Synthetic Glycolipids as Molecular Vaccine Adjuvants: Mechanism of Action in Human Cells and In Vivo Activity

Modern adjuvants for vaccine formulations are immunostimulating agents whose action is based on the activation of pattern recognition receptors (PRRs) by well-defined ligands to boost innate and adaptive immune responses. Monophosphoryl lipid A (MPLA), a detoxified analogue of lipid A, is a clinically approved adjuvant that stimulates toll-like receptor 4 (TLR4). The synthesis of MPLA poses manufacturing and quality assessment challenges. Bridging this gap, we report here the development and preclinical testing of chemically simplified TLR4 agonists that could sustainably be produced in high purity and on a large scale. Underpinned by computational and biological experiments, we show that synthetic monosaccharide-based molecules (FP compounds) bind to the TLR4/MD-2 dimer with submicromolar affinities stabilizing the active receptor conformation. This results in the activation of MyD88- and TRIF-dependent TLR4 signaling and the NLRP3 inflammasome. FP compounds lack in vivo toxicity and exhibit adjuvant activity by stimulating antibody responses with a potency comparable to MPLA.


Computational Methods
Ligands 3D structure: construction, optimization and parameters derivation. The 3D structures of FP11, FP18 and FP111 were built with PyMOL molecular graphics and modeling package (www.pymol.org), using as a template the 6YA monosaccharide (GLYCAM database, www.glycam.org). The structures of the ligands were first refined at the AM1 level of theory, and then optimized at the Hartree-Fock level (HF/6-311G**) with Gaussian09. 1 The parameters needed for MD simulations were obtained using the standard Antechamber procedure in Amber14. 2 The partial charges were derived from the HF calculations, and formatted for AmberTools15 and Amber14 with Antechamber, assigning the general AMBER force field (GAFF) atom types, and GLYCAM force field atom types for the saccharide atoms.
Macromolecule preparation. 3D coordinates from the X-ray structure of the human (TLR4/MD-2/E. coli LPS)2 ectodomain (PDB ID 3FXI) were retrieved from the Protein Data Bank (www.rcsb.org), and the chains A (TLR4) and C (MD-2) were extracted and considered as TLR4/MD-2 heterodimer in agonist conformation. Ligands and ions were removed. Hydrogen atoms were added to the X-ray structure using the preprocessing tool of the Protein Preparation Wizard of the Maestro package (www.schrodinger.com). The protein structure went through a restrained minimization under the OPLS3 force field with a convergence parameter to RMSD for heavy atoms kept default at 0.3 Å.
Docking calculations. AutoDockTools 1.5.6 program 3 was used to assign the Gasteiger-Marsili empirical atomic partial charges to the atoms of both the ligands and the receptor. Nonpolar hydrogens were merged for the ligands. The structure of the receptor and the ligands were set rigid and flexible, respectively.
Preliminary docked poses were obtained with AutoDock Vina 1.1.2. 4 The box spacing was set to the default value of 1 Å; the size of the box was set to 33.00, 40.50, and 35.25 Å in the x, y, z-axes respectively, with the box center located equidistant to the mass center of residues Arg90 (MD-2), Lys122 (MD-2) and Arg264 (TLR4). The best-predicted docked poses from AutoDock Vina were redocked with AutoDock 4.2 3 using the Lamarckian evolutionary algorithm; all parameters were kept default except that the number of genetic algorithm runs was set to 100 to enhance the sampling. Docking box spacing was set to 0.375 Å and box size was set to the same dimensions as for AutoDock Vina.
MD simulations. Selected docked complexes were submitted to all-atom MD simulations for 50 ns in the Amber14 suite using the force field ff14SB to describe the protein system. The simulation box was designed such that the edges were distant at least 10 Å of any atom. The system was solvated with the TIP3P water molecules model; Na + ions were added to counterbalance the charges of the protein-ligand systems. All the simulations were performed with the same equilibration and production protocol. First, the system was S5 submitted to 1000 steps of the steepest descent algorithm followed by 7000 steps of the conjugate gradient algorithm. A 100 kcal·mol -1 ·Å -2 harmonic potential constraint was applied to both the proteins and the ligand.
