N-¹³Cα projections from 3D hCαNH spectra of WT GlpG (blue) and crosslinked GlpG F153C + W236C (red). Examples of residues for which peaks are only present in one of the samples and residues for which peaks exhibit considerable chemical shift changes between the samples are indicated. ppm, parts per million. (B) Structure of GlpG with residues for which peaks are only present in WT GlpG indicated in blue and residues for which peaks are only present in crosslinked GlpG F153C + W236C indicated in red, residues for which peaks are present in both samples are indicated in dark gray, and residues for which no information is available are shown in light gray. (C) Absolute chemical shift differences for ¹⁵N (dark gray) and ¹³Cα (orange) between WT GlpG and crosslinked GlpG F153C + W236C. The spectra were recorded on 600-MHz (WT GlpG) and 900-MHz (crosslinked mutant) spectrometers using 1.9-mm probes operating at 40-kHz magic angle spinning and at a sample temperature of ca. 20°C.

Note that because our solid-state NMR experiments were conducted on perdeuterated samples that were subsequently back-exchanged in water-based buffers, we could only extract the information of those solvent exposed residues. Therefore, we propose here two potential explanations for why the visibility of the peaks of the two different samples is distinctive: (i) the protein dynamics is considerably different between WT GlpG and the mutant; and (ii) hydrogen/deuterium (H/D) exchange is different due to variation of the protein-lipid interactions. Moreover, the fact that residues in TM5 cannot be observed in the crosslinked sample (H/D exchange should have taken place before the crosslink was established) suggests that the crosslinked GlpG variant is not in a rigid closed conformation as we previously anticipated, but rather with substantial structural heterogeneity in TM5.

The relatively large chemical shift changes observed for several residues in L1 suggest an allosteric coupling between the TM2/TM5 dynamics and the conformation of loop L1. A comparison between spectra of the F153C + W236C mutant with and without crosslink unexpectedly did not reveal any noticeable chemical shift differences (fig. S2), but rather a strong decline in sensitivity for the mutant without crosslink (fig. S4). This suggests that crosslinking TM2 and TM5 does not lead to a considerable change in conformation. However, because we cannot detect any residues in TM5 for the cysteine mutants, regardless of whether they are crosslinked or not, we cannot conclude if the conformation of TM5 is affected by the crosslink. The considerable increase in sensitivity after the establishment of the crosslink suggests that the overall dynamics of the protein is strongly affected by the crosslink and that the crosslink leads to a more rigid protein conformation. This observation is further supported by a comparison of bulk ¹⁵N R₁ and R_1ρ relaxation rates between the samples (fig. S4). The R₁ rates, dominated by fast ps-ns motion, are slightly lower in the cysteine mutants (0.035 s⁻¹ without and 0.029 s⁻¹ with crosslink) compared to WT (0.040 s⁻¹). R_1ρ, dominated by slower (high ns-ms) motion, is increased in the cysteine mutant without crosslink (30.5 s⁻¹) and decreased in the cysteine mutant with crosslink (21.2 s⁻¹) compared to WT (25.7 s⁻¹). While the differences in bulk relaxation rates are not very strong, they show a clear trend toward increased slow motion in the double cysteine mutant that is reversed when the crosslink is added.

TM5 as a dynamic hotspot suggested by MD simulations

To complement the results of the solid-state NMR measurements, we performed a series of atomistic MD simulations of WT GlpG, the double mutants F153A + W236A, F153C + W236C, and of the M2M crosslinked F153C + W236C variant. All simulations were started from both, the open [Protein Data Bank (PDB) ID: 2NRF, chain A] and closed (PDB ID: 2IC8) states of GlpG and were repeated five times for 500 ns.

While no full opening-closing dynamics were observed in the MD simulations at the current time scale (fig. S8, E and F), the lateral gate became partially closed in the simulations of the M2M crosslinked F153C + W236C variant starting from the open structure (Fig. 5A). The corresponding end snapshot is noticeably different from the end snapshot of the closed simulations of WT GlpG (Fig. 5, A and B). Besides the crosslinked variant, the F153A + W236A and F153C + W236C mutants also exhibited heterogeneous conformations of TM5 when comparing the final coordinates of the individual replicas (fig. S8, A to D). These results are in very good agreement with the solid-state NMR data, showing conformational heterogeneity in TM5 of the crosslinked and noncrosslinked F153C + W236C variants.

