An obesity model and corona multiomics analysis reveal high-density lipoprotein effects on lipid nanoparticle function


 Lipid nanoparticles (LNPs) are currently of great interest for therapeutic nucleic acid delivery. Lipid-based nanoparticles are, however, difficult to study analytically and so our understanding of the interaction between LNPs and biological systems remains obscure, particularly in terms of biomolecular corona formation and the impact this has on therapeutic efficacy and targeting. Ideally, we would like to engineer particles to acquire corona components that facilitate targeting, cargo delivery and improved safety. This requires studying the relationship between LNPs, biofluids and the resulting coronas. The particle-corona complexes are, however, fragile and biofluids also contain particles (such as lipoproteins) with sizes and biochemical characteristics similar to lipid nanoparticles, so contamination with biological components is a challenge. Here we describe a rapid, automated, and unbiased isolation method for LNP biomolecular coronas, coupled with proteomic and lipidomic analysis. Using these analytical methods, we systematically studied, in lean and obese rats, the variation in LNP-mediated mRNA delivery caused by individual physiology. A comprehensive multivariate model links LNP corona content to efficacy, identifying and validating high-density lipoproteins as a previously unidentified factor affecting particle efficacy.

Comparing LP and OP samples, we observed the most noticeable difference in eGFP expression 131 when recipient cells were supplemented with 1% lean or obese plasma prior to LNP dosing. As 132 shown in Fig 1b, the overall eGFP expression in the 1% lean condition was reduced, compared to 133 higher plasma concentrations, although all hepatocyte cell lines expressed an increasing amount 134 of eGFP with higher LNP doses and longer times and, the differences between lean and obese 135 plasma became more apparent. The most significant difference between lean and obese conditions 136 appeared at the combination of 1% plasma and the 200 ng mRNA/well dose (Fig. 1c), where LNP 137 efficacy was 5-10 fold improved by obese plasma supplementation. At lower LNP doses, such as 138 the 50 ng mRNA/well dose, the contrast between lean and obese plasma effects on LNP efficacy 139 was less apparent (Fig. 1d). The significant difference at 200 ng mRNA/well emerged as early as 140 4 h post-dosing and persisted until the endpoint at 10 h ( Supplementary Fig. 6). The expression 141 difference between obese and lean was less pronounced in kidney fibroblasts than in hepatocytes 142 (Fig 1c). The relationship between the LNP and plasma dosing is crucial because it reveals that 143 higher protein production is not simply because of the obese plasma's general anabolic effect (this 144 is explored in more detail below). 145 To further explore how individual physiological variations can affect LNP efficacy, we then dosed 146 cells with 50 ng and 200 ng eGFP mRNA containing LNPs, supplemented with individual plasmas 147 at a 1% ratio, as the 1% plasma supplementation seemed to drive higher efficacy. The difference 148 between lean and obese conditions remained when cell cultures were supplemented with individual 149 lean or obese plasmas. In contrast, individual variation now emerged ( Fig. 1e and Supplementary 150 Fig. 7), although the intragroup variation (among lean or obese individuals) was less significant 151 than the intergroup variation (lean v.s. obese).
