Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2023 Apr 7.
Published in final edited form as: J Struct Biol. 2022 Apr 26;214(2):107860. doi: 10.1016/j.jsb.2022.107860

Montage electron tomography of vitrified specimens

Ariana Peck a,1, Stephen D Carter a,2, Huanghao Mai b, Songye Chen a,c, Alister Burt d, Grant J Jensen a,e,*
PMCID: PMC10081539  NIHMSID: NIHMS1871136  PMID: 35487464

Abstract

Cryo-electron tomography provides detailed views of macromolecules in situ. However, imaging a large field of view to provide more cellular context requires reducing magnification during data collection, which in turn restricts the resolution. To circumvent this trade-off between field of view and resolution, we have developed a montage data collection scheme that uniformly distributes the dose throughout the specimen. In this approach, sets of slightly overlapping circular tiles are collected at high magnification and stitched to form a composite projection image at each tilt angle. These montage tilt-series are then reconstructed into massive tomograms with a small pixel size but a large field of view. For proof-of-principle, we applied this method to the thin edge of HeLa cells. Thon rings to better than 10 Å were detected in the montaged tilt-series, and diverse cellular features were observed in the resulting tomograms. These results indicate that the additional dose required by this technique is not prohibitive to performing structural analysis to intermediate resolution across a large field of view. We anticipate that montage tomography will prove particularly useful for lamellae, increase the likelihood of imaging rare cellular events, and facilitate visual proteomics.

Keywords: Cryo-electron microscopy, Montage tomography, Image processing, Ultrastructural analysis, Cellular biology

1. Introduction

Cryo-electron tomography (cryo-ET) is a powerful tool for studying macromolecular structures in the near-native context of frozen-hydrated cells, unperturbed by stains or fixatives (Böhning and Bharat, 2021; Tocheva et al., 2010). In this cryo-electron microscopy (cryo-EM) technique, a series of projection images is recorded as a vitrified specimen is tilted in an electron microscope. The resulting tilt-series is reconstructed into a tomogram, or volumetric map of the specimen’s electrostatic potential. Cryo-ET has revealed important details of cellular ultrastructure, and subtomogram averaging algorithms enable determining the structures of macromolecular complexes at a resolution of 0.4–4 nm (Tocheva et al., 2010). Recent technical advances have extended the high-resolution potential of this technique, yielding subtomogram averages of purified HIV-1 Gag particles to 3.1 Å and in situ ribosomes to 3.7 Å (Himes and Zhang, 2018; Tegunov et al., 2021). However, data must be collected at high magnification to retain this high-resolution signal. The trade-off is a smaller field of view, limiting the region that can be imaged to a tiny fraction of a cell.

In principle, montage tomography could permit imaging a large field of view without sacrificing high-resolution details. For montage data collection, the beam is tiled across a specimen and the recorded images are computationally stitched together during reconstruction (Mastronarde, 2005). To date, montage tomography has only been performed on resin-embedded samples, which resist radiation damage but suffer from artifacts induced by chemical fixation (Koning et al., 2008; Noske et al., 2008; Rog-Zielinska et al., 2016). By contrast, cryo-preserved specimens can tolerate only a limited electron dose before being destroyed (Baker and Rubinstein, 2010). Historically this dose sensitivity has been considered prohibitive to collecting montage data from vitrified samples, as portions of the sample must be exposed multiple times to facilitate stitching images during reconstruction.

Recent technical developments, however, motivate revisiting the potential of montage cryo-tomography. First, the highly stable optics of modern microscopes enable precise control over the region being exposed (Chreifi et al., 2019). Such precision is critical both to prevent gaps between neighboring images and to avoid accidentally enlarging the overlap region. Second, the ability to collect data using a circular beam with fringe-free illumination allows for more efficient tiling strategies and reduces loss of information due to corruption by Fresnel fringes (Konings et al., 2019; Weis and Hagen, 2020). Third, the increased sensitivity of modern detectors permits decreasing the exposure while achieving the same signal-to-noise ratio (Bammes et al., 2012; Kuijper et al., 2015; McMullan et al., 2016; Milazzo et al., 2011). Fourth, focused ion beam (FIB) milling has significantly expanded the range of specimens that can be studied by cryo-ET but is highly time-consuming, motivating efforts to image as much of the lamellae as possible (Mahamid et al., 2016; Marko et al., 2007; Rigort et al., 2012). In addition, there is growing recognition in the field that radiation damage is a progressive phenomenon (Bartesaghi et al., 2018; Glaeser, 2016; Hattne et al., 2018). As a result, important features of cellular biology remain observable even after receiving what was previously considered an intolerably high dose for vitrified samples. While other approaches like serial blockface scanning electron microscopy (SBF-SEM) (Lippens et al., 2019), soft X-ray tomography (Le Gros et al., 2014; Loconte et al., 2022), and 3d focused ion beam scanning electron microscopy (FIB-SEM) (Xu et al., 2017) also permit large volume imaging of cellular ultrastructure, montage cryo-ET has the greatest potential to achieve reconstructions with subnanometer resolution.

