Development of hematopoietic stem cell−targeted lipid nanoparticles through lipid composition optimization

Efficiently delivering gene-editing tools into hematopoietic stem cells (HSCs) is a major hurdle for treating hereditary blood disorders. While lipid nanoparticles (LNPs) are a promising delivery vehicle, existing formulations often struggle to achieve the high transfection efficiency required for therapeutic success in these sensitive cells. This study aimed to develop a more effective LNP formulation by optimizing lipid composition using machine learning.

The authors utilized a Bayesian optimization framework to systematically refine the lipid composition of LNPs. By iteratively testing 68 different molecular ratios of functional amino lipids (FFT-10 and FFT-20) alongside standard LNP components, they used Gaussian process regression to predict and synthesize formulations that maximize transfection efficiency in cord blood CD34+ cells while minimizing toxicity. They validated these optimized LNPs by performing ex vivo CRISPR-Cas9 gene editing of the TP53 gene and the BCL11A enhancer, and by testing RNA delivery in primary patient-derived leukemia cells.

The optimization process yielded several high-performing LNPs, most notably LNP-384. This formulation demonstrated significantly improved transfection efficiency compared to previous iterations. In functional assays, LNP-384 successfully facilitated up to 40% on-target gene editing of TP53 in human CD34+ cells, allowing the cells to bypass drug-induced growth inhibition. Additionally, the researchers achieved approximately 20% allelic modification at the BCL11A enhancer, a key target for treating hemoglobinopathies. While the LNPs showed variable performance in patient-derived leukemia samples, one formulation (LNP-413) achieved 90% transfection in a specific monocytic leukemia model, highlighting the potential of this platform for both hereditary and malignant hematopoietic conditions.