Natural Phenolic Inhibitors of Trichothecene Biosynthesis by the Wheat Fungal Pathogen Fusarium culmorum: A Computational Insight into the Structure-Activity Relationship

A model of the trichodiene synthase (TRI5) of the wheat fungal pathogen and type-B trichothecene producer Fusarium culmorum was developed based on homology modelling with the crystallized protein of F. sporotrichioides. Eight phenolic molecules, namely ferulic acid 1, apocynin 2, propyl gallate 3, eugenol 4, Me-dehydrozingerone 5, eugenol dimer 6, magnolol 7, and ellagic acid 8, were selected for their ability to inhibit trichothecene production and/or fungal vegetative growth in F. culmorum. The chemical structures of phenols were constructed and partially optimised based on Molecular Mechanics (MM) studies and energy minimisation by Density Functional Theory (DFT). Docking analysis of the phenolic molecules was run on the 3D model of F. culmorum TRI5. Experimental biological activity, molecular descriptors and interacting-structures obtained from computational analysis were compared. Besides the catalytic domain, three privileged sites in the interaction with the inhibitory molecules were identified on the protein surface. The TRI5-ligand interactions highlighted in this study represent a powerful tool to the identification of new Fusarium-targeted molecules with potential as trichothecene inhibitors.


Introduction
Trichothecene mycotoxins are produced by filamentous fungi belonging to the Fusarium genus and are among the major causes of crop loss in cereals [1,2]. Trichothecenes are heat resistant and very stable, and may cause severe toxicosis in mammals when contaminated grain or their derivatives are ingested. Besides affecting intestinal, immune endocrine and
The experiments were carried out as described by Pani et al. (2014) [21], with some modification as follows. Each phenolic compound (0.5 mM) was added along with β-CD (3 mM) to Vogel's medium [25]. After a 60 min sonication step, the amended substrates were poured into Petri plates (60 mm diameter, 8 mL per plate, 5 replicate plates for each treatment) and inseminated with a spore suspension of strain FcUK99 (final concentration 10 4 conidia mL -1 ). The evolution of pH in media amended with the different compounds was recorded at 0, 1, 2, 4, 8 and 14 days after inoculation (S1 Table). After 14 days incubation (at 25°C and in the dark) the culture broth and the mycelium were separated by vacuum filtration. After washing with sterile water, mycelia were harvested, dried at 80°C for 24 h and weighted. Culture broth was vortexed and an aliquot of 5 mL was collected from each replicate plate for mycotoxin extraction with ethyl acetate (3 mL), 1 min vortex, and 10 min centrifugation. Qualitative and quantitative mycotoxin detection was carried out by liquid chromatography interfaced to mass spectrometry (HP 1100, Agilent Technologies, Palo Alto, USA).

Statistical analysis
Data on total trichothecene production (ng mL -1 ) obtained from two separate experiments were expressed as percent value of the untreated control. Data were analysed by one-way analysis of variance (ANOVA) with Dunnett's test, using Minitab Express for Mac version 1.3.0.
Substitution of 33 amino acids from the B chain of 1JFG was carried out with Swiss-Model, an automated comparative protein model server interconnected with Swiss-PdbViewer 4.0.4 program [28][29][30]. The crystallographic molecules of water and glycerol ions present in the starting TRI5 protein of F. sporotrichioides were stripped. Hydrogen atoms were added using the ADT module of MGLTools 1.5.7rc1 [31]. Molecular docking: in silico studies on the binding of selected phenols to TRI5 using the Monte Carlo/energy minimisation [33] (Ei-E min <5 Kcal/mole, energy difference between the generated conformation and the current minimum). The atomic charges were assigned using the Gasteiger-Marsili method [34]. Representative minimum energy conformations of each compound were optimised using the ab initio quantum chemistry program Gaussian 09W by Density Functional Theory (DFT) with method B3LYP/6-311G basis set [35]. Visual analysis was carried out with GaussView version 5.0 [36]. Binding of the compounds was analysed using MGLTools 1.5.7rc1 [31] and AutoDock 4.2 docking programs [37,38]. The Gasteiger charges [34] for the ligands and proteins were used. The structures were docked using the Lamarckian genetic algorithm (LGA) defined through a centred grid (coordinates: X) -21.0, Y) 111.0, Z) 8.0, with 120, 120, 120 grid points in X, Y, Z dimensions, respectively. All ligands were docked with all bonds free to rotate. The Lamarckian genetic algorithm (LGA) of up to 100 runs was set to the population size of 150 individuals, maximum number of generations and energy evaluations of 27,000 and 15,000,000, respectively. From the estimated where ΔG is the docking energy, R (gas constant) is 1.98719 cal K −1 mol −1 and T (Temperature) is 298.15 K. Graphical representation of the hypothetical positions derived from the docking calculation was obtained using the software Chimera [39].

