DPPC (Sigma-Aldrich, Inc., St. Louis, MO, USA) and DOPC (Tokyo Chemical Industry Co., Ltd., Tokyo, Japan) were dissolved in chloroform at concentrations of 5 mg/mL. A 5 nm gold nanocolloid solution (BBI Solutions, Crumlin, UK) was sonicated for 10 min in a Branson M1800-J Ultrasonic Bath (Branson Ultrasonics, Brookfield, CT, USA). A 5 μL aliquot of the lipid solution was applied to a Quantifoil Cu #200 R2/2 grid (Quantifoil Micro Tools GmbH, Großlöbichau, Germany) and air-dried. Subsequently, 2 μL of the dispersed gold nanocolloid solution was applied, incubated for 3 min, blotted with filter paper, and air-dried at room temperature.

TEM observation was performed using a JEM-1230 (JEOL Ltd., Tokyo, Japan) operated at 100 kV with a magnification of 40,000×. The current density during imaging was maintained at 1–10 pA/cm². Images were acquired with a FastScan-F114 CCD camera (TVIPS GmbH, Gauting, Germany; 1,024 × 1,024 pixels, 14 × 14 μm² pixel size) controlled by EM-MENU software (TVIPS GmbH). Real-time TEM videos were recorded at 12 fps using OBS Studio (Open Broadcaster Software, OBS Project, https://obsproject.com) by capturing the EM-MENU display window.

Temperature control was achieved with a Peltier heating/cooling sample holder (KITANO SEIKI Co., Ltd., Tokyo, Japan) under PID regulation, maintaining temperature change rates ≤ 0.1°C/s. The holder allowed specimen temperatures between –30°C and 100°C. For video recording of DPPC dynamics, grids were first cooled from room temperature to 0°C, then heated stepwise to 98°C, followed by cooled back to 0°C. This heating–cooling cycle was repeated three times. For DOPC, specimens were first cooled from room temperature to –26°C, then heated to 60°C, followed by cooled again to –26°C. The experimental temperature ranges were selected based on reported phase transition points for DPPC and DOPC [14–18].

Video files were converted into continuous image sequences using FFmpeg (https://ffmpeg.org). Each frame was processed using ImageJ (Rasband, W.S., U.S. National Institutes of Health, Bethesda, MD, USA; https://imagej.net/ij/) for background correction, black/white inversion, and brightness and contrast adjustment to optimize particle identification.

Particle identification and mean squared displacement (MSD) calculation were performed using TrackPy (v0.3.2, https://doi.org/10.5281/zenodo.60550). Individual gold nanoparticles were detected in each frame and linked across consecutive frames using the link_df() function, which associates particle positions based on a maximum allowed displacement between frames, enabling simultaneous tracking of multiple particles while preserving unique trajectories. The MSD curves were further analyzed using custom routines implemented in IGOR Pro version 9 (Wavemetrics, Lake Oswego, OR, USA). Near the transition temperatures, nanoparticle motion increased, and some particles moved out of the field of view or exhibited reduced contrast, resulting in lower numbers of valid displacement events (n) compared with other temperature regions. The effective pixel size was calibrated based on the apparent diameter of the gold nanoparticles measured in ImageJ, yielding a pixel size of 0.392 nm.

Probability density (PD) analysis was conducted at a fixed lag time of Δt = 8 s. This lag time was selected based on a previous study in which MSD values were evaluated at one-second intervals from Δt = 1 s to Δt = 9 s, identifying Δt = 8 s as an appropriate timescale for comparing temperature-dependent differences and reported material properties [13]. Each experiment corresponds to a measurement at a single, fixed temperature. For each experiment, 121 frames were recorded at a frame rate of 12 frames per second, corresponding to a total acquisition time of 10 s. From each trajectory, 25 non-overlapping displacements corresponding to Δt = 8 s were extracted to avoid correlation between successive measurements. Displacements from all tracked particles were pooled to construct a single PD histogram for each experiment. Accordingly, the reported values (e.g., 2,784–7,171 for DPPC and 125–267 for DOPC) represent the total number of individual displacement events accumulated over all particles, rather than the number of particles.

The PD histograms of nanoparticle displacements at Δt = 8 s were plotted and fitted with single or multiple Gaussian functions using IGOR Pro to describe the distribution of nanoparticle motion. When multiple Gaussian components were required, this was interpreted as the coexistence of distinct mobility populations near the phase transition. Box plots were generated directly from the pooled displacement data (Δr), where the box boundaries represent the 25th and 75th percentiles, the central line indicates the median, and the whiskers denote the minimum and maximum values.

