Machine Learning Accelerates Lipid Discovery for Versatile mRNA Delivery in Nanomedicine

Although there have been advancements with mRNA delivery using LNPs, finding ionizable lipids that can work across a variety of therapeutic uses has been a roadblock. While some lipids have been effective—like those used in COVID-19 vaccines—they’re often optimized for specific applications, such as targeting the liver. To make mRNA therapies viable for treating more diseases, we need lipids that perform well in different tissues without causing unwanted immune responses. The problem is, there’s an enormous number of possible lipid structures, and manually exploring each one is both time-consuming and costly. So, there’s a real need for a way to speed up the discovery of these essential lipids.

Recent paper published in Nature Materials and performed by Assistant Professor Roy van der Meel, Francesca Grisoni & Professor Willem Mulder from the Eindhoven University of Technology, tackles a crucial aspect of mRNA-based therapies: figuring out how to get mRNA safely and effectively into cells. mRNA therapies have shown a lot of promise, especially for vaccines and gene treatments, but they rely on delivering fragile mRNA molecules without breaking them down along the way. That’s where lipid nanoparticles (LNPs) come in. These tiny carriers protect mRNA as it navigates through the body’s barriers to reach target cells. A vital part of these nanoparticles is the ionizable cationic lipids, which bind to mRNA and help it get into cells. But, finding the right lipids that work well and are safe is a tough challenge.

To tackle this, the researchers turned to machine learning and high-throughput synthesis. By harnessing the predictive power of computational models, they aimed to save time and resources when identifying promising lipid candidates. Their approach combined machine learning with a method called the Ugi four-component reaction, which allowed them to quickly create a large library of lipids. This method enabled them to predict which lipids might offer high transfection efficiency and a low immune reaction before even stepping into the lab for extensive testing. The motivation here was straightforward: by streamlining the discovery of new ionizable lipids, the researchers hoped to unlock mRNA’s full potential, making it more accessible and adaptable for treating various diseases. Through this innovative strategy, they not only aimed to enhance delivery methods but also to expand mRNA’s therapeutic reach. This could potentially open up mRNA therapies for targeting more than just the liver, extending to the lungs, muscles, and other tissues as well.

The researchers carried out a series of experiments to find better ionizable lipids for mRNA delivery, focusing on both the discovery and effectiveness of these lipids. To start, they used a chemical reaction process known as the Ugi four-component reaction. This technique allowed them to quickly generate a library of 384 unique ionizable cationic lipids. Once they had these lipids, they formulated them into lipid nanoparticles (LNPs) containing luciferase mRNA—a marker that makes it easy to track transfection efficiency in cells. Testing these formulations in vitro, they assessed how well each lipid could deliver mRNA by observing bioluminescence, a glow that indicates successful delivery. The experiments revealed several lipids with high transfection efficiency, showing the researchers which candidates might be worth exploring further. Building on these initial findings, the team wanted to broaden the chemical diversity of their lipid library. They synthesized an additional 200 lipids and then evaluated these in the same way, bringing the total number of tested lipids to 584. This expanded dataset enabled them to train a machine learning model to predict which lipid structures might work best for mRNA delivery. By using the transfection results from all these trials, they developed a model that could estimate how effective untested lipids might be based solely on their chemical makeup. The model’s predictions then guided them to screen a vast virtual library of 40,000 lipids. They synthesized 16 promising lipids based on these predictions, which was a much smaller and more manageable number to test in actual experiments.

Among the newly synthesized lipids, one candidate—referred to as 119-23—consistently showed impressive results. When the team tested LNPs formulated with 119-23 in mice, they compared its performance to well-established lipids like MC3 and SM102, which are commonly used in clinical mRNA and siRNA delivery. The experiments revealed that LNPs containing 119-23 led to higher gene expression levels after intramuscular injection than the formulations based on MC3 or SM102. Specifically, they found that 119-23 did a better job of delivering mRNA encoding for human erythropoietin (hEPO), suggesting its potential as a powerful tool for therapeutic applications.

Further experiments evaluated how well 119-23 could target different tissues. The researchers delivered LNPs containing 119-23 to the lungs via intravenous injection. In these trials, 119-23 outperformed the standard lipids once again, showing higher transfection efficiency in lung tissue. The findings suggested that 119-23 was not only effective for localized delivery but could also be suitable for systemic treatments, making it a versatile candidate for mRNA-based therapies. The team further tested its capability for organ targeting using reporter mice models. These tests confirmed that 119-23 could transfect a range of cell types within tissues like the liver, spleen, and lungs, underscoring its potential adaptability for diverse therapeutic needs.

Overall, the researchers’ experiments demonstrated that 119-23 could provide a more effective and versatile solution for mRNA delivery compared to conventional lipids. The use of machine learning, coupled with high-throughput synthesis, streamlined the process of finding such an impactful lipid, underscoring the value of this approach for future nanomedicine development. The study highlighted how computational tools and experimental work could come together to accelerate the discovery of crucial materials in medical research.  The significance of this study lies in its potential to revolutionize how we approach mRNA-based therapies. By leveraging machine learning and high-throughput synthesis, the researchers not only sped up the discovery process for ionizable lipids but also broadened the range of therapeutic applications for mRNA delivery. This is especially important given that mRNA technology has vast potential beyond the liver-focused applications we commonly see, such as in COVID-19 vaccines. With the discovery of lipid 119-23, which demonstrated superior delivery capabilities across multiple tissues, the study suggests that it’s possible to develop more adaptable and effective mRNA therapies for a broader spectrum of diseases. Another major implication is the way this approach could impact drug development. Traditionally, identifying viable lipid candidates for mRNA delivery involved lengthy, resource-intensive trials. The integration of machine learning models into this process means that researchers can now predict the effectiveness of new lipids much more efficiently, saving both time and resources. This not only accelerates the discovery process but also makes it more feasible to tailor lipid nanoparticles for specific treatments, possibly even personalizing mRNA therapies for individual patients in the future. Moreover, the success of this study highlights the growing importance of computational tools in medical research. Machine learning can handle complex datasets and detect patterns that may be overlooked in traditional experimental setups. As more high-quality data becomes available, such computational approaches could allow us to reverse-engineer LNP formulations to achieve desired effects, such as targeting specific organs or minimizing immune responses. This suggests that the field of nanomedicine is moving toward a future where we can fine-tune mRNA delivery systems with unprecedented precision, ultimately making treatments safer and more effective for patients. Finally, the broader implications of this study extend to the potential for mRNA therapies to address a wider array of diseases. The ability of lipid 119-23 to deliver mRNA to different tissues means that we may be able to target conditions beyond just genetic and infectious diseases, potentially expanding into areas like cancer, metabolic disorders, and even regenerative medicine. This versatility could transform the way we think about and develop treatments, as it opens the door for mRNA therapies to become a cornerstone of personalized medicine.