How mRNA Lipid Nanoparticles Interact with Cell-Like Membranes Depending on pH, Protein Coating, and Lipoprotein Levels

We developed an assay designed for temporal monitoring of single LNP binding events to a SLB designed to mimic the endosomal membrane. In this strategy, the ncSLB was formulated with POPC as the major lipid component and 6 mol % negatively charged POPS lipids to represent the anionic character of early to late endosomal membrane conditions.^34,46 LNP binding was continuously monitored under gentle flow conditions at a constant LNP concentration using time-lapse TIRF microscopy of Cy5 labeled mRNA (the LNP cargo) as the pH was reduced in steps from 7.5 to 4.6, followed by a final rinsing step at pH 7.5, as schematically illustrated in FigureA. To characterize the ncSLB, a small fraction (0.5 mol %) of fluorescently labeled PE lipids was incorporated, and fluorescent recovery after photobleaching (FRAP) was performed as previously described.⁴⁷ The analysis revealed a lipid diffusivity of 3.43 ± 0.5 μm²/s and an immobile fraction of 0.01 ± 0.02 (n = 3), with slightly increasing diffusivity and immobile fraction upon stepwise reduction in the pH (Supporting Information↗, Figure S1), which is consistent with formation of a continuous SLB across the entire pH range (FigureB).

To facilitate time-resolved visualization of the binding of individual LNPs (diameter ∼80 nm), 20% of the total mRNA cargo (∼16 mRNAs per LNP) was fluorescently labeled with Cy5, and the LNP concentration was adjusted to ∼0.7 × 10⁹ particles/mL. The LNP suspension was initially injected into a rectangular microfluidic channel (height 400 μm, width 3.8 mm) at pH 7.5 for 20 min at a flow rate of 150 μL/min using a syringe pump, chosen to provide laminar flow but still sufficiently fast liquid exchange over the sensing area (on the order of seconds) without inducing appreciable shear force (<1 fN) on the adsorbed LNPs (see Supporting Information↗, section 10 for estimate). This step was followed by consecutive LNP injections at continuous flow (150 μL/min) for 20 min at each consecutively decreased pH: 6.6, 6.0, 5.6, 5.0, and 4.6. Note that these LNPs had not been exposed to any serum proteins and that the PEGylated state is thus not representative of what would occur in a biological context, in which case the LNPs are coated with a protein corona, as analyzed further below.

While there were virtually no LNP binding events at pH 7.5, a gradual increase in the rate of binding was observed as an increase in the number of Cy5-fluorescent objects at the SLB surface following acidification, which approached saturation at a coverage of ∼0.05 particles/μm² at pH 5.0 (FigureA). In previous work an apparent isoelectric point of around 6 was estimated from zeta potential measurements for similar LNPs containing the same ionizable lipids, albeit at lower ionic strength (25 mM compared with 150 mM used in this study).⁴⁵ While this coincides with the observed onset of LNP binding, both the zeta potential and the nature of electrostatic interactions depend on ionic strength. We therefore measured the relative surface charge of our LNPs at a representative ionic strength using the anionic fluorescent dye 2-(p-toluidino)-6-napthalene sulfonic acid (TNS), which undergoes a significant fluorescent enhancement when binding to positively charged lipids but that is nonfluorescent in solution.⁴⁸ This approach has been previously used to characterize LNPs of the type used in this work.⁴⁹ The TNS assay shows a relatively sharp transition around an inflection point at pH 6.35, with 20 and 80% ionization at around pH 7 and 5.5, respectively (Figure B). In the LNP binding experiments, the electrostatic attraction continues to increase with decreasing pH down to pH 5.0, with an inflection point of the binding curve around pH 5.6 (Figure). This suggests that a significant amount of MC3 must be ionized before the electrostatic attraction becomes sufficiently large to overcome the steric repulsion expected between the shell of PEGylated lipids on the surface of the LNP and the ncSLB. Of note, LNPs preincubated in pH 4.6, which may induce structural alterations,^50,51 displayed insignificant binding when exposed to the ncSLB at pH 7, demonstrating that the ionization is fully reversible and that the ncSLB is free from defects that induce nonspecific LNP binding (Supporting Movie S1↗).

