Abstract
Lipid nanoparticles (LNPs) are a leading platform for nucleic acid delivery, yet current formulations typically suffer from rapid in vivo clearance and limited extrahepatic delivery. Recent experiments demonstrate that increasing the helper lipid content (e.g., to 40% 1,2-distearoyl-sn-glycero-3-phosphocholine [DSPC]) can significantly improve mRNA stability, prolong circulation, and enhance transfection efficiency. In this study, we used coarse-grained molecular dynamics simulations to provide mechanistic insight into the structural changes driving these improvements. Our simulations confirm that an elevated helper lipid content fundamentally alters the internal morphology of the LNP. Consistent with experimental observations, at high DSPC levels, the core transitions from a compact, disordered, predominantly lipid state to an extended structure with distinct lamellar-like elements encompassing lipid-rich and water-rich regions. This structural reorganization can be used to rationalize the enhanced stability and delivery properties reported in the experiments. These findings provide a molecular-level understanding of the composition-morphology relationship and suggest that coarse-grained modeling can inform the design of novel LNP formulations.


Background
The successful delivery of mRNA using lipid nanoparticles (LNPs) has revolutionized vaccinology and stands as a foundational technology for precision medicine. The rapid clinical success of LNP-mRNA COVID-19 vaccines (e.g., Pfizer-BioNTech’s Comirnaty and Moderna’s Spikevax) spurred a remarkable increase in mRNA/LNP research, leading to hundreds of active clinical and preclinical programs in diverse areas, including infectious diseases, oncology, immunology, and gene replacement therapies. , Despite this momentum and the apparent speed with which LNP systems can be translated to clinical trials, challenges remain in achieving efficient expression, enabling robust systemic delivery beyond the liver, , and preventing rapid degradation of the nucleic acid cargo during storage and in vivo.
These limitations are inherently related to the physicochemical properties of mRNA and LNPs. , Fundamentally, an LNP’s efficacy is dictated by its composition, typically a mixture of an ionizable lipid (IL, often with positive IL+ and neutral IL0 states), a helper lipid such as 1,2-distearoyl-sn-glycero-3-phosphocholine (DSPC) or Egg Sphingomyelin (ESM), cholesterol (Chol), and a PEG lipid. Many formulations use a standard molar ratio of these lipids (e.g., ∼50:10:38.5:1.5 mol %), , which generally self-assemble into LNPs with complex, disordered internal structures (see Figure ). Within this environment, the mRNA is subjected to a number of detrimental (e.g., high curvature) and chemical stresses. ,,
1.

Illustrative cross-sectional snapshot of an LNP containing mRNA. The components are colored: IL+: navy blue, IL0: green, DSPC: light gray, cholesterol: medium gray, PEG-lipid: orange, mRNA: red, and water: cyan. The mRNA resides within the core in contact with lipids, especially IL+ and water. The outer shell, enriched in DSPC and containing nearly all of the PEG-lipid molecules, provides structural stability and shields the mRNA from environmental factors. This shell can be a bilayer or a monolayer depending on the nature of the LNP core. LNPs can adopt a wide range of structures; see ref for a broader perspective.
Cullis and colleagues recently proposed a promising departure from standard LNP compositions (typically 50 mol % ionizable lipid, 10 mol % helper lipid, 38.5 mol % cholesterol, and 1.5 mol % PEG lipid) by significantly increasing the helper lipid content from the canonical 10 mol % to much higher levels (e.g., 40 mol % DSPC or ESM), and reducing ionizable lipid and/or cholesterol contents. , Their experiments ,,− demonstrated that this compositional shift resulted in enhanced mRNA stability in serum, prolonged systemic circulation, and improved gene expression across a wider range of tissues, overcoming several commonly encountered delivery bottlenecks. Similar results were found by others for ESM. However, the molecular mechanisms governing this superior performance, specifically how a higher proportion of helper lipid influences LNP self-assembly and internal mRNA architecture, remain incompletely understood. Closing this gap in our knowledge will provide chemical insights to guide next-generation formulation strategies.
Computational simulations can help to address this gap by characterizing the composition-dependent structure shifts at the molecular and supramolecular levels. , Given the intricate mechanisms driving LNP self-assembly, these studies can have a timely and pivotal impact on formulation strategies for the delivery of nucleic acids, particularly mRNA. Here, we apply coarse-grained molecular dynamics (CG-MD) simulations to investigate the impact of helper lipid concentration on LNP morphology, focusing on the Cullis-style high DSPC formulations with components from the Comirnaty COVID vaccine. Our simulations mimic the LNP production process, beginning with initial LNP formation at low pH in water/ethanol solutions, followed by dialysis, where the ethanol is removed and the pH is raised to physiological pH (7.4). We report that the elevated helper lipid content drives a fundamental transition in the LNP core structure from a disordered lipid-rich, nonlamellar arrangement to one with large, separate, aqueous, and lipid-dominated volumes along with lamellar structural elements to effectively sequester the mRNA cargo. These are the key structural elements noted in experiments that have been used to rationalize the performance of these high helper lipid formulations. Additionally, we provide structural insights into the nature of the molecular-level structure of these LNPs, which may prove instrumental to guiding future LNP formulation development. This study, combined with our previously published work on simulating LNP self-assembly, demonstrates the utility of coarse-grained modeling. This serves as an example of studying large-scale structural organization in a fluid utilizing a relatively high mapping ratio of 10 heavy atoms to 1 CG particle combined with soft potentials to reduce the computer time needed while enabling the study of larger systems by reducing memory requirements.
Methods
CG Model Mapping and Parametrization
We created the CG model for the mRNA-loaded LNPs using the Coarse-Grained Force Field Builder (CGFFB) in the Schrödinger Materials Science Suite (MS suite version 24-4), focusing on the Comirnaty vaccine components: ALC-0315 as the IL, cholesterol, DSPC and ALC-0159 as the PEG lipid (see Figure S1 for the structures of these lipids). The CGFFB maps the all-atom structure onto the corresponding CG model and subsequently determines the CG force field parameters by fitting to structural features from a reference all-atom MD simulation trajectory. The mapping scheme creates a CG particle for approximately every 10 heavy atoms (see Figure S1). The reference all-atom trajectory was generated by combining the trajectories from multiple independent simulations of identical composition. These all-atom simulations utilized the OPLS4 (Optimized Potentials for Liquid Simulations, version 4) force field in combination with the SPC (Simple Point Charge) water model.
The fitting protocol in the CGFFB determines initial CG parameter values and then refines those values via an approach analogous to iterative Boltzmann inversion (IBI). Each IBI iteration consists of a dissipative particle dynamics (DPD) simulation, structural analysis, and parameter adjustment to match features such as average CG particle distances, angles, and integrals of radial distribution functions. This physics- and structure-based process aims to produce a CG model that reproduces key molecular structural and self-assembly behaviors in larger systems containing molecules that can be represented by the same CG particles used in training. A prerelease version of this workflow previously produced a model that performed well for mRNA-LNPs. Details of the fitting workflow are provided in the Supporting Information. Our CG model was parametrized at body temperature (310.15 K) for use in future studies of endosomal escape, and we apply it here at room temperature (298.15 K) to model LNP production. On the basis of our experience, CG models of this type are transferable over a range of approximately 20 K.
Compositions of the Production Simulations
Lipid Composition
To systematically investigate the role of the helper lipid following the strategy described in Cullis et al., we constructed four LNP formulations based upon the Comirnaty composition with varying DSPC content (10, 20, 30, and 40 mol %). The molar ratio of IL to cholesterol (1.085) and the PEG lipid content (1.6 mol %) were held constant across all of the formulations. The 20 and 30 mol % DSPC systems were constructed based upon the 40 mol % system by varying only the number of DSPC molecules while keeping the number of mRNA, cholesterol, and IL molecules the same. Extending that approach to the 10% DSPC formulation and assuming all of the lipids and mRNA would assemble into a single LNP, the resulting particle would have been significantly smaller than those formed at higher DSPC levels. Consequently, we elected to use larger amounts of all of the components for that system. The compositions for all formulations are provided in Table . Three replicates with different randomly constructed arrangements of the molecules were generated for each formulation. In these initial solutions, we used a 90:10 IL+:IL0 ratio, which, according to the Henderson–Hasselbalch equation, corresponds to a pH of 5.14 given the apparent pK a value of 6.09 for ALC-0315.
