Structural characterization of LsrK to target quorum sensing and comparison between X-ray and homology model

Quorum sensing is being investigated as an alternative therapeutic strategy in antibacterial drug discovery to combat bacterial resistance. LsrK is an autoinducer-2 kinase, playing a key role in the phosphorylation of autoinducer-2 (AI-2) signalling molecules involved in quorum sensing. Inhibiting LsrK could result in reduced pathogenicity by interfering with the quorum sensing signalling. Previously, we have generated homology models to identify LsrK inhibitors using structure-based virtual screening and successfully found the first class of LsrK inhibitors. While conducting these studies, the crystal structure of LsrK was released providing us an opportunity to inspect the reliability and quality of our models. Structural analysis of crystal structure and homology models revealed the consistencies of constructed models with crystal structure in the structural fold and binding site. Further, binding characteristics and conformational changes are investigated using molecular dynamics. These simulations provided us insights into the protein function and flexibility that need to be considered during the structure-based drug design studies targeting LsrK.


INTRODUCTION
Increased bacterial resistance has become a global health threat that urges the development of novel therapeutics. One of the main reasons for antibiotic resistance is the conventional mechanism of action of existing drugs i.e. targeting the protein synthesis, cell wall synthesis, and DNA replication 1,2 . Currently, novel strategies such as targeting virulence are prominent in antibacterial drug discovery 3,4,5 . Virulence strategies reduce the selection pressure on bacteria to develop resistance as these processes are not essential for bacterial growth 6,7,8 . The major focus of these strategies is on disrupting the host-pathogen interactions and reducing the pathogenicity by inhibiting the adhesion and toxin release, biofilm formation, and quorum sensing 9,10,11 . Quorum sensing (QS) is a process used by bacteria to communicate between the species and among the species. This communication controls the population-based behaviours and functions such as virulence factor secretion, biofilm formation, motility, bioluminescence, sporulation, and development of genetic competence 12,13 . QS process is mediated by signalling molecules called autoinducers (AIs). These signalling molecules can be devided into three major groups: Acylated Homoserine Lactones (AHL), Autoinducer peptides (AIPs), and Autoinducer-2 (AI-2). AHLs are N-Acyl-L-homoserine lactones varying in their acyl chain length between 4 to 18 carbon atoms while AIPs are oligopeptides.
Generally, AIPs are utilized by gram-positive bacteria whereas AHLs are used by gram-negative bacteria 14 . In contrast, AI-2 molecules are the universal signalling molecules that are used by both gram-positive and gram-negative bacteria. AI-2 produced in bacteria will be internalized from the extracellular environment into the cells by an ATP binding cassette (ABC) transporter system called the Lsr transporter. Further, AI-2 is phosphorylated by LsrK (encoded by the lsrK gene) inside the cell and undergoes further modifications by LsrF and LsrG. The isomerized Phospho-AI-2 is responsible for the lsr operon activation and inactivation of a repressor protein, LsrR 15 .
Thus, hindering the phosphorylation of AI-2 can be a promising strategy in the design of antibacterial drugs. First reports by Zhu et al. provided details into the role of LsrK in AI-2 phosphorylation and its mechanism 16 . LsrK phosphorylates the AI-2 precursor, DPD (4,5dihydroxy-2,3-Pentanedione) and thus regulates the AI-2 signalling and QS process. Implying the role of LsrK in QS signalling and virulence regulation, we explored LsrK to identify anti-virulence agents by employing the homology modelling and virtual screening approaches 17 . Recently, LsrK 3 crystal structure (E. coli) was published by Ha JH et al. with a phosphocarrier protein, HPr. These studies revealed the role of LsrK kinase activity and how its activity is modulated by HPr protein 18 .
It provides an opportunity to evaluate the quality of our homology models and gain insights into the details of protein flexibility and inhibitor design targeting the LsrK.
In this study, the main focus is to inspect the quality of homology models using the crystal structure and gain insights into protein flexibility. Molecular dynamics simulations were employed to understand the protein conformational changes of LsrK using crystal structures and constructed homology models. The structural details provided an understanding of the LsrK structure and the conformational changes occuring during the ATP and substrate binding. These details are helpful to guide the structure-based inhibitor design targeting the LsrK kinase to interfere with the quorum sensing process.

