Fast Methods for Speeding Up Lipid Nanoparticle Development

A widely adopted approach for in vitro screening of LNPs involves multi‐well plate cell culture assays, which offer parallelized evaluation of their transfection efficiency, cellular uptake, and cytotoxicity. These assays can be conducted in plates with varying well densities, from 6 to 1534 wells, permitting large‐scale formulation screening with high reproducibility and throughput.^[62^] Early LNP screening efforts relied on low‐density multi‐well plate assays and simple cell lines, which, although informative, have limited throughput. To increase the efficiency and scale of LNP screening, combinatorial libraries consisting of diverse LNPs (e.g., varying ionizable lipid structures, helper lipids, cholesterol, and PEG‐lipid ratios) are screened in 96‐well or higher density plates using appropriate culture conditions.^[52, 63, 64, 65^] Assay design in these platforms can be tailored for functional readouts relevant to nucleic acid delivery. For instance, reporter mRNAs (e.g., luciferase) are encapsulated in LNPs and applied to cultured cells, where transfection efficiency is quantified via luminescence or fluorescence‐based assays, providing a quantitative measure of delivery efficiency. Xue et al., screened 180 LNP formulations encapsulating firefly luciferase (Fluc) mRNA in a 96‐well plate setup to rapidly identify formulations with superior transfection efficiency and low cytotoxicity in HeLa cells.^[64^] The successful translation of mRNA into protein was quantified through luminescence assays, demonstrating the use of this platform for rapidly refining large formulation libraries to a small subset of formulations with superior transfection efficiency across various cell types for downstream in vivo validation.^[66, 67^] When combined with combinatorial lipid synthesis, these in vitro screening approaches have facilitated multidimensional screening of ionizable lipids to uncover lipid structure‐function relationships crucial for mRNA delivery. For example, systematic variation of lipid tail unsaturation (degree, type, and position) was shown to influence IL‐27 protein expression profiles in the liver in vivo.^[41^] Beyond tail modification, changes in structural features like methylene linker length and headgroup rigidity (linear, cyclic, or aromatic) have been linked to LNP delivery potency in the liver, with some candidates surpassing benchmark ionizable lipids such as SM‐102.^[55^] Notably, incorporation of a benzene ring, as in the 12T‐O14 AID‐lipid, conferred cone‐shaped geometry that enhanced endosomal membrane disruption.^[60^] Moreover, mixing 12T‐O14 with benchmark lipid MC3 in varying ratios allowed tuning of organ‐selective delivery between lung and spleen. Together, these findings reveal how in vitro HTS platforms allow rapid screening of LNP design parameters by mapping LNP structural variations to biological outcomes.

To further improve throughput, researchers have developed combinatorial libraries of ionizable lipids and other components that, when integrated with robotic and microfluidic platforms, allow the rapid synthesis and parallel evaluation of hundreds to thousands of LNP variants under uniform conditions. For instance, Cui et al., reported an automated 384‐well microplate workflow capable of formulating up to 384 distinct LNPs per plate via precise liquid handling, followed by in vitro transfection assays.^[36^] Notably, the top LNP candidates identified by this high‐throughput screening match those from conventional screening, validating the accuracy of the platform. Such automation has enabled the systematic investigation of formulation parameters to map structure‐function relationships and accelerated the iterative design‐build‐test cycle central to LNP development. Overall, multi‐well plate‐based HTS platforms have been used for guiding rational LNP formulation design. These include: i) the relevance of tuning PEG‐lipid content to balance size, stability, and cellular uptake efficiency; ii) the predominant role of ionizable headgroup chemistry in mediating endosomal escape, and iii) the sensitivity of biodistribution to minor compositional changes in LNP. However, the use of simple cell lines and static conditions limits the physiological relevance of these platforms, limiting their predictive accuracy concerning in vivo performance.

To address the limitation of low physiological relevance and enhance the predictive capability of in vitro LNP screening, advanced cell culture models and organ‐on‐a‐chip (OoC) systems have recently emerged as more physiologically relevant alternatives with improved biomimicry of tissue microenvironment and reliability.^[37, 68, 69, 70, 71^] One such example is the development of an advanced HTS blood‐brain barrier (BBB) platform incorporating endothelial monolayers with functional tight and adherens junction proteins in a high‐density multi‐well format (Figure3a).^[70^] While conventional 96‐well brain endothelial monolayers enable HTS, they lack the capacity to measure BBB transport, which requires transwell compartments.^[70^] Transwell systems, typically available in 12‐ or 24‐well formats, can be used to measure transfection efficiency but are impractical for screening large formulation libraries. In contrast, the HTS‐BBB Transwell platform (96‐well format) has been developed to reliably predict LNP transport across the BBB and mRNA transfection in a monolayer of endothelial cells.^[70^] Using this platform, 14 LNP formulations with varying ionizable lipid structures were screened, revealing that ionizable lipid headgroup chemistry had a greater impact on transport performance than LNP size and zeta potential. This highlights the critical role of lipid design in enabling effective delivery to the brain. Furthermore, HTS‐BBB does not directly predict the transfection of deeper brain tissue because the standard model includes only endothelial cells and lacks downstream target brain cell types. However, the predictive accuracy of the model was improved by co‐culturing endothelial monolayers with neurons, suggesting that increasing the complexity of HTS models may enhance physiological relevance for more tailored applications. These findings highlight key principles for brain‐targeted LNP design: i) the role of ionizable lipid chemistry in governing BBB transport; ii) independent assessment of transcytosis and transfection in parallel, and multi‐parametric readouts for refining candidate selection. The HTS‐BBB platform offers a scalable and physiologically relevant approach to systematically map structure–function relationships for tissue‐specific delivery.

