Ionizable Lipids with Optimized Linkers Enable Lung-Specific, Lipid Nanoparticle-Mediated mRNA Delivery for Treatment of Metastatic Lung Tumors

Our primary objective was to design and synthesize new biocompatible ionizable lipids for extrahepatic delivery of mRNA. Over the past few years, we have developed several ionizable lipids for various RNA-based therapeutic and vaccine applications.^{34,41 −48} Lipid 14 stood out as one of the top-performing lipids in our library, with its ester-branched aliphatic tails enabling the development of COVID-19 and bacterial vaccines.^{46 −48} However, lipid 14 could be susceptible to hydrolytic degradation in ethanol and aqueous environments due to its ester linker, which hinders its utility following long-term storage. Thus, the use of different linkers could provide more stability to the structure of the ionizable lipid. Therefore, keeping lipid 14 with its backbone, we designed a combinatorial library of ionizable lipids using different biodegradable linkers (FigureB). In particular, for lipid 14 and lipid 34, an ester linker was used, whereas a reverse-ester linker was employed for lipid 32 and lipid 38. In the case of lipid 33 and lipid 39, a carbonate linker was selected, while an amide linker was used for lipid 34 and lipid 40. For lipid 35 and lipid 41, a urea linker was employed, whereas a reverse-amide linker was used for lipid 36 and lipid 42. All lipids were synthesized using standard organic synthesis procedures and underwent characterization by NMR and mass spectroscopic techniques (see the Supporting Information↗).

Upon synthesizing the lipids, we investigated their stability in different solvents. To this end, we dissolved the lipids in either ethanol or a water/ethanol mixture (1:3) and monitored their stability for 3 months using a high-performance liquid chromatography (HPLC) device equipped with a charged aerosol detector (CAD). We observed that while the amide (lipid 34), urea (lipid 35), and reverse-amide (lipid 36) linkers were stable and did not undergo hydrolysis in both solvent systems, the ester (lipid 14), reverse-ester (lipid 32), and carbonate (lipid 33) linkers rapidly degraded in the water–ethanol mixture as compared ethanol (FigureC,D, respectively).

LNPs’ composition was optimized by performing a small design of experiment (DoE)-based screening. Briefly, different lipid critical process parameters (CPPs) and critical quality attributes (CQAs) were used to generate a custom design (Figure S1A↗) containing 20 formulations. After formulating and characterizing these LNPs, statistical analysis elucidated how the CPPs influencing LNPs’ features were the different lipid proportions as well as the ionizable lipids’ identity (Figure S1B↗). We used these CPPs to interpolate a mathematical model correlating them to CQAs and simulated 10,000 theoretical formulations using a Monte Carlo simulation. Among these LNPs, a series of thresholds were applied to select the formulations with the highest theoretical performance (top 10%). As shown in Figure S1C↗, the best-performing theoretical formulations presented a very similar composition, with the average composition being ionizable lipid, cholesterol, DSPC, and PEG-DMG in molar ratios of 40:48.5:10:1.5, respectively. Thus, LNPs were formulated with these molar ratios using the synthesized lipids. As model cargo, firefly luciferase mRNA (mLuc) was encapsulated using the NanoAssemblr microfluidic mixing device, as described in the Experimental Section.

The pK_a of LNPs formulated with different lipids was also estimated by using a 6-(p-toluidino)-2-naphthalenesulfonic acid (TNS)-based fluorometric assay. Most of the new lipids had a pK_a between 5.9 and 6.5, which is considered suitable for most ionizable lipid-based LNP formulations (Figure S2↗).⁴⁹ Interestingly, LNPs formulated with lipids containing either urea or reverse-amide linkers (lipids 35, 36, 41, and 42) had a pK_a close to or higher than 7, suggesting that the number and orientation of the amide linkers can influence the lipid environmental response.

Noticeable variations were observed in the hydrodynamic diameter and zeta potential (ζ) of the produced LNPs (FigureE,F). The N-methylpiperazine head-containing lipids (lipid 37–lipid 42) yielded higher-sized LNPs as compared to dimethylamine head lipids (lipid 14 and lipid 32–lipid 36). Also, LNPs formulated with ester, reverse-ester, and carbonate linker-containing lipids (lipid 14, lipid 32, lipid 33, and lipid 37–39) had a negative ζ, whereas LNPs consisting of amide, urea, and reverse-amide-containing lipids had a positive ζ (lipid 34–lipid 36 and lipid 40–lipid 42). The polydispersity index (PDI) was less than 0.1 for all LNPs, making the formulations monodisperse. The encapsulation efficiency was >95% for all the LNP formulations (FigureG), and mRNA encapsulation as well as retention of its integrity was further confirmed by agarose gel electrophoresis (Figure S3↗).

Upon confirming the LNPs’ physio-chemical properties, we screened the formulations’ transfection efficiency in HeLa, RAW 264.7, and HepG2 cell lines by encapsulating mLuc as model cargo (Figure). HepG2 and RAW 264.7 cells were selected to represent the liver and the mononuclear phagocytic system as compartments that are often targeted by nanoparticles after systemic administration. In addition, HeLa cells were selected as a more general, fibroblast-like cell line.