1. Compositions for mRNA-LNP Simulations at 10, 20, 30, and 40 mol % DSPC .
| DSPC, mol % | 10% | 20% | 30% | 40% |
|---|---|---|---|---|
| mRNA | 2865 | 1910 | 1910 | 1910 |
| N/P ratio | 6.3 | 5.6 | 5.6 | 5.6 |
| IL | 17,946 | 10,788 | 10,788 | 10,788 |
| IL0 | 1795 | 1079 | 1079 | 1079 |
| IL+ | 16151 | 9709 | 9709 | 9709 |
| DSPC | 3643 | 5287 | 9089 | 14,195 |
| PEG-lipid | 619 | 422 | 484 | 567 |
| cholesterol | 16,549 | 9936 | 9936 | 9936 |
| water | 1,222,235 | 1,271,881 | 1,257,015 | 1,237,053 |
| ethanol | 417,145 | 434,089 | 429,015 | 422,202 |
| all lipids | 38,757 | 16,497 | 20,361 | 25,550 |
Values represent the number of each type of molecule, or for mRNA, the number of residues, incorporated in the initial randomly constructed systems. The IL is included as a mixture of 90% IL+ (positively charged) and 10% IL0 (charge neutral) forms consistent with acidic conditions with a pH of ∼5.14.
Total mRNA residues were determined for each LNP formulation.
mRNA Construct and N/P Ratio
We used truncated mRNA strands consisting of the first 955 nucleotides from the 5′ end of the Comirnaty vaccine sequence, comprising approximately one-quarter of the full-length mRNA sequence, including modified residues. , To maintain a target IL nitrogen-to-mRNA phosphate (N/P) ratio of ∼6, the 10 mol % DSPC system contained three mRNA strands, while the 20, 30, and 40 mol % systems contained two.
Solvent Content
LNPs are typically created by mixing an aqueous mRNA solution with an ethanolic lipid solution at a 3:1 volume ratio. In our CG model, one CG solvent particle represents either 10 water molecules or 3 ethanol molecules. Accounting for these mapping ratios and relative molecular volumes of water and ethanol, we used a 2.93:1 ratio of water-to-ethanol CG particles to approximate the solution used in the experiments. While actual therapeutic LNP formulations are overwhelmingly composed of solvents, simulating such dilute solutions requires an impractical amount of computational resources. To reduce the simulation time, we simulated at concentrations roughly 12 times higher than those of the actual formulations. Even under these more concentrated conditions, solvent particles comprise roughly 90% of the system.
Simulation Scale and Rationale
Our experience with simulating LNPs suggests that somewhat smaller LNPs display much of the same structure and behavior as full-sized LNPs. The computational cost to simulate the evolution of an LNP scales with ∼d 4.5 where d is the nanoparticle diameter. Consequently, to reduce computational overhead and wall-clock time, we simulated systems containing enough lipids to form a single LNP with a diameter of ∼45 nm. While this is smaller than the ∼65 nm particles often found in LNP vaccines, this size is closer to the 50 nm diameters reported in Cullis’ studies. , Even at this reduced scale, the initial formulation and dialysis simulations require ∼4 months of continuous GPU time.
LNP Assembly and Dialysis Protocol
Initial system configurations for LNP self-assembly were constructed by randomly placing molecules within the simulation box. Larger molecules, which are more difficult to place in dense systems, were placed first. Specifically, mRNA strands were incorporated first using the tangled chain option for the Disordered System Builder in the Schrödinger Materials Science Suite (MS suite version 24-4). This method randomly places the initial mRNA CG particle for the chain and sequentially adds bonded particles one at a time to grow the chain while avoiding strong steric clashes with previously placed particles. Subsequently, the lipid molecules were introduced and grown using a similar procedure. Consistent with an acidic formulation environment, a 9:1 ratio of positively charged to neutral ionizable lipids, corresponding to a pH of approximately 5.14, was used in this step.
To avoid the computational expense of adding a large number of solvent particles sequentially, a tile and trim approach was employed. First, a small box containing a randomly placed, nonoverlapping mixture of water and ethanol particles at a 2.93:1 ratio was created using the Disordered System Builder. This small box was relaxed by using Brownian dynamics in the NVT ensemble at 10 K for 100 ps. The relaxed solvent box was then tiled over the system in all three dimensions. Subsequently, solvent particles overlapping with lipid or mRNA molecules were deleted, and excess solvent particles were randomly removed to attain the targeted number of water and ethanol particles given in Table . Please see the Supporting Information for additional details.
Following a brief 500 ps Brownian Dynamics relaxation simulation at 10 K, each system was simulated using Langevin dynamics for 4000 ns at 298.15 K in the NVT ensemble to permit the initial partial assembly of the LNP nanostructures before dialysis. The input files for these and all other dynamics simulations in the current work were created using the MD Multistage Workflow panel in Schrödinger Materials Science Suite (MS suite version 24–4) and run with Desmond. , The critical step for LNP formation is dialysis, during which the pH is typically adjusted from acidic conditions (pH of 5.14 for our system) up to the physiological level (pH 7.4), while ethanol is simultaneously replaced by water. To model this process, the partially assembled LNP systems underwent a seven-stage protocol. Each stage consisted of two steps: (1) adjustments to the composition, and (2) an NVT Langevin Dynamics simulation for 1000 ns at 298.15 K. Compositional adjustments involved replacing randomly chosen IL+ molecules with IL0 molecules to mimic the continuous shift in pH from 5.14 to 7.4, and replacing chosen ethanol particles with water particles randomly. The amount of ethanol present, the pH, and the fraction of IL+ for each stage are detailed in Table . For a more complete description of the dialysis process, please see the Supporting Information.
2. Staged Adjustment of LNP Solution Composition Simulating Dialysis .
| stage # | 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 |
|---|---|---|---|---|---|---|---|---|
| % of initial ethanol retained | 100 | 66.67 | 33.33 | 16.33 | 8.66 | 4.24 | 1.44 | 0 |
| pH | 5.14 | 5.31 | 5.61 | 5.91 | 6.18 | 6.46 | 6.84 | 7.4 |
| protonated fraction of IL | 0.9 | 0.858 | 0.752 | 0.6 | 0.45 | 0.3 | 0.15 | 0.047 |
The system transitions from the initial self-assembly state (Stage 0) through seven subsequent stages, progressively exchanging ethanol for water while IL+ molecules are neutralized. Throughout this process, the solution pH increases from 5.14 to 7.4, with complete ethanol replacement by the final Stage. By Stage 4, the vast majority of the ethanol has been replaced, and the protonated fraction of IL (IL+) has dropped to 45%, indicating that the solution pH is approximately the apparent pK a value for the IL.
Force Field Adjustment to Improve Encapsulation
In our initial test simulations at low pH (5.14) in the presence of ethanol, the mRNA was embedded in the lipid globules in the 10 and 20 mol % DSPC simulations. However, at 30 and 40 mol % of DSPC, the mRNA largely remained in the aqueous phase. Because this lack of early encapsulation would likely prevent mRNA incorporation during subsequent dialysis, we adjusted the automatically generated CG force field, a common practice when creating or extending CG force fields to capture specific behaviors. − We tested a number of modifications to the nonbonded parameters and then determined the degree of mRNA encapsulation for stage 0 in the 40 mol % DSPC system (see the Supporting Information for a detailed description and rationale for this process). Based upon this testing, we reduced the repulsive nonbonded interactions between the charged IL headgroup and mRNA particles by 20%. This specific reduction was the minimum amount necessary to ensure mRNA incorporation prior to dialysis, correctly mimicking the strong electrostatic complexation that leads to mRNA encapsulation.
Because this change is restricted to the interaction of the IL+ headgroup with mRNA, it is not expected to introduce effects beyond those associated with having increased initial encapsulation. Since experiments show a tendency for encapsulated mRNA to remain inside LNP structures, the higher level of encapsulation in the adjusted model is expected to persist beyond stage 0, potentially up to stage 7 even as the IL+ concentration drops considerably by that stage.
Results
Morphology Analysis: After Initial Self-Assembly in the Presence of Ethanol
Figure presents snapshots taken at the end of the 4 μs self-assembly simulations, before dialysis. All systems contain relatively irregular and dynamic structures approximately 60 nm in size, with the structures in the higher DSPC content systems appearing less compact. Despite displaying local bilayer-like characteristics, particularly at elevated DSPC levels, these assemblies do not form regular bilayers. Morphologically, structures at low DSPC levels are perfused by aqueous channels, whereas those at high DSPC levels have more continuous bilayers that enclose predominantly aqueous volumes. Consistent with all-atom MD simulations, the presence of ethanol results in more dynamic LNPs. In all cases, these assemblies encapsulate the mRNA with transient exposure at the particle surface.
2.

Final structures of the mRNA-LNPs from the 4000 ns CG simulations at pH 5.14 with ethanol present. The systems contain the initial ethanol concentration at varying levels of DSPC, shown across three independent replicates for each composition. The components are colored as follows: navy blue: IL+, green: IL0, light gray: DSPC, medium gray: cholesterol, orange: PEG-lipid, red: mRNA. Water and ethanol are present in the simulations but omitted from these images.
Morphology Changes during Dialysis
Overall Morphology
As illustrated in Figures and S3, substituting ethanol with water and neutralizing ionizable groups in the ILs resulted in a transition to a more compact, less dynamic LNP morphology consistent with experimental results. Most of this structural compaction occurred during Stages 1 and 2, coinciding with the removal of the majority of the ethanol and a significant decrease in unaggregated lipids. Overall, this structural evolution was strongly dependent on the DSPC content.
