Zooming into lipid droplet biology through the lens of electron microscopy

Electron microscopy (EM), in its various flavors, has significantly contributed to our understanding of lipid droplets (LD) as central organelles in cellular metabolism. For example, EM has illuminated that LDs, in contrast to all other cellular organelles, are uniquely enclosed by a single phospholipid monolayer, revealed the architecture of LD contact sites with different organelles, and provided near‐atomic resolution maps of key enzymes that regulate neutral lipid biosynthesis and LD biogenesis. In this review, we first provide a brief history of pivotal findings in LD biology unveiled through the lens of an electron microscope. We describe the main EM techniques used in the context of LD research and discuss their current capabilities and limitations, thereby providing a foundation for utilizing suitable EM methodology to address LD‐related questions with sufficient level of structural preservation, detail, and resolution. Finally, we highlight examples where EM has recently been and is expected to be instrumental in expanding the frontiers of LD biology.


Historic perspective of EM imaging in LD research
Since the first electron microscopes became available in the 1940s to researchers in the Life Sciences [1], cells of various origins have been imaged.The first micrographs of cultured cells already pictured LDs, albeit at resolutions comparable with modern light microscopy (Fig. 1A) [2].Based on such early EM observations and despite important improvements in EM sample preparation that allow for better lipid preservation over the decades that followed (Fig. 1B) [3,4], cellular LDs were long considered to be passive lipid inclusions, described to have 'no bounding membrane and appear to be held together by their hydrophobic interaction with the aqueous environment' [5].In fact, their lack of a delimiting membrane was considered a 'distinguishing feature' [5].However, in the early 1990s, LD-specific proteins, perilipins, were discovered [6].EM revealed that LD-targeted proteins localize to the LD surface (Fig. 1C,D), which exhibits a single, electron opaque line.Freeze-fracture EM suggested an occasional membrane continuity between the endoplasmic reticulum (ER) and LDs [7].
These observations lead to the notion that the surface layer of LDs may be a 'specialized area of the endoplasmic reticulum membrane leaflet' [6] or a 'novel membrane domain' [8].The ultimate proof that LDs are truly surrounded by a monolayer of phospholipids (as opposed to bilayers bounding all other cellular organelles) was provided in 2002 by cryogenic-EM images of isolated LDs (Fig. 1E) [9].Interestingly, plant cytologists had already proposed in 1972 that LDs or 'intracellular oil-containing particles', called 'spherosomes' in peanuts, show the presence of an 'atypical, single-line [. ..] biological membranes that correspond to half unit-membranes [. ..] whose polar surfaces face the hyaloplasm and whose lipoidal nonpolar surfaces contact internal storage lipid' [10].These fundamental discoveries enabled by EM were instrumental in shaping the definition and our current understanding of oil bodies, oil droplets, spherosomes, oleosomes, and adiposomes, unified today under the common name of LDs [11].

