Sample preparation methods

A requirement to visualise biological specimens using cryo-ET (or cryo-EM) is that the specimen must be vitrified, meaning that the aqueous environment of the specimen of interest should form an amorphous, glass-like arrangement (Dubochet & McDowall, Reference Dubochet and McDowall1981; McDowall et al., Reference McDowall, Chang, Freeman, Lepault, Walter and Dubochet1983). Vitrification preserves the sample in a near-native state, providing ideal conditions for imaging while minimising radiation damage, with no artefacts in imaging caused by crystalline ice, by avoiding electron diffraction from ice crystals that corrupt the acquired data (Dubochet et al., Reference Dubochet, Adrian, Chang, Homo, Lepault, McDowall and Schultz1988; Henderson, Reference Henderson1992). It is worth mentioning that while vitrification has been deemed as a necessity for cryo-EM, recent work has demonstrated reduction in beam induced motion and better reconstructions from the initial frames of the movie acquisition in a specimen devitrified in a controlled manner (Wieferig et al., Reference Wieferig, Mills and Kühlbrandt2021). Nonetheless, to prepare a suitably vitrified sample, the specimen must be cooled at a rate faster than the rate of crystalline ice formation (Dubochet et al., Reference Dubochet, Adrian, Chang, Homo, Lepault, McDowall and Schultz1988). While most biological specimens are present in an aqueous solution, the peculiarities of each individual specimen being studied in any particular experiment determines how, practically, vitrification is performed to ensure suitable preservation for cryo-ET. Broadly, there are two major techniques that can be utilised to produce a vitrified sample – plunge freezing for thin specimens, up to ~10 μm in thickness (Fuest et al., Reference Fuest, Schaffer, Nocera, Galilea-Kleinsteuber, Messling, Heymann, Plitzko and Burg2019), and high-pressure freezing (HPF) for thicker specimens. For both plunge frozen and high-pressure frozen samples, the quality of vitrification must be investigated experimentally (for instance, by using electron diffraction), as this cannot be reliably assumed a priori, because vitrification intimately depends on the chemical characteristics of the sample.

Vitrification of thick samples

An alternative to the approaches listed above for thin specimens is available, termed high-pressure freezing (HPF), which was developed several decades ago (Moor & Riehle, Reference Moor, Riehle and Bocciarelli1968), and is particularly suitable for thicker samples up to ~200 μm (Kelley et al., Reference Kelley, Raczkowski, Klykov, Jaroenlak, Bobe, Kopylov, Eng, Bhabha, Potter, Carragher and Noble2022; Studer et al., Reference Studer, Humbel and Chiquet2008). During HPF, a pressure of ~2100 bar is applied to biological specimens clasped between two metal planchettes during freezing. As ice is less dense than water, the high pressure hinders crystalline ice formation, thereby reducing the cooling rate requirement for vitrification (Moor, Reference Moor1987). To further prevent crystalline ice from forming and improve vitrification, a cryoprotectant can be added to the sample such as glycerol (Dahl & Staehelin, Reference Dahl and Staehelin1989), glycans (I. Y. Chang et al., Reference Chang, Rahman, Harned, Cohen-Fix and Narayan2021), polyvinyl compounds (Weil et al., Reference Weil, Ruhwedel, Meschkat, Sadowski and Möbius2019) and 1-hexadecene (McDonald et al., Reference McDonald, Schwarz, Müller-Reichert, Webb, Buser and Morphew2010). These cryoprotectants prevent the formation of crystalline ice by increasing the global concentration of all solutes present in the aqueous sample (Pegg, Reference Pegg2007). Another special cryoprotectant is 2-methylpentane, which can be sublimed from the vitrified specimen by heating to −150 °C, allowing additional advantages such as post-addition of fiducials, as well as for exposing the surface topography of specimens to reduce the volume that needs to be removed in the downstream thinning step (Harapin et al., Reference Harapin, Börmel, Sapra, Brunner, Kaech and Medalia2015; S. Wang et al., Reference Wang, Zhou, Chen, Jiang, Yan, You and Li2023).

Another route to accessing thicker volumes is to use the so-called waffle method (Figure 1a) following earlier reports of a similar nature (Weiner et al., Reference Weiner, Kapishnikov, Shimoni, Cordes, Guttmann, Schneider and Elbaum2013), where a grid is sandwiched between two planchettes and high-pressure frozen using the grid bars as a spacer (Kelley et al., Reference Kelley, Raczkowski, Klykov, Jaroenlak, Bobe, Kopylov, Eng, Bhabha, Potter, Carragher and Noble2022). This approach is designed to accommodate various samples at a thickness compatible with maximal reasonable gallium milling depth, which is ~50 μm (Schaffer et al., Reference Schaffer, Pfeffer, Mahamid, Kleindiek, Laugks, Albert, Engel, Rummel, Smith, Baumeister and Plitzko2019). This approach is applicable to cellular or multicellular samples, sometimes made possible by concentrating the cells (by skipping blotting), thus circumventing preferred orientation of the cells, and could be useful for purified particles as well (Kelley et al., Reference Kelley, Raczkowski, Klykov, Jaroenlak, Bobe, Kopylov, Eng, Bhabha, Potter, Carragher and Noble2022).