It is also notable that there is a decrease in the rate of binding at pH 5.6 and 4.6. A plausible explanation of the reduction in the rate of LNP binding at pH 5.6 and 4.6 is that negatively charged lipids in the SLB become accumulated in the contact zones between the LNP and the SLB, while simultaneously, LNPs disintegrate and the corresponding positively charged MC3 (the MC3 structure together with the structure of the other lipid components in the LNP is shown in Figure S2↗) diffuse out from the contact zones, which would, in turn, balance negative charges in the ncSLB and suppress further attachment of LNPs.

To scrutinize these scenarios, it is worth noting that the saturated LNP coverage of ∼0.05 particles/μm² is several orders of magnitude lower than the coverage of ∼100 particles/μm² that corresponds to the jamming limit (∼54% surface coverage) of LNPs with a radius of 40 nm. To examine the magnitude of accumulation of negatively charged lipids in the LNP-SLB contact regions, we recall that with a fraction of POPS lipids η = 0.06 and an area per lipid s = 0.6 nm², the average concentration of charged lipids in the SLB is c = η/s = 10⁵ μm^–2. At an LNP coverage of ∼0.05 particles/μm², this relatively high concentration of charged lipids in the SLB cannot be appreciably reduced if the LNPs remain nearly intact. If the LNPs instead undergo a significant collapse, but remain compact and confined near the attachment spot, the scale of the area of the LNP-SLB contact can be roughly estimated as S ≈ 4R² = 6.4 × 10⁴ nm², where R = 40 nm is the average LNP radius in solution. Lipids can to a first approximation be assumed to form a triangular lattice both in the SLB and the LNP. Due to steric repulsion, charged lipids are not expected to be located in the nearest-neighbor sites of such a lattice. This means that the minimal area per charged lipid should not be smaller than 3s = 1.8 nm². Thus, the maximum number of charged lipids in the contact area can be estimated as n_∗ = S/3s = 1.2 × 10³. Further, the average concentration of charged lipids (per μm² of the SLB) associated with attached LNPs is consequently given by c_∗ = n_∗C = 60 μm^–2, where C = 0.05 μm^–2 is the LNP coverage at which the rate of LNP binding is suppressed. Under these conditions c_∗ is around 3 orders of magnitude lower than c (the average concentration of charged lipids in the SLB) even for compactly collapsed LNPs. Hence, for the reduced rate of binding to be solely due to accumulation of negatively charged lipids near the site of LNP attachment, the relaxation of the LNP shape should be dramatic and engage the majority of charged MC3 lipids in an LNP (∼5.5 × 10⁴ MC3 lipids per LNP; see Supporting Information↗, section 2 for estimation). Of note, in experiments with more than 1 order of magnitude higher LNP concentration (∼25 × 10⁹ particles/mL) and a lower fraction of charged lipids η = 0.02 the LNP binding saturated at C = 0.3 μm^–2 at pH 4.6 (Supporting Information↗, Figure S3) which is also far below the jamming limit of 100 μm^–2 for LNPs with a radius of 40 nm. Under the assumption that the LNP collapse is dramatic and that a majority of MC3 (n_∗ ∼ 5.5 × 10⁴) is accumulated in the contact zone, we have c_∗ = n_∗C = 2 × 10⁴ μm^–2. This number is close to the average concentration c of charged lipids in the SLB at η= 0.02 (c = η/s = 2.4 × 10⁴ μm^–2), which suggests that the pH-induced LNP attachment may indeed induce a dramatic collapse of the LNPs.

