PREreview of Energetic and structural control of polyspecificity in a multidrug transporter

Review For: Energetic and structural control of polyspecificity in a multidrug transporter

https://www.biorxiv.org/content/10.1101/2025.04.09.647630v2.full

Summary

Multidrug transport proteins are clinically important proteins responsible for multidrug resistance, but the mechanisms directing their action and promiscuous transport behavior are poorly understood. This paper studies NorA, a protein-coupled efflux pump, by performing a deep mutational scan (DMS) to interpret the functional role of each residue on the export of eight substrates. The authors find that residues across the full protein, not only those in and surrounding the active site, are responsible for determining substrate specificity. Further, the paper analyzes the role of each residue in exporter efficiency by comparing exporter function in different pH conditions, which provide different proton gradients to drive transport. The authors find a correlation between promiscuous and efficient transport behavior.

The paper provides significant insight into structural mechanisms of NorA function and substrate selectivity. Their DMS data verifies amino acids known to play roles in NorA function and identifies sixteen novel functional hotspots in the protein responsible for function. Their screens across multiple substrates indicate specific residues and interactions responsible for binding and transporting specific substrates. Complementing this, their screen in multiple pH conditions provides a straightforward method to assay protein efficiency and indicates residues and regions responsible for efficient transport. 

A central assumption underlying several conclusions is that ΔF_pH values derived from norfloxacin can be extrapolated to represent energy efficiency for all substrates tested. While the observed trends are compelling, the lack of direct ΔF_pH measurements across multiple drugs leaves this generalization only partially supported.

Overall, this work provides improved understanding into the functional and structural characteristics of NorA, in both specificity and efficiency metrics. The assay using pH to modulate available energy is a simple and powerful method to measure transporter efficiency, and the paper’s methodology provides a good approach for understanding other multidrug transporters in the future. 

Major Points:

• The abstract and introduction pose the question of the relationship between transporter promiscuity and efficiency very abstractly in a way that implies a common relationship or structural trend governing all multidrug transporters. However, this paper’s introduction does not provide a reason to suspect such a general trend exists. While this paper’s results are insightful for NorA, they do not demonstrate its structural findings or promiscuity-efficiency relationships apply to other transporters. Framing this conclusion as a trend observed in this system, rather than a general rule, would more accurately reflect the data and prevent overextension of the model’s applicability.

• The last two paragraphs in the section “Unsupervised Hierarchical clustering reveals sequence determinants of specificity” attempt to identify “functional hotspots” and use Rosetta predicted Δ⁢ΔG to determine mutations which cause misfolding. However, Rosetta predicted Δ⁢ΔG is not an accurate predictor of either true Δ⁢ΔG or misfolding tendency; the cited Tiemann et al paper “Interpreting the molecular mechanisms of disease variants in human transmembrane proteins” states that “when we compared experimental and computational Δ⁢ΔG values, we obtained a Spearman rank correlation coefficient of 0.46, leaving uncertainty about the predictability of the extent of loss of stability.” To improve confidence in the proposed mutational hotspots, the authors should use an alternative mechanism to determine which mutations are folding or misfolding. 

  • In addition, the authors should elaborate on their cutoff choices for determining if a protein is misfolded (Δ⁢ΔG exceeding 1 standard deviation in all conformations) and if a position is a mutational hotspot (5 or more missense mutations that disrupt function without causing misfolding). While they acknowledge these thresholds are arbitrary and their choice “balances strictness with practical sensitivity”, they should elaborate on how they chose these values and why they provide the best results (either in the text or in Supplemental Figure 12).

