Branched Ionizable Lipids May Improve mRNA Stability, Fusion with Cells, and Delivery Efficiency

The branching structure creates a large chemical space. Consequently, branched compounds have many structural parameters such as the number and position of branched chains, the length of each hydrocarbon chain, the symmetry of the main and branched chains, the total carbon number, and the number of heteroatoms. This complexity has previously prevented accurate comparisons among these branched compounds. The preparation of a rationally designed lipid library has also been difficult due to the limited number of commercially available branched chemicals. Thus, there remains a need to improve our understanding of branched chains and how they affect the quality of mRNA‐LNPs.

In this investigation, we have focused on the α‐branched tail lipid and reduced its structural descriptors to two simple parameters: symmetry and total carbon number. α‐branched fatty acids can also be synthesized from simple commercially available chemicals, and the length of each chain can be independently controlled, which is ideal for the design of a systematic lipid library to assess a structure–activity relationship (SAR). Here, we have generated an α‐branched tail lipid library in which the head structure is the same, but the α‐branched tails have 32 different structural diversities.

Alkylating the α‐carbon of the carbonyl compounds using enolates yields α‐branched fatty acids. Utilizing two different synthetic routes depending on the difficulty of purification, targeted α‐branched tails were obtained by separating them from the raw materials and unintended byproducts. One route for this is the alkylation of dimethyl malonate with one or two iodoalkanes (C1–C15), the other is the alkylation of linear fatty acids with iodoalkanes.

The headgroups of the novel lipid library members were fixed with a CL4 structure, which is an ionizable lipid that was previously developed.^[19^] The lipids were synthesized by the esterification of α‐branched fatty acids (termed m − n, where “m” is the main chain length and “n” is the side chain length), and they yielded α‐branched‐tail lipids (termed CL4F m − n) (Figure1a).

Ionizable lipids were purified using both normal‐phase and reversed‐phase chromatography. The lipids were all identified using proton nuclear magnetic resonance (¹H‐NMR) and positive mode liquid chromatography with mass spectrometry (LC/MS). The purity was measured by liquid chromatography with charged aerosol detection (LC/CAD) and confirmed to be over 83.9%.

The α‐branching tail is described based on its symmetry and total carbon number (Figure 1b). The symmetry of the main chain and side chain is calculated as the ratio of n to m − 2. The total carbon number for the α‐branched tail was calculated as m + n and it can range from C8–C26. Each of the 32 lipid types assessed was plotted according to their tail structures (Figure 1c). For example, CL4F 9–7, CL4F 10–6, CL4F 12–4, and CL4F 16–0 have the same total carbon number (16) in each tail, but their symmetry (y‐axis) varied, and were fully symmetrical (1.0), highly symmetrical (0.75), moderately symmetrical (0.4), and linear (0.0), respectively. In contrast, CL4F 6–4, CL4F 9–7, and CL4F 14–12 have fully symmetrical (1.0) branched tails, but the total carbon number for their tails (x‐axis) varied, and was 10, 16, and 26, respectively. This simple description of the structure facilitates an improved understanding of the “structure–property–function” relationships among the branched‐tail lipids.

LNPs were formulated by mixing ionizable lipids, cholesterol, 1,2‐distearoyl‐sn‐glycero‐3‐phosphocholine (DSPC), and 1,2‐dimyristoyl‐rac‐glycero‐3‐methoxypolyethylene glycol‐2000 (DMG‐PEG2k) (50:38.5:10:1.5 mol%) with mRNA using an ethanol dilution method. Despite the same composition and formulation process, the physicochemical properties of the LNPs immediately after preparation varied depending on the tail structure of the ionizable lipids (Table S1, Supporting Information). Particle size ranged from 65–220 nm, the polydispersity index (PDI) from 0.002–0.225, and the encapsulation efficiency from 40% to 100%.

The apparent pKa of the LNPs helps to explain their in vivo fate and it was reported to shift dynamically if the head structure changed.^[9^] However, despite the 32 ionizable lipids in the constructed library having the same headgroup, the apparent pKa values were found to range from 5.85 to 6.81. This indicates that not only the head but also the tail structure contributes to the apparent pKa.

