Engineered internal architecture of core-shell lipid nanoparticles promotes efficient mRNA endosomal release

Abstract

Messenger RNA (mRNA) therapeutics rely on lipid nanoparticles (LNPs) for delivery, yet inefficient endosomal escape remains a major bottleneck, with only a small fraction of internalized cargo reaching the cytoplasm. Conventional LNPs encapsulate mRNA in amorphous lipid cores, where partial charge neutralization and lack of structural order limit protonation-driven membrane disruption. Here, we present an architectural strategy that engineers LNP internal structure using ionizable lipid–coated gold nanoparticles (IC-AuNPs) as rigid, pH-responsive cores. The Au cores template the formation of radially ordered core–shell architectures that stabilize particles at physiological pH while amplifying charge segregation and curvature stress under acidic endosomal conditions. As a result, Au-LNPs achieve a twofold increase in endosomal escape and ~100-fold greater cytoplasmic mRNA diffusion compared to conventional LNPs. Functionally, Au-LNPs enhance mRNA expression in vitro, increases in vivo protein production up to sevenfold, boost antibody responses to SARS-CoV-2 vaccines, and improve therapeutic efficacy in a triple-negative breast cancer model.

Lipid Nanoparticles (LNPs) effectively deliver mRNA to cells but suffer have low levels of endosomal release. Here the authors report on core-shell LNPs with ionizable lipid–coated gold nanoparticle cores with enhanced pH-responsive membrane disruption, endosomal escape, and cytosolic mRNA delivery improving therapeutic efficiency.

Introduction

Messenger RNA (mRNA)-based therapeutics are poised to transform preventive and curative medicine, offering promising solutions for a broad arrange of diseases, including infectious disorders, cancers, and genetic conditions^1–3. Central to the clinical success of mRNA therapies is the development of efficient, safe, and versatile delivery systems that can protect the mRNA cargo, navigate biological barriers, and ultimately release mRNA into the cytoplasm for effective translation^4–7. Among the various non-viral vectors, lipid nanoparticles (LNPs) have emerged as the leading platform for mRNA encapsulation and delivery, exemplified by the success of LNP-formulated mRNA vaccines against COVID-19^8–12.

Despite these advances, a critical bottleneck in mRNA delivery remains the inefficient escape of LNPs-encapsulated mRNA from endosomes into the cytosol, with current cytoplasmic delivery rates estimated at around 2%^5,13,14. Once internalized, LNPs become trapped in the endo-lysosomal system, where mRNA is prone to degradation if not promptly released^15,16. Over the past decade, researchers have explored a wide range of strategies to enhance endosomal escape^17,18. Chemical modifications of ionizable lipids, incorporation of pH-responsive groups, and careful tuning of lipid compositions have all expanded the design toolkit for LNPs and yielded valuable improvements^19–21. Nevertheless, achieving robust and reliable endosomal escape remains an ongoing challenge, underscoring the need to explore new design principles that move beyond established paradigms.

Conventional LNPs encapsulate mRNA in an amorphous mixture of ionizable lipids, helper lipids, cholesterol, and polyethylene glycol (PEG)–lipids¹. Upon exposure to the acidic endosomal environment, ionizable lipids become protonated and can, in principle, promote membrane fusion by forming non-bilayer hexagonal (H_II) phases²². In practice, however, the effective charge density of ionizable lipids is partially neutralized by mRNA within the LNP core, attenuating their membrane-disruptive capacity²³. Moreover, the lack of structural order within conventional LNPs may limit the extent and coordination of these phase transitions, resulting in suboptimal cytosolic release⁵. These intrinsic architectural constraints highlight the need for alternative design principles that address not only lipid chemistry but also the internal organization of LNPs.

In this work, we present a distinct architectural approach that shifts the focus from solely chemical modifications to the internal structure of LNPs. Specifically, we leverage ionizable lipid–coated gold nanoparticles (IC-AuNPs) as rigid, pH-responsive cores that template the formation of radially ordered core–shell architectures. This design stabilizes nanoparticles under physiological conditions (pH 7.4), preserving their integrity during circulation, while amplifying membrane-disruptive forces in the acidic endosomal environment (pH ~5.5–6.0) through enhanced charge segregation and curvature stress (Fig. 1).

Fig. 1. Schematic of the enhanced endosomal escape mechanism of Au-LNPs.

In conventional LNPs, the efficacy of ionizable lipids is hindered as they are neutralized by electrostatic interaction with the mRNA cargo. The unique internal architecture of our Au-LNPs resolves this issue. The positively charged gold core attracts and sequesters the mRNA, which in turn electrostatically repels the ionizable lipids in the surrounding shell. This repulsion liberates the lipids, maximizing their availability to interact with and disrupt the endosomal membrane, leading to superior cytoplasmic mRNA delivery.

By systematically optimizing this architecture, we achieved a twofold increase in endosomal escape efficiency and a ~100-fold enhancement in cytoplasmic mRNA diffusion compared to conventional LNPs. The resulting Au-LNPs exhibited markedly higher mRNA expression across multiple cell types in vitro and up to a fivefold increase in protein expression in vivo. Functionally, Au-LNPs enhanced the immunogenicity of a SARS-CoV-2 spike mRNA vaccine and improved therapeutic efficacy in a triple-negative breast cancer model. Importantly, gold was used here as a proof-of-concept scaffold; the underlying design principle is generalizable to other chemically compatible cores. By shifting the focus from lipid chemistry alone to the internal architecture of LNPs, this work establishes a new paradigm for overcoming the endosomal escape bottleneck and expands the design toolkit for next-generation mRNA therapeutics.

Results and discussion

Design and physicochemical characterization of gold- lipid nanoparticles

AuNPs were selected as a proof-of-concept core material primarily due to their chemical versatility, including high surface reactivity for versatile modifications and precise size control (Fig. 2a). To explore whether internal architecture could enhance mRNA delivery, we synthesized IC-AuNPs as structural cores for LNP assembly. AuNPs were first PEGylated and then coated with ionizable lipids such as MC3 or SM-102 via phase transfer mediated by hydrophobic forces, yielding hydrophobic IC-AuNPs that readily integrated into organic solvents (Fig. 2a, Methods). Successful coating was confirmed by Fourier transform infrared (FTIR) spectroscopy and nuclear magnetic resonance (NMR), which revealed diminished PEG-specific peaks alongside the emergence of MC3 signatures (Supplementary Fig. 1a, b).

Fig. 2. Morphological and characterization of Au-LNPs.

a Schematic representation of IC-AuNPs modification and the potential Au core structure. b TEM image of gold nanoparticles (13 nm diameter), repeated independently at least 3 times with similar results. c IC-AuNPs showing a corona-like morphology, repeated independently at least 3 times with similar results. d Au-LNPs incorporating 13 nm gold nanoparticles exhibiting a “single-sided fried egg” morphology, repeated independently at least 3 times with similar results. e Average hydrodynamic sizes of LNPs and Au-LNPs measured by DLS (ethanol injection method; a. u.: arbitrary units). f Lipid membrane fluidity of LNPs and Au-LNPs under varying pH conditions(Data are presented as mean ± SEM). g Disruption efficiency of artificial liposomes under varying pH conditions for LNPs and Au-LNPs.

