Perturbed structural dynamics underlie inhibition and altered efflux of the multidrug resistance pump AcrB

Resistance–nodulation–division efflux pumps play a key role in inherent and evolved multidrug resistance in bacteria. AcrB, a prototypical member of this protein family, extrudes a wide range of antimicrobial agents out of bacteria. Although high-resolution structures exist for AcrB, its conformational fluctuations and their putative role in function are largely unknown. Here, we determine these structural dynamics in the presence of substrates using hydrogen/deuterium exchange mass spectrometry, complemented by molecular dynamics simulations, and bacterial susceptibility studies. We show that an efflux pump inhibitor potentiates antibiotic activity by restraining drug-binding pocket dynamics, rather than preventing antibiotic binding. We also reveal that a drug-binding pocket substitution discovered within a multidrug resistant clinical isolate modifies the plasticity of the transport pathway, which could explain its altered substrate efflux. Our results provide insight into the molecular mechanism of drug export and inhibition of a major multidrug efflux pump and the directive role of its dynamics.

The hydrogen/deuterium exchange mass spectrometry were determined for AcrB purified and solubilised in the detergent DDM. Detergent solubilised protein is not necessarily the best representation of the native protein state. The authors should discuss if they would expect differences if the protein were solubilised with SMA technology or reconstituted in a lipid membrane.
The result with PAßN and PAßN-Ciprofloxacin combination is very interesting. PAßN does not have a particularly high binding affinity for AcrB in comparison with other inhibitors or even substrates though. Would the result be different if another, more high affinity, inhibitor was used instead of PAßN?
It is interesting that PAßN completely abolish ethidium bromide resistance in AcrB from Escherichia coli. However, PAßN does not compete with ethidium bromide in transport assays in Pseudomonas aeruginosa. What would be the explanation for this difference?
Reviewer #2 (Remarks to the Author): Reading et al. examine closely the structural dynamics of AcrB by means of molecular dynamics simulations and H/D exchange mass spectrometry. The bottom line conclusion from their studies is that the PAβN efflux pump inhibitor restricts the intrinsic motions of the drug-binding pockets, indicating that structural dynamics play a critical role in the inhibition and substrate specificity of AcrB. The idea is that PAβN does not prevent antibiotic binding but rather inhibits efflux by enforcing a restrained state that reduces the frequency and magnitude of the conformational changes in the substrate translocation path. The impressive amount of work reported appears to be competently done and accurately reported.
The work presents a new point of view that specialists will be interested in. I doubt that more casual readers will be willing to slug through the paper to understand the findings. The paper is complex and relies heavily upon supplementary materials. For example, reading carefully the 'supplementary discussion' is mandatory for appreciating the paper. The paper is more appropriate for a specialist journal.
A minor but irritating point is the overuse of the word 'impact', which the authors use both as a verb and a noun. This has become rampant in the literature. The better usage is affect (verb) and effect (noun).
Reviewer #3 (Remarks to the Author): In this work, the authors use HDX-MS and MD simulations to investigate the synergistic effect of substrate (CIP) and inhibitor (PAβN) binding to the MDR efflux pump AcrB, together with a clinically relevant mutation (G288D). The three main finding are clearly stated in the introduction: • PAβN EPI restricts the intrinsic motions of the drug-binding pockets as part of its mechanism of action and is effective against both AcrBWT and AcrBG288D • an EPI can dually bind to AcrB alongside an antibiotic, without affecting its inhibitory action • an MDR mutation in acrB impacts upon the structural dynamics of the efflux translocation pathway likely contributing to its modified substrate specificity The authors used adequate methods (i.e CD and thermal protein unfolding) to confirm that both recombinant constructs (WT and G288D) have similar overall structural and thermal stability. The article is particularly well written, adapted to the broad readership of Nature Communication and yet giving enough details in the supplementary material for structural biologists (concerning the analysis of the MD simulations in particular).
I believe that this article should be considered for publication in Nature communications with editorial revisions.
Suggestions below are designed to enhance its communication and impact for the non-specialist.
Major comments: • Supplementary Fig.2c-d: Different MWs were obtained native MS before and after delipidation. Could the authors comment on that?
• l.138: "antibiotic binding only subtly alters the structural motions of AcrBWT": how can the authors make sure that they have a stoichiometric binding and that the binding is not just transient…this sentence is possibly overstated or at least too general as the results might be different at different concentrations and for different antibiotics… The reviewer suggests to replace "antibiotics" by "CIP".
• The sequence coverage seems different in the different conditions (CIP or PAβN), could the authors comment on that? On the same line, did the mutation induce a change in the sequence coverage of the region carrying the G288D mutation? Were the authors able to compare the peptide(s) carrying the mutation?
• Simulations show the formation of hydrophobic bonds between PAβN and E130/K131/Q176/A619 (pose 1), E130/D276 (pose 2) and S126/S46/D174 (pose 3). How do the HDX-MS protection upon PAβN binding relate to these different proposed H-bonds? Are there any encompassing peptide significantly protected by PAβN? If yes this should be clearly stated in the text. If not, do the authors think that this might be due to a partial and/or transient binding of the drug? This point should be addressed in the manuscript.
