Gating interactions steer loop conformational changes in the active site of the L1 metallo-β-lactamase

Read the full article See related articles


β-lactam antibiotics are the most important and widely used antibacterial agents across the world. However, the widespread dissemination of β-lactamases among pathogenic bacteria limits the efficacy of β-lactam antibiotics. This has created a major public health crisis. The use of β-lactamase inhibitors has proven useful in restoring the activity of β-lactam antibiotics, yet, effective clinically approved inhibitors against class B metallo-β-lactamases (MBLs) are not available. L1, a class B3 enzyme expressed by Stenotrophomonas maltophilia , is a significant contributor to the β-lactam resistance displayed by this opportunistic pathogen. Structurally, L1 is a tetramer with two elongated loops, α3-β7 and β12-α5, present around the active site of each monomer. Residues in these two loops influence substrate/inhibitor binding. To study how the conformational changes of the elongated loops affect the active site in each monomer, enhanced sampling molecular dynamics (MD) simulations were performed, Markov State Models (MSM) were built, and convolutional variational autoencoder (CVAE)-based deep learning was applied. The key identified residues (D150a, H151, P225, Y227, R236) were mutated and the activity of the generated L1 variants was evaluated in cell-based experiments. The results demonstrate that there are extremely significant gating interactions between α3-β7 and β12-α5 loops. Taken together, the gating interactions with the conformational changes of the key residues play an important role in the structural remodeling of the active site. These observations offer insights into the potential for novel drug development exploiting these gating interactions.

Article activity feed

  1. eLife assessment

    In this useful study, the authors utilize state-of-the-art computational methods complemented with some experimental validation to investigate the dynamics of flexible loops of the L1 Metallo-β-lactamase enzyme, resulting in a better understanding of the various conformational states useful for the rational design of superior β-lactamase inhibitors/antibiotics. The evidence supporting the claims is solid, and the work will be of interest to computational, experimental biologists, and drug designers working on antibiotic resistance.

  2. Reviewer #1 (Public Review):

    In the study, Zhao et al. investigated loop conformational changes in the active site of L1 Metallo-beta-lactamase. Antibiotic resistance is on the rise and beta-lactamases are enzymes that cleave a lactam ring. Authors investigate class B3 MBLs since these could be used for designing drugs for treating antibacterial resistance. Authors find specific loops that act as gates to the shape and access to the active site of the enzymes. They study these loops via MD simulations, Markov state models, and CVAE-based deep learning to experimentally reveal how each residue affects activity as well as remodeling of the active site.

    - The authors make a good case for why MD is important for this scaffold and protein class. The study performs MD simulations coupled with Markov State Models - this coupled with CVAE to understand the different states the protein exists in shapes the state-of-the-art study. Authors are able to isolate three different states that the protein exists in and pinpoint which interactions cause a reshaping of the active site.
    - Furthermore, they isolate the likely states that also correspond with lower free energy indicating why these states might be more populated. This study adds to the depth of their work.

    - Overall, the impact of work on the currently used antibiotic classes is unclear since the total market presence of all antibiotics is discussed not the carbapenem-based antibiotics class. Statistics related to broad antibiotic class reduce the impact statement instead of improving it.
    - Finally in the experimental testing only a few variants at each position were tested, leading to limited learning of the impact of active site interactions.
    - Authors state from previous studies on TEM-1 that disruption of the salt bridge between the two loops would alter the binding site, thus reducing antibiotic resistance. The authors continue on to hypothesize that this would hold true for the structure in consideration for this paper as well. While a good hypothesis, this cannot be inferred until we see experimental evidence for the same or a sequence comparison discussing how similar TEM1 is to the L1 MBL in question.
    - The authors do not explain how different splits of this data in terms of splitting (80:20 vs 70:30 or others) and reducing interaction matrix lower than 22 x 22 residues can impact their results. Also, the effect of changing the distance shell (8A) for matrix generation is not described. This variation is unaccounted for and can enable authors to pressure test their method and learnings.