For LNP assembly, both conventional LNPs and Au-LNPs were prepared from the same pre-mixed lipid master solution (Methods), ensuring identical lipid composition across groups. The critical difference lay in the assembly sequence. In conventional LNPs, mRNA was directly mixed with the lipid solution, resulting in amorphous lipid–mRNA condensates. In contrast, for Au-LNPs, mRNA was first condensed onto IC-AuNPs to form a stable mRNA–Au core complex. This condensation step was confirmed by gel electrophoresis, which showed complete retardation of mRNA migration and no detectable free mRNA in solution (Supplementary Fig. S1c), indicating efficient binding to the IC-AuNPs. The resulting mRNA–Au complexes were then encapsulated by the lipid master mix to generate ordered core–shell nanoparticles. This two-step assembly process positioned mRNA and ionizable lipids radially around the Au core, establishing a distinct internal architecture compared with conventional LNPs.

Transmission electron microscopy (TEM) visualized ~13 nm Au cores (Fig. 2b) coated with MC3 (Au-MC3) and surrounded by a 2.5 nm lipid layer, confirming successful core–shell formation (Fig. 2c, Methods). When assembled into Au-LNPs, these MC3-coated cores yielded nanoparticles with an average dry-state diameter of 42.37 ± 0.25 nm (Fig. 2d and Supplementary Fig. S1d). In contrast, dynamic light scattering (DLS) indicated a hydrated hydrodynamic size of ~70.2 nm (Fig. 2e, “Methods”), while preliminary cryo-EM measurements confirmed a hydrated size of 69.32 ± 0.34 nm with intact spherical morphology (Supplementary Fig. S1e, Methods). The smaller values from TEM are consistent with the expected shrinkage of lipid nanoparticles during negative staining and drying, as noted in previous LNP studies²⁴. Importantly, TEM of conventional LNPs was also included for direct comparison (Supplementary Fig. S1f). To confirm that lipid coating is generalizable, we extended the phase transfer method to three different ionizable/cationic lipids and verified successful attachment to the AuNPs surface in all cases (Supplementary Fig. S1g). Similar core–shell morphology was also observed when ALC-0315 or DOTAP were used in place of MC3 (Supplementary Fig. S1h), demonstrating that the architecture is reproducible across different lipid chemistries and not an artifact of a single lipid.

Notably, Au-LNPs exhibited significantly higher uniformity than conventional LNPs, with no oversized particles detected by DLS and a markedly lower polydispersity index (PDI) of 0.073 compared to 0.128 for LNPs (Fig. 2e). Mechanistically, the absence of larger particles in Au-LNPs can be attributed to enhanced hydrophobic free energy around the gold core, which promotes tight lipid assembly and prevents the formation of poorly hydrated lipid aggregates that otherwise yield oversized structures.

We further investigated the impact of IC-AuNPs incorporation on LNP structural stability through molecular dynamics (MD) simulations and membrane fluidity assays. MD simulations revealed that Au-LNPs exhibited a significantly lower assembly enthalpy (–354.3 kJ/mol) compared to conventional LNPs (–210.7 kJ/mol), indicating that the two-step templated process yields thermodynamically more stable structures (Supplementary Fig S1i). Membrane viscosity was assessed using the fluorescent probe 1,6-diphenyl-1,3,5-hexatriene (DPH), which showed higher fluorescence intensity at acidic pH in Au-LNPs, suggesting an increased presence of hydrophobic domains²⁵. These findings support the conclusion that Au-MC3 increases hydrophobic free energy, thereby promoting lipid encapsulation around the mRNA–Au complex and stabilizing Au-LNPs assembly (Fig. 2f).

Liquid Chromatography–Mass Spectrometry (LC-MS) analysis confirmed that ionizable lipid content was indistinguishable between LNPs and Au-LNPs formulations (Supplementary Fig. 1j, k, “Methods”), demonstrating that incorporation of the Au core does not alter overall lipid composition and ruling out formulation bias as a confounding variable. To validate the functional contribution of the Au core, we characterized the Zeta potential of Au-MC3 compared with free MC3 lipids across a range of pH values (Methods). Both exhibited characteristic pH-dependent charge shifts, but Au-MC3 showed a significantly larger change in surface potential (Supplementary Fig. 1l), indicating that the gold core amplifies the protonation-driven responsiveness of ionizable lipids.

Together, these results establish that incorporating IC-AuNPs as scaffolds drives the formation of stable, structurally ordered Au-LNPs, while maintaining identical lipid composition compared to conventional LNPs, thereby isolating internal architecture as the key variable underlying their distinct behavior.

AuNP size dictates assembly stability and delivery efficiency

We next investigated how the size of the AuNPs core influences LNP assembly and mRNA delivery. IC-AuNPs with diameters of 13, 18, 25, and 50 nm were synthesized and coated with MC3, then incorporated into LNPs using the same lipid master solution as above. TEM and DLS analysis revealed that 13 nm Au cores produced monodisperse Au-LNPs with narrow size distributions, whereas larger cores disrupted lipid organization and yielded broader distributions with oversized aggregates (Fig. 2e and Supplementary Fig 1l–n, “Methods”).

Functional studies using luciferase (Fluc) mRNA demonstrated a strong correlation between Au core size, assembly stability, and transfection efficiency. Au-LNPs with 13 nm cores enhanced luciferase expression by approximately threefold relative to conventional LNPs, whereas 18 nm cores provided only a modest 1.5-fold increase. By contrast, 25 nm and 50 nm cores showed no improvement or even reduced expression efficiency (Supplementary Fig. S1m). These findings indicate that only small Au cores are compatible with ordered LNPs assembly, while larger cores occupy excessive internal volume and destabilize lipid packing.

Consistent with these observations, DLS confirmed that only 13 nm Au-MC3 cores produced stable nanoparticles with uniform size distributions, whereas larger cores resulted in heterogeneous particles with higher polydispersity (Supplementary Fig. 1n). Together, these results identify ~13 nm as an optimal Au core size for templating ordered Au-LNPs, directly linking nanoscale architecture to functional mRNA delivery efficiency.

Enhanced pH-responsive membrane disruption by Au-LNPs

Efficient endosomal escape requires ionizable lipids to become protonated in acidic compartments and destabilize the endosomal membrane. Having established that Au-LNPs exhibit increased thermodynamic stability and amplified pH responsiveness, we next assessed their membrane fusion potential under endosomal-like conditions.

We first employed a sulforhodamine B (SRB)–loaded liposome assay as a widely used proxy for pH-dependent membrane fusion of lipid nanoparticles (Methods)²⁶. Both Au-LNPs and conventional LNPs showed minimal leakage at physiological pH (7.4) and robust activity near endosomal pH (~5.5), consistent with ionizable lipid protonation. Across the acidic range, Au-LNPs consistently induced higher leakage than conventional LNPs, with a significant enhancement at pH 5.5 (Fig. 2g).