• Fig.2 shows the ΔHDX comparisons of (AcrBWT+CIP) vs. (AcrBWT) and (AcrBWT+PAβN) vs. (AcrBWT). We can clearly see that CIP mainly destabilizes the N-ter & C-ter of AcrBWT and stabilizes/protects the connecting-loop, while binding of PAβN additionally protects other regions, including the Switch loop. However, a proper statistical comparison between (AcrBWT+CIP) vs. (AcrBWT+PAβN) would be needed before stating that "However, in stark contrast to CIP, inhibitor binding led to HDX reduction throughout extensive parts of the PC1/PC2 cleft of the drug-binding pockets and within the connecting-loop ( Fig.2a-b). This could signify inhibitor induced structural stabilization of the drug-binding pocket entrances" (l.139-140).
• Fluorescence polarization binding and competition assays indicate that both molecule can interact independently with AcrB. This was further supported by MD simulations of AcrB in the Tstate with both CIP and PAβN nicely showing that PAβN is able to bind to the HT region while CIP lies next to the DBP-PBP interface. I just wonder how the CIP docking is affected by the additional presence of PAβN? Is CIP binding to the same site in the absence of PAβN? Could it be that the presence of PAβN stabilizes one particular bound state of CIP, preventing its efflux? This seems to be suggested by the MD simulation analysis (supplementary material) but maybe not enough clearly stated in the main text.
• The authors then present in Fig.2 the ΔHDX comparison of (AcrBWT+CIP+PAβN) vs. (AcrBWT), which indeed shows very similar behavior as the previous comparison (AcrBWT+PAβN) vs. (AcrBWT), suggesting that most fluctuations are due to PAβN and that there isn't much synergy between the substrate and the EPI. However, to see the additional effect of each molecule, identify which regions are statistically altered synergistically and reach these conclusions, proper statistical comparisons between the states (AcrBWT+CIP+PAβN) vs. (AcrBWT+CIP) and (AcrBWT+CIP+ PAβN) vs. (AcrBWT+PAβN) are required (at least in the supplementary material).
• The local and allosteric effects of the G288D mutation on the dynamics of AcrB are quite impressive, as shown by the changes in HDX observed on Fig.4b (no drugs). Irrespective of the drug(s), the G288D mutation induces local flexibility (increase HDX) and a long-range stabilization (decrease HDX) of the connecting loop, as seen in Fig.4c. The effects of the drugs are first represented by comparing the HDX of AcrBG288D+CIP or PAβN with AcrBWT, meaning that two parameters (mutation and presence of the drug) are changing and that the contribution of each parameter is hard to apprehend. Actually, it seems to me that the local destabilization of the PN2 region is only due to the G288D mutation rather than by the drugs as it seems to be the case in Fig.4c. The authors then rightfully compare (AcrBG288D+CIP) vs. (AcrBG288D) and (AcrBG288D+PAβN) vs. (AcrBG288D) (supplementary Fig.9), which is better, I think to see the sole effect of each drug on the mutant. In this representation, the destabilization of the PN2 region by the drugs is not visible anymore. I would thus recommend to present these comparisons (AcrBG288D+drug) vs. (AcrBG288D) in the main text and the one from Fig.4c in the supplementary data (just keeping the AcrBG288D vs. AcrBWT comparison from Fig.4c), in order to avoid any misleading interpretation of the data. Alternatively, the authors could represent the comparison of (AcrBG288D+drug) vs. (AcrBWT+drugs) to see the effect of the mutation on the drug-bound AcrB (as mentioned later on lines 295-302).
• The comparison throughout the manuscript of the first hydration shell and HDX levels is very interesting. The authors claim several times that these results of simulations are in good agreement with their HDX data. I think it would be a good idea, for a non-expert readership, to detail a little bit more in what sense these results agree for eg. by saying (in the main text) that a reduced hydration shell should imply reduced HDX and why.
• On the same line, the fact that hydrogen bounds were found in the MD ternary complex simulations, between the two drugs (and not only with the HT and switch-loop) is, I think, important to state in the main text.
• The antibiotics used for MIC assays are different in Fig.2c and Fig.5c (apart from CIP), could the authors explain why?
• The last sentence of the results states that "G288D has a profound impact on AcrB substrate specificity within both E. coli and Salmonella", however the MD simulations clearly showed that both WT and G288D AcrB were able to bind to both CIP (with and w/o PAβN). I don't clearly see how the data presented here infers directly a difference in substrate "specificity" per se (in terms of Kd for eg.) but rather on substrate export (indirectly with the MIC assays) or stability (HDX and MD simulations). Could the authors use fluorescence polarization to determine and compare the Kds of CIP and PAβN to AcrBG288D as they did for the WT construct?
• The authors use a ~1000:1 ligand to AcrB ratio for the HDX experiments: how to make sure that in these conditions, non-specific binding does not occur? This could obviously affect the regions pinpointed by the HDX experiments. Do the authors expect to see a single drug bound to AcrB under such conditions in native MS?
• The HDX-MS data was made available through on Pride (#416950) but I could not find it. The authors need to make sure that the data is either publicly available or provide login and password to the reviewers. The deposited data should include .raw files, DynamX .csv output files, .fasta files and PLGS output files.
Minor Comments: • Supplementary Fig.2e: please map the TM domains to the sequence.
• Supplementary Fig.3 shows the HDX heatmap obtained for AcrBWT alone. The reviewer suggests to include (panel C) a 3D structure color-coded with this HDX-MS data (last time point for eg.). • l.283: "how the mutation effects" should read "how the mutation affects". • The top legend of Supplementary Fig.9 should read (AcrBG288D+drug(s)) instead of AcrBG288D+/drug(s)). In the Material and Methods section: • the flow rates for the protein purification (elution on HisTrap and SEC) should be indicated.
• the MWCO of the spin column for the concentration step should be indicated.
• the argon trap collision gas flow rate should be indicated. • l.670: "Bipharma" should read "Biopharma". • HDX: the volume of sample transferred to the quench solution, the volume of quench solution and the volume of sample injected (as well as the volume of the loop) should be indicated.