To visualize the influence of tail chemistry on mRNA‐LNP formulations, contour plots were prepared with the two parameters describing the branched structure, symmetry, and total carbon number (Figure2a–d). The increase in total carbon number contributes to the smaller particle size with narrow PDI and higher encapsulation efficiency, probably due to the hydrophobicity of the lipids as indicated by the logP value. It was shown that higher symmetry contributes to well‐controlled particle size with higher encapsulation efficiency. However, this tendency cannot be explained by the hydrophobicity of lipids and it is hard to understand the mechanism. Figure 2d suggests that there is a regular pattern of change in pKa values related to the tail structure, but the underlying mechanism is unclear. Currently, there are very limited studies on how their physicochemical properties are controlled by the chemical structure of the lipids. For example, M. Cornebise et al. suggested that chemical structure influences the size and heterogeneity of LNPs via head‐mRNA interplay.^[20^] This study will guide future lipid design but was mainly a comparison of head structures. Therefore, we tried to perform a mechanism‐based study of how tail chemistry impacts physicochemical properties.

Hydrophobic tails play an important role in particle formation, as their hydrophobicity is the driving forces behind self‐assembly.^[21^] To understand the impacts of branched tail chemistry on the physicochemical properties of LNPs, we focused on the relationship between the molecular conformation and rotational motion of the lipids in LNPs. Several in silico studies have suggested that branched chains limit the rotational motion of the molecule in the lipid bilayer, as indicated by an increase in the areas elastic moduli and the axial relaxation time of the lipids.^[22, 23^] On a molecular level, a high gauche probability due to the branched chains brought about chain bending and the slow conformational motion of the hydrophobic chain (Figure3a).^[24^] Taking this into account, we hypothesized that the restriction of the molecular rotation by the branched chain would stabilize the nanoparticles, and this would result in a smaller particle size with lower PDI and higher encapsulation efficiency.

To experimentally evaluate the role of the branched chains on the LNPs, we have focused on microviscosity, which is important for membrane transport, membrane‐bound enzyme function, and receptor binding. Viscosity on the macroscale is typically measured by mechanical methods, while at the microscale it can be monitored using fluorescent probes or markers. Fluorescent techniques include single molecule tracking,^[25^] fluorescence recovery after photodegradation,^[26^] fluorescence correlation spectroscopy (FCS),^[27^] and fluorescence anisotropy.^[28^] Among the techniques used to measure microviscosity, the molecular rotor is one of the most powerful.^[29, 30^] The molecular rotor is a fluorescent probe based on a twisted intramolecular charge transfer (TICT), in which the quantum yield and fluorescence lifetime both respond to microviscosity. This mechanism enables molecular rotors to respond to nano‐scale viscosities with higher sensitivity. Of the molecular rotors, Push–pull type molecular rotors such as 9‐(dicyanovinyl)‐julolidine (DCVJ) and 9‐(2‐carboxy‐2‐cyano)vinyljulolidine (CCVJ) and BODIPY‐based molecular rotors such as BODIPY‐C₁₂ are widely used for monitoring microviscosity. DCVJ, which is not chargeable, was selected to eliminate the influence of electrostatic interactions of the molecular rotor with ionizable lipids on the measurement values. The intramolecular rotation of DCVJ occurred around the single bond between the julolidine and dicyanovinyl groups. This caused nonradiative thermal relaxation of the excitation energy, resulting in the quenching of the fluorescence intensity (Figure S1a, Supporting Information).^[31^] The DCVJ has been shown to be predominantly partitioned in the lipid membrane and detects differences in microviscosity between DPPC/cholesterol liposomes and DOPC/cholesterol liposomes.^[32^] Typically, LNPs are formed from a core which contains mRNA, water, ionizable lipid, and cholesterol, as well as a DSPC‐enriched layer and a PEG layer.^[33^] The hydrophobic nature of DCVJ allows it to locate in the hydrophobic region of LNPs, making it suitable for microviscosity measurements of internal LNPs as well as the liposome lipid membrane.

