IMod

Vitrified samples were imaged using a FEI Titan Krios microscope operated in EFTEM mode at 300 kV with a 70-μm C2 aperture and a 70-μm objective aperture. Cryo-ET data were collected using SerialEM (Mastronarde, 2005 (link)) on a K2 summit direct electron detector fitted behind an energy filter (Gatan Quantum) at a nominal magnification of 53,000× with a calibrated pixel size of 2.17 Å. The energy filter was set to remove electrons > ±10 eV from the zero loss peak energy. The K2 summit camera was operated in counting mode at a dose rate of ∼5–8 electrons/pixel/s on the camera except for the pre-illuminated images (in Figure 4A) where super-resolution mode was used. Each tilt image was dose-fractionated into three image frames, each with ∼0.5 e⁻/Ų electron dose, and aligned during the acquisition itself using the “Combined Filter” in the Gatan Digital Micrograph software. Based on Thon ring appearance, we observed that this alignment becomes unreliable for high tilt angles. Therefore, we did not use frame alignment for tilt angles higher than 45°. For the HBV capsid sample, tilt series data were collected between ±60° with 3° tilt increments. Data were collected at 3.2–5.6 μm underfocus with a cumulative dose of 60 e⁻/Ų equally fractionated over the tilt series. For DNA-origami molecules, data were collected between ±30° with 5° tilt increments. A defocus of −7 μm and a cumulative electron dose of 70 e⁻/Ų was applied, and was equally fractionated over each tilt angle in the series.
During data collection on ultrastable gold supports, the hole was illuminated symmetrically with the center of the beam coinciding with the center of the hole, and where the beam encompassed the region of the gold support adjacent to the hole. Since the gold support scatters more electrons than the specimen, it is visible at very low electron doses. Therefore, a low-dose image (0.05 e⁻/Ų) was collected and used to center the hole using image shift before commencing data collection. Finally, because gold does not show Thon rings, the extra images for defocus estimation (cf. Figure 1C) were collected by imaging adjacent specimen holes on either side of the region of interest.
Fiducial particles in tilt series data were tracked automatically using IMOD (Kremer et al., 1996 (link)) and then inspected manually for errors. Corrections of tracking errors were made manually within IMOD, and the final aligned tilt series were produced in IMOD. Only in the comparison with strip-based CTF, the aligned tilt series data were CTF corrected by phase flipping in IMOD. Tomogram reconstructions from the aligned tilt series were conducted using weighted back-projection implemented in Tomo3D (Agulleiro and Fernandez, 2011 (link)). All sub-tomogram averaging and single-particle analysis refinements were performed in RELION with icosahedral symmetry imposed for HBV capsid particles. Additional details are provided in the Approach section.

Cryo-ET data of FIB-milled cells was acquired on three different Titan Krios microscopes (Thermo Scientific), all equipped with a K2 direct electron detector and a BioQuantum energy filter (Gatan). SerialEM was used for acquisition.²⁸ (link) To identify the positions of the lamellae, low-magnification montages of the central part of the grid were acquired , with the detector operated in linear mode. Then2.3, 5.1 or 5.5 nm pixel size montages of individual lamellae were taken and used for finding interfaces between LDs. Tilt series were acquired from 0° to ±60° (maximum) using a dose-symmetric acquisition scheme,³⁴ (link) 1° increment and a pixel size of 3.7, 3.5 or 3.4 Å. The detector was operated in counting mode. Images were acquired in a tilt group size of 4. The target defocus was set to −5 μm. The dose per tilt series image was adjusted to 1–1.2 e⁻/A² and the target dose rate at the detector was kept around 4 e⁻/px/s. For a subset of the tilt series, we used exposure fractionationation into 3 frames per tilt image. Alignment of exposure fractionated frames was performed in IMOD using alignframes. Tilt series were aligned in IMOD using patch tracking.²⁶ (link)^,27 (link) Final tomograms were reconstructed at 7.5, 7.1 or 6.7 Å pixel size using SIRT reconstruction with 10 iterations. For presentation in figure panels and movies, a median filter was applied to virtual slices. The cryo-EM data presented in the manuscript for Cidec-EGFP was obtained from 30 cells expressing Cidec-EGFP, plunge-frozen on at least 8 different days, and 15 tomograms containing LD interfaces were acquired on 10 of these cells. The cryo-EM data presented for Cidec was obtained from 15 cells expressing Cidec, frozen on 3 days, and 7 tomograms containing LD interfaces were acquired on 7 of these cells. The cryo-EM data presented of cells not expressing Cidec-EGFP was obtained from 5 cells frozen on 2 different days, and one representative tomogram, showing an LD but not containing LD interfaces, is included in the manuscript. Three of the tomograms of cells expressing Cidec-EGFP were previously used in unrelated projects published before.¹⁶ (link)^,31 (link)

LD diameters used to calculate the ratios shown in Figure 3D were estimated in IMOD applying imodcurvature to a model consisting of points picked along the LD monolayer in a single virtual tomographic slice. We assumed that LDs are spherical in shape as in the analysis of FM images (see above). If the equatorial plane of the LD appeared to be included in the tomogram, a single imodcurvature readout from the corresponding virtual slice was used to estimate the LD diameter. If the LD segment contained in the tomographic volume did not include the equatorial plane, two imodcurvature radii a and b were determined on two different virtual slices spaced in z-direction by z nm. The LD radius r was then calculated using the formula: $r=\surd \left(a^2+{\left(\frac{a^2-b^2-z^2}{2\ast z}\right)}^2\right)$ . In one case of a large LD forming two interfaces with smaller LDs, the radius of the large LD could not be estimated because the fraction of the LD contained in the tomogram was too small. The two corresponding interfaces were therefore excluded from the analysis in Figure 3D.

Frames were aligned on the fly in SerialEM; CTF estimation, phase flipping and dose-weighting was performed in IMOD^{64 (link)}. Tilt-series’ were aligned in IMOD either using patch-tracking or by using nanoparticles (likely gold or platinum) on lamella surfaces as fiducial markers. Tomograms were binned 4x and filtered in IMOD or by using Bsoft^{65 (link)}.