The three-dimensional (3D) structures of tetraspanin CD81, both the full-length protein (PDB: 5TCX) and the truncated protein corresponding to the LEL/EC2 domain (PDB: 5M3T), were retrieved from the PDB (https://www.rcsb.org/). They refer both to high-resolution structures (2.96 Å and 2.02 Å resolution) obtained by X-ray diffraction [35, 36]. The GOLD 5.3 software (Cambridge Crystallographic Data Centre, Cambridge, UK) was used to perform the docking analyses. The Biovia 2020 platform (Dassault Systèmes BIOVIA Discovery Studio Visualizer 2020; Dassault Systèmes, 2020) was used for molecular graphics and analyses. The web server Computed Atlas of Surface Topography of proteins (CASTp) 3.0 was used to identify potential ligand-binding sites on proteins. The molecular modeling software Chimera 1.15 was used for visualization [37].

Prior to docking, the protein and the ligand are independently prepared. The eventually missing atoms in the protein according to the PDB file are added and the protein structure is minimized using the Spectroscopic Potential Algorithm for Simulating Biomolecular conformational Adaptability (SPASIBA) force field (taking into account all parameters and atomic charges). A Monte Carlo (MC) conformational searching is performed for the free ligand. The final conformer is the one used in the induced fit docking procedure. The multi-step docking procedure includes an MC conformational analysis of the ligand, performed using the BOSS software. We rely on a previously described MC conformational search procedure with BOSS [38] to optimize the ligand structure. The conformational analysis serves to define the starting geometries for the ligand. With this procedure, all minimum-energy conformers for the ligand are characterized and then the energy minimization defines a unique conformer for the free ligand. The MC simulations were performed within the constant number of particles, constant-pressure, constant-temperature ensemble (NPT), as defined in BOSS. Then, an evaluation of the free energy of hydration (ΔG) is performed for the defined structure of the ligand, using the molecular mechanics (MM)/generalized Born surface area (GBSA) procedure [39]. The calculation of ΔG values was performed with the MC search, within BOSS using the xMCGB script, as previously described [39, 40]. According to this process, the best ligand structure is defined and then used for the docking analysis. The next step corresponds to a definition of the sites of interaction between the tetraspanin and the ligand, using the CASTp 3.0 program. A sphere of 10 Å radius is superposed to the CASTp region in order to define the active site. Within the sphere, the amino-acids side chains are considered fully flexible together with the ligand. This allows a mutual conformational adaptability of both the ligand and the protein. Once the docking is completed, the ligand in its environment is the subject of an MM/GBSA calculation to evaluate the hydration-free energy using the MC BOSS program.

The flexible amino acids are: (i) Thr161, Leu165, Ser168, Asn172, Ser179, Ile181, Asn184, Leu185, Asp189, and His191 for CD81 full-length (PDB: 5TCX), (ii) Lys148, His151, Glu152, Ser160, Leu170, Leu174, Asn184, Leu185, Lys187, and Asp189 for the LEL fragment of CD81 (PDB: 5M3T). The docking grid is centered in the volume defined by the central amino acid (C157 for CD81/LEL) and within the binding site, the side chains of these amino acids are fully flexible during docking. The docking procedure is performed using GOLD. Our typical process considers 100 energetically reasonable poses [defined with the Chem Piecewise Linear Potential (PLP) scoring function] which are screened to search for the optimized ligand binding mode. The PLP fitness scoring function of GOLD v5.3 is used to define and rank the binding poses [41]. In general, 6 poses are selected per analysis. Then 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 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 MM force field parameters [42]. SPASIBA (integrated into CHARMM) empirical energies of interaction are calculated as described [43–45]. It is an excellent system for reproducing crystal phase infrared data. We used similar operational procedures to construct molecular models for CD81 and CD9.

Two crystallographic structures of CD81 were considered and an overlapping model of the paired structure is given (Figure 2B). 5TCX is a full-length structure of CD81 (246 amino acids) with a molecule of cholesterol bound to the four TM helices [35]. In this case, the LEL extracellular portion of the tetraspanin adopts a compact (closed) configuration that restricts access to the central cavity of the loop, defined as the drug binding site (see below). 5M3T corresponds to the crystal structure of the LEL fragment of the truncated protein comprising 101 amino acids [36]. In this case, the LEL adopts a more extended (open) conformation, allowing wider access to the central cavity. The superimposition of the two structures is good, with a satisfactory overlap between the non-EC2 portions (Figure 2). The superimposition between the two PDB structures 5TCX and 5M3T is excellent with an average C-alpha atom root-mean-square deviation (RMSD) of 0.118 Å. But the subtle deviation is marked at the level of the loop positions 154–190 where the calculated RMSD value amounts to 2.85 Å, with a significant local opening of the loop. The central cavity centered around residue Cys157 is well-defined, with two surrounding walls, more or less open, as shown in Figure 2C.

