Detailed Structure Analysis of Tiny Lipid Nanoparticles Using Advanced Light and Neutron Scattering Techniques

With respect to previous studies on non‐spherical LNP based on carnauba wax [8], the LNP systems depicted in Table 1 were chosen for further structural investigations by the coupling of AF4 to SAXS and SANS for online analysis.

Hereby, LNP‐p resembles the placebo particle and LNP‐Q the quinine loaded one, both based on LNPs patented as CellInject for drug delivery applications [29]. Variation of the surfactant composition in LNP‐CW‐Tween and LNP‐CW‐TPGS further elucidates the influence of Polysorbate 40 (Tween) and vitamin E‐based d‐α‐tocopherol‐polyethylenglycol‐1000‐succinate (TPGS) on the particle morphology. These LNP consist solely of the lipid carnauba wax and the respective surfactant.

Key limitations of conventional AF4 in coupling with SAXS are the restricted injection volume due to channel overloading and the sample dilution during separation. These constraints significantly limit the sample concentration, thereby reducing SAXS signal intensity. To address excessive dilution during AF4‐SAXS coupling, we implemented an AF4 mode incorporating a dedicated Dilution Control Module (DCM). The DCM enables a controlled flow split directly at the channel outlet, i.e., after the size separation is complete. Under conventional AF4 operation, the entire cross‐section of the channel outlet, including both the particle‐rich lower region and the solvent‐rich upper region, is reaching the detectors, resulting in substantial dilution, particularly for low‐concentration samples. In the DCM mode, the outlet stream is divided into two adjustable fractions. The lower, analyte‐rich layer of the parabolic channel flow profile is directed to the detectors, whereas the upper, mostly particle‐free layer is removed as waste. This selective removal of the solvent‐rich portion preserves the peak concentration and delivers a stable stream to downstream detectors, leading to enhanced SAXS signal intensity without altering separation conditions inside the AF4 channel (Figure 1, details on the DCM approach can be found in Section S1).

As shown in Figure 2a for the quinine‐loaded LNPs, dilution‐controlled AF4 results in substantially higher average SAXS intensities compared to standard AF4 and measurements in semi‐batch mode (flow without separation), enabling more sensitive and precise structural characterization. The improved SAXS statistics extend the accessible scattering vector (q) range (Figure 2b), which is critical for the accurate determination of particle shape. SAXS data in the elution maximum with signal‐to‐noise ratio (SNR) > 2 can be obtained up to q‐values of 0.04 Å⁻¹ as shown for LNP‐Q which corresponds to length scales of approx. 15 nm. For SAXS data in semi‐batch mode and using standard AF4 q‐limits for similar SNR would be only 0.03 Å⁻¹ (20 nm) and 0.025 Å⁻¹ (25 nm) respectively.

Notably, the broader SAXS peak after dilution‐controlled AF4 results from a lower detector flow of 0.2 mL/min as compared to 0.5 mL/min for standard AF4. For a comprehensive description of AF4 separation approaches and data see also Supporting Information (AF4 systems: Figuresand, AF4 data: Figuresand, SAXS data: Figure). S1 S2 S5 S6 S13

The higher SAXS quality after dilution‐controlled AF4 enhances the contrast along the elution profile. This increased contrast also expands the range of the fractogram for which high‐resolution SAXS data are obtained (Figure 2c), thereby allowing for more comprehensive structural analysis over a larger portion of the sample population. This improvement is particularly evident in the derived shape parameters of the investigated LNP‐Q. The dimensions of separated LNPs were calculated using a form factor model of an ellipsoid of revolution with rotational symmetry around the polar (short) axis [30]. Figure 2c compares the polar (R_p) and equatorial (R_e) radii extracted from SAXS. Clearly, a standard AF4 limits the range of fractions which can be interpreted via online SAXS. In contrast, SAXS data collected after dilution‐controlled AF4 reveal smaller R_p for a wide range of elution times, indicating a pronounced elliptical shape (Figure 2d). On the one hand, the optimized separation increases the quality of the SAXS data, leading to higher resolution plateau‐values for polar and equatorial radius (RpSAXS=(9.3±0.1) nm and ReSAXS=(20.8±0.2) nm). On the other hand, the expanded SAXS‐fractogram unveils a distinct size progression for R_e from 20.4 to 29.8 nm vs almost constant R_p—a significant distribution in particle shape, not visible in batch SAXS or standard AF4‐SAXS.