Montage tomography of cryo-preserved specimens would further the potential of cryo-ET by increasing the likelihood of imaging transient events and providing significantly more cellular context for macromolecules of interest. Here we use simulations to optimize a montage data collection scheme and develop strategies to stitch tiles at each tilt angle into a composite projection image. We then harness modern cryo-ET algorithms to reconstruct tomograms from these montage tilt-series with a small pixel size but a large field of view. To demonstrate proof-of-principle, we applied this technique to the thin edges of HeLa cells. Thon rings to better than 15 Å were observed at both low and intermediate tilt angles in the individual tiles and detected to under 10 Å in the untilted projection images after stitching. The reconstructed tomograms spanned a 3.3 μm2 field of view and contained diverse cellular features, including mitochondria, multilammelar vesicles, and microtubules. These results indicate that despite the additional dose required by this method, montage tomography enables capturing large fields of view for structural analysis at the intermediate resolutions typical of cryo-ET data.

2. Optimization of tiling strategies

Given the sensitivity of biological samples to radiation damage (Bartesaghi et al., 2018; Glaeser, 2016; Hattne et al., 2018), the success of montage tomography depends on efficiently distributing the total exposure both at each tilt angle and across the full tilt-series. The former is readily addressed for a circular beam: the optimal strategy to pack circles in a plane uses a hexagonal tiling scheme, in which circles are centered on the vertices of a regular hexagonal grid and three neighboring circles intersect at a point (Fig. 1A) (Kershner, 1939). The question then remains how to displace these hexagonally-packed circular tiles between tilt angles to most uniformly spread the dose across the tilt-series. Applying global translations and rotations to the hexagonal array of circles between tilt angles changes which regions of the sample lie in an overlap region at each tilt angle, thereby reducing the amount of sample that receives excess dose (Fig. 1B–C).

Fig. 1. Optimization of a montage tiling strategy.

Fig. 1.

Open in a new tab

(A) At each tilt angle, a hexagonally-packed set of circular tiles is imaged. Applying a global (B) rotation or (C) translation to the tiles between tilt angles changes which regions of the specimen lie in an overlap region to more uniformly spread the dose. To systematically introduce translational offsets between tilt angles, the global displacements applied to the tiles between tilt angles followed one of three spiral patterns: (D) an Archimedean spiral, (E) a sunflower spiral, or (F) a “snowflake” spiral. The positions of the central tile are indicated by black dots and spiral outwards during the course of the tilt-series. For the snowflake spiral, the pattern is repeated starting from the center if the outermost position is reached before the final tilt angle. One of the tunable parameters for all three spiral patterns is the maximum translation of the central tile (t_max); a second parameter for the Archimedean and snowflake patterns is the number of revolutions (n_rev or n_steps). (G) The dose distributions received by voxels of a discretized specimen during a simulated tilt-series are compared for four tiling strategies. The spatial distribution of the dose is mapped on the specimen (upper). Each tiling strategy was scored by the variance (σ2) of the dose distribution, which is noted at the upper right of each histogram (lower). For the “no offsets” strategy, the hexagonally-packed tiles were rotated by 10° relative to the plane of the detector but no offsets were applied to tile positions between tilt angles. For the “rotations only” strategy, a 20° clockwise rotation was applied to all tiles between tilt angles. For the “translations only” strategy, tile positions were translated along an Archimedean spiral with three revolutions and a maximum translation of 80% of the beam radius for each tile. The best-ranked strategy (right) applies both of these translational and rotational offsets to efficiently distribute the dose.

Since a priori it is unclear which combination of offsets would most efficiently distribute dose, we used simulations to characterize hundreds of different tiling strategies. These simulations used a right-handed coordinate system with the detector oriented in the xy-plane and the incoming electron beam directed along the z-axis (Fig. S1). A rectangular specimen with a width of 3,420 nm and depth of 400 nm was discretized into 4 nm cubic voxels and tilted about the x-axis following a standard dose-symmetric tilt-scheme (Hagen et al., 2017). At each tilt angle, a 1 μm diameter beam was used to illuminate a set of hexagonally-packed circular tiles. Even with fringe-free illumination, residual fringes were observed to affect up to 2% of the outer edge of each tile (see below). To prevent gaps in the montage after discarding this corrupted region, the overlap between tiles was increased relative to optimal hexagonal packing such that a small fraction of voxels was exposed up to three times at any tilt angle. We then computed the accumulated dose received by each voxel of the specimen under tiling strategies that differed in the translational and rotational offsets applied between tilt angles.

To systematically introduce translational offsets, we examined three basic spiral patterns: an Archimedean spiral, in which adjacent points are equidistant along the curve of the spiral (Fig. 1D); a sunflower or Fibonacci spiral, in which points are distributed in concentric shells of equal area, and successive points are placed in the largest angular gap between previous points (Swinbank and James Purser, 2006) (Fig. 1E); and a “snowflake” spiral, in which points are positioned on the vertices of concentric hexagons that mirror the 6-fold symmetry of the packed circular tiles (Fig. 1F). The positions of all tiles were uniformly shifted between successive tilt angles to follow the path of the spiral while maintaining hexagonal packing at each tilt angle. For all three spiral types, one of the adjustable parameters was the maximum translation permitted for each tile across the entire tilt-series (Fig. S2A); this was capped at a distance of one beam radius to ensure that all but the outermost tiles remained fully in the field of view. The second translational parameter was the number of revolutions or radial steps respectively for the Archimedean spiral and snowflake pattern (Fig. S2B). For the sunflower pattern, positions were dictated exclusively by the number of points and maximum translation. A third translational parameter was an optional scaling of the x-axis displacements by 1/cosα, where α is the tilt angle (Fig. S2C). This scaling mimicked the y-axis elongation of circular tiles into ellipses at high tilt angles (Fig. S1), with the intent of preserving efficient circular symmetry.