Samples inside the electron microscope are effectively dehydrated due to the high-vacuum environment. To compare dynamics under similar conditions, DSC measurements were performed on fully dehydrated samples. DPPC and DOPC samples were dehydrated at 80°C for 18 h under vacuum (CVE-2100, TOKYO RIKAKIKAI CO., LTD., Tokyo, Japan). Dehydrated DPPC (4.0 mg) and DOPC (12.2 mg) were sealed in 25 μL aluminum pans with lids, and DSC measurements were carried out using a DSC 3 instrument (Mettler Toledo International Inc., Greifensee, Switzerland). Thermal cycles were conducted from 0°C to 100°C for DPPC and from –80°C to 60°C for DOPC at a heating/cooling rate of 5°C/min under a nitrogen atmosphere. Each heating–cooling cycle was repeated multiple times to ensure reproducibility.

EBMD was applied to biologically relevant lipid membranes to assess its versatility beyond synthetic polymers, previously studied for PC₈FA (poly{2-(perfluorooctyl)ethyl acrylate}) and PSA (poly(stearyl acrylate)) [13]. Because TEM observations were conducted under high vacuum, the lipid membranes were effectively dehydrated. This dehydration is known to shift the main phase transition temperature (Tm) of DPPC from approximately 41°C in fully hydrated bilayers to 50–75°C or higher in dehydrated states [14, 15]. To account for this effect, DSC measurements were also performed on dehydrated samples for direct comparison with EBMD results. Lipid films of DPPC and DOPC, dissolved in chloroform, were deposited onto Quantifoil Cu #200 R2/2 grids with 2‑μm–diameter holes. 5‑nm gold nanoparticles were dispersed on the films as tracer probes. Both DPPC and DOPC formed stable membranes spanning the holes, similar to previously reported polymer systems [13]. Representative TEM images show continuous lipid membranes spanning the Quantifoil holes (Figure 1c, Figure S1). These observations confirm that the membranes indeed extend across unsupported regions rather than adhering solely to the carbon film. However, the membrane thickness and lamellarity cannot be strictly determined from conventional TEM contrast alone, and single-bilayer formation cannot be unequivocally concluded. Membrane dynamics were recorded using a Peltier-type heating/cooling TEM holder (KITANO SEIKI Co., Ltd., Tokyo, Japan), which allows precise temperature control from −30°C to 100°C. Video recordings were captured in a conventional JEOL1230 electron microscope during heating and cooling cycles, and MSD of gold nanoparticles was calculated at each temperature point to quantify membrane dynamics (Figure 1a–e). Representative raw EBMD image movies recorded at different temperatures are provided (Movies S1–S6).

DPPC contains two saturated 16-carbon chains (palmitic acid, C16:0), whereas DOPC contains two monounsaturated 18-carbon chains (oleic acid, C18:1 cis-Δ9) (Figure 1f). Despite their structural similarity, these lipids exhibit markedly different membrane properties due to the presence or absence of unsaturated acyl chains. We examined their main phase transition temperatures (melting temperatures: Tm) and crystallization temperatures (Tc) using a DSC. It is well established that the transition temperature of phospholipids largely depends on hydration state [14–18]. Because samples inside the electron microscope are effectively dehydrated under high vacuum, DSC measurements were conducted on dehydrated specimens for direct comparison. DSC demonstrated a Tm and Tc of dehydrated DPPC were 64°C and 62°C, respectively, while 3°C and −20°C for dehydrated DOPC (Figure 1g, 1h). These values confirm low hydration conditions of the specimens used in our experiments (Figure S4). In this experiment, DSC experiments were performed at a heating/cooling rate of 5°C/min under a nitrogen atmosphere. Although slower DSC scan rates may yield sharper transition features, the aim of the present DSC measurements was to identify approximate phase transition temperatures rather than to perform a detailed thermodynamic analysis. As the reproducibility of the observed transitions and their consistency with reported values (approximately 41°C in fully hydrated bilayers to 50–75°C or higher), a scan rate of 5°C/min was considered sufficient for defining the temperature window relevant to the EBMD experiments (Figure S4).

To evaluate the temperature-dependent dynamic of DPPC, MSD measurements using EBMD were repeated three times over a heating–cooling cycle (0–98°C) by maintaining temperature change rates ≤ 0.1°C/s (Figure 2, Figure S5). The normalized MSD curves for the heating-cooling cycles are shown (Figure 2). In the first heating cycle, multiple MSD fluctuations were observed between 6°C and 40°C, and a prominent MSD peak appeared around 52.5°C, corresponding to phase-transition-related fluctuations. The discrepancy in Tm observed between EBMD and DSC, which was also noted in our previous polymer analysis [13], likely arises from EBMD’s sensitivity to local, non-equilibrium molecular dynamics preceding the bulk phase transition, as well as differences in sample environment, including thin-film geometry and partial hydration. Low-temperature fluctuations between 6°C and 40°C may reflect heterogeneity in membrane domains or residual hysteresis from sample preparation.