3.

Structural evolution of LNPs containing 10%, 20%, 30%, or 40% DSPC during dialysis. Snapshots depict the end of initial assembly (Stage 0) and the subsequent seven stages of ethanol–water exchange. One replicate is shown for each DSPC concentration. Images of the other replicates are provided in Figure S4. Color details are as provided in Figure .
With Low DSPC Content
Systems containing 10% and 20% DSPC formed bumpy spherical structures after the first ethanol-exchange stage. From stage 5 through the end of dialysis, all replicas matured into smooth compact particles with structural characteristics closely resembling those observed experimentally at low helper lipid concentrations by Cullis and co-workers ,,, and others, as well as our previous computational results for 10 wt % DSPC.
With High DSPC Content
Formulations with higher DSPC levels, 30 or 40 mol %, exhibited distinct morphologies that were more complex. For LNPs with 30% DSPC, bumpy circular structures appeared at Stage 4 and gradually smoothed by Stage 7, forming moderate-sized internal aqueous cavities, some containing mRNA, that resembled blebs. LNPs containing 40% DSPC exhibited asymmetric, elongated (“rice-like”) morphologies by Stage 6, characterized by multiple large aqueous cavities and distinct lipid-rich regions consistent with experimental observations at high levels of helper lipids. , This elongated structure persisted into Stage 7 after complete replacement of ethanol with water and the neutralization of most of the ILs. Notably, during elongation, nearly all of the LNPs established connections with periodic images (Two of the 40 mol % DSPC systems retained these connections at the end of Stage 7) due to the high concentrations of LNP components in our simulation box. However, the internal morphology of the LNPs retained the same DSPC-concentration-dependent trends regardless of these periodic connections.
Comparison with Experiment
Our DSPC simulation results align well with the structural changes observed in the experiment. ,, After dialysis, Cullis and co-workers found that while 10 and 20 wt % helper lipid formulations gave fairly compact spherical LNP structures, 30 mol % formulations produced a mixture of morphologies, including compact spheres, structures with blebs (as seen in our simulations), and complex structures with lipid cores surrounded by a large amount of water and embedded in spherical or elongated (rice-like) liposomal bilayer. At 40 mol % helper lipid, nearly all of the LNPs adopted the complex liposomal structures. While our 40 mol % simulations also capture significant elongation (please see the Supporting Information for a discussion of the potential influence of periodic boundary conditions) and water internalization, the volume of internalized water is less dramatic than in the experiments, ,,− and the segregated lipid-dominated regions remain in contact with the outer LNP wall. Additionally, while experimental LNPs contain only one very large internal aqueous compartment, multiple large aqueous cavities are present in our simulations (calculated cavity size distributions are available in Figure S15). We expect that with long enough simulations, likely orders of magnitude longer than presented here, these internal compartments would eventually merge. This is supported by exploratory simulations, ten times longer, of larger LNPs (see the Supporting Information for more information). We estimate that the average aspect ratio of experimentally observed LNPs at 40 wt % ESM is 1.7±0.2 (see the Supporting Information for more information). In our original simulations, the 40 mol % DSPC replicate that is not periodically connected at the end of stage 7 had an aspect ratio of 3.2; however, that value was decreasing during stage 7 and would be expected to decrease more if the simulation were continued (please see the discussion in the nonequilibrium state of LNPs below). Consistent with this, in the simulations of the larger LNPs reported in the Supporting Information, Replicate (#2) was the only simulation that formed a single, nonperiodically connected LNP, and that LNP had an aspect ratio of 1.4.
The 30 and 40 mol % formulations result in more irregular LNP morphologies and internal segregated aqueous regions, consistent with DSPC’s preference for planar lipid packing. , These formulations also produced larger LNPs at the end of dialysis than the corresponding low DSPC formulations. Despite some differences in gross morphology at the highest DSPC levels, our simulations captured the underlying physical driving forces and reproduced composition-dependent trends in LNP cavitation, size, and shape as observed in experiments. Interestingly, the structures we observe are similar to the Type III morphology found by Ling et al. at high ESM levels, in which the internal lipids adsorbed on the inner surface of a liposome-like structure.
LNP Solvent Content
We analyzed the LNP solvent content across the seven stages of dialysis (Figure ). As shown in Figure a, ethanol content dropped steadily and similarly across all DSPC levels, with the majority of it removed by Stage 4. The only exception to this steady drop is an increase in ethanol content for 40 mol % DSPC between stages 1 and 2. This is discussed below. For the 10 and 20 mol % DSPC formulations, the ethanol fraction within the LNP was enriched by approximately 2-fold relative to the bulk solution throughout stages 0 through 6. In contrast, while the 30 and 40 mol % formulations exhibited similar initial enrichment at stage 0, this factor dropped to approximately one by stage 6.
4.

Fractional (a) ethanol and (b) water content of the LNPs during dialysis for formulations containing 10, 20, 30, and 40 mol % DSPC. Fractional content is defined as the number of ethanol or water particles inside the LNP divided by the total number of particles (including lipid, mRNA, and solvent particles) inside the LNP.
Figure b illustrates that the LNP water content gradually increased similarly across all formulations up to Stage 4, with the exception of the 30 and 40 mol % DSPC formulations, which experienced dramatic increases (∼50% and >3-fold, respectively) in the amount of buried water between Stages 1 and 2. This increase was driven by the formation of bilayers with relatively little curvature. This dramatic increase in solvent content for the 40% DSPC LNPS also led to the peak in the ethanol content at Stage 2. The water content in the 10, 20 and 30 mol % DSPC LNPs steadily decreased in the later stages. In contrast, the amount of buried water in the 40 mol % DSPC formulation continued to increase until Stage 4 before leveling off at 40% of the particles present in the LNPs. These final water contents in the 30 and 40 mol % DSPC LNPs are approximately 2 and 7 times higher than the initial Stage 0 levels. Interestingly, this DSPC-dependent divergence in water content coincides with the removal of most of the ethanol and neutralization of ∼50% of the IL (i.e., near the apparent pK a value of the IL). This transition of low water content at low 10 and 20% helper lipid concentrations to high water content at high concentrations at pH 7 mirrors that found in experiments. ,,−
An examination of LNP morphologies at intermediate stages indicates that this increase arises from the formation of water-filled internal cavities. This behavior is most apparent in the LNP cross sections (Figure ) for the last stage in which the mRNA remained encapsulated at each DSPC level (see Figure S5 for cross sections of all systems at the end of Stage 7). This interpretation is supported by the fact that after Stage 1, nearly all of the lipids are associated with the LNPs. Consequently, the increase in the total water content of the LNPs for 30 and 40 mol % DSPC shown in Figure (as well as the related plot Figure S8 showing the total number of lipid and water particles in the LNP at the end of Stage 7) clearly reflects progressive incorporation of water and the subsequent swelling of the LNPs in the later stages of dialysis. From our analysis of the most closely related experimental system with suitable data, 40 mol % ESM, we estimate that the LNP is ∼80% water by volume (see the Supporting Information for additional information). Because the particles in our simulations were chosen to have similar volumes, the fraction of particles inside the LNP that are water should approximate the volume fraction of water inside the LNP. While the experimental value is considerably higher than our simulated value of ∼40% water for 40 mol % DSPC, this difference may arise from the differing lipids (ESM vs DSPC) and that, at least partially, our simulated LNPs are smaller and thus would be expected to have a smaller volume fraction of water. Nevertheless, the 40% water level in the 40 mol % DSPC system is qualitatively much higher than in more typical 10 mol % DSPC simulations (3.5% water). Additionally, the presence of water cavities in the simulated 40 mol % DSPC systems, far larger than necessary to just contain mRNA in narrow channels, demonstrates that the general morphological behavior in our simulations corresponds well with experiment.
5.

Snapshots of whole LNPs (top) and corresponding cross sections (bottom) for selected replicates at one of the last stages during dialysis in which mRNA remains encapsulated. Colors are as given in Figure .
At lower DSPC concentrations, Cullis et al. ,, observed compact LNPs with a disordered but overall homogeneous structure without water cavities, while at high concentrations, the LNPs were found to contain a lipid core floating in an aqueous environment, enclosed in a bilayer. The overall morphological dependence of our LNPs on DSPC concentration, depicted in Figure , aligns with the trends observed in experiments. ,,, We found compact LNPs at low DSPC concentrations and with large water cavities at higher concentrations. The main difference is that in our study, at the highest DSPC loadings, internal bilayers persist, and the internal lipid-dominated region remains attached to the outer shell of the LNP. This morphological deviation could be attributable to the smaller size of the LNPs in our study or the possibility that the systems have not reached a mature structural equilibrium within the time simulated.