Overview of EM methods in LD research
Versatility of EM imaging techniques EM offers high resolution, down to the Angstrom range, owing to the extremely short wavelength of electrons, but at the cost of having to image the biological object in a high vacuum chamber (to prevent scattering of the electrons by air molecules).Most traditional EM methods therefore deal with dehydrated specimens, typically treated with chemical fixatives, which are meant to retain cellular ultrastructure but can also cause severe alterations.EM also comes in different flavors.Depending on the biological question at hand, the EM imaging method should be carefully selected based on considerations that include: (a) the required size of the imaging area, (b) the desired resolution, (c) the level of preservation of cellular features, (d) the compromise between 2D high throughput imaging and comprehensive 3D visualization, and (e) the available equipment and expertise.Specifically, in relation to the study of LDs by EM, the preservation of lipids in the sample further requires special considerations [12], which we discuss in more detail in Section "EM sample preparation specifics for LDs".We provide an overview of EM imaging and related preparation methods in the context of LD research, their current capabilities, and their limitations in Table 1.
As a general rule, room temperature (RT) EM procedures start with live, hydrated specimens and end with water-free, heavy-metal-stained, and resin-embedded material [5].To gain a general understanding of LD sizes, numbers, and the immediate cellular context of surrounding organelles, conventional transmission EM (TEM) imaging of thin sections (typically of 70 nm thickness generated in an ultramicrotome by mechanical sectioning with a diamond knife) represents a good choice [13][14][15][16].However, being 2D projections, they lack 3D volumetric information.For high-resolution 3D imaging of LDs, TEM tomography of thick sections (200-300 nm) is more suitable, albeit with the data acquisition and processing being more time-consuming (Fig. 1B,F) [17][18][19].This technique can also be applied to consecutive serial sections (serial section TEM), which significantly increases the contextual information and volume covered, but further reduces throughput when considering the need to investigate multiple samples across different conditions [17,18,20,21,22,23].The resolution in these TEM techniques is in the order of a few nanometers, whereas the resolution in the z direction (parallel to the beam direction) in the 3D reconstructed volumes from tomography is further distorted by the incomplete angular sampling, resulting in an anisotropic 3D representation of the specimen [24].The signal-to-noise in such data is dictated by the thickness of the section, but ultimately, the apparent nanometer scale resolution is a result of the limited structural preservation in traditional RT preparations (detailed below) and the fact that the signal comes from heavy metal staining of the cellular structures rather than from the biomolecules themselves.
For the 3D analysis of entire cells or tissue subregions, a number of recently developed methods allow for automatic sectioning and scanning EM (SEM) of large volumes, ranging from 100s to 100 000s lm 3 [25,26].In array tomography, ultra-thin sections are obtained using an ultramicrotome and deposited manually on a slide or automatically on tape using ATUM (Automatic Tape-collecting Ultra-Microtome) to be imaged by SEM [27].In serial block face SEM (SBEM) [28][29][30][31] and focused ion beam SEM (FIB-SEM) (Fig. 1G) [32][33][34][35], the sample is sequentially imaged by SEM followed by repeated material removal by an ultramicrotome knife or a FIB column integrated inside the SEM chamber, respectively.Currently, available knife-based approaches are limited in z resolution, dictated by slicing thickness to ~30 nm, whereas ion beam ablation can deplete material with 5 nm precision [24].Furthermore, it is important to note that while the theoretical resolution in a TEM is in the Angstrom range, the SEM image resolution falls in the range of a few nanometers.Therefore, the maximal resolution of 5 nm achievable with these volume EM techniques precludes fine ultrastructural information on, e.g. the details of direct LD-organelle interactions, but parameters such as LD number, cellular localization, and clustering can be precisely analyzed in relatively large volumes (Fig. 1G).
Another flavor of EM that has been successfully used for visualizing LDs is freeze-fracture EM, wherein a frozen (multi) cellular specimen is mechanically fractured to expose the cell interior.Here, fractures in frozen biological specimens commonly and spontaneously occur along the membrane plane.After further processing, the specimen can be imaged in TEM (by generating a metal replica of the fracture surface), SEM (following sublimation to achieve dehydration), or directly in a SEM equipped with a cryogenic stage (cryo-SEM).In combination with immunogold labeling of specific proteins on the metal replica (Fig. 1C), this technique provides detailed insight on LD membrane topology and protein localization [7,[36][37][38][39].
While RT EM techniques provide diverse means to investigate LDs, emerging cryogenic EM techniques allow the visualization of frozen-hydrated biological material preserved in a near-native state.Cryo-FIB-SEM volume imaging is similar to its RT counterpart in terms of its ability to image multicellular specimens in 3D, but typically suffers from somewhat poor contrast and resolution in comparison to its RT parallels that employ heavy-metal staining combined with backscattered electron detection [40].In cryo-FIB-SEM, charging of the nonconductive frozen-hydrated sample, which is also particularly affected by the neutral lipid-rich LDs, significantly decreases imaging quality (Fig. 2A).Nevertheless, cellular LDs are typically well distinguishable, suggesting that cryo-FIB-SEM has the potential to become a useful technique for their study in complex scenarios and tissues [41].

