The lipid samples were prepared using Capmul GMO-50, diglycerol monooleate (DGMO) and polyoxyethylene (20) sorbitan monooleate (P80). Capmul GMO-50 (Lot # 190111-9, Abitec, United States) was composed of 53.8% monoglycerides, 15–35% diglycerides, and 2–10% triglycerides with the following fatty acid composition: 86.6% oleic (C18:1), 4.9% linoleic, 4.1% palmitic and 3.3% stearic. DGMO (DGMO-90V, Lot # 9147) was purchased from Nikkol Chemicals (Japan) and P80 (Lot # 1102YP0133) from Croda (Limhamn, Sweden) was kindly provided by Camurus (Lund, Sweden).

Equine skeletal muscle myoglobin (Mb) (Sigma Aldrich) and purified BvPgb1.2 (see 2.2) were dissolved in 50 mM Tris-HCl pH 8.5 or milli-Q purified water (MQ, 18 MΩ cm) to make a stock solution of 60 mg/mL.

Wild type (WT) BvPgb1.2 (GenBank: KF549981) was expressed in BL21-DE3 Escherichia coli using a pET-DEST42 vector as described previously (Leiva Eriksson et al., 2019). The production of BvPgb 1.2 was conducted following previously described methods (Leiva Eriksson et al., 2019), but here the Q-Sepharose HP step was omitted. The final BvPgb1.2 fractions had an absorbance ratio at 412 nm and 280 nm (A₄₁₂/A₂₈₀) in the range 2.5-3, corresponding to >95% purity. BvPgb1.2 preparation was lastly concentrated using 10 kDa Vivaspin^® 20 mL ultrafiltration units (Vivascience) and stored at −80°C before use.

Hemeprotein concentration was determined by recording the spectra between 500 and 600 nm using a Cary 60 UV-Vis system (Agilent Technologies) by adding 0.2 mL Pyridine (Fischer Scientific), 0.1 mL 1 M NaOH, H₂O and sample up to 1 mL in a cuvette. By supplementing 0.1 mg of sodium dithionite to the protein, the reduced spectrum was obtained and Hb concentration was determined from the absorbance difference between peaks 556 to 540 nm and the millimolar extinction coefficient for heme (20.7).

The final concentrations used for BvPgb1.2 and equine skeletal Mb (Sigma) were 60 mg/mL. Both proteins were diluted/dissolved in 50 mM Tris-HCl pH 8.5 or MQ water for further characterization.

Dynamic light scattering (DLS) measurements were performed on a Zetasizer Nano ZS (Malvern Instruments Ltd., United Kingdom) using ZEN0040 disposable micro cuvettes filled with 1 mL of sample. Before measuring, dispersions were diluted to 0.1 wt% and proteins were diluted to 6 mg/mL with their dispersing solvent, which had been filtered with a 0.2 μm syringe filters (PALL^® Acrodisc^® 13 mm 0.2 μm PVDF). The temperature was set to 25°C and all samples were allowed to equilibrate for 120 s before measurement. The apparent hydrodynamic diameter was calculated using the Einstein-Stokes equation, with the viscosity of water, η, as 0.8872 cP and assuming spherical particles. For each sample, 5 measurements were collected and the standard deviation calculated.

To examine possible effects on and leakage from the nanoparticles, BvPgb1.2 and Mb-containing nanoparticles with varying protein loading concentrations (0–60 mg/mL) were used. Analytical size exclusion chromatography (SEC) was carried out with a Superdex 7,510/300 GL column (CV = 24 mL) on an ÄKTA™ Avant system (Cytiva Life Science, Uppsala, Sweden) calibrated with conalbumin (75 kDa), carbonic anhydrase (29 kDa), ribonuclease A (13.7 kDa), and aprotinin (6.512 kDa) as molecular weight (Mw) standards (Cytiva Life Science) and 50 mM NaP+150 mM NaCl pH 7.2 was used as running buffer (Christensen et al., 2022). All standard protein solutions were run in volumes of 100 µL with a protein concentration of 1 mg/mL. The same sample volume was applied for all nanoparticles. Different filter pore sizes (0.45, 1.2 and 5 µm) (Sarstedt) were tested for nanoparticles. The largest pore size (5 µm) was used in the final preparations of the particles before column loading. The Mw of the BvPbg and Mb solutions were estimated by fitting the measured retention volume (Vr) to a semi-log plotted calibration curve (Christensen et al., 2022).

