Search
Close this search box.

In situ combinatorial synthesis of degradable branched lipidoids for systemic delivery of mRNA therapeutics and gene editors – Nature Communications

Design and construction of DB-lipidoids

A tandem and in situ construction method based on a one-pot, two-step, 3-CR was developed (Fig. 1a), which enabled rapid and parallel synthesis of DB-lipidoids. In this approach, an epoxide serves as a hydroxyl group precursor: a primary amine first undergoes epoxide-mediated ring-opening reaction twice under neat conditions to afford an aminoalcohol lipidoid with two body tails, and the in situ generated hydroxyl groups further undergo acylation with acyl chlorides to attach two branch tails via ester bonds. Since the headgroup, body tail, and branch tail can be independently controlled by using a variety of inexpensive and commercially available amines, epoxides, and acyl chlorides, this 3-CR is ideal for high-throughput and cost-efficient establishment of a large and systematic combinatorial library of branched lipidoids for screening and SAR analysis.

Moreover, since i) both reactions are highly efficient with an overall yield above 80%, ii) the solvent dichloromethane (DCM) and base triethylamine (TEA) can be easily removed, and iii) the leftover reactive acyl chloride is quenched in ethanol, the crude DB-lipidoids can be directly used for LNP formulation and screening. To the best of our knowledge, this construction method represents the simplest way to produce structurally diverse degradable branched lipidoids, in comparison to previous studies involving tedious synthesis and purification (Fig. S1)19,20,27.

Optimizing the tail region of DB-lipidoids

We aimed to systematically optimize the structure of DB-lipidoids and investigate SARs. Since there are three moieties (headgroup, body tail, and branch tail) that can be altered in DB-lipidoids (Fig. 1a), we decided to optimize the tail regions first and then optimize the headgroup in order to screen a large combinatorial library of DB-lipidoids in a resource-effective manner. Therefore, in the first library (Library 1), the headgroup was kept constant as amine 1 (i.e., 3-(dimethylamino)−1-propylamine) and the tail regions—both body tail and branch tail—were varied (Fig. 2a, S2 and Table S1). Amine 1 was chosen based on the studies of MC3 lipidoid, which suggest that the dimethylamino moiety with a spacer of three methylene units is effective for siRNA delivery29,30. In addition, epoxides with short alkyl chains (five variations between 6 and 14 carbons) were used as body tails to minimize the molecular weights (<500 Da) of non-degradable metabolites (Fig. 1a and S2), since previous studies have suggested that small-molecule metabolites tend to undergo rapid elimination16,20. Correspondingly, acyl chlorides with short alkyl chains (five variations between 6 and 14 carbons) were selected as branch tails. DB-lipidoids were denoted as 1-m-n (Fig. 2a), where m and n are the number of carbons in the body tail and branch tail, respectively.

Fig. 2: Optimization of the tail region and screening of Library 1.
figure 2

a A workflow for the synthesis and evaluation of Library 1. b In vitro mLuc expression shown in a heat map (n = 3 biologically independent samples). HepG2 cells were treated with mLuc-loaded LNPs at an mRNA dose of 15 ng/well (0.24 nM) for 24 h. RLU, relative light unit. Data are presented as mean. c In vivo mLuc expression shown in a heat map (n = 3 biologically independent samples). Mice were i.v. injected with mLuc-loaded LNPs at an mRNA dose of 0.1 mg/kg. Bioluminescence imaging (BLI) was performed at 4 h post-treatment and total flux was quantified. Data are presented as mean. d Correlation between in vitro and in vivo results of DB-lipidoids. The black dashed line indicates 10,000 RLU in vitro. The blue dashed line indicates the performance of MC3 LNP in vivo. Statistical significance was evaluated by a two-tailed correlation analysis using GraphPad Prism 8.0. e Body tail-activity relationship (n = 3 biologically independent samples). f Branch tail-activity relationship (n = 3 biologically independent samples). g Total carbon number-activity relationship. h Total carbon number-symmetry-activity relationship. The grey shadows indicate background levels. Source data are provided as a Source Data file.

