Low-flux scanning electron diffraction reveals substructures inside the ordered membrane domain

Ordered/disordered phase separation occurring in bio-membranes has piqued researchers’ interest because these ordered domains, called lipid rafts, regulate important biological functions. The structure of the ordered domain has been examined with artificial membranes, which undergo macroscopic ordered/disordered phase separation. However, owing to technical difficulties, the local structure inside ordered domains remains unknown. In this study, we employed electron diffraction to examine the packing structure of the lipid carbon chains in the ordered domain. First, we prepared dehydrated monolayer samples using a rapid-freezing and sublimation protocol, which attenuates the shrinkage of the chain-packing lattice in the dehydration process. Then, we optimised the electron flux to minimise beam damage to the monolayer sample. Finally, we developed low-flux scanning electron diffraction and assessed the chain packing structure inside the ordered domain formed in a distearoylphosphatidylcholine/dioleoylphosphatidylcholine binary monolayer. Consequently, we discovered that the ordered domain contains multiple subdomains with different crystallographic axes. Moreover, the size of the subdomain is larger in the domain centre than that near the phase boundary. To our knowledge, this is the first study to reveal the chain packing structures inside an ordered domain.

there are several lines of evidence for the existence of subdomains 20 , the techniques used could not address lipid packing structures inside the ordered membrane domain.
Wide angle X-ray diffraction (WAXD) has been used for the investigation of lipid packing structures in artificial membranes [21][22][23][24] because the WAXD shows diffraction peaks corresponding to lattice spacings of the lipid carbon chains. However, the diameter of the X-rays frequently used for those experiments is more than 100 μm, which is much larger than the size of the ordered domain. Consequently, a WAXD pattern contains integrated information on a number of ordered and disordered domains residing within the beam size. In addition, owing to the limitation of the scattering efficiency of the X-rays, the WAXD peaks from a single membrane domain are too weak to obtain reliable results. Therefore, another method is needed to examine the local structures inside a single ordered domain.
Instead of WAXD, electron diffraction (ED) is a promising method for local-structure analysis due to the following reasons: First, because ED is performed with an electron microscope, we can direct the electron beam to a micro-size target under visual observation. Second, using a selector aperture, we can select the ED patterns only from the desired regions, whose size is less than 1 µm in diameter. Additionally, due to its significantly greater diffraction potency than X-rays, ED likely provides strong diffraction peaks even from a single membrane domain. While ED experiments were sometimes used for structural analysis of lipid membranes from the 1970s to the 1990s 25−32 , the techniques progression has been hampered by some technical difficulties. For example, ED requires a dry and fixed sample because it is performed under a high vacuum. In addition, compared with X-rays, the electron beam causes significant damage to the lipid membranes 33,34 .
In the present study, we prepare dehydrated samples employing a rapid-freezing and sublimation (RFS) protocol, hampering the shrinkage of the chain-packing lattice in the dehydration process. In addition, we optimised electron flux to minimise beam damage to the lipid membranes. Finally, we revealed the local structures inside a single ordered domain by newly developed low-flux scanning electron diffraction (LFSED).

Results
Influence of the rapid-freezing and sublimation (RFS) on the chain packing in lipid monolayer. It has previously been reported that dehydration of lipid membranes causes shrinkage of the chain packing lattice effectively in the disordered membranes 25 . To overcome this problem, we employed the RFS method to prepare dehydrated samples (see "Materials and methods" for details). Because lipid diffusion is fixed through the RFS protocol, it should create dehydrated samples without the shrinkage of the chain-packing lattice.
This could be confirmed by fluorescent energy transfer (FRET) measurements. Since FRET sensitively depends on the distance between the FRET donor and acceptor in membranes, shrinkage of the chain-packing lattice causes effective FRET-quenching of the donor fluorescence. Figure 1 shows normalised donor intensity I donor in DOPC supported monolayers, which form a homogeneous disordered membrane on the air-water interface 35 . The dehydration of the sample at atmospheric pressure and room temperature caused effective FRET-quenching of the donor intensity (compare Fig. 1a and c), indicating that the packing shrinkage occurs in the dehydration process. On the other hand, the I donor -value of the RFS sample is consistent with that of the wet sample within the error range (compare Fig. 1a and b). While the RFS method has been widely used for electron microscopy and mass-based imaging 36−38 , we demonstrated for the first time that it does not cause significant shrinkage of the chain packing lattice in the disordered monolayer.