Looking at individual lipids, in HeLa cells (FigureA), all of the LNPs formulated with amide or urea linkers were the best-performing ones. Conversely, in RAW 264.7 cells (FigureB), the best-performing LNPs were the ones formulated with lipid 36, lipid 38, and lipids 40–42, while HepG2 cells (FigureC) were transfected best by lipids 33–36 and lipids 40–42.

However, to elucidate the structure–activity relationship of the new ionizable lipids, it is important to account for how these molecules can influence the LNP features, which in turn can change the particles’ biological behavior, constituting noise factors. Thus, accounting for the effect of ionizable lipids on particles’ properties such as size, apparent pK_a, and ζ would help us to discern the direct effect of these molecules on the LNP activity.

Since it is technically challenging to control for these variables in an experimental setting, we assessed the possible correlations between LNP size, ζ, and Luc signal from the tested cell lines, demonstrating how there was no correlation between the LNP size and their Luc signal in vitro (Pearson’s correlation coefficient (PCC) < 0.4 for all cell lines, Figure S4A↗).

However, the data demonstrated a positive correlation between the ζ and the Luc signal in HeLa and HepG2 cells (PCC = 0.69 and 0.65; p = 0.0132 and 0.0220, respectively) with positively charged LNPs (lipids 34, 35, 36, 40, 41, 42) showing slightly higher transfection efficacy (Figure S4B↗). These observations are reinforced by a similar positive correlation between these two cell lines and the LNPs’ pK_a (PCC = 0.74 and 0.72 and p = 0.0064 and 0.0079, respectively, Figure S4C↗). This can be explained by the higher electrostatic interaction of positive LNPs with the cellular surface, leading to higher uptake.⁵¹ Interestingly, all the lipids with N-methyl-piperazine groups yielded particles with slightly higher transfection efficiency, although the difference was not significant compared to the dimethylamine head lipids (Figure S4D–F↗).

Overall, these in vitro studies allow us to elucidate some fundamental interactions between LNPs and different cell lines, providing an inkling for further in vivo studies.

To test these ionizable lipids in vivo, C57BL/6J mice were systemically injected with a volume of particles corresponding to 10 μg of mLuc per animal via retro-orbital injection. The luminescence signal in major organs was assessed after 6 h (FigureA). Once again, to account for the presence of possible correlations between LNP features and organ biodistribution, we performed the same correlation analysis discussed in the previous section.

Notably, the size of LNPs was reversely correlated with the liver Luc signal, underlining how smaller particles tended to accumulate more in this tissue (PCC = −0.6741, p = 0.0162, Figure S5A↗). Conversely, LNPs with positive ζ (lipids 34–36 and lipids 40–42) showed higher luminescence in the lung (PCC = 0.6345, p = 0.0267) and lower luminescence in the liver (PCC = −0.5815, p = 0.0474) and kidneys (PCC = −0.7378, p = 0.0062, Figure S5B↗). Furthermore, the apparent TNS-measured LNPs’ pK_a values had a strong positive correlation with the lungs’ expression (PCC = 0.8417, p = 0.0006) and a negative correlation with the kidneys’ signal (PCC = −0.6978, p = 0.0116, Figure S5C↗).

When looking at the possible correlations between in vitro and in vivo results, it is interesting to note how the Luc signal in RAW 264.7 cells and the one detected in the animal spleens were positively correlated with each other (PCC = 0.7746, p = 0.0031, Figure S5D↗). From a biological standpoint, this observation could be attributed to the presence of phagocytic cells in the spleen that have somewhat similar behavior to the RAW 264.7 macrophage-like cells. It is also important to note how both the behaviors of HeLa and HepG2 hepatocyte cells were positively correlated with the lung signal (PCC = 0.8843 and 0.8428 and p = 0.0006 and p = 0.0001, respectively, Figure S5E,F↗). Ultimately, these correlations can be easily explained by the LNP positive ζ that increases both cellular expression and lung expression.

Despite these correlations, the positive ζ LNPs formulated using ionizable lipids with amide and urea linkers behaved heterogeneously for both head groups. Lipids 40–42 were not well tolerated by the animals, and some of them displayed obvious behavioral signs of discomfort and had to be sacrificed soon after injection for ethical reasons. This acute adverse reaction could be attributed to the combination of larger LNP size (above 200 nm) derived from the N-methylpiperazine headgroup and positive ζ, which could have resulted in LNP aggregation after injection, causing embolism in the mice. Thus, lipids 40–42 were excluded from further studies. The other positively charged ionizable lipids (lipids 34–36) were well tolerated by animals and showed high lung specificity, with lipids 35 and 36 showing both a slightly higher lung signal (FigureC) and high lung/liver and lung/spleen ratios when compared with ester, reverse-ester, and carbonate linkers (FigureF,G). Since the particles formulated with these three lipids had analogous size and surface charge, the difference between lipid 34 and lipids 35 and 36 can be strictly attributed to the lipid structures themselves. Indeed, the presence of amide groups in the lipid linker resulted in increased LNPs ζ and, therefore, lung transfection. However, lung tropism is further improved by either positioning this nitrogen closer to the dimethyl-amine head (lipids 36) or having nitrogen on both sides of the linker (lipid 35), suggesting the nuanced effect of even minor structural changes on LNP tropism. Thus, lipids 35 and 36 were selected as lead lipids for delivery of mRNA to the lungs.