The resulting 25 DB-lipidoids in Library 1 were formulated into 25 DB-LNP formulations by pipette mixing along with 1,2-dioleoyl-sn-glycero-3-phosphoethanolamine (DOPE), Chol, and 1,2-dimyristoyl-rac-glycero-3-methoxypolyethylene glycol-2000 (DMG-PEG). The weight ratio of DB-lipidoid/DOPE/Chol/DMG-PEG was fixed at 16/10/10/3 for initial screening. To evaluate the mRNA delivery efficiency of DB-LNPs, 1-methylpseudouridine (m1ψ)-modified firefly luciferase mRNA (mLuc) was encapsulated during LNP formulation. It is should be noted that due to the divergent structures, not all lipidoids were equally suited for LNP formation and some LNP formulations might be suboptimal. In general, aminoalcohol lipidoids were inferior for mRNA encapsulation, while DB-lipidoids showed enhanced capability to encapsulate mRNA (Table S2), presumably due to the increased hydrophobicity and self-assembling ability after the attachment of two branch tails. All LNPs were between 120–230 nm in size with polydispersity index (PDI) between 0.14–0.23 and surface charge between ± 4 mV. Notably, DB-lipidoids with long body tails and branch tails (i.e., 1-12-14, 1-14-12 and 1-14-14) tended to form larger and more polydispersed DB-LNPs (Table S2), suggesting the detrimental consequence of a bulky tail region.

Afterwards, Luc expression in vitro (HepG2 cells) and in vivo after intravenous (i.v.) administration into C57BL/6 mice was determined (Fig. 2a–c). To be noted, we used low doses of mRNA for initial screening to avoid the toxicity of LNPs that could potentially affect protein synthesis (Fig. S3). For all DB-LNPs, Luc expression was mainly observed in the upper abdomen by in vivo bioluminescence imaging (BLI), which was identified to come from the liver by ex vivo BLI (Fig. S4). Interestingly, the attachment of two branch tails to aminoalcohol lipidoids boosted delivery efficiencies both in vitro and in vivo (Fig. 2b, c). Generally, the in vitro potencies of DB-lipidoids were affected by the length of the body tail (Fig. 2b), while their in vivo potencies were affected by the length of both tails (Fig. 2c). However, there was no correlation between in vitro and in vivo potencies of DB-lipidoids (Fig. 2d), a phenomenon also reported by others31.

Next, we investigated the relationship between tail structure and in vivo potency. While all aminoalcohol lipidoids tested were inefficacious in vivo (total flux <107 p/s), appending branch tails dramatically boosted mRNA delivery efficiency by one to three orders of magnitude (Fig. 2e). Specifically, 8- or 10-carbon epoxide and 8-carbon acyl chloride were most effective (Fig. 2e, f), while further modulation of tail length reduced the overall potency benefit. Notably, the top three DB-lipidoids (1-8-10, 1-10-8, 1-12-6) in this library exhibited comparable potency to MC3, all of which bear a total of 18 carbons in each tail (m + n, Fig. 2d, g). Further increase or decrease in total carbon number of body and branch tails generally led to the loss of activity (Fig. 2g). We termed this phenomenon as the “18-Carbon Rule.” Interestingly, we found this “Rule” to be necessary but not adequate to afford potent DB-lipidoids, as the 1-6-12 DB-lipidoid displayed suboptimal performance even though it contained a total of 18 carbons in each tail (Fig. 2g).