We then used these two structures to perform a binding site analysis using the web server CASTp 3.0 well-adapted and convenient to predict the position of drug binding areas (Figure 3) [37]. The analysis with the full-length protein (5TCX) revealed a major interaction site in the TM part of the protein, evidently corresponding to the cholesterol-binding site (area in red in Figure 3A). Another much smaller site is detected in the LEL portion. It is a small site (purple area in Figure 3A) with a volume of 11.30 ų and a surface of 43.55 Ų. There is also a cavity toward the intracellular portion of the protein (areas in green/blue/orange in Figure 3A) which was not considered for drug binding because it has been shown that Hz-A and -B bind to the LEL domain of CD81 [22]. The situation is significantly different when considering the open conformation of the LEL domain (5M3T). In this case, the LEL-binding area is much larger, with a volume of 81.17 ų and a surface of 127.90 Ų. The volume of the central cavity suitable to accommodate a small molecule has expanded considerably. It is now 7-time bigger with the open cavity compared to the closed one (Figure 3B). It can be appreciated also by comparing the distance between two residues on each wall of the cavity, Leu165, and Leu185, as depicted in Figure 3C. The L165-L185 distance increases from 6.70 Å to 17.50 Å in the closed (5TCX) and open (5M3T) structures, respectively. Similarly, the distance between residues Ile181 and Val169 increased from 6.38 Å to 18.85 Å, in the closed and open structures, respectively. The local widening of the cavity is significant. The structural variation is very localized and confined to a portion of the LEL domain. This local variation is exploited for drug binding.

The two conformations of the LEL domain were considered for the binding study with compounds Hz-A and -B. With closed structure 5TCX, the docking was performed with the cholesterol molecule in place, leaving free access of the test molecule to the small cavity centered around position Cys157. The same drug binding site (C157) was used with the open structure 5M3T. In both cases, energies of interaction were calculated (Table 2). The calculated empirical energy of interaction (ΔE) and free energy of hydration (ΔG) are much more negative with the LEL structure compared to the full-length protein. The open conformation of the EC2 loop allows easier access of the natural product to the binding cavity compared to the closed conformation. The more compact structure of the loop in the full protein does not permit drug binding; the energies are weak (ΔE < 40 kcal/mol). In contrast, the LEL alone presents a wider groove to which the natural product can bind. The best compound is Hz-B with a ΔE value a little more negative than that calculated with Hz-A. Our analysis is essentially based on the calculated ΔE values; the predicted molecule’s solvation-free energy (ΔG) provides essentially complementary information to verify the conformity of the calculations and the lack of major difference between the two test compounds. ΔG represents the solute−solvent interactions and no large variations have been observed in the present cases.

A molecular model of Hz-B bound to the LEL structure is presented in Figure 4A. The compound fully enters the cavity in the center of the structure, approaching the central Cys157 residue (Figure 4B). The cavity presents a relatively large solvent-accessible surface (SAS) and the compound engages its elongated carboxypentenyl side chain deep into the cavity (Figure 4C). The ligand-protein complex is stabilized by a set of H-bonds, alkyl/π-alkyl interactions, and van der Waals contacts (Figure 5). There are two key H-bonds with residues Asp155 and Lys187 common to Hz-A and Hz-B. A third H-bond with Cys175 is specific to Hz-B and explains the higher affinity of this compound for the LEL structure compared to Hz-A. Altogether, we identified 14 possible contact points between Hz-B and the LEL protein. This is significant to maintain the product in place in the cavity. The three side chains attached to the 2-methylcyclobutane core of Hz-B contribute to the protein interaction, notably the 3-[(1S)-1-carboxy-1-hydroxyethyl] side chain which is specific to Hz-B, engaged into an H-bind with Cys175 (Figure 5).

There are 33 tetraspanins in humans, plus non-conventional forms generated by alternative splicing [46, 47]. The majority of them present an organization similar to that of CD81, with an intracellular portion, a TM-4 domain, and a LEL. Several tetraspanins are involved in cancers, notably CD9, CD37, CD63, CD81, CD82, and CD151 [27]. Among them, the tetraspanin CD9 is of prime interest because it plays a major role in cell motility and adhesion, and contributes to head and neck cancer and other advanced solid tumors [48–50]. Like CD81, CD9 is considered as a target for cancer therapy [51]. The 3D structures of both the full-length CD9 protein (6K4J) [52] and the LEL portion (6Z1V) [53] are available from the PDB (https://www.rcsb.org/). The protein organization of CD9 is very close to that of CD81 [54]. These two tetraspanins are often considered surface markers for extracellular vesicles [55, 56]. For these different reasons, we investigated the potential binding of Hz-A and -B to CD9, in comparison to CD81.