Fungicide and inhibitory activity of phenols against Fusarium culmorum FcUK99
Six out of the 8 compounds (ferulic acid 1, apocynin 2, propyl gallate 3, eugenol 4, Me-dehydrozingerone 5, and eugenol dimer 6) demonstrated their trichothecene inhibitory activity when tested at 0.5 mM, providing a reduction in the total mycotoxin production ranging from 31% to 98%. One compound (magnolol 7) had a fungicidal action, hence complete inhibition of mycelium growth and no trichothecene production were observed. On the contrary, ellagic acid 8 induced a significantly higher (159%) trichothecene biosynthesis compared to the untreated control ( Table 1).
Monitoring of medium pH amended with compounds 1-8 showed a progressive decreasing that, after 4 days, reached a plateau around 2.2 in the presence of all compounds except for magnolol 7 that stabilised the pH at 3.38 between the 4 th and the 8 th day (S1 Table). After 14 days the medium pH was comprised between 2.41 and 2.94 for all the tested compounds.
Computational studies: homology modelling of TRI5 As the X-ray crystal structure of F. culmorum TRI5 was not available, a 3D structure at atomic level by homology modelling was generated on the TRI5 crystal structure of F. sporotrichioides [18] and the starting protein sequence for F. culmorum (amino acid sequence code Q8NIG9). In the RCSB protein data bank (http://www.pdb.org) the trichodiene synthase of F. sporotrichioides is represented in a wide array of crystals, including wild type and mutated structures. The 3D structure homology model of F. culmorum was prepared starting from the chain B of the F. sporotrichioides crystal (1JFG), which contains the three magnesium atoms and the pyrophosphate (PPi), both essential for the enzyme activity. The model was refined by substitution of 33 amino acids from the B chain of 1JFG and by comparing the obtained result and the existent homology model sequence of F. culmorum TRI5 ( Table 2).
Among all the possible rotamers, we focused on the mutation achieving the best score. In the crystal structure of 1JFG, residues from Arg355 to Glu374 are not resolved as in other differently resolved crystals of F. sporotrichioides. For this reason, also in the reconstruction of F. culmorum TRI5 we decided not to consider these amino acids in order to set two comparable systems.

Docking of phenols and hydroxylated biphenyls to TRI5
All docking tests were performed considering a very large grid that included nearly the complete TRI5 enzyme of F. sporotrichioides and F. culmorum. The centre of the grid was located close to the catalytic site containing the three Mg 2+ ions. As a result, a large number of interactions in ligands with different docking sites were evaluated. The default grid spacing (0.375 Å) was adopted, treating the docking active site as a rigid system and the ligands as flexible, i.e., all non-ring torsions were considered active and free to rotate. In addition to the active site, five further sites showing significant ligand affinities were identified on the TRI5 protein surface. These five sites, overlapping in the F. sporotrichioides and F. culmorum enzymes, are numbered from 1 to 5 (Fig 2).
A greater scattering was observed in the different sites of the F. culmorum TRI5, although the orders of magnitude of Ki and ΔG are very similar to the F. sporotrichioides TRI5 protein.