EM methods Notes
Conventional plastic embedded RT EM: requires fixation, dehydration, staining and resin embedding 2D TEM [13,14,16,65,87] Thin sections; High throughput; XY resolution of few nm TEM tomography [17][18][19] Thick sections; 3D information; Laborious data acquisition and analysis; XYZ of few nm Serial section TEM and tomography [17,18,20,21,22,23] Good resolution in XYZ for larger volumes; Laborious data acquisition and analysis Pre-embedding immunolabeling EM [7,30,149] Good localization precision of proteins of interest; Requires good antibodies On-section CLEM [104] Direct correlation of fluorescence and TEM on the embedded material; Good localization precision of proteins of interest; Requires genetic engineering when using fluorescently tagged target Genetically encoded tags (e.g.Apex, Apex2) [84,101,102] Good localization in TEM; Requires genetic engineering and is not readily suitable for low abundance targets Alternative RT 2D EM sample preparation HPF and Freeze Substitution [61,66,98] Often better structural preservation of LDs in comparison to chemical fixation at room temperature, but some protocols result in low membrane contrast Freeze-fracture EM [7,37,39] Especially suited for questions related to the fine structure of membranes; Combinable with immunolabeling in replica preparations Tokuyasu method [8,97,150] Good membrane visibility and antibody labeling sensitivity RT volumetric EM: 3D Imaging of large volumes (cells to tissues), a compromise between acquisition time, area, and level of detail.LD numbers, sizes, localization, and juxtaposition with other organelles can be analyzed for large volumes, but with more limited detail/ resolution compared to TEM Array Tomography [27] Useful for large volumes (e.g.multiple cells); XY resolution of 5-20 nm; Z resolution dictated by the section thickness (~30-70 nm) Serial Block Face SEM [30] Useful for large volumes (e.g.multiple cells); XY resolution of 10-20 nm; Z resolution dictated by the thickness of the sectioned area (~30 nm) RT Focused Ion Beam SEM [32] Typically used for volumes of one cell, isotropic pixel size XYZ with a maximum resolution of ~5 9 5 9 5 nm Cryo-EM: optimal structural preservation of hydrated samples, close to native state.Fine details, such as LD core structural features and molecular tethers, can be directly observed In situ cryo-ET [46,47,76] High resolution in XY in the nm range; Currently low throughput due to non-routine multistep sample preparation Cryo-FIB-SEM [40,151,152] Imaging still requires improvement due to charging-related artifacts and a low signal-to-noise ratio; 3D reconstruction of large volumes (whole single or multiple cells) is possible Cryo-FIB has also been refined over the last decade as a minimally perturbing preparation method for single-cell cultures or suspensions for cellular cryogenic transmission electron microscopy and tomography (cryo-EM/ET).Cells frozen on EM grids are thinned using cryo-FIB-assisted micromachining at a shallow angle to generate cellular sections of 200-300 nm thickness, called lamellae, that remain attached to the biological material on the grid, and the entire grid is then transferred to the TEM for cryo-EM/ET [42][43][44].This technique preserves LDs in their native milieu in a potentially unaltered state, allowing imaging at sub-nm resolution (Fig. 2B,C) [45][46][47].Cellular cryo-ET is currently relatively low-throughput and requires expensive equipment not commonly available in EM laboratories.
However, recent advances in instrumentation [48,49], automation of sample thinning and data acquisition [41,50,51], as well as computational analysis [52,53], have begun to broaden the scope and breadth of applications of cellular cryo-ET [54].LDs are, on the one hand, easily distinguishable in cryo-TEM and cryo-FIB-SEM imaging of unstained specimens.On the other hand, increased LD abundance (e.g. in cells with high LD levels, such as adipocytes, or after lipid loading) introduces challenges during lamella preparation; the high density of material in LDs, compared to the surrounding cytoplasm, significantly extends the FIB ablation time and can lead to 'curtaining artifacts', which negatively influence lamella quality and the subsequent cryo-TEM data quality [41].