Kinetics of structural decay caused by electron beam irradiation. We optimised the flux of the electron beam to minimise beam damage to the sample. Here, we used a distearoylphosphatidylcholine (DSPC)  www.nature.com/scientificreports/ monolayer, which forms an ordered phase at air-water interface 39 . Because the DSPC monolayer provides a clear reflection, it is convenient for optimisation. We set the wavelength and image acquisition time to be 0.0037 nm (acceleration voltage V acc = 100 keV) and five s, respectively. The sample was kept at − 180 °C with a cooling holder and the beam damage was reduced. Figure 2 shows the diffraction patterns from a DSPC monolayer at different electron fluxes. The irradiation at 30.0 e/nm 2 ·s gave rise to a broad ring pattern (indicated by arrowheads in Fig. 2e), indicating immediate disruption of the chain-packing structure upon such strong irradiation. Some strong spots appearing in the wide angular regions are unlikely to have originated from the carbon chains because their peak positions (> 3.0 nm −1 ; Supplementary Fig. S1) are far from the lattice spacing of the chain packing. Next, we reduced the electron flux to less than 15.9 e/nm 2 ·s and successfully obtained a sharp Debye-Scherrer diffraction at 2.40 ± 0.01 nm −1 (indicated by arrowheads in Fig. 2a-d and their one-dimensional profiles are shown in the bottom of Fig. 3a-d). By applying the lattice spacing d = 0.42 nm (= 1/2.40 nm) to Eq. (2) that is described in "Materials and methods", we calculated the lateral molecular area of the DSPC to be 0.41 nm 2 . This value is close to the molecular area of the DSPC on the water subphase obtained in the π-A isotherm measurement (0.45 nm 2 at π = 30 mN/m; a black isotherm in Fig. 4a). Also, this result is likely reasonable because it was previously reported that that the chainpacking structure of an ordered membrane (the gel phase) is hardly affected by dehydration 25 . Figure 3a-d shows one-dimensional ED profiles upon continuous irradiation with electron beams, demonstrating that the peak intensity decreases as the irradiation duration t irr increases. To quantitatively analyse the structural decay, we plotted the normalised intensity I norm as a function of t irr (Fig. 3e). Fitting these data to a linear function (dashed lines in Fig. 3e) revealed that the peak intensity decreases in the initial five seconds by 0.4 ± 5%, 2.8 ± 2%, 5.8 ± 4%, and 15.0 ± 7% at electron fluxes of 0.9 e/nm 2 ·s, 2.3 e/nm 2 ·s, 4.7 e/nm 2 ·s, and 15.9 e/ nm 2 ·s, respectively. Considering their errors, we concluded that the structural decay is sufficiently small to be considered negligible upon irradiation with an electron flux of less than 4.7 e/nm 2 ·s.
Moreover, we examined the effect of the wavelength λ e on the structural decay by comparing the red plots in Fig. 3e (λ e = 0.0037 nm; V acc = 100 keV) and in Supplementary Fig. S2 (λ e = 0.0021 nm; V acc = 200 keV). Consequently, the decrease in the I norm -value became moderate at the shorter wavelength. However, such shorter wavelength caused a large deviation in the time curve of the I norm -value ( Supplementary Fig. S2). Probably, our TEM could not supply stable electron beam under the high acceleration voltage (V acc = 200 keV) and, thus, the structural decay of the sample did not show a linear correlation with t irr . Taking account of these results, we used a beam flux of less than 4.7 e/nm 2 ·s and a wavelength of 0.0037 nm (V acc = 100 keV) in the following ED experiments.
Macroscopic phase separation in DSPC/DOPC binary monolayers. While there are some previous studies using DSPC/DOPC binary monolayers 39,40 , their phase behaviour remains unknown. Here, we investigated the composition-dependent phase behaviour of the DSPC/DOPC monolayers by surface pressure-molecu- www.nature.com/scientificreports/ lar area isotherm (π-A isotherm) measurements. Figure 4a shows the π-A isotherms of monolayers consisting of DSPC, DOPC, and their mixtures. The surface pressure of the pure DSPC and pure DOPC monolayers increased steadily as the monolayers were laterally compressed (black and pink isotherms, respectively, in Fig. 4a). On the other hand, DSPC/DOPC monolayers (0.2 ≤ x DSPC ≤ 0.7) show a stepwise increase in surface pressure. This result indicates that the DSPC/DOPC monolayers undergo macroscopic phase separation; namely, the DOPC-rich disordered domains collapse at lower surface pressure (indicated by an arrow in Fig. 4a) and the DSPC-rich www.nature.com/scientificreports/ ordered domains at higher surface pressure (π > 40 mN/m) 5 . Here, the disordered domain is formed by almost pure DOPC because the collapse pressure of the disordered domain (π ~ 40 mN/m) is consistent with that of the pure DOPC monolayer across the experimental compositional range (crosses in Fig. 4b). On the other hand, in the region of 0.2 ≤ x DSPC < 0.4, the collapse pressure of the ordered domain is significantly smaller than that of pure DSPC (circles in Fig. 4b), indicating that the ordered domain does not consist of pure DSPC, but contains small amounts of DOPC. In the following experiments, we analyse the lipid packing structure of the ordered domain formed in the DSPC/DOPC (x DSPC = 0.3) monolayer because the structure of the pure DSPC monolayer has already been obtained in Figs. 2 and 3.