Among sequence differences (33 amino acids) occurring between the two enzymes, only four amino acids are located in the privileged binding site 2 (Lys103Ser, Arg303Ala, Ser308His, Tyr329Phe), whereas the others are in its vicinity or far away ( Table 2). The difference between the four amino acids involved in the binding site has an effect on lipophilicity, slightly increasing in F. culmorum TRI5. These modifications may lead to a variation in the space available to the ligand, therefore influencing hydrophobicity, charge and polarity of the environment.
The scoring results of docking calculations deduced for ligands, as well as the major ligand receptor interactions, are listed in Table 3.
Docking studies indicate that sites 1, 2, and 4 of F. culmorum Q8NIG9 are privileged sites for ligands inhibiting trichothecene biosynthesis in vitro. Ligands with a long aliphatic chain and dimeric molecules (hydroxylated biphenyls) interact simultaneously with sites 1 and 2 (Fig 3).
In docking calculations, PPi was always considered as an additional fixed residue. The reliability of the docking approach was further verified by extracting the PPi from the catalytic site and by considering it as a normal ligand. After repositioning of PPi into the protein, the new PPi location was the same as in the original X-ray structure of TRI5, with only minimal conformational changes (RMSD 1.27 Å), hence confirming the reliability of the system (Fig 4).
Ferulic acid 1, the sole charged phenol among the eight ligands tested in in vitro, interacts with the catalytic site in the proximity of the Mg 2+ ions with significant docking score (Ki 1.99 μM) Table 1. In vitro effect of tested compounds on total trichothecene (DON) production by Fusarium culmorum FcUK99.

Compound
Dry fungal biomass DON (relative yield ± SD) a (relative yield ± SD) b and with amino acids of three sites located in the surface of TRI5, identified as site 3, 4 and 5 ( Fig  3a; Table 3). The estimated free energy of binding (E.F.E.B.) at sites 3, 4 and 5 was calculated over a range between -5.19 and -5.58 Kcal/mol (Table 3). Generally, monomers interact with sites 1 and 2, favouring site 1. Apocynin 2, which inhibited trichothecene biosynthesis by approximately 30% in in vitro assays, displayed 71% of docking runs in site 1 (Table 3), with an energy of -5.35 Kcal/mole and Ki of 119.64 μM. In Fig  5, the coloured areas represent the atoms of the different amino acids interacting with apocynin 2. In virtue of its structural and chemical functional features, the apocynin molecule is fully positioned within the cavity formed by these amino acids.
Docking experiments with propyl gallate 3 (trichothecene production reduced by 98%) showed a similar behaviour. Forty per cent of docking runs occurred at the lowest calculated energy of -5.86 Kcal/mole and Ki 50.71 μM (Table 3). Furthermore, propyl gallate 3 fits in 13% of the docking runs (-5.28 Kcal/mol) to site 2 with a Ki 133.84 μM (Table 3). Eugenol 4 (able to reduce trichothecene production by approximately 78%), binds in 22% of the cases to site 1 of the TRI5 protein (-4.98 Kcal/mole). Despite the low inhibition constant Ki (224.72 μM) calculated for site 1, eugenol 4 binds also (50% of docking run sum) to site 2, with two different conformations. Eugenol 4 shows affinity for this site with energies of -5.12 Kcal/ mole and -5.07 Kcal/mole, according to its conformations. Eugenol 4 and apocynin 2 interact also with the catalytic domain with negligible docking run, therefore, these interactions were ruled out.   Me-dehydrozingerone 5, a molecule that inhibited 92% of the trichothecene production, interacts with site 2 with two distinct conformations of similar energy with a docking run sum of 54%. Moreover, Me-dehydrozingerone 5 also interacts with site 4 with 14% of docking run and significant Ki (51.63 μM) and estimated free energy of binding (-5.85 Kcal/mol).