First, we confirmed that the fluorescence intensity from DCVJ strongly responds to viscosity in the glycerol/methanol mixtures (Figure S1b, Supporting Information). In contrast, the DCVJ absorption spectrum was consistently independent of the viscosity and could thus be used to correct the dye concentration (Figure S1c, Supporting Information). To convert measured values into viscosity, we utilized an integrated fluorescence emission and absorbance ratio (denoted as “R”) instead of the quantum yield.^[32^] When the measured R values were plotted against the viscosity values of the glycerol/methanol mixture at each temperature (30–60 °C), it enabled the fitting of a calibration curve based on the power law (R² = 0.833) (Figure S1d, Supporting Information). The value of the exponent (0.581) determined by fitting the power law is within the range (0.51–0.59) of previous reports.^[34, 35^]

Next, we measured the R values for each of the 15 LNP types labeled with DCVJ at each temperature (30–60 °C) and converted them into microviscosity. When the microviscosity of each LNP was plotted against temperature, a data profile could then be fitted using the Arrhenius law within the range of 30.3–50.3 kJ mol⁻¹ as activation energy, which is consistent with a previous study.^[32^] No transition point was observed since 38.5% cholesterol prevents a rapid phase transition. Considering the absence of a transition point and its association with the thermostability data, the microviscosity of each LNP was compared at 40 °C.

We compared the microviscosity of 8 different LNPs prepared from each ionizable lipid with the same carbon number (C16) but different symmetries (0.0–1.0) (Figure 3b). The results showed that highly symmetric lipids, i.e., those with longer side chains, tended to form more viscous LNPs. For example, CL4F 9–7 (longer side chain) and CL4F 14–2 (shorter side chain) had 1.76‐ and 1.29‐fold increases in microviscosity, respectively, when compared with the linear lipid CL4F 16–0 at 40 °C. Longer side chains are thus thought to restrict the rotational motion of lipid molecules. In addition, we compared the microviscosity of eight different LNPs prepared from fully symmetric ionizable lipids with different total carbon numbers (C10–C26) (Figure 3c). The results confirmed that lipids with a higher total carbon number tended to form more viscous LNPs when compared to CL4F 6–4 (C10) at 40 °C. For example, CL4F 14–12 (C26) and CL4F 9–7 (C16) were 1.96‐fold and 1.47‐fold higher in microviscosity, respectively, due to their increased hydrophobicity.

Thus, microviscosity follows a highly ordered pattern depending on the branched chemical structure (symmetry and total carbon number), which was regressed using multiple regression analysis (R² = 0.837) (Figure 3d). Furthermore, the factor analysis results made it clear that both symmetry and total carbon number significantly contribute to microviscosity (Table S2, Supporting Information). For 32 types of lipids, regressed microviscosity was visualized on a contour map (Figure 3e). This figure clearly illustrates that both symmetry and total carbon number contribute to increases in microviscosity. When the regressed microviscosity values were plotted against each of the physicochemical characteristics, a clear trend was identified as the LNPs with higher microviscosity had smaller particle sizes with lower PDI and improved encapsulation efficiency (Figure 3f–h). When the microviscosity falls below 120 mPa s, the deterioration of each physicochemical property is observed.

Surprisingly, a strong correlation was also found between microviscosity and the apparent pKa of the LNPs (Figure 3i). This is consistent with our previous study in which branched tails (assumed to be viscous) were found to cause a decrease in pKa when compared to the linear tails.^[19^] The apparent pKa was strongly influenced by the local dielectric constant as the ionizable head group is located at the lipid membrane.^[36, 37^] It was thus speculated that more viscous LNPs may require a higher proton concentration to ionize, as the amine headgroups are less likely to escape the charge repulsion in the viscous membrane.