Solvent-Exposed and Buried Regions
The interesting pattern of hydration and morphological evolution motivates a more detailed examination of the water–lipid interactions and the composition of the lipids within distinct LNP regions across different DSPC levels. We classified lipids as either exposed or buried, based upon the number of nearby lipid particles (Figure ; see the Supporting Information for details of the classification scheme). There is a clear separation between exposed regions, which consist of water-facing bilayers or monolayers, and the buried regions. In most LNPs, a single buried region tends to dominate (see Figure S5). To facilitate future comparisons, we estimate that the thickness of the bilayers in the outer surface of the LNPs at 40 mol % DSPC is approximately 5.1 ± 1.0 nm.
6.

Cross section of a 40% DSPC LNP postdialysis. Lipids are classified and rendered based upon solvent exposure: exposed lipids are shown with a space-filling representation, while buried lipids are depicted with thick tubes. Colors are as given in Figure .
The mol percentage of water-exposed lipids across Stages 0–7 is plotted in Figure . As expected, when the initial maximum level of ethanol is present, nearly all lipids are exposed across all DSPC levels, since the ethanol acts as a compatibilizer between lipids and water. The removal of the majority of the ethanol by Stages 3 and 4 drives reduction in the fraction of exposed lipids, particularly for lower DSPC concentrations. Superimposed on this process is the general burial of ILs as they undergo neutralization – a process that is roughly 50% complete around Stage 4, and also reduces overall solvent exposure. Lipid–water exposure correlates with the DSPC concentration, driven by the lipid’s surface activity and bilayer-forming propensity; , notably, the majority of the lipids remain solvent-exposed in the 40 mol % DSPC formulation. After dialysis, the fraction of exposed lipids decreases, corresponding to the experimentally observed increase in core size, as DSPC levels are lowered.
7.

Percentage of lipids that are solvent exposed during the dialysis process for formulation stages containing 10, 20, 30, and 40 mol % DSPC.
Figure details the composition of buried versus exposed regions, broken down by lipid type, for each DSPC loading level at the end of Stage 7. It is clear that the buried (IL0 > cholesterol > DSPC) and exposed (DSPC > cholesterol > IL0) regions differ significantly in composition. As may be expected, given its preference for bilayer-like structures, DSPC concentrates in the exposed regions. In the 10 mol % DSPC formulation, DSPC constitutes ∼41 mol % of the exposed lipids, representing a concentration enhancement of over 4-fold. A common assumption for these “normal” LNP compositions is that most of the DSPC resides in the shell, ,− although some modeling studies suggest that there is significant DSPC content in the LNP core. , Based upon the fraction of all lipids located in the shell (13.3%) and the DSPC fractions in the core (4.5 mol %) and the shell (41 mol %), we calculate that 58% of all DSPC molecules are located in the shell. Thus, our calculations suggest that while the shell is very enriched in DSPC and contains the majority of these molecules, a significant fraction (42%) is distributed within the core even though they constitute a small percentage of the core lipids.
8.

Percentage of exposed and buried lipid molecules, categorized by lipid type at the end of the dialysis process (Stage 7).
As the overall DSPC level increases to 40 mol %, the DSPC content of the exposed lipids slowly rises to just over 50%. Simultaneously, the cholesterol and net IL (IL+ + IL0) contents remain fairly constant, while the fraction of PEG lipids steadily drops because the exposed lipids constitute a progressively larger fraction of the LNP. In the buried region, the IL0 makes up just under half of the lipids present, followed by cholesterol and DSPC for all formulations. As the total DSPC content rises, the fraction of the buried lipids that are IL0 or cholesterol steadily drops, while the buried DSPC fraction steadily rises from 4 mol % to 26 mol %. The ratio of the fraction of lipids that are DSPC in the exposed and buried regions is more than 10:1 in the traditional (10 mol % DSPC overall) formulation, consistent with a common assumption in the literature. ,− However, this ratio drops to approximately 2:1 at the highest overall DSPC level. Under all conditions, the fraction of the exposed lipids that are IL+ is consistently higher than the IL+ fraction of the buried regions.
Figure S6 depicts the compositional variation of the exposed and buried lipid regions across all stages. Before dialysis, the compositions of the exposed and buried regions are qualitatively similar. As dialysis proceeds, several notable trends emerge, which are strongest at low DSPC levels, including:
The gradual transfer of IL (the sum of IL0 and IL+) present in the exposed regions occurs into the buried regions.
An increase in the exposed DSPC content.
Reduction of exposed cholesterol.
Lipid–Water Contacts
Figure a displays the average number of water particles in contact with each lipid particle, evaluated for all lipids as well as the exposed or buried subsets, at the end of the dialysis process as a function of the DSPC level. The steady drop in the hydration of exposed lipids as the DSPC content increases reflects the growing number of internal water cavities within the LNPs. Because these internal cavities are concave, they geometrically limit the water exposure for the surrounding lipids. Additionally, the tendency for DSPC to form compact bilayers likely also restricts the water contact. Interestingly, the water content of the buried region follows the inverse trend, rising as the total DSPC concentration increases. Figure b shows that this increase is driven by the formation of progressively larger water clusters within the buried regions. Figure reveals that these water clusters are largely surrounded by DSPC. Similar results were found by Garaizar et al. This suggests that a higher buried DSPC concentration, which favors lower curvature interfaces, directly facilitates increased water incorporation within the buried region.
9.

Water particles in contact with lipid particles at the end of the dialysis process (Stage 7). (a) Average number of contacting water particles per lipid particle, evaluated for all lipids and broken down by the exposed or buried fractions. (b) Snapshots highlighting the water particles (cyan) in direct contact with buried lipid particles.
10.

DSPC lipid structure and internal water distribution within the LNPs. Colors are as given in Figure except that the DSPC headgroups are colored purple and water is colored cyan. (a) Renderings of system 40–3. The mRNA is hidden for clarity, water is represented by small spheres, and DSPC is shown in space-filling representation. The left image displays the whole LNP, while the right provides an expanded view of the predominant buried lipid region. (b) Similar whole LNP views for one replicate at each DSPC loading level with the mRNA visible.
Monolayer versus Bilayer Shell
Different studies indicate that the outer shell can be either a monolayer ,,, or a bilayer ,, or more complex structures. Figure highlights the DSPC molecules and colors the headgroup purple. For DSPC located in narrow lipid layers between aqueous regions, present at 20, 30, and 40 mol % DSPC, the structure is clearly bilayer-like. Conversely, for surfaces separating aqueous regions from lipid regions, the DSPC arrangement is consistent with a monolayer, with the headgroups facing the aqueous side. These monolayers are most noticeable in the 10 mol % DSPC system; however, such interfaces are present at all of the DSPC levels. Our results indicate that the nature of the outer shell can be complex, with both forms present in the same LNP, and that it is dependent on the lipid composition, consistent with experiments.
mRNA Retention
Figure plots the percentage of encapsulated mRNA for all DSPC concentrations across the dialysis stages (raw data in Table S3). Because this study utilizes 3 replicates, each containing only 2–3 mRNA molecules (depending on the DSPC level), the data have substantial interreplicate variation, which limits the precision of the average values. As noted previously, our initial automatically generated force field resulted in predialysis encapsulation for the 10 and 20 mol % DSPC systems but failed to encapsulate for the 30 and 40 mol % formulations. To resolve this, we minimally adjusted the interactions between the headgroup of IL+ and mRNA particles to effectively make them more attractive to attain 100% encapsulation in the predialysis simulations. However, as dialysis progressed, premature mRNA release still occurred, dropping to an average of 46% encapsulation by stage 7, well below the approximately >95% encapsulation rate observed in the experiment.
11.

Percent of mRNA encapsulated averaged over the replicates as a function of DSPC content and dialysis stage.
Despite this model limitation, the release levels in Figure offer valuable physical insights. Initially, formulations with higher DSPC concentrations lose mRNA more rapidly than those with lower concentrations, suggesting an early-stage incompatibility between the DSPC and mRNA.
Yet, after stage 4, mRNA retention largely stabilizes in the 30 and 40 mol % systems, likely due to the formation of robust, DSPC-dominated bilayers that more securely enclose mRNA in aqueous compartments. Conversely, the 10 and 20 mol % formulations, which exhibited robust retention early on, experience near-complete mRNA loss during the final stages of dialysis. Because these final stages are characterized by the neutralization of the ILs, this late-stage release indicates that our force field might be improved by effectively making the interactions between the IL0 molecules and the mRNA more attractive.
Nonequilibrium State of LNPs
Experimental LNP systems have been described as metastable ,, and can evolve chemically and physically beyond the production window, which is why they are often stored at reduced temperatures. Additionally, the literature notes variation in the LNP structures as a function of processing conditions. − Surfactant and lipid systems inherently evolve via a range of phenomena, each having a characteristic time scale ranging from fast (e.g., local conformational and association) through intermediate (e.g., LNP initial formation) and slow (e.g., LNP internal structure organization) to very slow (e.g., LNP size distribution evolution), overall spanning many orders of magnitude of time from ns to hours or longer. Dialysis is a nonequilibrium process, and experimental and manufacturing time scales postdialysis during which the LNP structure can evolve are typically much longer than simulated in this study. Because some experimental time scales can be much longer than practical simulations can follow, our computational results should be regarded as snapshots of system behavior rather than representations of a final equilibrium state.