Future advancements in developing physiologically relevant models for different applications are expected to further enhance the predictive accuracy of in vitro HTS platforms for LNP screening. One promising example is the use of vascularized microfluidic systems, such as the iFlowPlate (Figure 3b), which was originally developed for vascularized organoid cultures.^[72^] This platform incorporates perfusable endothelial networks and recapitulates dynamic flow conditions, offering a more biomimetic alternative to conventional static assays. The iFlowPlate was shown to maintain stable endothelial monolayers under flow, support vascularized organoid cultures, and enable high content, real‐time imaging across hundreds of wells, demonstrating its capability for physiological relevance and throughput compatibility.^[72^] Although the platform has not yet been applied directly to LNP discovery, its key features, such as fluorescent‐based real‐time tracking, live cell imaging, and perfusable vascular networks that are relevant for evaluating LNP uptake, biodistribution, and transfection across vascular barriers. Nevertheless, realizing the full potential of the iFlowPlate for LNP discovery will require addressing several practical challenges regarding scalability, specifically in terms of fabrication consistency, compatibility with automated liquid handling, and the need for standardized protocols to ensure reproducibility across laboratories. Additionally, endothelial cells within perfusable networks can exhibit sensitivity to flow rate, matrix stiffness, and media composition, all of which may introduce variability in LNP interaction readouts. These limitations must be addressed through standardized protocols and modular assay designs before such platforms can be widely adopted for routine LNP screening. Despite these limitations, the iFlowPlate is a promising physiologically relevant HTS platform that can bridge the gap between traditional static in vitro assays and in vivo models.

Although OoC platforms may be less scalable than conventional platforms and require more complex handling, their ability to simulate human organ environments in vitro (e.g., organ‐specific functions, receptor profiles, and 3D tissue architecture) makes them indispensable for refining lead LNP formulations prior to animal studies. For instance, Moderna researchers integrated the Emulate human Liver‐Chip into their screening workflow to identify LNPs for treating liver fibrosis.^[73, 74^] Traditional animal models often fail to predict human‐specific toxicity, particularly with novel LNP chemistries. The Liver‐Chip enabled early detection of gene signatures correlated with adverse outcomes in non‐human primates, as well as collagen remodeling indicative of pro‐fibrotic responses. Critically, it also distinguished between LNP‐ and mRNA‐mediated toxicity and uncovered mechanistic pathways such as necroptosis. Over 18 months, the team screened 35 LNP formulations for approximately USD 325K, over four times faster and at a fraction of the cost of equivalent non‐human primate studies, which would have required more than USD 5 M and taken over five years. These results highlight the potential of humanized OoC platforms to deliver human‐relevant data at a throughput and cost that make them suitable for screening LNP candidates before advancing to in vivo evaluation. Therefore, future development should focus on improving the scalability and automation of such platforms to enable their integration into parallelized HTS workflows, thereby enhancing physiological relevance, reducing dependence on animal models with limited predictive capability for human outcomes, and accelerating decision‐making for downstream in vivo screening.

The nanoPRISM method is a multiplexed screening method that combines nucleic acid barcoding with multi‐well plate‐based LNP screening. This technique is based on the PRISM method, which has previously been used to reliably screen a massive library of existing and potential anticancer drugs across multiple cancer cell lines.^[75, 76, 77^] The nanoPRISM method uses nucleic acid barcoding to simultaneously screen NP interaction with different cell types in a single experiment. Specifically, lentiviral barcoding plasmids containing unique 24‐base pair DNA sequences (barcodes) are transfected into host cells (e.g., HEK‐293T). The barcoded cells are pooled together in equal proportions, seeded into multi‐well plates, and subsequently exposed to various NP formulations, including multiple liposomal variants. Using this strategy, Boehnke et al., investigated the interactions of 35 distinct NP formulations, including various liposomal NPs, across 488 cancer cell lines (Figure4).^[78^] The screening process involved the following steps: 1) transfecting each cancer cell line with a plasmid containing a unique oligonucleotide barcode, 2) pooling the barcoded cells in equal proportions, 3) seeding the pooled cells on microplates, with each well treated with a different NP formulation, 4) sorting cells based on their fluorescence intensity using fluorescence‐activated cell sorting (FACS), and 5) extracting, amplifying, sequencing, and analyzing the DNA barcodes to determine NP uptake and response across cell types. Their methodology demonstrated that nucleic acid barcoding significantly enhanced screening throughout, overcoming the limitations of conventional methods, which typically assess only a small subset of formulations per experiment. Notably, this method allows the rapid, high‐throughput assessment of NP‐cell interactions at an unprecedented scale, revealing critical formulation‐dependent effects on cellular uptake and providing robust data to inform formulation optimization.

The nanoPRISM platform, when integrated with multiomics and ML approaches, enables high‐throughput profiling of NP‐cell interactions across a pool of 488 different cancer cell lines. This large‐scale screening revealed that the NP core materials (liposome, PLGA, or polystyrene) were the primary determinant of cell‐type‐specific association rather than the influence of surface chemistry. A comparative screening of liposomes with different surface chemistries showed that natural polysaccharide coatings increased receptor‐mediated interactions, while PEGylation broadly reduced both specific and non‐specific binding. By linking NP uptake profiles with multiomics data using a random forest model across 488 cell lines, the study identified SLC46A3 as a negative regulator of both liposomal and LNP uptake. Notably, cell lines with low SLC46A3 expression exhibited markedly higher cellular uptake and transfection efficiency. These findings were validated using GFP were experimentally validated using GFP‐encoding LNPs, establishing a mechanistic link between cellular genotype and functional delivery. This study demonstrates the power of integrating nanoPRISM with multiomic analysis to uncover predictive biomarkers and design rules that would remain inaccessible through conventional low‐throughput screening. Despite its advantages, nanoPRISM has limitations. It primarily assesses cellular association or NP uptake rather than intracellular payload delivery or therapeutic efficacy. Cellular internalization alone does not guarantee biological activity or therapeutic benefit, potentially limiting the predictive accuracy of this screening method. Furthermore, variations in barcoding efficiency across different cell lines could introduce biases, requiring careful interpretation of the obtained data. This method is restricted to in vitro conditions, thus neglecting the complex in vivo tissue microenvironment. Future improvements could involve integrating barcoding approaches that can measure therapeutic payload and intracellular delivery while mimicking physiologically relevant in vitro models to improve the predictive accuracy of nanoPRISM.