In addition, dilution‐controlled AF4 allows an insight into a broad peak at 0.09 Å⁻¹ corresponding to a length of ≈ 7 nm (Figure 2b), which is detected for all LNPs (see Figure S17a). Briefly, the peak could either stem from a surfactant‐bilayer structure factor [9] or the lamellar stacking of crystalline wax esters within the carnauba wax [31]. The internal structure of the particle is discussed more closely in our previous report [8]. In the current study, we focus on the overall shape and size distribution of the LNPs.

In the same previous work [8], we observed deviations between the LNPs aspect ratios (R_e: R_p) obtained from orthogonal characterization via batch laboratory SAXS (2:1) and cryo‐TEM experiments (4:1). It is obvious that batch lab SAXS delivers average values (in addition to lower resolution compared to synchrotron SAXS), while cryo‐TEM gives only discrete values of a very limited number of particles. Thus, separation prior to scattering detection is the only reliable method to get the full distribution picture: an aspect ratio of 2:1 during the first 4 min of the separation, which increases to approximately 4:1 during subsequent elution until complete separation (Figure 2c).

Compared to SAXS, SANS offers a variation of contrast which has the potential to increase the information density of an AF4‐online experiment. A first‐of‐its‐kind online dilution‐controlled AF4‐SANS experiment was thus performed on the same LNPs analyzed with AF4‐SAXS. We obtained a sufficient SANS contrast in the range 0.007 < q < 0.1 Å⁻¹ (90> d > 6 nm) as shown for LNP‐Q (Figure 3a; Figure S21a), which significantly extends to higher q as compared to SAXS, allowing deeper insight into the LNP structure. The scattering intensities vs. q vary in the first 5 min of the separation (Figure 3b), while I(q) remains constant beyond the elution maximum (Figure S21b). This is opposite to the observed trends in AF4‐SAXS with constant I(q) before the elution maximum and changes in the SAXS scattering profile at a later elution time (Figure S14).

At the elution maximum, the low q‐range SANS data (0.007 < q < 0.1 Å⁻¹) was combined with the integrated data of the high‐q‐range (0.045 < q < 0.6 Å⁻¹). The integration over the complete peak was required due to insufficient intensity of the single data point (Figure 3a, green fractogram). The resulting scattering curve was modelled with a core–shell ellipsoid FF (Figure 3c; Figure S24a). The overall dimensions derived from the FF model at the maximum were RpSANS=(7.7±0.4) nm, and ReSANS=(18.2±0.9) nm aligning well with AF4‐SAXS‐derived sizes (Figure 2c). Normalization of the SANS data, i.e. determining the intensity I₀ (Figures S22 and S23), allows to evaluate the scattering length densities of the core and the shell. For LNP‐Q, we find a core with slightly lower scattering length density SLDcoreSANS=(6.351±0.001) 10⁻⁶/ Ų as compared to the shell SLDshellSANS=(6.359±0.001) 10⁻⁶/ Ų. Although the data is not corrected for the (unknown) absolute particle concentration, these values allow for a relative comparison and confirm that the shell and core of the LNP exhibit different contrasts in D₂O. Based on the theoretical SLDs of the LNP components (Table S5), one interpretation of the data could be a particle internal structure of a more solid core and a more swollen shell in D₂O. However, previous SAXS and cryo‐TEM studies suggest rather incorporated water in the core [8]. This is a more plausible explanation of the SLD contrast between core and shell, considering (i) a proton‐rich core, due to encapsulated H₂O from the particle formation step and (ii) the fact that the hydrophobic shell is most probably not penetrable for D₂O. Considering that the non‐loaded LNP‐p has a similar core–shell structure (Figure S24c), we calculate the same core SLD but higher SLDshellSANS=(6.365±0.001) 10⁻⁶/ Ų. The composition of LNP‐Q and LNP‐p differs only in the quinine content. This fact and the lower SLD of LNP‐Q in the shell is a clear indication, that the quinine (SLD in Table S5) is preferentially located in the shell rather than in the core.