While our simulations reproduce critical aspects of LNP structure evolution, it is worth characterizing their evolution during the final stage, stage 7. To assess potential continued system evolution, Figures S13, S14, and S16 plot the LNP radius of gyration (R g), LNP water content, and largest water cluster within the LNP as a function of time for Stage 7. During the last half of Stage 7, most of the replicates at all levels of DSPC display no discernible trend. One exception is a weak downward trend in R g in all three replicates for 40 mol % of DSPC. The other exception is a weak and decaying downward trend in the water content of the LNPs for 30 and 40 mol % DSPC. The downward trend in R g combined with the relatively stable water content and size of the largest water cluster is consistent with the LNPs slowly evolving toward a less asymmetric form with time.
Discussion/Outlook
Our simulations reproduce a variety of structural features consistent with prior experimental observations and computational predictions, including:
Morphology change during dialysis: The transition from dynamic, irregular globules of lipids and mRNA to more well-defined, less dynamic structures
DSPC localization: The enrichment of DSPC on the surfaces of water-exposed regions, both internally and on the LNP exterior. ,−
Lipid burial: The proportion of the lipids buried drops with an increasing DSPC level.
The nature of the lipid shell: The outer shell is a monolayer for water–lipid interfaces and bilayers for narrow lipid membranes between aqueous regions.
Internal hydration: Significant amounts of LNP-encapsulated water, alongside the formation of enclosed aqueous regions at high DSPC that exceed the amount needed to house the mRNA. ,,−
Postdialysis morphology: Elongated LNP geometries at high DSPC concentrations, as well as the formation of bleb-like structures (some of which contain mRNA ,,− ).
Compositional stratification: Distinct compositions for exposed vs. buried lipids. ,,,
We also find evidence for:
DSPC-mRNA Incompatibility: We observed a distinct spatial incompatibility between DSPC and mRNA in the same phase.
Late-stage swelling: significant swelling of LNPs late in the dialysis process due to incorporation of large amounts of water.
Higher levels at all stages and large increases in the DSPC content in the exposed regions late in dialysis for all formulations.
Encapsulation stability: evidence for enhanced encapsulation stability of mRNA in high DSPC formulations due to the influence of the enclosing shell, which is expected to form a more robust barrier as DSPC increases.
Core–shell partitioning during dialysis: DSPC migrates to the exposed LNP regions, while cholesterol and IL0 accumulate in the buried regions as dialysis progresses.
We observe a greater degree of mRNA loss from the LNPs at higher pH values, particularly at lower DSPC concentrations (10 and 20 mol %), than found in experiments. , This is likely due to compositional and sampling limitations in the all-atom training system used as a reference when constructing the CG model. Although this limits our ability to study specific mRNA behaviors, mRNA constitutes only a small fraction of the LNP composition (about 1.5% of the nonsolvent particles in our simulations). As well, at the high DSPC levels, retained mRNA is dissolved in aqueous enclosures. Given this small mass fraction, alongside the model’s successful replication of many experimental LNP features, we suggest that our simulations capture sufficient overall LNP behavior to support useful qualitative characterizations and, in some cases, provide molecular-level insights that are difficult to observe directly in experiment. In principle, the discrete dialysis process may also contribute to excessive mRNA release; however, our previous observations in other studies showed that mRNA release is insensitive to the number of steps used to change the IL+ to IL0 ratio from values consistent with pH ∼5–7.4 in the absence of ethanol.
The water content of the LNP after dialysis changes dramatically as the DSPC content rises, particularly for 30 and 40 mol % DSPC formulations. This leads to the formation of large water-filled cavities, consistent with the experimental findings. ,,− This swelling occurs when the majority of the ethanol has been removed and the fraction of IL that is positively charged drops below 50% (i.e., pH ≃ the apparent pK a). At 40 mol % DSPC, the change during dialysis amounts to roughly an 8-fold increase in internal water content. Interestingly, as the overall DSPC content rises, both its concentration and that of water increase inside the LNP. Much of this behavior can be attributed to DSPC’s strong tendency to form lamellar, particularly bilayer, structures in water. This tendency, combined with DSPC’s high melting temperature (54 °C), yields highly stable structures. Furthermore, helper lipids, such as DSPC, are known to be mildly incompatible with neutralized ionizable lipids ,− (IL0), meaning that only a limited amount of it can be accommodated in the outer shell of a conventional LNP. This incompatibility drives the formation of standalone bilayers characterized by high DSPC content and low IL0 content. The high DSPC content is expected to favor bilayers with relatively small curvature. Experimentally, those stiff bilayers manifest as a large liposomal-like shell enclosing a volume that is too large to consist solely of lipids and mRNA with limited hydration; this results in the formation of a largely aqueous core containing a separated IL0-rich aggregate. In these experimental systems, the encapsulated mRNA resides in the aqueous region, whose volume far exceeds that of typical blebs. ,,,,
In our simulations under similar formulation conditions, much of the LNP surface is composed of a DSPC-dominated bilayer with water on both sides, directly capturing the formation of large, bilayer-enclosed internal water cavities. Furthermore, the mRNA that was not released early in dialysis remains well encapsulated within these aqueous regions. We also observe the formation of an IL0-rich lipid “core” inside the LNP; however, unlike in the experimental studies, this core and the outer bilayer shell do not fully separate. Additionally, we observe internal DSPC-rich bilayers subdividing the aqueous regions, and where these bilayers contact the lipid core, they generally do so perpendicularly to the surface of the core. Overall, our results successfully capture the significant swelling driven by DSPC-dominated bilayer formation and the large-scale incorporation of water inside the LNP (an 8-fold increase) where the mRNA resides. Crucially, these large-scale behaviors emerged spontaneously and were not explicitly parametrized into the model. Exploratory simulations of larger systems with 40 mol % DSPC with longer Stage 7 reported in the Supporting Information are largely consistent with those examined in detail in the paper. Given this, it seems unlikely that the discrepancy regarding the persistence of contacts between the lipid core and the bilayer, as well as the subdivision of the core by bilayers, may arise because our simulated systems contain less lipid and mRNA than needed to form the fully sized LNPs observed experimentally. Instead, these discrepancies may reflect limitations in the current CG model or a need for simulation times orders of magnitude longer to allow for the progressively slower relaxation processes that emerge as an LNP matures, or both, which is consistent with the trends in LNP R g and water content noted in Stage 7 for higher DSPC levels. We intend to explore size and temporal scale effects in future studies.
The CG model was trained at 310.15 K, since we intend to use it to study endosomal escape. Since LNP production is typically performed close to room temperature, the current study examines LNP formation during dialysis at 298.15 K. While we have not explicitly explored the model’s sensitivity to this temperature difference, we expect that the underlying behavior would remain consistent if the model were trained at 298.15 K. This assumption is supported by the observation of dramatically higher LNP water content for 30 and 40 mol % DSPC, consistent with experiments.
With current computational resources, calculations using this model typically require months of GPU time, even for LNPs somewhat smaller than the full-sized particles observed in experiments. Furthermore, simulations extending significantly beyond 10 μs, or systems containing multiple LNPs, consume prohibitively large amounts of wall-clock time for routine use. While training against atomistic models provides a general mechanism for parametrizing models for systems with a broad range of components, it is computationally expensive and relies on the accuracy of the atomistic force field to capture fairly subtle interactions in solution. Consequently, the parameters used for training (such as number of independent reference runs, system size, composition, simulation time, the number of iterations in the CG fitting process) should be chosen carefully to ensure sufficient exploration of the relevant structural variations and adequate refinement of the CG model.
Additionally, further work to establish the best practices for training CG models is warranted. As demonstrated in the current study, the original force field required adjustment to achieve mRNA encapsulation at low pH, and those adjustments were ultimately insufficient to reproduce the observed high levels of mRNA sequestration inside LNPs at pH 7.4. The soft nonbonded potentials and large mapping scale may limit transferability, particularly for temperature changes (exceeding 20 K or more). The current model has been tested and employed only at 298.15 K; applying it to significantly different temperatures would likely require substantial retraining. Furthermore, our simplified model for RNA cannot predict secondary structure; instead, it focuses on capturing broader conformational differences such as extended versus compact states. Transferability to other environments, such as those lacking water, may also be a challenge. Ultimately, as with all CG models, it is critical to benchmark the results against known quantities to establish the reliability of the model before using it to predict new phenomena.
Despite the limitations outlined above, our results, combined with those of an earlier study, demonstrate the ability of this modeling approach to reproduce large-scale changes in the LNP behavior as a function of composition. Beyond the characterization of these morphologies, this approach also enables the characterization of the local composition and structure within distinct regions of the LNP, providing insight into the factors governing these behaviors.
