Two tridimensional structures of a drug-bound α/β-tubulin dimer were retrieved from the Protein Data Bank (PDB; www.rcsb.org) under the PDB codes 1TVK [34] and 5FNV [35]. The GOLD 5.3 software (Cambridge Crystallographic Data Centre, Cambridge, UK) was used to perform molecular docking analysis. Prior to the docking operations, the structure of each ligand was optimized using a classical Monte Carlo (MC) conformational searching procedure via the Biochemical and Organic Simulation System (BOSS) software [36]. For the ligands preparation, the 2D and 3D structures of the natural products (SDF files then converted into PDB or MOL2 files) were obtained from the PubChem compound database (https://pubchem.ncbi.nlm.nih.gov/) and from the original publications describing the compounds (Figure 1). Molecular graphics and analysis were performed using Discovery Studio Visualizer, Biovia 2020 (Dassault Systèmes BIOVIA Discovery Studio Visualizer 2020, San Diego, Dassault Systèmes).

The web server Computed Atlas of Surface Topography of proteins (CASTp) 3.0 was used to identify potential ligand-binding sites on the tubulin dimer (1TVK). CASTp is a convenient geometry-based approach to predict the position of ligand binding sites, with an estimated success rate of 74%, equivalent to that of other programs [37]. The method locates and measures pockets and voids on 3D protein structures based on the alpha shape and the pocket algorithm. It is a convenient method of reproducing the secondary structure distribution of key residues in proteins and identifying residues implicated in cavities or clefts [38, 39]. The molecular modeling software Chimera 1.15 was used for visualization [40].

The colchicine/pironetin binding area within the tubulin dimer structure (5FNV) was considered the potential binding site for the studied alkaloids. During the process, the side chains of the following amino acids within the binding site were rendered fully flexible: Phe135, Ser165, Phe169, Cys200, Phe202, Ser241, Leu242, Phe255, Cys316, Lys352. A docking grid centered in the volume defined by the central amino acid has been defined based on shape complementarity and geometry considerations. In general, up to 100 poses considered energetically reasonable are selected during the search for the correct binding mode of the ligand. The decision to select a trial pose is based on ranked poses, using the fitness scoring function [piecewise linear potential (PLP) score incorporated in GOLD 5.3] [41]. The same procedure was used to establish molecular models for all studied alkaloids.

In general, 6 poses are selected per analysis. The ranking leads to the evaluation of the empirical potential energy of the interaction (ΔE), defined using the expression ΔE(interaction) = E(complex) – [E(protein) + E(ligand)]. The SPASIBA (Spectroscopic Potential Algorithm for Simulating Biomolecular conformational Adaptability) spectroscopic force field is used to calculate the final energy. The required parameters are derived from vibrational wavenumbers obtained in the infrared and Raman spectra of a large series of compounds of diverse chemical nature (organic molecules, amino acids, saccharides, nucleic acids, and lipids). The last step corresponds to a validation using the SPASIBA force field, an essential step to determine the best protein-ligand structure. This force field has been specifically developed to provide refined empirical molecular mechanical (MM) force field parameters [42]. SPASIBA (integrated into CHARMM) empirical energies of interaction are calculated. It is an excellent system for reproducing crystal phase infrared data. SPASIBA has been specifically developed to provide refined empirical MM force field parameters, as described in other studies [43–45]. Using this specific force field for MC simulations achieves the same level of convergence as molecular dynamics (MD), with less computer time [46]. The BOSS program and the MM/generalized Born surface area (MM/GBSA) procedure were used to evaluate free energies of hydration (ΔG), in relation to aqueous solubility [47].

Several crystallographic structures of α/β-tubulin heterodimers are available from the PDB. In their study of securinine binding to tubulin, the authors reported a computational docking analysis of the compound bound to tubulin based on the crystallographic structure of the α/β-tubulin dimer in interaction with the drug epothilone A (PDB: 1TVK). They proposed binding of securinine at the interface of the α/β-tubulin dimer, with the alkaloid engaging a key H-bond with residue Asn228 of the α-tubulin unit [26]. We performed a similar docking analysis with the same structure (1TVK) positioning the alkaloid at the described site to allow the interaction between the compound and Asn228. In this case, the calculated empirical energy of interaction (ΔE) and free energy of hydration (ΔG) were –40.40 kcal/mol and –13.80 kcal/mol, respectively. This model is satisfactory but we consider that it presents a limitation, that is the necessity to remove the GDP molecule present in the cavity and hindering access to Asn228. The crystallographic structure includes a GDP molecule deeply inserted into the cavity and stacked over the tyrosine residue Tyr172 of α-tubulin, as represented in Figure 2. This GDP molecule precludes access to the site. Securinine can be positioned at this site, only if the GDP is removed. For this reason, we considered this binding option as non-optimal and this prompted us to search for an alternative model.