Figure 1.

Overview of sample preparation by HPF and FIB milling of cellular and tissue specimens. (a) Cartoon description of the waffle assembly – the EM grid is placed between two planchettes and subsequently vitrified using HPF. Adapted from Kelley et al. (Reference Kelley, Raczkowski, Klykov, Jaroenlak, Bobe, Kopylov, Eng, Bhabha, Potter, Carragher and Noble2022). Image is CC BY, license link: http://creativecommons.org/licenses/by/4.0/. (b) Schematic showing the geometry of the focused ion beam, SEM and the grid containing the sample (top). SEM image of a plunge-frozen sample with the milling direction marked, and myofibrils are marked with red arrows (bottom left). Polished lamella images, top-view imaged with the SEM, and side-view imaged with the FIB (bottom right). Bottom left scale bar 50 µm; Bottom right panel 5 µm. Adapted from Z. Wang et al. (Reference Wang, Grange, Wagner, Kho, Gautel and Raunser2021). (c) Serial lift-out workflow: After the region of interest was identified using fluorescent labelling (green), the micromanipulator was mounted, and the area was milled in preparation for lift-out (top left). The slab removed in the previous step is positioned for subsequent thinning (top right). Overview of the milled sections prior to cryo-ET data collection (bottom). Adapted from Schiøtz et al. (Reference Schiøtz, Kaiser, Klumpe, Morado, Poege, Schneider, Beck, Klebl, Thompson and Plitzko2023). Image is CC BY, license link: http://creativecommons.org/licenses/by/4.0/.

Yet even HPF is limited by the sample thickness and is typically useful only up to 100–200 μm (Kelley et al., Reference Kelley, Raczkowski, Klykov, Jaroenlak, Bobe, Kopylov, Eng, Bhabha, Potter, Carragher and Noble2022; Studer et al., Reference Studer, Humbel and Chiquet2008). Accessing thick tissues is currently made possible by initial mechanical sectioning using a vibratome prior to vitrification. Typically, the sample is immersed in buffer or embedded in agar, then sliced using a blade and placed on a grid for HPF (Creekmore et al., Reference Creekmore, Kixmoeller, Black, Lee and Chang2024; Matsui et al., Reference Matsui, Spangler, Elferich, Shiozaki, Jean, Zhao, Qin, Zhong, Yu and Gouaux2024; J. Zhang et al., Reference Zhang, Zhang, Sun, Ji, Huang, Niu, Xu, Ma, Zhu, Gao, Xu and Sun2021). However, this step prolongs the period between sample isolation and freezing and can introduce cutting artefacts at the surface of the sample, which could hinder the preservation of the native, physiologically relevant state of interest. Thus, at the sample preparation stage, there is an urgent need to develop novel techniques for obtaining vitrified samples of much larger volumes, in particular when aiming to image entire tissues and organisms (Baumeister, Reference Baumeister2005; Hutchings & Zanetti, Reference Hutchings and Zanetti2018).

Thinning procedures

For cryo-ET data acquisition, since the electrons must be transmitted through the biological specimen, to be able to contribute to image formation at the detector, the mean free path of electrons, or removal of inelastically scattered electrons by an energy filter, limits specimen thickness usable for cryo-ET. This thickness limit has been estimated by different studies that reported slightly different values, with some studies reporting this limit to be as low as 300 nm, because the effective thickness of the specimen increases significantly at high tilt angles during cryo-ET data acquisition, and the signal-to-noise ratio is thus greatly diminished (Petrov et al., Reference Petrov, Müller and Glaeser2022). Even for a 200 nm-thick specimen, it has been reported that the total transmitted electrons are roughly half of the illuminated dose, as they interact with the sample resulting in decoherence and energy loss (Elbaum, Reference Elbaum2018). As most cellular specimens, apart from a few examples of smaller microbial cells (O’Reilly et al., Reference O’Reilly, Xue, Graziadei, Sinn, Lenz, Tegunov, Blötz, Singh, Hagen, Cramer, Stülke, Mahamid and Rappsilber2020; von Kügelgen et al., Reference von Kügelgen, Cassidy, Van Dorst, Pagani, Batters, Ford, Löwe, Alva, Stansfeld and Tanmay Bharat2024), are usually thicker than 200–300 nm (Melia & Bharat, Reference Melia and Bharat2018; Oikonomou et al., Reference Oikonomou, Chang and Jensen2016), they must be thinned before cryo-ET can be performed. Previously, sample thinning for cryo-EM was only possible using cryo-ultramicrotomy, where a diamond knife is used to produce thin sections of the specimen at cryogenic temperatures. These sections are subsequently placed on an EM grid for imaging (Al-Amoudi et al., Reference Al-Amoudi, Norlen and Dubochet2004; Bharat et al., Reference Bharat, Hoffmann and Kukulski2018; Eltsov et al., Reference Eltsov, Grewe, Lemercier, Frangakis, Livolant and Leforestier2018; Gilbert et al., Reference Gilbert, Fatima, Jenkins, O’Sullivan, Schertel, Halfon, Wilkinson, Morrema, Geibel, Read, Ranson, Radford, Hoozemans and Frank2024; Leistner et al., Reference Leistner, Wilkinson, Burgess, Lovatt, Goodbody, Xu, Deuchars, Radford, Ranson and Frank2023; McDowall et al., Reference McDowall, Chang, Freeman, Lepault, Walter and Dubochet1983). This method, termed cryo-electron microscopy of vitrified sections (CEMOVIS), might lead to distortions in the specimen including expansion and compression due to the mechanical action of the knife. Even though CEMOVIS can be used successfully to study cellular and tissue samples in situ (Bharat et al., Reference Bharat, Hoffmann and Kukulski2018; Gilbert et al., Reference Gilbert, Fatima, Jenkins, O’Sullivan, Schertel, Halfon, Wilkinson, Morrema, Geibel, Read, Ranson, Radford, Hoozemans and Frank2024; Ma et al., Reference Ma, Chong, Lee, Cai, Siebert, Howe, Zhang, Shi, Surana and Gan2022), practical application of CEMOVIS tends to be quite tedious as the sample is prepared manually with low throughput, with the skill of the experimentalist critically determining the outcome of the procedure (Al-Amoudi et al., Reference Al-Amoudi, Studer and Dubochet2005).