In the subsequent steps, the harmonic potential was progressively lowered (respectively to 10, 5, and 2.5 kcal·mol -1 ·Å -2 ) for 600 steps of the conjugate gradient algorithm each time, and then the whole system was minimized uniformly. Next, the system was heated from 0 to 100 K using the Langevin thermostat in the canonical ensemble (NVT) while applying a 20 kcal·mol -1 ·Å -2 harmonic potential restraint on the protein and the ligand. Finally, the system was heated from 100 to 300 K in the isothermal-isobaric ensemble (NPT) under the same restraint condition as in the previous step and followed by simulation for 100 ps with no harmonic restraint. At this point, the system was ready for the production run, which was performed using the Langevin thermostat under the NPT ensemble, at a 2 fs time step. All production runs were performed for 150 ns. The analysis was performed using the cpptraj module of AmberTools. 5 Atomic coordinates of the TLR4 complexes (TLR4-FP11-01.pdb, TLR4-FP11-02.pdb, TLR4-FP18.pdb, TLR4-FP111-A.pdb and TLR4-FP111-B.pdb) after simulation are available.

Docking calculations of the binding of FP11, FP18, and FP111 to TLR4/MD-2
A deeper analysis of the docked binding poses revealed, in the case of FP18, a tendency for all poses to bury the three lipid acyl tails deep inside the hydrophobic cavity of MD-2, occupying the full pocket. In the case of FP11, different docked poses were obtained, most of them with all the FA chains allocated inside the MD-2 pocket, and other poses with one acyl chain located into the hydrophobic channel of MD-2, delimited by residues Arg90 and Phe126, and the other FA chains inside the MD-2 cavity (Fig. 1B, top). This second binding pose is similar to that observed for E. coli LPS in human and murine (TLR4/MD-2)2 heterotetramers (PDB ID 3FXI 6 and 3VQ2 7 ), where one FA chain is found occupying this channel. The presence of longer acyl chains in FP11 (C14 vs C12 in FP18) may favor the placement of one of them protruding into the MD-2 channel, whereas this phenomenon was not observed for FP18. Regarding FP111, in all AutoDock4 predicted poses, at least one FA tail was left outside the MD-2 pocket, without interacting either in the MD-2 Phe126 channel, and remaining exposed to the outer (Fig. 1B, bottom, left). These binding modes justify the unfavorable predicted binding energies.
The best docked clusters from each compound were visually inspected, with special attention to the ligand/receptor interactions, to establish a relationship between the chemical structure of the ligands and their effect on functional activity of TLR4 receptor. For FP11 and FP18, common features were observed. The polar head groups were placed at the entrance of MD-2 cavity, interacting with the polar residues present in that region, whereas the FA chains established contacts with many hydrophobic residues from the MD-2 pocket, specifically, Val24, Ile32, Ile46, Val48, Ile52, Leu54, Leu61, Ile63, Leu74, Phe76, Leu78, Ile80, Val82, Leu87, Ile94, Tyr102, Phe104, Ile117, Phe119, Phe121, Ile124, Phe126, Ser127, Tyr131, Val135, Phe147, Leu149, Phe151, and Ile153. The ligands phosphate groups were often placed at the rim of MD-2 where they are exposed to the solvent, in agreement with the reported X-ray crystallographic complexes of TLR4/MD-2   with glycolipids (for example, complex with Eritoran, PDB ID 2Z65, 8 or with lipid IVa, PDB ID 2E59 9 ). Two different orientations have been reported for TLR4 binders, rotated 180° between them 10 : type A (antagonistlike binding mode), similar to that found for lipid IVa in PDB ID 2E59; 7 and type B (agonist-like binding mode), similar to that found for E. coli lipid A in PDB ID 3FXI. 6 These two different binding types lead to opposed biological activities. 11 Remarkably, the obtained docked poses for FP11 and FP18 showed a type B like binding mode, in agreement with their reported in vitro agonist activity.