Nucleic acid barcoding combined with next‐generation sequencing (NGS) allows a high‐throughput multiplexed evaluation of LNP delivery within a single animal^[21, 36, 64, 79, 80, 83, 84, 85, 86, 87, 88^] (Figure5b). By sequencing unique barcodes associated with each LNP formulation, NGS allows the parallel analysis of multiple formulations in a single experiment, significantly enhancing screening efficiency. This high‐throughput technique follows four key steps: 1) LNPs are formulated with barcoded DNA (b‐DNA) or mRNA (b‐mRNA) using microfluidic mixing, assigning each formulation a unique nucleic acid identifier; 2) pooled LNPs are administrated into a single animal at a fixed nucleic acid dose; 3) at predetermined time points, target tissues or cells are collected, barcode sequences are extracted and polymerase chain reaction (PCR) amplification is performed to ensure sufficient quantity of barcode sequences for analysis; and 4) deep sequencing with NGS to quantify LNP distribution in different tissues (Figure 5b). The sequencing data are normalized across different tissues within the same animal, providing a relative quantification of LNP accumulations in the cells or tissues of interest.^[79, 85, 87^] For example, if the sequencing reads for barcode 1, barcode 2, and barcode 3 in a given cell type are 5000, 3000, and 2000, the normalized deliveries of LNP 1, LNP 2, and LNP 3 are 50%, 30%, and 20%, respectively.

Nucleic acid barcodes can be subdivided into different segments (Figure 5a). b‐DNA consists of two universal primer binding sites, a barcode region unique to each LNP composition, and a unique molecular identifier (UMI). The universal sites, shared across all barcodes, serve as primer binding sites for initiating PCR amplification^[64, 85, 86^] and undergo phosphorothioate modifications to increase barcode stability and decrease exonuclease degradation.^[79, 85^] The barcode region ensures that all LNPs with the same composition carry the same barcode (Figure 5b). For instance, LNPs with composition 1 contain barcode sequence 1, LNPs with composition 2 contain barcode sequence 2, and so on. UMIs, in contrast, are randomly generated sequences incorporated into each barcode molecule so that each molecule can be uniquely identified.^[79, 83, 86, 89^] This allows one to distinguish true molecule counts from PCR amplification artifacts and correct for unequal amplification across barcodes. Because each UMI is counted only once regardless of amplification, the absolute number of original barcodes is preserved (Figure6).^[90^] This prevents overrepresentation of barcode sequences from PCR bias, enabling more accurate quantification of LNP distribution.

b‐mRNA has a more complex design with five segments: a functional mRNA region, a barcode region, a UMI, a PCR docking site, and a long chain of adenosine called the poly(A) tail (Figure 5a).^[86^] The functional mRNA sequence is translated into protein, while the barcode tracks delivery, and UMI controls PCR amplification bias. The PCR docking site acts as a universal primer‐binding site for PCR amplification.^[91^] The poly(A) tail supports mRNA translation and improves its stability by restricting exonuclease access, and provides a binding site for reverse transcriptase to generate complementary DNA (cDNA) for subsequent PCR amplification.^[86, 92^]

Nucleic acid barcoding‐based in vivo HTS approach offers significant advantages over conventional in vivo screening. First, multiplexed screening reduces the number of animals required to screen hundreds of formulations compared to the larger animal cohorts required in low‐throughput techniques, screening only a few LNPs at a time. Second, PCR amplification enables high‐sensitivity detection of nucleic acid doses as low as 0.0001 mg kg⁻¹,^[85^] minimizing the amount of LNPs required for screening, thereby reducing both time and cost. Conventional in vivo studies, by comparison, have lower limits of detection and require larger amounts of LNPs per formulation.

In recent years, nucleic acid barcoding has led to important insights into the relationship between LNP design and site‐specific accumulation. For example, subtle modifications in LNP composition, ranging from changes in individual constituents to subtle molecular‐level alterations, influence their biodistribution and delivery efficiency.^[21, 36, 64, 79, 80, 83, 84, 85, 87, 88^] While it is widely assumed that changes in LNP composition influence their size and hence delivery, in vivo HTS studies have revealed no correlation between LNP size and delivery efficacy.^[87, 88^] This finding suggests that size may not be a confounding factor for LNP delivery, reinforcing the validity of observed structure‐function relationships. Traditionally, ligand conjugation has been used for selective targeting. Strong evidence suggests that lipid structure and LNP composition alone can influence tissue‐specific accumulation without targeting ligands.^[93, 94, 95, 96, 97^] The integration of nucleic acid barcoding and NGS has enabled rapid identification of several key parameters affecting site‐specific LNP accumulation in a high‐throughput manner.^[65, 80, 84, 85, 88^] For example, LNP biodistribution is significantly affected by the administration route,^[83^] ionizable lipid,^[64, 83, 88^] PEG‐lipid,^[21, 64, 79, 84, 88^] helper lipid,^[79^] cholesterol,^[87^] and constituent molar ratios.^[64, 79, 83, 85, 88^] A remarkable discovery is that molecular‐level modifications to ionizable lipids, such as changing the amine head of an ionizable lipid, appear to significantly impact biodistribution.^[64^] These promising results highlight the potential of nucleic acid barcoding to systematically identify key determinants of LNP delivery, establishing the rational design rules for efficient delivery systems.

Beyond identifying LNP biodistribution, nucleic acid barcoding can also uncover the molecular pathways that govern LNP tropism. The protein corona itself can be leveraged to control LNP organ tropism,^[98, 99^] a property that can be tuned through LNP composition modifications.^[93, 100, 101, 102^] For example, certain LNPs have been proposed to enter hepatocytes via apolipoprotein E (ApoE) and low‐density lipoprotein receptor (LDLR) mediated endocytosis.^[103^] However, alternative pathways have also been identified; serum albumin, for instance, can facilitate LNP uptake via an ApoE‐independent pathway.^[104^] A study investigating the LAMP1 gene silencing by siRNA‐carrying LNPs in wild‐type as well as ApoE, LDLR, and very low‐density lipoprotein receptor (VLDLR) knockout mice showed that LNPs can achieve non‐hepatocyte tropism through a mechanism independent of LDLR, VLDLR, or APoE.^[88^] These findings demonstrate the effectiveness of nucleic acid barcoding in elucidating the molecular mechanism driving LNP tropism across diverse formulations.