  3. Reviewer #2 (Public Review):

    Zhao, Shen et al. ran molecular dynamics simulations, followed by the application of Markov State Model analysis and deep machine learning dimensional reduction, to study the dynamical behavior of two loops close to the catalytic site of L1 Metallo-β-lactamase (MBL).

    The simulations are carefully executed and of sufficient length to build a representative kinetic model. Using a dimensional reduction of the loop conformational sampling based on backbone dihedral features followed by tICA embedding, the authors obtain a Markov state model that identifies the main conformational states of the loops in the absence of bound ligands and provides estimates of the timescales for the transitions between them. Next, the authors employ an alternative way to cluster the conformations of the loops, using unsupervised dimensional reduction, implemented as a convolutional VAE applied to residue distances followed by tSNE embedding. This second step gives results that are not consistent with the clustering used for the kinetic modeling (for instance, supplement 2 of figure 4 shows that the 7 macrostates obtained by the MSM analysis don't always correspond to different areas in the CVAE+tSNE embedding).

    Moreover, an inspection of the results from both analysis techniques just confirms the role of interactions that are readily observed in the available crystal structure stabilising the most populated, closed conformation of the loops. The sophisticated computational analysis does not elucidate much of the role of the loop dynamics beyond the intuitive conclusion that disruption of the key interactions keeping the loops in the closed state would affect the function. For instance, it does not clarify what is the role of the other observed metastable states.

    Finally, the authors propose and test mutations that would likely disrupt the stability of the closed state and find that they have variable effects on the ability of the enzyme to contrast the antibiotic effect of a panel of substrates. These experimental results look useful and can potentially be used to elucidate the role of the loops in the recognition and activity of the enzyme, and for the design of inhibitors. However, no additional attempt is made to clarify the experimental results based on the mechanistic model of loop dynamics: why do different mutations have different effects? why do some mutations affect all substrates, other mutations only some substrates, and for others, no substrate is affected? What is the role of the tetrameric arrangement?

  4. Reviewer #3 (Public Review):

    The authors provide a molecular dynamics (MD)-based detailed evaluation of the contribution of the two elongated loops (alpha3-beta7 and beta12-alpha5), present near each active site of the tetrameric Stenotrophomonas maltophilia class B Metallo-beta-lactamase (MBL) L1, towards the L1's lactamase activity with the premise that a better understanding of the categorical conformational states sampled by the loops would ultimately help in the design of a better lactamase inhibitor. This is to then ultimately alleviate the public health crisis arising from β-lactam antibiotic resistance. Using enhanced sampling MD, Markov state modeling (MSM), and convolutional variation autoencoder (CAVE)-based deep learning, the authors identify five key interacting residues in these two loops which contribute to the conformational states of loops.

    The major strength of the study is that the authors carry out a detailed study (e.g., enhanced sampling MD, Markov state modeling, and convolutional variation autoencoder-based deep learning) of the conformational landscape of an important enzyme as these findings would help further experimental studies (e.g., NMR dynamics) for ligand binding, better design of inhibitory ligands of an important class of enzyme. One weakness would be that MBL L1 is a good representative of the class of MBL enzymes or not needs clarification.

    The authors achieve the goal of capturing the various conformational states of the L1 enzyme loops and their computational results support the conclusion about the various loop conformations sampled during the dynamics. However, how the mutagenesis experiment supports the existence of different conformational states will likely benefit from more clarification. Further clarification on how detecting the existence of multiple conformers benefits better inhibitor design will be very beneficial.

    Since details on macromolecular motion are often neglected in macromolecular experimental studies, the detailed MD methods described here will be a very useful companion in experimental studies of proteins and their interactions.

    A discussion on how the study of one particular enzyme could benefit in understanding the molecular properties of a class of enzymes would enhance the generality of the study.