EM sample preparation specifics for LDs
Biological specimens aimed at RT EM must undergo procedures consisting of several major steps: (a) fixation, (b) staining, (c) dehydration, and (d) embedding.Each of these steps can be optimized to improve LD visualization.
Fixation immobilizes and arrests all cellular processes, making cellular structures rigid and stable for the following steps.Fixation can be chemical or physical (i.e.cryopreservation; detailed below).For chemical fixation, paraformaldehyde (PFA), glutaraldehyde (GA), or osmium tetroxide (Os) are most commonly used.PFA and GA cross-link mainly proteins and nucleic acids, while lipids and membranes are stabilized to a lesser extent [5].To prevent the washing away of unfixed LDs by organic solvent during the following dehydration steps, the use of Os is crucial.Os reacts primarily with lipids and their unsaturated hydrocarbon bonds, and protects LDs from extraction during dehydration.It has been shown that the use of Os in multiple steps, first together with GA at the initial fixation and later alternately with thiocarbohydrazide [3,55] at the staining stage, significantly protects LDs from extraction.Moreover, the addition of malachite green to the primary fixative enhances LD staining [56,57].
Fixation is typically followed by the dehydration step, wherein all the water in the sample is gradually substituted with an organic solvent.This step is critical, and LDs, especially their hydrophobic cores, are often dissolved and washed away by commonly used solvents like acetone and ethanol [58].To minimize this unwanted effect, it is vital to optimize the preceding fixation step as described above, but it is important to note that acetone is a more powerful solvent for lipids than ethanol.Thus, the use of ethanol in LD EM studies is recommended over acetone for dehydration.At the embedding stage, omitting the transitional solvent, typically propylene oxide, is beneficial.No significant benefits from specific resins have been reported for LD preservation.Before imaging, sections may be additionally stained with lead citrate and uranyl acetate to enhance LD visibility.
RT fixation often induces LD deformation (Fig. 1F).This can be circumvented using cryopreservation or vitrification (Table 1).Here, molecules are immobilized by transferring the specimen to cryogenic temperature (below À143 °C) rapidly enough to avoid the formation of ice crystals [59], which would be detrimental to fine cellular structure.Cells up to 10 lm thickness can be seeded or deposited on EM grids and plunge-frozen into a cryogenic liquid, such as liquid ethane.Thicker cells, small multicellular organisms, and up to 200 lm-thick tissue sections can be frozen by high-pressure freezing (HPF), where the sample is cooled while being pressurized to 2100 bar.Cryopreservation, followed by the slow introduction of the fixative, stains and dehydration at sub-zero temperatures (À90 to À20) (freeze substitution) where thermal motion is minimal [60], substantially mitigates large-scale structural distortions and artifacts like LD and membrane deformations observed in RT fixation (see Fig. 1B vs. Fig.1F) [17,61].Importantly, both HPF-freeze substitution and RT preparations very often result in significant removal of the LD core lipids, the extent of which is a result of a multifactor combination of the type of biological material, form, and timing of the fixatives, stains, and solvents introduced at specific temperatures.The surface phospholipid monolayer, with the embedded proteins, may remain intact and can be used as a labeling target with specific antibodies.Thus, alternatively, cryopreserved samples can be imaged directly in their frozenhydrated state using cryo-FIB-SEM and in situ cryo-EM/ET, as described in the previous section, which allow for more pristine sample preservation compared to RT approaches (Fig. 2A-C).These preparations, however, preclude the use of specific immunolabeling.In summary, due to their high lipid content, LDs are exceptionally sensitive organelles to RT EM sample preparation, and special care must be taken to preserve their presence and structure.

Correlative EM methods for LD research
Since most EM methods only allow detailed imaging of a small fraction of the entire specimen, it is often challenging to capture specific biological events such as LD-organelle interactions, or even LDs altogether if LD abundance is low [46].To increase the success rate in specific LD-targeting and visualization by EM, correlative light and EM (CLEM) approaches are instrumental.In a typical CLEM experiment, specimens are imaged using fluorescence microscopy after fixation, and this information is used for site-specific targeting (pre-embedding CLEM).Correlation can also be performed after embedding, but it requires sample processing that takes into account minimal heavy metal staining to preserve the fluorescence signal that is used to guide the subsequent EM acquisition [62][63][64].For example, cells may be grown on specialized finder EM grids or sapphire discs with engraved markings that can be located in both light microscopy and EM, allowing the electron microscope to target the cell-of-interest.Fluorescence imaging can be performed on the embedded resin block or at the level of thin sections (on-section CLEM).In such workflows, commonly available neutral lipid stains such as Bodipy, fluorescently labeled lipids, or expression of fluorescently tagged LD-resident proteins can be used to guide the selection of specific targets [65].It is of note that neutral lipid stains suffer from the same pitfalls in the fixation and dehydration steps as do lipids, owing to their similar hydrophobic nature.
To increase localization precision in the TEM, pre-embedding immuno-EM with gold-particle-conjugated antibodies or immunoperoxidase that are detectable in the EM due to the high scattering power of the metal is also possible [66,67].However, these protocols, similar to immunolabeling for light microscopy experiments, usually require permeabilization to facilitate reagent entry into cells, which will inevitably somewhat compromise structural integrity [68,69].To avoid permeabilization, Tokuyasu (Fig. 1D) and post-embedding on-section CLEM are viable alternatives.The Tokuyasu technique (cryosectioned, solvent-and resin-free) shows excellent antibody specificity and a high signal-to-noise ratio [8,70].However, it is also not free from structural artifacts, and LDs require special considerations like the use of uranyl acetate and osmium tetroxide to preserve and visualize LDs (Os is incompatible with immunocytochemistry) [12].In on-section CLEM, a high-pressure frozen and freezesubstituted sample is embedded in a specific hydrophobic resin that preserves the immunoreactivity of epitopes, permitting antibody labeling and light microscopy imaging on the section [62].In Apex2-EM, a peroxidase-derived genetic tag is expressed in live cells, allowing for signal development after fixation by reaction with DAB (3,3 0 -Diaminobenzidine) and H 2 O 2 enhanced by osmium tetroxide, resulting in electron-dense contrast agent deposition at the tagged protein location [71,72].Moreover, pretreatment of cells with the unsaturated fatty acid DHA (docosahexaenoic acid), which is incorporated into LDs, increases reactivity with Os and enhances LDs contrast [73].
In cryo-CLEM workflows, cryopreserved cells may be imaged before or after cryo-FIB micromachining (when cell thinning is required) using widefield or confocal fluorescence microscopes equipped with cryogenic stages [74][75][76][77] and a similar palette of fluorescent molecules as in RT approaches [33,78].Immunolabeling and other chemical reactions deteriorate the pristine structural preservation offered by cryogenic approaches, and cryogenic alternatives for the specific labeling of macromolecules-of-interest in cellular cryotomograms are being actively developed [79,80].
In summary, to increase the success rate in specific LD-targeting and visualization by EM, it is worth considering correlative strategies already at the sample preparation stage.

