Beyond The Liver: How Machine Learning-Optimized LNPs Are Expanding RNA Delivery Into Adipose Tissue

By Jagesh V. Shah, Ph.D., SVP, head of platform, Mirai Bio

For years, one of the defining limitations of RNA therapeutics has not been the cargo itself but where that cargo can go. The next phase of the field will depend on delivery systems that are not only potent but also selective, translatable, and designed around the specific cell- and tissue-specific biology they are intended to reach.

Lipid nanoparticles (LNPs) have enabled major breakthroughs in nucleic acid medicine, from siRNA therapeutics to mRNA vaccines, yet most clinically validated delivery systems still exhibit strong liver tropism. That reality has shaped much of the field’s development strategy. Many RNA programs have been built around diseases where hepatic delivery is either desirable or at least acceptable.

But some of the most commercially and biologically important diseases — including obesity, cardiometabolic disease, and metabolic dysfunction — depend on accessing tissues beyond the liver.

At the 2026 American Society of Gene & Cell Therapy (ASGCT) Annual Meeting, Mirai Bio researchers presented new preclinical findings demonstrating that machine learning-guided LNP optimization may help overcome one of the field’s most persistent challenges: delivering RNA therapeutics into adipocytes while minimizing off-target accumulation in the liver and spleen.

The work highlights a broader shift underway in RNA therapeutics, where delivery is increasingly being treated not simply as a formulation problem but as a data-driven engineering discipline.

Why Adipocyte Delivery Matters

Adipose tissue has rapidly emerged as one of the most strategically important targets in modern drug development.

The rapid growth of obesity therapeutics, driven largely by GLP-1 receptor agonists, has transformed metabolic disease into one of the largest markets in pharmaceutical history. Yet most current therapies still rely on peptides or protein-based mechanisms that require chronic administration and systemic exposure.

RNA therapeutics offer the possibility of a very different approach: transiently programming cells to produce therapeutic proteins or modulate metabolic pathways directly within target tissues. The challenge is that adipocytes have historically been difficult to reach with conventional LNP systems.

Most LNP formulations preferentially accumulate in the liver because of their interaction with serum proteins and endogenous lipid transport pathways. While this hepatic tropism has been advantageous for liver-directed therapies, it creates a major obstacle for extrahepatic applications. Achieving meaningful adipocyte transfection while limiting liver exposure therefore represents a significant delivery milestone for the development of tissue-specific RNA delivery systems.

Moving Beyond Trial-And-Error LNP Design

Traditional LNP development has often relied on iterative screening: formulate particles, test biodistribution, empirically modify chemistry and composition, and repeat.

The newer approach presented at ASGCT applies machine learning to that process. Researchers screened more than 400 proprietary ionizable lipids in vivo and used biodistribution data to guide iterative optimization of LNP composition and component selection. Rather than evaluating single variables in isolation, the machine learning algorithm analyzed multiple delivery attributes simultaneously, including adipose targeting efficiency and off-target tissue distribution.

In this setting, machine learning is not a substitute for in vivo testing. It is a way to learn from in vivo data more efficiently, prioritize increasingly more effective formulations, and refine the design space based on the delivery attributes that matter most. This type of multiparameter optimization reflects the growing complexity of modern RNA delivery science.

In practice, successful delivery depends on a tightly interconnected set of variables, including:

• ionizable lipid chemistry
• particle composition
• RNA encapsulation behavior
• serum interactions
• tissue biodistribution
• intracellular trafficking
• endosomal escape efficiency.

Machine learning platforms can help navigate this multidimensional design space more efficiently than conventional formulation screening alone.

What The Data Showed

The optimized LNP formulations demonstrated strong white adipose tissue expression alongside reduced off-target signal in the liver and spleen in preclinical models.

According to the presented data, cellular profiling confirmed meaningful adipocyte transfection, while immunofluorescence imaging showed selective adipocyte delivery. Liver immunohistochemistry demonstrated comparatively low off-target hepatic accumulation in lead formulations. Flow cytometry further suggested limited delivery into non-adipocyte cell populations.

Collectively, the findings suggest that adipose-enriched RNA delivery may be achievable through systematic optimization of LNP composition rather than relying solely on ligand-based targeting approaches. Importantly, the work also reinforces a growing realization across the field: bulk formulation metrics alone are increasingly insufficient for understanding functional delivery performance.

As RNA therapeutics become more tissue-specific and biologically sophisticated, nanoparticle architecture, payload organization, biodistribution patterns, and cell-type specificity are becoming central engineering variables.

Delivery As The Next Competitive Layer In RNA Therapeutics

The broader significance of this work extends beyond obesity.

RNA therapeutics have already demonstrated that programmable nucleic acid cargo can generate clinically meaningful biology. The next major competitive frontier is likely to center on tissue access.

Many of the most attractive future applications for RNA medicine — including CNS disease, immunology, fibrosis, regenerative medicine, and metabolic disease — require reliable extrahepatic delivery. This is why delivery platforms are increasingly becoming strategic assets rather than secondary enabling technologies.

The ability to repeatedly engineer LNPs capable of selectively reaching distinct tissues could dramatically expand the therapeutic addressability of nucleic acid medicines. In that sense, delivery may ultimately become one of the most important determinants of platform value in RNA therapeutics.

A Shift Toward Precision Delivery Engineering

Historically, RNA delivery has often been framed as a binary problem: did the payload reach the tissue or not? That framework is evolving. Modern RNA therapeutics increasingly require precision delivery — not simply tissue exposure but controlled delivery into the correct cell populations, at the appropriate levels, while minimizing off-target effects and immune activation. This represents a shift from generalized biodistribution toward cell- and tissue-specific delivery engineering.

The integration of machine learning, in vivo biodistribution data sets, and advanced lipid chemistry suggests the field may be entering a new phase where delivery systems themselves become increasingly rationally engineered and computationally optimized. For RNA medicine, that evolution could prove just as important as advances in the therapeutic cargo itself.

About The Author:

Jagesh Shah, Ph.D., is senior vice president, Head of Platform at Mirai Bio where he leverages his background in quantitative biology, biophysics, and systems biology, as well as his experience working on the development of in vivo delivery approaches for novel therapeutics. Before joining Mirai, Shah served as vice president, Gene Therapy Technologies at Sana Biotechnology and prior to that as a senior fellow at Flagship Pioneering and vice president of research at Cobalt Biomedicine. Before holding positions in the biotech industry, Shah spent over a decade as faculty at Harvard Medical School and Brigham and Women’s Hospital, leading a laboratory focused on systems biology research. Shah received his doctorate in medical engineering from MIT and completed his post-doctoral fellowship at UCSD.