Analyses using FRAP and NMR have reported that above the transition temperature (Tm), molecular mobility increases markedly, with diffusion constants rising by one to two orders of magnitude [19, 20]. In contrast, our TEM-based MSD analysis did not show a pronounced change, possibly due to differences in detection sensitivity or the motion tracking index, here represented by gold nanoparticle movement. It should be noted that the MSD of the gold nanoparticles does not directly correspond to the lateral diffusion coefficient of individual lipid molecules. Because each particle is coupled to multiple lipids, the measured Δr reflects the effective, collective mobility of the membrane over nanometer length scales, which is sensitive to membrane heterogeneity, viscoelasticity, and domain formation.

When comparing the results of successive heating cycles, the MSD profiles changed markedly. Low-temperature peaks disappeared in the first and second heating cycles, while a dominant peak appeared around 38°C in the second cycle and shifted to 34°C in the third cycle. These changes suggest that initially heterogeneous DPPC domains became more homogeneous multilamellar after thermal annealing, as reported in previous papers [21, 22]. However, the progressive downward shift in transition temperatures across cycles may reflect potential electron-beam damage or structural destabilization [23, 24]. Despite the annealing effect, large gaps in the MSD curves were observed between 22.5–30°C in the second cycle and 25–30°C in the third cycle, which may correspond to the “pretransition shift” of DPPC from the Lamellar tilted gel phase (Lβ′) to the Ripple gel phase (Pβ′) [25]. To evaluate the possibility that lipid films preferentially adhere to the carbon support, as reported for aqueous lipid samples [26], we examined the carbon–hole boundaries in our preparations. TEM images show that gold nanoparticles are present on both carbon and hole regions, suggesting that organic‑solvent deposition distributes lipids across both areas, although some aggregation near carbon edges is observed (Figure S1).

In the cooling cycles, MSD remained relatively high at higher temperatures but declined sharply near 54°C in the first cycle, corresponding to membrane recrystallization (Figure 2). At temperature below Tc, the MSD levels were significantly reduced compared to pre-transition values. Fluctuations diminished across repeated cooling cycles, supporting the idea that initial heterogeneity was resolved by thermal annealing. The sharp MSD decline observed at 54°C in the first cycle shifted to 50°C in the second cycle and 32°C in the third cycle, likely reflecting cumulative electron-beam damage again. Based on these observations, subsequent analyses focused primarily on the first heating–cooling cycle, where the original membrane structure was best preserved.

Molecular dynamics of the DPPC membrane near Tm and Tc were analyzed in detail (Figures 3 and 4). Just below Tm in the heating cycle 1, motion histograms showed broad distributions fitted by multiple Gaussian components (Figure 3b), indicating coexistence of distinct molecular motion modes, likely due to structural heterogeneity or fabrication-induced hysteresis. At 52.5°C, both overall mobility and histogram complexity increased, requiring four Gaussian components for fitting. After exceeding Tm, dynamics became more uniform, with a single Gaussian adequately describing the distribution, reflecting a homogeneous membrane.

During cooling, no major fluctuations were observed aside from a ~10% drop in mobility between 54°C and 52°C (Figure 4a). Histogram distributions before and after Tc indicated that recrystallization proceeded relatively uniformly, transitioning from two Gaussian components to one (Figure 4b).

To further validate the versatility of EBMD, DOPC was examined. Although DOPC shares the same headgroup as DPPC, its unsaturated acyl chains confer distinct physical properties (Figure 5). DSC measurements of dehydrated DOPC confirmed a Tm of 3°C and a Tc of −20°C (Figure 1h).

In contrast to DPPC, DOPC exhibited no distinct peaks in the MSD profile during either the heating (Figure 5a) or cooling (Figure 5b) cycles. The cis double bond–induced disorder in DOPC may disrupt tight molecular packing and prevent sharp phase transitions. During heating, membrane mobility gradually increased up to −10°C and plateaued near Tm (3°C), followed by irregular fluctuations indicative of dynamic disorder rather than discrete phase transitions (Figure 5a). Motion histograms showed minimal changes around Tm with slight variation after the transition (Figure 5c). During cooling, mobility decreased continuously, with a marked reduction below −20°C, consistent with Tc, and histogram distributions remained largely unchanged (Figure 5b, 5d).