Bulk phase samples with two different P80 fractions (30/45/25 DGMO/GMO-50/P80 and 28/42/30 DGMO/GMO-50/P80), but a constant DGMO/GMO-50 ratio of 40/60, were prepared with a range of BvPgb1.2 and Mb concentrations in 50 mM Tris-HCl buffer pH 8.5. Clear physical differences between the samples were observed, as the 25% P80 samples for both proteins had a rigid gel-like consistency, whereas all of the 30% P80 samples were fluid. The colour intensity increased with protein concentration; all the BvPgb and the 25% P80 Mb samples were red, whereas the 30% P80 Mb series were brown, indicating strong oxidation. SAXS measurements of these bulk samples showed that the P80 content strongly affected the phase behaviour, while mostly small changes were observed with increasing protein concentration. In the 25% P80 samples (Figures 1A, B), the structure changed from inverse bicontinuous cubic with Im3m space group to Pn3m space group with increasing protein concentrations. This transition occurred at a lower protein concentration for Mb than BvPgb and in the trace for the highest concentration of Mb there was an additional peak at q = 0.046 Å⁻¹, indicating coexistence with another phase.

In the 30% P80 samples (Figures 1C, D), the two diffuse peaks characteristic of the sponge phase were observed, where the lower q peak of the sponge phase (q ≈ 0.05 Å⁻¹) relates to the cell-cell correlation distance and the higher q peak (q ≈ 0.11 Å⁻¹) relates to the bilayer thickness (Valldeperas et al., 2016). These were present for all concentrations of protein, with only very minor shifts of the peaks. In Figure 1C, there is a sharp upturn at low q, which increases in magnitude with increasing [BvPgb]. This low q upturn in scattering intensity indicates the formation of aggregates with a larger length scale than is covered by the available q range. It should be noted that aggregates could be observed visually in these samples (see Supplementary Figure S1). In some of the 25% P80 samples with Mb and BvPgb1.2, these upturns can also be observed, but without such a clear effect of concentration. We also note that the upturn was associated with a shift of the peaks to lower q that reflects dehydration of the phase. Thus, the visible aggregates likely reflect excess of water in the sample. Conversely, in Figure 1D, increasing [Mb] results in a minor broadening and shift to higher q of the higher q peak. This indicates that additional Mb causes the bilayer to become more disordered, which likely results in higher head group hydration.

From these SAXS patterns, it is possible to estimate the diameter of the water channels in the structure, as plotted in Figure 2. For the inverse bicontinuous cubic structures, the calculation was made using the infinite periodic minimal surface (IPMS) that the structure is based on. For the sponge phase, it was made by extrapolating from the the characteristics of the cubic phase formed at the same lipid ratio along the dilution line, but at higher lipid content, the calculation was made by comparison to the cubic phase immediately preceding it. For the 30% P80 samples, all of which retained the sponge phase structure, there was very little change in water channel diameter with both proteins. For the 30% P80 with BvPgb, there was a slight decrease in water channel diameter, whereas for Mb, the water channel diameter stayed at approximately 9 nm for all the Mb concentrations measured. For the 25% P80 samples, however, very different behaviours were observed. For the BvPgb, there was little change in water channel diameter even with the phase change from Im3m to Pn3m. For Mb, the size of the water channel decreased with increasing protein concentration until the highest concentration, where we observed an additional structure.

Some interesting behaviour was observed for the highest concentration of BvPgb in the 30% P80 samples. The sample initially appeared homogenous when loaded into the SAXS sandwich cells. After leaving the sample in a vertical position for 24 h in the holder, a distinct region of higher BvPgb concentration could be observed (see inset of Figure 3). SAXS measurements of the protein rich and poor regions showed that the sponge phase structure was retained in each region and peak positions were similar to the sample before separation. However, the low q upturn was sharper in the protein rich region even than the pre-separation 60 mg/mL BvPgb sample. After placing the sample in a horizontal position, the sample returned to the pre-separation appearance (see Supplementary Figure S2). This cycle was repeated, indicating that this protein redistribution is reversible. This behaviour might indicate that the sample contained one phase rich in protein and another where the protein is depleted. Both phases had a similar structure of the lipid matrix.