We expect this capability to extend to understanding how processing conditions affect the LNP structure. This understanding could aid in improving morphological control in LNP production (e.g., decreasing the diversity of structures found in a given formulation), optimization of encapsulation efficiency, and enhancing LNP stability during processing and storage (e.g., freeze–thaw cycling or lyophilization and reconstitution). Active LNP targeting approaches , may also benefit, concerning the timing at which targeting entities are introduced during production, their retention on the particle, and the extent of their exposure on the LNP surface. This class of simulation could inform our understanding of LNP-biological interactions, including targeting-ligand engagement with cell-surface receptors, interactions with endogenous serum proteins, ,, and endosomal escape. ,
The delineation of the local composition within an LNP provided by this approach is complementary to that of other classes of modeling. For instance, more detailed CG or AA models can be used to provide higher-resolution characterizations and more accurate property predictions, particularly when starting from a more informed selection of initial conditions, perhaps via back-mapping approaches.
LNP technology has had a profound impact, ,,, and its potential for further applications to human health is enormous. − However, the LNP design remains a complex and challenging endeavor. Consequently, advancing the field requires integrating complementary characterizations; combining input from multiple experimental techniques (DLS, SANS, SAXS, NMR, electron microscopy, etc. , ) with a diverse range of modeling approaches will likely be the most effective strategy to rapidly advance this field. ,,,
We successfully simulated the structural shift in LNPs at higher DSPC concentrations. Our analysis suggests that the distinct change in behavior observed as the ethanol is mostly removed and the pH rises from 5.14 to physiological pH (7.4) is driven by the formation of DSPC-rich bilayers. Future studies are planned to refine the best practices for constructing CG models, simulate larger LNPs, and investigate endosomal escape.
Supplementary Material
Input data files and analysis scripts are available on Zenodo: 10.5281/zenodo.20454629 and DOI: 10.5281/zenodo.18636861. A complete list of these files is given in the Supporting Information.
The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jcim.6c02185.
Descriptions of the simulation and analysis methods as well as additional results. (PDF)
The authors declare the following competing financial interest(s): Desmond, Maestro Materials Science, and Materials Coarse-Grained are products sold by Schrodinger, Inc. All authors performed this research as Schrodinger Employees and may own Schrodinger stock or stock options.
References
- Watson O. J., Barnsley G., Toor J., Hogan A. B., Winskill P., Ghani A. C.. Global Impact of the First Year of COVID-19 Vaccination: A Mathematical Modelling Study. Lancet Infect. Dis. 2022;22(9):1293–1302. doi: 10.1016/S1473-3099(22)00320-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- The Untapped Potential of mRNA–Lipid Nanoparticles. Nat. Rev. Bioeng. 2025, 3 (9), 715 10.1038/s44222-025-00362-x. [DOI] [Google Scholar]
- Huang X., Ma Y., Ma G., Xia Y.. Unlocking the Therapeutic Applicability of LNP-mRNA: Chemistry, Formulation, and Clinical Strategies. Research. 2024;7:0370. doi: 10.34133/research.0370. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Wu Z., Li T.. Nanoparticle-Mediated Cytoplasmic Delivery of Messenger RNA Vaccines: Challenges and Future Perspectives. Pharm. Res. 2021;38(3):473–478. doi: 10.1007/s11095-021-03015-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Bitounis D., Jacquinet E., Rogers M. A., Amiji M. M.. Strategies to Reduce the Risks of mRNA Drug and Vaccine Toxicity. Nat. Rev. Drug Discovery. 2024;23(4):281–300. doi: 10.1038/s41573-023-00859-3. [DOI] [PubMed] [Google Scholar]
- Cheng M. H. Y., Zhang Y., Fox K., Leung J., Strong C., Kang E., Chen Y., Tong M., Bommadevara H., Jan E., Ip O. Y. L., Rodríguez-Rodríguez C., Saatchi K., Häfeli U. O., Abdolahzadeh A., Witzigmann D., Cullis P. R.. Liposomal Lipid Nanoparticles for Extrahepatic Delivery of mRNA. Nat. Commun. 2025;16(1):4135. doi: 10.1038/s41467-025-58523-w. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Schoenmaker L., Witzigmann D., Kulkarni J. A., Verbeke R., Kersten G., Jiskoot W., Crommelin D. J. A.. mRNA-Lipid Nanoparticle COVID-19 Vaccines: Structure and Stability. Int. J. Pharm. 2021;601:120586. doi: 10.1016/j.ijpharm.2021.120586. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Packer M., Gyawali D., Yerabolu R., Schariter J., White P.. A Novel Mechanism for the Loss of mRNA Activity in Lipid Nanoparticle Delivery Systems. Nat. Commun. 2021;12(1):6777. doi: 10.1038/s41467-021-26926-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Chander N., Basha G., Yan Cheng M. H., Witzigmann D., Cullis P. R.. Lipid Nanoparticle mRNA Systems Containing High Levels of Sphingomyelin Engender Higher Protein Expression in Hepatic and Extra-Hepatic Tissues. Mol. Ther. Methods Clin. Dev. 2023;30:235–245. doi: 10.1016/j.omtm.2023.06.005. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Jayaraman M., Ansell S. M., Mui B. L., Tam Y. K., Chen J., Du X., Butler D., Eltepu L., Matsuda S., Narayanannair J. K., Rajeev K. G., Hafez I. M., Akinc A., Maier M. A., Tracy M. A., Cullis P. R., Madden T. D., Manoharan M., Hope M. J.. Maximizing the Potency of siRNA Lipid Nanoparticles for Hepatic Gene Silencing in Vivo. Angew. Chem., Int. Ed. 2012;51(34):8529–8533. doi: 10.1002/anie.201203263. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Tenchov R., Bird R., Curtze A. E., Zhou Q.. Lipid Nanoparticles – From Liposomes to mRNA Vaccine Delivery, a Landscape of Research Diversity and Advancement. ACS Nano. 2021;15(11):16982–17015. doi: 10.1021/acsnano.1c04996. [DOI] [PubMed] [Google Scholar]
- Udepurkar A., Devos C., Sagmeister P., Destro F., Inguva P., Ahmadi S., Boulais E., Quan Y., Braatz R. D., Myerson A. S.. Structure and Morphology of Lipid Nanoparticles for Nucleic Acid Drug Delivery: A Review. ACS Nano. 2025;19(23):21206–21242. doi: 10.1021/acsnano.4c18274. [DOI] [PubMed] [Google Scholar]
- Kafle U., Truong H. Q., Nguyen C. T. G., Meng F.. Development of Thermally Stable mRNA-LNP Delivery Systems: Current Progress and Future Prospects. Mol. Pharmaceutics. 2024;21(12):5944–5959. doi: 10.1021/acs.molpharmaceut.4c00826. [DOI] [PubMed] [Google Scholar]
- Mo Y., Zheng G.. From Morphology to Mechanism: Cryo-Electron Microscopy Insights into Lipid Nanoparticles for RNA Delivery. ACS Nano. 2026;20(28):19906–19929. doi: 10.1021/acsnano.6c09354. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Kulkarni J. A., Witzigmann D., Leung J., Tam Y. Y. C., Cullis P. R.. On the Role of Helper Lipids in Lipid Nanoparticle Formulations of siRNA. Nanoscale. 2019;11(45):21733–21739. doi: 10.1039/C9NR09347H. [DOI] [PubMed] [Google Scholar]
- Ling D., Li N., Pan X., Gao W., Li H., Tian Q., Li J., Cheng Q., Tang L., Zhang J., Gao P., Ying B., Wang C.. A New Type of Lipid Nanoparticles with Sphingomyelin-Induced Active Lipid Raft Structures for Improved mRNA Cellular Uptake. J. Controlled Release. 2026;391:114640. doi: 10.1016/j.jconrel.2026.114640. [DOI] [PubMed] [Google Scholar]
- Kulkarni J. A., Darjuan M. M., Mercer J. E., Chen S., van der Meel R., Thewalt J. L., Tam Y. Y. C., Cullis P. R.. On the Formation and Morphology of Lipid Nanoparticles Containing Ionizable Cationic Lipids and siRNA. ACS Nano. 2018;12(5):4787–4795. doi: 10.1021/acsnano.8b01516. [DOI] [PubMed] [Google Scholar]
- Grzetic D. J., Hamilton N. B., Shelley J. C.. Coarse-Grained Simulation of mRNA-Loaded Lipid Nanoparticle Self-Assembly. Mol. Pharmaceutics. 2024;21(9):4747–4753. doi: 10.1021/acs.molpharmaceut.4c00216. [DOI] [PubMed] [Google Scholar]
- Massotti V., Buffo A., Pisano R.. Molecular Drivers of Self-Assembly in RNA-Loaded Lipid Nanoparticles Revealed by Coarse-Grained Simulations. Int. J. Pharm. 2026;688:126449. doi: 10.1016/j.ijpharm.2025.126449. [DOI] [PubMed] [Google Scholar]
- Garaizar A., Díaz-Oviedo D., Zablowsky N., Rissanen S., Köbberling J., Sun J., Steiger C., Steigemann P., Mann F. A., Meier K.. Toward Understanding Lipid Reorganization in RNA Lipid Nanoparticles in Acidic Environments. Proc. Natl. Acad. Sci. U.S.A. 2024;121(45):e2404555121. doi: 10.1073/pnas.2404555121. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Lu C., Wu C., Ghoreishi D., Chen W., Wang L., Damm W., Ross G. A., Dahlgren M. K., Russell E., Von Bargen C. D., Abel R., Friesner R. A., Harder E. D.. OPLS4: Improving Force Field Accuracy on Challenging Regimes of Chemical Space. J. Chem. Theory Comput. 2021;17(7):4291–4300. doi: 10.1021/acs.jctc.1c00302. [DOI] [PubMed] [Google Scholar]
- Berendsen, H. J. C. et al. Interaction Models for Water in Relation to Protein Hydration. In Intermolecular Forces; Pullman, B. , Ed.; Springer: Dordrecht, 1981; Vol. 14. [Google Scholar]
- Reith D., Pütz M., Müller-Plathe F.. Deriving Effective Mesoscale Potentials from Atomistic Simulations. J. Comput. Chem. 2003;24(13):1624–1636. doi: 10.1002/jcc.10307. [DOI] [PubMed] [Google Scholar]
- Groot R. D., Warren P. B.. Dissipative Particle Dynamics: Bridging the Gap between Atomistic and Mesoscopic Simulation. J. Chem. Phys. 1997;107(11):4423–4435. doi: 10.1063/1.474784. [DOI] [Google Scholar]
- PubChem . Novel lipids and lipid nanoparticle formulations for delivery of nucleic acids. WIPO Patent WO2017075531A1, 2026.