Recently, Harashima et al. suggested that the symmetry of a branched tail could impact potency of the lipidoid19. Therefore, we prepared a contour plot with both parameters—total carbon number and tail symmetry—to better visualize their influence on in vivo performance (Fig. 2h). As expected, the potency of DB-lipidoids was dependent not only on total carbon number, but also on the symmetry of body to branch tails. The symmetry (calculated as n/(m-2)) of the lead 1-10-8 DB-lipidoid was defined as 1 due to the relatively symmetrical structure of the body and branch tails, while the symmetries of the slightly less potent 1-8-10 and 1-12-6 DB-lipidoids as well as the significantly less potent 1-6-12 DB-lipidoid were determined to be 1.7, 0.6, and 3, respectively. This finding suggests that, for DB-lipidoids with a total carbon number of 18, less symmetry (i.e., more deviation from 1) is associated with less in vivo mRNA delivery efficiency. Taken together, these results indicate that the optimal tail region should follow the “18-Carbon Rule” and have a symmetry of 1, which leads to the combination of a 10-carbon body tail and an 8-carbon branch tail.

Next, we tested the in vivo potency of purified DB-lipidoid containing two branch tails and compared it to DB-lipidoid containing one branch tail (an intermediate metabolite of the standard DB-lipidoid, Fig. S5). Purified 1-10-8 with two branch tails (denoted as 1-10-8(2)) demonstrated comparable potency to the crude product, confirming this to be the active transfection agent. In contrast, synthesized 1-10-8 with one branch tail (denoted as 1-10-8(1)) was unable to encapsulate or deliver mRNA. We then assessed the degradability of DB-lipidoids. After treatment of 1-10-8(2) with esterase (an enzyme that hydrolyzes esters) for 1 h, its metabolite 1-10-8(1) was detected (Fig. S6). Together, these results suggest that both branch tails are required for the potency of DB-lipidoid and they can be detached following degradation.

Optimizing the headgroup of DB-lipidoids

In the second stage of our optimization scheme, we kept the tail region constant at the optimal parameters identified above and varied the headgroup to generate a second library of DB-lipidoids (i.e., Library 2, Fig. 3a, S7 and Table S1). To maintain their structural similarity, only amines that can attach two body tails were chosen. In total 20 chemically diverse amines were tested, including monoamines, diamines, polyamines, and hydrazines (Fig. S7). The majority of these amines were selected from previous publications32,33. DB-lipidoids were denoted as x-10-8, where x stands for amine number identifier. The in vitro and in vivo performance of DB-lipidoids in Library 2 was determined using the same experimental conditions as Library 1 (Fig. 3b, c and S8). Again, these DB-lipidoids demonstrated a poor correlation between their in vitro and in vivo potencies (Fig. 3d). In this library, six DB-lipidoids—containing headgroup 1, 7, 9, 10, 11, or 12—exhibited comparable or superior in vivo performance compared to MC3, despite that none of them outperformed MC3 in vitro (Fig. 3b-d). Interestingly, the results from Libraries 1 and 2 demonstrate that although potent in vivo DB-lipidoids do not necessarily perform well in vitro, all of them surpass a baseline mRNA transfection threshold (i.e., 10,000 RLU, Figs. 2d and 3d).

Fig. 3: Optimization of the headgroup and screening of Library 2.
figure 3

a A workflow for the synthesis and evaluation of Library 2. b In vitro mLuc expression (n = 3 biologically independent samples). HepG2 cells were treated with mLuc-loaded LNPs at an mRNA dose of 15 ng/well (0.24 nM) for 24 h. Data are presented as mean ± SD. c In vivo mLuc expression (n = 3 biologically independent samples). Mice were i.v. injected with mLuc-loaded LNPs at an mRNA dose of 0.1 mg/kg. BLI was performed at 4 h post-treatment and total flux was quantified. Efficacious DB-lipidoids and their amines are highlighted in red. The grey shadow indicates background level. Data are presented as mean ± SD. Statistical significance was evaluated by a one-way ANOVA with Tukey’s correction. d Correlation between in vitro and in vivo results of DB-lipidoids. The black dashed line indicates 10,000 RLU in vitro. The blue dashed line indicates the performance of MC3 LNP in vivo. Statistical significance was evaluated by a two-tailed correlation analysis using GraphPad Prism 8.0. e Structures of efficacious amines and DB-lipidoids. The chemical structure of lead DB-lipidoid 11-10-8 is shown. 11-10-8 demonstrates a total carbon number of 18, a symmetry of 1, and a packing parameter (P) of 4.1. Source data are provided as a Source Data file.