In addition to translational offsets, rotational offsets were systematically introduced by varying three parameters. The first parameter was the starting angle, which dictated the initial orientation of the hexagonal array of circular tiles in the plane of the detector. The second parameter was the rotational step size, which determined the magnitude of the global rotation applied to the hexagonally-packed tiles between each tilt angle. Third, this global rotation was either applied continuously or in an alternating fashion. For the continuous scheme, the hexagonally-arranged tiles were rotated by the same amount and in the same direction between each tilt angle. For the alternating scheme, global counterclockwise and clockwise rotations of the same rotation step size were applied between successive tilt angles.

In total we simulated 576 snowflake, 546 spiral, and 286 sunflower patterns by systematically varying the parameters described above. Each pattern was scored by the variance of the distributed dose, with superior patterns characterized by low variance (Fig. 1G). Across all spiral types, we found that translational offsets were more critical than rotational offsets to uniformly spread the dose throughout the specimen (Fig. 1G, S3). Although the top-ranked variant for the (Archimedean) spiral, sunflower, and snowflake patterns achieved similar scores, we found that variants of the first consistently scored well for translational offsets of the same magnitude (Fig. S3). All variants were also observed to perform similarly across different data collection schemes (Fig. S4). We thus chose the best-performing spiral variant for experimental data collection. This pattern was characterized by a maximum tile translation of 80% of the beam radius, 3 revolutions, no scaling of the x-axis translational component, a starting offset angle of 10° of the hexagonally-arranged tiles in the plane of the detector, and continuous rotations of 20° between tilt angles (Fig. 1G, right). These offsets reduced the variance of the distributed dose by 20-fold compared to the corresponding pattern without any offsets (σ2 = 0.003 versus σ2 = 0.063).

3. Montage data collection and processing

3.1. Sample preparation

HeLa cells (cell line 25) were cultured in DMEM without phenol red (Gibco) supplemented with 10% FBS, 100 U/mL penicillin, and 100 μg/mL streptomycin at 37 °C with 5% CO2. Cells were plated at a density of 2 × 105 cells/mL on laminin-coated 200-mesh gold R3.5/1 London finder Quantifoil grids (Quantifoil Micro Tools) and incubated for ~12 h. For a subset of grids, 3 μL of a colloidal solution of 20 nm gold fiducials (Sigma-Aldrich) pre-treated with bovine serum albumin were applied immediately before vitrification. The grids were then plunge-frozen in a mixture of liquid ethane/propane using an FEI Vitrobot Mark IV (Iancu et al., 2006) and stored in liquid nitrogen at <−150 °C.

3.2. Montage tilt-series collection

Montage tilt-series were collected on a Titan Krios G3i (Thermo Fisher Scientific) equipped with fringe-free illumination (Konings et al., 2019), a Gatan imaging filter, and a K3 Summit direct electron detector (Gatan). Data acquisition was performed using SerialEM in electron-counting mode (Mastronarde, 2003) by providing the software a custom script (see Code Availability). Each tile was acquired using a circular beam of diameter 1.08 μm, such that the beam spanned the short edge of the detector at a pixel size of 2.65 Å. A beam-centering step was performed after collecting each tile to reduce drift without applying additional exposure. At each tilt angle, 37 tiles were acquired: one central tile, with three surrounding rings of hexagonally-packed tiles. The beam coordinates were updated using SerialEM’s image shift function to follow a spiral pattern over the course of the tilt-series as described in Section 2. Once the full complement of 37 tiles was collected at a particular tilt angle, the stage was rotated to the next tilt angle following a grouped dose-symmetric tilt-scheme with a group size of 6°, 2° increments between tilt images, and a tilt-range of ± 60° (Chreifi et al., 2021). Applying the above data collection scheme at a constant defocus was predicted to yield a ~8 μm defocus gradient at the highest tilt angles (Fig. 2A, left). To avoid this spread, we adjusted the defocus with which each tile was collected based on that tile’s estimated z-height in the microscope. Defocus values ranged from −5 to −11 μm for different tilt-series, and a total dose of 60–106 e−/Å2 wasmontage dataset took 4 h to collect, yielding 100 GB of raw data.

Fig. 2. CTF estimation reveals a stable defocus throughout the tilt-series and Thon rings to better than 10 Å in the untilted stitched images.

Fig. 2.

Open in a new tab

(A) Defocus values were predicted based on the tiles’ estimated heights in the microscope and are shown as a function of tilt angle (left). By contrast, the per-tile defocus values estimated by CTFFIND4 showed a relatively stable defocus gradient, as plotted for two representative tilt-series (middle and right). This was accomplished by performing an autofocusing step prior to collecting each tile to compensate for the predicted defocus gradient across each tilt angle. (B) The 2d experimental spectrum (upper) and rotationally-averaged 1d CTF fits (lower) are shown for representative tiles (left) or CTF-uncorrected stitched projection images (right) at the indicated tilt angle.