The PDB structure 5FNV provides another model for tubulin, corresponding also to a noncovalent heterodimer composed of one α- and one β-tubulin monomer. The original crystallographic structure refers to the interaction of α/β-tubulin with the α,β-unsaturated lactone derivative pironetin [35]. It is a convenient model to analyze drug binding to tubulin, notably for compounds which bind to the interface between the two facing monomers. There is a large cavity between the two monomers, accessible to small molecules. An analysis of the model with the web server CASTp 3.0 provides a clear definition of the potential binding areas (Figure 3). The main binding area is located between the two monomers, delimiting a large zone of 1,595 ų in which bulky molecules can bind. There are also smaller cavities on each subunit, seven in total but they are too small to accommodate dimeric compounds. Securinine can easily enter into the interfacial cavity to engage in contact with the two-facing tubulin units. In this case, the calculated empirical energy of interaction (ΔE) and free energy of hydration (ΔG) were –44.60 kcal/mol and –17.10 kcal/mol, respectively. Therefore, this model offers a better representation of the capacity of securinine to bind to the α/β-tubulin compared to the initial (1TVK) model. For this reason, we selected this second model (5FNV) for the subsequent docking analysis with all studied compounds. There are more than 20 high-resolution tubulin crystal structures available from the PDB, corresponding to the free protein or the protein bound with different types of small molecules, mostly natural products [48]. We considered the pironetin bonding site as a suitable cavity for this type of molecules, as shown recently with dihydropyrone derivatives [24]. Other drug-binding sites on tubulin may be considered as well, but we favored the pironetin site for its capacity to accommodate large, extended molecules.

Next, we performed a comparative analysis of the interaction of a dimeric alkaloid with the two tubulin models, 1TVK and 5FNV. The dimeric ligand flueggine B was selected as a test ligand. This dimeric indolizidine alkaloid has revealed potent antiproliferative activities against cultured cancer cells [48]. Molecular models of flueggine B bound to tubulin dimers were elaborated using both structures. The calculated ΔE values were –54.40 kcal/mol and –63.40 kcal/mol with structures 1TVK and 5FNV, respectively. Much more stable complexes were obtained with 5FNV compared to 1TVK. This comparative analysis reinforced our choice of structure 5FNV for the analysis of the flueggenines.

The six fluevirines were docked into the interfacial cavity, as done with securinine. Fluevirine B is a direct analogue of securinine with an additional ethoxyl substituent at position 14. Fluevirines C and D both bear the same 14-ethoxyl group in the norsecurinine series. They differ by the orientation of the hydrogen atom at position 2; fluevirine C is the β anomer whereas fluevirine D corresponds to the α anomer. The tubulin binding capacity of these three compounds is relatively modest, as observed also with the related N-oxide product fluevirine F. This latter product is not well adapted to bind to the target, with a very low hydration free energy (ΔG). The situation is significantly better with the unrelated product fluevirine E which is a totally distinct bis-indolyl compound. Its interaction with tubulin is stabilized by 22 molecular interactions with the protein dimer. The two essential contacts are an H-bond between Asn258 and an indolic NH group, and an aromatic stacking interaction between the other indole group and residue Phe255 (Figure 4). It provided a first observation that dimeric molecules were better adapted for tubulin binding than monomeric compounds. The other dimer in the series is fluevirine A with a norsecurinine-type tricyclic unit connected to a bicyclic unit equipped with an acid side chain (Figure 1). This atypical compound turned out to be a potent tubulin binder, with the implication of its two units in the binding process. There are 25 interactions between the alkaloid and the protein, with in particular two H-bonds with Lys352 (Figure 4). The free energy of interaction (ΔE) calculated with this compound is about 40% superior (more negative) than that measured with the corresponding monomer fluevirine C (Table 1). This observation prompted us to investigate other dimeric alkaloids from F. virosa.

The series of flueggenines includes 9 compounds (A–I) and each of them was docked onto the same drug binding site at the interface of the tubulin dimer. The calculated energies are collated in Table 2. Flueggenine A is a dimeric archetype consisting of two norsecurinine units. Its interaction with the tubulin dimer is reinforced compared to the norsecurinine precursor, but it is not as strong as fluevirine A. In this series, the best compound is the atypical polycyclic compound flueggenine B which seems to be well adapted to fit into the pironetin site of tubulin. The tubulin-flueggenine B complex is stabilized by a network of about 20 molecular contacts, including two essential H-bonds with residue Gln256 and Lys352 (Figure 5A, B). The hybrid compound flueggenine E which combines a norsecurinine-type unit with an aminoethyl-indole side chain provided also a good tubulin binder. In this case, the number of ligand-protein contacts was even more important (27 contacts identified). A π-π stacking interaction between the ligand indole ring and residue Phe255 can be evidenced (Figure 5C, D), as observed previously with the bis-indole compound fluevirine E (Figure 4).

Another interesting compound in the series is flueggenine I which is also a C-C linked norsecurinine-type dimer. Its binding to the tubulin site chiefly implicates the two lactone units of the ligand and H-bonding interaction with residues Lys352 and Leu 242. The natural product showed the best comprise in terms of binding free energy (ΔE) and free energy of hydration (ΔG). Its shape is well adapted to fit into the hydrophobic pocket and here again, there are over 20 molecular contacts between the ligand and the protein (Figure 6). The molecule is relatively compact for a dimer, with a linear shape suitable for inserting between two long α-helices of the protein. Upon binding to the tubulin α-subunit, it is positioned just in front of the GDP binding site on the opposite β-tubulin subunit. Signal transmission between the drug-binding site and the nucleotide site of tubulin can occur [49]. Its molecular interaction with the protein dimer is profound and maximal, as represented in Figure 7. The other flueggenines provided fewer stable complexes than flueggenine I, our favorite ligand in the series (Table 2). The binding configuration observed with all compounds, including the reference compounds securinine and norsecurinine, is shown in Figure S1.