To provide a more biologically relevant validation, we complemented the SRB assay with a pH-dependent hemolysis assay using red blood cells (RBCs, Methods). In line with the SRB results, Au-LNPs triggered greater membrane disruption at acidic pH (65% hemolysis at pH 5.5 vs. ~57% for conventional LNPs; Supplementary Fig. 2a), while remaining inert at neutral pH. Together, these assays confirm that Au-LNPs possess superior pH-triggered membrane-lytic activity compared with conventional LNPs.

To mechanistically link this enhanced activity to lipid chemistry, we measured the Zeta potential of Au-MC3 versus free MC3 across a pH gradient. Both systems exhibited the expected protonation-dependent charge shift, but Au-MC3 underwent a substantially larger change in surface potential (Supplementary Fig. S1l). This finding supports the model that the Au core amplifies protonation-driven charge segregation, thereby increasing the effective charge density of ionizable lipids at the particle–membrane interface. The enhanced membrane destabilization near pH 5.5 is also consistent with prior reports showing that protonated ionizable lipids promote non-bilayer hexagonal (H_II) phases, which are highly disruptive to membranes^5,22.

Together, these results demonstrate that Au-LNPs not only stabilize nanoparticle architecture under physiological conditions but also potentiate the pH-triggered membrane-disruptive behavior of ionizable lipids under acidic conditions, a critical prerequisite for efficient cytosolic mRNA release.

In vitro assessment of mRNA expression efficiency of Au-LNPs

We conducted a series of in vitro experiments that analyzed mRNA expression in cells to evaluate the actual mRNA delivery efficiency of Au-LNPs and the TEM imaging of endosome escape of based on unique electro density of Au-LNPs (Methods). We first assessed the impact of different lipids coatings on Au-LNPs by using Fluc as a reporter gene. HeLa and 293 cells were transfected with Au-LNPs, each functionalized with different ionizable/cationic lipids, and containing Fluc-mRNA. Following transfection, luciferase activity of cell lysate was measured to quantify mRNA expression. Among the tested coatings, Au-MC3 demonstrated the most significant enhancement in mRNA expression compared to conventional LNPs (Fig. 3a). This enhancement is possibly attributed to the favorable zeta potential shift of Au-MC3 under acidic conditions, which facilitates better interaction between the nanoparticles and the cell membrane, leading to improved endosome escape and subsequent mRNA translation (Fig. 3b). The toxicity of Au-LNPs with MC3 were verified by cell-counting-kit-8(CCK8) assay (Methods), showing no statistically significant difference between the Au-LNPs treated and LNPs treated HeLa cells (Fig. 3c). MC3 was selected as surface modification of IC-AuNPs for subsequent research.

Fig. 3. In vitro assessment of mRNA delivery efficiency and endosomal escape.

a Luciferase (FLuc) expression in HeLa cells 6 h after transfection with LNPs, Au-DOTAP-LNPs, Au-MC3-LNPs, or Au-SM102-LNPs. Data are presented as mean ± SEM. Sample size n = 3 biologically independent samples, a. u.: arbitrary units. b Zeta potential of Au-DOTAP, Au-MC3, and Au-SM102 measured at pH 4.0 (50 mM acetic acid) and pH 7.4 (PBS). Data are presented as mean ± SEM. Sample size n = 3 technically independent samples. c Cell viability of HeLa cells assessed by CCK-8 assay 24 h after transfection with LNPs or Au-LNPs at different mRNA doses. Data are presented as mean ± SEM. Sample size n = 3 biologically independent samples. d Dose–response of FLuc expression in HeLa cells transfected with LNPs or Au-LNPs at four different mRNA concentrations. Data are presented as mean ± SEM. Sample size n = 3 biologically independent samples. e Confocal fluorescence images of GFP expression in HeLa cells 12 h after transfection with GFP-mRNA encapsulated LNPs or Au-LNPs. f Flow cytometry analysis of GFP-positive HeLa cells 12 h after transfection with increasing doses of LNPs or Au-LNPs carrying GFP-mRNA. Data are presented as mean ± SEM. Sample size n = 3 biologically independent samples. Statistical significance was determined by one-way ANOVA by Fisher’s LSD multiple comparisons test by Graph Pad Prism 9.4.

The interaction between Au-MC3 and mRNA was characterized using gel electrophoresis, isothermal titration calorimetry (ITC), and DLS (Supplementary Figs. 1c and 2b, c). Gel electrophoresis demonstrated that Au-MC3 efficiently retarded mRNA migration, confirming strong binding between the cationic MC3-coated cores and the anionic mRNA. ITC further revealed an exothermic binding profile with favorable enthalpy, consistent with electrostatic interactions and hydrogen bonding driving mRNA condensation on the Au-MC3 surface. Correspondingly, DLS analysis showed an increase in hydrodynamic diameter upon mRNA loading, indicating the formation of stable Au–mRNA complexes. Together, these results confirm that Au-MC3 cores act as effective nucleation sites for mRNA condensation, enabling subsequent lipid encapsulation into ordered core–shell Au-LNPs.

We then assessed the optimal dosage for maximal FLuc-mRNA expression delivered by LNPs and Au-LNPs (Fig. 3d and Supplementary Fig. 2d). In both systems, luciferase intensity increased as the mRNA dosage increased. However, in the conventional LNPs group, protein expression peaked at 0.5 μg/mL; beyond this dose, at 1 μg/mL, protein expression decreased. Au-LNPs showed continued increase without peaking up to 2 μg/mL, possibly attributable to enhanced escape enabling higher dose tolerance without saturation/degradation (Supplementary Fig. S2d). The improved expression efficiency mRNA delivered by Au-LNPs may be attributed to improved endosomal escape efficiency.

To further evaluate the mRNA delivery performance of Au-LNPs, we employed confocal microscopy (CLSM) and flow cytometry to measure GFP-mRNA expression in HeLa and 293 cells using both conventional LNPs and Au-LNPs delivery systems at two different dosages (Methods). As expected, cells treated with Au-LNPs exhibited substantially higher GFP transfection efficiency and mean GFP expression intensity compared to those treated with conventional LNPs at both high and low dosages (Fig. 3e, f and Supplementary Fig. 2e, f). Notably, at the low dosage, Au-LNPs demonstrated dominant superiority, achieving a transfection efficiency of 91.4% compared to 6.72% with conventional LNPs. This enhanced performance could be due to improved endosomal escape efficiency. Similar trends were observed using Fluc as a reporter gene, confirmed across five different cell types and with three LNPs formulations (Supplementary Fig. 2g–l).