Cryo-ET of thinned specimens

Once the multicellular specimen has been appropriately thinned, it is ready for cryo-ET data collection for structural and cell biology. One of the major factors limiting the throughput of cryo-ET is the long acquisition time of tilt-series, compared to cryo-EM single particle analysis (Böhning & Bharat, Reference Böhning and Bharat2021). Different data collection schemes have sought to overcome this hurdle to support widespread application of cryo-ET. One such approach accelerates the speed of a single tilt-series acquisition by implementing a continuous data collection (Chreifi et al., Reference Chreifi, Chen, Metskas, Kaplan and Jensen2019, Reference Chreifi, Chen and Jensen2021; Eisenstein et al., Reference Eisenstein, Danev and Pilhofer2019). In this scheme, the specimen is exposed and tilted continuously (in a single movement) without the need to track and correct stage shifts, required in standard cryo-ET data collection (Mastronarde, Reference Mastronarde2005). Abandoning these constant adjustments, which require slow mechanical stage movements in the microscope, increases the speed of tilt-series acquisition up to an order of magnitude, but limits the overall quality of the reconstructed tomograms since the tilt angle of each image must be estimated experimentally (Chreifi et al., Reference Chreifi, Chen, Metskas, Kaplan and Jensen2019). Other approaches aimed at optimally imaging each square nanometre of the valuable milled area of the specimen include the use of overlapping tiles that are stitched together, thereby forming mosaic images that can eventually be merged and reconstructed as a highly detailed, large tomographic volume (Peck et al., Reference Peck, Carter, Mai, Chen, Burt and Jensen2022). Alternatively, the beam shape could be changed to a square to maximise the collection area within the lamella and permit data acquisition of neighbouring areas without losing high-resolution features due to overlapping, unnecessary exposure during data collection (Brown et al., Reference Brown, Smith, Wardle and Hanssen2024; Chua et al., Reference Chua, Alink, Kopylov, Johnston, Eisenstein and de Marco2024).

Perhaps the most widely applicable acquisition strategy parallelises cryo-ET data collection by defining a geometric model describing the lamella surface (or any specimen surface) relative to the tilt axis. This geometric model helps in parallelised data collection by utilising beam image shifts combined with a single tracking area, hence allowing multiple tilt-series acquisition in a nearly simultaneous manner. This facilitates the collection of hundreds of tilt series in a single session, substantially increasing throughput compared to the traditional collection schemes (Bouvette et al., Reference Bouvette, Liu, Du, Zhou, Sikkema, da Fonseca Rezende e Mello, Klemm, Huang, Schaaper, Borgnia and Bartesaghi2021; Eisenstein et al., Reference Eisenstein, Yanagisawa, Kashihara, Kikkawa, Tsukita and Danev2022; J. E. Yang et al., Reference Yang, Larson, Sibert, Kim, Parrell, Sanchez, Pappas, Kumar, Cai, Thompson and Wright2023). To overcome errors introduced either by misaligned lamellae, specimen rotation caused by the mechanical autoloader system, and inaccurate measurement of the lamella’s eucentric position, the geometric model is used to compensate for these errors and is updated throughout sample tilting in the PACE-tomo (parallel cryo electron tomography) workflow (Figure 2a; Eisenstein et al., Reference Eisenstein, Yanagisawa, Kashihara, Kikkawa, Tsukita and Danev2022). To complement this approach and introduce further automation, a machine learning model dubbed SPACE-tomo (smart parallel automated cryo electron tomography) was trained to facilitate unattended lamella definition, identification and segmentation of regions of interest within the lamella, and data acquisition set-up (Eisenstein et al., Reference Eisenstein, Fukuda and Danev2023). As these methods become more widespread, we expect that such unsupervised approaches will become an indispensable part of the cryo-ET data collection pipelines.