Overall, the intracellular mRNA expression difference between lean and obese conditions is likely 153 not due to the absolute plasma component abundance (for example, ApoE) or plasma's impact on 154 cellular metabolism alone. For example, the higher ApoE content (i.e., obese v.s. lean, or 10% v.s 155 5% plasma) did not introduce significantly higher eGFP expression levels in recipient hepatocytes 156 at the tested dose range. In contrast, the 1% obese supplementation resulted in the highest eGFP 157 expression. This suggests that ApoE is not the sole determinant of LNP hepatic potency. The 158 imaging results (Fig. 1b) also implied that the difference in eGFP expression did not directly 159 correlate to the overall amount of ApoE or other nutrients provided by plasma. 160 Based on these findings, we hypothesized that the altered mRNA cargo expression is, at least 161 partially, a result of different biomolecular corona formations. The binding of the corona 162 components to LNPs can cause structural and compositional rearrangement 24 , and this can affect 163 potency. Thus, in the 1% lean condition, the attenuation in the eGFP expression was likely due to 164 the limited availability of biomolecules necessary to form the biomolecular corona, especially at 165 doses higher than 50 ng mRNA/well. The obese plasma condition preserved the eGFP expression 166 and enhanced the expression compared to high plasma ratios because, in this case, a corona formed 167 that was more compatible with hepatocytes. Alternatively, lowering the plasma ratio may also 168 unmask corona interactions with recipient cells, as previously reported 36 . Therefore, isolating LNP 169 coronas and identifying the corona determinant driving altered eGFP mRNA expression is 170 necessary to understand how to modulate corona composition and adapt to individual physiology 171 and pathology variations. 173 Given the great variation in abundance and complexity of plasma biomolecules, the LNP-corona 174 complex (LNPcor) must be separated from unbound biomolecules and endogenous nanosized 175 structures in the plasma, as illustrated in Fig 2a. Preservation of corona function and integrity is 176 crucial for studying the corona and revealing the identity of its components. So aggressive 177 chemical and mechanical separation procedures are not suitable for LNPcor isolation. We assessed 178 various methods to retrieve the LNP-corona complexes from plasma prior to corona analysis. 179

Ultrafast affinity isolation enables unbiased biomolecular corona phenotyping
Retrieving these complexes is technically challenging as plasma components overlap in size, 180 composition, and physicochemical characteristics. The commonly used size exclusion 181 chromatography (SEC) is prone to co-isolate unbound protein and endogenous particles together 182 with nanoparticles 37 . Latest techniques, such as asymmetric flow field-flow fractionation (AF4), 183 is an improved size-based separation technique that utilizes a liquid cross-flow to generate a 184 parabolic flow profile in a narrow channel 38 . It has previously been utilized in the characterization 185 of lipid-based nanoparticles 39  LNPcor from antibodies. Nevertheless, a pH lower than the pKa of ionizable cationic included in 208 the LNPs (i.e., DLin-MC3-DMA) usually results in LNPs disassembly 40 . The DoE optimization 209 revealed that a combination of antibody against PEG backbone, PBS wash and basic pH elution 210 resulted in robust LNPcor recovery in the final elution. 211 We then tested this miniaturized, high-throughput setting to identify optimal incubation and elution 212 conditions to retrieve maximal LNPs without introducing impurities by prolonged incubation and 213 elution period ( Supplementary Fig. 10). Then, the optimized approach is used to retrieve LNPcor 214 incubated in 1% lean pool and obese pool plasmas. As shown in Fig. 2c, this method successfully 215 between LNPcor formed in the two plasma conditions. The number and recovery rate of nanosized 217 particles within the elution was examined by nanoparticle tracking analysis (NTA) to confirm that 218 LNPcor was harvested in a particle-like format. The recovery rate of lipid and mRNA content was 219 calculated by Rhod and Cy5 fluorescent intensity associated with LNPs. Overall, the recovery rate Cryogenic transmission electron microscopy (Cryo-EM) was also employed to reveal the 239 macroscopic effects that the corona exerts on LNPs ( Supplementary Fig. 12). Cryo-EM showed 240 that the lean LNPcor remained mostly as electron-dense spheres as described earlier 3 . Intriguingly, 241 the obese LNPcor demonstrated multi-laminar structures, likely resulting from the incorporation 242 of biomolecules with detergent-like properties, such as apolipoproteins 41 . 