We further investigated the relationship between headgroup structure and in vivo potency. Amines with only one primary amine (2, 3, and 6), two secondary amines (16–20) or a hydrazine group (4 and 5) were unable to generate potent DB-lipidoids (Fig. 3c). Efficacious amines (1, 7, 9, 10, 11, and 12) shared certain structural similarities: diamines with one primary amine and one tertiary amine spaced by two or three carbons. Notably, both the length of spacer (e.g., 1, 7, and 8) and the form of tertiary amine (e.g., 9, 11, and 14) critically impacted in vivo performance. The optimal spacer length was identified to be two or three methylene units, while the ideal tertiary amine structure was determined to be dimethylamino, diethylamino or pyrrolidinyl (Fig. 3e).

Together, after two rounds of optimization, several structural criteria of potent DB-lipidoids were identified: (1) total carbon number = 18; (2) symmetry = 1; (3) diamines with one primary amine and one dimethylamino-, diethylamino- or pyrrolidinyl-based tertiary amine spaced by two or three carbons. The lead DB-lipidoid 11-10-8, roughly six-fold more potent than MC3, was identified and used for subsequent studies. Benefitting from two branch tails, the packing parameter (P) of this molecule was determined to be 4.1 according to molecular dynamics simulations (Fig. S9), suggesting that 11-10-8 adopts a more cone-shaped structure than MC3 (4.1 versus 2.11)34.

Predicting performance and optimizing formulation of DB-LNPs

To further validate our structural criteria for DB-lipidoids, we sought to predict the in vivo performance of unidentified DB-lipidoids. As a case study, we synthesized four amine 11-based DB-lipidoids that either violated the “18-Carbon Rule” or defied the optimal symmetry of the tail region: 11-6-12 (total carbon number = 18, symmetry = 3), 11-8-10 (total carbon number = 18, symmetry = 1.7), 11-12-6 (total carbon number = 18, symmetry = 0.6) and 11-12-10 (total carbon number = 22, symmetry = 1). We predicted that these DB-lipidoids would have less in vivo mRNA transfection potency relative to 11-10-8. Indeed, our experimental results confirmed our prediction (Fig. 4a); all four DB-lipidoids that breached the aforementioned design criteria were less potent than 11-10-8. Moreover, despite having the same total carbon number of 18, 11-8-10 and 11-12-6 with more symmetry exhibited similar or higher potency compared to MC3, but 11-6-12 with less symmetry dramatically lost potency as expected. Therefore, our structural criteria can be used to predict the in vivo performance of DB-lipidoids, which could be useful for the future design and discovery of potent branched lipidoids.

Fig. 4: Prediction, optimization, and characterization of DB-LNPs.
figure 4

a Prediction and verification of in vivo performance for unidentified DB-lipidoids (n = 3 biologically independent samples). Mice were i.v. injected with mLuc-loaded LNPs at an mRNA dose of 0.1 mg/kg. BLI was performed at 4 h post-treatment and total flux was quantified. The red dashed line indicates the performance of 11-10-8 LNP, while the blue dashed line indicates the performance of MC3 LNP. The grey shadow indicates background level. b Optimization of 11-10-8 LNP formulation (n = 3 biologically independent samples). Mice were injected i.v. with mLuc-loaded LNPs at an mRNA dose of 0.1 mg/kg. BLI was performed at 4 h post-treatment and total flux was quantified. The grey shadow indicates background level. Statistical significance was evaluated by a one-way ANOVA with Tukey’s correction. c A representative cryo-EM image of 11-10-8 LNP from three independent experiments. Scale bar = 50 nm. d Physicochemical properties of 11-10-8 LNP (n = 3). e Inhibition of 11-10-8 LNP uptake by various endocytic inhibitors (n = 3 biologically independent samples). Amiloride is an inhibitor of macropinocytosis; Chlorpromazine is an inhibitor of clathrin-mediated endocytosis; Genistein is an inhibitor of caveolae-mediated endocytosis; Methyl-β-cyclodextrin (Mβ-CD) is an inhibitor of lipid raft-mediated endocytosis. Statistical significance was evaluated by a one-way ANOVA with Tukey’s correction. Data are presented as mean ± SD. Source data are provided as a Source Data file.