Eugenol dimer 6, magnolol 7 and ellagic acid 8 are dimeric phenols belonging to the family of hydroxylated biphenyls. These dimeric molecules are structurally different from the other two groups as they display a more hindered structure due to the presence of two aromatic rings linked by a single C-C bond. Two lactone groups fused in the C-C single bond make the structure of ellagic acid 8 quite planar, whereas eugenol dimer 6 and magnolol 7 may freely rotate along with the C-C single bond. These latter two ligands enter in site 1 with one of their two aromatic rings, positioning the second one perpendicularly to the first one that then interacts with the amino acids located in site 2, in proximity to site 1. This conformation allows magnolol 7 and eugenol dimer 6 to interact simultaneously with the amino acids of site 1 and 2 by each aromatic ring. Magnolol 7 interacts with sites 1 and 2 with two conformations that are quite equal in energy (29% and 11% of docking runs, respectively), and with site 4 in 14% of the docking runs. The energy ranges between -6.91 and -5.91 Kcal/mole with significant Ki up to 8.56 μM ( Table 3). The eugenol dimer 6 behaves similarly to magnolol 7, binding at the same time to sites 1 and 2 with a probability of 52% as sum of the three conformations similar in energy, while interactions with site 4 were assessed to 14% (Table 3).
On the contrary, ellagic acid 8 (the sole dimeric compound inducing a significant increase in the production of trichothecene by F. culmorum FcUK99 when added to the substrate at the tested concentration; Table 1), interacts separately and with significant Ki with site 1 (36% of docking runs) and site 2 (27% of docking runs; Table 3). The lowest energy (-7.00 Kcal/mole) and the best Ki (7.44 μM), respectively, were predicted for ellagic acid 8 when interacting with site 4 ( Table 3). The flat conformation of ellagic acid 8 forces the molecule to assume an intercalating position in the groove. Discussion A wide array of natural and natural-like phenols and dimers belonging to cinnamic acids, acetophenones, benzaldehydes, benzoic acids, phenylpropanoids, and hydroxylated biphenyls has already proven able to inhibit vegetative growth and/or trichothecene production when tested on a 3-acetyl-4-deoxynivalenol producing strain of F. culmorum [21]. In our previous study, no linear correlation was observed between antioxidant properties of the tested compounds and their inhibitory effect on fungal growth and mycotoxin production. Nonetheless, a guaiacyl unit in the structure was provisionally hypothesized to play a key role in trichothecene inhibition.
In the present investigation, we aimed to provide further insight into the understanding of structure-activity relationship of trichothecene inhibitors and the trichodiene synthase protein TRI5. Hence, we selected eight compounds based on their structural properties and on their inhibitory activity towards the F. culmorum wild-type strain FcUK99, already used in a number of genomics and biological studies [22,40,41].
Preliminary docking experiments with selected phenols onto the F. culmorum TRI5 protein model predicted that these compounds occupy five binding pockets that are different from the catalytic site. Sites 1, 2, and 4 are privileged sites of no-charged ligands where non-covalent interactions would take place, whereas ferulic acid 1 activates interactions with the catalytic domain with significant Ki, preferentially with magnesium atoms and the PPi, although noncovalent interactions with amino acids of sites 3, 4 and 5 were also estimated with significant docking score (Table 3).
In the tested experimental conditions, all compounds, among which ferulic acid 1 presents the lowest pKa (4.48), are likely to be in the protonated form, as confirmed by the low pH value detected in the medium since the first days of fungal growth.