The storage stability of the mRNA‐LNPs is an important issue. For mRNA‐LNPs to remain functional after storage, the chemical degradation of mRNA and lipids, as well as the physical degradation of LNPs, must be prevented. Chemical degradation can result in mRNAs losing their activity due to hydrolysis,^[38^] oxidation, and transesterification.^[39, 40, 41^] Furthermore lipids can undergo hydrolysis or oxidation.

To prevent physical degradation, it is recognized that heating and repeated freeze–thaw cycles should be avoided, but the impacts of lipid chemical structures on LNP stability are poorly understood. As microviscosity is related to energy that inhibits the lipids attempting to diffuse freely within LNPs, we have speculated that microviscosity will affect the physical stability of LNPs as well as their physicochemical properties immediately after preparation.

To investigate these relationships, LNPs were stored at various temperatures (4, 25, or 40 °C) and their physicochemical characteristics were assessed for up to 90 d (Table S3, Supporting Information). As expected, less viscous LNPs such as CL4F 15–1 and CL4F 14–2 gradually increased in particle size and polydispersity and over time, leaked mRNA. To better clarify the impacts of microviscosity on stability, microviscosity was plotted against physicochemical properties after 1 month of storage at 40 °C (Figure 3j–l). The results showed that LNPs with microviscosity below ≈150 mPa s have poor storage stability. These data suggest that the lipid tail structure determines the properties of LNPs, and thus influences their physical stability.

After the cellular uptake of LNPs, ionizable lipids are responsible for the release of mRNA into the cytoplasm as they facilitate membrane fusion through protonation upon endosomal acidification. Hajj et al. experimentally suggested that the methyl branch at the end of the hydrophobic tail enhanced the ionization ability at pH 5.0 which was correlated with in vivo efficacy.^[17^] Based on an in silico approach, branched chains are known to disrupt lipid packing, as indicated by an increase in surface area per lipid and a decrease in lipid chain order (Figure4a).^[22, 23^] Shinoda et al. produced a simulation to show that the branched lipid DPhPC was packed in a more disrupted manner than the linear DPPC, due to chain bending.^[24^] Overall, we have hypothesized that the additional space created by the branched‐chain maximizes the ionization capacity. The branched tails in the lipid library that was generated are a variety of carbon lengths and thus an ideal tool with which to investigate the aforementioned hypothesis.

We compared the ionization ability of five lipids with different lengths of the branched chain (linear or methyl branched ‐ heptyl branched) by utilizing the anionic fluorescent dye 6‐p‐toluidino‐2‐naphthalenesulfonic acid (TNS). To eliminate the influence of the total carbon number, each lipid tail used had the same total carbon (C16), but different symmetry. Consequently, the methyl‐branched CL4F 15–1 was found to have a higher ionization ability when compared to the linear CL4F 16–0, which is consistent with an earlier study on the methyl‐branched lipid (Figure 4b).^[17^] Additionally, we have gained a new understanding that the longer the branched side chain, the more effective its ionization (Figure 4b). This is possible because longer side chains generate more additional space between the lipid molecules and maximize the protonatable capacities of the amine headgroups. We further analyzed five fully symmetric lipids with different total carbon numbers (C10–C26). Almost all the fully symmetric lipids, except CL4F 14–12 (C26), were found to have a high ionization capacity due to their long side chains (Figure 4c). CL4F 14–12, which has the highest total carbon number in our lipid library, only had 80% protonation when compared to CL4F 6–4, and this may be because strong van der Waals interactions have filled the space.^[42, 43^]

To further support the hypothesis that the space created by the branched tails maximizes the ionization ability, tightly packed LNPs such as CL4F 14–12 and CL4F 15–1 were solubilized with a surfactant to create the space. To avoid the influence of charge on the assay, TritonX‐100, as a nonionic surfactant was added to the LNPs and the ionization ability at pH 5.0, was measured using TNS. The addition of TritonX‐100 improved the ionization in CL4F 14–12 and CL4F 15–1, which was expected, while for CL4F 6–4 and CL4F 10–6 it remained unchanged (Figure 4d). This data strongly suggests that the tightly packed conformation of the LNPs restricts the protonation of ionizable lipids.