Characterization of LDs at the organelle and sub-organelle level by EM
With this versatile range of EM methodologies, we proceed to highlight some examples where EM has played a pivotal role in broadening our understanding of LD biology at different scales of spatial resolution.

LD lifecycle
EM has offered important snapshots into the different stages of the LD life cycle, from biogenesis to breakdown.During LD formation, neutral lipids are speculated to form phase-separated lenses in the ER membrane, followed by their budding into the cytoplasm [81,82].Putative neutral lipid lenses of 30 to 60 nm diameter and enclosed within the ER bilayer have been observed in yeast cells in HPF, freezesubstituted, resin-embedded sections imaged by RT tomography (Fig. 1H) [83].RT EM of chemically fixed specimens has also been used to visualize early LD formation in mammalian cells, with observations of putative nascent LD structures that are 30 to 100 nm in diameter with thread-like bridges to the ER [22], or papillary ER protrusions and vesicular structures closely associated with the ER but lacking a distinct neutral lipid core [84].ER phospholipid composition and membrane asymmetry are key factors contributing to LD budding, and EM has allowed for the direct visualization of aberrantly ER-enwrapped LDs upon manipulation of these conditions [85,86].Despite these intriguing observations, the detailed membrane architecture at early LD biogenesis remains unclear, mainly due to the small size and expected metastability of early LD intermediates, rendering them difficult to preserve and observe using traditional RT EM techniques.
EM has also offered fascinating glimpses into LD degradation, including snapshots of direct interactions of LDs with autophagosomal membranes [87], piecemeal engulfment of LD constituents directly into lysosomes in human hepatocytes (Fig. 1I) [88], and detailed characterization of LD-vacuole interactions in yeast [89][90][91].In situ cryo-EM has further revealed surprising dynamic alterations in the structural arrangement of LD core lipids upon mobilization of triglycerides for membrane synthesis or energy production, leading to increased cholesterol ester levels and their phase transition into a crystalline organization in LDs (Fig. 2C) [45,46].Freeze-fracture EM of cells incubated with cholesterol-rich lipoproteins has similarly hinted at structuring within LDs, showing the presence of multiple lamellae enclosing amorphous areas in the LD cores [92].Dynamic fatty acid fluxes have been investigated using EM, with the fatty acid DHA-enhanced neutral lipid contrast allowing for analysis of lipid incorporation into pre-existing LDs, revealing that most preexisting LDs can attain newly synthesized neutral lipids, presumably from the ER [73].