Reviewer #1 (Remarks to the Author):
In this manuscript the authors describe their efforts to determine the structural dynamics of the drug efflux pump AcrB in the presence of substrates and the efflux pump inhibitor PAßN.
Complimentary methods of hydrogen/deuterium exchange mass spectrometry and molecular modelling were used to determine the structural dynamics of inhibition and transport. This is a novel approach and yielded interesting data. The structural dynamics were also investigated for AcrB from a MDR clinical isolate with a G288D substitution in the drug binding pocket. This is an accomplished work that would make an important contribution to the field.
The hydrogen/deuterium exchange mass spectrometry were determined for AcrB purified and solubilised in the detergent DDM. Detergent solubilised protein is not necessarily the best representation of the native protein state. The authors should discuss if they would expect differences if the protein were solubilised with SMA technology or reconstituted in a lipid membrane.

(2017) -reference 16 in this manuscript). However, this is an extremely interesting point, which forms part of future work but is outside the scope of this study.
It is interesting that PAßN completely abolish ethidium bromide resistance in AcrB from Escherichia coli. However, PAßN does not compete with ethidium bromide in transport assays in Pseudomonas aeruginosa. What would be the explanation for this difference?

Reviewer #2 (Remarks to the Author):
Reading et al. examine closely the structural dynamics of AcrB by means of molecular dynamics simulations and H/D exchange mass spectrometry. The bottom line conclusion from their studies is that the PAβN efflux pump inhibitor restricts the intrinsic motions of the drugbinding pockets, indicating that structural dynamics play a critical role in the inhibition and substrate specificity of AcrB. The idea is that PAβN does not prevent antibiotic binding but rather inhibits efflux by enforcing a restrained state that reduces the frequency and magnitude of the conformational changes in the substrate translocation path. The impressive amount of work reported appears to be competently done and accurately reported.
The work presents a new point of view that specialists will be interested in. I doubt that more casual readers will be willing to slug through the paper to understand the findings. The paper is complex and relies heavily upon supplementary materials. For example, reading carefully the 'supplementary discussion' is mandatory for appreciating the paper. The paper is more appropriate for a specialist journal.