Next, we optimized the formula of 11-10-8 LNP and screened a series of well-established LNP formulations (F1-6, Fig. 4b and Table S3)29,35,36. F1 was the formulation used in initial screening, which comprised crude 11-10-8, while purified 11-10-8 was used in F2-6 (Figs. S10 and S11). Crude 11-10-8 and purified 11-10-8 had similar in vivo potency when the same LNP formula was used (F1 versus F2, Fig. 4b), which was in line with the previous observation (Fig. S5d). Moreover, all 11-10-8 LNP formulations outperformed the benchmark MC3 LNP formulation except F4, which possessed a higher molar percentage of DMG-PEG compared to other formulations (Fig. 4b and Table S3). The top-performing formulation F5 (11-10-8/DOPE/Chol/DMG-PEG at a molar ratio of 40/10/48.5/1.5), roughly seven-fold more potent than MC3, was chosen for further studies.

Characterization and in vitro studies of DB-LNPs

We next characterized the physicochemical properties of this optimized 11-10-8 LNP. Cryogenic electron microscopy (Cryo-EM) showed that 11-10-8 LNP possessed a dense spherical structure with a multi-lamellar shell and an amorphous core (Fig. 4c). The mRNA encapsulation efficiency (EE) of 11-10-8 LNP was determined to be ~95% (Fig. 4d). The hydrodynamic size of 11-10-8 LNP was approximately 71.2 nm with a low polydispersity index (PDI = 0.087) and a neutral surface charge (ζ = −1.09 mV). The apparent pKa of 11-10-8 LNP was determined to be 6.22 (Fig. S12). Due to its ionization ability, 11-10-8 LNP induced minimal hemolysis at pH 7.4, but increased hemolysis at pH 6.0 (Fig. S13). This ionization behavior is critical for reducing the toxicity of LNPs and enhancing membrane disruption in the acidic endosome8,37.

In HepG2 cells, 11-10-8 LNP showed dose-dependent mLuc delivery with minimal toxicity at doses ranging from 0 to 240 ng/well (Fig. S14). Moreover, 11-10-8 LNP exhibited dose-dependent delivery of GFP mRNA (Fig. S15), and could achieve nearly 100% transfection efficiency at a dose as low as 150 ng/mL. Next, we examined the cellular uptake routes of 11-10-8 LNP. The endocytosis of 11-10-8 LNP was highly dependent on lipid rafts (Fig. 4e), as pre-treatment with methyl-β cyclodextrin (Mβ-CD), an inhibitor of lipid raft-mediated endocytosis, completely suppressed mLuc delivery. Other internalization pathways (e.g., macropinocytosis, clathrin-mediated endocytosis and caveolae-mediated endocytosis) were also involved but contributed less to overall LNP internalization, since their inhibitors only slightly suppressed mLuc delivery (Fig. 4e).

Hepatic delivery of mRNA-based gene editors using DB-LNPs

Hepatic delivery of mRNA-based gene editors holds great promise for genome editing therapies. Therefore, we first demonstrated the potential of DB-LNPs for this application by comparing 11-10-8 LNP with MC3 LNP (Table S4). Ex vivo BLI results confirmed that 11-10-8 LNP predominantly transfected the liver, resulting in 4.8-fold higher Luc expression in the liver than MC3 LNP (Fig. 5a). These results were consistent with previous in vivo BLI results (Fig. 4b). Similarly, 11-10-8 LNP also outperformed MC3 LNP in terms of hepatic delivery of GFP mRNA (Fig. 5b). Immunofluorescence staining results further confirmed that 11-10-8 LNP mediated greater GFP expression than MC3 LNP in major liver cells, including hepatocytes, Kupffer cells and liver sinusoidal endothelial cells (Fig. S16).