Mechanistic insights into Au-LNP–mediated endosomal escape

To elucidate the endosomal escape mechanisms of Au-LNPs compared to conventional LNPs, we employed CLSM and TEM (Fig. 4, Methods). Using Cy5-labeled mRNA, we quantified endosomal escape by measuring the cytosolic diffusion area of the Cy5 signal and calculating the endosomal escape efficiency (Fig. 4a–d and Supplementary Fig. 3a–e, Methods). After 6 h of incubation, a substantial portion of Cy5-mRNA delivered by Au-LNPs reached the cytoplasm, as evidenced by diffuse Cy5 fluorescence (Fig. 4c). Notably, the cytoplasmic diffusion area of Cy5-mRNA in Au-LNP–treated cells was 100-fold greater than that in cells treated with conventional LNPs (Fig. 4d)¹³, and the endosomal escape efficiency of Au-LNPs was twice as high (Supplementary Fig. S3d. To assess endosomal maturation, we quantified colocalization between Cy5-mRNA (green) and LysoTracker (Red, Methods) via CLSM. Reduced overlap (~50% lower; Supplementary Fig. S3e) across doses (0.25–1 μg/mL) and times (4–6 h) in Au-LNP-treated cells indicates less mRNA in acidified compartments, implying Au-LNPs reduced endosomal acidification (lower LysoTracker colocalization; Fig. 4a–c and Supplementary Fig. S3e), favoring early escape at pH ~5.5–6.0 where MC3 is protonated, while avoiding progression to highly degradative lysosomes (pH ~4.5). This balance enhances cytoplasmic release and protects the mRNA cargo. This aligns with enhanced escape (Supplementary Fig. S3d). This observation, consistent with our membrane disruption studies under different pH conditions, supports the notion that Au-LNPs inhibit endosome acidification and maturation, thereby maintaining conditions more favorable for mRNA release.

Fig. 4. Intracellular trafficking and endosomal escape of Au-LNPs.

Confocal images showing colocalization of lysosomes (red) and Cy5-mRNA (green) in HeLa cells treated with LNPs or Au-LNPs for 6 h at mRNA doses of 0.25 μg/mL (a), 0.5 μg/mL (b), and 1 μg/mL (c). Scale bar, 20 μm. d Quantification of cytosolic Cy5-mRNA signal in HeLa cells treated with LNPs or Au-LNPs for 4 and 6 h. Data are presented as mean ± SEM. Sample size n = 5 biologically independent samples. e TEM images of HeLa cells treated with LNPs or Au-LNPs at mRNA doses of 0.25 μg/mL (left), 0.5 μg/mL (middle), and 1 μg/mL (right) for 4 h. Scale bar, 2 μm. f Quantification of vesicle size distribution Data are presented as mean ± SEM. Sample size n = 13 biologically independent samples. g, h Representative TEM images showing endosomal escape events of Au-LNPs, repeated independently at least two times with similar results. Statistical significance was determined by two-way ANOVA by Fisher’s LSD multiple comparisons test by Graph Pad Prism 9.4.

TEM provided additional insights into how Au-LNPs influence endosomal morphology (Fig. 4e–h). Au-LNPs exhibit favorable stable characteristics within biological environments, as demonstrated in Fig. 2f, g. The high electron density and unique nanostructure of Au-LNPs enabled clear visualization^27–30. Au-LNPs were found in vesicles ranging from 0.2 to 5 µm in diameter, encompassing early endosomes (EE, 0.1–0.5 µm) and micropinosomes (MS, 0.5–5 µm). Both LNPs- and Au-LNPs-treated cells displayed more large vesicles (>1 µm) than the negative control (Supplementary Fig. S3f), a hallmark of disrupted trafficking between early and late endosomes/lysosomes³¹. Importantly, Au-LNP–treated cells contained significantly more enlarged vesicles than LNP-treated cells (Fig. 4f), suggesting a stronger membrane disruption capability. We observed that these enlarged vesicles often fused with existing MS (Fig. 4g, h), and in some cases, vesicles with partially compromised membranes also underwent fusion. The exact cause of these membrane defects—whether membrane fusion events, proton sponge effects, or other factors—remains unclear. Moreover, Au-LNPs tended to cluster near damaged areas of partially disrupted vesicles, possibly due to osmotic pressure gradients (Fig. 4g, h).

To further investigate whether Au-LNPs modulate endosomal maturation, we examined the expression and localization of canonical endosomal markers^32,33. Confocal immunofluorescence staining revealed that LNP-treated cells exhibited strong colocalization of mRNA with the late endosomal/lysosomal marker LAMP1, whereas Au-LNPs–treated cells showed markedly reduced LAMP1 signal, with mRNA distributed more diffusely in the cytoplasm (Fig. 5a). In contrast, no significant differences were observed in the early endosomal marker RAB5 between LNPs- and Au-LNPs–treated groups (Fig. 5b). Quantification confirmed that LAMP1 intensity was significantly reduced in Au-LNP–treated cells, while RAB5 intensity remained unchanged (Fig. 5c).

Fig. 5. Mechanistic study of Au-LNP–mediated endosomal escape via inhibition of endosomal maturation.

Confocal fluorescence images of 293 T cells treated with 0.5 or 1 µg/mL LNPs or Au-LNPs. Nuclei (blue), mRNA (red), and late endosome/lysosome marker LAMP1 (green, a) or early endosome marker RAB5 (green, b) were immunostained. Scale bars, 10 µm. c Quantification of LAMP1 and RAB5 mean fluorescence intensity from images in (a, b). Data are presented as mean ± SEM. Sample size n = 3 biologically independent samples. a. u.: arbitrary units (d). Western blot analysis and densitometric quantification of LAMP1 and RAB5 protein expression after treatment with LNPs or Au-LNPs for 4 and 6 h. Data are presented as individual data points from n = 2 independent experiments. e. qPCR analysis of LAMP1 and RAB5 mRNA levels in 293 T cells after treatment with LNPs or Au-LNPs for 4 and 6 h. Data are presented as individual data points from n = 2 independent experiments. Statistical significance was assessed using two-way ANOVA (for >2 groups) by Graph Pad Prism 9.4.

These observations were corroborated by Western blot and qPCR analyses. LAMP1 protein and mRNA levels were consistently decreased following Au-LNP treatment at both 4 and 6 h, whereas RAB5 expression showed no significant differences compared with conventional LNPs (Fig. 5d, e). Together, these results indicate that Au-LNPs selectively interfere with late endosomal maturation without altering early endosome formation. By attenuating progression to LAMP1 degradative compartments, Au-LNPs create a temporal window that favors endosomal escape of mRNA at the early-to-late transition stage, consistent with the enhanced cytosolic release observed in our colocalization and TEM studies.

Collectively, our findings reveal fundamental differences between Au-LNPs and conventional LNPs in promoting endosomal escape (Fig. 1). Previous research has shown that ionizable lipids in acidic endo-lysosomal compartments become positively charged and interact with the negatively charged endosomal membrane, encouraging the formation of a metastable hexagonal (H_II) phase that facilitates mRNA release^15,16. However, in conventional LNPs, the uniform distribution of ionizable lipids and mRNA can diminish the lipids’ effective charge density, as they partially neutralize each other’s charges, reducing their membrane-disruptive potential. By contrast, Au-LNPs employ a pH-sensitive core that becomes strongly positively charged at low pH. This positively charged core attracts negatively charged mRNA, drawing it closer to the core region, while simultaneously repelling positively charged ionizable lipids and pushing them toward the endosomal membrane. This strategic segregation ensures that the ionizable lipids maintain a higher effective charge density and more readily induce the H_II phase. Consequently, membrane destabilization is enhanced, enabling more efficient mRNA release into the cytoplasm. Once mRNA diffuses into the cytosol, the local environment reverts to near-neutral pH (~7), decreasing the core’s positive charge and weakening its attraction to mRNA. Freed from the core’s electrostatic constraints, the mRNA diffuses more readily into the cytoplasm. Thus, the pH-sensitive core in Au-LNPs serves a dual function: directing ionizable lipids toward the endosomal membrane to promote escape and orchestrating a charge-driven redistribution of mRNA and lipids for more efficient cytosolic release. This mechanistic model explains the superior endosomal escape efficiency and mRNA diffusion achieved by Au-LNPs, demonstrating how architectural engineering and charge responsiveness can overcome the intrinsic limitations of conventional LNPs. By offering a deeper understanding of the interplay between nanoparticle structure, lipid chemistry, and endosomal microenvironment, our findings provide a blueprint for future design strategies aimed at enhancing mRNA delivery efficacy.