Lipid 33, despite showing a similar Luc signal in the liver (FigureD) to lipid 14, displayed much lower signals in other organs, especially the lungs and spleen, resulting in a higher specificity in the liver, as highlighted by the low lungs/liver and the high liver/spleen ratios (FigureF,H). Comparing the behavior of lipid 33 to lipids 14, 32, and 37–39, for each of these lipids with dimethylamine heads, their N-methyl-piperazine equivalent always resulted in a lower liver signal (FigureD). This corroborates the effect of LNP size on liver biodistribution since all the piperazine lipids yielded bigger particles than their counterparts. Similarly, lipid 38 displayed higher spleen specificity in the form of a slightly better spleen luminescence (FigureE) and Lung/spleen ratio (FigureG). Thus, lipids 33 and 38 were selected, respectively, as lead lipids for the liver and spleen.

It is especially notable to observe how the combination of a carbonate linker and an N-methylpiperazine head in lipid 39 seems to abolish lipid activity in all the tested cell lines and organs, making it the worst-performing lipid. The mechanism behind this could warrant further investigation and the study of additional organs.

Finally, the heart and kidney signal measured was barely detectable compared to the background radiance, and no clear trend was visible among ionizable lipids (Figure S6A,B↗, respectively).

As presented in Figure S7↗, in which the different lipids are arranged in a ternary plot based on their different proportions of organ radiance, also allows us to evidence some clear trends: all the lipids with ester, reverse-ester, or carbonate linkers are mostly clustered between the liver and spleen, with the dimethylamine lipids among these being the ones oriented toward the liver while the N-methyl-piperazine clustered toward the spleen. All the lipids having amide, reverse-amide, or urea linkers were distributed between the lungs and spleen, with the N-methyl-piperazine lipid accumulation proportionally higher in the spleen compared to the dimethylamine lipids. Furthermore, this graphical summary helps us observe how lipid 33, lipids 35–36, and lipid 38 have the highest proportional signal for the liver, lungs, and spleen, respectively, since they are the ones closer to the vertex of each organ, representing higher luminescence signal. These data make them the most selective lipids in our library for these tissues, confirming them as lead compounds to deliver mRNA toward them for further investigation.

Taken together, these results show how small differences in the linker structure of ionizable lipids can have radical effects on the resulting particles’ transfection and tropism in different organs.

We next investigated the cellular specificity of LNPs’ transfection formulated with the selected leads lipid 35 and lipid 36. To elucidate this, we formulated LNPs with Cre recombinase mRNA (mCre) and injected them in Ai9-tdTomato mice as previously described (Figure S8↗).^5,43 As can be seen in Figure S9A,B↗, the mCre-LNPs displayed analogous size distribution and mRNA encapsulation compared to their mLuc-loaded equivalents. We quantified the functional mRNA delivery as a percentage of tdTomato⁺ cells in different organs 72 h post-systemic administration (FigureA). Using flow cytometry, we quantified successful mRNA transfection and expression in different cell types, including myeloid immune cells (CD45⁺ CD326^– CD11b⁺), dendritic cells (CD45⁺ CD326^– CD11c⁺), endothelial cells (CD45^– CD11b^– CD326^– CD31⁺), and epithelial cells (CD45^– CD326⁺).

In concurrence with the in vivo mLuc expression data, the lungs of mice treated with lipids 35 and 36 resulted in significantly more tdTomato⁺ cells than the lungs of mice injected with lipids 14, 33, and 38. Comparing lipids 35 and 36 for their transfection efficiency in different lung cells, we observed that lipid 35 efficiently reached nonimmune cells, with >60% of endothelial cells and ∼40% of epithelial cells expressing tdTomato (FigureB,C). Albeit to a lesser extent, this lipid also showed higher transfection levels in myeloid cells and dendritic cells (FigureD,E). It should be noted that despite lipids 35 and 36 having similar physicochemical properties and mLuc expression profiles, lipid 35 displayed a significantly higher transfection efficiency in all the analyzed lung cell populations. The lung specificity of lipids 35 and 36 is further confirmed by the complete absence of tdTomato⁺ cells in the spleen and the liver (Figure S10E–M↗) of mice injected with these LNPs. Therefore, we decided to focus our efforts on further assessment of lipids 35 and 36 as lead lipids for mRNA delivery to the lungs. However, lipids 33 and 38 did not confirm their liver and spleen selectively in this model (Figure S10↗).

When comparing the mLuc results with the Cre-tdTomato murine models, lipids 35 and 36 strongly validated the previous mLuc results by showing efficient transfection in all of the lung cell populations analyzed, with lipid 35 showing a remarkably higher transfection compared to all other formulations and across all of the lung cells.