Focusing again on the size evolution from the SANS fractogram collected with sufficient SNR at 0.007 < q < 0.1 Å⁻¹, FF modeling of an ellipsoid of revolution was successfully applied. Hereby, a different trend compared to AF4‐SAXS was observed (Figure 3d vs. Figure 2c): The equatorial radius modeled from SANS increased during early elution before reaching the plateau (ReSANS∼18 nm), while constant values of ReSAXS∼20.7 nm from SAXS were obtained before the elution maximum. The polar radius remained constant during the separation for both SANS and SAXS, but with ∼2‐3 nm smaller SANS‐values.

The general discrepancy between SANS and SAXS derived sizes, with SANS sizes being systematically smaller than those from SAXS, may stem from screening of the outer polar layer by deuteration in D₂O, potentially reducing the apparent shell thickness in the SANS measurements.

Comparing the time progression of average detected intensities after dilution‐controlled AF4 from SAXS (Figure 2a, UV‐signal in Figure S6) and SANS (Figure 3a, UV‐signal in Figure S7), it is evident that the main elution peak in the SAXS experiment was considerably narrower than for SANS. An obvious reason for this is that the AF4‐SANS separation takes place in D₂O, which possesses higher viscosity compared to H₂O, which influences the separation efficiency. In addition, SAXS and SANS allow for detection in different size ranges, with SANS resolving a broader size range (Figure 3). To understand these effects, a precise comparison to other detectors coupled within the AF4 multidetector system is considered in the next section.

To critically assess the accuracy of particle size determination, we compared radii obtained from light scattering (MALS, DLS) and those derived from SAXS and SANS following dilution‐controlled AF4. For this purpose, we focus on the non‐loaded, placebo LNP‐p.

For better comparison, the zero‐point of time axes of concentration‐dependent UV signals and scattering intensities (SAXS, Figure 4a; SANS, Figure 4b) was shifted to the elution peak maximum. Notably, SAXS and SANS exhibited signal broadening as compared to the UV signal, attributed mainly to a large inter‐detector volume after AF4. In both cases, band broadening at high elution times occurs due to the low flow rate of 0.2 mL/min. R_g values from SAXS and SANS (Figure 4a,b) remained relatively constant, though SANS‐derived R_g was systematically lower—consistent with prior LNP‐Q model data. Remarkably, in the elution maximum the same R_g of 14.4 ± 0.6 nm is obtained from SAXS and MALS data. This value also corresponds to the lower theoretical limit of static light scattering (∼16 nm).

Before the elution maximum, R_g from SAXS and SANS are higher than MALS‐derived R_g. Figure S9 indicates that the Zimm plot fit quality is below 90% up to the maximum of the elution curve, which confirms that static light scattering (λ = 660 nm) is limited for sizes below λ/20. In contrast, SAXS and SANS deliver reliable sizes in this range.

The increase in MALS‐derived R_g after the elution maximum cannot be found in the data from SAXS and SANS. MALS probes close to zero angle and is therefore highly sensitive to a small fraction of larger species (> 20 nm). In contrast, our SAXS and SANS (q > 0.003 and 0.007 Å⁻¹, respectively) data only partly cover, or completely miss, the Guinier regime of these aggregates, so that they contribute only a weak low‐q signal. At the late elution times where MALS detects these species, their volume fraction is low and the primary LNPs (< 20 nm) dominate the SAXS/SANS signal, placing any aggregate contribution within the noise level. This difference rather indicates contributions from loosely bound aggregates/associates with diluted interconnected shells, detectable by MALS, but not by SAXS/SANS. The relatively high concentrations used to obtain reliable SNRs (1 mg/mL (SAXS) and 2.7 mg/mL (SANS)) could promote aggregate formation, which in fact, was not observed at lower concentrations relevant in drug delivery systems [8]. Given the low scattering contrast between the interconnecting shell and the eluent in SAXS and SANS, these techniques primarily resolve the individual particles. Co‐elution of differently sized species can be ruled out, as evidenced by the consistent fits of the DLS correlation curves across the fractograms, which confirm narrow size‐distributed fractions (Figures S10 and S11). In summary, MALS reliably maps the size distribution beyond the elution maximum (for values larger than approximately λ/20), whereas SAXS and SANS provide precise measurements of the individual LNP dimensions. This interpretation is further supported by the distribution of R_h across the entire peak obtained by online DLS, which, unlike MALS, does not depend on anisotropic scattering but rather on fluctuations in scattering intensity. The R_h trend aligns well with the above conclusions. However, DLS remains limited to probing the global hydrodynamic size associated with diffusion properties and does not provide structural information.