Recently, the FIND (Fast Identification of Nanoparticle Delivery) was developed by combining DNA barcoding with Cre mRNA to assess functional mRNA delivery in addition to biodistribution.^[79^] In this approach, successfully transfected Ai14 cells undergo Cre‐mediated recombination, triggering robust tdTomato fluorescence that colocalizes with barcoded DNA to quantify functional delivery.^[79, 106^] Although FIND measures functional delivery more accurately than DNA barcoding alone, it cannot identify which specific LNP mediates mRNA translation in a given cell. When multiple LNPs enter the same cell, any one of them (or several in combination) may be responsible for the observed fluorescence. mRNA barcoding addresses this by linking barcode identity to the translatable mRNA sequence while maintaining structural similarity to native mRNA,^[86^] potentially predicting delivery more accurately. Yet, successful delivery of barcoded mRNA does not guarantee its translation into functional protein, and this approach is not suitable for gene silencing or knockdown efficiency. Furthermore, FIND only reports tissue‐level data from the bulk‐sorted tdTomato+ cells, capturing cell‐type‐level differences but not heterogeneity within those populations. Other limitations include fluorescent signal saturation and reliance on the transgenic Ai14 mouse model, restricting its applicability in larger animal or disease models where Cre‐Lox reporters are not available.^[79, 105^]

To overcome the limitations of transgenic Cre‐Lox systems and improve translational relevance, more versatile platforms such as SANDS (Species‐Agnostic Nanoparticle Delivery Screening) and SENT‐seq (Single‐cell Nanoparticle Targeting‐sequencing) have been developed. Traditional in vivo HTS relies on rodent models, which are genetically and physiologically distinct from primates, leading to species‐dependent differences in LNP biodistribution and cellular uptake that limit the predictability for non‐human primates (NHPs) and humans.^[107, 108, 109, 110^] The SANDS platform addresses this gap by evaluating functional LNP delivery across humanized, primatized, and murinized liver types within the same animal model, without requiring transgenic fluorescence reporters.^[107^] In this approach, each LNP co‐delivers both b‐DNA and mRNA encoding glycosylphosphatidylinositol (GPI)‐anchored camelid single‐domain antibody (referred to as aVHH). Following pooled LNP administration into mice engrafted with human, NHP, or native hepatocytes, aVHH+ cells are isolated via FACS (using a fluorescently tagged anti‐VHH antibody that binds aVHH expressed on the cell surface). Sequencing the barcodes in these cells reveals species‐dependent differences in functional delivery between human, primate, or mouse hepatocytes. SANDS data have shown that rodent models often overestimate or misrepresent LNP efficacy in primates and humans, likely due to endocytic and mRNA translation pathways. This platform also highlighted that PEG‐lipid tail length and ionizable lipid type affect delivery in a species‐specific manner, emphasizing the need for tailored LNP design criteria. While humanized animal models have become more widespread for LNP screening,^[111, 112, 113, 114^] their integration into high‐throughput workflows remains limited. Furthermore, like FIND, SANDS reports tissue‐level data only with limited cellular resolution.

The SENT‐seq platform goes a step further by achieving single‐cell resolution without the need for transgenic mouse models. Like SANDS, each LNP carries both b‐DNA and aVHH‐encoding mRNA. After pooled administration, the liver is harvested and dissociated into a single aVHH+ cell in suspension. Rather than requiring a transgenic fluorescent protein, SENT‐seq uses magnetic polymer beads conjugated to oligonucleotides with two orthogonal capture sequences: one for b‐DNA, and another for endogenous mRNA and DNA‐tagged anti‐aVHH antibodies.^[115^] These beads simultaneously label each aVHH‐expressing cell, capture the co‐delivered b‐DNA, and record a sample of the cell's mRNA transcriptome. Subsequently, single‐cell RNA sequencing (scRNA‐seq) provides three readouts: i) each cell identity and state from endogenous mRNA profiles, ii) functional delivery from DNA‐tagged antibody readouts, and iii) evidence of mRNA translation from detection of the aVHH protein. Using SENT‐seq, Dobrowolski et al., identified liver cell subpopulations with high or low LNP uptake and linked cell‐intrinsic factors, such as CDK13 and CDK14 expression, to effective mRNA delivery and translation.^[115^] More importantly, its application to NHPs showed that LNPs performing well in mice often exhibit distinct biodistribution and delivery profiles in primate cells,^[110^] highlighting the relevance of cross‐species functional datasets to guide clinical translation.

Complementary to single‐cell sequencing, spatial transcriptomics (ST) provides microenvironment‐level resolution for mapping LNP delivery within intact tissues. By preserving tissue architecture while profiling gene expression in situ, ST can reveal which cell types and regions within organs receive functional delivery. For example, a recent high‐resolution in situ transcriptomic analysis showed that intravenous LNP‐mediated mRNA delivery in the brain occurred predominantly in endothelial cells, especially those adjacent to vascular smooth muscle cells in blood vessels.^[116^] This suggests that local microenvironment features (such as proximity to vasculature) can strongly influence LNP uptake. Combined with barcoding, ST can map where functional delivery events happen within the tissue, directly linking LNP performance to specific anatomical regions. In hepatocellular carcinoma models, ST and scRNA‐seq together revealed how siRNA‐loaded LNPs reprogrammed the specific tumor region, activating immune interferon signaling and promoting immune infiltration.^[117^] As single‐cell and ST approaches gain wider adoption, they are reshaping how targeting specificity is evaluated. ST in particular can complement barcoding techniques such as SANDS and SET‐seq by resolving LNP delivery at single‐cell and sub‐tissue resolution that goes beyond traditional tissue‐level biodistribution readouts.

