MLRA BACKGROUND

MlrA: The microcystin degrading enzyme

MlrA is a member of the gene cluster mlrABCD, which is required for the full degradation and subsequent inactivation of the microcystin cyclic peptide mclr. A critical step in the inactivation of this toxin is carried out by mlrA; and the subsequent B, C, and D clusters are used in further degradation. Due to its specificity to toxins and known function, it is a promising candidate for combatting water toxicity due to cyanoblooms via cellular uptake (Maseda et al., 2012).

Homology model of mlrA based on template 4cad.1.C. Constructed using the SWISS-MODEL engine (Bienert et al., 2017).

MCLR BACKGROUND

MC-LR: The harmful cyanotoxin

MC-LR is a hepatotoxin, i.e. a toxin that has been shown to be incredibly harmful to the liver tissue of animals. MC-LR adopts a cyclic structure with non-standard amino acids (Greer et al., 2018). The mechanism of MC-LR interacting with its target protein in epithelial cells, phosphatase 2A, leading to functional inhibition as studied by Liang et al. (2011).

MC–LR cyclic hepatotoxin structure (PDB ID 1LCM).

Degradation pathway of MC-LR by the mlrA peptidase (Bourne et al, 2001) The cyclic heptapeptide with the highest degree of toxicity is degraded by mlrA. Firstly, mlrA cleaves at the ADDA–Arg peptide bond of cyclic MC–LR (MW = 994 u), resulting in MC–LR linearization (MW = 1012 u) and 160–fold loss of MC–LR reactivity with target phosphatase 2A.

MlrA contains a metal–binding motif characteristic of metalloproteases, HEXXH, thought to interact with MC–LR. Xu et al. (2019) previously studied this interaction, and based on the H260AIH263NE265 motif, thought to be a variant of the HEXXH motif. Their study examined this as the active site for the linearization of the cyclic peptide, but in their studies found that this active site was not possible due to the inability of the Glu265 to act in catalysis with the mlrA H260AIH263NE265 sequence.

Previously hypothesized metalloendopeptidase MC-LR active site. Residues previously thought to be a part of the zinc-binding domain shown, with H263, and H260 as zinc-binding residues, and N264 as the nucleophilic water hydrogen-bonding residue.

Mechanism of MC-LR ADDA-Arg cleavage by mlrA proposed by Xu et al. (2019)

We had the following overall objective:

To improve the characterization of the mlrA and MC—LR interaction using computational modeling. Currently, the homology model is static and accuracy of structural prediction as well as dynamic behaviour is greatly improved by using MD simulations. We hypothesized that through using AMBER force field-based MD simulations, the structural behaviour of mlrA can be better characterized. To ensure reliability in prediction and reproducibility of convergence in the conformational ensemble, we chose a biologically relevant time-scale of 500 ns for the MD production runs with a consistent starting minimized, heated, and equilibrated structure run in duplicate. This allowed us to observe general trends and minimize the effect of statistically irrelevant fluctuations in behaviour. Our plan based on this year was summarized in the following diagram:

The SWISS-MODEL homology model provides an initial crude predicted structure based on the glutamate protease. Allowing us to obtain approximate atomic coordinates of the enzyme. Molecular dynamics simulations (MD) then allowed us to build on the biases of structure prediction solely using homology modeling by firstly introducing a water box which gives a much better prediction of the different interactions affecting the main and side-chain folding of the protein. Once the equilibrated model was first obtained, which acts as a standard first stage in order to obtain input coordinates for MD production simulations of the protein that has been acclimated to solvent conditions by restraining. Next, the production MD is carried out allowing the protein to move without restraints to obtain simulated behavioural data.

Different technical aspects during trajectory setup are highly dependent on the objective. Molecular dynamics is concerned with atomic motion and therefore also simulates the convergence of the enzyme as well as how it arrives at its preferred structure from the homology model starting point. Without crystallographic information of the protein being available, our model provides an ensemble of minimized structures as well as motion of atoms over time calculated with the best computational tools available to capture the structural bases of biochemical behaviour.

Following the MD production simulations, analysis was completed using CPPTRAJ. This program reads our production trajectories and outputs analysis data. Using CPPTRAJ, we completed both visual analysis of enzyme structure as well as different quantitative analyses. Firstly, inter-replicate structure validation was performed due to the highly chaotic nature of simulations containing the motion of thousands of atoms with consistent initial coordinates but randomized initial velocities.
Our trajectories of the mlrA enzyme provide a look into the hypothesis from Xu et al. (2019), as they propose water coordinates between Glu172 and His205. By examining the interactions between these residues and waters first through radial distribution and followed by AMBER Grid Inhomogeneous Solvation theory (GIST) analysis, not only can we test this hypothesis, but we can also obtain the optimal water geometry for the following docking simulations. From our docking simulations of the best predicted possible structures that are improved from previous homology models, and our understanding of the active site location, the chemical interactions, structural behaviour, and energetics of binding can be better understood. When the coordinates of initial active site orientations are modelled, MD trajectories can be run with the bound ligand-enzyme structure.

With the previously computed 500 ns production simulations of mlrA, we analyzed the probability density of hypothetical nucleophilic waters in the postulated active site of the mlrA enzyme. This served to both confirm a water molecule acts as a nucleophile during catalysis as well as providing its coordinates for docking simulations. This tool uses a manually drawn grid to calculate probability densities of all waters occupied in this region during the simulation (Figure 6). Expanding on the modelling work from last year’s project, this year’s objective was further characterization of the binding action of mlrA. Firstly, Grid Inhomogeneous Solvation Theory (GIST) was used to build on previous findings in the radial distribution of water analysis.

Radial distribution boxes of water on two different replicates of 500 ns production simulations of mlrA. Both distributions are not showing similar results of previous findings in the radial distribution of water analysis.

Unfortunately, our results are currently inconclusive as our runs did not display expected radial distributions of water analysis. Currently we are trying to troubleshoot what went wrong with our GIST runs and are hoping that this can be fixed in a future project. NEW SECTION AUTODOCKING We ran autodocking simulations using the open web server ClusPro (https://cluspro.bu.edu/publications.php). This webserver allowed us to understand how the mclr toxin would interact with the mlrA protein. This web server provided us with 4 main replicate jobs, each with 6 sub-replicates included. The images shown below are our top hits when running both our molecular-dynamics-ran protein model, and our control non-dynamics protein model.

The left image is the autodocking results of the mclr toxin docking to an mlrA protein that was run through molecular dynamics simulation. The right image is the autodocking results of the mclr toxin docking to an mlrA protein that wasn’t ran through molecular dynamics simulation. Both toxin ligands docked near the active site, shown in white in the left image, and shown in blue in the right image, of the mlrA protein. Both autodocking results seem to match the area that the mclr ligands would theoretically approach, the HAIHNE amino acid active site. One concern is that the outputted-ligand structure does not match our current models for the mclr toxin. This is most likely due to the non-standard residues not being properly parameterized in the third-party web servers used to run autodocking jobs. A future project would be to undergo parameterization of the non-standard residues and run autodocking again to conclude. This would involve following a procedure similar to Xu et al, 2019, where Gaussian 09 was first used with the Hartree-Fock level of theory and 6-31G* basis set to fit electrostatic potentials to MC-LR residues. After that, 100 ns MD Simulations using the general AMBER force field (GAFF) would be run to supply ligand poses for subsequent docking simulations.

REFERENCES