3.3. Tile pre-processing

Before stitching tiles into a composite image, Fresnel fringes and non-uniform illumination must be accounted for. Fringe-free illumination (FFI) reduces but does not entirely eliminate Fresnel fringes during image formation (Konings et al., 2019; Weis and Hagen, 2020). Residual fringes are particularly evident when data are collected at a defocus and magnification typical for tomography, since current FFI set-ups are optimized for single-particle cryo-EM applications instead. The number of observed fringes depends on both defocus and the signal intensity of the specimen, so removing a fixed fraction of the outer edge of each tile does not adequately eliminate the fringe-contaminated region.

Given this variability, we developed the following heuristic approach to mask residual fringes. A 2-dimensional Gaussian bandpass filter was applied to each tile, using kernel sizes of 2 and 6 nm. In the filtered tile, a high-intensity ring that spanned the Fresnel fringes was observed at the edge of the illuminated region (Fig. S5A). A circle was fit to this ring of pixels using least-squares optimization, and the start of the Fresnel fringes was estimated as the ring’s inner radius. All pixels outside this radius were masked (Fig. S5B–E).

In addition to residual fringes, we observed a consistent reduction in radial intensity by ~15% between the center and edge of each tile (Fig. S5B). Left uncorrected, this would artificially depress the intensity of the overlap regions during stitching. We therefore applied a radial gain correction as follows. At each tilt angle, the tiles’ radial intensity profiles were normalized to a value of 1 in the central region of the tile and merged to generate a single intensity profile for the tilt angle. The resulting radial intensity profile was median-filtered and applied as a gain reference, with linear interpolation used to compute the correction factor at each pixel (Fig. S5B–E).

3.4. Correcting for the contrast transfer function

Despite adjusting the focus on a per-tile basis to avoid an excessive defocus gradient across the tilt-series, some variation in defocus between tiles was expected due to microscope error. We therefore performed a contrast transfer function (CTF) correction on individual tiles. We used CTFFIND4 (version 4.1.13) to estimate each tile’s defocus (Rohou and Grigorieff, 2015) and ctfphaseflip to correct for the CTF (Xiong et al., 2009). The estimated defocus values confirmed that the per-tile focus adjustment compensated for the large changes in z-height due to montage collection and yielded a relatively stable defocus throughout the tilt-series (Fig. 2A). The high-resolution limit of detected Thon rings for most tiles ranged from 10 to 20 Å, with better resolution at lower tilt angles as expected (Fig. 2B, left). To verify that the stitching procedure described below did not degrade resolution, CTFFIND4 was also applied to the CTF-uncorrected stitched images to assess the high-resolution limit of detectable Thon rings (Fig. 2B, right). In the untilted stitched images, CTFFIND4 detected Thon rings to resolutions of 8.2, 9.3, and 12.1 Å for the three montage datasets presented in this work (Fig. S6).

3.5. Tile registration

We developed an image registration workflow tailored for montage cryo-tomograms collected using a circular beam. Current software for processing montage data is designed for resin-embedded specimens and unsuitable for two reasons. First, resin-embedded samples are typically acquired using the full area of the detector, yielding a large and rectangular overlap region between tiles (Mastronarde, 2005). By contrast, we employ a more efficient tiling strategy of hexagonally-tiled circular beams to overcome radiation sensitivity. This scheme results in lemon-shaped overlap regions that change positions and orientations relative to the specimen at each tilt angle (Fig. 1B–C). Second, the algorithms used to perform automated landmark extraction and alignment rely on high-contrast features that typify resin-embedded specimens but are absent in cryo-tomograms even when collected at high defocus (Saalfeld et al., 2012). Although gold fiducials can be added to the sample to provide high-contrast features, a high concentration would be needed to ensure that sufficient markers are present in the overlap regions for use during tile registration. Further, these fiducials reside on the sample’s surface so may experience more severe warping or doming effects than the cellular matter below.

To overcome low contrast, data collected with a pixel size of 2.65 Å were first binned to 10.6 Å. After applying a correction for uneven radial illumination, tiles were bandpass-filtered and masked to remove the unilluminated and Fresnel fringe-corrupted regions. For bandpass-filtering, we found that kernel sizes of roughly 6 and 19 nm enhanced features such as gold beads, membranes, and grid hole boundaries that serve as useful landmarks for image registration. These features were further selected for by thresholding; specifically, only pixels with intensities in the bottom 15th percentile for each tile (belonging to high contrast features) were retained.

Despite the stable optics of modern microscopes, some drift is expected during data collection. To refine the tile positions from the beam coordinates supplied to the microscope, we computed the translational shifts that maximized the normalized cross-correlation between pairs of overlapping tiles, T and T’, at each tilt angle:

argx,ymax∑i,jT(i,j)T′(i+x,j+y)∑i,jT(i,j)2∑i,jT′(i+x,j+y)2 (1)

where the sum is over all pixels at coordinates (i,j) that are in register when T′ is translated by (x,y))while the position of T remains fixed. The initial search was performed in a box of length 30 nm centered on the beam coordinates used during data collection. If the translational shift that maximized the cross-correlation score was located at the edge of this box, the search box was re-centered to this position and another search was performed. This calculation yielded the relative positions for each pair of overlapping tiles.