Dispersions were prepared from the bulk samples with varying concentrations of Mb and BvPgb with different appearance depending on the starting bulk structure. For the 25% P80 sample series, there was no clear trend of change in colour with concentration (see Supplementary Figure S3). The dispersions with Mb appeared homogeneous and there were no visible aggregates. For the BvPgb samples, however, large translucent red/orange aggregates were observed at all concentrations and the number of aggregates increased with concentration. For both the 30% P80 sample series, the intensity of the colour increased with concentration and the dispersions appeared homogeneous (see Supplementary Figure S4). However, for the BvPgb series, a small number of opaque dark red aggregates were observed at the edges of the vials, most likely the protein aggregates observed at the edges of the corresponding bulk phases.

Broad peaks were observed in the SAXS patterns for all samples (Figure 4), with more well-defined and rounded peaks in the 30% P80 samples. The concentration dependent low q upturn observed in the bulk samples were also present in both the 25% and 30% P80 dispersions with BvPgb. However, for the 25% P80 samples with Mb the low q upturn was found to be independent of protein concentration.

Due to the large size of the aggregates in the 25% P80 BvPbg sample series, the filtration required to run size exclusion chromatography was not possible. For the 30% P80 samples, however, size exclusion chromatography was performed to analyse the amount of free protein in the dispersion, either from leakage or protein that was not originally encapsulated.

Firstly, empty nanoparticles (no loaded protein) were run as a control and this is shown in Figure 5. Here, the particles eluted at ∼7.5 mL, corresponding to the maximal separation capacity of the column (75 kDa). As expected, no additional significant peak around 11–13 mL was observed, where both proteins would be detected. Thus, this spectrum can be compared to the protein-containing nanoparticles to evaluate the degree of free protein present in the samples.

Nanoparticles containing BvPgb1.2 or Mb were analysed with different proteins loads. The results for the lowest and highest concentrations are shown in Figure 6. No free Mb was detected for 15 mg/mL (Figure 6A), showing similar results as the empty particles. However, a clear peak at 12.8 mL elution volume was observed for the highest protein load (60 mg/mL, Figure 6B). This is the expected elution volume for the 16.9 kDa protein, indicating presence of free Mb in the sample. In addition, the strong peak at 412 nm indicates a heme-containing protein was eluted. This was not observed for BvPgb1.2, either for 15 or 60 mg/mL (Figures 6C, D, respectively). The lower concentration showed no free protein at all, whereas the highest concentration showed a minor peak at 11.9 mL. However, the expected elution volume for the dimeric BvPgb1.2 corresponds to ∼10.9 mL (38.4 kDa). Thus, the observed peak was presumably a minor contamination, probably from the encapsulation, rather than monomeric, dimeric or oligomeric BvPgb1.2.

In the 25% P80 dispersions, aggregation of the proteins clearly plays a role, as apparent from the low q upturn observed in the SAXS curves for the Mb concentration series and the extensive aggregation of BvPgb at the bottom of the vials in the BvPgb series. As protein stability is highly dependent on the environment, there are a range of factors that could affect the protein stability in the sponge phase, including interactions with the sponge phase lipids, and pH and ionic strength (Talley and Alexov, 2010; Deller et al., 2016). Although the dispersing solvent easily diffuses in and out of the water channels in the bicontinuous phases, the buffering effect of the Tris-HCl buffer in the bulk systems is affected by the large area of lipid aqueous interface. The “local pH” close to the lipid interfaces can therefore drop close to or below the isoelectric point, pI, of the proteins, which may result in aggregation. The pI of Mb is 7.3 (major component) and 6.8 (minor component) (Zhao et al., 2017) whereas the corresponding pI values for BvPgb are 7.85 (monomer) and 8.32 (dimer) (see Supplementary Material for calculation), making the BvPgb more sensitive to any pH fluctuations. We therefore prepared an additional set of samples with Mb of different concentrations included in a lipid dispersion with 25% P80 in 50 mM Tris-HCl buffer pH 8.5 instead of MQ.