Figure 2.

Cryo-ET data collection and high resolution subtomogram averaged structures. (a) FIB-milled lamella with defined regions for parallel cryo-ET data acquisition using beam image shifts. The tilt axis is marked with a dashed line. Adapted from Eisenstein et al. (Reference Eisenstein, Yanagisawa, Kashihara, Kikkawa, Tsukita and Danev2022). Reproduced with permission from SNCSC. (b) Slice through a tomogram of an entire microbial cell where ribosomes, nucleoid and the surface layer (S-layer) encapsulating the cell are all visible. Inset - the subtomogram averaged map of the S-layer, scale bar 50 nm. Adapted from von Kügelgen et al. (Reference von Kügelgen, Cassidy, Van Dorst, Pagani, Batters, Ford, Löwe, Alva, Stansfeld and Tanmay Bharat2024). Image is CC BY, license link: http://creativecommons.org/licenses/by/4.0/. (c) Slice through a tomogram of the sarcomere thin and thick filaments along with the subtomogram averaged map of the thin filament with a bound myosin. Scale bar 20 nm. Adapted from Z. Wang et al. (Reference Wang, Grange, Pospich, Wagner, Kho, Gautel and Raunser2022). Reprinted with permission from AAAS.

While cryo-ET data collection has advanced significantly since the advent of direct electron detectors, there is still room to substantially improve the quality of cryo-ET data, which will potentially have a huge impact on reducing the amount of data needed to solve structures inside cells and tissues. Lamella preparation requires a lot of time and effort; therefore, it is imperative that the cryo-ET data collected is of the highest possible quality (Ochner & Bharat, Reference Ochner and Bharat2023). From a hardware standpoint, the laser phase plate is drawing significant attention and holds the potential to substantially improve the signal-to-noise ratio in cryo-ET by modulating the phase contrast difference between scattered and unscattered electrons, during in-focus specimen imaging (Müller et al., Reference Müller, Jin, Danev, Spence, Padmore and Glaeser2010). The unscattered electrons are focused and passed through an electric field generated by an ultrafast continuous (Schwartz et al., Reference Schwartz, Axelrod, Campbell, Turnbaugh, Glaeser and Müller2019) or pulsed (Du & Fitzpatrick, Reference Du and Fitzpatrick2023) laser. The passage through the field induces a phase shift caused by the ponderomotive force, resulting in a phase difference (π/2 in the case for a quarter phase plate), thus boosting contrast significantly even when the specimen is in focus (Du & Fitzpatrick, Reference Du and Fitzpatrick2023; Müller et al., Reference Müller, Jin, Danev, Spence, Padmore and Glaeser2010; Schwartz et al., Reference Schwartz, Axelrod, Campbell, Turnbaugh, Glaeser and Müller2019). The phase contrast of the detected electrons results in better image quality by converting the sine oscillation to cosine, thus increasing the low-frequency signal, although CTF correction is still required for both low and high resolution (Campbell et al., Reference Campbell, Schwartz, Axelrod, Turnbaugh, Glaeser and Müller2018; Müller et al., Reference Müller, Jin, Danev, Spence, Padmore and Glaeser2010; Petrov et al., Reference Petrov, Müller and Glaeser2022; Schwartz et al., Reference Schwartz, Axelrod, Campbell, Turnbaugh, Glaeser and Müller2019). The drawback of the current design was described as a resolution loss due to magnetic field fluctuations caused by the laser pulses, which are currently being investigated for future improvements (Axelrod, Petrov, Zhang, Remis, et al., Reference Axelrod, Petrov, Zhang, Remis, Buijsse, Glaeser and Mȕller2023; Axelrod, Petrov, Zhang, Sandhaus, et al., Reference Axelrod, Petrov, Zhang, Sandhaus, Remis, Glaeser and Müller2023).