When focusing on the lead formulations, LNPs with higher pK_a resulted in LNPs with higher ζ, which in turn appeared to have higher lung tropism. Indeed, the lung transfection observed in vivo mLuc expression starts to significantly increase only for LNPs with pK_a values above 6.5 and seems to reach a plateau for mLuc transfection of the lungs at a pK_a of 7.2 (lipid 36, FigureE). However, when comparing the mLuc results with tdTomato expression, the percentage of positive cells steeply increased among cell populations for LNPs containing lipid 35 (pK_a = 7.6), compared to ones formulated with lipid 36. Thus, there seems to be a threshold value for the pK_a above which the transfection efficiency in lung cells significantly improved as reported previously.⁵²

We also analyzed the biodistribution of LNPs loaded with fluorescently labeled RNA, assessing the possible correlations between the transfection efficiency observed in the tdTomato model and simple particle accumulation. In this instance, we investigated lipid 14 as a benchmark and lipids 35 and 36 as our main leads for lung targeting. As summarized in Figure S11A–D↗, lipids 35 and 36 accumulated more than lipid 14 in all the lung cell populations analyzed but with a percentage of positive cells that was significantly lower compared to the tdTomato model. Furthermore, unlike the lung data in Figure, there was no clear difference between lipids 35 and 36. Thus, lipid 35 and lipid 36 particles had similar lung targeting capacity. However, lipid 35 displayed higher transfection efficiency in tdTomato animals, suggesting a possible alternative mechanism behind lipid 35 specificity that is not strictly dependent on particle accumulation. When considering instead the LNP biodistribution in the liver (Figure S11E–H↗) and spleen (Figure S11I–M↗), we observed an overall similar accumulation for all three lipids, with lipids 35 and 36 accumulating slightly more in the liver endothelial cells, and at the same time significantly less in the spleen B-cells. Despite these small differences, the different ionizable lipids yielded LNPs with overall similar biodistribution, in contrast with the data in Figure S10↗. This reinforces the possible contribution of a different, perhaps intracellular, mechanism behind the observed transfection selectivity. Furthermore, the comparable liver deposition allows us to exclude first-pass lung accumulation as the mechanism behind transfection in these organs.

Taken together, these results underline how minor changes in the structure of ionizable lipids can radically affect the resulting LNP organ transfection. In particular, the addition of nitrogen atoms to the lipid linker enables the fine-tuning of the LNPs pK_a, which can be leveraged to achieve not only high levels of protein expressions selectively in the lungs without using additional targeting moieties but also a more widespread transfection across all the major lung cell populations. Based on these results, we selected lipid 35 as the definitive lead for lung targeting.

Focusing on the best-performing lipid 35, we extensively tested its biocompatibility after a single bolus intravenous injection. To this end, a volume of LNPs equivalent to 10 μg of mRNA was injected through the retro-orbital vein. To assess a possible acute immune response to LNP injection, mice blood was collected at 2 and 24 h after injection and the concentrations of Monocyte Chemoattractant Protein-1 (MCP-1), interleukin 6 (IL-6), tumor necrosis factor α (TNF-α), and interleukin 10 (IL-10) in mice plasma were measured by ELISA (FigureA). As a control, we selected the clinically approved SM-102 ionizable lipid.

At 2 h postinjection, all cytokines were slightly elevated in the serum. Pro-inflammatory MCP-1, IL-6, and TNF-α were similarly elevated compared to the control SM-102 (FigureB,D,F). The elevation was comparable between lipid 14 and lipid 35 and resulted in significantly higher levels of IL-6 compared to SM-102 but not for any other cytokine. Conversely, the immunomodulatory cytokine IL-10 was only slightly induced by lipid 35 (FigureH). Nevertheless, the increase in serum cytokines was transient as all cytokines dropped to the level of untreated animals within 24 h postinjection, with only a very small amount of MCP-1 detectable (FigureC,E,G,I).

After 24 h of injection, we also measured the concentration of key serum markers, including alkaline phosphatase (ALP) and liver transaminases: serum aspartate aminotransferase (AST) and alanine transaminase (ALT). All markers were comparable between the untreated mice, SM-102, and lipid 35-treated animals. AST and ALT were slightly not significantly increased by lipid 14 (FigureJ–M).

Finally, we investigated whether any tissue damage is caused by the different lipids in the lungs, the livers, and the spleens of injected animals. To this end, we harvested these organs 24 h post-systemic administration and performed histological assessment using hematoxylin/eosin-stained sections. As shown in FigureN, no visible sign of tissue inflammation, infiltration of leukocytes, necrosis, or fibrosis was detectable in any treatment group.

Taken together, these results highlight that lipid 35 after bolus injection is well tolerated with no liver or other major organ damage detected, as confirmed by blood markers and histological analysis.

The therapeutic potential of LNPs formulated with lipid 35 was tested in a murine model of metastatic lung cancer. To this end, C57BL/6J mice were intravenously injected with B16F10.9 melanoma cells (as detailed in the Experimental Section in FigureA). LNPs loaded with mRNA encoding for domain III of the pseudomonas exotoxin A domain (mmPE), a toxic subunit produced by the bacteria Pseudomonas aeruginosa were systemically administrated (as detailed in the Experimental section).⁵³ As a control group for these experiments, we used lipid SM-102 as a commercial benchmark ionizable lipid formulated with mmPE.