SAXS data from this dispersion series containing the lowest and highest Mb concentration are plotted with the SAXS data for the corresponding dispersions in MQ water for comparison in Figure 7 (SAXS data for the full concentration series are shown in Supplementary Figure S7). The SAXS curves showed broad peaks for both Mb concentrations in MQ and Tris-HCl. The peaks in MQ were more well defined than for the Tris-HCl samples. Additionally, the sharp upturn at low q seen in the MQ dispersion data was either not present or greatly reduced for all of the dispersions in Tris-HCl.

Figure 8 shows the correlation curves from the corresponding DLS measurements for these samples. The data for the dispersions in MQ have a similar shape for all Mb concentrations. However, with increasing Mb concentration in Tris-HCl buffer, the y intercept of the curve decreases and the growth of a shoulder at longer relaxation time indicates the increasing contribution of larger objects, which most likely were sedimenting (Stetefeld et al., 2016; Hawe et al., 2020).

Although there was a clear case for the buffer effect on the encapsulated proteins, it should be noted that there is a possible buffer effect on the lipids too. In previous work, the DLS data for these samples typically had a polydispersity index (PDI) of <0.26 (Valldeperas et al., 2016; Gilbert et al., 2019) for 30% P80 without protein, while here we observed a typical PDI of >0.4. This value increased to >0.5 with addition of the protein. In order to investigate this effect, lipid mixtures containing 25, 27.5% and 30% P80 were dispersed directly into MQ or 50 mM Tris-HCl buffer pH 8.5.

The mean hydrodynamic diameter of the dispersions was very similar for all samples (summarised in Figures 9A, B), while the PDIs for the dispersions in Tris-HCl were significantly lower than those in MQ. Although they cannot by directly compared due to the difference in dispersion methods, as described in the Supplementary Material, the PDI of the dispersions in MQ here are similar to those observed in previous work (Gilbert et al., 2019).

In the corresponding SAXS data (Figure 9C), the same trend as for the 25% P80 Mb concentration series dispersed in MQ and Tris-HCl was observed. The dispersions in both solvents showed broad peaks, but with a less well defined, lower intensity peak for the dispersions in Tris-HCl.

Self-assembly of lipids into different structures can be affected by many factors, including lipid composition, ionic strength, pH and presence of additives, such as proteins (Lindblom and Rilfors, 1992; Seddon and Templer, 1995; Hyde and Schröder, 2003; Boyd et al., 2009). Both Mb and BvPgb have relatively hydrophilic amino acids at the surface and hydrodynamic diameters that are smaller than the calculated water channel diameters, 4.3 ± 0.1 nm for Mb and 6.7 ± 0.3 nm for BvPgb (see Supplementary Table S1). This indicates that they would most likely be situated in the water channels of the bulk phase and/or interacting with the hydrophilic head groups of the lipids.

As seen when comparing Figures 1A–D, increasing the amount of P80 in the lipid mixture changed the structure formed from a higher curvature (inverse bicontinuous cubic phase) to one with lower curvature due to the large head group of P80 (Valldeperas et al., 2016). This effect has previously been observed in work comparing the ternary DGMO/GMO-50/water system and quaternary DGMO/GMO-50/P80/water systems (Valldeperas et al., 2016). For both proteins studied a transition from the bicontinuous cubic Im3m to Pn3m structure occurred for the samples with 25% P80 with increasing protein concentration (Figures 1A, B). This cubic-cubic phase transition has typically been associated with a decrease in hydration (Mezzenga et al., 2019), indicating that there are protein-lipid head group interactions for both proteins. This has previously been shown for similar systems in which water-lipid head group bonds are replaced by protein-lipid head group bonds, decreasing hydration of the head groups (Gilbert et al., 2019; Gilbert et al., 2022). For both 30% P80 protein series, the sponge phase was retained for both protein concentration ranges measured here with some minor changes in peak position.

For both BvPgb sample series (Figure 2, left), little change in water channel size was observed with either percentage of P80, indicating that there are no particularly strong lipid-protein interactions. In literature, electrostatic interactions have been observed between Hb and negatively charged lipids (PS) below physiological pH (7.4), but not between zwitterionic and neutral lipids at physiological pH (Shviro et al., 1982; Szebeni et al., 1985). Although, due to the difference in quaternary structure between tetrameric Hb and dimeric BvPgb1.2, the interactions between these proteins and lipids might differ. Moreover, the hexacoordinated heme in BvPgb1.2 would make the protein less prone to cause lipid oxidation or similar fatty acid interactions. Additionally, both BvPgb sample series were red, indicating that BvPgb was able to retain O₂ in its active ferrous (Fe²⁺) oxidation state, probably due to its increased thermal and redox stability (Christensen et al., 2022).