Subtomogram averaging structure determination

Once cryo-ET data on the multicellular specimen has been collected, subtomogram averaging (STA) can be applied to obtain structural information from the macromolecules present within the specimen. Subtomogram averaging approaches have been described in other reviews focused on this method (Briggs, Reference Briggs2013; Lučić et al., Reference Lučić, Rigort and Baumeister2013); therefore, it is only considered here briefly for completeness. Following tilt-series acquisition, tomograms can be reconstructed in a variety of software (Kremer et al., Reference Kremer, Mastronarde and McIntosh1996; Zheng et al., Reference Zheng, Wolff, Greenan, Chen, Faas, Bárcena, Koster, Cheng and Agard2022) using a variety of pipelines (Burt et al., Reference Burt, Toader, Warshamanage, von Kügelgen, Pyle, Zivanov, Kimanius, Bharat and Scheres2024; Himes & Zhang, Reference Himes and Zhang2018; H. F. Liu et al., Reference Liu, Zhou, Huang, Piland, Jin, Mandel, Du, Martin and Bartesaghi2023). Next, typically tomographic volumes are denoised to improve contrast (Buchholz et al., Reference Buchholz, Jordan, Pigino and Jug2018; Y. T. Liu et al., Reference Liu, Zhang, Wang, Tao, Bi and Zhou2022), after which researchers can use a variety of tools for manual picking, template matching or other feature identification tasks (Chaillet et al., Reference Chaillet, Roet, Veltkamp and Förster2024; Cruz-León et al., Reference Cruz-León, Majtner, Hoffmann, Kreysing, Kehl, Tuijtel, Schaefer, Geißler, Beck, Turoňová and Hummer2024; de Teresa-Trueba et al., Reference de Teresa-Trueba, Goetz, Mattausch, Stojanovska, Zimmerli, Toro-Nahuelpan, Cheng, Tollervey, Pape, Beck, Diz-Muñoz, Kreshuk, Mahamid and Zaugg2023; Lucas et al., Reference Lucas, Himes and Grigorieff2023; Moebel et al., Reference Moebel, Martinez-Sanchez, Lamm, Righetto, Wietrzynski, Albert, Larivière, Fourmentin, Pfeffer, Ortiz, Baumeister, Peng, Engel and Kervrann2021; Rice et al., Reference Rice, Wagner, Stabrin, Sitsel, Prumbaum and Raunser2023; T. Wagner et al., Reference Wagner, Merino, Stabrin, Moriya, Antoni, Apelbaum, Hagel, Sitsel, Raisch, Prumbaum, Quentin, Roderer, Tacke, Siebolds, Schubert, Shaikh, Lill, Gatsogiannis and Raunser2019, Reference Wagner, Lusnig, Pospich, Stabrin, Schonfeld and Raunser2020; Wan et al., Reference Wan, Khavnekar and Wagner2024). These subtomogram selection tasks can be followed by classification and subtomogram averaging (Burt et al., Reference Burt, Toader, Warshamanage, von Kügelgen, Pyle, Zivanov, Kimanius, Bharat and Scheres2024; M. Chen et al., Reference Chen, Bell, Shi, Sun, Wang and Ludtke2019; Förster et al., Reference Förster, Medalia, Zauberman, Baumeister and Fass2005; H. F. Liu et al., Reference Liu, Zhou, Huang, Piland, Jin, Mandel, Du, Martin and Bartesaghi2023; Tegunov et al., Reference Tegunov, Xue, Dienemann, Cramer and Mahamid2021). Subtomogram averaging allows structure determination of macromolecules in their native environment (Figure 2b-c; Allegretti et al., Reference Allegretti, Zimmerli, Rantos, Wilfling, Ronchi, Fung, Lee, Hagen, Turoňová, Karius, Börmel, Zhang, Müller, Schwab, Mahamid, Pfander, Kosinski and Beck2020; S. Chen et al., Reference Chen, Basiashvili, Hutchings, Murillo, Suarez, Louro, Leschziner and Villa2024; Z. Chen et al., Reference Chen, Shiozaki, Haas, Skinner, Zhao, Guo, Polacco, Yu, Krogan, Lishko, Kaake, Vale and Agard2023; Fedry et al., Reference Fedry, Silva, Vanevic, Fronik, Mechulam, Schmitt, des Georges, Faller and Förster2024; Gemmer et al., Reference Gemmer, Chaillet, van Loenhout, Cuevas Arenas, Vismpas, Gröllers-Mulderij, Koh, Albanese, Scheltema, Howes, Kotecha, Fedry and Förster2023; Q. Guo et al., Reference