To investigate whether maximizing the ionization ability of the amine headgroups could contribute to endosomal escape, the membrane‐disrupting activity was evaluated using a hemolysis assay in an acidic environment. Different concentrations of the LNPs were incubated with red blood cells (RBCs) to estimate the concentration at which 50% hemolysis was induced. As expected, lipids with a higher symmetry and smaller total carbon number induced 50% hemolysis at concentrations ≤10 μm (Figure 4e,f). Furthermore, a strong correlation was confirmed between ionization ability and hemolysis at pH 5.0 (Figure 4g). Hemolysis activity and ionization under physiological pH were so low that there is little need to consider nonspecific interactions in the blood (Figure S2b–d, Supporting Information). These data have demonstrated that branched side chains disrupt tightly packed membranes, which creates space for protonation in acidic environments and thus facilitates endosomal escape. It is important to note that the chain length itself of the lipids may also affect endosomal escape in terms of adaptation in the cellular lipid bilayer. Thus, the low hemolysis activity of CL4F 14–12 and CL4F 16–0 may be attributed not only to their low level of ionization ability but also to their low adaptation in the cellular lipid bilayer with too long chain length.

The in vivo efficacy of all 32 ionizable lipids was assessed. LNPs carrying firefly luciferase (Fluc)‐encoded mRNA were administered intravenously at a dose of 0.1 mg kg⁻¹. The luminescence levels in the liver and spleen, and lung were measured 6 h after i.v. administration, and the intensity was found to vary dramatically, depending on the tail structure (Figure S3a–c, Supporting Information). Since the data indicated that our lipid library is promising for efficient mRNA delivery to the liver and spleen, we focused on those organs in this study. Data was visualized as a contour plot for symmetry and the total carbon number (Figure5a,b). These figures clearly indicate that highly symmetric lipids promote mRNA expression in the liver and spleen. Additionally, the mRNA was strongly expressed in the liver with total carbon numbers of C14–C18 and in the spleen with a total carbon number of C11–C14, suggesting that the total carbon number of the branched tail controls organ selectivity.

To understand the factors that determine the mRNA expression levels in each organ, 12 LNPs carrying Cy5‐mRNA were administered and their biodistribution was evaluated (Figure S3d,e, Supporting Information). In the branched lipid library, a 7500‐fold difference in luminescence intensity was observed in the liver, whereas the Cy5 fluorescence intensity in the liver only differed by twofold and there was no confirmed correlation (Figure 5c). This indicates that biodistribution alone cannot explain mRNA expression in the liver. For the spleen, a correlation between luminescence intensity and biodistribution was identified, except for CL4F 6–4 (Figure 5d). This indicated that biodistribution was an important factor for splenic mRNA expression.

To understand the significant gap between mRNA expression and distribution in the liver, we explored the impact of LNPs’ properties in relation to the experimental parameters on efficacy. It has been established that apparent pKa is an important parameter that is used to determine the in vivo fate of LNPs. For example, a pKa value of 6–7 is optimal for effective siRNA/mRNA delivery to the hepatocytes in the liver.^[9, 44^] When plotting the measured pKa versus luminescence intensity in the liver for 32 LNPs, LNPs with an apparent pKa between 5.9 and 6.3 effectively induced mRNA expression in the branched lipid library (Figure 5e). The apparent pKa is thus a useful parameter by which to predict in vivo efficacy and is in agreement with previous knowledge. However, it alone cannot explain the mechanism, as a 10‐fold difference in luminescence is observed even if it was within the optimal pKa range.

Further analysis revealed that hemolytic activity in an acidic environment was also correlated with efficacy (Figure 5f), which is consistent with the knowledge that endosomal escape is a rate‐limiting step.^[45, 46^] Exceptionally, CL4F 6–4 induces hemolysis at low concentrations despite its low efficacy in vivo. The cellular uptake of CL4F 6–4 LNPs was very low in vitro (Figure S4, Supporting Information), suggesting that they cannot introduce mRNA into the cytoplasm even if its hemolytic activity is higher.