To shed light on the LNP structure, next to R_g determination, FF modeling was performed for the LNP‐p and discussed in relation to R_h (Figure 4c,d). If we compare equatorial and polar radii from SAXS and SANS again, systematically smaller radii from SANS are evident. As for the LNP‐Q particles, ReSANS increases up to the elution maximum, while ReSAXS is constant before reaching the maximum. Since these findings for LNP‐p and LNP‐Q could, in principle, also stem from composition‐dependent deuteration effects, we now have a closer look at the hydrodynamic radius derived from DLS data. The detected hydrodynamic radii differ slightly between AF4‐DLS‐SAXS and AF4‐DLS‐SANS, but follow the trends observed in SAXS and SANS at lower elution times (Figure 4c,d). R_h follows remarkably well the trend observed for ReSANS up to the maximum of the elution curve, which as a hydrodynamics‐based value matches the expectations for ellipsoidal particles. Beyond the maximum, R_h again indicates larger particle sizes, while sizes derived from SANS stay constant. Hereby, the same trends in R_h and ReSANS before the elution maximum confirm improved separation efficiency of AF4 in D₂O compared to H₂O. We propose that the experiments conducted in D₂O provide improved resolution in separating smaller particles compared with those performed in a standard water‐based buffer. The higher viscosity of D₂O reduces the diffusion of all particles. Combined with the high sensitivity of SANS in the lower size range, this enables the detection of the smallest particle populations.

In summary, particle separation and size detection by different scattering detectors lead to complementary insights into size distributions, elucidating particle shape and composition progression in a comprehensive manner.

Surfactants play a decisive role in self‐assembly, size distribution, and shape of LNPs. In a previous report, we found that the dual‐surfactant system comprising TPGS and Tween yields LNPs with narrow size distributions and low polydispersity. In contrast, single‐surfactant systems (TPGS or Tween alone) produce broader size distributions, indicative of less controlled assembly. The stabilization provided by both surfactants was sufficient to prevent aggregation [8].

To probe the effect of the type of surfactant on particle geometry, we evaluated LNPs composed solely of carnauba wax and stabilized by either Tween 40 (LNP‐CW‐Tween) or TPGS (LNP‐CW‐TPGS), alongside dual‐surfactant and drug‐loaded systems (LNP‐p and LNP‐Q). SAXS‐ and SANS‐derived ellipsoidal parameters at the maximum of the elution curve revealed a clear surfactant‐dependent trend (Figure 5a): the polar radius (R_p) increased from LNP‐CW‐Tween < LNP‐Q ≈ LNP‐p < LNP‐CW‐TPGS, while the R_e was highest for LNP‐CW‐Tween, and lower but similar for all other particles containing TPGS (Figure 5a). These trends indicate that Tween 40 favors more elongated, elliptical particles, whereas TPGS promotes less elongated shapes, an important consideration for drug delivery applications where particle shape can influence biodistribution and cellular uptake [32]. Notably, all carnauba‐wax‐only formulations still displayed ellipsoidal geometries (R_e/R_p = 1.6 for LNP‐CW‐TPGS), highlighting the intrinsic tendency of carnauba wax to form non‐spherical particles. For the dual‐surfactant formulations loaded with (LNP‐Q) or without drug (LNP‐p), the aspect ratio R_e/R_p was 2.2 to 2.5. The anisotropic shape of all particles is also confirmed by the shape parameter calculated from the ratio R_g/R_h in the range 0.65–1.3 (Figure S12 and Table S4).

As discussed before, the R_e values derived from SANS were systematically smaller (3.2 ± 0.2 nm) than R_e derived from SAXS. For LNP‐p and LNP‐Q, also the R_p from SANS were smaller (2.0 ± 0.2 nm). Since both Tween 40 and TPGS include PEG segments (on average 20–22 units), we infer that the outer shell is formed by the surfactant headgroups, corresponding to approx. 3.4 nm thickness as indicated in previous SAXS studies [9]. Notably, LNP‐CW‐Tween showed broad R_e (Figure 5b) and R_p distributions as compared to the narrow distribution of R_p for TPGS containing particles (Figures S18 and S25), indicating a polydisperse mixture of elliptical particles. Here, the R_h correlates well with R_e from SANS, suggesting a genuine size progression of the elliptical LNP‐CW‐Tween particle shape, with an aspect ratio of R_e/R_p = 3.3 ± 0.4 in the maximum of the elution curve (Figure 5a).