We first assessed the mmPE-A-loaded LNPs’ (mPE-A LNP) tolerability after systemic administration. Interestingly, the treatment with SM-102 mPE-A LNPs caused an increase in the serum levels of AST, ALT, and ALP 24 h of injection (FiguresB,C and S12A↗) that remained elevated at 72h, possibly because of accumulation of particles in the liver. However, enzyme levels in lipid 35 mPE-A LNP-treated individuals, despite being elevated in the first 24 h, went down to baseline levels after 72 h from treatment, and ALP levels were not affected by lipid 35. Similarly, blood chemistry 72 h after injection (Figure S12B–E↗) displayed a slight increase of total protein and urea levels only for SM-102 mPE-A LNP. This could be attributed to the increased lung specificity of lipid 35, which reduces its off-target accumulation compared to the benchmark and results in improved tolerance of the treatment.

Next, the therapeutic efficacy of the LNPs was assessed. As shown in FigureC, after sacrifice, tumor-bearing animals treated with lipid 35 LNPs loaded with the mock mLuc mRNA did not show any decrease in lung weight or number of metastases. However, treatment with mmPE-loaded LNPs did result in a decrease of lung weight for both SM-102 (124 ± 17%) and lipid 35, which was slightly, albeit not significantly, more effective than the benchmark (69 ± 7%). Similarly, the number of metastases was decreased further by lipid 35 (12 ± 2) than SM-102 (21 ± 5) compared to both untreated (47 ± 6) and mLuc LNP-treated mice (42 ± 9), albeit the difference was not statistically significant.

When animal survival was assessed after treatment (FigureD), both SM-102 and lipid 35 mmPE LNPs displayed a protective effect. However, animals injected with lipid 35 mm PE LNPs had a remarkably higher median survival of 34 days compared to the 24 days of those injected with SM-102 mmPE-LNPs.

Taken together, these results show how the use of lung-targeting lipid 35 can be translated to a significant improvement in the therapeutic profile of LNPs.

To determine whether lipid 35 provided LNPs with improved colloidal stability and reliable transfection efficiency, we tested LNPs when stored in mild conditions and dispersed in 1× PBS at 4 °C. As summarized in FigureA, all of the tested LNPs retained their average size distribution over 2 months of storage. The PDI of lipid 14 gradually increased over time, while lipid 35 showed only a small increase in PDI (around 0.1) on day 63 (FigureB). Thus, positively charged lipid 35 provided a high ζ, preventing LNP aggregation by electrostatic repulsion. The ζ of the LNPs formulated with lipid 14 decreased over time, going from an average of −6 mV to almost −20 mV at day 63 (FigureC). This phenomenon could be attributed to the aggregation of LNP. Of note, there were no changes in the mRNA encapsulation efficiency in both tested formulations (FigureD).

In summary, these results indicate that using lipid 35 provides the resulting LNPs with colloidal stability. Combined with the longer chemical stability discussed above, this lipid appears very suitable to produce a stable LNP.

1,2-Distearoyl-sn-glycero-3-phosphocholine (DSPC, cat. no. 850365P-1g), cholesterol (cat. no. 700000P-5g), and 1,2-dimyristoyl-rac-glycero-3-methoxypolyethylene glycol-2000 (PEG-DMG 2000, cat. no. 880151P-5g) were purchased from Avanti Polar lipids. Ethanol absolute (cat. no. 000525052100) and 2-propanol (cat. no. 001626052100) were provided by Bio-Lab. DPBS 1× was purchased from Gibco (Cat# 14190–169), PBS 10X from Hylabs (Cat# BP507/500D), DEPC-treated water from BioPrep (Cat# DPH20-500 ML), and 0.5 M citrate buffer solution from Thermo Scientific (pH = 4.5, Cat# J60024.AK).

Fetal bovine serum was purchased from Biowest (FBS, heat-inactivated, EU origin, Cat# S140H-500). l-Glutamine 200 mM (100×, cat. no. 25030-024), 0.25% trypsin–EDTA (1×, cat. no. 25200-114), penicillin–streptomycin solution (10,000 U/mL of penicillin +10,000 μg/mL of streptomycin, cat. no. 15140-122), DMEM (1×, cat. no. 41965-039), and RPMI Medium 1640 (1×, cat. no. 21875-034) were provided by Gibco.

RAW264.7 cells (cat. no. TIB-71), HeLa cells (cat. no. CCL-2), and HepG2 cells (cat. no. HB-8065) were provided by ATCC. Cell cultures were maintained in an incubator at 37 °C in a controlled atmosphere (5% CO₂, 95% humidity) using the culture media required by the producer using T25 and T75 flasks.

See the Supporting Information↗.