Au-LNPs enhanced mRNA delivery across administration routes and organ targeting compatibility

We then applied Au-LNPs in mouse models to evaluate its in vivo potential (Methods). Mice were administered Au-LNPs or conventional LNPs prepared via both ethanol injection and microfluidic methods. In vivo chemiluminescence imaging, reflecting mRNA expression levels, revealed that mice treated with Au-LNPs exhibited significantly higher signals, showing up to a fivefold increase compared to those treated with conventional LNPs (Fig. 6a–c). This result aligns with our in vitro findings, highlighting the robustness of the Au-LNP system in enhancing mRNA expression across diverse biological environments. Notably, the chemiluminescence intensity from microfluidic preparations was an order of magnitude higher than that from ethanol injection, consistent with our nanoflow cytometry results (Fig. 6b, c).

Fig. 6. In vivo mRNA delivery using LNPs and Au-LNPs.

a Schematic illustrating incorporation of IC-AuNPs into conventional and SORT formulations. Bioluminescence imaging of mice 24 h after intravenous (i.v.) injection of 5 μg FLuc mRNA delivered by LNPs or Au-LNPs prepared by ethanol injection (b) or microfluidics (c). Data are presented as mean ± SEM. Sample size n = 3 biologically independent samples. a. u.: arbitrary units (d) Bioluminescence imaging of mice 24 h after intramuscular (i.m.) injection of 5 μg FLuc mRNA delivered by LNPs or Au-LNPs prepared by ethanol injection. Data are presented as mean ± SEM. Sample size n = 3 biologically independent samples. e Bioluminescence imaging of mice 24 h after i.v. injection of lung-targeting SORT LNPs or Au-LNPs encapsulating 5 μg FLuc mRNA. Data are presented as mean ± SEM. Sample size n = 2 biologically independent samples. Statistical significance was determined by Two-tailed t test by Graph Pad Prism 9.4.

We also demonstrated that Au-LNPs are effective across different administration routes. As shown in Fig. 5d, intramuscular administration of Au-LNPs led to a 7-fold increase in luminescence intensity compared to conventional LNPs. We investigated mRNA expression and organ distribution from both Au-LNPs and LNPs. Due to the similar surface properties of Au-LNPs and LNPs, we expected comparable biodistribution profiles across major organs such as the liver and spleen (Supplementary Fig. S4a–c). This similarity in biodistribution suggests that Au-LNPs can maintain the organ-targeting characteristics of LNPs while significantly enhancing mRNA delivery and expression.

As IC-AuNPs do not influence the native chemistries of lipids in LNPs, our promoting mRNA endosome escape structure design can incorporate lipids with different chemistries to achieve targeted organ delivery while still enhancing mRNA delivery efficiency. We incorporated the Selective Organ Targeting (SORT) system—known for its affinity to particular organs—to allow preferential accumulation of Au-LNPs in the liver or lungs, respectively. This level of control over organ targeting was further confirmed by enhanced mRNA expression in the lungs, as indicated by increased chemiluminescence in those regions (Fig. 6e and Supplementary Fig. S4d)^34,35. These findings suggest that IC-AuNPs-assisted enhance endosome escape not only enhances delivery efficiency across different preparation methods but also improves the efficacy of various LNPs formulations to achieve specific organ-targeted delivery.

Enhanced immunogenicity of Au-LNPs in SARS-CoV-2 spike-mRNA vaccines

The potential of Au-LNPs in enhancing the efficacy of preventive vaccines was thoroughly investigated using Spike-mRNA for SARS-CoV-2 as a model antigen (Methods). To assess the immune response induced by the Au-LNPs-based vaccine, we measured the geometric titer of antibodies produced following immunization. The results were promising, demonstrating that Au-LNPs were significantly more effective in eliciting a robust immune response compared to conventional LNPs. Specifically, after the initial immunization, the geometric titer of antibodies in the group vaccinated with Au-LNPs was double that observed in the group receiving the conventional LNPs-based vaccine (Fig. 7a, b). This remarkable increase in antibody titers is likely attributable to the enhanced mRNA endosome escape and expression facilitated by the Au-LNPs system. The increased efficiency in structure design and subsequent antigen production likely led to a more potent activation of the immune system, resulting in a stronger antibody response.

Fig. 7. Au-LNPs enhance the immunogenicity of an mRNA vaccine against SARS-CoV-2.

a Timeline of immunization and sample collection. Anti-Spike RBD-specific IgG titers measured by ELISA 14 days after the priming dose (b) and 7 days after the booster dose (c). Data are presented as mean ± SEM. Sample size n = 5 biologically independent samples d, e IC₅₀ neutralization titers against prototype SARS-CoV-2 pseudovirus and mouse serum variants following two-dose immunization with LNPs (d) or Au-LNPs (e). Data are presented as mean ± SEM. Sample size n = 5 biologically independent samples. f Neutralization titers against pseudoviruses of the prototype SARS-CoV-2 and variants elicited by two-dose immunization with LNPs or Au-LNPs. Data are presented as mean ± SEM. Sample size n = 5 biologically independent samples. Statistical significance was determined by one-way ANOVA by Fisher’s LSD multiple comparisons test by Graph Pad Prism 9.4.

The enhanced immune response observed after the initial dose was further amplified following a booster immunization. In the Au-LNPs group, the geometric antibody titer increased by an additional 68% compared to the conventional LNPs group after the booster dose (Fig. 7c), further highlighting the superior efficacy of Au-LNPs in vaccine applications. This suggests that the Au-LNPs platform not only enhances the primary immune response but also significantly boosts the secondary response, which is critical for achieving long-lasting immunity. In addition to the overall increase in antibody titers, the quality and functionality of the antibodies produced were also significantly improved with Au-LNPs. Virus neutralization assays were conducted to evaluate the effectiveness of the antibodies in preventing viral entry into host cells. The results showed a 46% increase in the half-maximal inhibitory concentration (IC50) for the antibodies generated by the Au-LNPs-based vaccine compared to those from the conventional LNPs group (Fig. 7d, f). This higher IC50 value indicates that the antibodies induced by the Au-LNPs vaccine were more effective at neutralizing the virus, providing stronger protection against SARS-CoV-2 infection.

