(A-15) A novel molecular model for the identification of nanoparticle forming drug-excipient pairs

Introduction: Coformulations of hydrophobic, insoluble drugs with inactive excipient molecules have shown to produce spherical nanoparticles with high drug loading capacity. However, the internal structure and physical mechanism of formation of these nanoparticles remains unknown. Here we propose a novel computational model that aims to simulate nanoparticle stability to improve our understanding of the molecular underpinnings of nanoparticle formation. We hypothesize that the nanoparticles consist of a hydrophobic drug core surrounded by a thin shell of stabilizing excipient molecules, like the candy jawbreakers. Molecular assemblies of drug cores and excipient surfaces were parametrized as all-atom models and were simulated using the Molecular Dynamics engine OpenMM. Over the course of the trajectory, Solvent Accessible Surface Area (SASA) of the drug core was calculated as a proxy measurement for nanoparticle stability. We hypothesized that nanoparticles that were stable in this biased jawbreaker configuration should be stable when synthesized. Our model correctly predicts one of the most stable excipients from our previous work as being conducive towards nanoparticle formation. Furthermore, we simulated and successfully synthesized a nanoparticle using this excipient with a previously untested chemotherapeutic drug.

Materials and Methods: A library of 9 drugs and 8 excipients were identified from our previous work. We then generated molecular representations of each molecule for simulation using an in-house Python pipeline. We derived a mathematical expression for the number of excipients needed to fully coat a core of 100 drug molecules using geometric approximations, based on the excluded volume of each atom inside of the excipient molecule according to its Van der Waals radius. Using the MD system generation software, Packmol, we placed 100 drug molecules into a spherical core, and coated it with the calculated number of excipients to result in full core coverage. 25 nanoseconds of NPT molecular dynamics simulation were performed using the software OpenMM. Simulations were visualized and rendered into graphics using PyMol. SASA was calculated using the Shrake-Rupley algorithm implemented in VMD. SASA values for simulations of the same drug were ranked, with the lowest SASA score corresponding to 1 and the highest 8. Nanoparticles were prepared by adding a 1:1 ratio of drug and excipient stock in DMSO and mixing via pipette. Phosphate buffered saline (PBS) was added to dilute the stock by a factor of 100 and create nanoaggregates. Particle size was characterized using Dynamic Light Scattering (DLS). For the novel pair, Nilotinib and Candesartan Cilexetil, we acquired transmission electron microscopy (TEM) images of the nanoparticles.

Results, Conclusions, and Discussions: Overall, we performed a total of 72 simulations in triplicate. From these, we correctly predict that the excipient Candesartan Cilexetil (CDS) is likely to form nanoparticles, regardless of the drug used in the core (average SASA ranking of 2.0). This is commensurate with our previous findings, where we observed nanoparticle formation when using CDS as an excipient with 62.5% of tested drugs. Our second-best predicted stabilizing molecule, Indomethacin (average SASA ranking of 2.67), was also observed to form nanoparticles with 31.3% of tested drugs when used as an excipient. Our model also correctly predicts generally poor excipients, namely Neotame and Ursodiol. These excipients formed nanoparticles with 12.5% and 0% of tested drugs and had average SASA ranking of 7.0 and 5.2. Interestingly, the excipient tannic acid was predicted to be the best at forming nanoparticles (average SASA ranking of 1.13). Despite this near perfect SASA ranking, tannic acid has only been able to form nanoparticles with 18.8% of tested drugs. There are many reasons for why this may have happened, for example, the force field parameters generated for this molecule are incorrect. Tannic acid is a very large, pentameric molecule rich in hydrogen bonding sites. The effects of these sites might not have been captured with our force field, as we are using a classical force field that doesn’t account for polarizability of molecules. Further investigation with a force field that explicitly accounts for electron polarizability like AMOEBA or Drude could remedy this misclassification. Additionally, we simulated and synthesized a novel pairing, CDS with the Leukemia drug Nilotinib, and observed stable nanoparticle formation both in silico and in experimental measurements. Simulations of molecular aggregates and SASA determination serve as a promising means of screening potential excipients before experiments are performed. Our original work selected molecules from a machine learning prediction. Now, we can take these predicted excipients and screen them with a molecular dynamics simulation as a second method of filtering. This allows the saving of both time and money as we will only purchase combinations of molecules that are favorable under both criteria.