Fig. 5. Molecular dynamics (MD) simulation of open and closed GlpG conformations with and without 1,2-bis(methylsulfonylsulfanyl)ethane (M2M) crosslinks.

(A) Overlays of the crystal structure of the open wild-type (WT) conformation (PDB ID: 2NRF, chain A) and a selected final snapshot of the M2M crosslinked F153C + W236C variant after 500-ns simulation starting from the open conformation. (B) Overlays of the crystal structure of the closed conformation (PDB ID: 2IC8) and a selected final snapshot of the M2M crosslinked F153C + W236C variant after 500-ns simulation starting from the closed conformation. (C and D) Comparison of the residue-wise time-averaged root mean square fluctuation (RMSF) for WT, F153A + W236A, F153C + W236C, and F153C + W236C M2M GlpG derived from five runs of 500-ns simulations for (C) the open and (D) closed states. (E and F) Comparison of averaged minimum distance between the gate residues X153 and X236 for (E) the open and (F) closed states of the aforementioned GlpG variants. All averages are calculated by taking five independent simulations of 500 ns with the Amber99SB-ILDN force field into account. The shaded areas highlight the SD of the five replicas. The red lines represent the distance between the two crosslinked sulfur atoms that remains constant during the simulations.

Furthermore, we calculated the residue-wise time-averaged root mean square fluctuation (RMSF) for the simulations of different GlpG variants to identify regions with high structural flexibility. The results shown in Fig. 5 (C and D) suggest substantially higher dynamics for TM5 in all GlpG variants compared to other transmembrane helices, as expected from previous studies. The RMSF values of TM5 are almost comparable to the loop regions, underlining the unusually high mobility of this helix.

Calculation of the minimum distance between X153 and X236 (where X stands for the amino acid variation in the simulated models) showed that the distance between X153 and X236 remained large in all variants during the simulations of both the open and closed conformations (0.6 to 1.0 nm). This finding suggests a weak interaction between these two residues throughout all of the simulations. In strong contrast, WT GlpG shows a large difference in the distance of the two gating residues between closed (0.2 to 0.3 nm) and open simulations (0.6 to 0.8 nm) (Fig. 5, E and F). Hence, open and closed conformations of WT GlpG remained rather stable during the simulations (fig. S6), while final snapshots of open and closed simulations become indistinguishable in both F153C(A) + W236C(A) double mutants but without extreme bending of the top section of TM5 for the open conformation (fig. S8). Here, the entire TM5 helix becomes relatively mobile, most probably due to the diminishment of the TM2/TM5 interaction.

Docking of substrate TatA in the lateral gate

Last, we investigated whether the cavity formed by the opening of the TM5 helix allows binding of TatA, a well-characterized substrate of GlpG (10). For this, we used protein-protein docking using the program ZDOCK (26) to find possible docking poses of TatA in the open and closed states of WT GlpG. For the open-state of WT GlpG, the program predicted 92 clusters of poses. In the top three ranked clusters, one cluster showed TatA binding to the interface of TM2 and TM5. In this cluster, we could identify poses where the cleavage site—peptide bond of Ala8 and Ala9—is adjacent to the catalytic center (Fig. 6, A and B). In contrast, for the closed state, no docking pose could be predicted in the same region. It should be noted that here we only energy-minimized the docking pose of the substrate-bound GlpG, which was not further relaxed by MD simulations. Therefore, the docking poses shown in Fig. 6 should be considered only as an approximate model for the GlpG-TatA complex and do not show the unwinding of the helix that was previously predicted (27, 28). However, these poses are sufficient to correlate substrate binding with the cavity between TM2 and TM5 because the exact time of the helix unwinding remains unknown.

Fig. 6. Docking of twin arginine transporter A (TatA) substrate to WT GlpG and GlpG F153A + W236A.