Au-LNPs in therapeutic mRNA vaccines efficacy in triple-negative breast cancer model

After confirming the excellent performance of Au-LNPs in preventive vaccines, we aimed to investigate the potential application of Au-LNPs in therapeutic vaccines (Methods). The triple-negative breast cancer (TNBC) is chosen as therapeutic mRNA vaccines model for its aggressive nature and lack of targeted therapies, making it one of the most challenging forms of cancer to treat. The Wilms’ tumor 1 gene (WT1) focally amplifying in TNBC is chosen as mRNA coding region, which induces T cells to recognize tumor antigens through vaccination, activating an immune response to produce tumor antibodies^36–41. Seven days after the 1 × 10⁶ 4T1-Fluc tumor cells orthotopic injection, the mice were randomly divided into three groups: phosphate-buffered saline (PBS), WT1-mRNA loaded by LNPs, or Au-LNPs. The mice were intraperitoneally injected three times with different groups every 2 days (Fig. 8a). The TNBC volume growth rate and survival rate are evaluation indicators. The results were encouraging, demonstrating that Au-LNPs significantly enhanced the efficacy of the WT1-mRNA vaccine in reducing tumor burden and extending survival time. Compared to PBS group and conventional LNPs group, all mice of the Au-LNPs group survived for 30 days (Fig. 8b, c).

Fig. 8. Au-LNPs enhance the therapeutic efficacy of an mRNA cancer vaccine in a TNBC model.

a Schematic of 4T1 tumor model establishment and treatment schedule. Tumor growth curves (b) and survival of tumor-bearing mice (c) treated with PBS, LNPs, or Au-LNPs (n = 5 per group). Mice were euthanized when tumor volume exceeded 1500 mm³. Data are presented as mean ± SEM. Sample size n = 5 biologically independent samples. d Representative bioluminescence images of tumor-bearing mice 16 days after treatment. Data are presented as mean ± SEM. Sample size n = 5 biologically independent samples. e Quantification of tumor bioluminescence intensity at days 8, 12, and 16. Data are presented as mean ± SEM. Sample size n = 5 biologically independent samples a. u.: arbitrary units. f Images of excised TNBC tumors and corresponding tumor weights at day 16. Data are presented as mean ± SEM. Sample size n = 5 biologically independent samples. g Flow cytometry analysis of mature dendritic cells (DCs) in draining lymph nodes following the indicated treatments. Data are presented as mean ± SEM. Sample size n = 5 biologically independent samples. Statistical significance was determined by one-way ANOVA by Fisher’s LSD multiple comparisons test by Graph Pad Prism 9.4.

In addition to the significant reduction in tumor volume and extension in survival rate, the bioluminescence intensity, used as an indicator of tumor activity, was markedly lower in mice treated with Au-LNPs compared to those treated with conventional LNPs (Fig. 8d, e). Bioluminescence imaging provides a non-invasive way to monitor the metabolic activity and growth of tumors in vivo. The lower bioluminescence intensity observed in the Au-LNPs-treated group indicates not only a reduction in tumor size but also a decrease in the overall metabolic activity of the tumor cells. This reduction in metabolic activity suggests that the tumor cells are less viable and less capable of sustaining growth and proliferation, which could lead to improved long-term outcomes for patients. Specifically, Au-LNPs treatment led to reductions of 87% in tumor size and 86% in tumor weight in comparison with PBS group at 16 days (Fig. 8f and Supplementary Fig. S5a). This substantial decrease in tumor size and tumor weight suggests that the enhanced mRNA delivery and expression capabilities of Au-LNPs can be effectively harnessed to improve the outcomes of cancer immunotherapies.

The enhanced efficacy of Au-LNPs in reducing both tumor size and metabolic activity highlights their potential to improve therapeutic outcomes in cancer treatment, particularly in difficult-to-treat cancers such as TNBC. The ability of Au-LNPs to deliver mRNA more effectively, promote stronger antigen expression, and stimulate DC cell maturation may lead to more robust and sustained immune responses against cancer cells (Fig. 8g). This could translate into better clinical outcomes, including prolonged survival and potentially lower recurrence rates for patients with aggressive cancers. Moreover, the success of the Au-LNPs in this TNBC model suggests that this platform could be broadly applicable to other types of cancers, especially those that are similarly challenging to treat. By enabling more efficient mRNA delivery and enhancing immune activation, Au-LNPs have the potential to become a key component in the development of next-generation cancer vaccines and immunotherapies.

Biocompatibility and biosafety evaluation of Au-LNPs

Ensuring the biocompatibility and biosafety of new nanomaterials is critical for their successful translation into clinical applications. The biocompatibility and biosafety of Au-LNPs were assessed through a series of in vivo tests, including weight monitoring, cytokine assays, and histopathological examination of major organs (Methods). Weight monitoring and cytokine assays indicated that Au-LNPs were well-tolerated in mice, with no significant adverse effects observed during study (Supplementary Fig. S5b, c). The stability in body weight and the absence of abnormal cytokine levels suggest that Au-LNPs do not elicit significant inflammatory or toxic responses, which is a promising indicator of their safety profile.

Hematoxylin and eosin (H&E) staining of major organs further supported the favorable biosafety profile of Au-LNPs (Methods). The histopathological analysis revealed no apparent damage or inflammation in key organs such as the liver, kidneys, and lungs, confirming that Au-LNPs are bio-tolerant and do not cause significant tissue damage compared to conventional LNPs (Supplementary Fig. S5d, Fig. S6). Metabolism in vivo was conducted using Inductively Coupled Plasma Mass Spectrometry (ICP-MS). Results indicated that 48 h post-administration, the levels of Au elements in vivo significantly decreased and returned to baseline, exhibiting no significant difference from the PBS control group (Supplementary Fig. S7). These findings are critical as they demonstrate that the inclusion of gold nanoparticles in the LNPs system does not compromise the safety of the formulation, making Au-LNPs a viable option for clinical use.

In this study, we introduced a novel structural design strategy for LNPs that leverages an engineered core-shell architecture to significantly improve mRNA delivery. By incorporating a representative MC3-AuNPs core, we created a pH-sensitive environment within the LNPs that enhances their stability at physiological pH and facilitates more effective membrane disruption under the acidic conditions of endosomes. Comprehensive characterization, including molecular dynamics simulations, membrane fluidity measurements, and pH-responsive membrane disruption assays, demonstrated that this internal architecture optimization yields thermodynamically favorable assemblies and promotes hexagonal-phase lipid arrangements conducive to efficient endosomal escape.

Compared to conventional LNPs, our Au-LNPs achieved a twofold increase in endosomal escape efficiency and a 100-fold enhancement in cytoplasmic mRNA diffusion, ultimately leading to higher mRNA expression in both in vitro and in vivo models. This improved performance was further validated in applications ranging from enhanced immunogenicity with SARS-CoV-2 spike-mRNA vaccines to therapeutic efficacy in a triple-negative breast cancer model.

Notably, while the biodistribution and clearance data indicate that Au cores in our formulation do not cause long-term retention under the conditions tested, suggesting that gold itself is not a safety concern in this context. Nevertheless, for eventual clinical translation, the same architectural principle can be readily extended to FDA-approved or biodegradable core materials that offer well-established safety profiles. Thus, while Au-LNPs validate the concept, future work will focus on adapting this internal core–shell strategy to clinically preferred materials to accelerate translational potential.