A superimposition of the full-length CD81 and CD9 proteins and the corresponding LEL structures is shown in Figure 6A. The pairs of molecular models superimpose very well. The two tetraspanins display the same configuration, and the superimposition of the LEL portions is almost perfect (Figure 6B). Nevertheless, subtle differences can be seen at the level of the binding cavity in the EC2 loop, as shown in Figure 6C. These models were used to perform the docking of Hz-A and Hz-B, and the calculated energies are given in Table 3. The empirical energies of interaction (ΔE) calculated with CD9 are significantly weaker (less negative) than those measured with CD81. In other words, the two natural products exhibit a significant preference for CD81 over CD9, despite the similarity of the two protein structures. The difference is very important when comparing the two LEL structures. The ΔE values amount to –61.40 kcal/mol and –43.30 kcal/mol for Hz-B bound to CD81 and CD9, respectively. The difference in energy (–18.1 kcal/mol) is major, and it is even more pronounced (–21.8 kcal/mol) in the case of Hz-A (Table 3). There is no doubt that the two compounds are better adapted for binding to CD81 compared to CD9. The LEL portions of CD9 and CD81 are very similar (Figure 6B) but there are local variations that can be exploited for drug binding. Apparently, CD81 offers a suitable binding cavity for Hz-A but not CD9. This tetraspanin selectivity aspect warrants further examination.

There are few small molecules known to bind to CD81. In the natural product class, the only two compounds for which binding to CD81 has been evidenced at the experimental level are Hz-A and -B, no other ones. Recent computational analyses have suggested the possible binding of the pentacyclic triterpene maslinic acid [57] and diverse flavonoids to the EC2 loop of CD81, such as puerarin, quercetin, and (–)-epicatechin [58]. This is entirely plausible, but these polyphenolic compounds tend to bind to many proteins in cells. We searched the natural products database for analogues of Hz, but we did not identify structurally close compounds. The only two products with a certain degree of similarity are the monoterpenes grandisol and the diastereomer fragranol (Figure 7A) which can be found in essential oils from the aerial parts of different plants, such as sagebrush (Artemisia spp.) [59] and the Mediterranean plant Achillea ligustica All. [60]. Their chemical synthesis has been described recently [61]. Fragranol is a rare monoterpene alcohol endowed with an anxiolytic action [62] but the compound has been little studied. Grandisol is a little more known, acting as a pheromone for curculionid beetles, such as Polygraphus punctifrons and Sternechus subsignatus (Coleoptera) [63, 64]. We have modeled the interaction of these two natural products with the LEL domain of CD81 but found only a weak interaction (ΔE = –40.35 kcal/mol and –35.50 kcal/mol for grandisol and fragranol, respectively; Figure 7B). The binding capacity of grandisol to CD81-LEL is modest but it relies on the same type of contacts as with Hz-B, notably the key H-bond with Lys187 and Cys175 (Figure 7C). The molecule is accessible chemically, with different total syntheses described [61, 65, 66]. Therefore, it is conceivable to design grandisol analogues with improved CD81-binding properties.

In recent years, new insights have been gained into the structure and function of tetraspanin proteins and CD81 in particular. This TM protein is a target of interest for the treatment of viral diseases and cancers. The identification of small molecules binding to CD81 remains little studied and the design of compounds targeting selectively the LEL domain has been largely untouched. There is a need for new compounds and scaffolds interacting with CD81-LEL. Our study supports the use of the terpenoid Hz-A as a possible template to elaborate new CD81 binders. The study underlines the importance of considering the dynamic conformation of the LEL binding domain. The binding of Hz-B to CD81 varies markedly according to the open/closed conformation of the protein loop, and the nature of the flexible loop. Hz-B exhibits a net preference for binding to CD81-LEL over CD9-LEL, suggesting that a tetraspanin selectivity can be achieved. A cyclobutane-containing fragment (grandisol) has been identified as a possible starting element for drug design. The identification of the fungal sesquiterpene Hz-B as a ligand for CD81-LEL offers perspective to the synthesis analogues. The drug-based modulation of tetraspanin biology opens novel avenues for the treatment of advanced cancers and viral diseases.