- Jeong, D.-E. ; McCoy, M. ; Artiles, K. L. ; Ilbay, O. ; Fire, A. ; Nadeau, K. ; Park, H. R. ; Betts, B. ; Boyd, S. ; Hoh, R. ; Shoura, M. . Assemblies of Putative SARS-CoV2-Spike-Encoding mRNA Sequences for Vaccines BNT-162b2 and mRNA-1273.
- Bowers, K. J. ; Chow, D. E. ; Xu, H. ; Dror, R. O. ; Eastwood, M. P. ; Gregersen, B. A. ; Klepeis, J. L. ; Kolossvary, I. ; Moraes, M. A. ; Sacerdoti, F. D. ; Salmon, J. K. ; Shan, Y. ; Shaw, D. E. In Scalable Algorithms for Molecular Dynamics Simulations on Commodity Clusters, SC’06: Proceedings of the 2006 ACM/IEEE Conference on Supercomputing, 2006; pp 43–43. [Google Scholar]
- Schrödinger . Maestro-Desmond Interoperability Tools, Schrödinger Release 2026-2: Desmond Molecular Dynamics System; D. E. Shaw Research: New York, NY, 2024. [Google Scholar]
- Javanainen M., Martinez-Seara H., Vattulainen I.. Excessive Aggregation of Membrane Proteins in the Martini Model. PLoS One. 2017;12(11):e0187936. doi: 10.1371/journal.pone.0187936. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Thomasen F. E., Pesce F., Roesgaard M. A., Tesei G., Lindorff-Larsen K.. Improving Martini 3 for Disordered and Multidomain Proteins. J. Chem. Theory Comput. 2022;18(4):2033–2041. doi: 10.1021/acs.jctc.1c01042. [DOI] [PubMed] [Google Scholar]
- Grünewald F., Punt M. H., Jefferys E. E., Vainikka P. A., König M., Virtanen V., Meyer T. A., Pezeshkian W., Gormley A. J., Karonen M., Sansom M. S. P., Souza P. C. T., Marrink S. J.. Martini 3 Coarse-Grained Force Field for Carbohydrates. J. Chem. Theory Comput. 2022;18(12):7555–7569. doi: 10.1021/acs.jctc.2c00757. [DOI] [PMC free article] [PubMed] [Google Scholar]
- R Tejedor A., Aguirre Gonzalez A., Maristany M. J., Chew P. Y., Russell K., Ramirez J., Espinosa J. R., Collepardo-Guevara R.. Chemically Informed Coarse-Graining of Electrostatic Forces in Charge-Rich Biomolecular Condensates. ACS Cent. Sci. 2025;11(2):302–321. doi: 10.1021/acscentsci.4c01617. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Hardianto A., Muscifa Z. S., Widayat W., Yusuf M., Subroto T.. The Effect of Ethanol on Lipid Nanoparticle Stabilization from a Molecular Dynamics Simulation Perspective. Molecules. 2023;28(12):4836. doi: 10.3390/molecules28124836. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Kamanzi A., Zhang Y., Gu Y., Liu F., Berti R., Wang B., Saadati F., Ciufolini M. A., Kulkarni J., Cullis P., Leslie S.. Quantitative Visualization of Lipid Nanoparticle Fusion as a Function of Formulation and Process Parameters. ACS Nano. 2024;18(28):18191–18201. doi: 10.1021/acsnano.3c12981. [DOI] [PubMed] [Google Scholar]
- Leung A. K. K., Hafez I. M., Baoukina S., Belliveau N. M., Zhigaltsev I. V., Afshinmanesh E., Tieleman D. P., Hansen C. L., Hope M. J., Cullis P. R.. Lipid Nanoparticles Containing siRNA Synthesized by Microfluidic Mixing Exhibit an Electron-Dense Nanostructured Core. J. Phys. Chem. C. 2012;116(34):18440–18450. doi: 10.1021/jp303267y. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Cheng M. H. Y., Leung J., Zhang Y., Strong C., Basha G., Momeni A., Chen Y., Jan E., Abdolahzadeh A., Wang X., Kulkarni J. A., Witzigmann D., Cullis P. R.. Induction of Bleb Structures in Lipid Nanoparticle Formulations of mRNA Leads to Improved Transfection Potency. Adv. Mater. 2023;35(31):e2303370. doi: 10.1002/adma.202303370. [DOI] [PubMed] [Google Scholar]
- Mabrey S., Sturtevant J. M.. Investigation of Phase Transitions of Lipids and Lipid Mixtures by Sensitivity Differential Scanning Calorimetry. Proc. Natl. Acad. Sci. U.S.A. 1976;73(11):3862–3866. doi: 10.1073/pnas.73.11.3862. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Sebastiani F.. et al. Apolipoprotein E Binding Drives Structural and Compositional Rearrangement of mRNA-Containing Lipid Nanoparticles. ACS Nano. 2021;15:6709–6722. doi: 10.1021/acsnano.0c10064. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Yanez Arteta M., Kjellman T., Bartesaghi S., Wallin S., Wu X., Kvist A. J., Dabkowska A., Székely N., Radulescu A., Bergenholtz J., Lindfors L.. Successful Reprogramming of Cellular Protein Production through mRNA Delivered by Functionalized Lipid Nanoparticles. Proc. Natl. Acad. Sci. U.S.A. 2018;115(15):E3351–E3360. doi: 10.1073/pnas.1720542115. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Li M., Schroder R., Ozuguzel U., Corts T. M., Liu Y., Zhao Y., Xu W., Ling J., Templeton A. C., Chaudhuri B., Gindy M., Wagner A., Su Y.. Molecular Insight into Lipid Nanoparticle Assembly from NMR Spectroscopy and Molecular Dynamics Simulation. Mol. Pharmaceutics. 2025;22(4):2193–2212. doi: 10.1021/acs.molpharmaceut.4c01437. [DOI] [PubMed] [Google Scholar]
- Guevara M. L., Persano F., Persano S.. Advances in Lipid Nanoparticles for mRNA-Based Cancer Immunotherapy. Front Chem. 2020;8:589959. doi: 10.3389/fchem.2020.589959. [DOI] [PMC free article] [PubMed] [Google Scholar]
- PubChem . Limit size lipid nanoparticles and related methods. WIPO Patent WO2013059922A1, 2026.