EM of LD-organelle interactions
Owing to the unique advantages of heavy metals as a general stain for biomolecules or cryo-EM as a labelfree imaging method, EM offers a holistic view of LDs in their native milieu.This has expedited the discovery and characterization of many LD-organelle interactions.Indeed, the special relationship between LDs and their mother organelle, the ER, was already evident in early EM studies [7].There appear to be at least two main morphological categories of ER-LD contact sites [95]: (a) discrete ER-LD necks with membrane continuity that are regulated by seipin (Fig. 1K) [19,22,66,[96][97][98], and (b) more-extensive, egg-in-a-cuplike contact sites, where LD and ribosome-free ER membranes are closely apposed but lack direct membrane continuity (Fig. 1B) [39,61,99].ER-LD necks may be a consequence of LD biogenesis occurring at the seipin foci [100], whereas the more extensive contact sites may be formed by ER-LD tethering proteins such as Rab18, Snx14, VPS13, and MOSPD2, possibly functioning in LD expansion [101][102][103][104]. RT EM has provided direct observations of proteinaceous ER-LD tethers [18,22], paving the way for future work to decipher the in situ structures of such tethers, similarly to those recently elucidated for the tunnel-like lipid transport protein VPS13 at ER-lysosomal contact sites [105,106].It should be noted that, due to the ubiquitous nature of the ER, detailed investigations of ER-LD contact sites truly necessitate the enhanced resolution offered by EM.Importantly, EM has been instrumental in unambiguously visualizing the membrane continuities between these two organelles.
Early EM studies also noted the high prevalence of LD-mitochondria contacts in differentiating adipocytes (Fig. 1B) [107].More recent studies with in situ cryo-ET are starting to unravel the molecular machinery acting at these membrane contact sites (Fig. 2B) [108][109][110][111][112][113], including direct visualizations of molecular tethers [45].An emerging theme is that mitochondria associated with LDs may have distinct bioenergetic and enzymatic properties compared to other mitochondria, possibly fine-tuned to promote LD expansion or breakdown as dictated by the metabolic needs of the cell.EM has indeed been instrumental in quantifying LD-mitochondria interactions under various metabolic conditions [114][115][116][117][118].
Less-well studied intracellular interactions, such as those of LDs with peroxisomes [33,119], pathogens [120][121][122][123], and the cytoskeleton [124,125], have been supported by EM observations.Recent investigations into LD-LD contact sites support a model where a potentially phase-separated Cidec-condensate facilitates the flux of neutral lipids along an internal pressure gradient from smaller to larger LDs [47,113,126,150].Cryo-ET showed that at these contact sites, the apposed LD monolayers remain separate, maintaining an approximate 10 nm distance from each other.Importantly, large LDs were often indented by smaller LDs, directly informing on their respective relative internal pressures, the difference of which drives the lipid transfer process [47,126].This exemplifies how EM can help bridge the gap between the biophysical, structural, and cell biology perspectives on LDs.

Molecular machinery in LDs
The life cycle of LDs is regulated by an array of lipidmetabolic enzymes and molecular machineries that control biogenesis and membrane contact sites.Cryo-EM of purified macromolecular complexes has been used in recent years to provide atomistic structural models that inform on the molecular mechanisms involved for many key players in LD biology.Cryo-EM single-particle analysis (SPA) entails capturing tens-to-hundreds of thousands of snapshots of such purified macromolecules frozen in many different orientations and computationally averaging them to provide a high-resolution 3D reconstruction of the target molecule (Fig. 2D,E).Compared to traditional structural biology methods such as X-ray crystallography, SPA can be especially useful in the structural characterization of challenging targets such as integral membrane proteins, as exemplified by the recent structures of key triglyceride and cholesterol ester synthesis enzymes, DGAT1 (diacylglycerol Oacyltransferase 1) and ACAT1 (acyl-coenzyme A: cholesterol acyltransferase 1) (Fig. 2D-F) [127][128][129][130].The structures of these intricate multi-transmembrane ERresident proteins strongly suggest that the newly synthesized neutral lipids are deposited within the leaflets of the ER, consistent with the lens model of LD biogenesis.Following neutral lipid synthesis within the ER leaflets, the lipodystrophy protein seipin is postulated to catalyze LD nucleation, a model suggested by the partial cryo-EM structures of seipin [131][132][133][134] and molecular dynamics simulations supported by cell biology experiments [135][136][137][138].For a more detailed discussion on seipin and its structures, we point the reader to a thorough review included in this issue of FEBS letters [139].Altogether, these recent works start to build a molecular view of LD assembly.
Classical immuno-EM approaches have provided a detailed view on the localization of perilipins on the LD surface (Fig. 1C) [7,39,140], where they mainly function as a protective barrier against lipolysis [141].Interestingly, some freeze fracture immunolabeling EM also indicated perilipins could reside in the LD core [92], although this intriguing finding has yet to be confirmed using other techniques.Perilipins are defined by a domain architecture composed of an N-terminal PAT domain (named after the historical names of perilipins 1-3: Perilipin1, Adipophilin, and TIP47), followed by variable stretches of amphipathic helix forming 11-mer repeats and a C-terminal 4-helix bundle [142].There is currently limited direct structural data on perilipins, with only the perilipin 4-helix bundle solved by X-ray crystallography.Emerging evidence from in vitro, cell biology, and structural studies indicates that the unique biophysical properties of the LD phospholipid monolayer are key to perilipin recruitment to the LD surface [143][144][145], with their binding to the monolayer imposing order on the otherwise disordered domains [143,146].Such structural dynamics may be the reason why perilipins have thus far been a challenging target for structural studies, as most methods typically require structurally defined and stable complexes to arrive at high-resolution models.Finally, a detailed molecular understanding of the LD breakdown machinery is also still largely lacking.