Chemically modified mRNAs encoding firefly luciferase (62 kDa) and Cre recombinase (39 kDa) were provided by BioNTech or purchased from TriLink Biotechnologies. Custom mRNA encoding mmPE-A was acquired from Trilink with the following open reading frame:

5′-ATGgccgaagaagctttcctcggcgacggcggcgacgtcagcttcagcacccgcggcacgcagaactggacggtgg agcggctgctccaggcgcaccgccaactggaggagcgcggctatgtgttcgtcggctaccacggcaccttcctcgaagc ggcgcaaagcatcgtcttcggcggggtgcgcgcgcgcagccaggacctcgacgcgatctggcgcggtttctatatcgcc ggcgatccggcgctggcctacggctacgcccaggaccaggaacccgacgcacgcggccggatccgcaacggtgccct gctgcgggtctatgtgccgcgctcgagcctgccgggcttctaccgcaccagcctgaccctggccgcgccggaggcggc gggcgaggtcgaacggctgatcggccatccgctgccgctgcgcctggacgccatcaccggccccgaggaggaaggcg ggcgcctggagaccattctcggctggccgctggccgagcgcaccgtggtgattccctcggcgatccccaccgacccgcg caacgtcggcggcgacctcgacccgtccagcatccccgacaaggaacaggcgatcagcgccctgccggactacgcca gccagcccggcaaaccgccgcgcgaggacctgaagTAA-3′

The stability of lipids was assessed by HPLC (Agilent, 1260 Infinity II) equipped with Thermo Fisher’s CAD using a Poroshell 120 EC-C18 (2.7 μm, 3 × 150 mm) column following the reverse phase method (Solvent-A: 0.1% TFA in 55% MeOH, 15% IPA and 30% water; solvent-B: 0.1% TFA in 70% MeOH and 30% IPA). Briefly, 3 mg of lipid was dissolved in 1 mL of ethanol and a water–ethanol mixture (1:3) solvent and then filtered into Agilent scintillation vials. Three μL of the sample was injected into the HPLC-CAD instrument each time. The study was carried out over a period of 90 days at specific intervals (days 0, 1, 2, 3, 7, 14, 21, 28, 35, 60, and 90). The samples were stored at room temperature throughout the study. The stability study graph was plotted by taking time (days) on the x-axis and the percentage of lipid purity on the y-axis.

The screening was designed by using JMP Pro. The CPP ranges selected were: 30 to 45% ionizable lipid molar proportion; 20% to 45% cholesterol molar proportion, 5% to 45% helper lipid molar proportion; PEG-DMG/PEG/OME ratio from 0 to 1; total lipid concentrations between 1 mM and 7 mM; formulation temperature between 20 and 60 °C; as possible previously published ionizable lipids; as helper lipids DSPC or DOPE, and an N/P ratio between 5 and 12 (Figure S1A↗). These ranges were used to create a custom design with 20 experimental runs, including center points and repeated measures. The selected CQAs were the particle size, PDI, and zeta potential measured by DLS, mRNA encapsulation measured by RiboGreen, as well as in vitro and in vivo Luc expression. After formulating and characterizing these formulations, JMP Pro was used to understand which CPPs were statistically significant in determining the CQAs. Then, the critical CPPs were selected to simulate in silico 10⁴ possible formulations and their relative projected features and transfection efficiency using Monte Carlo Simulation. Finally, ad hoc quality thresholds were applied (Figure S1C↗) to select a small range of possible optimal formulations.

LNPs were formulated using a NanoAssemblr Benchtop device (catalog no. NIT0055) equipped with a heating block (catalog no. NIT0026) and relative cartridges (catalog no. NIS0009, Precision Nanosystems). The lipids were dissolved in absolute ethanol and kept at 55 °C. mRNA was diluted in a 25 mM citrate buffer (pH 4.5, Thermo Scientific, Cat# J60024.AK). To prepare the LNP organic phase, lipids were mixed in the following molar ratios: 40% ionizable lipid, 48.5% cholesterol, 1.5% PEG-DMG, and 10% DSPC to a final concentration of 6 mM. The N/P ratio of ionizable lipids to mRNA was 9. The lipid and aqueous phases were loaded in the NanoAssemblr in 1 and 3 mL syringes, respectively. The particles were assembled at 45 °C, using a flow rate ratio (FRR) of 3:1 (aqueous: ethanol), a total flow rate (TFR) of 12 mL/min, using a pre- and postwaste of 50 μL. Particles were subsequently dialyzed using MAXI GeBaFlex-tubes, 14 kDa MWCO (Gene Bio-Application LTD, cat. no. D050-100), against 1× PBS (Hylabs, cat. no. BP507/500D), which was replaced after 4 h. Particles were recovered after dialysis the next day.

The nanosize and ζ-potential of prepared mRNA-LNPs were analyzed by dynamic light scattering (DLS) using a Malvern Zetasizer (Malvern Instruments). Briefly, mRNA-LNPs were diluted in double-distilled water (1:50, volume ratio) and PBS (1:50, volume ratio) for ζ-potential and size measurements, respectively.

The Quant-iT RiboGreen RNA assay kit (Life Technologies) was used to measure the mRNA encapsulation in LNP. In summary, 2 μL of LNPs was diluted in a final volume of 350 μL of TE buffer (20 mM EDTA, 10 μM Tris–HCL) with or without Triton X-100 (0.5%, Sigma-Aldrich). The samples with Triton were incubated at 55 °C for 5 min. Following this, 100 μL of each sample was added in triplicate to a black 96-well plate (Costar, Corning), and 100 μL of TE buffer (0.5% v/v, RiboGreen reagent) was added to each well. The fluorescence was detected using a microplate reader (Biotek Industries) following the manufacturer’s protocol. To estimate the percentage of encapsulated RNA,was used eq

where E is the encapsulation percentage of mRNA; FluoLNPs and FluoLNPTr are the fluorescence signals without and with Triton, respectively; Blank and BlankTr are the fluorescence signals of blanks without and with Triton, respectively.