To scrutinize the diffusion of lipids to and out from the contact zone, it is instructive to first estimate the time scale, τ, corresponding to association of n negatively charged lipids to the LNP-SLB contact area. Roughly, the diffusion-limited flux J of these lipids is given by J ≈ Dc, and accordingly, τ = n/J ≈ n/Dc, where D = 3 μm²/s is the lipid diffusion coefficient. Setting n to 5.5 × 10⁴, i.e., engagement of the majority of the MC3 lipids, the time scale corresponding to lipid diffusion limitations τ becomes ∼0.2 s at c = 10⁵ μm^–2. This is appreciably shorter than the time scale during which the rate of LNP binding diminishes at pH 5 and pH 4.6 (Figure), suggesting that if this hypothesis holds, the process is controlled by the time scale of LNP relaxation. This analysis also suggests that an additional contributing factor to the reduction in the rate of LNP binding may be attributed to escape of positively charged MC3 from collapsed LNPs into the SLB, being consistent with lipid transfer observed upon pH induced binding of MC3 containing LNPs to anionic lipid monolayers formed at the air water interface.²⁰ Transfer of cationic lipids into the anionic lipid bilayer would lead to a reduction of net negative charge of the SLB, a process that was previously observed to influence electrostatically controlled liposome binding to oppositely charged SLBs.^52,53

These findings are potentially relevant in the context of endosomal maturation arrest recently observed for endosomes containing multiple LNPs,⁵⁴ since they suggest that both the charge balance between the total number of ionized MC3 and the limited number of negatively charged lipids available in the endosomal membrane as well as the time scale of LNP collapse may be crucial parameters for successful endosomal escape. Further, it is also worth noting that upon stepwise reversal of pH up to 7.5, ∼50% of the Cy5-labeled mRNA remained visible on the ncSLB, with most of the release events occurring when the pH was switched from pH 5.6 to 6.6 (Supporting Information↗, Figure S4). This observation further supports that a significant proportion of the LNPs undergo structural changes that are not reversed upon charge neutralization of MC3. It also shows that mRNA can be released upon increasing the pH above the isoelectric point of the LNPs, which is representative of mRNA coming in contact with the cellular cytosol after endosomal rupture.

The above-described function of pristine LNPs is of interest from a biophysics perspective but not fully representative in a biological context. In particular, LNPs are known to acquire protein coronas that are, in many cases, crucial to promote their cellular uptake. We therefore next asked whether such a corona would influence the magnitude and nature of the LNP interaction with the endosomal model membrane. To address this question, the experiments described above were repeated for LNPs that had been preincubated with serum-containing media prior to injection in the flow cell. Inspired by recent reports, demonstrating the important role of lipoprotein binding to LNPs for successful uptake and endosomal cargo escape,^12,13,55,56 we specifically compared the pH-induced LNP binding to the endosomal membrane mimic for LNPs preincubated with diluted untreated fetal bovine serum (FBS 10%) with the corresponding binding patterns of LNPs preincubated with 10% lipoprotein saturated (Lipo-S) or lipoprotein-depleted (Lipo-D) fetal bovine serum (Figure). The relative presence of lipoproteins in the 10% FBS, Lipo-D and Lipo-S samples, obtained by LC/MS-MS, validated the depletion and enrichment of lipoproteins in the respective fractions (Supporting Information↗, Figure S5). Compared with pristine LNPs (Figure, red symbols), preincubation in 10% untreated FBS for 10 min prior to injection did not induce any detectable binding to the ncSLB above pH 6.6 (Figure, green). Instead, there was a significant shift in the onset of LNP binding to lower pH, and a more than 50% reduction in the rate of LNP binding at all pH conditions assayed. These results suggest that the FBS-induced protein corona formation on the surface of the LNPs, addressed in detail elsewhere,¹⁴ reduces the MC3-mediated pH-induced electrostatic attraction to the negatively charged SLB. However, there was no reduction in the rate of LNP binding even at the lowest pH (as observed for pristine LNPs, Figure), which suggests that protein corona formation at least in part prevents appreciable LNP collapse and lipid accumulation in the contact zone and/or exchange with the endosomal membrane mimic. We also found that the lipoprotein content of the FBS, and hence the formed protein corona, had significant effects on the interaction of the LNPs with the endosomal membrane mimic. Upon removal of lipoproteins from the preincubation serum by KBr-induced flotation (Lipo-D), the pH-induced LNP binding was still hampered, but eventually reached the same coverage as for pristine LNPs at pH 4.6. In contrast, for LNPs preincubated in lipoprotein supplemented serum (Lipo-S), the LNP binding was further reduced compared with the FBS 10%-treated sample at all pH. This suggests that lipoprotein association to LNPs, while reportedly often beneficial for cell uptake, may influence the crucial endosomal escape event.