• A central conclusion of the study is that energy efficiency, measured as ΔF_pH using norfloxacin, broadly predicts the substrate promiscuity of NorA variants. While their usage of functional difference between pH conditions is an innovative and practical approach, their analysis rests on the assumption that ΔF_pH for the single substrate norfloxacin is a valid proxy for the general energy-coupling capacity of the transporter across all drugs. The authors partially support this assumption by demonstrating a correlation between ΔF_pH and substrate promiscuity (Figures 4C–E), and by referencing similar trends observed using acriflavine in Supplementary Figure 14. However, this evidence is limited, and if energy-efficient variants for norfloxacin are not the same as those for acriflavine, it would challenge the assumption that ΔFpH measured from norfloxacin generalizes across substrates. This concern is particularly relevant in Figure 6B, where the model is generated using the ΔF_pH values from norfloxacin. Direct ΔF_pH comparisons across multiple substrates, or clear evidence of shared efficiency profiles (for instance, presenting a scatter plot of  ΔF_pH  from norfloxacin and  acriflavine), would strengthen the authors’ conclusion. If such assays or analyses are not feasible, it may still help readers if the authors explicitly acknowledge that using norfloxacin-based ΔF_pH as a global efficiency metric is an assumption, and the plots in Figures 4C–E are not based on ΔF_pH values derived from experiments conducted at two pH conditions for each of the eight substrates.

• The authors cloned 14 NorA variants and measured clonal ΔF_pH to validate the accuracy of values from high-throughput sequencing. However, the plot in Figure 5B shows the overall set of mutations has a positive and fairly good Spearman R of 0.77, it also indicates the significant limitations in using high-throughput ΔF_pH as a proxy for clonal ΔF_pH: mutations across a wide range of low clonal ΔF_pH scores, from -0.1 to -0.6, are mapped to a small range of high-throughput ΔF_pH scores, from -1.75 to -2.5, and these set of points do not follow a positive trend line; conversely, mutations across a small range of positive clonal ΔF_pH scores, from -0.1 to 0.2, are mapped to a large range of high-throughput ΔF_pH scores, from -1.25 to 1.0. These results indicate high-throughput ΔF_pH values have low sensitivity for clonal ΔF_pH in loss of efficiency mutations and high-throughput ΔF_pH values introduce additional variance from clonal ΔF_pH in wild type-like mutations. These trends should be mentioned when analyzing Figure 4B and considered when interpreting high-throughput ΔF_pH data.

  • This analysis has further uncertainty because the authors do not explain how these 14 variants were selected. Without clarity on whether they represent the full spectrum of functional, structural, and substrate-specific diversity, it is difficult to assess the strength of this validation. Highlighting a selection strategy - such as including mutations near proton-coupling residues (e.g., Glu222, Asp307), others in distant structural regions, and some from substrate-specificity clusters - would strengthen confidence in the robustness of the pooled screening method. 

• In Figure 5C, the authors use RT-qPCR to measure mRNA levels of NorA variants and report minimal variation across mutants. However, they do not directly assess protein expression. While they supplement this with Rosetta-based ΔΔG predictions to estimate structural stability, the absence of empirical protein-level measurements leaves uncertainty as to whether functional differences could stem from variation in translation, folding, or membrane localization.  Moreover, the correlation between RNA abundance, ΔΔG predictions, and DMS-derived fitness scores appears weak. It is also important to note that Rosetta ΔΔG is known to be unreliable for membrane proteins due to limitations in modeling transmembrane environments and conformational flexibility. Given these concerns, the conclusion’s reference to ΔΔG as one of the three fitness dimensions in the “multiparametric screen” framework is generally misleading. Unlike drug resistance and pH sensitivity, Rosetta ΔΔG is not experimentally derived but is instead a computational prediction. For clarity, the authors should reframe this metric accordingly, and the model in Figure 6A should be revised to reflect ΔΔG as an inferred rather than a screened parameter.

Minor Points:

• In Figure 1A, it would be beneficial to note the binding pocket either by showing a small molecule at that position or by coloring the implicated residues.

• Figure 1C presents fitness distributions across all tested substrates, which is helpful for visualizing the overall range of mutational effects. However, additional analysis—such as stratifying variants by functional class or highlighting partial loss-of-function variants—would help assess whether the screen has sufficient resolution to distinguish intermediate phenotypes. This is particularly relevant given that some IC₅₀ curves suggest step-like behavior rather than gradual shifts in transporter function.

• Figure 2E is referenced to state “permitted mutations are common in unstructured loops and distal support helices.” However, the plot shows these mutations have the most frequency specifically at the unstructured regions surrounding these distal support helices. Clarifying or otherwise nuancing this claim would be beneficial.