Other material presented in the Supporting Information include essential protein characterization and additional differential HDX (∆HDX) analysis (as requested by Reviewer #3 and/or requested by the terms set out in the recent HDX-MS 'white paper': Masson, G. R. et al. Recommendations for performing, interpreting and reporting hydrogen deuterium exchange mass spectrometry (HDX-MS) experiments. Nat. Methods 16, 595-602, doi:10.1038/s41592-019-0459-y (2019)).
A minor but irritating point is the overuse of the word 'impact', which the authors use both as a verb and a noun. This has become rampant in the literature. The better usage is affect (verb) and effect (noun).

Reviewer #3 (Remarks to the Author):
In this work, the authors use HDX-MS and MD simulations to investigate the synergistic effect of substrate (CIP) and inhibitor (PAβN) binding to the MDR efflux pump AcrB, together with a clinically relevant mutation (G288D).
The three main finding are clearly stated in the introduction: • PAβN EPI restricts the intrinsic motions of the drug-binding pockets as part of its mechanism of action and is effective against both AcrBWT and AcrBG288D • an EPI can dually bind to AcrB alongside an antibiotic, without affecting its inhibitory action • an MDR mutation in acrB impacts upon the structural dynamics of the efflux translocation pathway likely contributing to its modified substrate specificity The authors used adequate methods (i.e CD and thermal protein unfolding) to confirm that both recombinant constructs (WT and G288D) have similar overall structural and thermal stability. The article is particularly well written, adapted to the broad readership of Nature Communication and yet giving enough details in the supplementary material for structural biologists (concerning the analysis of the MD simulations in particular).
I believe that this article should be considered for publication in Nature communications with editorial revisions.
Suggestions below are designed to enhance its communication and impact for the nonspecialist.
Major comments: • Supplementary Fig.2c-d: Different MWs were obtained native MS before and after delipidation. Could the authors comment on that?  . LPS binding to membrane proteins during purification has been previously identified by additional masses of 3.2-3.7 kDa from the apo form 45 . Data was collected on a Synapt G2-Si mass spectrometer with AcrB WT within Triton X-100 detergent micelles. (d) Native mass spectra of AcrB WT within after optimized purification. Data was collected on a Thermo Scientific Q Exactive UHMR hybrid Quadrupole-orbitrap mass spectrometer with AcrB WT within DDM detergent micelles. (e) Native mass spectra of AcrB G288D after optimized purification. Data was collected on a Synapt G2-Si mass spectrometer with AcrB G288D within Triton X-100 detergent micelles. (f) HDX-MS sequence coverage and peptide redundancy map of AcrB in DDM detergent micelles.

This optimized purification regime produced well-resolved mass spectra with LPS removed, with some phospholipid binding maintained, which closely matches the expected mass for homotrimeric AcrB WT -6xHis (Supplementary figure 2d). We have also included a mass spectra of AcrB G288D in the Supplementary Figure 2e (using the same optimized purification protocol) with a measured mass corresponding to homotrimeric AcrB G288D -6xHis. We have added that the G288D mutation did not affect the homotrimeric oligomeric state of AcrB in the main text also and present the new Supplementary
• l.138: "antibiotic binding only subtly alters the structural motions of AcrBWT": how can the authors make sure that they have a stoichiometric binding and that the binding is not just transient…this sentence is possibly overstated or at least too general as the results might be different at different concentrations and for different antibiotics… The reviewer suggests to replace "antibiotics" by "CIP".
The binding of different antibiotics could alter the structural dynamics of AcrB in distinct ways. We therefore agree with the reviewer that this statement is overstatedespecially when considering the wide range of chemically distinct antibiotics which are substrates for AcrB efflux -and has been corrected.  (kD of 74.1 and 15.7 μM,  • The sequence coverage seems different in the different conditions (CIP or PAβN), could the authors comment on that? On the same line, did the mutation induce a change in the sequence coverage of the region carrying the G288D mutation? Were the authors able to compare the peptide(s) carrying the mutation? Data Tables 1-2, which are provided with the manuscript. G288D region (peptides 281-289, 281-290, 282-289) which allowed us to compare the data within ΔHDX = AcrB G288D + drug(s) -AcrB G288D (Supplementary Figure 11b), as the sequences were identical between the two states. Figure 11b)." G288D . All data reported as in Fig. 2 Table 2. • Simulations show the formation of hydrophobic bonds between PAβN and E130/K131/Q176/A619 (pose 1), E130/D276 (pose 2) and S126/S46/D174 (pose 3). How do the HDX-MS protection upon PAβN binding relate to these different proposed H-bonds? Are there any encompassing peptide significantly protected by PAβN? If yes this should be clearly stated in the text. If not, do the authors think that this might be due to a partial and/or transient binding of the drug? This point should be addressed in the manuscript. 138-149, 162-177 and 263-274 (such  as 128-133, 174-176 and 273-276) Table 6). (Fig. 2a-b and Supplementary Fig. 10; Fig. 9)." G288D -PAβN (Supplementary Fig. 13)