243 Overall, the result confirmed that the ultrafast affinity isolation can harvest intact LNPcor complex 244 with functionality preserved, minimizing impurities that biased conventional isolation methods. 245 The individual coronas were also evaluated using quantitative proteomics, revealing significant 254 differences between lean and obese coronas. On average, obese-plasma-derived biomolecular 255 coronas harbored 1.8-fold more protein than those derived from lean plasma (Fig. 3a). Control 256 experiments were conducted without LNPs to test for unspecific protein and lipid adsorption onto 257 the magnetic microbeads and plastic surfaces. The amount of unspecifically bound proteins (Fig.  258 3a, BG) and endogenous plasma particles (Supplementary videos) were substantially lower compared to the LNPcor samples. The relative protein abundance (RPA) of identified proteins was 260 also calculated by comparing to the total number of detected protein sequences. The group average 261 RPA of corona apolipoproteins increased from 76.04% to 94.40%, when moving from lean to 262 obese plasmas (Fig. 3b). This increase of lipid-binding apolipoproteins posed a potential 263 explanation for the Cryo-EM morphology change of obese derived LNPcor due to their 264 amphipathic properties ( Supplementary Fig. 12). Other consistently detected corona proteins were 265 involved in the acute-phase response, redox homeostasis, iron transfer, complement and 266 coagulation processes, and immune response-related proteins. 267

Individual lean or obese conditions altered LNP biomolecular corona
In general, the LNP protein corona did not mimic the plasma protein composition, implying 268 selective absorption of corona proteins. Fig. 3c illustrates the breakdown of major corona proteins 269 (relative protein abundance, RPA>0.1%, summarized from individual coronas) and their 270 abundance in original blood plasma. We found that in obese-plasma-derived coronas, 271 apolipoprotein A-II (Apo A-II), apolipoprotein C-II (Apo C-II), apolipoprotein C-III (Apo C-III), 272 and apolipoprotein M (Apo M) were elevated. In contrast, the ApoE abundance was moderately 273 reduced in all obese-plasma-derived coronas compared to corona samples derived from lean 274 plasmas, despite the improved cargo eGFP mRNA expression following LNP exposure to obese 275 plasma (Fig. 1b). Antibodies against apolipoproteins were employed to confirm the presence of 276 Apo A-II, C-II, C-III, and E in lean and obese-plasma-derived coronas ( Supplementary Fig. 15). 277  Multi-omics analysis generated comprehensive data on the coronas formed following exposure to 301 the individual lean and obese plasmas. Corona components function cooperatively, so it is 302 problematic to consider them in isolation. Multivariate regression models have been shown to out-303 perform single-component models when predicting the effects of corona composition on 304 nanoparticle performance 44 . Therefore, we utilized an orthogonal partial least squares (OPLS) 305 analysis to reveal the latent multivariant correlation between the individual plasma-derived 306 biomolecular coronas and eGFP expression. In OPLS plots (Fig. 5a), the distance on the Y-axis 307 indicates orthogonality to the OPLS model. Omics hits that localize closer to 0 on the y-axis have 308 lower orthogonality and are more relevant. While the distance along the x-axis reveals the relative 309 correlation. A hit with a positive x value, for example, is positively correlated to the cellular eGFP 310 expression. Overall, the OPLS correlation model can reveal the potential correlation between the 311 corona content resulting from lean and obese plasma exposure and LNP performance. 