We have thus proposed that the chemical “structure” of the branched tail determines the “properties” of the LNPs such as apparent pKa and ionization ability (hemolysis activity), which in turn affects their in vivo “function.” To elucidate if branched hydrophobic scaffolds regulate the properties and function of LNPs, two types of symmetric ionizable lipid (CL4F 6–4 and CL4F 14–12) were mixed in molar ratios ranging from 65:35 to 20:80. As expected, an increase in the ratio of CL4F 6–4 was attributed to a higher pKa value and increase in hemolytic activity due to increased molecular rotation and the protonatable space available (Figure 5f,g). When CL4F 6–4 and CL4F 14–12 were mixed at 50% each, the LNP properties approached optimal values and may have contributed to a 10‐fold higher efficient delivery than those alone (Figure 5g). The study shows that the properties of the LNPs can be precisely controlled by varying the mixing ratio of ionizable lipids with different tail structures to maximize their functions.

CL4F 8–6 is a fully symmetric lipid that forms LNPs with favorable apparent pKa and membrane disruption abilities; CL4F 8–6 LNP achieved the most efficient Fluc‐mRNA expression in the liver among the branched lipid library. After 6 months of storage at 4 °C, the particle size and encapsulation efficiency of CL4F 8–6 LNPs were unchanged when compared to CL4F 14–2 LNPs (short, branched chain) (Figure S5a–c, Supporting Information), due to the restrictions in the rotational motion by the branched‐chain, as described above (Figure 3a). It is reported that LNPs targeted to the liver are adsorbed by protein coronas, mainly Apolipoprotein E (ApoE), in the blood and accumulate in liver hepatocytes via a low‐density lipoprotein receptor (LDLR).^[47, 48^] To determine which cells the mRNA expressed, intrahepatic enhanced green fluorescent protein (EGFP) expression was observed following the administration of EGFP mRNA/CL4F8–6 LNPs. The microscopic observations revealed a clearly lower localization of EGFP fluorescence on blood vessels, so we consider that mRNA expression occurs mainly in hepatocytes in a dose‐dependent manner (Figure6a,b, and S6, Supporting Information). To investigate the applicability of our lipids for genome editing in hepatocytes, LNPs co‐encapsulated with Cas9 mRNA and single guide ribonucleic acid (RNA) against transthyretin (sgTTR) were prepared. Transthyretin (TTR) is a protein mainly produced by hepatocytes and secreted into the blood. TTR knockdown by siRNA is approved for ATTR amyloidosis treatment,^[49^] and genome editing by CRISPR/Cas9 is also in clinical trials.^[18^] CL4F 8–6 LNPs carrying Cas9 mRNA and sgTTR with different ratios from 2:1 to 1:2 were administered, with 1:2 being optimal for efficient TTR protein reduction (Figure S7, Supporting Information). To ensure the sequence specificity and eliminate the unexpected effects of lipids themselves, LNPs encapsulating sgTTR or sgGFP were compared in the same experiment (Table S4, Supporting Information). ELISA experiments showed a 77% reduction in serum TTR after 30 days of treatment with 2.5 mg kg⁻¹ total RNA (mRNA/sgTTR = 1:2) (Figure 6c). In addition, mouse DNA sequencing was performed and the genome editing outcomes were analyzed. Genome editing efficiency was 54% in the liver at a single dose of 2.5 mg kg⁻¹ (Figure 6d). sgTTR‐treated or sgGFP‐treated animals showed no editing in the spleen (Figure S8, Supporting Information). The distribution of alleles identified around the targeted locus is dominated by single nucleotide insertions, with the rest being deletions of a few bases (Figure 6e), consistent with the characteristic pattern of spCas9 reported in several studies.^[50, 51^] These mutations lead to frameshifts and loss of function. These data illustrate the potential of CL4F 8–6 for long RNA delivery and genome editing.

The data that support the findings of this study are available from the corresponding author upon reasonable request.