Aided by the single LNP resolution, the time lapse movies recorded during the TIRF experiments provided another curious observation: the bound LNPs displayed slow intensity variations (on the time scale of ∼200 s) which, for a significant fraction, was combined with one or sometimes two appreciable stepwise increases (on the time scale of ∼5 to 10 s) of the intensity at the locations of LNP binding (FigureA; Figure S6↗). Between ∼20 and 30% of the LNP binding events were accompanied by sudden intensity increases at pH 5.6 and below for pristine LNPs and at pH 5.0 and below for LNPs preincubation in serum (FigureB). Further, the intensity distribution at the end of each incubation ranged over more than 2 orders of magnitude, which is significantly broader than expected for LNPs with diameters ranging from 40 to 120 nm (Supporting Information↗, Figure S2). It is worth noting that there is a tendency that sudden intensity increase events are more likely to occur for LNPs with higher intensity. These observations suggest that an appreciable number of adsorbed LNPs undergo a significant collapse, since such events would transfer the fluorescent labels closer to the surface where the intensity of the evanescent field utilized in TIRF imaging is highest. This interpretation is further supported by a complementary experiment in which biotin-modified LNPs containing a small fraction of fluorescently labeled lipids were immobilized at neutral pH to a biotin-modified ncSLB using streptavidin as a linker. Upon reduction of pH, clear signatures of lipid transfer to the ncSLB, which is consistent with LNP collapse, was observed for a fraction of the LNPs (Supporting Information↗, Figure S7).

As detailed in section 8 of the Supporting Information↗, a complete LNP collapse would, for LNPs with diameters ranging from 40 to 120 nm, result in an intensity increase between 1.2 and 1.8, i.e., as observed, the effect is predicted to increase with increasing LNP size. This is in reasonable agreement with most events observed, which ranged between 1.1 and 2.5 (FigureA, left and central panels, and Figure S6A–G↗). Still some occasional events with significantly more pronounced intensity increases were also observed (FigureA, right panel and Figures S6H–J↗). This suggests that the interaction between the LNPs and the ncSLB may not only induce rapid collapse events, but also that later arriving LNPs can, in some instances, colocalize and aggregate with already bound LNPs. In principle, LNP aggregation might also occur in the bulk of the solution but this is unlikely because the interaction between LNPs is weak even in the absence of electrostatic repulsion (see section 9 of the Supporting Information↗ for an estimate of the interaction strength between suspended LNPs). Thus, LNP-LNP aggregation is not expected to occur unless LNPs bound to the ncSLB would expose hydrophobic defects or negatively charged mRNA, which adds further support to a structural change of the LNP being induced when bound to the ncSLB. This interpretation is further supported by the fact that LNP collapse and/or aggregation was not observed at pH 6.6, displayed a peak at pH 5.6 for pristine LNPs and did not occur until pH 5.0 for LNPs preincubated in normal and lipoprotein-depleted serum (FigureB), where the protein corona may serve to reduce electrostatic attraction and possibly also to stabilize local defects on the surface of the LNPs. Further, mRNA was not observed to diffuse into the ncSLB, which is attributed to electrostatic attraction between the positively charged headgroup of MC3 and the negatively charged mRNA, which is expected to ensure firm association to the lipid membrane or the underlaying silica support.