Guo, Lehmer, Martínez-Sánchez, Rudack, Beck, Hartmann, Pérez-Berlanga, Frottin, Hipp, Hartl, Edbauer, Baumeister and Fernández-Busnadiego2018; Held et al., Reference Held, Liang and Brunger2024; Hoffmann et al., Reference Hoffmann, Kreysing, Khusainov, Tuijtel, Welsch and Beck2022; Hutchings et al., Reference Hutchings, Stancheva, Miller and Zanetti2018; Kravčenko et al., Reference Kravčenko, Ruwolt, Kroll, Yushkevich, Zenkner, Ruta, Lotfy, Wanker, Rosenmund, Liu and Kudryashev2024; Leung et al., Reference Leung, Zeng, Wang, Roelofs, Huang, Zenezini Chiozzi, Hevler, Heck, Dutcher, Brown, Zhang and Zeev-Ben-Mordehai2023; Mattei et al., Reference Mattei, Glass, Hagen, Kräusslich and Briggs2016; Ni et al., Reference Ni, Sun, Burn, Al-Hazeem, Zhu, Yu, Liu and Zhang2022; Obr et al., Reference Obr, Keizer, Righetto, Zhang, Kelley, Khavnekar, Franken, Engel, Plitzko and Kotecha2024; Pyle et al., Reference Pyle, Miller and Zanetti2024; Santos et al., Reference dos Santos, Knowles, Dendooven, Hale, Hale, Burt, Kolata, Cannone, Bellini, Barford and Allegretti2024; Schur et al., Reference Schur, Obr, Hagen, Wan, Jakobi, Kirkpatrick, Sachse, Kräusslich and Briggs2016; Tamborrini et al., Reference Tamborrini, Wang, Wagner, Tacke, Stabrin, Grange, Lin Kho, Rees, Bennett, Gautel and Raunser2023; Turoňová et al., Reference Turoňová, Sikora, Schürmann, Hagen, Welsch, Blanc, von Bülow, Gecht, Bagola, Hörner, van Zandbergen, Landry, de Azevedo, Mosalaganti, Schwarz, Covino, Mühlebach, Hummer, Locker and Beck2020; von Kügelgen et al., Reference von Kügelgen, Tang, Hardy, Kureisaite-Ciziene, Brun, Stansfeld, Robinson and Bharat2020, Reference von Kügelgen, Cassidy, Van Dorst, Pagani, Batters, Ford, Löwe, Alva, Stansfeld and Tanmay Bharat2024; J. Wagner et al., Reference Wagner, Carvajal, Bracher, Beck, Wan, Bohn, Körner, Baumeister, Fernandez-Busnadiego and Hartl2024; Waltz et al., Reference Waltz, Righetto, Kelley, Zhang, Obr, Khavnekar, Kotecha and Engel2024; Z. Wang et al., Reference Wang, Grange, Wagner, Kho, Gautel and Raunser2021, Reference Wang, Grange, Pospich, Wagner, Kho, Gautel and Raunser2022; Watanabe et al., Reference Watanabe, Buschauer, Böhning, Audagnotto, Lasker, Lu, Boassa, Taylor and Villa2020, Reference Watanabe, Zyla, Parekh, Hong, Jones, Schendel, Wan, Castillon and Saphire2024; Wozny et al., Reference Wozny, Di Luca, Morado, Picco, Khaddaj, Campomanes, Ivanović, Hoffmann, Miller, Vanni and Kukulski2023; Xue et al., Reference Xue, Lenz, Zimmermann-Kogadeeva, Tegunov, Cramer, Bork, Rappsilber and Mahamid2022; You et al., Reference You, Zhang, Cheng, Xiao, Ma, Sun, Zhang, Wang and Sui2023; X. Zhang et al., Reference Zhang, Sridharan, Zagoriy, Eugster Oegema, Ching, Pflaesterer, Fung, Becher, Poser, Müller, Hyman, Savitski and Mahamid2023; Zimmerli et al., Reference Zimmerli, Allegretti, Rantos, Goetz, Obarska-Kosinska, Zagoriy, Halavatyi, Hummer, Mahamid, Kosinski and Beck2021), using image processing algorithms that support high-resolution structure determination (Bharat et al., Reference Bharat, Russo, Löwe, Passmore and Scheres2015; Burt et al., Reference Burt, Toader, Warshamanage, von Kügelgen, Pyle, Zivanov, Kimanius, Bharat and Scheres2024; Tegunov et al., Reference Tegunov, Xue, Dienemann, Cramer and Mahamid2021; Zivanov et al., Reference Zivanov, Otón, Ke, von Kügelgen, Pyle, Qu, Morado, Castaño-Díez, Zanetti, Bharat, Briggs and Scheres2022). The resulting structures provide valuable insights on the mode of action of macromolecules in tissues, along with their interactions with drugs, ligands, or accessory molecules in situ. These interactions are often transient or disrupted by protein purification techniques and thus cannot be easily reconstituted and visualised in vitro.