312 Hits derived from the -omics analysis (see the proteomic and lipidomic methods for lead 313 identification methods) were normalized using z-score normalization for unbiased correlation 314 (Supplementary data 3). In Fig. 5a, the correlation between corona contents (protein and lipid 315 contents, X variables) and the eGFP expression obtained by quantitative imaging (AUC10h, Y 316 variables) for the OPLS model was illustrated as cumulative R2X, R2Y, and Q2 (R2Xcum, R2Ycum, 317 Q2cum, range 0-1). The correlation model explained 74.1% (R2Xcum=0.741) of the X variable 318 behavior and 93.9% (R2Ycum=0.939) of the Y variable behavior. An R2 value greater than 0.66 is 319 considered indicative of a good model fitness 45 . Furthermore, the overall model predictive power 320 (X variables' impact on Y variables) of the OPLS is 68.6% (Q2cum=68.6). Overall, the OPLS 321 correlation model can reveal the potential correlation, between the corona content resulting from 322 lean and obese plasmas and, LNP performance. 323 The OPLS analysis revealed two groups of corona proteins that were of particular interest 324 concerning LNP efficacy: immune response proteins (exemplified by complement proteins) and 325 proteins related to lipid physiology (apolipoproteins) (Fig. 5a). The immune response proteins are 326 not considered here and are the focus of separate, ongoing experiments. The presence of many 327 apolipoproteins in LNP coronas correlated with LNP efficacy (-0.1<Orthogonality<0.1). In 328 particular, Apo A-II, B, C-II, and C-IV, the structural proteins found in HDL, VLDL, and 329 chylomicron (CM) lipoproteins, were positively correlated with LNPs performance. Apo M, which 330 is found primarily on HDL 46 , also exhibited a high positive correlation to LNP performance. In 331 contrast, Apo E and Apo A-IV 47 , which are primarily found in VLDL and CM demonstrated a 332 moderate negative correlation to LNPs performance (Fig. 5a). While Apo E is often considered 333 highly correlative to LNP hepatic delivery efficacy, the amount of corona Apo E in obese-plasma-334 derived coronas declined modestly compared to lean-derived counterparts ( Supplementary Fig.  335 14). Remarkably, quantitative imaging revealed that LNPs stimulated higher cellular eGFP mRNA 336 expression when supplemented with obese plasmas that resulted in coronas with less ApoE (Fig.  337 1e). As will be shown later, this is also not simply because the obese serum stimulates protein 338 production, but rather that specific plasma components accumulate in LNP coronas and this leads 339 to enhanced LNP function. 340 While not all protein components of LNP coronas are strongly correlated with LNP function, the 341 majority of lipids associated with LNPs did correlate with enhanced LNP performance. A lipid-342 rich corona, in general, is favorable for LNP mediated mRNA delivery. However, the 343 orthogonality of lipid hits was higher than most of the protein hits (Orthogonality>0.1 or <-0.1), 344 suggesting that the corona proteins are more relevant for explaining the differences between LNP 345 function in lean and obese contexts. The plasma lipids became more relevant (orthagonality 346 dropped below 0.1 in most cases) when considering particle uptake as opposed to function. While 347 particle uptake was not our primary readout, it is interesting that PC, for example, has previously 348 been shown to be important for particle uptake in an SR-BI-dependent manner 48 . 349 In addition to the intergroup correlation between X and Y variables, OPLS also revealed the 350 intragroup correlation within X variables and Y variables separately. Within the Y variables group 351 (eGFP expression), the human and rat hepatocytes were located in close proximity to each other, 352 indicating the LNP performance with the three hepatocyte lines were similar when supplemented 353 with either lean or obese plasmas. On the contrary, the non-hepatocytes were located further away 354 from the hepatocytes cluster. This indicates that the LNPs efficacy is likely cell-type and tissue-355 origin specific. Within the X variables, the apolipoprotein clusters (Apo A-II, C-II, C-III, C-IV, 356 and ApoM) are adjacent to the corona lipids, suggesting the apolipoproteins and lipids, which are 357 commonly co-assembled into lipoprotein particles, formed LNP coronas collectively. 