and are involved in high-occurrence interactions with PAβN, similar to those formed within AcrB WT -PAβN (Supplementary Tables 3 and 6)."
• Fig.2 shows the ΔHDX comparisons of (AcrBWT+CIP) vs. (AcrBWT) and (AcrBWT+PAβN) vs. (AcrBWT). We can clearly see that CIP mainly destabilizes the N-ter & Cter of AcrBWT and stabilizes/protects the connecting-loop, while binding of PAβN additionally protects other regions, including the Switch loop. However, a proper statistical comparison between (AcrBWT+CIP) vs. (AcrBWT+PAβN) would be needed before stating that "However, in stark contrast to CIP, inhibitor binding led to HDX reduction throughout extensive parts of the PC1/PC2 cleft of the drug-binding pockets and within the connecting-loop (Fig.2a-b). This could signify inhibitor induced structural stabilization of the drug-binding pocket entrances" (l.139-140). Fig.  2 Table 2.

. All HDX-MS peptide data can be found in Supporting Data
• Fluorescence polarization binding and competition assays indicate that both molecule can interact independently with AcrB. This was further supported by MD simulations of AcrB in the Tstate with both CIP and PAβN nicely showing that PAβN is able to bind to the HT region while CIP lies next to the DBP-PBP interface. I just wonder how the CIP docking is affected by the additional presence of PAβN? Is CIP binding to the same site in the absence of PAβN? Could it be that the presence of PAβN stabilizes one particular bound state of CIP, preventing its efflux? This seems to be suggested by the MD simulation analysis (supplementary material) but maybe not enough clearly stated in the main text. Fig. S6 and Table 1 in Vargiu & Nikaido, PNAS 109 (50), 20637). These results were compared with two additional MD simulations of AcrB WT +CIP, performed following the protocol described the Methods section.  ). See Supplementary Fig. 6 for further details on the representation of the binding poses. As stated in the Supplementary Methods, the binding pose of AcrB WT -CIP features the substrate within the same region as in Ref. [Vargiu and Nikaido, PNAS, 2012], although the orientation is opposite. Hydrogen bond analyses were performed on the last 300 ns of each simulation (see Methods). Only direct or water-mediated bonds with occupancy higher than 20% have been reported.

Supplementary Figure 16. Representative binding poses of AcrB WT -CIP and AcrB G288D -CIP (upper panel) and high-occupancy hydrogen bonds (H-bonds) established between CIP and the protein (lower table
• The authors then present in Fig.2 the ΔHDX comparison of (AcrBWT+CIP+PAβN) vs. (AcrBWT), which indeed shows very similar behavior as the previous comparison (AcrBWT+PAβN) vs. (AcrBWT), suggesting that most fluctuations are due to PAβN and that there isn't much synergy between the substrate and the EPI. However, to see the additional effect of each molecule, identify which regions are statistically altered synergistically and reach these conclusions, proper statistical comparisons between the states (AcrBWT+CIP+PAβN) vs. (AcrBWT+CIP) and (AcrBWT+CIP+ PAβN) vs. (AcrBWT+PAβN) are required (at least in the supplementary material).