As previously described, the pK_a values of LNPs were measured using the 2-(p-toluidino)-6-naphthalenesulfonic acid (TNS) assay.³³ Briefly, the master buffer was prepared using 10 mM 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES), 10 mM 4-morpholineethanesulfonic acid (MES), 10 mM ammonium acetate, and 130 mM sodium chloride (NaCl). Sixteen buffers with pH ranging from 2.5 to 10 were prepared using 1.0 M sodium hydroxide and 1.0 M hydrochloric acid based on the master buffer. The TNS reagent was prepared as a 0.1 mM stock solution in Milli-Q water. 90 μL of each buffer was added in triplicate to a black 96-well plate, and then 6 μL of 0.1 mM total lipid LNPs were added to each well. Then, 5 μL of the TNS stock solution was added to each well and kept on the shaker to mix properly for 10 min by covering the plate with aluminum foil. Fluorescence intensity was measured using excitation and emission wavelengths of 322 and 431 nm, respectively. pK_a curves were prepared by plotting the pH values on the x-axis and the normalized fluorescence values on the y-axis. Estimation of the pK_a values was performed using GraphPad Prism to perform nonlinear regression with variable slope.

HeLa, HepG2, and RAW 264.7 cells were cultured according to the producer’s instructions, periodically checked to ensure the absence of Mycoplasma infections, and kept below passage 10. To test LNP activity in vitro, these different cell lines were seeded at a density of 10,000 cells per well in a 96-well plate. The day after seeding, cells were treated with LNPs at an mLuc mRNA concentration of 0.5 μg/mL, 0.25 μg/mL, and 0.125 μg/mL diluted in a complete cell culture medium. Every treatment was performed in quadruplicate. 24 h after treatment, cells were lysed using 50 μL of passive lysis buffer 1× (Promega) per well. Afterward, 30 μL of the cell lysates was pipetted in a white 96-well plate (Costar, White flat bottom, nontreated no lid, cat. no. 3912) and their luminescence was read using a GloMax plate reader equipped with dual injectors. 50 μL of the Luciferase Assay System (Promega) substrate was injected per well and the exposure time was set to 10 s.

All animal studies were performed in accordance with ethical guidelines and were approved by the Tel Aviv University Ethics Committee (Protocol # TAU-LS-IL—2201-108-3).

Healthy, 8–10 weeks old, female C57BL/6J mice were injected retro-orbitally with a volume of LNPs loaded with luciferase-encoding mRNA (mLuc) corresponding to 10 μg of mRNA. 6 h after the injection, mice were injected intraperitoneally with 200 μL of 15 mg/mL of IVISbrite D-luciferin diluted in 1× PBS (PerkinElmer Cat#122799). After 5 min, mice were anesthetized with isoflurane and sacrificed by cervical dislocation. Their hearts, lungs, livers, spleens, and kidneys were harvested, and the Luc signal was measured using an IVIS Lumina device with automatic settings. For analysis, the pictures were processed using the Living Image Software (version 4.1) to measure the average radiance (measured in photons/sec/cm²/sr) within ad hoc defined ROIs. Every experimental group included 4 mice. Mice that displayed obvious signs of discomfort after LNP injections were sacrificed for ethical reasons and removed from the experiment.

For this application, LNPs were loaded with mRNA encoding the Cre recombinase enzyme (GeneScript). Female, 8–10 weeks old B6g.Cg-Gt(ROSA)26Sortm9(CAG-tdTomato)/Hze/j mice (referred to as Cre-tdTomato mice) were injected retro-orbitally with a volume of particles corresponding to 20 μg of Cre mRNA. After 72 h, mice were anesthetized and sacrificed by cervical dislocation, and lungs, livers, and spleens were harvested.

Spleens were ruptured through 70 μm cutoff strainers, and the obtained single-cell suspension was centrifuged at 300g for 5 min at 4 °C. After pipetting the supernatant, cells were resuspended in red blood cell lysis buffer (Sigma-Aldrich) for 1 min quenched by adding 10 mL of 1× PBS and centrifuged at 300g for 5 min at 4 °C. After decanting the supernatant again, cells were resuspended in 0.5 mL of FACS buffer for cell counting and antibody staining.

Livers and lungs were processed using the Miltenyi mouse liver dissociation kit (Cat# 130-105-807) or the Miltenyi mouse lungs dissociation kit (Cat# 130-095-927), respectively, following the manufacturer’s instructions.

Cells were then incubated with mouse FcR blocker reagent (Cat# 130-092-575 Miltenyi) according to manufacturer’s instructions and then stained with different antibody panels as reported in Tables and 2.

After 30 min of incubation at 4 °C, cells were centrifuged at 500g for 5 min and finally resuspended in 100 μL of FACS buffer supplemented with DAPI 5 μg/mL before analyzing them on a Cytoflex Flow cytometer (Beckman) using the gating strategy summarized in Figures S13 and 14↗.

