An AI-guided Data-driven Methodology to Design Thermostable Long-acting HIV-1 Immunogen

Introduction: : Products based on RNAs formulated in lipid nanoparticles (LNP), including HIV-1 immunogens, hold great promise and are all in many stages of clinical development, but face significant challenges in stability. Preserving the bioactivity of RNA-LNP formulations requires ultra-cold storage, posing major barriers to rapid global distribution, especially in resource-limited settings. Primary wastage in the total cost of immunization comes from improper vaccine storage and expiration. Here, we aim to develop highly scalable vacuum-drying protocols and harness the power of Artificial Intelligent (AI) to enable rapid excipient formulation optimization for solid-state HIV mRNA-LNP, ensuring their integrity and stability under long-term storage at ambient temperature.

Materials and

Methods: : We first used Fluc mRNA as a model and high throughput in vitro cell assays to streamline the process of identifying suitable excipient compositions. We implemented AI-guided Bayesian optimization models to design experimental groups. Materials including sugars, amino acids, buffers, and polymers to stabilize LNP were selected based on preliminary experiments and existing literatures. Solid-state mRNA-LNPs were made by mixing soluble LNP with excipient compounds and vacuum-drying for 48-hour. Samples were then rehydrated and treated to cells. Bioluminescence data from in vitro assay was input into the AI model. The model optimized the data set and prompted the next conditions to evaluate. This interactive feedback loop was repeated until we obtained the most optimized formulations. After identifying the optimized formulations, solid-state mRNA encoded for N332-GT2 HIV Env trimer loaded LNPs were fabricated and tested in vivo in mouse model.

Results, Conclusions, and Discussions:: Solid-state and water-free RNA-LNP formulations enhance thermostability by reducing risks of degradation, aggregation, and hydrolysis. The selection of appropriate protectants or excipients plays a critical role in preserving essential properties and bioactivity of LNP within the solid matrix. Yet, conventional manual screening methods are inefficient and resource-heavy, given the wide range of factors involved. Thus, searching for an optimal solution while minimizing the number of experiments performed is crucial. Figure 1 describes the implemented high-throughput experimental framework to screening to identify optimized excipient composition. After solid-state formation of mRNA-LNP with excipients via vaccum drying at ambient temperature, samples were rehydrated in PBS then treated to cells for 24 hours. Bioluminescence signals were measured and normalized to unprocessed LNPs. The data generated from experiments are then input to the AI models and the model makes predictions of the next sets of conditions to test. After 2 iterations, we were able to identify optimal excipient candidates that were able to preserve 100% bioactivity of mRNA-LNP after drying (Figure 2A). Solid-state N332-GT2 HIV Env trimer mRNA-LNP stabilized with this optimized formulation was able to elicit similar levels of immune responses compared to fresh LNPs (Figure 2B). Through this AI-guided high throughput approach, we can quickly identify excipient combinations that could preserve LNP bioactivities after water-removal process. The AI-guided model will accelerate discovery by prioritizing candidates for experimental validation, reducing trial-and-error cycles and enhancing precision. This methodology could accelerate the designing process of a thermostable long-acting HIV-1 immunogen delivery system.

Acknowledgements and/or References (Optional):: Khanh Tran is a recipient of a postdoctoral fellowship from the Ludwig Center at MIT’s Koch Institute for Integrative Cancer Research. Jinbi Tian is a recipient of Siebel Scholars Foundation Fellowship. This abstract is based on research funded by (or in part by) the Gates Foundation. The findings and conclusions contained within are those of the authors and do not necessarily reflect positions or policies of the Gates Foundation.