Amine headgroups in ionizable lipids drive immune responses to lipid nanoparticles by binding to the receptors TLR4 and CD1d

Abstract

Lipid nanoparticles (LNPs) are the most clinically advanced delivery vehicle for RNA therapeutics, partly because of established lipid structure-activity relationships focused on formulation potency. Yet such knowledge has not extended to LNP immunogenicity. Here we show that the innate and adaptive immune responses elicited by LNPs are linked to their ionizable lipid chemistry. Specifically, we show that the amine headgroups in ionizable lipids drive LNP immunogenicity by binding to Toll-like receptor 4 and CD1d and by promoting lipid-raft formation. Immunogenic LNPs favour a type-1 T-helper-cell-biased immune response marked by increases in the immunoglobulins IgG2c and IgG1 and in the pro-inflammatory cytokines tumour necrosis factor, interferon γ and the interleukins IL-6 and IL-2. Notably, the inflammatory signals originating from these receptors inhibit the production of anti-poly(ethylene glycol) IgM antibodies, preventing the often-observed loss of efficacy in the LNP-mediated delivery of siRNA and mRNA. Moreover, we identified computational methods for the prediction of the structure-dependent innate and adaptive responses of LNPs. Our findings may help accelerate the discovery of well-tolerated ionizable lipids suitable for repeated dosing.

Figure 1:. Lipid nanoparticles elicit structure-dependent innate immune responses in vivo.

(A) A combinatorial library of 15 degradable ionizable lipids was synthesized from three amine headgroups (red) and five acrylate tails (blue). (B) In experiments throughout this study, mice were intravenously injected with two doses of 1 mg/kg siGFP-LNPs one month apart. Blood samples were collected for ELISA analysis, and spleens were collected for cellular analysis by flow cytometry. (C, D) Four hours after each dose, the inflammatory cytokines TNFα and IL-6 were measured in the serum. Error bars represent s.e.m. with n = 4 and *, **, ***, and **** representing p < 0.05, 0.01, 0.001, and 0.0001, respectively, according to two-way ANOVA with Tukey’s post-hoc analysis between LNP groups and nested t-test between doses.

Figure 2:. Ionizable lipids activate TLR4 when their amine headgroups protrude from the MD-2 binding cavity.

(A) TLR activation by LNPs was examined by incubating Raw-Blue^™ macrophage cells with either positive control TLR stimulators or LNPs and measuring resultant NF-κB expression. The positive control stimulators for TLR4, TLR2/6, and TLR1/2 were LPS, FSL-1, and Pam₃CSK₄. TLR4 and TLR2 effect was inhibited by TAK-242 and mTLR2 monoclonal antibodies, respectively. LNPs incorporating the 313 headgroup activated TLR4. (B) Ionizable lipid binding with mouse TLR4 was analyzed using AutoDock Vina. Lipids bound the MD2 binding cavity (white) and interacted with the amino acids of the TLR4 protein (black and red) through hydrophobic interactions. The binding mode of (C) DLin-MC3-DMA, (D) 304O₁₃, (E) 306O₁₃, (F) 313O₁₃, and (G) 200O₁₃ in the binding cavity and the amino acids of TLR4 (black and red) that interact with the lipids are shown. (H) Raw-Blue^™ cells were incubated with MC3 and LNPs containing the O₁₃ tail, and NF-κB activity was measured after 24 hrs. Error bars represent s.e.m. (n = 3–4), with *, **, ***, and **** representing p < 0.05, 0.01, 0.001, and 0.0001, respectively, according to two-way ANOVA with Sidak’s post-hoc analysis (A), and one-way ANOVA with Dunnett’s post-hoc analysis (H).

Figure 3:. Ionizable lipids cause an innate interference with phase separation and lipid raft formation.

A) The effect of ionizable lipid headgroups on lipid raft formation was assessed by incubating formulated LNPs or unformulated lipids with Raw 264.7 macrophage cells. The cells were then incubated with Alexa Fluor-555 Cholera Toxin-B (CTB-AF555), a marker that binds lipid rafts, and the cells were analyzed by flow cytometry. Higher mean fluorescent intensity (MFI) correlates to more phase separation and lipid raft formation. Sample histograms for two formulated LNPs and average MFI values for six formulated LNPs are shown along with negative controls (gray) and a positive control (green). (B) MFI values normalized to PBS controls are shown for all 15 unformulated lipids, grouped by amine headgroup. (C-E) Lipid raft formation was simulated in a bilayer containing DIPC (green), DPPC (grey), and cholesterol (orange) in the presence of three ionizable lipids using molecular dynamics. (F) The position of the lipid headgroup in the bilayer was analyzed by calculating the average density of the headgroup relative to the bilayer center from 2–4.5 μs. Phase separation occurs when the lipid position is near the center of the bilayer. (G) Phase separation and lipid raft formation were analyzed by calculating the average distance of the lipids from cholesterol over the course of the simulation, with average distances negatively correlating with phase separation. Error bars represent s.e.m. (n = 5), with *, **, ***, and **** representing p < 0.05, 0.01, 0.001, and 0.0001, respectively, according to two-way ANOVA with Tukey’s post-hoc analysis between lipid groups.