(A) Energy-minimized docking pose of substrate TatA (gray) in the intramembrane exosite of the open x-ray structure of GlpG (blue; PDB ID: 2NRF, chain A). (B) Zoom into the catalytic site of (A). (C) Energy-minimized docking pose of substrate TatA (gray) in the intramembrane exosite of a selected final snapshot of F153A + W236A from a 500-ns molecular dynamics (MD) simulation (green). (D) Zoom into the catalytic site of (C). In (A) and (C), red sticks represent phosphate lipid groups, while the hydrophobic tails of the lipids are displayed as gray sticks. In (B) and (D), important residues are highlighted and shown as sticks. Docking was performed with ZDOCK.

As discussed above, for the double mutants F153A + W236A and F153C + W236C, the distance between TM5 and TM2 remains relatively large during both open and closed simulations. Consequently, we chose one final snapshot from the open simulations of F153A + F236A, where the extreme bending of the top section of TM5 was not observed. Here, we repeated the docking to investigate whether bending of TM5 away from TM2 is indeed necessary for the substrate binding in this variant. Similar docking poses compared to open WT GlpG could be predicted for the F153A + W236A mutant, where the cleavage site of TatA is in proximity to the active center (Fig. 6, C and D). This result suggests that bending of the top part of TM5 as revealed in the open-state x-ray structure is not required for substrate processing in these variants, while without steric hindrance between TM2 and TM5, substrates such as TatA are ready to bind at the intramembrane exosite of GlpG. It should, however, be noted that the L5 cap remains open in the F153A + W236A and F153C + W236C double mutants.

DISCUSSION

Conflicting theories have been proposed regarding how the substrate accesses the catalytic center of GlpG. In this study, we used a combination of functional assays including crosslinking experiments, solid-state NMR, MD simulations, and protein-protein docking to study GlpG in its native lipid bilayer environment. We found that introducing mutations that diminish the interaction between TM2 and TM5 helices promotes substrate access through the lateral gate and thereby increases protease activity. We were also able to show that stable M2M crosslinks between TM2 and TM5 block most of the substrate processing. Our results are therefore more in line with the lateral gate theory proposed by Urban and colleagues (12). However, because our data show that the bottom and middle TM2/TM5 interactions are less important than the top part, a model is conceivable where the substrate binds to TM2 and TM5 in the lower part and bends over in the top part close to the active site.

Available x-ray data show both open and closed conformations of WT GlpG, where the L5 loop adopts different conformational states, and the position of the TM5 helix varies between apo- (5, 6, 29) and inhibitor-bound (8, 9, 30–32) structures. In one of our earlier studies, we observed a kink in TM5 at W236, which indicates that TM5 does not adopt a fully helical structure when situated in native-like liposomes, and thus GlpG in liposomes has a slightly different conformation than reported in the closed-state structure by x-ray crystallography on detergent-solubilized protein (15). In another previous study of GlpG with inhibitors bound in the active site (33), we observed that the peaks corresponding to TM5 were missing in solid-state NMR spectra of the bound state, indicating that ligand binding may also affect the stability of TM5. In the present study, our NMR data show disappearing peaks for TM5 upon mutation, suggesting that weakened interactions between TM2 and TM5 lead to increased dynamics. Unexpectedly, the same observation also occurred when GlpG variants were crosslinked with M2M. The comparison of the RMSF for different GlpG variants in the MD simulations revealed TM5 nearly as dynamic as some of the loop regions, e.g., the L5 cap. Furthermore, AlphaFold2 predictions suggested a lower confidence level (74 to 85 per residue confidence score) for TM5 compared to the other transmembrane helices (see fig. S7) (34, 35). All of these observations support TM5 as a dynamic hotspot in GlpG.

We generated three different cysteine double mutants at different positions of TM2/TM5 and observed the strongest increase of substrate processing in the F153C + W236C variant. This increase is diminished when mutations are introduced in the middle part of the helices (W157C + F232C) and weakest for the mutation pair in the bottom section of the helices (Y160C + L229C). This result indicates that weakening the interaction at the top TM2 and TM5 region is essential for substrate processing. Previous MD simulations and our docking result suggest that substrates interact with the exosite (mostly residues of TM2/TM5) of the protease before being unwound and processed (14). It is possible that the mutated residues are also important for recognizing and binding the substrate and that M2M impedes this action through steric hindrance. This could explain as well why residues far away from the active site and at the bottom of the substrate gate (Y160 + L229) will also lead to reduced activity when crosslinked.