To assess the LNPs’ biodistribution, LNPs were loaded with noncoding siRNA labeled with Cy5 and nonmodified Luc mRNA in a 50:50 (w/w) ratio. The particles were then injected IV in 10 week old female C57BL6/j mice at a dose of 10 μg of total mRNA per mouse. 18 h after administration, mice were sacrificed and their lungs, livers, and spleens were harvested and processed for flow cytometry as previously described. The extracted cells were then stained using the antibody panels summarized in Tables and 4 and analyzed using the gating strategies analogous to the ones presented in Figures S13 and 14↗. LNP accumulation was measured as a percentage of Cy5-positive cells.

Female 8–10 weeks old, C57BL/6J mice were injected retro-orbitally with a volume corresponding to 10 μg of mRNA of LNPs loaded with luciferase-encoding mRNA (mLuc). 2 and 24 h after injection, approximately 300 μL of blood per mouse was recovered via cheek bleed in a Microtainer SST Blood collection tube (ref# 365,968, BD). To isolate the plasma from whole blood, the collection tubes were centrifuged at 3500 rpm for 10 min. The supernatant plasma was recovered in 1.5 mL Eppendorf tubes and stored at −80 °C for future analysis.

The concentrations of TNF-α, IL-6, MCP-1, and IL-6 in the plasma were measured using ad hoc ELISA kits from R&D Systems: Mouse CCL2/JE/MCP-1 DuoSet ELISA (ref # DY 479), Mouse IL-6 DuoSet ELISA (ref # DY 406), Mouse IL-10 DuoSet ELISA (ref # DT 417), Mouse TNF-alpha DuoSet ELISA (ref # DY410).

The plasma concentrations of alkaline phosphatase (ALP), serum glutamic oxaloacetic transaminase (SGOT), and serum glutamate pyruvate transaminase (SGPT) were analyzed by AML Lab Services.

24 h after injection, the livers, spleens, and lungs were also harvested and fixed in a paraffin solution (Sigma-Aldrich). The tissue processing, embedding in paraffin, sectioning, and hematoxylin/eosin staining were performed by Histospek.

The protocols were adapted from ref (54). B16F10.9 cells (5 × 10⁵ diluted in 100 μL of PBS) were administered intravenously into 10- to 12 week old male C57BL/6 mice. Treatment groups (n = 7 mice/group) included (1) mock treatment (1× PBS), (2) mLuc-loaded lipid 35 LNP, (3) mmPE-loaded SM-102 LNPs, and (4) mmPE-loaded lipid 35 LNP. An additional group of untreated tumor-free mice served as control (n = 7). The dose in each formulation was 0.09 mg/kg mRNA encoding PE.⁵² Treatments were administered on days 1, 5, and 9 from tumor inoculation. Administration was by intravenous injection of 100 μL of the selected formulation to the lateral tail vein using 26-gauge needles.

Two independent experiments were run: one to evaluate lung metastatic burden and the other to evaluate survival.

For mmPE-LNP, the general toxicity was evaluated using liver enzyme levels (AST and ALT), as well as ALP, urea, bilirubin, total protein, and creatinine measured in the blood. Mice bearing B16F10.9 tumors (B16F10.9) received three doses of either PBS, mLuc-loaded lipid 35 LNP, mmPE-loaded SM-102 LNPs, or mmPE-loaded lipid 35 LNPs at a dose of 0.09 mg/kg mRNA (n = 4 mice/group). AST and ALT levels were assessed 24 and 72 h from the end of treatment while the other markers were measured 72 h post-treatment.

For evaluation of lung metastatic burden, the experiment was terminated 20 days post-tumor injection. The lungs of all animals in the experiment were removed, weighed, and fixed in Blouin’s solution. The increase in lung weight was calculated using eq 2(53)

Surface metastases were counted, using a dissecting microscope, by a pathologist blinded to the experimental groups involved. For evaluation of survival, 7 animals were included in each experimental group. After the end of treatment, they were monitored every other day and the experiment was terminated on day 38. Statistical significance for the difference in survival was assessed using the Log-rank (Mantel–Cox) Test.

After assembly, LNPs formulated using the lead lipids (lipids 14, 35, and 36) were stored as suspension in 1× PBS in sealed 2 mL clear glass vials (Merck, catalog no. 27265) and kept at 4 °C. A small nanoparticle aliquot was withdrawn, and DLS, Ribogreen, and in vitro Luciferase assay were performed as previously presented at day 1 and then every 7 days until 2 months.

Data are presented as the data mean ± the standard error. Before analysis, normality tests were performed to assess the data Gaussian distributions. Statistical comparisons of two different groups were performed using a two-tailed paired t-test. For experiments involving multiple groups, one-way ANOVA with multiple comparison post-hoc tests was employed instead. All analyses were performed on Graph Pad Prism version 5.00 (GraphPad Software, San Diego, California, USA, www.graphpad.com↗). To elucidate possible correlations between the LNP sizes, zeta potentials, pK_a, and in vitro and in vivo signals, we performed correlation analysis using GraphPad Prism. The correlations were assessed by plotting variables in pairs and assessing the function in GraphPad Prism. P values below 0.05 were considered statistically significant. Significance intervals for p were designed as follows: * for p < 0.05, ** for p < 0.01, *** for p < 0.001.