- Li H., Song P., Li Y., Tu S., Mehmood M., Chen L., Li N., Tian Q.. Mesoscopic Structure of Lipid Nanoparticles Studied by Small-Angle X-Ray Scattering: A Spherical Core-Triple Shell Model Analysis. Membranes. 2025;15(5):153. doi: 10.3390/membranes15050153. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Larson N. R., Hu G., Wei Y., Tuesca A. D., Forrest M. L., Middaugh C. R.. pH-Dependent Phase Behavior and Stability of Cationic Lipid–mRNA Nanoparticles. J. Pharm. Sci. 2022;111(3):690–698. doi: 10.1016/j.xphs.2021.11.004. [DOI] [PubMed] [Google Scholar]
- Sanchez-Fernandez, A. ; Donnelly, K. A. ; Bolinsson, H. ; Börjesdotter, A.-M. ; Rønnemoes Bobak, T. ; Erlendsson, S. ; Mohammadi-Zerankeshi, M. ; AboulFotouh, K. ; Kawelah, M. R. ; Herranz-Trillo, F. ; Schagerlöf, H. ; Capasso Palmiero, U. ; Huus, K. ; Johnston, K. P. ; Marras, A. E. ; Nilsson, L. . Structural Heterogeneity in mRNA-LNP Subpopulations Revealed by AF4-SAXS: Implications for Cargo Loading and Cell Transfection, 2026. 10.64898/2026.01.16.699683. [DOI]
- Terada T., Kulkarni J. A., Huynh A., Chen S., van der Meel R., Tam Y. Y. C., Cullis P. R.. Characterization of Lipid Nanoparticles Containing Ionizable Cationic Lipids Using Design-of-Experiments Approach. Langmuir. 2021;37(3):1120–1128. doi: 10.1021/acs.langmuir.0c03039. [DOI] [PubMed] [Google Scholar]
- Ripoll M., Martin E., Enot M., Robbe O., Rapisarda C., Nicolai M.-C., Deliot A., Tabeling P., Authelin J.-R., Nakach M., Wils P.. Optimal Self-Assembly of Lipid Nanoparticles (LNP) in a Ring Micromixer. Sci. Rep. 2022;12(1):9483. doi: 10.1038/s41598-022-13112-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- O’Brien Laramy M. N., Costa A. P., Cebrero Y. M., Joseph J., Sarode A., Zang N., Kim L. J., Hofmann K., Wang S., Goyon A., Koenig S. G., Hammel M., Hura G. L.. Process Robustness in Lipid Nanoparticle Production: A Comparison of Microfluidic and Turbulent Jet Mixing. Mol. Pharmaceutics. 2023;20(8):4285–4296. doi: 10.1021/acs.molpharmaceut.3c00390. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Hammel M., Fan Y., Sarode A., Byrnes A. E., Zang N., Kou P., Nagapudi K., Leung D., Hoogenraad C. C., Chen T., Yen C.-W., Hura G. L.. Correlating the Structure and Gene Silencing Activity of Oligonucleotide-Loaded Lipid Nanoparticles Using Small-Angle X-ray Scattering. ACS Nano. 2023;17(12):11454–11465. doi: 10.1021/acsnano.3c01186. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Devos C., Udepurkar A., Sagmeister P., Hodlewsky A. S., Chen J., Hatas A., Ostrovsky N., Al-Jazrawe M., Ren J. I., Liu A. Y., Braatz R. D., Myerson A. S.. Manufacturing mRNA-Loaded Lipid Nanoparticles with Precise Size and Morphology Control. ACS Nano. 2025;19(38):33991–34002. doi: 10.1021/acsnano.5c09800. [DOI] [PubMed] [Google Scholar]
- Ramezanpour M., Schmidt M. L., Bodnariuc I., Kulkarni J. A., Leung S. S. W., Cullis P. R., Thewalt J. L., Tieleman D. P.. Ionizable Amino Lipid Interactions with POPC: Implications for Lipid Nanoparticle Function. Nanoscale. 2019;11(30):14141–14146. doi: 10.1039/C9NR02297J. [DOI] [PubMed] [Google Scholar]
- Hamilton N. B., Arns S., Shelley M., Bechis I., Shelley J. C.. Calculating Apparent pKa Values of Ionizable Lipids in Lipid Nanoparticles. Mol. Pharmaceutics. 2025;22(1):588–593. doi: 10.1021/acs.molpharmaceut.4c00426. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Dehghani-Ghahnaviyeh S., Smith M., Xia Y., Dousis A., Grossfield A., Sur S.. Ionizable Amino Lipids Distribution and Effects on DSPC/Cholesterol Membranes: Implications for Lipid Nanoparticle Structure. J. Phys. Chem. B. 2023;127(31):6928–6939. doi: 10.1021/acs.jpcb.3c01296. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Zhigaltsev I. V., Cullis P. R.. Morphological Behavior of Liposomes and Lipid Nanoparticles. Langmuir. 2023;39(9):3185–3193. doi: 10.1021/acs.langmuir.2c02794. [DOI] [PubMed] [Google Scholar]
- Sorkin R., Kampf N., Zhu L., Klein J.. Hydration Lubrication and Shear-Induced Self-Healing of Lipid Bilayer Boundary Lubricants in Phosphatidylcholine Dispersions. Soft Matter. 2016;12(10):2773–2784. doi: 10.1039/C5SM02475G. [DOI] [PubMed] [Google Scholar]
- Kang D. D., Marks A., Morla-Folch J., Dong Y., Brown B. D., Teunissen A. J. P.. Targeting and Tracking mRNA Lipid Nanoparticles at the Particle, Transcript and Protein Level. Nat. Biomed. Eng. 2025;9(10):1591–1609. doi: 10.1038/s41551-025-01511-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Greco A., Corbo C.. Therapeutic Implications of Biomolecular Corona on Lipid Nanoparticle-Based Gene Delivery Systems. Small Sci. 2025;5(9):2500206. doi: 10.1002/smsc.202500206. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Chatterjee S., Kon E., Sharma P., Peer D.. Endosomal Escape: A Bottleneck for LNP-Mediated Therapeutics. Proc. Natl. Acad. Sci. U.S.A. 2024;121(11):e2307800120. doi: 10.1073/pnas.2307800120. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Chahal G. S.. et al. The Biology of Endosomal Escape: Strategies for Enhanced Delivery of Therapeutics. ACS Nano. 2026;20:1789–1813. doi: 10.1021/acsnano.5c18112. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Wassenaar T. A., Pluhackova K., Böckmann R. A., Marrink S. J., Tieleman D. P.. Going Backward: A Flexible Geometric Approach to Reverse Transformation from Coarse Grained to Atomistic Models. J. Chem. Theory Comput. 2014;10(2):676–690. doi: 10.1021/ct400617g. [DOI] [PubMed] [Google Scholar]
- Cárdenas M., Campbell R. A., Yanez Arteta M., Lawrence M. J., Sebastiani F.. Review of Structural Design Guiding the Development of Lipid Nanoparticles for Nucleic Acid Delivery. Curr. Opin. Colloid Interface Sci. 2023;66:101705. doi: 10.1016/j.cocis.2023.101705. [DOI] [Google Scholar]
- Swetha K., Kotla N. G., Tunki L., Jayaraj A., Bhargava S. K., Hu H., Bonam S. R., Kurapati R.. Recent Advances in the Lipid Nanoparticle-Mediated Delivery of mRNA Vaccines. Vaccines. 2023;11(3):658. doi: 10.3390/vaccines11030658. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Moschioni M., Siraji R. A., Dissard R., Segafredo G., Mutungi H., Jain A., Thakur R., Maurya N., James I.. mRNA Vaccines and Therapeutics beyond COVID-19: A Review of the Global Clinical Development Landscape, Low- and Middle-Income Countries Involvement and Relevance to Their Contexts. Hum. Vaccines Immunother. 2026;22(1):2628424. doi: 10.1080/21645515.2026.2628424. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Joshi R., Sharma S., Kapoor D. U., Prajapati B. G., Huanbutta K., Sriamornsak P.. mRNA-Based Therapeutics: Advances in Drug Delivery, Comparative Innovations, and Biomedical Applications. Mol. Pharmaceutics. 2026;23(2):583–621. doi: 10.1021/acs.molpharmaceut.5c00774. [DOI] [PubMed] [Google Scholar]
- Farsani A. M., Mokhtari N., Nooraei S., Bahrulolum H., Akbari A., Farsani Z. M., Khatami S., Ebadi M. S., Ahmadian G.. Lipid Nanoparticles: The Game-Changer in CRISPR-Cas9 Genome Editing. Heliyon. 2024;10(2):e24606. doi: 10.1016/j.heliyon.2024.e24606. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Chan C., Du S., Dong Y., Cheng X.. Computational and Experimental Approaches to Investigate Lipid Nanoparticles as Drug and Gene Delivery Systems. Curr. Top Med. Chem. 2021;21(2):92–114. doi: 10.2174/1568026620666201126162945. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Tesei G., Hsiao Y.-W., Dabkowska A., Grönberg G., Yanez Arteta M., Ulkoski D., Bray D. J., Trulsson M., Ulander J., Lund M., Lindfors L.. Lipid Shape and Packing Are Key for Optimal Design of pH-Sensitive mRNA Lipid Nanoparticles. Proc. Natl. Acad. Sci. U.S.A. 2024;121(2):e2311700120. doi: 10.1073/pnas.2311700120. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Paloncýová M., Valério M., Dos Santos R. N., Kührová P., Šrejber M., Čechová P., Dobchev D. A., Balsubramani A., Banáš P., Agarwal V., Souza P. C. T., Otyepka M.. Computational Methods for Modeling Lipid-Mediated Active Pharmaceutical Ingredient Delivery. Mol. Pharmaceutics. 2025;22(3):1110–1141. doi: 10.1021/acs.molpharmaceut.4c00744. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
Input data files and analysis scripts are available on Zenodo: 10.5281/zenodo.20454629 and DOI: 10.5281/zenodo.18636861. A complete list of these files is given in the Supporting Information.