These observations are curious in the context of the reported significance of nanoparticle clustering for modulating cellular uptake and endosomal escape.^57,58 Further, even though lipoprotein depletion was not observed to influence the tendency of LNPs to undergo collapse and/or aggregation, lipoproteins seem to hamper the electrostatic attraction of LNPs to the endosomal membrane, which is, in turn, expected to drive the critical endosomal escape event. Since protein corona formation precedes and has even been shown to be required for efficient cellular uptake,^{11 −14} these findings call for complementary live cell studies to gain deeper insight into the influence of protein corona formation in general, and lipoproteins in particular, not only on uptake efficiency, but also with respect to maintained coronation in the endosomal lumen and the impact of coronal proteins on endosomal escape.

To investigate how the presence of lipoproteins in the preincubation serum influences the functional cellular delivery of the LNPs, we exposed cultures of hepatic Huh7 cells to MC3 LNPs containing mRNA encoding eGFP and monitored cell uptake and protein translation. Within a minute before cell exposure, the LNPs were diluted in cell culture media with 10% untreated FBS or 10% Lipo-D to represent lipoprotein-rich and -depleted protein coronas corresponding to the ones in the biophysical studies (Figures–4). The LNP uptake was visualized by Cy5 fluorescence from the labeled mRNA and functional delivery by the fluorescence of the expressed eGFP protein expression using confocal microscopy (FigureA) and quantified, after harvesting of the cells, with flow cytometry (FigureB–D). The results show that the uptake of the LNPs pretreated with FBS 10% and Lipo-D 10% (FigureA,B) is largely comparable (∼7% higher with Lipo-D). This indicates that a general depletion of lipoproteins in the serum does not prevent cellular uptake and is consistent with that the depletion method used only resulted in minor removal of ApoE (Supporting Information↗, Figure S5B), known to be crucial for cellular uptake.¹⁰ However, the cells that were treated with Lipo-D pretreated LNPs (and continuously kept in Lipo-D culture media throughout the experiments) displayed a significantly higher level of eGFP fluorescence (FigureC). Although protein expression is an aggregate readout that can be influenced by multiple processes, these results suggest improved delivery for LNPs preincubated in serum with lower lipoprotein content. Under the assumption that the endocytic mechanisms that control LNP uptake are not directly coupled to the mechanisms that dictate endosomal escape (pH-buffering effect, flip-flop mechanism, fusion or destabilization mechanism etc.),⁵⁹ and the assumption that protein expression is not affected by the lipoprotein composition of the cell media, an indicative measure of relative endosomal escape efficiency can be obtained by normalizing the eGFP signal to the Cy5 signal (FigureD). The higher eGFP/Cy5 ratio for cells incubated in Lipo-D FBS compared to FBS suggests that endosomal escape could be promoted under lipoprotein-depleted conditions. This result agrees with the biophysical data (Figure), where LNPs preincubated with Lipo-D solution presented a more favorable interaction with the ncSLB.

Time-resolved TIRF microscopy with single LNP resolution was used to provide biophysical insights with respect to how gradual acidification influences the interaction between ionizable lipid-containing LNPs and a supported lipid bilayer mimic of the endosomal membrane. Although the onset of LNP binding occurs at high pH (∼6.6), efficient binding does not occur until the pH is reduced to between 6.0 and 5.6, and appears to eventually be severely limited by suppression of the driving force for multivalent bond formation during LNP attachment or, more specifically, by accumulation of anionic lipids in the contact zone between the LNP and the membrane mimic accompanied by charge neutralization of anionic lipids in the model membrane due to their association with cationic lipids escaping from the earlier-attached LNPs upon their disintegration. This finding suggests that the nature of the interaction between LNPs and endosomal membranes may change significantly as LNPs mature along the endosomal pathway.