The requested differential HDX (∆HDX) plots have been created and are presented in the new Supplementary Figure 5. The new plots show that AcrB WT +CIP+PAβN largely acts like AcrB WT +PAβN. This has strengthened our conclusions and we thank the reviewer for the suggestion for its inclusion. A reference to the figure has also been included into the main text where appropriate.
• The local and allosteric effects of the G288D mutation on the dynamics of AcrB are quite impressive, as shown by the changes in HDX observed on Fig.4b (no drugs). Irrespective of the drug(s), the G288D mutation induces local flexibility (increase HDX) and a long-range stabilization (decrease HDX) of the connecting loop, as seen in Fig.4c. The effects of the drugs are first represented by comparing the HDX of AcrBG288D+CIP or PAβN with AcrBWT, meaning that two parameters (mutation and presence of the drug) are changing and that the contribution of each parameter is hard to apprehend. Actually, it seems to me that the local destabilization of the PN2 region is only due to the G288D mutation rather than by the drugs as it seems to be the case in Fig.4c. The authors then rightfully compare (AcrBG288D+CIP) vs. (AcrBG288D) and (AcrBG288D+PAβN) vs. (AcrBG288D) (supplementary Fig.9), which is better, I think to see the sole effect of each drug on the mutant. In this representation, the destabilization of the PN2 region by the drugs is not visible anymore. I would thus recommend to present these comparisons (AcrBG288D+drug) vs. (AcrBG288D) in the main text and the one from Fig.4c in the supplementary data (just keeping the AcrBG288D vs. AcrBWT comparison from Fig.4c), in order to avoid any misleading interpretation of the data. Alternatively, the authors could represent the comparison of (AcrBG288D+drug) vs. (AcrBWT+drugs) to see the effect of the mutation on the drug-bound AcrB (as mentioned later on lines 295-302). Fig. 4. In Fig. 4b

we present both (AcrB G288D vs. AcrB WT -both apo forms) and ((AcrB G288D +drug) vs. (AcrB WT +drugs) -both holo forms), as was recommended by the reviewer. We have now corrected the labelling within Fig. 4b, annotating the top panel as:
 ΔHDX = AcrB G288D -AcrB WT and the bottom three panels as: Fig. 4c and within the dashed boxes in Fig. 4b.

"Upon substrate binding, PAβN caused reduced HDX within the PC1/PC2 regions for
AcrB G288D (Supplementary Figure 11), as was observed for AcrB WT (Fig. 2a), with AcrB G288D -PAβN having increased HDX reduction within PC1 and R2 (TM 7-12) domains in comparison to AcrB WT -PAβN (Fig. 4b). This supports that the PAβN EPI affects the dynamics of the different AcrB genotypes in a similar manner, with AcrB G288D -PAβN possibly undergoing further restraint than AcrB WT -PAβN. Where CIP caused increased HDX throughout extensive regions of AcrB G288D (Supplementary Figure 11) PAβN conditions (Fig. 4b). (Fig. 4b). Signifying that these effects are retained even upon substrate binding and, due to their close structural proximity, may relate to concerted changes to the dynamics of the substrate translocation pathway (Fig.  4c)."

Markedly, in the apo form and for all three substrate conditions tested (CIP, PAβN, and CIP-PAβN), the G288D substitution consistently caused increased HDX within the PN2 region and decreased HDX of the connecting-loop
• The comparison throughout the manuscript of the first hydration shell and HDX levels is very interesting. The authors claim several times that these results of simulations are in good agreement with their HDX data. I think it would be a good idea, for a non-expert readership, to detail a little bit more in what sense these results agree for eg. by saying (in the main text) that a reduced hydration shell should imply reduced HDX and why.

Further to this we have also reworded our introduction to HDX-MS within the 'Hydrogen/Deuterium eXchange Mass Spectrometry of AcrB' section to aid the understanding of the HDX process, and what this measures, for a non-expert readership: "HDX-MS is a solution-based method which can provide molecular level information on local protein structure, stability, and dynamics 31-33 . HDX occurs when backbone amide hydrogens are made accessible to D2O solvent through structure unfolding and H-bond breakage; HDX is fast within unfolded regions and slow within stably folded regions (i.e. αhelices, β-sheet interiors), where local structural fluctuations which expose an otherwise protected amide hydrogen to solvent transiently are required for HDX to occur."
• On the same line, the fact that hydrogen bounds were found in the MD ternary complex simulations, between the two drugs (and not only with the HT and switch-loop) is, I think, important to state in the main text. Fig. 9)." (Fig.  5a). As in AcrB WT -CIP-PAβN, stabilizing interactions include several substrate contacts with the AcrB G288D protein, also involving D288 (Supplementary Fig. 14-15

Table 7), as well as intermolecular hydrogen bonds between the two ligands (Supplementary Table 8)."
• The antibiotics used for MIC assays are different in Fig.2c and Fig.5c (apart from CIP), could the authors explain why?
The MIC data presented in Fig.2c was to show the effect of increasing PABN  concentration on the MIC of three different classes of antibiotics (CIP, TET, and CHL), all known AcrB substrates, in a wildtype E. coli with wildtype AcrB.