Fig. 5: DB-LNP-mediated hepatic delivery of mRNA-based gene editors.
figure 5

a Ex vivo BLI of major organs from treated mice and their quantification (n = 3 biologically independent samples). Mice were i.v. injected with mLuc-loaded LNPs at an mRNA dose of 0.1 mg/kg. Images were taken at 4 h post-treatment. Statistical significance was evaluated by a one-way ANOVA with Tukey’s correction. b Ex vivo fluorescence imaging of major organs from treated mice and their quantification (n = 3 biologically independent samples). Mice were i.v. injected with GFP mRNA-loaded LNPs at an mRNA dose of 0.25 mg/kg. Images were taken at 4 h post-treatment. Statistical significance was evaluated by a one-way ANOVA with Tukey’s correction. c, d LNP-mediated Cas9 mRNA/TTR sgRNA co-delivery and gene editing. Mice were i.v. injected with LNPs co-delivering Cas9 mRNA/TTR sgRNA (4:1, wt:wt) at a total RNA dose of 1 mg/kg. Mice were euthanized on day 7, and DNA was extracted from the liver to determine on-target indel frequency by next-generation sequencing (c, n = 5 biologically independent samples). Statistical significance was evaluated by a one-way ANOVA with Tukey’s correction. Serum was collected at the indicated time points for ELISA analysis of TTR (d, n = 5 biologically independent samples). Statistical significance was evaluated by a one-way ANOVA with Tukey’s correction. Data are presented as mean ± SD. Source data are provided as a Source Data file.

Given the growing interest in CRISPR/Cas9-mediated liver gene editing as exemplified by Intellia’s LNP-mediated transthyretin (TTR) knockout38, we chose TTR as a model target and co-delivered Cas9 mRNA and TTR sgRNA. A single injection of 11-10-8 LNP co-delivering these two components at a clinically relevant dose (1 mg RNA/kg) led to ~30% insertions and deletions (indels) at the TTR locus and ~50% reduction of serum TTR (Fig. 5c, d). In contrast, MC3 LNP only achieved ~7% indels and ~17% reduction of serum TTR. Notably, no observable hepatotoxicity was induced by both LNP treatments as demonstrated by normal levels of alanine transaminase (ALT) and aspartate aminotransferase (AST) in treated mice (Fig. S17). Moreover, no abnormal innate immune responses were observed after examining 13 cytokines (Table S5). Together, these results strongly demonstrate the potential of our DB-LNPs in the hepatic delivery of gene editors.

Hepatic delivery of mRNA-based therapeutics using DB-LNPs

Hepatic delivery of mRNA-based therapeutics holds great promise for protein supplementation therapies. FGF21 is a pleiotropic metabolic hormone primarily secreted by the liver, which is a promising therapeutic agent for obesity, type 2 diabetes and non-alcoholic steatohepatitis39,40. We next evaluated 11-10-8 LNP for the delivery of human FGF21-encoded mRNA in high fat diet (HFD)-induced obese mice. Male mice were fed a HFD for three weeks to induce obesity (body weight ~30 g) and fatty liver41, and then i.v. injected with FGF21 mRNA-loaded LNPs. The expression of FGF21 peaked (58.8 ± 1.1 ng/mL) at 12 h post-administration of FGF21 mRNA-loaded 11-10-8 LNP, which was five-fold higher than that in FGF21 mRNA-loaded MC3 LNP-treated mice (11.8 ± 1.1 ng/mL, Fig. 6a). Moreover, the area under curve (AUC)—a pharmacokinetic metric of therapeutic exposure—of FGF21 was 4.6-fold greater in 11-10-8 LNP-treated mice than that in MC3 LNP-treated mice (1104.1 ± 12.2 ng·h/mL versus 240.4 ± 7.8 ng·h/mL, Fig. 6a).