Future opportunities and challenges in connecting structures to LD cell biology
An overreaching goal in modern structural biology is to resolve macromolecular structures in action and in their native functional environment.Recent technological advances in cryo-ET of FIB-milled cells have begun to realize this goal [41,48,49,50,51,52,54].In combination with advanced structure determination algorithms being implemented in subtomogram averaging (STA) workflows [53], an analysis technique analogous to SPA, the structures of molecular machines such as ribosomes are starting to be resolved to atomistic detail and in a number of functional states directly within intact cells [147].It will be exciting to start leveraging these tools in the context of LD biology.However, several bottlenecks remain.Despite significant recent improvements, cryo-FIB and cryo-ET are still relatively low-throughput techniques requiring highly specialized infrastructure and expertise.Cryo-FIB preparations are also not completely artifact-free and can introduce damage to the sample.Furthermore, successfully attaining high resolutions to visualize protein secondary structure elements in STA requires capturing thousands of instances of the macromolecule-of-interest, which is difficult to attain for low-abundance proteins (e.g.seipin at ER-LD contacts) and is further complicated by the large structural flexibility expected for macromolecular complexes in action in a live cell.Nevertheless, even medium-tolow-resolution maps obtained in situ, especially when interpreted with integrative structural modeling, can lead to important biological insights [75,106,148].Finally, a challenging aspect is how to identify specific macromolecules-of-interest in the cryo-tomograms, especially if the targets are small in size, and/or lack welldefined structural features, and are embedded in dense lipid environments, as is likely the case for many LDrelated proteins.Current cryo-CLEM approaches do not provide the resolution and precision required to localize individual macromolecular complexes.Encouragingly, labeling of proteins-of-interest with probes designed to assemble into structurally well-defined particles, allowing for their identification in cryo-tomograms and localization with nanometer precision, shows promise in this regard [79,80].For example, using bacterialderived protein probes as localization markers enabled pinpointing the precise location of seipin at ER-LD contacts in cryo-FIB-milled human cells [80].Considering the biophysically unique membrane environment of LDs, which may be somewhat challenging to mimic in vitro, it is conceivable that approaches such as in situ cryo-ET will be required and may be able in the future to resolve the structures of LD-related proteins.

Conclusions and perspectives
Here, we have summarized the basic principles of EM techniques employed in LD research.The unique lipidrich nature of LDs amongst cellular organelles necessitates careful consideration when choosing a suitable EM method to minimize specimen preparation artifacts but also to gain sufficient resolution and potentially volumetric information to answer the questions under study.No single EM technique alone will be suitable to address all questions, and most key EM studies in the LD field have employed a rich palette of complementary techniques.
The label-free nature of cryo-EM holds promise to provide unprecedented, possibly even hypothesisfueling, snapshots into the intracellular structure and diverse interactions of LDs.On the other hand, atomistic-level insight into LD assembly and breakdown can be attained with meticulous characterization of the relevant purified or reconstituted macromolecular complexes.Continuous advances in these technologies, which allow the acquisition and processing of large datasets, are likely to transform cellular cryo-EM from a descriptive to a quantitative imaging tool, providing far more than just beautiful pictures.In combination with complementary methods, including advanced light microscopy, cell biological approaches and perturbations, molecular modeling, and dynamic simulations, these techniques will deliver a detailed mechanistic understanding of the different stages of the LD lifecycle in health and disease.

Table 1 .
An overview of the basic technical details and requirements of cellular EM imaging and preparation methods.References are provided as examples for studies related to LD research.