For LNP‐CW‐TPGS, a core could be modelled from SAXS data in the elution maximum utilizing a core–shell ellipsoidal FF (sketch Figure 5c; Figure S17). The resolved core has an almost spherical shape with a radius of 10–12 nm, comparable to the extended length of the LNP‐Q and LNP‐p cores (Figures 3c and 5c). Since LNP‐CW‐TPGS consists solely of carnauba wax and TPGS, this implies that the core–shell structure is defined by these two components in the LNP. In contrast, a core–shell structure could not be found for the LNP‐CW‐Tween system (Tween 40 + carnauba wax, SAXS data Figure S17a, SANS data Figure S24b). SAXS intensity normalization enabled fitting of the SLDs for these two LNPs: For the LNP‐CW‐Tween particle, without core–shell characteristics, we obtain an SLD^SAXS = 9.419 × 10⁻⁶/Ų, which is an average of the SLD values for the core and the shell in LNP‐CW‐TPGS (SLDcoreSAXS=(9.428±0.001)×10−6/Å2 and SLDshellSAXS=(9.409±0.001)×10−6/Å2). Compared to theoretical SLDs (Table S5), this result suggests a diluted core and a more compact shell for LNP‐CW‐TPGS, which is in contrast to reported LNP structures [32] but in agreement with our conclusions made from the SANS analysis of LNP‐Q and LNP‐p and confirmed by batch lab SAXS analysis combined with cryo‐TEM [8].

LNPs, patented as CellInject, were prepared as described elsewhere [8, 29]. Briefly, carnauba wax (1.00–5.00 w/w parts), red palm oil concentrate 0.10–0.50 w/w parts), d‐α‐tocopheryl polyethylene glycol 1000 succinate (TPGS) (0.70–3.50 w/w parts) and Polysorbate 40 (Tween) (0.50–2.50 w/w parts) are mixed and heated up to 90°C under stirring until a clear melt mixture is obtained. Quinine (0.1–0.5 w/w parts) is added to the mixture at 90°C for LNP‐Q. Pre‐heated NaCl (0.9 w/w parts) in deionized water (90°C) is added dropwise to the mixture. Finally, the dispersion is cooled down to 20°C under stirring. For LNP‐CW‐Tween and LNP‐CW‐TPGS only carnauba wax and Tween 40/TPGS are used respectively in the initial mixture.

All AF4 experiments were performed using an identical channel configuration comprising a trapezoidal short channel (tip‐to‐tip length 175 mm, maximum width 21 mm) with a 350 µm spacer thickness, fitted with a 10 kDa regenerated cellulose (RC) membrane. The specific AF4 systems used are detailed below. A detailed description of the Dilution Control Module (DCM) is given in the Supporting Information.

AF4 separations were conducted using an Eclipse 3+ field‐flow fractionation system (Wyatt Technology, Santa Barbara, CA, USA. The separation system was coupled to the following detectors: A DAWN EOS MALS detector (Wyatt Technology, λ = 660 nm) and a diode array UV–vis detector 1260 (Agilent Technologies, USA, λ = 280 nm). The eluent consisted of ultrapure water. Semi‐batch measurements were performed by direct channel injection without applied focus and cross‐flow at a flow rate of 0.5 mL/min. A total of 0.260 mg of sample (1.04 mg/mL) was loaded by injecting 250 µL. See Figurefor the optimized separation profile. S1

AF4 separation was performed using an Eclipse NEON system (Waters | Wyatt Technology, USA) with Dilution Control Module (DCM) option coupled to a multi‐detector setup consisting of a (1) DAWN NEON MALS detector (Waters | Wyatt Technology, USA, λ = 660 nm) with DLS option, (2) an Optilab T‐rEX differential refractive index (dRI) detector (Waters | Wyatt Technology, USA, λ = 660 nm), and (3) UV detector 1260 (Agilent Technologies, USA) (system 2)/UV detector SPD‐30 (Shimadzu, Japan) (system 3). The integrated Dilution Control Module (DCM) minimizes sample dilution without compromising resolution, improving fraction concentration and sensitivity. The split ratio (SR) of DCM flow and detector flow was set to 8.