Most of the FP11 and FP18 poses presented the hydroxyl group of the saccharide moiety interacting with the Glu92 side chain from MD-2, and the amide CO group was often close to the Arg90 of MD-2, establishing hydrogen bonds and electrostatic interactions, while the ester oxygen of the 4-acyl chain interacted with the hydroxyl group of the MD-2 Ser120. The main difference observed between FP11 and FP18 ligands was related to the interactions established by the phosphate group; in FP11 poses, the phosphate interacted with the backbone of MD-2 Lys122, whereas in FP18 interacted with MD-2 Arg90 (Fig. 1B, top, right). The relevance of the interactions between the studied ligands and the above-mentioned MD-2 residues, which are key residues in the recognition of E. coli LPS by the TLR4/MD-2 complex according to the literature data must be highlighted. 6,12 As for FP111, although AutoDock Vina was able to produce plausible docked poses inside the TLR4/MD-2 system, AutoDock did not lead to poses with favorable predicted binding energies. It is well-known the better performance correlation between predicted binding free energy and experimental value for AutoDock versus AutoDock Vina, as well as better precision and success rates. 13 Remarkably, FP111 poses showed the two above mentioned types of ligand poses, A and B, along the results generated by means of both docking programs (Fig. 1B, bottom, right). In any case, we also analyzed the docking results for FP111 from AutoDock. FP111 poses were predicted to be anchored through one phosphate group to MD-2, and the second phosphate to positively charged residues present in TLR4; in the type A poses, the 6-phosphate group interacted with TLR4 residues, equivalent to the interaction of the 1-phosphate in the type B poses (Fig. 1B,   bottom). The 6-phosphate should mimic the interaction of the hydroxyl group present in FP11 and FP18 but, given a bigger size and different electronic distribution, this interaction is not possible and tries to reach positive residues at TLR4. As a consequence of these contacts between FP111 and TLR4, the generated poses remained not as deep inside the MD-2 pocket as in the FP11 and FP18 cases, and more exposed to the solvent. These observations could explain the unfavorable predicted binding energy. We selected two of the best binding poses for further study, one type A (antagonist-like binding pose, Fig. 1B, bottom, right), and one type B (agonist-like binding pose, Fig. 1B, bottom, right). In the type A docked pose, rotated 180° with respect to FP11 and FP18 predicted poses, the 1-phosphate interacted with TLR4 Arg264, establishing two S7 hydrogen bonds, and with MD-2 Tyr102 residue, whereas the 1-phosphate interacted with MD-2 residues Ser118 and Ser120. The 4-acyl chain was found outside MD-2, interacting with unexpected residues, such as MD-2 Gly56, Ser57 and Lys58. These residues have not been reported before in the interaction between TLR4 modulators and the receptor complex. The other two FP111 acyl chains, buried inside MD-2, established contacts with Leu61, Phe76, Leu78, Ile94, Phe121, Lys122, Ile124, Phe126, Cys133, Val135 and Phe151. In the type B orientation, the 1-phosphate interacted with MD-2 Arg90 and Glu92 residues, establishing a hydrogen bond with the latter one, similar to the interaction previously described for the hydroxyl group of FP11 and FP18, and the 6-phosphate interacted with the Lys362 of TLR4. The disposition adopted by FP111 due to this last contact, forces to place the 4-acyl chain outside the MD-2 cavity, establishing a hydrogen bond between the ester CO group and the TLR4 Arg264 side chain. Additionally, other electrostatic and hydrophobic interactions were observed between this acyl tail and TLR4 residues, namely Asp101, Tyr292, Leu293, Tyr296, Ser317, Val318, Thr319, Asn339, Cys340 and Lys341, and between the remaining FA chains, located inside MD-2 protein, and MD-2 residues, such as Val24, Ile32, Ile46, Val48, Leu61, Ile63, Phe76, Leu78, Ile94, Tyr102, Phe104, Phe117, Phe119, Val135, Phe147 and Phe151. It is well known that the determining factors of the immunostimulatory activity of TLR4 modulators, are the number and the distribution of acyl chains and the phosphorylation pattern. As example, the orientation of the lipid IVa is rotated by 180° in the di-saccharide plan thus lipid IVa presents two different molecular patterns of interaction for human and mouse TLR4 receptors. These different binding modes of lipid IVa determine how the phosphate groups interact with the TLR4/MD-2 complex and are crucial for the observed different behavior among both species. 11 Interestingly, the distances between the two phosphates in all the FP111 docked solutions ranged from 5.4 to 8.2 Å, values lower to the distance between the two phosphates of the agonist E. coli lipid A at C1 and C4' positions (distance of 12.4 Å in PDB ID 3FXI 6 ). Specifically, the two phosphate groups were placed at a distance of 6.5 Å in the selected FP111 type A pose, and of 8 Å in the type B. Overall, we can conclude there is a different binding pattern for FP111 in comparison to FP11 and FP118, not including extensively reported and well-known TLR4/MD-2/ligand interactions, 10 as well as unfavorable predicted binding energies. This anomalous behavior can be explained by the presence of two phosphate groups in 1, 6-positions simultaneously, that cannot allow a proper docking into the MD-2 rim and need to anchor to TLR4, unlike the mono (1-or 6-) phosphate pattern, accounting for the lack of activity observed for this compound.