The full mosaic for each tilt angle was then generated by fixing the central tile at the origin and determining the coordinates of the surrounding six tiles relative to this anchor tile (Fig. 1A). The position of each tile in the surrounding layer was estimated as the mean of all pairwise positions weighted by the cross-correlation scores between the tile of interest and each of its neighbors. Using the cross-correlation coefficients as weights ensured that overlap regions with the highest contrast features contributed most to tile positioning. The coordinates of the tiles in this first ring were then fixed. Tiles in each successive concentric ring were positioned using the same strategy, based on the consensus coordinates from pairwise registrations between neighboring tiles in the ring under consideration and anchor tiles in the previously fixed ring. Once the positions of the tiles in the outermost ring were determined, the full registration procedure was repeated using the optimized positions as the tiles’ starting coordinates. Tile registration was then performed on the unbinned data, using the optimized coordinates as the starting tile positions and a smaller search space.

Tiles were then stitched based on these optimized positions to generate a mosaic (Fig. 3, inset). For pixels lying in the overlap regions, the intensity values were selected from the tile whose center was nearest to the pixel. Continuous cellular features were observed in the overlap regions of the montage projection images, indicating minimal radiation damage and robust stitching (Figs. 3 and S6). Occasionally the combination of beam shift error and masking resulted in small gaps between tiles; these missing pixels were filled by randomly sampling intensity values in the surrounding region to prevent holes in the stitch. Each montage projection image was then shifted to compensate for the spiraling translational offsets applied between tilt angles, cropped to the maximal region imaged at all angles, and stacked to generate a tilt-series. The size of each tilt-series was approximately 33 GB.

Fig. 3. Continuity of cellular features in the overlap regions indicates successful stitching.

Fig. 3.

Open in a new tab

The montaged projection image at 0° from a representative tilt-series is shown in the upper left inset. The region boxed in red is visualized at higher detail in the main image, with the boundaries of the circular tiles drawn in black. The clear and continuous membranous features visible in the overlap regions between adjacent tiles suggests both successful stitching and minimal radiation damage despite the extra dose. The diameter of each circular tile is 1.08 μm.

3.6. Tomogram reconstruction

Tomograms were reconstructed from the montage tilt-series using the EMAN2 (version 2.3) software package, with the projection model generated by patch-tracking (Chen et al., 2019; Tang et al., 2007). The tilt-series data were binned during reconstruction to a pixel size of 7.95 Å to improve contrast. Tomograms were visualized and movies generated using the 3dmod module from IMOD (version 4.10.3) (Kremer et al., 1996).

4. Example montage cryotomograms

For proof-of-principle, we collected montage tilt-series from the thin edge of HeLa cells and reconstructed them into tomograms. During data processing, Thon rings to better than 15 Å were frequently observed on individual tiles (Fig. 2B) and detected to better than 10 Å in the untilted stitched projection images for two of the three tomograms (Fig. S6). Distinct cellular features that spanned the overlap regions were visible throughout the tilt-series (Figs. 3 and S7). These observations suggested that radiation damage was not severe despite the additional dose required for stitching. Consistent with this, there was no evidence of bubbling, a hallmark of radiation damage, or unevenly distributed dose in the reconstructed volumes, which were visually comparable to cryotomograms collected by standard data collection protocols. Inspection of the montage tomograms revealed rich and diverse cellular structures, including microtubules, multilamellar vesicles, mitochondria with calcium granules (Wolf et al., 2017), actin bundles, and ribosomes (Figs. 4–6 and S8, Movies S1–6).

Fig. 4. Diverse cellular features are observed in an example cryotomogram reconstructed from montage tilt-series.

Fig. 4.

Open in a new tab

A slice is shown from a representative montage tomogram that spans a 3.3 μm2 field of view.

Fig. 6. Insets from representative montage cryotomograms.

Fig. 6.

Open in a new tab

Close-up views from the cryotomograms presented in Figs. 4 and 5 visualize the mitochondria (upper) and microtubules (lower) in greater detail.

5. Discussion

Here we present a tomographic data collection and processing workflow to acquire montage tilt-series with a small pixel size but a large field of view. We used simulations to determine an acquisition strategy that efficiently distributed additional dose throughout the specimen and developed algorithms to stitch the recorded data into seamless projection images. We then assessed the efficacy of this pipeline by applying it to the thin edge of HeLa cells, yielding tomograms that spanned a 3.3 μm2 field of view with a pixel size of 7.95 Å. The pixel size of the montage tilt-series, which would determine the Nyquist frequency of a subtomogram average, was 2.65 Å. The observation of Thon rings to better than 10 Å in the untilted stitched images indicated that the montage tilt-series retained signal to a resolution typical of cryo-ET, despite the extra dose required for stitching that historically has been considered prohibitive to this method’s success.

