Structure-Based Modeling of Environment-Dependent Protonation States Across LNP Formulations with Atomistic CpHMD

The p<i>K</i>_a values and associated protonation states of ionizable lipids in lipid nanoparticle (LNP) formulations are strongly dependent on their chemical environment. This phenomenon leads to poorly understood structure-function relationships that influence payload delivery, tissue selective biodistribution, and manufacturing. For example, the charge- and biodistribution of an mRNA-loaded LNP can vary based on the type of ionizable lipid used, the molar ratio of its components, and its cargo. Yet, the spatial resolution of experimental protonation-state measurements is currently limited to the apparent charge of an ionizable lipid averaged over all environments/conformations of the LNP─best represented by the apparent p<i>K</i>_a value. Such measurements are too coarse to capture the heterogeneous charge distributions of ionizable lipids in LNPs, which influence biocorona formation and interactions with the payload. Similar limitations are inherent to classical, fixed-protonation-state, in silico models that cannot capture the environment-dependent protonation states determining local p<i>K</i>_a. To address this gap in experimental and computational tools available to accurately determine the local charge distributions in LNPs, this work now incorporates a scalable continuous constant pH molecular dynamics (C<i>p</i>HMD) model to simulate the self-assembly processes of five reported distinct LNP formulations. Parameters for ionizable lipids were generated from fitting fixed λ-state calculations performed with Hamiltonian replica exchange (HREX) to improve conformational sampling during parametrization. Simulated systems were composed of 100 ionizable lipids (50 mol %), cholesterol (40 mol %), distearoylphosphatidylcholine (10 mol %), and mRNA (20 nucleotides) to model the interior of an LNP. Self-assembly was simulated for 100 ns at different pH values and integrated with bilayer models for neutral/basic pH conditions to validate the apparent p<i>K</i>_a for each system. For our most accurate mixed self-assembly/bilayer model, the theoretically calculated apparent p<i>K</i>_a values matched well with the experimental values (mean absolute error (MAE) = 0.32 p<i>K</i>_a units (<i>R</i>² = 0.52)), and all systems exhibited pH-dependent structures. Overall, this work provides a new computational platform technology to (i) predict the p<i>K</i>_a values of ionizable lipids in different chemical environments and (ii) enable a structure-based way to model the heterogeneous, environment-dependent charge distributions of ionizable lipids in LNP systems typically encountered during LNP manufacturing and delivery.