The opening of the TM2/TM5 gate in WT GlpG is necessary for substrate entry, as the interaction between F153 and W236 appears to stabilize the closed conformation that is inaccessible for the substrate. In contrast, for the F153C + W236C and F153A + W236A double mutants, MD simulations suggested an open conformation where bending of the top of TM5 is not stable, but rather the interaction between TM2 and TM5 becomes weak and consequently the entire TM5 helix is relatively mobile. This finding is in line with solid-state NMR data, showing a high conformational heterogeneity for the double mutants. The decreasing interaction of TM2 and TM5 allows an easier access for substrate processing and consequently a substantially higher protease activity was observed for these two mutants. For both WT GlpG and F153A + W236A in their open states, docking could predict reasonable binding poses for the substrate TatA. In conclusion, we propose here that extreme bending of the top section of the TM5 helix as observed in the original open x-ray structure (PDB ID: 2NRF, chain A) is not required for substrate processing.

On the other hand, when the F153C + W236C mutant is crosslinked with M2M, MD data suggest different dynamics not only for TM5 but also for the L5 loop. The L5 loop is connected to TM5 and caps the active site, which might play an important role in substrate positioning during proteolysis (31). M2M binding might also impede this interaction, and therefore, the activity is reduced. Unfortunately, we cannot observe residues of the L5 loop in our solid-state NMR spectra to complement these observations. Furthermore, it was shown previously that the cleavage site of the substrate shifts when cleaved by a “gate-open” mutant, most likely because the positioning of the substrate in the enzyme changes (36). The cleavage site also differs depending on whether GlpG cleaves substrates in detergent or in membranes highlighting the role of lipid interactions in the substrate-protease complex (36).

Furthermore, a previous study has suggested that the L1 loop is involved in the positioning of the enzyme in the membrane (37). Our current and previous NMR data and MD simulations show structural changes of this loop, especially upon binding of inhibitors (14, 33). When crosslinked, we observe previously invisible and unassigned residues in our NMR spectra, and MD simulations (of the closed form) show that fluctuations in this region decline, suggesting a more rigid L1 loop. Note that mutations (e.g., R137A) in L1 result in activity decline (12). Changing the loop might result in a better accessibility of the core to lipids, which would disturb the overall processing of the substrate (38). In addition, the L1 loop is deeply submerged in the lipid bilayer, even more so when the cytosolic domain is present and it likely facilitates lipid distortion and therefore faster lateral diffusion through the E. coli membrane (39).

The solid-state NMR spectra of WT and crosslinked GlpG show chemical shift changes, in particular around TM5, the L1 loop, and the active site. For both, only a single set of resonances is observed (15), indicating one dominant, unperturbed conformation for each, the WT and crosslinked variant. We therefore propose that the WT GlpG conformation is more open than the crosslinked variant. Though the conformational dynamics changed, the active site remains accessible and reactive to fluorophosphonate. The binding is still possible because fluorophosphonate reaches the active center from the aqueous phase, in contrast to regular substrates such as TatA binding laterally through the TM5 gate (24). Small inhibitors and peptides might take a different path to the active site (31). It should also be considered that unwinding of the substrate helix needs space, which might be blocked by crosslinking. Last, MD simulations show that the introduction of the crosslinker does not lead to major changes in the overall structure or to a perturbation of the geometry of the active site, which may potentially result in deactivation.

In conclusion, with the current study, we could confirm that substrates access the active center of the rhomboid protease GlpG through the lateral gate. A stable conformation is the most commonly populated state for WT GlpG, whereas for mutants with diminished TM2 and TM5 interaction, dynamics of TM5 becomes increased and thereby facilitates substrate processing. Crosslinking TM2/TM5 renders the cavity between TM2 and TM5 inaccessible for the substrate and therefore impedes substrate processing.

MATERIALS AND METHODS

Expression and purification of isotopically labeled GlpG

The GlpG core domain (GlpGΔN, i.e., residues 87 to 276) with a hexahistidine tag was expressed and purified as described before (15). Briefly, glpGΔN in pet15b (Novagen) was transformed into Tuner(DE3)pLysS competent cells (Novagen). To produce the [²H, ¹³C, ¹⁵N]-labeled GlpGΔN samples, ¹⁵NH₄Cl and ¹³C₆-d₇-glucose (Cambridge Isotope Laboratories, USA) were used as the sole nitrogen and carbon sources in a perdeuterated M9 medium.