These results are a pioneering demonstration of montage cryo-ET that we hope will motivate further development of this technique. On the data acquisition side, we anticipate that the next phase will involve expanding the spiral pattern to encompass hundreds rather than just tens of tiles. This would dramatically expand the field of view, which in principle is only limited by aberrations associated with extreme beam shifts and, for FIB-milled specimens, the accessible areas of lamellae. Such massive montages would in turn require the development of new software to optimize alignment and reconstruction. We anticipate that the optimal strategy will involve cropping out subvolumes from the initial reconstruction and then using these as fiducials to refine the tile alignments and estimated defocus, while also modeling the global warping of the sample that occurs during data collection. This approach would be an extension of the per-particle, per-tilt refinement employed by some subtomogram averaging packages like emClarity and EMAN2 (Chen et al., 2019; Himes and Zhang, 2018; Ni et al., 2022), and similar to the multi-particle refinement scheme used by the software tool M to model spatial deformations of the specimen (Tegunov et al., 2021). Extending these algorithms to montage data would not only maximize the signal extracted from the small montages shown in this work, but also be invaluable for much larger montages, where warping and doming are more severe.

Once such software is available to maximally exploit the montage technique, a pressing question will be whether montage tomograms offer more information than the corresponding single-exposure tomogram acquired at lower magnification. We anticipate that the montage technique will retain higher resolution information, in part because one of the principal resolution-limiting factors in cryoET is the precision to which the defocus can be estimated for CTF correction. In principle, montage tomography should permit more precise defocus determination because the geometrical relationship between all tiles is known a priori, such that CTF corrections can leverage all the available data, which far exceeds the amount of data acquired for a single-exposure tomogram. The disadvantage of montage cryoET is the additional exposure received by the overlap regions, but these are minimized by an efficient tiling scheme and can still be integrated during reconstruction with appropriate dose-weighting. We observed that only a small fraction (~4%) of the imaged sample must be discarded due to corruption by residual Fresnel fringes, but fine-tuning of the FFI set-up may further reduce this. We expect that the benefits provided by a more precise defocus estimation will outweigh the cost of this small amount of wasted dose, though further software development is needed.

We also anticipate that montage tomography could prove particularly useful for cellular lamellae. The FIB-milling required to produce these samples is challenging and time-consuming, so maximizing the yield during data collection is a critical concern (Mahamid et al., 2016; Marko et al., 2007; Rigort et al., 2012). By imaging a large field of view, this technique also increases the chances of capturing transient or infrequent cellular events. Finally, data collected by this method will be a valuable resource for visual proteomics, which seeks to build atlases of cellular structure at molecular resolution (Moebel et al., 2021; Nickell et al., 2006). As technical advances continue to improve the high-resolution limit of cryo-ET, montage tomography offers a way to provide more cellular context while retaining detailed views of macromolecules in situ.

Supplementary Material

Supplementary Material
MMC6
Download video file (13.8MB, mp4)
MMC7
Download video file (18.7MB, mp4)
MMC4
Download video file (22.1MB, mp4)
MMC3
Download video file (20.1MB, mp4)
MMC2
Download video file (16.8MB, mp4)
MMC1
MMC5
Download video file (34.3MB, mp4)

Fig. 5. Representative montage cryotomogram from the thin edge of a HeLa cell.

Fig. 5.

Open in a new tab

A tomographic slice spanning a 3.3 μm2 field of view is visualized, with cellular features of interest annotated.

Acknowledgements

Data collection and analysis were respectively performed at the Beckman Institute Resource Center for Transmission Electron Microscopy and the Resnick High Performance Computing Center at Caltech. We thank Wei Zhao for preliminary samples, David Mastronarde for valuable discussions, and Tom Morrell for help with the Caltech Data Repository. A.P. is The Mark Foundation for Cancer Research Fellow of the Damon Runyon Cancer Research Foundation (DRG 2361-19). This work was supported by NIH grant AI127401 to G.J.J.

Abbreviations:

cryo-EM

cryo-electron microscopy

cryo-ET

cryo-electron tomography

CTF

contrast transfer function

FFI

fringe-free illumination

FIB

focused ion beam

Footnotes

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

CRediT authorship contribution statement

Ariana Peck: Methodology, Software, Writing – original draft, Writing – review & editing. Stephen D. Carter: Conceptualization, Methodology, Investigation. Huanghao Mai: Methodology, Software. Songye Chen: Software, Methodology, Investigation. Alister Burt: Formal analysis, Writing – review & editing. Grant J. Jensen: Conceptualization, Writing – original draft, Writing – review & editing, Funding acquisition, Supervision.

Appendix A. Supplementary material

Supplementary data to this article can be found online at https://doi.org/10.1016/j.jsb.2022.107860.

Data and code availability.

Raw data are available at the Caltech Data Repository (https://data.caltech.edu) under accession IDs 2096, 2099, and 2103. The processed tilt-series binned to 5.3 Å can be found in the Caltech Electron Tomography Database (https://etdb.caltech.edu/). The code developed for data collection and processing is available at https://github.com/apeck12/montage.