Figure 4:. Lipid nanoparticles that promoted lipid raft formation and TLR4 binding produced the greatest cytokine responses.

Mice were dosed twice, 30 days apart, with LNPs containing siGFP (1 mg/kg). (A) Spleens were collected from mice two or seven days after the second LNP dose, and levels of T cells were assessed via flow cytometry. Negative controls corresponded to PBS. (B) Two and seven days after the second LNP dose, levels of inflammatory cytokines were measured in the serum. (C) Ionizable Lipid-CD1d (PDB: 2AKR) binding was visualized using Pymol. Lipid-CD1d binding was assessed by incubating splenocytes with unformulated ionizable lipids followed by treatment with the CD1d ligand –KRN7000. Cells were incubated with an antibody that binds to the CD1d-KRN7000 complex and analyzed via flow cytometry. Lower fluorescence intensity indicated stronger lipid binding and lower KRN7000 binding. ionizable lipids show (D) structure-dependent and (E) dose-dependent binding to CD1d. Error bars represent s.e.m. (n = 3–4), with *, **, ***, and **** representing p < 0.05, 0.01, 0.001, and 0.0001, respectively, according to two-way ANOVA with Tukey’s post-hoc analysis between LNP structures and nested t-test for analysis between day 2 and day 7 (A, B), and one-way ANOVA with Dunnett’s post-hoc analysis (D, E).

Figure 5:. Lipid nanoparticles induced structure-dependent changes in B cell populations and antibody levels.

Spleens were collected from mice two or seven days after the second LNP dose, and levels of (A) germinal center cells, (B) memory B cells, and (C) plasma cells were measured via flow cytometry. Cell population numbers are normalized to PBS-treated negative controls. Seven days after the second LNP dose, (D) levels of IgG subtypes, (E) IgG2c/IgG1 ratio, and (E) IgM levels were analyzed in the serum. Error bars represent s.e.m. (n = 3–4), with *, **, ***, and **** representing p < 0.05, 0.01, 0.001, and 0.0001, respectively. Significance between the LNP head groups was determined according to two-way ANOVA with Tukey’s post-hoc analysis. Differences between days 2 and 7 were determined using nested t-test.

Figure 6:. Immunogenic LNPs induced PEG-IgG formation that prevented loss of efficacy upon repeat dosing.

(A) Mice were injected twice with LNPs containing anti-Factor VII siRNA (dose 0.5 mg/kg) 30 days apart, and Factor VII levels were measured two days after each injection relative to PBS negative control. Only mice injected with LNPs containing the 304 amine group lost efficacy. Next, mice were injected five times with LNPs encapsulating luciferase mRNA 30 days apart. (B) Luminescence was visualized using Living Image software and (C) quantified using Region of Interest analysis. Mice were injected with 1 mg/kg 306O_i10 or 304O_i10 siRNA-LNPs with or without PEG in the formulation, and (D) anti-PEG IgG and (E) anti-PEG IgM levels were measured over the course of 21 days from serum samples using ELISA. In separate experiments, mice were injected twice with 1 mg/kg LNPs 30 days apart, and (F) anti-PEG IgG and (G) anti-PEG IgM levels were measured seven days after the second dose. (H) Splenocytes were incubated ex vivo with LNPs or PBS along with the TLR4 inhibitor – TAK-242 or the CD1d neutralizing monoclonal antibody – 20H2, and anti-PEG IgM levels were measured after five days. Mice were injected with 304O_i10 LNPs along with CD1d and TLR4 inhibitors and corresponding isotype and solvent controls 30 days apart, and (I) anti-PEG IgM and (J) anti-PEG IgG levels were measured weekly. Binding energies of ionizable lipids with (K) the mouse MD-2 binding cavity of TLR4 and (L) the mouse CD1d binding cavity were calculated using AutoDock Vina. Ionizable lipids with lower binding energy had higher levels of anti-PEG IgM seven days after the second dose. (M) A proposed mechanism of the immune response of LNPs is shown. Error bars represent s.e.m. *, **, ***, and **** representing p < 0.05, 0.01, 0.001, and 0.0001, respectively, as determined by paired t-tests with Holm-Sidak post-hoc analysis (n = 3) (A), two-way ANOVA with Sidak’s analysis (n = 4) (B, C), two-way ANOVA with Tukey’s post-hoc analysis (D-J), spearman’s correlation (K, L) (n = 3–4).