Chain packing structures inside a single ordered membrane domain. Figure 5a shows a fluorescent micrograph of the DSPC/DOPC monolayer, which is formed on the collodion-coated TEM grid. Since this sample contains 0.2 mol% Texas Red-DPPE, a disordered phase marker 41 , the darker and brighter regions correspond to the DSPC-rich ordered and DOPC-rich disordered domains, respectively. In the following experiments, we determined the beam positions, referring to the domain distribution shown in the fluorescent micrographs (see "Materials and methods" and Supplementary Fig. S3 for details). Figure 5b shows the ED patterns from the 0.79 μm 2 region inside the ordered domain, giving rise to sharp hexagonal spots. To gain further insight into the chain-packing structure, we represented the one-dimensional intensity of the hexagonal spots and found that the ordered domain gives a single sharp peak at 2.40 nm −1 (Fig. 5d). These results revealed that the carbon chains aligned perpendicular to the monolayer surface with a lattice spacing of 0.42 (= 1/2.40) nm. On the other hand, the disordered domain, which consists of almost pure DOPC, gave a broad Debye-Scherrer pattern centred at 2.26 nm −1 (Fig. 5c and a red profile in Fig. 5e). This ED pattern is similar to the WAXD pattern of the disordered DOPC bilayers 42 ; DOPC bilayers also give a broad WAXD peak and the peak position is roughly estimated to be 2.23 nm −1 . This fact indicates that the RFS protocol does not cause significant shrinkage of the chain-packing lattice in the disordered monolayer, being in line with FRET experiments (Fig. 1). www.nature.com/scientificreports/ Finally, taking advantage of the small size of the selected areas in the ED experiments, we compared the chain packing structure at seven different regions across a single ordered domain by low-flux scanning electron diffraction (LFSED). In the LFSED experiment, we reduced the electron flux and exposure time to be 0.8 e/ nm 2 ·s and 0.5 s, respectively, for further suppression of the beam damage. Then, the size of each region was set to 6.2 μm 2 . As in Fig. 5b, sharp hexagonal spots were obtained in the ordered domain (regions 3-6 in Fig. 6b,c), while a broad Debye-Scherrer pattern was observed in the disordered domain (regions 1, 2, and 7 in Fig. 6b,c). Moreover, the one-dimensional ED pattern showed that the peak positions (2.39 ± 0.02 nm −1 ) are almost the same between regions 3-6 ( Supplementary Fig. S4), indicating a similar lattice spacing of the chain packing in the ordered domain. To analyse substructures formed in the ordered domains, we plotted the peak intensity at 2.39 nm −1 along the azimuthal direction θ. Consequently, it was found that the centre regions of the ordered domain (regions 4 and 5) show diffraction peaks at almost the same angles (dashed line in Fig. 6d). On the other hand, the regions 3 and 6 showed diffraction peaks at slightly different angles from the centre regions: + 12° and − 14°, respectively, (arrowheads in Fig. 6d). These results suggest that a single ordered domain contains multiple subdomains with different crystallographic axes. Here, the regions 4 and 5 belong likely to a same subdomain because it is quite unlikely that two different subdomains are incidentally oriented in the same direction. In such a case, the size of the subdomain is larger at the domain centre than in the vicinity of the phase boundary (Fig. 7).