The choice of nucleic acid (e.g., siRNA or mRNA) cargo itself also influences the size and structure of LNPs as well as the spatial distribution of various components, such as cholesterol and helper lipids,^[118, 119^] thereby affecting their in vivo delivery. Guimaraes et al., demonstrated that replacing b‐mRNA with b‐DNA resulted in changes in LNP hydrodynamic diameter and polydispersity index (PDI), which in turn influenced in vivo delivery efficiency.^[86^] Given that b‐DNA is orders of magnitude smaller than mRNA, this size discrepancy likely contributed to changes in LNP size. However, this study assessed the delivery of b‐DNA and b‐mRNA separately but not the codelivery of b‐DNA and functional mRNA. In contrast, another study evaluating co‐delivery of functional mRNA and b‐DNA found that incorporating b‐DNA did not significantly alter LNP properties.^[64^] This suggests that the presence of a smaller b‐DNA alongside a larger mRNA does not drastically affect LNP size. Furthermore, a strong correlation was observed between b‐DNA and siRNA delivery, where LNPs with low b‐DNA accumulation in the liver also displayed lower gene silencing efficiency with siRNA.^[85^] This may be due to the comparable sizes of b‐DNA and siRNA,^[86^] minimizing cargo‐induced alterations in LNP size. However, the co‐delivery of b‐DNA and siRNA could still influence LNP biodistribution. To date, only one study has assessed the effects of b‐mRNA. Further work is required to determine the effects of b‐DNA vs b‐mRNA on LNP properties and in vivo fate. More importantly, the choice of nucleic acid barcode could be critical in predicting LNP delivery efficiency. While the size of b‐DNA is comparable to siRNA and sgRNA, it is substantially smaller than mRNA.^[86^] Thus, choosing a barcode with minimal influence on LNP properties will be essential for improving the accuracy of delivery efficacy predictions.

DNA and mRNA barcoding present several key limitations in assessing LNP delivery (Table2). A major challenge of DNA and mRNA barcoding is their inability to differentiate between LNPs that have functionally delivered a therapeutic payload, those that remain on the cell surface, and those internalized but lack functional activity (Figure7a). FIND, SANDS, and SENT‐seq cannot definitively attribute a functional delivery event to a single LNP when many formulations are tested in parallel. While mRNA barcoding may provide a more accurate measure of functional delivery compared to FIND or other barcoding methods, its validation has been limited to a small number of LNP formulations.^[86^] This raises the possibility that certain LNP formulations may accumulate at a target site without being translated into b‐mRNA. Ultimately, LNP efficacy should be evaluated based on both biodistribution and functional delivery rather than each factor separately. Given the inherent size differences between b‐DNA, b‐mRNA, siRNA, and sgRNA, the barcode type must be carefully selected according to the application to minimize unintended effects on LNP properties. Furthermore, some degree of bias from PCR or sequencing is inevitable. More broadly, these barcoding strategies apply only to NPs compatible with nucleic acid delivery, as they must accommodate barcodes without altering their properties. For NPs incompatible with nucleic acid delivery, alternative high‐throughput screening methods are required.

A peptide barcoding platform encoded by mRNA is a high‐throughput screening strategy to assess functional LNP delivery by tracking mRNA translation into proteins rather than just measuring mRNA biodistribution. Unlike nucleic acid barcoding, which identifies where LNPs accumulate, peptide barcoding provides a direct readout of functional protein expression in target cells (Figure 7a).^[105^] This approach overcomes a major limitation of FIND and other nucleic acid barcoding strategies, which detect LNP presence but do not confirm whether mRNA translation has occurred. In this approach, each LNP is barcoded with an mRNA encoding a unique peptide tag. Upon successful delivery and translation, this mRNA produces a distinct peptide sequence that serves as a marker for functional delivery. The relative abundance of each peptide barcode is then quantified using liquid chromatography followed by tandem mass spectrometry (LC‐MS/MS).

To design the platform, LNPs are formulated with mRNA‐encoding peptide barcode proteins. These proteins consist of two components: a carrier protein, monomeric streptavidin to reduce background noise from miscellaneous proteins, and a variable barcode sequence with an affinity tag fused to the carrier protein through a TEV protease cleavage site (Figure 7c). Similar to the previous nucleic acid barcoding method, the barcoded LNPs are pooled and administered to a single animal. After extracting and processing tissues, lysates are produced, and peptide barcode proteins are purified from the lysate by immobilization on biotinylated beads. After washing, the beads are treated with TEV protease to release the peptide barcodes, which are filtered and analyzed using LC‐MS/MS to quantify the relative amounts of each peptide and, hence, LNP delivery efficiency.

Peptide barcoding presents various advantages and limitations (Table 2). Unlike FIND, which may overestimate functional delivery by detecting LNP accumulation without confirming mRNA translation, peptide barcoding provides a direct measure of protein expression in target cells. By pairing RNA sequencing with peptide barcoding, both biodistribution and protein expression in target sites (or LNP delivery) can be evaluated in a single experiment, improving predictive accuracy over sequential nucleic acid barcoding and luciferase mRNA validation. Furthermore, peptide barcoding can be applied in any preclinical disease model where the targeted cell type can translate exogenous mRNA into functional protein.

Although peptide barcoding can detect hepatic protein expression at doses as low as 0.001 mg kg⁻¹, its sensitivity may be lower than that of DNA or mRNA barcoding due to the lack of a signal amplification step.^[105^] Initial experiments show that quantification of splenic expression of peptide barcodes is less reliable compared to liver lysates, highlighting the need for further improvement of the assay.^[105^] Since peptide barcoding relies on LC‐MS/MS to detect protein expression, it is unsuitable for gene knockdown or silencing applications, where DNA or mRNA barcoding remains preferable.

A key consideration is the potential for non‐linear effects when pooling LNPs with different ionizable lipid species, which may lead to differences in delivery efficiency compared to individual formulations.^[101, 105^] Specifically, Rhym et al., found that the ranking of the top three LNPs from peptide barcoding (pooled setting) did not entirely match hEPO assay results (individual setting). However, the top‐ranked LNPs consistently outperformed the original formulation, and the lowest‐ranked LNP produced the least protein.^[105^] Moreover, variation in mRNA translation efficiency across cell types could influence functional delivery predictions‐ an issue inherent to all nucleic acid‐based barcoding strategies.^[105, 120, 121^] Nonetheless, with broader adoption and improvements, peptide barcoding has the potential to surpass nucleic acid barcoding in evaluating in vivo nucleic acid delivery. Future advancements in proteomic techniques could enhance the sensitivity and precision of peptide barcode quantification, potentially making this approach superior to nucleic acid barcoding for assessing functional delivery.^[122^]

Dynamic light scattering is a widely adopted analytical technique for evaluating the size, size distribution, surface chemistry, and stability of LNPs. These physicochemical properties directly influence the pharmacokinetic profiles, biodistribution, and therapeutic efficacy of LNPs. DLS measures the hydrodynamic size of NPs in solution by analyzing the intensity fluctuations of scattered light arising from Brownian motion. This hydrodynamic size accounts for both the core particle size and any attached surface molecules. Tracking size changes over time provides insights into LNP aggregation behavior and stability, both of which are critical for maintaining consistent therapeutic performance. Unlike imaging‐based methods such as cryo‐EM, which require extensive imaging processing and individual particle analysis to measure the average size of LNPs, DLS offers a faster and more streamlined assessment of LNP formulation. Although DLS assumes particles are spherical, and this can introduce minor inaccuracies, this limitation is generally negligible since LNPs are predominantly spherical in structure.