This contrasts with the behaviour of the Mb sample series (Figure 2, right), where, although the water channel size for the 30% P80 series was relatively constant, the size for the 25% P80 series decreased with increasing Mb concentration. It has previously been reported that fatty acids can bind both specifically and non-specifically to Mb at physiological pH (Shih et al., 2015). In addition, particularly relevant for this work, oleic acid, which has the same lipid tail as all of the lipid components used in this system, can bind strongly to Mb (Sriram et al., 2008). The heme subunit has also been observed to partition into the hydrophobic core of phospholipid bilayers, due to hydrophobic interactions (Ginsburg and Demel, 1983; Cannon et al., 1984). This could result in a strong interaction between Mb and the hydrophobic part of the bilayer. This raises the question, however, of why there is a strong trend in the water channel size with Mb concentration for the 25% P80 series, but less so in the 30% P80 series. This can be an effect of the different structures, the added P80, the oxidation state of the Mb or a combination of these factors. The samples in the 25% P80 sample series were red, indicating that the Mb is in the oxyMb form, which has been shown to have a higher fatty acid binding affinity (Götz et al., 1994) than the metMb form, most likely present in the brown 30% P80 series. In the case that the Mb fatty acid binding is significant, it is also possible that the addition of P80, with its large head group, limits the Mb interaction with and access to the hydrophobic region, thereby reducing this effect.

It should also be noted that both Mb and BvPgb have the ability to autoxidize to metMb and metPgb (i.e., from Fe⁺² to Fe⁺³). Although the reduced form has a lower pro-oxidant ability, both forms have been shown to cause oxidation and peroxidation in lipids (Sterniša et al., 2018; Tatiyaborworntham et al., 2022; Wu et al., 2022). However, due to the stabilization of the hexacoordination ferrous form in BvPgb1.2; this protein would potentially be less reactive in comparison to the pentacoordniated Mb.

In previous work with the 30% P80 lipid system, similar effects on the lipid bulk structure have been observed with other encapsulated proteins over a similar concentration range. For beta-galactosidase (238 kDa, hydrodynamic diameter = 11.8 nm), the sponge phase structure was retained over the full range studied (Erskine et al., 2003; Pereira-Rodríguez et al., 2012), but a significant decrease in water channel size was observed, whereas for aspartic protease (34 kDa, hydrodynamic diameter = 6.0 nm), a transition from sponge phase to Pn3m cubic phase was observed at an intermediate protein concentration. For both proteins, a strong interaction with the lipid bilayer, including (partial) insertion was observed, even though both proteins have mostly hydrophilic surface amino acids (Erskine et al., 2003; Pereira-Rodríguez et al., 2012), indicating that there can be strong hydrophobic interactions even for hydrophilic proteins (Valldeperas Badell, 2019).

The additional observation of gravity induced phase separation in the highest concentration sample of the 30% P80 BvPgb sample into a protein rich and protein poor region indicates an interesting aggregation behaviour of the BvPgb in the bulk phase. The concentration dependent low q upturn observed for the 30% P80 BvPgb series most likely indicates a concentration dependent protein aggregation. However, this phase separation was only observed in the highest concentration sample of the series. For sedimentation, the size of the aggregates must be dense and sufficiently large. Therefore, a much stronger upturn at low q could be observed in the SAXS data for the protein rich region than both the protein poor regions and the pre-separation in the 60 mg/mL BvPgb sample (Figure 3). However, the separation was reversible and upon changing position, the protein appeared to diffuse through the sponge phase structure to the pre-separation state. This implies that these aggregates are transient (induced by confinement), weakly bound or can exist alongside the sponge phase structure without disrupting it.