Comparison of homology models and crystal structure
The release of E. coli LsrK structure (ecLsrK), crystallized with HPr protein, raised the interest to inspect the structural quality of our homology models (S. typhimurium:stLsrK). Both sequences were aligned to inspect the sequence differences using the ClustalW alignment server and depicted in Figure S1 using ENDScript 3.0. The sequence identities between ecLsrK and stLsrK were 82.64%. The major variation was found in Domain I of residues 76-85, and in Domain II of residues 419-424 and 496-503. For the structural comparison, homology models were aligned with the X-ray crystal structures using the protein structure alignment. The X-ray crystal structure is in the open state and thus the open state homology model was correlated with the X-ray crystal structure (CS) containing ATP, and a cryoprotectant (hexane-1,6-diol) (PDB ID: 5YA1). Alignment of the model and CS revealed that secondary structural elements (structure helices, strands and loops) are in good agreement with the X-ray structure except helix (a13) of residues 326-337 ( Figure 1 and see Supporting Information for numbering Figure S2). This region was predicted as a loop in the homology model whereas in the CS it is a helix and located in the vicinity of the ATP binding site. However, none of the residues in this helix are interacting with the ATP.
Further, RMSD was checked between the homology model and CS. The overall backbone RMSD was 2.89Å and the RMSD for binding site residues was 0.97Å.  However, the binding site residues around the substrate binding site are similar in CS and homology model (Figure 2). To understand this in better detail, we inspected the X-ray structure and associated electron density maps. Density was not fully visible for the hexanediol ( Figure SX).

Protein flexibility and molecular dynamics
Crystal structure is available in apo form (CS-Apo), with ATP (CS-ATP), and with ADP (CS-ADP) in PDB IDs 5YA0, 5YA1 and 5YA2 respectively. There are no major conformational differences observed between the CS-Apo, CS-ATP, and CS-ADP. To understand how the protein flexibility and conformational changes occur during the substrate and ATP binding, molecular dynamic simulations were carried out on both homology models and all the crystal structures.
Simulation trajectories were analysed for the Ca-atoms RMSD during the 500ns timescale ( Figure   3).  43  57  71  85  99  113  127  141  155  169  183  197  211  225  239  253  267  281  295  309  323  337  351  365  379  393  407  421  435  449  463  477  491  505  519  533  547  561  575  589  603  617  631  645  659  673  687  701  715  729  743  757  771     homology models. However, the loop1 (residues 35-45) stay in the proximity of catalytic cleft affecting the binding site size. This might be playing the role of gatekeeper during the substrate binding and the catalytic reaction in LsrK. Unfortunately, this is not established yet and this loop was not solved in the crystal structure (hence modelled using the Prime loop modelling for the simulations). Further, this analysis also revealed domain movements during the simulation and is also evident from the pocket shape and size analysis discussed in the next section.
Understanding the domain movements using pocket shape and size analysis.  (Table T1-

CONCLUSIONS
The necessity to design new antibacterials has been raised to fight the emerging microbial resistance for the existing drugs and antibiotics. Alternative therapies such as targeting virulence to address the resistance is gaining interest in the research areas. One of such efforts is targeting the bacterial quorum sensing process that controls virulence and pathogenesis. LsrK is the key kinase involved in the quorum sensing process that regulates the lsr operon. Implying the role of LsrK in QS and virulence, computational methods are employed to understand the binding characteristics of LsrK and conformational changes associated with it. Our previous virtual screening driven by homology models led to the identification of first class of inhibitors for LsrK.
The current study proves the quality of homology models and structural consistency with the crystal structure. Simulations are providing details about domain movements and structural flexibility that can help the structure-based drug design efforts to target LsrK binding site.
Experimental studies are needed to depict the phosphorylation events occurring in the LsrK active site as this information would help LsrK targeted drug design further.

Homology modelling
Structure of LsrK kinase of Salmonella typhimurium was modelled using FGGY carbohydrate with ≥ 95% probability of the correct fold. Based on these statistical parameters, homology models were found to be of optimum quality that can be used for virtual screening purposes. Thus, homology models were further used to identify LsrK inhibitors using structure-based virtual screening. The detailed procedures of the homology modelling, virtual screening and experimental bioassays can be found in our previous paper 17 .

Comparison of X-ray structure and Models
Sequence and Structure Analysis: LsrK sequences (E. coli and S. typhimurium) were retrieved from UniprotKB and analysed for the identity and similarity using ClustalW alignment server. The additional files are available as pdf.

Author Contributions
The original draft was written by P.M. Manuscript was edited and reviewed by T.L and A.P. All authors have given approval to the final version of the manuscript.

Funding Sources
This project has received funding from the European Union's Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie grant agreement No 642620 (INTEGRATE).

ACKNOWLEDGMENT
We also thank CSC -IT Center for Science Ltd. Finland for the use of their facilities, software licenses, computational resources and the Biocenter Finland/DDCB for financial support.