Conventional DLS, limited by single‐sample cuvette measurements, restricts high‐throughput characterization of LNPs. The multi‐well plate reader (Wyatt Technology, CA, USA) overcomes this limitation by offering high‐throughput DLS technology compatible with microplates of 96, 384, or 1536 wells.^[36, 79, 84, 94, 141, 142^] This is a technique performed by diluting different formulations of LNPs in PBS to the same concentration and transferring the formulations into individual wells, followed by microplate loading. A mobile laser illuminator and detector measure scattered light intensity from below each well, providing data on hydrodynamic size, size distribution, and stability over time for each sample (Figure9). This high‐throughput approach enables the collection of large datasets in a single run. This system can be interfaced with robotic liquid handling and automated plate loading systems, eliminating the need for manual intervention (Figure 9).^[141, 142, 143^] This automation not only improves throughput but also reduces human error and variability, which is essential for consistent data quality. A key advantage of the multi‐well plate reader DLS system is its temperature‐controlled operation, allowing rapid screening of LNP formulations under various storage or physiological conditions. This is particularly useful for investigating LNP stability and aggregation behavior under varying temperature conditions, which is critical for clinical translation. Furthermore, HT‐DLS enables rapid identification of LNP formulations outside of the desirable size range and polydispersity criteria, which are key factors in vivo performance. Oversized or highly polydisperse LNPs are more susceptible to rapid clearance by the mononuclear phagocyte system, limiting circulation time, biodistribution, and therapeutic efficacy. For example, the DynaPro Plate Reader was used to screen 128 LNP formulations in a single run, eliminating 63 candidates that exceeded the target size range (20–200 nm), showed a high polydispersity index (>0.5), or displayed multi‐modal scattering profiles.^[79, 84, 94^] This allowed the early elimination of suboptimal formulations, thereby streamlining subsequent screening steps and increasing overall experimental efficiency. Beyond size‐based filtering, HT‐DLS has been used to investigate how PEG‐lipid content, total lipid concentration, and nucleic acid loading affect LNP uniformity and stability. In one study, HT‐DLS performed in a fully automated 96‐well format revealed that increasing PEG‐lipid content reduced LNP size but increased polydispersity. An optimal formulation containing 1.5 mol% PEG‐lipid yielded a unimodal size distribution with high encapsulation efficiency.^[141^] Importantly, size and polydispersity index trends identified using HT‐DLS in 96‐well format were strongly correlated (R² >0.9) with data from scale‐up data from microfluidic synthesis, highlighting the predictive power of HT‐DLS in guiding formulation development. Similarly, Sarode et al., also used HT‐DLS to identify optimal formulation for nucleic acid delivery from a library of 54 antisense oligonucleotides (ASOs) loaded LNPs with varying PEG‐lipid composition.^[144^] The resulting dataset revealed clear correlations between PEG content and LNP diameter, and that more monodisperse LNPs were observed at higher PEG molar ratios. The results also showed that LNPs sized between 100 and 150 nm exhibited the highest mRNA knockdown in primary neurons. These finding links HT‐DLS‐based size selection to functional delivery outcomes. Collectively, these studies demonstrate that HT‐DLS serves as a powerful pre‐screening filter, enabling early elimination of suboptimal candidates before functional evaluation, and thereby accelerating the rational LNP design.

Nevertheless, the multi‐well plate reader‐based DLS technology lacks integrated electrophoretic light scattering (ELS) capabilities, limiting simultaneous measurement of surface charge–an important parameter of colloidal stability in a biological environment. Future technological advancements should focus on integrating DLS with ELS within a single high‐throughput analytical platform that enables concurrent characterization of size, stability, and surface charge.

Small‐angle scattering (SAXS) has become an essential analytical method for elucidating the internal structure of LNPs, providing insights into their nanoscale organization, including phase behavior, the degree of crystallinity, and core–shell architectures.^[14, 133, 145, 146, 147, 148, 149, 150, 151^] By probing variations in electron density within the LNP, SAXS generates scattering profiles that indicate the presence of ordered domains (e.g., lamellar or inverse hexagonal phase) through distinct diffraction peaks, while disordered or amorphous regions appear as broader and less defined scattering features. In addition to internal structure, SAXS can also evaluate the size and stability of LNPs under varying environmental conditions, as well as the loading and the spatial distribution of therapeutic agents.^[152^] In a typical SAXS experiment, exposing a sample to X‐rays that are scattered to different degrees due to electron density variations within the LNPs generates scattering profiles with distinct peaks and shoulders.^[153, 154^] The resulting scattering intensity (I) profile is plotted against the scattering wavevector (q), where q = (4π sin θ)/λ, with 2θ representing the scattering angle and λ the wavelength of the X‐rays (Figure10a).^[14^] The inverse relationship between q and the characteristic length of the scattering object (d), given by q = 2π/d, enables SAXS to resolve nanoscale structural features such as interlayer spacing, lipid ordering, and core–shell architecture.^[148^] Advanced model fitting of scattering profiles further allows the quantification of parameters such as shell thickness, internal lattice spacing, and radial density distributions, thereby linking structural features to LNP stability and functionality.^[155, 156, 157^] When complemented by real‐space imaging methods such as cryo‐EM, SAXS profiles can be validated and used to build robust structural models (Figure 10b).^[135^]