358 Additionally, the OPLS revealed acute phase, complement cascade, and coagulation function 359 associated proteins are less likely to be present in an apolipoprotein-abundant LNP corona. Overall, 360 the above intragroup correlations show that OPLS is a valuable method for elucidating the 361 multivariant relationship between corona content and LNP mRNA expression efficacy. 362 Our focus then turned to identify the mechanism that triggered different corona formation and 363 associated efficacy variation in lean and obese conditions. While OPLS analysis revealed complex 364 multivariant correlations, linear classifiers are a more intuitive way to explore the impact of corona 365 components on LNP function. There was a moderate correlation between individual corona ApoE 366 and cellular eGFP expression and, according to our data, both corona and plasma AopE were not the optimal indicators for hepatic mRNA expression following LNP delivery (Fig 5b and 5c). In 368 contrast, the correlation between individual ApoM and cellular eGFP expression was much 369 stronger. The corona ApoM also correlated more with LNP function than plasma ApoM (Fig 5b  370 and 5c). In addition, other major HDL-associated apolipoproteins, such as ApoA-II, were also 371 highly correlated with cellular eGFP expression (Supplementary Fig. 16). Lipoprotein fragments, 372 even intact lipoproteins, have previously been identified in nanoparticle coronas 49,50 , so we 373 hypothesized that LNP corona formation involved interactions with various lipoprotein particles. 374 In terms of composition, lipoproteins particles are, however, complex and dynamic. While our 375 analysis revealed that certain lipoproteins correlated more with increased LNP efficacy, it is 376 challenging to map this proteomic fingerprint onto a naturally occurring lipoprotein particle 377 population, although our data suggested that HDL particles were a likely source of the efficacious 378 corona components. 379 To explore LNP interactions with lipoprotein particles experimentally, Huh7 hepatocytes were 380 exposed to LNPs together with 1% lean plasma spiked with purified HDL, VLDL, and CM to 381 determine how these components affected cellular mRNA expression (recall that LNPs 382 supplemented with 1% lean plasma are not very functional in Fig. 1). The lipoprotein candidates 383 were spiked into lean plasma using a wide concentration range, either separately or combined, at 384 the 200 ng/well eGFP mRNA dose, and the resulting eGFP production was measured (Fig. 5d, and  385 supplementary Fig. 17). 386 We found that, unlike other lipoproteins, only corona HDL could potentiate cellular responses to 387 LNP exposure, and even at the lowest spike-in concentration (5x10 5 particles/well), HDL alone 388 could stimulate the expression of mRNA delivered by LNPs, with a maximum response at 1.6 x10 7 particle/well. At higher HDL concentrations, a decline in eGFP mRNA expression was observed. 390 In comparison, even at high concentrations, the addition of VLDL and CM did not affect LNP 391 performance compared to lean plasma alone (dotted line , Fig 5d). When HDL, VLDL, and CM 392 were spiked in an equal concentration together, the LNP performance enhancement was also not 393 different from HDL alone. In contrast, at higher spike-in doses, the lipoprotein combination 394 mitigated somewhat the eGFP expression reduction seen when using higher doses of HDL. 395 Next, we investigated whether the HDL-potentiated cellular mRNA expression was LNP/HDL 396 ratio-dependent. We dosed the LNPs to recipient cells at various combinations of LNP doses and 397 HDL concentrations. At lower LNP doses, the maximum mRNA expression occurred at lower 398 HDL concentrations, while high LNP doses required more HDL but resulted in higher mRNA 399 expression (Fig. 5e), suggesting a constant optimal ratio between corona HDL and the LNPs in 400 terms of productive cargo delivery. 