Aided by the single LNP resolution provided by TIRF imaging, we also conclude that a significant fraction of the LNPs undergo substantial stepwise collapse on the model membrane, a process that is likely a prerequisite for efficient cargo delivery across the endosomal membrane. The LNP collapse typically occurred within 5 to 15 min after initial LNP binding, being consistent with the previously observed lag phase associated with LNP-binding induced reorganization of anionic lipid monolayers formed at an air water interface.²⁰ From a physical perspective, this is consistent with the relatively large length scale of the structural heterogeneity of the LNP, with a core composed primarily of mRNA and ionizable lipids surrounded by a shell composed of helper lipids, both of which only about three to four times smaller than the LNP size.^42,60 Although a detailed mechanistic understanding of LNP collapse is lacking, our observations suggest that this process could potentially contribute to the endosomal escape and should therefore be considered when optimizing the design of LNPs containing pH-sensitive ionizable lipids. These insights may also be important to consider when evaluating endosomal escape capacity since single endosomes are likely to contain multiple LNPs and since the total number of negatively charged lipids available in a maturing endosome may be a limiting factor for attachment, fusion, and escape. A relevant extension of the work is therefore to investigate LNPs made using different ionizable lipids as well as different lipid compositions and cargo content.

We also observed that preincubation of the LNPs in serum solutions shifts the onset of LNP binding toward lower pH, suggesting that proteins adsorbed on the surface of the LNP may hamper the interaction between LNPs and the endosomal membrane. This effect might actually influence the endosomal escape efficiency, making this observation interesting in the context of previous studies demonstrating particularly productive endosomal escape early in the endolysosomal pathway.^61,62 Further, preincubation of LNPs in lipoprotein-depleted (Lipo-D) serum caused a weaker inhibitory effect on the electrostatic attraction between the LNPs and the supported endosomal model membrane. We also observed more protein (eGFP) translation in live cells treated with Lipo-D preincubated LNPs compared to the FBS 10% case. Although not possible to draw firm conclusions regarding endosomal escape from differences in the ratio between LNP uptake and protein production, these results suggest that although binding of lipoproteins to the surface of LNPs has been identified as crucial for efficient cellular uptake,^12,13 it may be equally important to consider how this class of proteins may impact the efficiency of the actual mRNA endosomal escape event as well as other mechanisms that control protein production.

Although further studies are required to verify the biological significance of the biophysical insights gained using a reductionist mimic of the endosomal membrane as done in this work, the approach is readily transferable to a wide range of drug-delivery vehicles and more realistic model membrane systems,^63,64 from which insights could be gained that should aid the design of next generation lipid-based nanoparticles and more efficient oligonucleotide delivery.

Authors would like to acknowledge the Swedish Foundation of Strategic Research for financing the project and all the members of Industrial Research Centre “FoRmulaEx” (IRC15-0065) for their support. Everyone in the group is grateful especially to AstraZeneca, Sweden, for their great help in terms of scientific guidance and providing LNPs.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsnano.2c04829↗.

^⊥ Division of Solid-State Physics and NanoLund, Physics Department, Lund University, P.O. Box 118, Lund 22100, Sweden

L.L., E.K.E., and F.H. conceived the project. N.A. did the microfluidic experiments. M.M. and N.A. performed the corresponding data analysis. A.G., E.W., and S.H. did the cell experiments and corresponding analysis. Y.J. produced and characterized the LNPs together with L.L. A.G., K.L., and A.S. performed the proteomics analysis. G.E. conducted the TNS measurements and analysis. V.P.Z. developed the theoretical basis for the estimates which was implemented by N.A., M.M., and F.H. All authors contributed to the writing of the manuscript.

The authors declare no competing financial interest.