  • Figure 2E additionally shows the distribution of universally disabling mutations along the sequence. These are not referenced and make the plot slightly difficult to interpret under the colors used.

• Figure 2F shows residue distance from the binding site, which was determined in the “inward-open AlphaFold2 structure” by calculating the distance of each residue from a list of binding site residues which were “determined by examining the structure in PyMol.” The analysis using these measurements would benefit from additional justification for the decisions in this process including the choice to use the AlphaFold2 structure rather than a cryo-EM structure, whether the structure has good resolution / confidence at the binding site, how the binding site region was determined to identify binding site residues, and whether the binding site residues might differ in the closed conformation.

• In the analysis of allosteric mutations, the authors focus on “third-shell” residues, defined as those not contacting the ligand or any ligand-contacting residue. However, second-shell residues are also not directly contacting the substrate, and may play key roles in allosteric regulation. The rationale for excluding second-shell residues from the allosteric analysis is unclear. 

• In Figures 3B and 3E, the authors show an enrichment in aromatic residues among mutations conferring puromycin-specific tolerance and cation-specific impairment. In Figures 3G, the authors show an enrichment in mutation to proline residues among mutations causing ethidium-specific tolerance. These changes should be discussed in the paragraphs interpreting these mutations to fully examine the mechanism behind these compounds’ binding (such as the role of proline mutations in breaking helices near the active site).

• The discussion of Figure 3D describes pentamidine as having a polar surface and so able to accommodate mutation of the basic R310 residue. The discussion of Figure 3F describes pentamidine as hydrophobic to explain why nonpolar to polar mutations would disrupt its binding. These explanations are in conflict, so one or both of these mechanisms should be reconsidered.

• In Figure 4, the study relies on norfloxacin resistance as a proxy for efflux activity at different energetic conditions, which are determined by adjusting pH and thus proton gradient. However, the change in pH may also affect norfloxacin’s passive uptake, solubility, or aggregation. Without direct measurement of intracellular drug levels, it is difficult to fully exclude these as confounding effects. The authors could acknowledge this limitation in the discussion.

• In Figure 5A, the IC₅₀ curves appear to be largely driven by a relatively small number of plateau values. This limits confidence in IC₅₀ quantification. A tighter titration around the expected IC₅₀ values would provide more confident and higher resolved information on variant-specific differences. Moreover, the high-throughput screen may lack sufficient sensitivity to detect partial loss-of-function variants, especially those with moderate reductions in transporter activity. This could bias the dataset toward either strongly functional or nonfunctional mutants and underestimate the true spectrum of intermediate phenotypes. Including additional intermediate-resistance variants in clonal validation would improve confidence in both the ΔF_pH score and the dynamic range of the screening method.

• The authors find that mutations in residues throughout the protein (not only those near the binding site) are responsible for controlling specificity. In contrast, they only discuss mutations near the active site as responsible for controlling efficiency. It would be beneficial to propose why they fail to see efficiency-regulation mutations in other regions of the protein.

• The model presented in Figure 6A illustrates that usable energy (ΔG_available) is a function of the strength of the driving force (ΔG_proton) and the coupling efficiency (η). While conceptually sound, the model is built directly on presenting classical thermodynamic principles in a transporter-specific context. It would strongly benefit from further mechanistic depth or empirical integration. 

  • The concept in Figure 6A might be better represented as a pair of lines in a 2D plot of ΔG_proton against available energy. Showing a 3D surface complicates this representation.

  • The model implies that some substrates require more energy to be transported (“expensive”) than others (“cheap”), but this classification is not explicitly defined. Proposing measurable criteria, such as charge, hydrophobicity, molecular weight, or binding affinity, would help ground these terms in physicochemical properties and strengthen the biological interpretation of the model.

• The final paragraph of the discussion refers to potential applications in biofuel and bioproduct production. This application is not introduced or supported elsewhere in the manuscript. Unless the authors intend to expand on this point, replacing it with a more relevant forward-looking statement, perhaps related to multidrug resistance mechanisms or transporter engineering, might make for a more coherent conclusion.

Competing interests

The authors declare that they have no competing interests.