401 We also verified the HDL-enhanced cellular mRNA expression in human lipoprotein deficient 402 serum ( Supplementary Fig. 18). When human Huh7 hepatocytes were exposed in the human serum 403 with lipoprotein removed, LNPs did not significantly evoke eGFP expression at 25, 50, and 200 404 ng/well mRNA doses. When HDL was added, the same relationship between HDL and LNP 405 numbers was seen, with lower plateaus at lower doses and higher plateaus at higher doses. Again, 406 further increasing the amount of HDL added inhibited cellular eGFP expression. In comparison, 407 VLDL and CM did not stimulate effective cellular eGFP expression. The decline of cellular eGFP 408 expression at high HDL doses is likely a result of increased competition, between corona 409 components and the unbound counterparts in plasma, for cell surface receptors, such as the 410 previously mentioned SR-BI 51 (Supplementary Fig. 19). This is interesting, as it suggests that LNPs interact with HDL at lower doses and it is only at relatively higher doses that cell surface 412 scavenger receptor blockade negatively impacts LNP performance. 413 Finally, thanks to our efficient method for isolating LNPs and their associated coronas, we could 414 allow the coronas to form, retrieve the LNPs and then dose them to Huh7 hepatocytes 415 ( Supplementary Fig. 20). The LNPcor was formed and retrieved from 1% lean pool (LeanPD) and 416 1% obese (ObesePD) as described above. The retrieved LNPcor were then dosed together with 1% 417 plasma from lean or obese animals. In agreement with our previous data, the obese plasma 418 containing higher HDL levels clearly inhibited the uptake of LNPs with pre-formed HDL-enriched 419 coronas. 420

421
Much effort has been devoted to the development of highly effective LNPs through sophisticated 422 chemical composition designs. However, most of these efforts proceed in a "trial and error" 423 manner. How these designs can affect delivery efficacy and tissue tropism is largely unpredictable, 424 particularly due to the tremendous diversity of pathophysiological conditions. This severely 425 hinders the development of personalized precision nanomedicine. Understanding corona formation 426 is, on the other hand, an emerging approach for optimizing nanomedicine design towards desirable 427 biological outcomes. 428 Our study demonstrated that individual physiological states can affect LNP function through 429 corona formation. The heterogeneity amongst patients/diseases can hinder the success of clinical 430 nanomedicines, whereas most nanomedicine studies are carried out in unstratified patient 431 populations 52 and, our data suggest that corona composition is more predictive than plasma biomarkers. For example, although animal 12 was classified as obese using relevant biomarkers, 433 its plasma resulted in an LNP corona composition closer to lean plasma-derived corona and eGFP 434 expression similar to the lean group. Our study is the first to show wew LNP designs, informed by 435 corona composition generated in a stratified patient populations or individuals, can improve the 436 delivery of and response to, nanomedicine therapies. 437 Therefore, unbiased corona analysis is essential for decoding the relationships between LNP 438 composition, corona composition, and LNP function. Using a streamlined process for corona 439 isolation, multi-omics corona assessment, and multivariate analysis, we successfully pinpointed 440 HDL as a determinate of LNP mRNA delivery and expression efficacy and, we confirmed this by 441 adding exogenous HDL. Interestingly, we uncovered a relationship between how HDL is 442 associated with the LNP coronas, the amount of free HDL and the cellular capacity for HDL-443 mediated uptake. At higher plasma concentrations, free HDL functioned antagonistically, whereas 444 reduced plasma concentrations resulted in lower free HDL, greater particle uptake and, 445 subsequently, improved mRNA expression. At the lowest lean plasma concentrations, there was 446 not enough HDL to populate LNP coronas, especially when using high doses of LNPs and this 447 degraded LNP performance. The number of HDL particles is, however, elevated in obese rat 448 plasma, so the LNPs remained functional even at low obese plasma concentrations, and under 449 these conditions, there was also less competition from unbound HDL. Taken together, these results 450 are first to indicate that LNP designs that encourage association with HDL or HDL components 451 will improve LNP performance by tipping the balance between LNP uptake and, free-HDL 452 blockade of cellular uptake mechanisms (such as SR-B1 53 ). The OPLS analysis also indicated that lipids (rather than lipoproteins) were contributive but less 465 decisive for LNPs' performance. Although one lipid hit LPC 20:1 did display a negative 466 correlation to LNP performance, and LNPs formulated with LPC demonstrated increased tropism 467 towards liver endothelial cells 56 , making LPC 20:1 a potential candidate for delivery to tissues 468 other than liver hepatocytes. 