Together, our findings thus establish a new conceptual framework for designing LNPs, shifting the focus from solely chemical modifications to architectural optimization. By doing so, we provide a versatile and scalable route to overcoming the longstanding bottleneck of endosomal escape, thereby advancing the clinical potential of mRNA therapeutics and vaccines.

Methods

Materials

A complete list of reagents, suppliers, and catalog numbers is provided in Supplementary Table S1. Key materials include: HAuCl₄ (Sinoreagent, Cat. 10010711), sodium citrate (Sinoreagent, Cat. 10019418), SH-PEG (Aladdin, Cat. T164400-200mg), DLin-MC3-DMA (MC3, AVT, Cat. 1224606-06-7), SM-102 (AVT, Cat. 2089251-47-6), ALC-0315 (AVT, Cat. 2036272-55-4), DSPC (Avanti, Cat. 850365 C), cholesterol (Merck, Cat. 1.04675), DMG-PEG2000 (Avanti, Cat. 880151 P), DOTAP (Qiyue, Cat. Q-0362051), DOPC (Psaitong, Cat. D10336), Cy5-mRNA (BiosynRNA, Cat. BR1RP015), and other reagents as listed. All water was nuclease-free, and consumables were endotoxin-free.

Synthesis of AuNPs

AuNPs were synthesized by the Turkevich method. HAuCl₄ (1 mM, 30 mL) was brought to vigorous boil, followed by injection of sodium citrate (1% w/v; 0.90 mL for ~13 nm or 0.45 mL for ~18 nm particles under 700 rpm/min). The solution color changed from pale yellow to deep red within 10 min. For larger AuNPs (25 and 50 nm), a seeded growth protocol using 18 nm AuNPs as seeds was employed. All glassware was cleaned with aqua regia (HCl:HNO₃ = 3:1) prior to use.

For 25 nm/50 nm AuNPs:

The as-prepared 18 nm/25 nm AuNPs (28.4 mL) served as the seed solution. The temperature of the seed solution was adjusted to 90 °C (or your specific temp) under continuous stirring (400 rpm/min). To promote growth, 1 mL of sodium citrate (1% w/v) was added, followed by the injection of 600 μL of HAuCl₄ (25 mM). The mixture was maintained at 90 °C for 30 min to ensure complete reduction of the gold precursor. A slight color shift was observed, indicating particle growth.

PEGylation of AuNPs (Au-PEG)

AuNPs (30 mL) without purification were incubated with SH-PEG (10 mg/mL in ethanol; 500 μL under 400 rpm/min) at pH 8.0 for 12 h at room temperature. Particles were pelleted by centrifugation (11,000 × g, 20 min), washed ×3 with ddH₂O, and resuspended for storage at 4 °C.

Ionizable lipid coating (IC-AuNPs)

Au-PEG (50 μL, 1 mg/mL, aqueous) was added onto MC3 in chloroform (200 μL, 25 mg/mL). After vortexing (3000 rpm, 10 min), the AuNPs transferred to the organic phase, confirming adsorption. To remove the organic solvent and unbound lipids, the Au-MC3 nanoparticles were precipitated by centrifugation (11,000 × g, 20 min). The supernatant containing chloroform and excess lipids was carefully discarded. The nanoparticle pellet was then washed once with ethanol to remove residual organic solvent, centrifuged again, and finally resuspended in 50 mM acetic acid buffer (pH 4.0). Identical procedures were used for SM-102, ALC-0315, or DOTAP formulation coating.

mRNA synthesis

GFP, Fluc, Spike, and WT1 mRNAs were transcribed in vitro (EasyCap T7 kit, Vazyme) and purified with VAHTS RNA Clean Beads. Products were dissolved in RNase-free water. Cy5-labeled mRNA was obtained from BiosynRNA (BR1RP015).

LNP and Au-LNP formulation

LNPs and Au-LNPs were prepared from the same lipid master mix (SM-102: DSPC: cholesterol:DMG-PEG2000 = 50:10:38.5:1.5 mol%). Lipids were dissolved in ethanol (22.1 mg/mL).

Au-MC3 nanoparticles, suspended in 50 mM acetic acid buffer (pH 4.0), were aliquoted to achieve a fixed mass ratio of 2:1 (Au-MC3: mRNA). Subsequently, the calculated amount of mRNA was added directly to the Au-MC3 suspension. The mixture was incubated at 4 °C for 10 min in a static state.

LNPs: aqueous phase = mRNA in 50 mM acetic acid (pH 4.0)

Au-LNPs: aqueous phase = mRNA condensed onto Au-MC3 cores in 50 mM acetic acid (pH 4.0).

Formulations were generated using a CD-01 microfluidic chip (INano) at total flow rate 12 mL/min (aqueous:organic ratio 3:1) at room temperature. Suspensions were dialyzed overnight against PBS (10 kDa MWCO), sterile-filtered (0.22 μm), and stored at 4 °C.

Encapsulation efficiency

Encapsulation efficiency was measured by RiboGreen assay. LNPs/Au-LNPs (10 μL) were added to PBS (90 μL) or PBS + 1% Triton X-100 (90 μL). Fluorescence (Ex 485 nm, Em 525 nm) was measured, and efficiency calculated:

Physicochemical characterization

DLS/ζ: Zetasizer Pro (Malvern, Zetasize software 7.12). For ζ-potential titration, particles (10 ng/mL mRNA) were measured in buffers prepared by mixing 50 mM citric acid/sodium citrate.

TEM: negatively stained (phosphotungstic acid) samples imaged on JEOL JEM-2100F. >100 particles from 10 images were analyzed in ImageJ 1.51.w⁴².

Cryo-EM: imaged on Talos Glacios (200 kV, Ceta CMOS, 57,000×, defocus –1.2 to –2.5 μm).

LC-MS: The concentration of SM-102 in the supernatant, obtained after precipitating Au-MC3 particles with acetonitrile and centrifugation, was determined by LC-MS on a SCIEX Triple Quad™ 4500 (C18 column, monitoring the m/z 710.7 → 472.3 ESI⁺ mode).

Membrane disruption assays

SRB leakage: DOPC liposomes loaded with SRB (3:1 v/v, SRB:DOPC) were prepared; free dye removed by Sepharose CL-2B column. Leakage quantified by dequenching; Triton X-100 = 100% release. Microplate reader: Softmax Pro 5.4.1

RBC hemolysis: Whole blood (SIPEIFU B006-02) was washed ×2. 2% RBC suspensions were incubated with nanoparticles (0.5 μg/mL mRNA) at varying pH for 2 h at 37 °C. Hemolysis quantified by A541; PBS = 0%, Triton = 100%.