Historically, SAXS was a relatively low‐throughput technique. Each sample had to be loaded and measured individually, often at specialized synchrotron beamlines or using lab‐scale instruments with lengthy exposure times. Recent advances in high‐throughput SAXS (HT‐SAXS), such as the SIBYLS beamline 12.3.1 at the Lawrence Berkeley National Laboratory, have accelerated the structural analysis of LNPs.^[158, 159^] Unlike conventional SAXS, which processes single samples per execution,^[14, 133, 149, 160^] HT‐SAXS can analyze 96 samples in less than 4 h using microplate‐based sample handling (<30 µL per well).^[158, 160, 161^] This platform incorporates a liquid‐handling robot that automates sample loading, exposure, retrieval, and cleaning cycles; significantly reducing both data acquisition time and human error compared to conventional SAXS workflows (Figure 10c).^[158, 160^] HT‐SAXS capabilities are available at several synchrotron facilities worldwide. The Australian Synchrotron houses the BioSAXS beamline, equipped with a liquid handling system (Coflow Autoloader) that enables high‐throughput analysis across a broad particle size range (≈0.001–3 Å⁻¹) while maintaining low background noise.^[162^] Another HT‐SAXS facility is the Synchrotron P12 beamline of the European Molecular Biology Laboratory (EMBL) located at the PETRA III storage ring (DESY, Hamburg),^[163, 164, 165^] which can collect scattering data with exposure times as short as 30–50 ms, spanning structural features from 0.5 to 300 nm. It also employs robotic sample changers and an integrated pipeline (SASFLOW) for intermediate data processing, delivering Guinier analyses and low‐resolution models within minutes. The facility allows researchers to send samples for analysis and interpretation.^[161, 166^]

The significance of HT‐SAXS in LNP screening is demonstrated by a recent study directly linking LNP internal structure to functional outcomes. Hammel et al., used HT‐SAXS (SIBYLS beamline 12.3.1) to demonstrate a direct correlation between the internal structure of antisense oligonucleotides (ASOs)‐loaded LNPs and their gene silencing activity.^[14^] By screening a large library of 54 LNP formulations (varying in PEG‐lipid content) using HT‐SAXS, they revealed important structural differences in ASOs‐loaded LNP formulations, notably the presence of hexagonal (H_II) and lamellar (L_α) phases, as well as distinction between ordered (crystalline) and disordered or fluid‐like behavior (Figure 10b, left). These structural characteristics, initially observed in SAXS profiles, were cross‐validated using cryo‐EM imaging of selected formulations (Figure 10b, right), confirming the presence of hexagonal lattices or multilamellar structures as indicated by the scattering patterns. Importantly, the study showed that the ratio of disordered to ordered lipid phase, as quantified from the SAXS profiles, was strongly predictive of gene silencing activity. Formulations with lower disordered/ordered ratios (i.e., more highly ordered ones) achieved greater mRNA knockdown, whereas those with predominantly disordered interiors were less effective. By screening many formulations, the study identified that certain PEG‐lipid compositions produced a highly ordered internal structure, which corresponded to the best‐performing formulations in biological assays. This highlighted HT‐SAXS as not just a structural screening tool, but a powerful method to predict and optimize LNP biological performance based on the internal architecture (structure‐function relationship). In practical terms, this can reduce the burden of in vivo testing. Instead of injecting dozens of candidates to see which works best, one can first use HT‐SAXS (alongside in vitro assays) to narrow the field to a few promising LNPs with optimal internal architecture for efficacy.

Despite its advantage, challenges remain in SAXS data interpretation, particularly in cases where broad NP size distributions and the coexistence of multiple internal phases can introduce signal variability,^[14, 154^] complicating the interpretation of subtle structural differences. Moreover, SAXS peak assignments may vary between studies due to differences in model assumptions and sample conditions.^[149, 150^] To address these issues, it is important to employ model‐based fitting strategies and complement SAXS with orthogonal techniques such as cryo‐EM or DLS. Cryo‐EM provides direct visualization of NP morphology and internal structure, confirming phase identities obtained from SAXS, while DLS serves as a rapid pre‐screening tool to assess NP size and aggregation status before SAXS measurement. Machine learning models trained on simulated and experimental SAXS data can enhance data reliability and accuracy.

Microplate‐based spectroscopic methods, particularly UV–vis and fluorescent assays, have emerged as indispensable tools for rapid and high‐throughput characterization of LNPs. These techniques greatly enhance the efficiency of iterative LNP formulation optimization. They allow rapid assessment of several samples, facilitated by automated liquid handling and multi‐well microplate readers (ranging from 96 to 1536 wells) and automated liquid handling systems to facilitate the rapid processing of large sample sets. These tools provide insights into the physicochemical characteristics of LNPs, such as encapsulation efficiency, cargo release kinetics, stability (or aggregation state), surface chemistry, and size and size distribution.^[141^] UV–vis spectrophotometry operates on the principle of molecular absorption at specific wavelengths, which is unique to different therapeutic molecules. For instance, nucleic acids exhibit a peak absorbance of ≈260 nm, enabling their quantification with high sensitivity.^[36, 141^] This method is particularly useful for determining encapsulation efficiency (EE), which is the ratio of encapsulated material to total (free and encapsulated) material, given by the formula EE = (Amount of encapsulated material/Total amount of material) × 100%. In practice, unencapsulated molecules are separated from LNPs (e.g., dialysis or centrifugation), and UV–vis absorbance is measured to determine EE. The method is easily scalable for high‐throughput characterization, enabling the analysis of up to 1536 samples per plate to guide the iterative refinement of LNP formulation with large EE.