469 In summary, we have created an efficient method for isolating LNPs, with intact coronas from 470 plasma followed by an automated and unbiased mass spectrometric analysis of corona protein and 471 lipid content. We used this method to evaluate how corona composition affects LNP function (in 472 terms of effective mRNA delivery) by forming coronas in plasma from individual lean and obese 473 animals. Our study identified a lipoprotein fingerprint that promotes LNP function, leading us to 474 examine the role of HDL-LNP interactions as an essential factor for LNP efficacy. With these methods, it will now be possible to explore a greater variety of particles, biofluids, tissues, and 476 physiological states in order to more fully explore the relationships between lipid nanoparticle 477 corona content and delivery efficacy. While designing LNPs to promote particular corona 478 compositions is a significant engineering challenge, the complexity of these interactions creates 479 many opportunities for improving safety, reducing cost, targeting tissues and, tuning therapeutic 480 particles for specific biological and pathobiological contexts.  Animals 519 Ten week-old male lean and obese Zucker rats (Crl:ZUC-Lepr fa ) were purchased from Charles 520 River Laboratories (Maryland, USA) and group-housed (n=4/cage) in an Association for 521 Assessment and Accreditation of Laboratory Animal Care (AAALAC) accredited facility at 20-522 22°C and relative humidity of 40-60 % with a 12-h day/night cycle. The rats had free access to 523 water and a standard rodent chow diet (R70, Lactamin AB). At 20 weeks of age, the rats fasted for 524 4 hours, and a tail vein blood sample was obtained for glucose (Accu-Chek® Mobile, Roche 525 Diagnostics) and glycosylated hemoglobin (HbA1c, PTS Diagnostics) analyses. Thereafter, the 526 rats were euthanized using isoflurane anesthesia (Forene®, Abbott) and blood was collected from 527 the heart and EDTA plasma was isolated and stored at −20 °C. The experimental procedures were 528 approved by the local Ethics Committee for Animal Experimentation (Gothenburg region, 529 Sweden). 530

Imaging experiments and quantification 536
Cells were seeded at appropriate densities into CellCarrier-384 Ultra plates (#6007558, 537 PerkinElmer) in complete media a minimum of 16 h prior to treatment. At the experimental start, 538 the media on cells was removed and replaced with media containing the experimental treatment as 539 denoted in the relevant figures. The lean and obese rat plasma used were withdrawn as described 540 above and characterized as shown in Supplementary Fig.2   96 magnetic rod heads was used to separate the free-proteins, endogenous nanosized particles from 584 the LNPcor. Briefly, the incubation, wash, and elution procedures were performed using the optimized conditions indicated in Supplementary Fig. 10. The antibody-conjugated Dynabeads 586 were incubated with LNPs within media for 20 min at an Antibody: mRNA (wt:wt) ratio of 1 with 587 gentle mixing. At the end of incubation, the Dynabeads were extracted using magnet rods and 588 washed four times with PBS. A basic pH elution buffer containing 0.5 M NH4OH and 0.5 mM 589 EDTA was utilized to release LNPcor from Dynabeads. LNP pull-down quantification was 590 performed by Cy5 fluorescence readout. 591

Cryo-electron microscopy 592
For cryo-electron microscopy experiments, lean and obese LNP samples at a concentration of 593 ~10 13 particles/ml were incubated with glow discharged carbon-coated copper grids (SPI supplies), 594 following vitrification at 10 °C and 99 % humidity by using a Leica EM GP automatic plunge 595 freezer (Leica Microsystems Company). The excess sample was removed by blotting dry the grid 596 for 2.5 sec with filter paper and plunging it into liquid ethane at -180 °C. Following vitrification, 597 grids were stored immersed in liquid nitrogen until use. Before imaging, the grids were mounted 598 in a Gatan 626 cryo-holder (Gatan Company) and analyzed using an FEI Tecnai G2 Spirit BioTwin 599 transmission electron microscope (Thermo Fisher Scientific). Pictures were taken with a Morada 600 digital camera (Olympus Soft Image Solutions) and iTEM image capture software. 601