Discussion
In the present study, we examined the packing structure of lipid carbon chains inside a single ordered domain using the LFSED and identified that the ordered domain consists of multiple subdomains with different crystallographic axes (Fig. 6d). Moreover, results indicated that the size of the subdomains is larger in the domain centre than in the vicinity of the phase boundary (Figs. 6 and 7). Because the edge of the ordered domain is exposed to show the one-dimensional ED profiles of (b) and (c), respectively. Although we subtracted the background using a monolayer-free TEM-grid (see "Materials and methods" for details), the diffraction peak from the collodion film could not be completely removed in the panel (e). This is because the thickness of the collodion film is not homogeneous on the TEM-grid. Thus, we deconvoluted those peaks using the two Lorentz functions; the red and blue profiles show the diffraction peaks corresponding to the carbon chains and the collodion film, respectively. www.nature.com/scientificreports/ a disordered matrix, the packing structure near the phase boundary should be directly affected by inter-domain interactions, such as electrostatic interaction 43 , line tension 44 , and transverse shear when the domains are in motion. It is probable that these interactions hamper the unidirectional arrangement of the chain packing, giving rise to the formation of small subdomains in the vicinity of the phase boundary. www.nature.com/scientificreports/ Some previous studies compared the size of the subdomains based on the halfwidth of the WAXD peak 45,46 . However, these studies also suggested that the width of the WAXD peak depends not only on the size of subdomains but also on the other factors such as the order of the chain packing, the instrumental setup, and so on. In that case, the deconvolution of these components is necessary for estimating the size of the subdomain 47 . On the other hand, the LFSED can assess the size of the subdomain without the deconvolution analysis and, thus, this point is remarkable merit for the LFSED when it is compared with WAXD.
Another interesting finding is that the lattice spacing of the carbon chains is almost the same between the pure DSPC monolayer (2.40 ± 0.01 nm −1 ; bottom profiles in Fig. 3a-d) and the DSPC-rich ordered domain (2.39 ± 0.02 nm −1 ; Supplementary Fig. S4), although the π-A isotherm measurements showed that the DSPC-rich macro-domain contains small amounts of DOPC (see Fig. 4b and Text). We speculated that, in the DSPC-rich domain, the subdomains are formed by pure DSPC, while small amounts of DOPC molecules reside in the intersubdomain space (Fig. 7). In this model, since DOPC molecules do not perturb the chain packing of DSPC, the lattice spacing of the DSPC-rich domain should be consistent with that of the pure DSPC monolayer.
Moreover, the pure DSPC monolayer showed a Debye-Scherrer pattern, while the DSPC-rich ordered domain gave the hexagonal spots (compare Figs. 2 and 5b). This means that the size of the subdomain is larger in the DSPC-rich domain than in the pure DSPC monolayer. It is not surprising that even in the pure monolayer, incidentally-created membrane defects often chop off the long-range alignment of carbon chains 48 , leading to the formation of subdomains. Here, the question arises as to why the size of the subdomains is different between the pure DSPC monolayer and the DSPC-rich ordered domain. In the pure system, the van der Waals interaction between the DSPC molecules could be the dominant force for DSPC assembly. On the other hand, in the DSPC-rich domain, molecular contact of the unfavourable pair, DSPC and DOPC, should cause energetic disadvantage 49 . We speculate that, besides the van der Waals interaction, avoidance of unfavourable pairing leads to the formation of a larger subdomain in the DSPC-rich domains. However, to confirm this speculation, we need further study on the correlation between the inter-lipid interactions and the size distribution of the subdomains.
It has been suggested that the substructures inside ordered membrane domains are involved in important biological functions. For example, Wu et al. demonstrated that molecular dynamics in the ordered domains are strongly affected by the substructure and, thus, proposed that the subdomains are responsible for determining the protein partitioning into the ordered domains 17 . In addition, our group reported that a minor structural change of a lipid molecule alters the size of the subdomains, affecting water permeability of the membranes 46 . However, these previous studies could not address the chain packing structure of each subdomain nor the determination of the subdomain size. Therefore, LFSED is promising methodology to understand mechanistic link between the substructure of the ordered domains and their biological functions.

Conclusion
Electron diffraction (ED) is promising methodology for examining local structures of the lipid membranes due to following reasons. Firstly, ED can assess local structures of lipid membranes. Secondly, due to great diffraction potency of the electron beam, ED can produce strong diffraction peaks even from a single membrane domain. Lastly, ED is easily performed with a usual transmission electron microscope. Compared with WAXD, these points are striking advantages of ED. However, some technical difficulties hamper application of the ED to lipid membranes. For example, the ED experiment requires a dry and fixed sample. In addition, the electron beam causes significant damage to the lipid membranes.