Cell culture and transfection

HEK293 and HeLa cells were maintained in DMEM + 10% FBS + 1% P/S. Cells were transfected in Opti-MEM containing LNPs or Au-LNPs. GFP expression was measured by flow cytometry, luciferase by luminescence assay, and cytotoxicity by CCK-8 assay. The HeLa cell line (1101HUM-PUMC000011) and 293 cell line (1101HUM-PUMC000010) was obtained from the Cell Resource Center, Peking Union Medical College (which is part of the National Science and Technology Infrastructure, the National Biomedical Cell-Line Resource, NSTI-BMCR. http://cellresource.cn) These 4T1, EMT6 and JC cell lines used in this study was kindly provided by Dr. Kai Miao at University of Macau.

Confocal imaging and lysosomal colocalization

Two hundered ninty three cells were seeded in glass-bottom dishes and incubated with LNPs or Au-LNPs containing Cy5-mRNA (0.25–1 μg/mL). After 4–6 h, lysosomes were stained with LysoTracker Green (75 nM, 1 h) and nuclei with Hoechst 33342 (1 μg/mL, 10 min). Live imaging was performed on Leica SP5 (Leica LASX 4.8.0.28989).

For quantification, two independent experiments with three fields each were analyzed. Colocalization (Pearson’s Correlation Coefficient) and cytosolic diffusion area were measured in Fiji (threshold 32–76).

Western Blot and qPCR

Cells treated with LNPs or Au-LNPs were lysed in RIPA buffer. Equal protein (30 μg) was resolved by SDS-PAGE and blotted to PVDF. Membranes were probed for RAB5, LAMP1, and GAPDH. Bands were quantified in ImageJ (1.51.w) and normalized. Uncropped scans of the blots and gels are provided in the Supplementary Information Figs. 8 and 9.

For qPCR, RNA was extracted (Vazyme RC112-01), reverse transcribed (Vazyme R323-01), and amplified using SYBR Green (APE K1170, 7500 software v2.3). RAB5 and LAMP1 expression was normalized to GAPDH. Primer sequences are listed in Supplementary Methods.

Immunofluorescence staining

Cells were fixed (4% PFA), permeabilized (0.2% Triton X-100), blocked (5% BSA), and incubated with anti-RAB5 or anti-LAMP1 antibodies overnight at 4 °C. After secondary antibody incubation, nuclei were counterstained with DAPI. Images were acquired on Leica SP5.

Cryo-EM sample preparation

Vitrification was performed using a Thermo Vitrobot at 8 °C, 100% humidity^43,44. Samples were blotted for 3.5 s, plunge-frozen in liquid ethane, and stored in liquid nitrogen.

DPH fluorescence anisotropy

LNPs or Au-LNPs were incubated with DPH (1 × 10⁻⁷ M) for 16 h in buffers at varying pH. Fluorescence anisotropy was recorded at Ex 350 nm/Em 428 nm (FluoroMax Plus). Apparent microviscosity was calculated using published formulas²⁵.

Molecular dynamics (MD) simulations

Molecular dynamics (MD) simulations were conducted using GROMACS 2021.7 with the CHARMM36 all-atom force field. Models of lipids and RNA fragments were generated using the Automated Topology Builder (ATB). Partial charges for RNA were refined using Multiwfn. Gold atom parameters were adapted from published CHARMM data.

Two systems were constructed: (i) a conventional LNP assembly (ionizable lipid + helper lipids + mRNA) and (ii) an Au-LNP assembly (Au core + MC3 coating + helper lipids + mRNA). Each system was solvated in a cubic water box and neutralized.

Energy minimization (steepest descent) was followed by equilibration under NVT and NPT. Production simulations were performed at 300 K, and 1 bar Long-range electrostatics were calculated using Particle Mesh Ewald (PME) with a real-space cutoff of 1.2 nm. All covalent bonds involving hydrogen atoms were constrained with LINCS.

Each production runs 50,000,000 steps with timestep 0.3 fs, corresponding to 15  ns per system. Enthalpy values were computed from trajectory averages. Au-LNPs exhibited lower assembly enthalpy (–354.3 kJ/mol) compared to conventional LNPs (–210.7 kJ/mol), consistent with enhanced thermodynamic stability observed experimentally.

Animal studies

Mice were housed in a specific pathogen-free (SPF) facility under controlled environmental conditions. The ambient temperature was maintained at 22 ± 2 °C, with a relative humidity of 40–60% and a 12-h light/12-hour dark cycle. The animals were kept in individually ventilated cages (IVC) with free access to standard rodent chow and water. All animal procedures were approved by NCNST IACUC (No. NCNST21-2208-YC01). Mice were randomized into groups. Female BALB/c mice (6–8 weeks, ~20 g) were randomized into groups (n = 5). Humane endpoints were defined as tumor >1500 mm³ or distress. Female BALB/cAnNCrl mice (6–8 weeks old) were purchased from Beijing Vital River Laboratory Animal Technology Co., Ltd.

Bioluminescence imaging: Mice received Fluc-mRNA (2.5 mg/kg) by i.m. or i.v. injection. Organs were imaged by IVIS Spectrum 24 h later with Living Image 4.3.1.

Vaccination: Spike-mRNA (2.5 mg/kg) was injected on days 0 and 14. ELISA plates were coated with Spike (300 ng/well). Sera dilutions were 1:40–1:5120 (prime) and 1:80–1:327,680 (boost). Secondary antibody was diluted 1:2000.

Pseudovirus neutralization: Vero E6 cells were seeded at 5 × 10⁴ cells/well. Serum dilutions (1:20, 11-point, 2-fold) were incubated with VSV-based pseudovirus for 1–2 h. Infection was quantified at 14–16 h using CQ1 imaging; IC50 values were fit with a logistic regression model.

TNBC model: 4T1-Fluc cells (1 × 10⁶) were implanted orthotopically. Mice were treated with PBS, LNPs, or Au-LNPs encoding WT1-mRNA i.p. every 2 days. Tumor burden and survival were monitored.

ICP-MS clearance: Organs were digested in aqua regia using M6 microwave digestion and analyzed on Thermo iCAP TQ ICP-MS. Calibration used a 5 ppm standard with QC every 10 samples.

Statistics and reproducibility

Statistical significance was determined using Student’s t-test (two groups) or one-/two-way ANOVA in GraphPad Prism 9.4.0.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

Y.C. conceived the project and supervised the study. T.L., J.Z., and J.G. designed and performed the majority of the experiments, analyzed the data. T.L. wrote the original draft. B.S., X.G., and L.L. contributed to the theoretical calculations and structural analysis. H.X assisted with the Cryo-EM data collection and processing. Y.H., L.D., M.L., D.W., and G.Z. performed the animal vaccination. Y.W., Q.C. assisted with biodistribution experiments. In vivo antitumor assays. Y.C. and T.L. revised the manuscript. All authors discussed the results and approved the final version of the manuscript.

Peer review

Peer review information

Nature Communications thanks Giuseppe De Rosa and the other anonymous reviewers for their contribution to the peer review of this work. [A peer review file is available].

Data availability

The Figs. 2–8 and Supplementary Figs. 1–7 data generated in this study have been deposited in the Figureshare database under accession under 10.6084/m9.figshare.31072687.

Competing interests

We have updated the Competing Interests statement to disclose a pending patent application. Specifically, we state that YC and TL are inventors on a patent application NMKX-25001 related to the technology described in this manuscript.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-69017-8.