Several modern studies not only report the cellular structures of macromolecules by STA but also map the resulting structures back into the original tomogram, providing additional ultrastructural information of the tissue. With this in mind, we must note that a thinned sample is taken out of the cellular or tissue context, because once thinned, it represents only a small slice from the initial intact specimen. We anticipate that in the next few years, more molecular structures will be characterised using a workflow combining cryo-FIB-SEM, cryo-ET, predictive algorithms (Jumper et al., Reference Jumper, Evans, Pritzel, Green, Figurnov, Ronneberger, Tunyasuvunakool, Bates, Žídek, Potapenko, Bridgland, Meyer, Kohl, Ballard, Cowie, Romera-Paredes, Nikolov, Jain, Adler and Hassabis2021) and cellular transcriptomics and proteomics approaches (Baumeister, Reference Baumeister2005; McCafferty et al., Reference McCafferty, Klumpe, Amaro, Kukulski, Collinson and Engel2024).

Complementary techniques for 3D in situ imaging

Cryo-ET provides molecular resolution in a limited sample volume, due to the requirement of thinning tissue specimens. This limitation can be partially alleviated by montage tomography (Peck et al., Reference Peck, Carter, Mai, Chen, Burt and Jensen2022; J. E. Yang et al., Reference Yang, Larson, Sibert, Kim, Parrell, Sanchez, Pappas, Kumar, Cai, Thompson and Wright2023), which expands the field-of-view in the “X-Y” dimension, and by serial lift-out approaches, which increases depth through fabrication of multiple lamellae from the same tissue (Nguyen et al., Reference Nguyen, Perone, Klena, Vazzana, Kaluthantrige Don, Silva, Sorrentino, Swuec, Leroux, Kalebic, Coscia and Erdmann2024; Schiøtz et al., Reference Schiøtz, Kaiser, Klumpe, Morado, Poege, Schneider, Beck, Klebl, Thompson and Plitzko2023). However, this loss of sample volume due to thinning is to an extent unavoidable in cryo-ET. To circumvent this issue, there are other in situ imaging techniques that provide an alternative option for imaging bulk volumes such as 3D FIB-SEM imaging (also termed serial surface imaging or “slice-and-view”), where a layer of biological material is removed using the FIB followed by imaging of the exposed surface using the SEM. By iterating the FIB-SEM process, a 3D volume can be generated with a nearly isotropic resolution of a few nanometres. This technique is extremely useful for cell biological investigations inside tissues, because it provides a large field-of-view, and depth information through the “Z”-axis of the tissue (Elbaum, Reference Elbaum2018). This serial FIB-SEM technique had previously been widely applied for room temperature specimens that were chemically fixed (Denk & Horstmann, Reference Denk and Horstmann2004; D’Imprima et al., Reference D’Imprima, Garcia Montero, Gawrzak, Ronchi, Zagoriy, Schwab, Jechlinger and Mahamid2023; Heymann et al., Reference Heymann, Hayles, Gestmann, Giannuzzi, Lich and Subramaniam2006; Xu et al., Reference Xu, Hayworth, Lu, Grob, Hassan, García-Cerdán, Niyogi, Nogales, Weinberg and Hess2017, Reference Xu, Pang, Shtengel, Müller, Ritter, Hoffman, ya Takemura, Lu, Pasolli, Iyer, Chung, Bennett, Weigel, Freeman, van Engelenburg, Walther, Farese, Lippincott-Schwartz, Mellman and Hess2021) and has been recently expanded to cryogenic temperature applications (Capua-Shenkar et al., Reference Capua-Shenkar, Varsano, Itzhak, Kaplan-Ashiri, Rechav, Jin, Niimi, Fan, Kruth and Addadi2022; Scher et al., Reference Scher, Rechav, Paul-Gilloteaux and Avinoam2021; Schertel et al., Reference Schertel, Snaidero, Han, Ruhwedel, Laue, Grabenbauer and Möbius2013; Sviben et al., Reference Sviben, Gal, Hood, Bertinetti, Politi, Bennet, Krishnamoorthy, Schertel, Wirth, Sorrentino, Pereiro, Faivre and Scheffel2016; Vidavsky et al., Reference Vidavsky, Masic, Schertel, Weiner and Addadi2015, Reference Vidavsky, Akiva, Kaplan-Ashiri, Rechav, Addadi, Weiner and Schertel2016). Despite the large potential applications, several challenges remain in the pipeline for imaging cryogenic, unstained biological specimen, such as problems with automatic focusing, automatic astigmatism and drift correction on these radiation sensitive samples that are imaged for several hours, and sometimes several days. Moreover, interpretation of the resulting images remains complex due to the incompletely understood mechanisms of contrast formation of cryogenic, unstained biological specimens. While the contrast is suggested to arise from differential surface potential and local charging, additional factors may also contribute (Schertel et al., Reference Schertel, Snaidero, Han, Ruhwedel, Laue, Grabenbauer and Möbius2013; Vidavsky et al., Reference Vidavsky, Akiva, Kaplan-Ashiri, Rechav, Addadi, Weiner and Schertel2016). With the growing attention on cryo-FIB-ET, the 3D FIB-SEM technique is expected to become more widely accessible, as it can be performed using the same instrumentation available in many laboratories for lamella production. Widespread application will likely require theoretical developments in understanding image formation, and in the development of streamlined strategies for data analysis. We hope further software and hardware advancements will address the current challenges, for example by reducing the ion beam size to allow finer slicing of the sample, as well as improved SEM detectors that can decrease the dwell time and allow faster imaging.

In the same vein as FIB-SEM imaging, another alternative method to investigate whole cells or tissues in 3D is cryo-scanning transmission electron microscopy (STEM), which uses a focused electron beam probe rather than flood beam used in TEM applications (Jones & Leonard, Reference Jones and Leonard1978; Kellenberger et al., Reference Kellenberger, Carlemalm, Villiger, Wurtz, Mory and Colliex1986). Cryo-STEM allows scanning over the sample in a tiled manner using multiple detectors that collect information for both transmitted and scattered electrons (Elbaum et al., Reference Elbaum, Seifer, Houben, Wolf and Rez2021; Wolf & Elbaum, Reference Wolf and Elbaum2019). While samples up to 2 μm in thickness can potentially be imaged using cryo-STEM, in practice to obtain data with a good contrast and a reasonable pixel size, the effective specimen thickness is usually less than 1 μm (Kirchweger et al., Reference Kirchweger, Mullick, Wolf and Elbaum2023; Wolf et al., Reference Wolf, Houben and Elbaum2014). Cryo-STEM has been successfully utilised to visualise whole cells (Wolf et al., Reference Wolf, Houben and Elbaum2014), organelles containing granular calcium structures (Kirchweger et al., Reference Kirchweger, Mullick, Wolf and Elbaum2023; Wolf et al., Reference Wolf, Mutsafi, Dadosh, Ilani, Lansky, Horowitz, Rubin, Elbaum and Fass2017), single particle reconstructions at sub-nanometre resolutions of purified proteins and virus particles (Lazić et al., Reference Lazić, Wirix, Leidl, de Haas, Mann, Beckers, Pechnikova, Müller-Caspary, Egoavil, Bosch and Sachse2022), as well as metal ion composition and localisation in purified proteins (Elad et al., Reference Elad, Bellapadrona, Houben, Sagi and Elbaum2017). Cryo-STEM is therefore a complementary technique for cellular imaging, providing another arrow in the quiver of the in situ structural cell biologist.