References

  1. Baker LA, Rubinstein JL, 2010. Radiation damage in electron cryomicroscopy. Methods Enzymol. 481, 371–388. 10.1016/S0076-6879(10)81015-8. [DOI] [PubMed] [Google Scholar]
  2. Bammes BE, Rochat RH, Jakana J, Chen D-H, Chiu W, 2012. Direct electron detection yields cryo-EM reconstructions at resolutions beyond 3/4 Nyquist frequency. J. Struct. Biol. 177, 589–601. 10.1016/j.jsb.2012.01.008. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Bartesaghi A, Aguerrebere C, Falconieri V, Banerjee S, Earl LA, Zhu X, Grigorieff N, Milne JLS, Sapiro G, Wu X, Subramaniam S, 2018. Atomic Resolution Cryo-EM Structure of β-Galactosidase. Struct. Lond. Engl. 1993 (26), 848–856.e3. 10.1016/j.str.2018.04.004. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Böhning J, Bharat TAM, 2021. Towards high-throughput in situ structural biology using electron cryotomography. Prog. Biophys. Mol. Biol. 160, 97–103. 10.1016/j.pbiomolbio.2020.05.010. [DOI] [PubMed] [Google Scholar]
  5. Chen M, Bell JM, Shi X, Sun SY, Wang Z, Ludtke SJ, 2019. A complete data processing workflow for cryo-ET and subtomogram averaging. Nat. Methods 16, 1161–1168. 10.1038/s41592-019-0591-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Chreifi G, Chen S, Jensen GJ, 2021. Rapid tilt-series method for cryo-electron tomography: Characterizing stage behavior during FISE acquisition. J. Struct. Biol. 213 (2), 107716. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Chreifi G, Chen S, Metskas LA, Kaplan M, Jensen GJ, 2019. Rapid tilt-series acquisition for electron cryotomography. J. Struct. Biol. 205, 163–169. 10.1016/j.jsb.2018.12.008. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Glaeser RM, 2016. Specimen Behavior in the Electron Beam. Methods Enzymol. 579, 19–50. 10.1016/bs.mie.2016.04.010. [DOI] [PubMed] [Google Scholar]
  9. Hagen WJH, Wan W, Briggs JAG, 2017. Implementation of a cryo-electron tomography tilt-scheme optimized for high resolution subtomogram averaging. J. Struct. Biol. Electron Tomography 197, 191–198. 10.1016/j.jsb.2016.06.007. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Hattne J, Shi D, Glynn C, Zee C-T, Gallagher-Jones M, Martynowycz MW, Rodriguez JA, Gonen T, 2018. Analysis of Global and Site-Specific Radiation Damage in Cryo-EM. Struct. Lond. Engl. 1993 (26), 759–766.e4. 10.1016/j.str.2018.03.021. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Himes BA, Zhang P, 2018. emClarity: software for high-resolution cryo-electron tomography and subtomogram averaging. Nat. Methods 15, 955–961. 10.1038/s41592-018-0167-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Iancu CV, Tivol WF, Schooler JB, Dias DP, Henderson GP, Murphy GE, Wright ER, Li Z, Yu Z, Briegel A, Gan L, He Y, Jensen GJ, 2006. Electron cryotomography sample preparation using the Vitrobot. Nat. Protoc. 1, 2813–2819. 10.1038/nprot.2006.432. [DOI] [PubMed] [Google Scholar]
  13. Kershner R, 1939. The Number of Circles Covering a Set. Am. J. Math. 61, 665–671. 10.2307/2371320. [DOI] [Google Scholar]
  14. Koning RI, Zovko S, Bárcena M, Oostergetel GT, Koerten HK, Galjart N, Koster AJ, Mieke Mommaas A, 2008. Cryo electron tomography of vitrified fibroblasts: microtubule plus ends in situ. J. Struct. Biol. 161, 459–468. 10.1016/j.jsb.2007.08.011. [DOI] [PubMed] [Google Scholar]
  15. Konings S, Kuijper M, Keizer J, Grollios F, Spanjer T, Tiemeijer P, 2019. Advances in Single Particle Analysis Data Acquisition. Microsc. Microanal. 25, 1012–1013. 10.1017/S1431927619005798. [DOI] [Google Scholar]
  16. Kremer JR, Mastronarde DN, McIntosh JR, 1996. Computer Visualization of Three-Dimensional Image Data Using IMOD. J. Struct. Biol. 116, 71–76. 10.1006/jsbi.1996.0013. [DOI] [PubMed] [Google Scholar]
  17. Kuijper M, van Hoften G, Janssen B, Geurink R, Carlo SD, Vos M, van Duinen G, van Haeringen B, Storms M, 2015. FEI’s direct electron detector developments: Embarking on a revolution in cryo-TEM. J. Struct. Biol. 192, 179–187. 10.1016/j.jsb.2015.09.014. [DOI] [PubMed] [Google Scholar]
  18. Le Gros MA, McDermott G, Cinquin BP, Smith EA, Do M, Chao WL, Naulleau PP, Larabell CA, 2014. Biological soft X-ray tomography on beamline 2.1 at the Advanced Light Source. J. Synchrotron Radiat. 21, 1370–1377. 10.1107/S1600577514015033. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Lippens S, Kremer A, Borghgraef P, Guérin CJ, 2019. Chapter 4 - Serial block face-scanning electron microscopy for volume electron microscopy. In: Müller-Reichert T, Pigino G (Eds.), Three-Dimensional Electron Microscopy, Methods in Cell Biology. Academic Press, pp. 69–85. 10.1016/bs.mcb.2019.04.002. [DOI] [PubMed] [Google Scholar]