"These findings were supported by bacterial susceptibility assays on E. coli containing overexpressed AcrB G288D . AcrB G288D was previously discovered within Salmonella clinical isolates 30 and found to have increased and decreased susceptibility to MIN and CIP antibiotics, respectively. We chose, therefore, to study these AcrB substrates in the presence of the inhibitor PAβN to observe what, if any, effect would be seen with the different AcrB genotypes. PAβN incubation led to increased MIN and CIP antibiotic susceptibility for both
AcrB WT and AcrB G288D (Fig. 5d). AcrB G288D being more susceptible to PAβN than AcrB WT . The decreased susceptibility of AcrB G288D to CIP, found in Salmonella 30 , was not recapitulated in our assays using the laboratory E. coli strain MG1655. This may be due to CIP efflux via another transporter found in E. coli but not in Salmonella. However, the associated increased susceptibility to MIN was observed (Fig. 5d)

, supporting that G288D has a profound impact on AcrB substrate efflux within both E. coli and Salmonella."
• The last sentence of the results states that "G288D has a profound impact on AcrB substrate specificity within both E. coli and Salmonella", however the MD simulations clearly showed that both WT and G288D AcrB were able to bind to both CIP (with and w/o PAβN). I don't clearly see how the data presented here infers directly a difference in substrate "specificity" per se (in terms of Kd for eg.) but rather on substrate export (indirectly with the MIC assays) or stability (HDX and MD simulations). Could the authors use fluorescence polarization to determine and compare the Kds of CIP and PAβN to AcrBG288D as they did for the WT construct? G288D and AcrB WT . Although AcrB G288D -PABN had a slightly higher binding affinity to CIP than AcrB WT -PABN (KD of 67.3 ± 13.2 µM (AcrB WT ) versus 22.3 ± 3.1  µM (AcrB G288D )). We argue that this demonstrates that the binding specificity and inhibitory action is similar between AcrB G288D and AcrB WT . At least for the CIP-PABN system investigated here. This supports the reviewer's statement that we are not directly observing altered specificity but rather differences in efflux and stability.

The reviewer makes a very good point and suggestion for further experiments. We have performed the analogous fluorescence polarization binding and competition assays used for AcrB WT on AcrB G288D . We found parity between the binding strengths of CIP to AcrB G288D -PABN and AcrB WT -PABN and observed that PABN could not outcompete CIP binding to both AcrB
Indeed, we consistently argue within the manuscript that the alteration in structural dynamics/stability explains why we observe differences in efflux; therefore, we agree with the reviewer that statements on specificity need to be changed. We have therefore changed the reference to "altered substrate specificity" throughout the manuscript and replaced with "altered substrate efflux", this has been changed within the below locations.: -Title: from "Perturbed structural dynamics underlie inhibition and altered specificity of the multidrug efflux pump AcrB" to "Perturbed structural dynamics underlie inhibition and altered efflux of the multidrug pump AcrB". -Abstract: from "altered substrate specificity" to "altered substrate efflux". -Introduction: from "substrate specificity" to "substrate efflux".

The new data figure and corresponding text have also been added to manuscript within the "PAβN EPI inhibits both wildtype and G288D AcrB" section:
"Fluorescence polarization binding and competition assays support that a ternary AcrB G288D -CIP-PAβN is also possible (Fig. 5b-c): i) CIP binds to a preformed AcrB G288D -PAβN complex (KD of 22.7 ± 2.9 µM) with similar, albeit slightly higher, affinity compared to CIP binding to AcrB WT -PAβN (KD of 67.3 ± 13.2 µM); ii) titration of the PAβN inhibitor could not effectively outcompete CIP binding from AcrB G288D -CIP, as was found for AcrB WT -CIP (Fig.  3b). Taken together, the fluorescence polarization binding assays and MD simulation data advocate that AcrB G288D is inhibited by PAβN in a similar manner as AcrB WT ." PAβN (cyan) to AcrB G288D T-state monomer and show their likely binding locations. The pose and its orientation are the same as shown for AcrB WT in Fig. 3c. EG = exit channel gate (blue spheres), SL = switch-loop (yellow), and HT = hydrophobic trap (purple). All computational data, including binding free energies can be found in Supplementary Table 5 and Supplementary Fig. 8, 14-16. (b) Binding of CIP by AcrB G288D in the presence of 150 μM of PAβN as determined by a fluorescence polarization assay performed by Su et al. 39 . All data are reported and fit as in Fig. 3a (R 2 = 0.99). (c) Binding competition assay between PAβN and CIP for AcrB WT . All data are reported as in Fig. 3b. (d) MIC assays of Escherichia coli containing AcrB WT or AcrB G288D in the presence of PAβN and antibiotics. AcrB was overexpressed in MG1655 ∆acrB from a pBR322 plasmid containing its corresponding acrAB genes, natural promoter and 'marbox' sequence. Minocycline = MIN, Ciprofloxacin = CIP, and phenylalanine-arginine-β-naphthylamide = PAβN. † PAβN was added at a concentration of 50 μg/ml. Fig. 3a.  • The authors use a ~1000:1 ligand to AcrB ratio for the HDX experiments: how to make sure that in these conditions, non-specific binding does not occur? This could obviously affect the regions pinpointed by the HDX experiments. Do the authors expect to see a single drug bound to AcrB under such conditions in native MS? Comput. Biol. 12,  e1004840, doi:10.1371/journal.pcbi.1004840 (2016)). In addition, blind molecular docking