Another cryo-tomography technique which has been recently used to investigate large cells and tissues, albeit at lower resolution, is cryo-soft X-ray tomography (cryo-SXT), which can provide information through specimens that are several microns thick (Larabell & Le Gros, Reference Larabell and Le Gros2004; Weiß et al., Reference Weiß, Schneider, Niemann, Guttmann, Rudolph and Schmahl2000). In cryo-SXT, contrast is naturally generated by the difference in the K-shell absorption of soft X-rays between carbon (or nitrogen) and oxygen in wavelengths ranging between 2.34–4.4 nm (Larabell & Nugent, Reference Larabell and Nugent2010). Imaging in this spectral range, also termed the ‘water window’, causes organic material, which is abundantly present in cells and organelles to absorb the X-rays, while water and other oxygen rich compounds are effectively transparent (Carzaniga et al., Reference Carzaniga, Domart, Collinson and Duke2014; Larabell & Le Gros, Reference Larabell and Le Gros2004). Cryo-SXT offers not only a large depth of field, which can reach 10–15 μm (Carzaniga et al., Reference Carzaniga, Domart, Collinson and Duke2014; J. Guo & Larabell, Reference Guo and Larabell2019; Uchida et al., Reference Uchida, McDermott, Wetzler, Le Gros, Myllys, Knoechel, Barron and Larabell2009), but also a large field of view together with fast data acquisition times, where unstained and unmodified cells nearly 50 μm in length can be imaged in 20 minutes with a resolution of about 50 nm (Larabell & Nugent, Reference Larabell and Nugent2010; Uchida et al., Reference Uchida, McDermott, Wetzler, Le Gros, Myllys, Knoechel, Barron and Larabell2009). This is much faster when compared to 3D FIB-SEM or cryo-STEM which take several hours or days to collect a dataset of a similar scale. Recent advancements in cryo-SXT include improvement of data collection schemes to increase the depth of field (Otón et al., Reference Otón, Pereiro, Conesa, Chichón, Luque, Rodríguez, Pérez-Berná, Sorzano, Klukowska, Herman, Vargas, Marabini, Carrascosa and Carazo2017); however, the most substantial is the transition from synchrotron-based microscopes into compact, standalone machines which can be operated in a typical laboratory (Fahy et al., Reference Fahy, Weinhardt, Vihinen-Ranta, Fletcher, Skoko, Pereiro, Gastaminza, Bartenschlager, Scholz, Ekman and McEnroe2021, Reference Fahy, Kapishnikov, Donnellan, McEnroe, O’Reilly, Fyans and Sheridan2024), which is expected to make this method available to a wider community.

Conclusions and outlook

In conclusion, we have reviewed recent advances pertaining to sample preparation, thinning strategies and cryo-ET data collection schemes, which are currently being used to investigate multicellular specimens and tissues in situ. From the sample preparation perspective, there is currently no method that can allow a reliable vitrification of specimens thicker than 100–200 μm, meaning that most tissues are currently not directly amenable for imaging by cryo-ET, and innovation in this aspect is urgently needed. This could be achieved by repurposing HPF to accommodate thicker specimens, or by devising alternative techniques for sample preparation. While metal ion beam sources have been used extensively in materials sciences as well as for biological cryo-FIB sample thinning, they are still limited by a low rate of material removal and hence prevent easy access to thicker tissue samples. Investigating different focused ion beams is required to allow faster and more reliable milling, ideally with the potential to reduce the damage the sample undergoes during thinning. As automation and increased throughput are introduced to the FIB milling process, cryo-ET data collection must also improve to allow tomogram acquisition from multiple lamellae across multiple grids. Beam-shift collection schemes could greatly increase the rate of data collection without compromising data quality, but there is room for improvement in making this available for all sorts of applications. To tackle the densely packed cellular environment, and increase the overall contrast of tomograms, the laser phase plate is expected to push the limits of macromolecular identification in tomograms. These and other approaches might help generate higher-resolution tomograms, where sub-nanometre-level details could be resolved and inferred directly from the reconstructed tomogram, without the need for subtomogram averaging. We envision that future cryo-EM instruments will include a combination of multimodal components such as cryo-FIB-SEM, light microscopy objectives and mass spectrometers, that will complement TEM data acquisition, with cryo-ET as the central method of choice linking information from these diverse sources together to help uncover new biological mechanisms.