SAXS data were less well-defined for the dispersions of the sponge phase structure, where the cell-cell correlation peak becomes more diffuse, partly due to surface effects of the particle morphology (Valldeperas et al., 2016). The peak due to the bilayer form factor, which is present in the scattering pattern for both sponge phase nanoparticles and vesicles, remains well-defined in the dispersion data. As a result, the bilayer peak tends to dominate the scattering curve and the cell-cell correlation peak becomes weak, making it hard to differentiate the contribution from the sponge phase nanoparticles and any vesicles in the sample. From complementary cryo-TEM images of the 30% P80 sample series (Supplementary Figure S8), a significant vesicle population could be observed in all samples, indicating that they can contribute to the dispersion SAXS data in Figure 4. However, we cannot rule out that some of the sponge phase particles are transferred to vesicles during the blotting process for the cryo-TEM samples.

For the 25% P80 sample series for both proteins, there seems to be no direct relation between the colour and the protein concentration. For the BvPgb, this is clearly due to extensive protein aggregation and sedimentation. However, this is less clear in the case of Mb, where there were no visible aggregates in the Mb dispersions. However, the protein concentration independent upturn in scattering intensity at low q observed in the SAXS data implies that there are similarly sized aggregates in all of the samples. This reflects the limited colloidal stability of the system in the presence of 25% P80. Additionally, in the presence of protein it is possible that the aggregate size is linked to the higher pI of BvPgb and therefore aggregation can be induced even by smaller changes in pH towards the pI.

In contrast, however, there was a clear concentration dependent color change for both 30% P80 sample series. Although the concentration dependent upturn in scattering intensity at low q was present for the BvPgb series, most of the protein appears to remain in the particles as observed when SEC (Figure 6) analyzed presence of free protein in the dispersions. For the highest concentration of BvPgb, no free protein was detected, showing that the protein remained within the lipid nanoparticles. A minor peak corresponding to a slightly larger mass than the protein was observed, however, which was possibly contamination or more likely a protein-lipid complex. In contrast, the presence of free monomeric Mb was observed for the higher Mb concentration, possibly due to leakage or incomplete encapsulation.

The two coordination states of the heme iron may have caused other types of interactions of the proteins with lipid matrix. The higher autoxidation rate and tendency to appear in the inert ferric (Fe³⁺) for pentacoordinated Hbs (Silkstone et al., 2016), like in Mb, might not impact the encapsulation process. However, it could cause oxidative reactions with lipids due to the rapid oxidation and increase the interactions with the lipids. Furthermore, even though the hexacoordinated BvPgb1.2 has increased thermal and redox stability in comparison (Christensen et al., 2022), the ability to retain O₂ in its active ferrous oxidation state might have affected the encapsulation capacity and lead to the formation of aggregates. The heme-containing proteins are known to interact with lipids in vivo (Ahmed et al., 2020) and differently between saturated/unsaturated lipid components (Szebeni et al., 1985). The clear difference in encapsulating capacity could indicate the inefficiency for hexacoordinated Hbs in this particular lipid phase system.

When comparing the data for the dispersions without proteins, it can be concluded that the presence of the Tris-HCl buffer also has an effect on the lipid self-assembly. When comparing the SAXS data for the dispersions in Tris-HCl and MQ (Figures 7, 9 and Supplementary Figure S6), the dispersions in Tris-HCl consistently had slightly less intense and well-defined peaks regardless of the protein or lipid composition. This indicates that the presence of Tris-HCl affects the order in the bilayer. For dispersions prepared from the bulk phase prepared with Tris-HCl buffer and dispersed in MQ, the PDI was consistently higher than in previous measurements where the bulk phase was prepared with the same concentration of phosphate buffer (pH7) or acetate buffer (pH5.5) (Gilbert et al., 2019; Valldeperas Badell, 2019) and dispersed in MQ. It is possible that if the bulk structure is formed and equilibrated with a higher concentration of Tris-HCl, it causes some kind of osmotic shock when the bulk phase is dispersed in MQ, where the reduced Tris-HCl concentration could lead to disrupted salt bridges and cause structural changes. However, when dispersing the lipids mixture directly via sonication or preparing the bulk phase with Tris-HCl buffer and dispersing it in Tris-HCl via the shaking method (Figure 9, Supplementary Table S2), the PDI of the dispersions in Tris-HCl have a lower PDI than the dispersions in MQ regardless of lipid composition, where the Tris-HCl ions may have a stabilising effect.