Cells were grown in M9 medium at 37°C until an optical density of 0.8 was reached and overexpressed after induction with 500 μM isopropyl-β-d-thiogalactopyranoside for 15 hours at 25°C. Cells were harvested, resuspended in lysis buffer, and lysed using an LM10 microfluidizer (Microfluidics, USA) with 15.000-psi working pressure. Insoluble parts were removed by centrifugation, and the supernatant was incubated with 2% (w/v) n-decyl-β-maltoside (DM; Glycon, Germany) for 2 hours at 4°C. The solubilized protein was purified via cobalt-based affinity chromatography on an ÄKTA Pure 25 System (GE Healthcare, Germany).

The same procedure was carried out for mutants of GlpGΔN. To allow for the formation of cysteine bridges, mutations between TM2 and TM5 were introduced at the following positions: F153C + W236C, W157C + F232C, or Y160C + L229C. In addition, the following alanine mutations were studied: F153A, W236A, and F153A + W236A.

Reconstitution

The rhomboid proteases were reconstituted into E. coli total lipid extract (Avanti Polar Lipids, USA) liposomes. For this purpose, E. coli total lipid extract in 3% DM detergent buffer was added to the purified GlpGΔN sample at a lipid/protein ratio of 30:1 (mol/mol). Detergent was removed by dialysis at 100 times dilution against dialysis buffer with additional Bio-Beads SM-2 resin (Bio-Rad) over the course of 10 days and buffer exchanges every 2 days until the sample was completely turbid.

Crosslinking

GlpGΔN mutants were diluted to 0.2 mg/ml. M2M (Santa Cruz Technology, USA) was diluted in dimethyl sulfoxide to a 20 mM stock solution. M2M was added to the GlpGΔN mutants in DM to a final concentration of 50 μM for the purpose of activity assessment. The sample was left for 30 min at room temperature while the crosslink was formed. If required for certain experiments, then the crosslink was subsequently dissolved by the addition of 50 mM DTT at 37°C for 2 hours. All samples were then diluted to 4 μM.

The same procedure was carried out for the NMR samples. A total of 50 μM M2M was added to the [²H, ¹³C, ¹⁵N]-labeled GlpGΔN F153C + W236C mutant, and the sample was incubated for 1 hour at room temperature. Afterward, it was reconstituted into E. coli total lipid extract and dialysed over the course of 10 days. Subsequently, the samples were collected by ultracentrifugation (2 hours at 300,000g) and transferred into NMR rotors while a few crystals of sodium trimethylsilylpropanesulfonate (DSS) were added.

Activity assessment

The TatA, the natural substrate of AarA in P. stuartii, was used as the substrate for the GlpG activity assay as described before (23). Briefly, a fluorescein isothiocyanate (FITC)–labeled peptide with the first 33 amino acids of TatA and a ß-alanine linker was produced (FITC-_βA-MESTIATAAFGSPWQLIIIALLIILIFGTKKLR). The peptide was dissolved in 50 mM tris, 150 mM NaCl, 0.2% (w/v) DM, and 0.2% (w/v) sarcosine to a final concentration of 400 μM.

E. coli total lipid extract was dissolved in 50 mM tris and 150 mM NaCl in a concentration of 10 mg/ml. Using 400-nm Nuclepore Track-Etched Polycarbonate filters (Whatman, USA), the lipids were extruded 31 times with an Avanti mini extruder (Avanti Polar Lipids, USA) to generate liposomes with a defined size.

A total of 4 pmol of DM-solubilized GlpGΔN or its respective mutants (native, crosslinked, or DTT uncrosslinked) were incubated with 20-fold excess of FITC-labeled TatA peptide in a solution of E. coli total liposomes (1 mg/ml) in 50 mM NaOAc (pH 4), 150 mM NaCl, and 0.2% (w/v) DM. The mixture was incubated for at least 20 min at room temperature. It was then diluted 20-fold with 12.5 mM NaOAc (pH 4.0) and 37.5 mM NaCl to reduce the detergent below its critical micelle concentration. After incubation for 10 min, the proteoliposomes were separated from the detergent by ultracentrifugation at 186,000g for 1 hour at room temperature. The samples were dissolved in neutral pH buffer (50 mM tris and 150 mM NaCl) to start the reaction and quickly transferred into a white small volume 384-well plate (784075, Greiner Bio-One, Germany). Fluorescence was read out every 5 min over the course of 120 min exciting at 490 nm and detecting emission at 525 nm at 25°C with a Tecan Infinite M Plex Microplate Reader (Tecan, Switzerland). The intensity of the fluorescent signal is measured in RFU. All 120 min were used for linear regression analysis because no plateau in fluorescence was reached (fig. S3). Slope and the SE of the slope were calculated with Prism GraphPad.