Fig. 6: DB-LNP-mediated hepatic delivery of FGF21 mRNA.
figure 6

a LNP-mediated FGF21 mRNA delivery (n = 4 biologically independent samples). Male mice were fed a HFD for three weeks to induce obesity (body weight ~30 g) and fatty liver. These obese mice were i.v. injected with FGF21 mRNA-loaded LNPs at an mRNA dose of 0.25 mg/kg. Serum was collected at the indicated time points for ELISA analysis of FGF21. AUC of FGF21 exposure during the time interval 0–48 h was determined. Statistical significance was evaluated by an unpaired two-tailed Student’s t-test. b A scheme of FGF21 mRNA therapy in HFD-induced obese mice. Obese mice were i.v. injected with various LNP formulations at an mRNA dose of 0.25 mg/kg every other day for three doses. Male mice fed a normal chow diet (NCD) were used as a control group. c Body weight growth curve (n = 4 biologically independent samples). d Body weight on Day 6 (n = 4 biologically independent samples). Statistical significance was evaluated by a one-way ANOVA with Tukey’s correction. e Liver weight (n = 4 biologically independent samples). Statistical significance was evaluated by a one-way ANOVA with Tukey’s correction. f Representative images of liver histological examinations (H&E staining). Scale bars: 100 μm. g Liver steatosis score (n = 12). A total of 12 images for each group (three random fields for each liver section, four mice per group) were analyzed semi-quantitatively for liver steatosis. Statistical significance was evaluated by a one-way ANOVA with Tukey’s correction. Data are presented as mean ± SD. Source data are provided as a Source Data file.

To further demonstrate the therapeutic potential of this FGF21 mRNA therapy, we examined its weight-reducing and lipid-lowering effects in obese mice39. Obese mice were treated with various LNP formulations every other day for three doses (Fig. 6b). While obese mice treated with PBS or mLuc-loaded LNPs gradually increased weight, obese mice treated with FGF21 mRNA-loaded 11-10-8 LNP or FGF21 mRNA-loaded MC3 LNP lost weight or maintained weight, respectively (Fig. 6c). At the end of this experiment, body weight as well as liver weight of obese mice treated with FGF21 mRNA-loaded 11-10-8 LNP was significantly reduced compared to that of obese mice treated with other formulations (Fig. 6d, e). Hematoxylin and eosin (H&E) staining of livers showed less vacuoles and reduced liver steatosis in the obese mice treated with FGF21 mRNA-loaded 11-10-8 LNP compared to those obese mice treated with other formulations (Fig. 6f, g). It is worth mentioning that although MC3 LNP-based FGF21 mRNA therapy exhibited weight-reducing and lipid-lowering effects to some extent in obese mice, they were less obvious than 11-10-8 LNP-based therapy, presumably due to less FGF21 expression (Fig. 6a). Together, these results suggest that 11-10-8 LNP-based FGF21 mRNA therapy could be a promising approach for treating obesity and fatty liver.

While 11-10-8 LNP was more effective at delivering large RNA constructs compared to MC3 LNP, we then assessed whether it still holds advantages in delivering siRNA, for which the MC3 LNP is approved42. Interestingly, when TTR siRNA was delivered, 11-10-8 LNP and MC3 LNP showed comparable potency with a similar median effective dose (ED50, 0.029 mg/kg vs 0.030 mg/kg, Fig. S18a). Owing to the potent silencing effect, a single injection of 11-10-8 LNP at a clinically relevant dose (1 mg siRNA/kg) resulted in ~100% reduction of serum TTR on day 3 and nearly 40% reduction on day 28 (Fig. S18b). These results suggest that our DB-LNPs could also be a robust platform for the delivery of siRNA.

Apart from systemic delivery, we further showed that 11-10-8 LNP mediated strong intramuscular (i.m.) mRNA delivery (Fig. S19), which outperformed the benchmark SM-102 LNP that has been approved by the FDA for mRNA vaccine delivery8. These results demonstrate the potential of 11-10-8 LNP for local mRNA delivery and mRNA vaccine development.