Advances in multi‐mode microplate readers have further expanded the capabilities of high‐throughput LNP characterization, allowing detection across multiple spectroscopic modalities, including absorbance, fluorescence, time‐resolved fluorescence, and luminescence. Fluorescence‐based detection in multimode readers is also widely employed to quantify encapsulation efficiency and release of payloads in LNPs using different fluorescent dyes (Figure 8).^[167, 168^] The fluorescence intensity of hundreds of samples can be measured in a single run.^[167^] The RiboGreen RNA assay has been widely used to determine RNA encapsulation efficiency in LNP formulations, using selective fluorescent signal enhancement of RiboGreen upon binding to nucleic acid.^[36, 52, 85, 134, 150^] In this assay, free RNA is measured by directly adding RiboGreen solution to buffer‐diluted LNPs, and total RNA is determined by adding Triton X‐100 solution, which destabilizes LNPs to release encapsulated RNA.^[168^] Alternatively, LNP samples can first be dialyzed to isolate free RNA before the assay, then treated with Triton X‐100 solution to release the encapsulated RNA.^[150^] The total amount of RNA can be obtained by summing the free and encapsulated material. The quantity of nucleic acids is then measured using standard calibration curves with known concentrations of nucleic acid solutions.^[52, 150^] Furthermore, the encapsulation efficiency is calculated as a percentage of encapsulated to total nucleic acids. A similar approach has been applied to quantify oligonucleotide encapsulation using the OliGreen ssDNA assay.^[14, 139, 169^] These fluorescence‐based assays offer high sensitivity (ng/mL detection limits) and compatibility with high‐throughput processes, making them an efficient method for iterative formulation refinement.^[170, 171^]

Recent advancements have led to the introduction of the single‐particle profiler (SPP), a high‐throughput analytical platform developed for the characterization of LNPs at the single NP level, ranging from 5 to 200 nm.^[140^] Unlike conventional methods that provide population‐averaged data, SPP uses fluorescence fluctuation intensity analysis, whereby individual fluorescently labeled LNPs diffuse through a confocal volume, allowing measurement of fluorescence intensity fluctuation and diffusion coefficients. When combined with a 96‐well microplate and an automated stage movement imaging program, the SPP technique facilitates simultaneous analysis of thousands of individual LNPs per well, allowing systematic multiparametric screening of size, encapsulation efficiency, and membrane fluidity. These properties are measured across two profiling modes, each requiring 15 s per technical replicate, with 40 replicates per well or sample, ultimately resulting in a total measurement time of ≈20 min per well/sample. Size estimates are derived by comparing DLS size measurements with diffusion coefficients measured by the SPP and fitting to the Stokes–Einstein equation. The number of full and empty LNPs is determined by measuring the co‐occurrence of mRNA and dyed‐lipid signals for each particle, while membrane fluidity is assessed by analyzing spectral fluctuations of the lipid‐dye. This technique with single NP level resolution significantly improves the precision and depth of characterization datasets, streamlining formulation optimization. Despite demonstrating high reliability in measurements such as LNP size (which fits the Stokes–Einstein equation) and mRNA encapsulation (with co‐occurrence approaching 100%), further cross‐validation with other characterization tools should be performed to confirm the technical capabilities of the SPP.

It should be noted that conventional quantification methods are limited in their applicability for high‐throughput quantification since they require the use of complex and time‐consuming sample preparation, such as separating unloaded molecules from cargo‐loaded LNPs. However, recent advances in machine learning‐driven spectroscopic data analysis have enabled nucleic acid quantification without separation steps. For example, Fan et al., introduced a locally weighted regression model to directly analyze full UV–vis spectral data from microplate readers, enabling accurate nucleic acid quantification without prior payload separation.^[139^] This strategy not only simplifies the analysis but also reduces sample handling errors and processing time, improving overall analytical reproducibility.

In addition to payload quantification, microplate‐based UV–vis spectroscopic measurements can assess LNP stability and aggregation states by monitoring optical density (OD) changes at wavelengths between 400 and 600 nm. A stable dispersion maintains constant OD values, whereas aggregation or fusion events induced by salts (e.g., Ca²⁺ and Mg²⁺) are readily detected by increased OD.^[172, 173^] Furthermore, it has been shown that the size of LNPs may also be evaluated by light scattering, with smaller LNPs displaying high absorption at lower wavelengths and larger LNPs scattering more light, shifting the absorbance peak to a higher wavelength.^[174^]

Another critical parameter, the apparent acid dissociation constant (pKa) of ionizable LNPs, influences LNP surface charge, stability, cellular uptake, and endosomal escape efficiency.^[128^] The term "apparent pKa" is used since the measured pKa value of LNPs differs from the intrinsic pKa of the same lipids in the solution. This shift occurs due to the complex microenvironment within LNPs, where factors such as lipid packing, surface charge, and interaction with surrounding lipids or encapsulated cargo can influence protonation behavior. To achieve the maximum therapeutic index, the apparent pKa value of LNPs should be optimized. The apparent pKa of LNPs can be measured in a high‐throughput manner using the 2‐(p‐toluidino)‐6‐naphthalenesulfonic acid (TNS) binding assay.^[36, 64, 84, 134^] In the free aqueous solution, the TNS molecule is non‐fluorescent; upon coming into contact with a cationic lipid species, it fluoresces.^[126, 175^] As TNS molecules potentially bind to the surface of LNPs without significant penetration into the inner compartments, values derived from this assay likely indicate the pKa properties of accessible ionizable lipids at the LNP surface, which is the apparent pKa.^[22^] LNPs have been tested in buffer solutions with varied pH. After mixing buffer solutions, LNPs, and TNS solution in 96‐ or 384‐well plates and incubating under gentle shaking, the fluorescent intensity was measured with a microplate reader at excitation wavelengths of ≈325 nm and emission wavelengths of 435 nm.^[84, 128^] By locating the inflection point on the generated fluorescence against pH curve,^[128, 176^] the apparent pKa of LNPs can be determined. Alternatively, the zeta potential of LNPs measured at different pH levels can be used to determine the pKa at the inflection point (Figure11). Nevertheless, the low data acquisition rate of electrophoretic light scattering (ELS) makes high‐throughput zeta potential measurement difficult. The TNS assays, in contrast, have been adapted for high‐throughput formats through the integration of acoustic liquid handling systems. For instance, Cui et al., demonstrated a streamlined workflow using a 384‐well plate format and the ECHO 550 system to automate LNP pipetting, pH titration, and TNS addition, significantly improving assay throughput.^[36^] Furthermore, previous studies have shown that TNS‐derived pKa values align more closely with the theoretically predicted pKa values than those obtained from zeta potential measurement.^[20^] Given its higher sensitivity, accuracy, and adaptability to automation, the TNS assay is the most reliable approach for determining the pKa of ionizable LNPs.^[20, 175^]