TAMRA-fluorophosphonate assay

For the labeling of the active site of GlpG and its respective mutants with M2M crosslinks, ActivX TAMRA-Fluorophosphonate Serine Hydrolase Probe (Thermo Fisher Scientific, Germany) was used as described before (25). Briefly, 0.5 μg of purified GlpGΔN was mixed with the reactive probe to a final concentration of 0.5 μM. Subsequently, it was incubated for 1 hour at 37°C protected from light. The reaction was stopped with 4x Laemmli buffer and subjected to SDS–polyacrylamide gel electrophoresis (SDS-PAGE). The gel was visualized with ultraviolet light and afterward stained with Coomassie brilliant blue dye.

Solid-state NMR spectroscopy and analysis

Proton-detected solid-state NMR experiments were performed on Bruker 700 and 900 MHz (¹H Larmor frequency) spectrometers equipped with four-channel (¹H, ²H, ¹³C, and ¹⁵N) 1.9-mm probes, operating at magic angle spinning (MAS) rates of 38 to 40 kHz. The sample temperature was kept around 20°C, estimated from the chemical shift of water relative to the DSS peak. D₂O was used for locking. 2D hNH spectra were recorded for all samples, and, in addition, a 3D hCαNH spectrum was recorded for the crosslinked F153C + W236C mutant using standard sampling on a 900-MHz spectrometer at 40-kHz MAS. Subsequently, four 3D experiments (hCαNH, hCONH, hCαcoNH, and hCOcαNH) were recorded for chemical shift assignments of the F153C + W236C mutant crosslinked with the M2M linker, and two 3D experiments were recorded for transfer of the assignments to the F153C + W236C mutant without crosslink. These assignment spectra were recorded on a 700-MHz spectrometer at 38-kHz MAS using 35% nonuniform sampling (NUS) with sampling schedules generated from http://gwagner.med.harvard.edu/intranet/hmsIST/ (40, 41). NUS spectra were reconstructed using the iterative reweighted least squares algorithm (20 iterations) in the qMDD software (42–44) and processed using nmrPipe (45). Assignments were performed using CcpNmr AnalysisAssign Version 3 (46). Spectra of WT GlpG (recorded on a 600-MHz spectrometer) were taken from our previously published study (33).

Bulk ¹⁵N R₁ and R_1ρ relaxation rates were recorded for GlpG WT, F153C + W236C without crosslink, and F153C + W236C crosslinked with the M2M linker using ¹H detected 1D hnH experiments with varying relaxation delays (0.01, 0.1, 0.5, 1.5, 4, 7, 14, and 24 s for R₁) or spin-lock pulse lengths (2, 4.5, 7.5, 12, 18, 35, and 50 ms at a nutation frequency of 12 kHz for R_1ρ). The relaxation experiments were recorded on a 900-MHz spectrometer equipped with a 1.9-mm probe operating at 40-kHz MAS and at a sample temperature of 21°C. R₁ and R_1ρ relaxation rates were extracted by fitting the intensities from the 1D spectra as a function of relaxation time to a monoexponential decaying function. Monte Carlo simulations based on the noise level of the spectra were used to estimate fit errors. The fits were repeated 1000 times with random noise (a random number between 0 and 1 was generated and multiplied with the average noise level of the spectra) added to the input data. Two times the SD was used for error bars in fig. S4.