Proteomics analysis 602
Corona protein digestion was performed on recovered LNPs from individual plasmas containing 603 an equal amount of mRNA. Briefly, sample denaturation and reduction were performed using a 30 604 min one-step 8M urea (#U1250, Merck) and TCEP bond-breaker solution (#77720, Thermo Fisher 605 Scientific), followed by a 30 min alkylation step using a 2-chloroacetamide reagent (#22790,  Figure 1 The e cacy of LNPs mRNA delivery is individual plasma and dose dependent. a, The experimental design for evaluating LNP potency under lean and obese conditions. The candidate mRNA doses and the lean (LP) and obese pooled (OP) plasma concentrations resulting in the largest difference between lean and obese states were identi ed prior to exploring individual plasma samples. b, A series of mRNA doses (25-400 ng/well) were tested in the presence of LP or OP plasma (0, 1, 5 and 10%) using H4-II-E-C3, McA-RH-7777, Huh7 and NRK-49F cell lines over a 10 h time course. The cellular eGFP mean uorescent intensity (MFI) was quanti ed using image analysis and visualized as contour maps (each black dot represents the mean of n=3). In general, the eGFP expression increased over time. The obese plasma induced higher eGFP expression at the 1% plasma concentration in hepatocytes. c,d, The fold change of mRNA cargo expression (eGFP) and LNP uptake (Rhodamine label for lipids; and Cy5 label for mRNA) at the 10 h endpoint was calculated using the data from panel b as indicated by triangles (solid triangle: 50ng/well dose; hollow triangle: 200ng/well dose). At the 50 ng/well mRNA dose, the lean and obese plasma-complexed LNPs resulted in comparable cellular uptake and eGFP expression. The uptake of LNPs, following lean and obese plasma supplementation, was mildly improved at the 200 ng mRNA/well, whereas the eGFP expression was signi cantly elevated, particularly in rat hepatocytes. e, The difference in LNP e cacy was assessed using a 200ng/well mRNA dose in vitro. An apparent variation was observed between lean and obese plasmas, and among every individual. The error bars represent standard deviation of the mean (n=3).

Figure 2
The development of a nity based magnetic isolation of LNP-corona complexes. a, Schematic illustration of the ultrafast a nity-based, 96-well isolation method. Anti-PEG antibody conjugated magnetic beads capture LNPcor from plasma containing free protein, extracellular vesicles and lipoprotein particles. b, The design of experiment (DoE) space of LNPcor capture (epitope), wash and elution. The circles indicate where the majority of LNPs were detected. Only when combining an antibody against PEG backbone with PBS washing and basic elution conditions, were the majority of LNPs was identi ed in nal elution. c, The recovery ratio of LNPcor in terms of particle number, rhodamine labelled lipids (lipid) and Cy5 labelled mRNA (mRNA). Fig. 2a is created with BioRender.   High-density lipoprotein modulated LNPs performance. a, OPLS analysis to illustrate the correlation between corona contents and cellular eGFP expression. The orthogonal axis (orthogonal loading vector po of the X-part and the projection onto Y (so), poso) indicates the corona contents' orthogonality to eGFP expression; The predictive axis (X loading weight p and Y loading weight q combined to one vector, pq) implies the corona content's impact on eGFP expression. b, The correlation between corona ApoM or ApoE (z-score normalized) and eGFP expression in a variety of cell lines with Pearson correlation coe cient r. c, The correlation coe cient r between plasma ApoM or ApoE (z-score normalized) and eGFP expression in a variety of cell lines. d, The spike-in of HDL, but not VLDL and CM stimulated LNPmediated eGFP expression at 200 ng/well mRNA dose in Huh7 hepatocytes. e, HDL spike-in with different doses of LNPs and HDL particles in Huh7 hepatocytes. The curves reveal the relationship between LNP and HDL particle numbers. The error bars represent standard deviation of the mean (n=3). *P<0.05, **P<0.01, ***P<0.001, ****P<0.0001.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.