In the present study, we first prepared dehydrated samples using the RFS method and demonstrated that it provides dehydrated monolayer samples without significant shrinkage of the chain-packing lattice in the monolayer samples. We next optimised the intensity of the electron beam to minimise the beam damage to the membrane sample. As a result, we found that the beam damage is sufficiently small to be considered negligible www.nature.com/scientificreports/ under the electron flux and irradiation durations of less than 4.7 e/nm 2 ·s and 5 s, respectively. Finally, we examined the chain packing structure inside the ordered domain, which is formed in the DSPC/DOPC (x DSPC = 0.3) phase separated monolayer. Consequently, our LFSED measurement indicated that the ordered domain consists of multiple subdomains with different crystallographic axes (Fig. 6d). Moreover, it is indicated that the size of the subdomain is larger in the domain centre than in the vicinity of the phase boundary (Fig. 7). To the best of our knowledge, this is the first example to address the chain packing structures inside a single ordered domain. In the preparation of the supported monolayer, the lipid solution in the presence of 0.2 mol% Texas Red-DPPE was spread on the water subphase. Then, the substrate such as a mica plate and a collodion coated TEM-grid (Okenshoji Co., Ltd, Tokyo, Japan) was horizontally dipped into the water subphase, and the lipid sample was compressed to 30 mN/m at the rate of 20 mm 2 /s. After the compression, the monolayer was transferred onto the substrate. Then, distribution of the ordered and disordered phases in the supported monolayers were observed with a fluorescent microscope BZ-X700 (Keyence, Osaka, Japan).

Rapid freeze and sublimation (RFS) methods.
The dehydrated sample for the electron diffraction experiment was prepared by the RFS method, which is derived from the methods employed in freeze-fracture electron microscopy and mass-based imaging 36−38 . Briefly, immediately after the preparation of the supported monolayer, the wet sample was rapidly frozen in slush N 2 . Then, the sample was stored in a vacuum chamber maintained at 6.7 × 10 −2 Pa, which is much lower than the triple point of water in the pressure vs. temperature phase diagram. The temperature was gradually increased from -180 to 25 °C to completely sublimate water molecules from the sample.

Fluorescent energy transfer (FRET) experiments.
Mica supported DOPC monolayers were prepared by aforementioned protocol. In the FRET measurements, the samples contained 0.2 mol% Bodipy-PC and 0.4 mol% Texas Red-DPPE as FRET donor and acceptor, respectively. Then, the samples were dehydrated by the RFS methods or dehydrated at the atmospheric pressure and the room temperature. Then, the intensity of the FRET-donor I donor was measured at more than thirty different locations (50 × 50 μm 2 for each) in the sample with fluorescent microscope BZ-X700 and the average value was shown. The I donor -value of the wet sample was measured immediately after the preparation of the supported monolayer. The excitation wavelength of 470 nm was applied to Bodipy-PC and the emission were detected at 525 nm using dichroic mirrors OP-87763 (Keyence, Osaka, Japan).

Electron diffraction (ED) experiments.
The local-structure analysis inside a single membrane domain was conducted with transmission electron microscope JEM1400 (JEOL, Tokyo, Japan), and other ED experiments were conducted with JEM2100HCKM (JEOL Ltd., Tokyo, Japan). The supported monolayer mounted on a cooling folder, such as Gatan 626 and Gatan 636 for JEM1400 and JEM2100HCKM, respectively, was set into an electron microscope. The sample was kept at − 180 °C, and the electron beam was irradiated perpendicular to the supported monolayer surface. The instrumental camera length was calibrated using gold particles. Electron diffraction experiments were performed at an accelerating voltage of 100 kV unless otherwise mentioned. The other conditions such as electron flux and exposure time were described in the text. Diffraction patterns were acquired using a CCD camera ES500W (Gatan Inc., Pleasanton, CA) and TemCam-F416 (TVIPS, Gauting, Germany) for JEM1400 and JEM2100HCKM, respectively. To improve visibility of the diffraction patterns, we showed black-and-white inverse images. The value of the electron flux was estimated from the brightness of the digitised image of the incident beam in the absence of a lipid sample (blank image) because the electron flux was too low to be detected by the equipped current meter 35 . Specifically, in the LFSED measurements, the electron flux was calculated by interpolating the brightness of the blank image at 0 pA and 1 pA. Before all experiments, we took a warming-up duration for more than two hours to supply stable electron flux. Referring to the fluorescent micrograph, we applied an electron beam to the target together with acquisition of the diffraction pattern. After the experiment, we determined the beam position using the debris and incidental breaks on the collodion film as landmarks (Supplementary Fig. S3). The diffraction patterns were linearised using Fit 2D and (1) A lipid = 2 × A chain cos θ . (2)