Protein modeling

For each, the open (PDB: 2NRF, chain A) and closed (PDB ID: 2IC8) structures of GlpG, four models were generated: (i) the WT, (ii) the F153A + W236A, (iii) the F153C + W236C, and (iv) the F153C + W236C crosslinked variants. All protein models were parameterized using the Amber99SB-ILDN force field (47) and inserted into a pre-equilibrated and solvated 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) membrane using the gmx_membed routine (48). The POPC lipid membrane was simulated with the parameters derived by Berger et al. (49). Water was modeled with the SPC/E potential (50) and ions by the Joung et al. parameters (51). A summary of the resulting models can be found in table S3.

Protein-linker modeling

For the crosslinked structure, the methylsulfonyl groups of the 3D structure of M2M were removed to solely represent the linker structure (fig. S5). Two cysteines were connected by a disulfide bridge in the trimmed linker. Both cysteine residues were capped with an acetyl (ACE) group and N-methyl amide (NME) capping group. The NME group was added to the C═O, and the ACE was added to the N side of the cysteines. The geometry of the linker was optimized, and the electrostatic potential was calculated at the Hartree-Fock/6-31G* level using Gaussian 16 (52). The generalized amber force field (53) topology of the linker was generated with the Antechamber software (54), incorporating partial charges from the preceding quantum mechanics calculations based on the restrained electrostatic potential approach (55). After geometry optimization, the cysteine and cap sections of the linker were removed and only the linker was manually inserted between cysteine 153 and cysteine 236 of GlpG. The disulfide bridges were generated at a distance of 2.05 Å.

MD simulations

All-atom MD simulations of the four models of the open and closed states were carried out with Gromacs 2019.3 (56). First, the models were energy minimized and equilibrated for 10 ns with position restraints using a force constant of 1000 kJ mol⁻¹ nm⁻² on the backbone atoms of the protein. Subsequently, 20 ns without position restraints was performed. During the complete equilibration, the isothermal-isobaric (NPT) ensemble was used. After the equilibration process, all models were simulated for 500 ns as a production run in an NPT ensemble. In total, five replicas of production runs were performed for each system.

A time step of 2 fs was enabled by constraining all bonds to hydrogen atoms with the Lincs algorithm (57). Short-ranged electrostatics and van der Waals interactions were truncated at 1.0 nm. Long-ranged electrostatics were calculated with the Particle-Mesh Ewald summation (58). Temperature and pressure coupling were treated with the V-rescale (59) scheme and the Parrinello-Rahman barostate (60), respectively. The temperature was set to 300 K, and the pressure was set to 1 bar. Fluctuations of the periodic cell were only allowed in z direction, normal to the membrane surface, keeping the density of the membrane unchanged.

Molecular docking

Molecular docking was conducted to dock the 33–amino acid peptide TatA onto randomly selected final structures from 500-ns MD simulations of the open (PDB ID: 2NRF, chain A) and closed (PDB ID: 2IC8) WT GlpG, and the open conformation of F153A + W236A was performed using the ZDOCK algorithm (61) integrated in the software DiscoveryStudio Modeling Environment (BIOVIA, Dassault Systèmes, DiscoveryStudio v22.1.0.21297, San Diego: Dassault Systèmes, 2022.). An angular step size of 6° was used resulting in 54,000 docked poses. After unfavorable poses were removed, a conformational clustering was performed. During the analysis, only the top three clusters were considered and analyzed by manual visualization. The selected docking poses shown in Fig. 6 were energy-minimized using the Charmm36 force field in DiscoveryStudio using the algorithm “Smart Minimizer” with a maximum of 1000 steps. No implicit solvent model was used.

Data analysis

The analysis was done on the basis of the MD production runs only. All MD simulations were analyzed by root mean square deviation (RMSD) of the protein backbone and RMSF per residue analysis using the routines g_rms and g_rmsf integrated in Gromacs 2019.3 (56), respectively. For the RMSD analysis, the corresponding x-ray structures were selected as a reference.

Graphical images depicting the structures were generated in ChimeraX (62, 63). Molecular graphics and analyses performed with UCSF ChimeraX, developed by the Resource for Biocomputing, Visualization, and Informatics at the University of California, San Francisco, with support from National Institutes of Health R01-GM129325 and the Office of Cyber Infrastructure and Computational Biology, National Institute of Allergy and Infectious Diseases.

Supplementary Materials

This PDF file includes:

Figs. S1 to S8

Table S1 to S3

View/request a protocol for this paper from Bio-protocol.