Optimization of Lipid Nanoparticles for saRNA Expression and Cellular Activation Using a Design-of-Experiment Approach

Abstract

Lipid nanoparticles (LNPs) are the leading technology for RNA delivery, given the success of the Pfizer/BioNTech and Moderna COVID-19 mRNA (mRNA) vaccines, and small interfering RNA (siRNA) therapies (patisiran). However, optimization of LNP process parameters and compositions for larger RNA payloads such as self-amplifying RNA (saRNA), which can have complex secondary structures, have not been carried out. Furthermore, the interactions between process parameters, critical quality attributes (CQAs), and function, such as protein expression and cellular activation, are not well understood. Here, we used two iterations of design of experiments (DoE) (definitive screening design and Box-Behnken design) to optimize saRNA formulations using the leading, FDA-approved ionizable lipids (MC3, ALC-0315, and SM-102). We observed that PEG is required to preserve the CQAs and that saRNA is more challenging to encapsulate and preserve than mRNA. We identified three formulations to minimize cellular activation, maximize cellular activation, or meet a CQA profile while maximizing protein expression. The significant parameters and design of the response surface modeling and multiple response optimization may be useful for designing formulations for a range of applications, such as vaccines or protein replacement therapies, for larger RNA cargoes.

Keywords: Box−Behnken Design; cytokine response; definitive screening design; design-of-experiment (DoE); lipid nanoparticle (LNP); mRNA (mRNA); protein expression; self-amplifying mRNA (saRNA).

Figure 1

Composition of LNPs including ionizable lipid, phospholipid, cholesterol, and PEGylated lipid encapsulating RNA of varying size; firefly luciferase mRNA (fLuc mRNA, ∼1700 nt) or self-amplifying RNA (fLuc saRNA, ∼9300 nt).

Figure 2

Statistical analysis from definite screening design experiments (Iteration A). (A, B) Box plot displaying LNP size and PDI as a function of PEG lipid content. (C) Heat map of correlations (p-value) from ANOVA carried out on the restricted data set. Explanatory variables with p-value < 0.1 are excluded from ANOVA model. (D) Heat map of regression coefficients for main and quadratic effects detected by ANOVA.

Figure 3

Response surface plot from Box–Benhken design experiments (Iteration B). IL-6 cytokine release (heat map) as a function of LNP size (A), encapsulation efficiency (B), and RNA integrity (C), predicted using second-order ordinary least square regression. Nondisplayed explanatory variables are fixed at center point, and ALC-0315 is herein selected (see Table 2). Black circles indicate the local maxima for protein expression, and black squares indicate the local minima for protein expression.

Figure 4

LNP protein expression profile and statistical analysis from BBD experiments (Iteration B). (A) Box plot displaying protein expression (RLU) as a function of ionizable lipid type. (B) Response surface plot for protein expression for LNP containing ALC-0315 ionizable lipid. (C) Response surface plot for protein expression for LNP containing SM-102. Heat map plot displaying Spearman correlation’s coefficient (D) and p-values (E) between CQAs. LNPs containing MC3 ionizable lipid are excluded from the protein expression analysis in B–E due to low expression levels.

Figure 5

Multiple response surface plot from BBD (Iteration B) based on desirability function optimization. IL-6 cytokine release is minimized in panels A–C (Desirability (1)) and maximized in panels D–F (Desirability (2)). In panels G–I, only physical CQAs are optimized (Desirability (3)). Nondisplayed explanatory variables are fixed at center point (see Table 2).