clearly shows that binding of CIP and PABN to the DBP and/or PBP of AcrB occurs with the highest frequency and is associated with the highest score (Supplementary Figure 8 and Supplementary Tables 2-3). Furthermore, state-of-the-art MD simulations confirm the formation of stable binary and ternary complexes between AcrB and CIP, PAβN or
PAβN+CIP. We, therefore, expect most binding events to be within the drug-binding pockets of AcrB.
To investigate the binding stoichiometry, we attempted to collect Native MS data of AcrB bound to the drugs, but we were unable to monitor the AcrB-drug complex in the gasphase. This is likely due to the high activation ( • The HDX-MS data was made available through on Pride (#416950) but I could not find it. The authors need to make sure that the data is either publicly available or provide login and password to the reviewers. The deposited data should include .raw files, DynamX .csv output files, .fasta files and PLGS output files.
Our dataset "HDX-MS reveals the perturbed structural dynamics underlie inhibition and altered specificity of the multidrug efflux pump AcrB" has been successfully submitted to ProteomeXchange via the PRIDE database. The data is currently private and can be accessed a single reviewer account that has been created. We have been informed by PRIDE that it is essential we notify them after the first (online) publication of the corresponding manuscript. Otherwise the data will remain inaccessible to readers.
For access to the account by the reviewer during the peer review process the Reviewer account details are provided below: Project accession: PXD019047 Reviewer account details: Username: reviewer33524@ebi.ac.uk Password: 961rnZhP Additionally, we have added the following sentence into the "Hydrogen/deuterium mass spectrometry" section of the Methods, as recommended by PRIDE: "The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository (https://www.ebi.ac.uk/pride/) with the dataset identifier PXD019047".
Minor Comments: • Supplementary Fig.2e: please map the TM domains to the sequence. This has been corrected but reassigned as Supplementary Fig. 2f. • Supplementary Fig.3 shows the HDX heatmap obtained for AcrBWT alone. The reviewer suggests to include (panel C) a 3D structure color-coded with this HDX-MS data (last time point for eg.).

We have generated an additional Supplementary Figure 4, which displays the HDX-MS heat maps from Supplementary Figure 3 translated onto the 3D structure of AcrB WT ; this has been cited in the main text where appropriate.
• l.283: "how the mutation effects" should read "how the mutation affects". This has been corrected within the text.
This has been corrected but has now been reassigned to Supplementary Fig. 11.
In the Material and Methods section: • the flow rates for the protein purification (elution on HisTrap and SEC) should be indicated.

A sentence has been added to the 'AcrB overexpression and purification' Methods section: "A flow rate of 1 ml/min was used during HiTrap and SEC purification."
• the MWCO of the spin column for the concentration step should be indicated.

The below sentence has been modified within the 'AcrB overexpression and purification' Methods section:
"Peak fractions eluted from the SEC column containing pure AcrB were pooled, spin concentrated using a 100K MWCO concentrator (Amicon®), and spin filtered before being flash frozen and stored at -80 °C." • the argon trap collision gas flow rate should be indicated.

A sentence has been added to the 'Hydrogen/deuterium mass spectrometry' Methods section:
"Argon was used as the trap collision gas at a flow rate of 2 mL/min." • l.670: "Bipharma" should read "Biopharma". This has been corrected within the text.
• HDX: the volume of sample transferred to the quench solution, the volume of quench solution and the volume of sample injected (as well as the volume of the loop) should be indicated.

This has been corrected within the 'Hydrogen/deuterium mass spectrometry' Methods section.
Finally, we thank the reviewers for their careful reading and constructive comments and hope that the manuscript is now suitable for publication in Nature Communications.