For system 2, a total of 0.260 mg of the sample (1.04 mg/mL) was loaded by injecting 250 µL. For system 3, the MALS detector was equipped with wide‐bore flow cells, attenuators, and fluorescence filters. The SANS flow cell was placed in between the MALS and the dRI detector. The eluent consisted of 99.9% D₂O. A total of 0.405 mg of sample (2.7 mg/mL) was loaded by injecting 150 µL. For more details on systems 2 and 3, and for the optimized separation profiles see the (Figures S2 and S3).

Astra 8.3 software (Waters | Wyatt Technology, USA) was used for the calculation of size distributions. The angle‐dependent scattering of the LNP systems was evaluated using the Zimm model which was found suitable in the regime of small particle radii (below 50 nm). It provides markedly higher fitting stability and often improved robustness compared to Berry or Debye fits. For LNPs, the angular dependence is weak, and the linear Zimm approximation is less sensitive to noise and less prone to overfitting in low singal to noise regimes than higher‐order Berry/Debye models.

Measurements were performed at the CoSAXS beamline at MAX IV (Lund, Sweden) at a wavelength of 0.99 Å. The data was collected on an Eiger2 4 M (Dectris) detector within an evacuated flight tube, positioned at 3.5 m distance (0.003 < q < 0.29 Å⁻¹), and at a scan rate of 1 s per frame. The flow cell consists of a 1.5 mm inner‐diameter quartz capillary, with a 10 µm wall thickness. The sample in 1 mm PBS buffer passed the SAXS cell at a flow rate of 0.2 mL/min after dilution‐controlled AF4. Reduction of the 2D scattering pattern to 1D scattering intensity was performed and the data corrected for the empty buffer background (integration of 300 frames), subsequently. 2D data was controlled for anisotropy (Figure S20). 1D SAXS data at low q was analyzed by Guinier fits, and R_g and I₀ extracted framewise by self‐written R‐scripts. The SAXS data was normalized to I₀ to correct for concentration dependency. Normalized data was fitted framewise to FF models of ellipsoid or core–shell ellipsoid of revolution (about the short axis) using SasView [30]. A cylindrical FF was also tested, which led to similar sizes (LNP‐Q), but had lower fit quality.

Small angle neutron scattering measurements [33] were carried out at the D11 instrument of the ILL (Grenoble, France), at a neutron wavelength of 5.24 Å (chosen for the highest available flux). Integration time was 30 s per frame. The sample‐to‐detector/collimation distances were at 10.2 m/10.2 m (0.007 < q < 0.1 Å⁻¹), and also at 1.6 m/4.1 m (0.045 < q < 0.6 Å⁻¹) for selected samples. A home‐built SANS flow‐through cell [28] was used with an exposed volume of 133 µL (11 mm diameter and 1 mm thickness). The sample in 99.9% D₂O (1 mm PBS) buffer passed the SANS cell at a flow rate of 0.2 mL/min after dilution‐controlled AF4. 2D SANS data was checked for anisotropy (Figure S27), scaled to absolute intensity using direct flux measurement and transmission‐corrected. As a background, the cell and buffer contributions were subtracted as well as the blocked beam contribution. Patterns were reduced to 1D scattering curves using the software GRASP [34]. For Figure 3a, intensity per pixel was averaged from 2D data over 2 circular boxes: one from q = 0.007 to q = 0.1 Å⁻¹, and one from q = 0.045 to q = 0.6 Å⁻¹. 1D SANS data at low q was analyzed by Guinier fits, and R_g and I₀ extracted framewise by self‐written R‐scripts. The SANS data was normalized to I₀ to correct for concentration dependency. Normalized data for 0.007 < q < 0.1 Å⁻¹ was fitted framewise to the FF of an ellipsoid of revolution (about the short axis) using SasView [30]. In the maximum of evolution, data for LNP‐Q and LNP‐p was stitched to the integrated data of 0.045 < q < 0.6 Å⁻¹ using PRIMUS [35], and fitted using a core–shell ellipsoid FF model using SasView [30].

All reported errors for the SAXS and SANS data correspond to one standard deviation of the respective fitted parameter.