Magnolol 7 and eugenol dimer 6 bind to sites 1 and 2 by partially or totally overlapping, in virtue of their conformational flexible structure that provides the best interaction with the set of surrounding amino acids. The best docking pose for biphenyls 7 and 8 included one aromatic moiety of biphenyl engaging amino acids that are present in site 1, while the other aromatic moiety interacts with amino acids in site 2. In the case of ellagic acid 8, though, significant docking runs were calculated (accounting for 36% and 27% for site 1 and 2, respectively): due to its almost flat structure, ellagic acid 8 does not overlap with sites 1 and 2, exerting a less intimate contact when compared with conformationally flexible magnolol 7 and eugenol dimer 6. These two molecules also bind to site 5 with a significant docking (E.F.E.B.) ( Table 3; Fig 3d and 3e). It is known that hydroxylated biphenyls are privileged molecules for protein binding in comparison to other aromatic compounds [44]. Such behaviour derives from the flexible structure of the biphenyl unit that can be accommodated, with high level of specificity, within a wide variety of protein pockets. The presence of a C 2 symmetry axis in the hydroxylated biphenyls 6-8 makes the two aromatic rings indistinguishable from the chemical and biological point of view [24,45,46]. This structural feature facilitates the docking, reducing the number of possible interactions between the two parts of the molecule because they are identical.
The energetically "best" poses for apocynin 2 involve only site 1 with significant docking run values, whereas Me-dehydrozingerone 5 engages sites 4 and 2, the latter with almost equal docking runs in two molecule conformations. We hypothesize that the far less efficient trichothecene inhibition observed in the case of apocynin 2 may be related to its limited number of non-covalent interactions with TRI5, whereas a large number of amino acids are involved in the interaction between TRI5 and Me-dehydrozingerone 5, which activates interactions with the hydrophobic amino acids present in site 2.
Another factor evidenced in the docking studies is the capacity of ligands with high trichothecene inhibitory properties to activate hydrogen bond (H-bond) interactions with amino acids of TRI5 (Table 4). Biphenyls 6-8 establish multiple H-bonds and hydrophobic interactions with the amino acids present in sites 1, 2 and 4 (Table 4; Fig 3d, 3e and 3g). Cross-bridge interactions with the same amino acids and the OH-phenol group of ferulic acid 1, propyl gallate 3, eugenol dimer 6, magnolol 7 and ellagic acid 8 were predicted (Table 4).
Cross-bridge H-bond interactions of biphenyls 6-8 and propyl gallate 3 involve mostly apolar and hydrophobic amino acids (Table 4), whereas polar and less hydrophobic amino acids activate interactions with ferulic acid 1. This type of interaction was not observed for eugenol 4, apocynin 2 and Me-dehydrozingerone 5, although the latter ligand interacts with sites 2 and 4 with a large number of hydrophobic amino acids.
Overall, the docking results confirmed what we had observed in a previous in vitro assay with a far larger collection of phenols [21]: lipophilicity and H-bonding capacity were postulated as key factors in the selection of a "good" trichothecene inhibitor. Although no π-π-stacking interactions between protein-ligands were predicted, these would not be ruled out.
Furthermore, the molecular size (i.e., the capacity of the ligand to interact with different amino acids at the same time) should be taken into account [47,48]. From this perspective, the scarce inhibitory activity of apocynin 2 might reflect its relatively small size, which limits extensive protein-ligand interactions.
As mentioned before, sites 1, 2 and 4 were identified as privileged sites of no-charged phenols. Quite interestingly, five amino acids (Cys301, Asp302, Ala303, Phe329, and Ala337) are represented in the range of 80-100% in sites 1 and 2, where most of the ligand-protein docking occurs. These amino acids have neutral and, except for Asp302, non-polar side chains, assuming a lipophilic character of the potentially binding ligand. A prevalence of hydrophobic amino acids also resulted in site 4.
To sum up, propyl gallate 3, Me-dehydrozingerone 5, eugenol dimer 6 and magnolol 7 proved the best trichothecene inhibitors, as predicted by docking studies, where these compounds bind to the same sites of TRI5, and further confirmed by the in vitro bioassay. The reduced inhibitory activity observed in vitro for apocynin 2, eugenol 4 and ellagic acid 8 might be related to restricted ligand-protein interaction.    As mentioned, magnolol 7 behaved as a fungicide in vitro when tested at 0.5 mM concentration. However, in the modelling study, ligand concentrations were not taken into account, and magnolol 7 may therefore switch from fungicide to trichothecene inhibitor at very low concentration, as shown previously [21]. The fungicidal activity of magnolol 7 should be further investigated since its properties are likely to affect other mechanisms of action related to both primary and secondary metabolism [49].