MD simulations of (TLR4/MD-2)2 complex with FP11 and FP18
Stability of the best FP11 and FP18 predicted binding modes was confirmed by molecular dynamics (  The relative orientation between the ligands and MD-2 was evaluated. We arbitrarily defined two vectors, one from the amide α-carbon atom to the ester α-carbon atom of the ligand, and another one from the αcarbon of residues Pro78 to Thr105 of MD-2 (Fig. S3a). The angle between these two vectors was plotted both over time, and it was observed that none of the ligands undergoes orientation flip during the 50 ns simulations, all remaining in the agonist type B orientation obtained from the docking calculations (Fig. S3b).
Furthermore, the motion of the TLR4 molecular switch, MD-2 Phe126 chain was also evaluated (Fig. S4). FP11 and FP18 ligands were able to retain the agonist conformation for MD-2 Phe126 along simulation time. S10 Therefore, we suggest these complexes as plausible binding modes for FP11 and FP18 accounting for their agonist activity in the TLR4/MD-2 system. During the MD simulations, the similar interactions were observed for FP11 (FP11-01 and FP11-02) and FP18 compounds, in both TLR4/MD-2 units. The ligands suffered a slightly reorientation of the saccharide moiety, which allowed them to establish a new electrostatic contact with each corresponding TLR4 chain, concretely between the hydroxyl group of the ligands and the TLR4 Arg264 residue (Fig. S1b, c).
Consequently, the interaction between the hydroxyl group of the compounds and MD-2 Glu92 was lost in all the cases, and this residue became to interact with both, the ester CO group or the oxygen of the ligands 2acyl chain, depending on the case. Interestingly, in the MD simulation of the (TLR4/MD-2/FP18)2 complex, the interaction between the phosphate group and MD-2 Arg90 was lost and two new contacts with MD-2 S11 residues were established, a hydrogen bond with the backbone of Lys122, as observed in the FP11 docked poses, and a polar contact with Ser120. Regarding FP11 poses, a new polar contact was displayed between the ligand 1-phosphate group and MD-2 Ser120, as in FP18 simulation, additional to the initially present interaction with Lys122, predicted by docking programs, which was maintained along the simulations. The interaction between Arg90 and the amide CO group of the 2-acyl chain was maintained in all the ligands.

In vitro evaluation of the cytotoxicity of compounds FP11 and FP18
THP1-derived macrophages were exposed to increasing concentration of compounds FP11 and FP18. FPs molecules do not affect THP-1 macrophage viability Fig. S18: THP1-derived macrophages were exposed to FP11 (A) and FP18 (B) at a concentration range of 0-20 µM. Cell viability was measured via MTT assay. Results were normalised to untreated controls and shown as mean ±SD of 5 independent samples.    Table S2: Log P and log S calculations. The values of QPlogPo/w and QPlogS parameters were computationally calculated within the QikProp tool implemented in the Maestro package (www.schroedinger.com). a Predicted octanol/water partition co-efficient log P. b Predicted aqueous solubility, log S. S in mol dm -3 is the concentration of the solute in a saturated solution that is in equilibrium with the crystalline solid.
Assessment of water solubility of FP11 and FP18 compounds as predicted by Qikprop. The highest log P value was obtained for FP11, indicating higher lipophilicity that might result in lower water solubility. This is in agreement with the lower log S value predicted for FP11, compared with FP18. In any case, this did not interfere with the performance of the cell assays.