Ferulic acid 1 is a cinnamic acid and was included in this study due to its well-known inhibitory activity towards trichothecene biosynthesis [50,51]. This compound demonstrated a high affinity with the catalytic domain of the TRI5 protein by strong metal-and ion-interactions (Fig 6), whereas the best docking scores were estimated with amino acids of sites 3, 4 and 5 where ferulic acid 1 likely activates non-covalent interactions.
Aiming to further investigate the interaction of F. culmorum TRI5 with phenolic acids, we compared the docking score of ferulic acid 1 with other cinnamic acids and benzoic acids whose biological activity has been previously tested in vitro [21]. 3-Hydroxycinnamic acid 9, 4-hydroxycinnamic acid 10, 2,5-dimethoxycinnamic acid 11 inhibited trichothecene production by F. culmorum strain MCf 21 (syn. INRA 117) in the range of 1.5 and 1.0 mM, whereas no trichothecene nor mycelium growth inhibition were observed for both 3-methoxybenzoic acid 12 and 3,4-dimethoxybenzoic acid 13 at 0.5 mM [21]. All phenolic acids interact significantly with the catalytic domain with higher docking scores compared to ferulic acid 1 (S2 Table). Contrary to benzoic acids 12 and 13, all cinnamic acids interact with sites 3, 4 and 5 as estimated in ferulic acid 1 evidencing an important role of the α,β-unsaturated chain in activating interactions with sites located on the protein surface. Substituents in the aromatic ring of cinnamic acids are likely to influence H-bond interactions, that increase when the hydroxyl group is in para position to the aliphatic chain, hence favouring electron delocalization as in compounds 1 and 10 (S3 Table). Nevertheless, ferulic acid 1 accounts for the highest H-bonds involving amino acids located on protein surface. Site 3 and, to a lesser extent, site 5 seem to be privileged sites for cinnamic acids (charged-phenols), mostly binding to polar and partially hydrophobic amino acids.

Conclusions
In the present study, a docking method was applied to explore and validate key interactions between natural and natural-like phenols with the F. culmorum TRI5 model. Apart from the catalytic domain, five binding sites were identified on the surface of TRI5 and among them sites 1, 2 and 4 are privileged binding sites for no-charged phenols. For charged phenols, and particularly ferulic acid 1 which was investigated as a prototype of cinnamic acids, sites 3 and 5 are privileged for binding, despite interaction with the catalytic domain should be also considered for more polar cinnamic acids. Docking data confirmed results obtained from in vitro assays with a collection of phenols, whose lipophilicity and H-bonding capacity were postulated as key factors in the selection of a good trichothecene inhibitor. An important role might be played by the molecular size, which governs the ability of the ligand to interact with different amino acids at the same time.
Hence, our model assumes that a phenolic molecule bearing substituents with high ability to activate protein-ligand interactions at the same time with a large number of amino acids would be an ideal candidate as trichothecene inhibitor.
It has to be born in mind that the modelling approach provides only an estimation of protein-ligand interaction, as it does not take into consideration a wide range of interfering factors (e.g., concentration, interactions with other proteins, interactions with water molecules, delivery system, membrane permeability, biocompatibility and bioavailability). Nonetheless, the TRI5-ligand interactions highlighted in this study, along with the availability of a crystal structure of the Fusarium TRI5, shall provide an additional tool to discover new molecules with potential as fungicides or as trichothecene inhibitors, and will guide the synthesis of novel Fusarium-targeted compounds by shortening the time of research and by reducing cost. The long-term goal of our project is to develop natural or natural-like phenolic compounds as effective, environmentally-friendly alternatives to synthetic fungicides. Based on in vitro and in silico characterisation, field testing is now being carried out with selected molecules in order to evaluate their efficacy in reducing FHB symptoms in durum wheat as well as trichothecene contamination in harvested grain.