  20. Loconte V, Singla J, Li A, Chen J-H, Ekman A, McDermott G, Sali A, Le Gros M, White KL, Larabell CA, 2022. Soft X-ray tomography to map and quantify organelle interactions at the mesoscale. Structure 30 (4), 510–521.e3. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Mahamid J, Pfeffer S, Schaffer M, Villa E, Danev R, Cuellar LK, Förster F, Hyman AA, Plitzko JM, Baumeister W, 2016. Visualizing the molecular sociology at the HeLa cell nuclear periphery. Science 351, 969–972. 10.1126/science.aad8857. [DOI] [PubMed] [Google Scholar]
  22. Marko M, Hsieh C, Schalek R, Frank J, Mannella C, 2007. Focused-ion-beam thinning of frozen-hydrated biological specimens for cryo-electron microscopy. Nat. Methods 4, 215–217. 10.1038/nmeth1014. [DOI] [PubMed] [Google Scholar]
  23. Mastronarde DN, 2005. Automated electron microscope tomography using robust prediction of specimen movements. J. Struct. Biol. 152, 36–51. 10.1016/j.jsb.2005.07.007. [DOI] [PubMed] [Google Scholar]
  24. Mastronarde DN, 2003. SerialEM: A Program for Automated Tilt Series Acquisition on Tecnai Microscopes Using Prediction of Specimen Position. Microsc. Microanal. 9, 1182–1183. 10.1017/S1431927603445911. [DOI] [Google Scholar]
  25. McMullan G, Faruqi AR, Henderson R, 2016. Direct Electron Detectors. Methods Enzymol. 579, 1–17. 10.1016/bs.mie.2016.05.056. [DOI] [PubMed] [Google Scholar]
  26. Milazzo A-C, Cheng A, Moeller A, Lyumkis D, Jacovetty E, Polukas J, Ellisman MH, Xuong N-H, Carragher B, Potter CS, 2011. Initial evaluation of a direct detection device detector for single particle cryo-electron microscopy. J. Struct. Biol. 176, 404–408. 10.1016/j.jsb.2011.09.002. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Moebel E, Martinez-Sanchez A, Lamm L, Righetto RD, Wietrzynski W, Albert S, Lariviére D, Fourmentin E, Pfeffer S, Ortiz J, Baumeister W, Peng T, Engel BD, Kervrann C, 2021. Deep learning improves macromolecule identification in 3D cellular cryo-electron tomograms. Nat. Methods 18, 1386–1394. 10.1038/s41592-021-01275-4. [DOI] [PubMed] [Google Scholar]
  28. Ni T, Frosio T, Mendonça L, Sheng Y, Clare D, Himes BA, Zhang P, 2022. High-resolution in situ structure determination by cryo-electron tomography and subtomogram averaging using emClarity. Nat. Protoc. 17, 421–444. 10.1038/s41596-021-00648-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Nickell S, Kofler C, Leis AP, Baumeister W, 2006. A visual approach to proteomics. Nat. Rev. Mol. Cell Biol. 7, 225–230. 10.1038/nrm1861. [DOI] [PubMed] [Google Scholar]
  30. Noske AB, Costin AJ, Morgan GP, Marsh BJ, 2008. Expedited approaches to whole cell electron tomography and organelle mark-up in situ in high-pressure frozen pancreatic islets. J. Struct. Biol. 161, 298–313. 10.1016/j.jsb.2007.09.015. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Rigort A, Bäuerlein FJB, Villa E, Eibauer M, Laugks T, Baumeister W, Plitzko JM, 2012. Focused ion beam micromachining of eukaryotic cells for cryoelectron tomography. Proc. Natl. Acad. Sci. 109, 4449–4454. 10.1073/pnas.1201333109. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Rog-Zielinska EA, Johnston CM, O’Toole ET, Morphew M, Hoenger A, Kohl P, 2016. Electron tomography of rabbit cardiomyocyte three-dimensional ultrastructure. Prog. Biophys. Mol. Biol., Focussed Issue on Past, Present and Future of Non-murine (Lagomorph) Model Systems as Experimental and Computational Drivers of Progress in Cardiac Biophysics 121, 77–84. 10.1016/j.pbiomolbio.2016.05.005. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Rohou A, Grigorieff N, 2015. CTFFIND4: Fast and accurate defocus estimation from electron micrographs. J. Struct. Biol., Recent Advances in Detector Technologies and Applications for Molecular TEM 192, 216–221. 10.1016/j.jsb.2015.08.008. [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Saalfeld S, Fetter R, Cardona A, Tomancak P, 2012. Elastic volume reconstruction from series of ultra-thin microscopy sections. Nat. Methods 9, 717–720. 10.1038/nmeth.2072. [DOI] [PubMed] [Google Scholar]
  35. Swinbank R, James Purser R, 2006. Fibonacci grids: A novel approach to global modelling. Q. J. R. Meteorol. Soc. 132, 1769–1793. 10.1256/qj.05.227. [DOI] [Google Scholar]
  36. Tang G, Peng L, Baldwin PR, Mann DS, Jiang W, Rees I, Ludtke SJ, 2007. EMAN2: An extensible image processing suite for electron microscopy. J. Struct. Biol. Software Tools Macromol. Microscopy 157, 38–46. 10.1016/j.jsb.2006.05.009. [DOI] [PubMed] [Google Scholar]
  37. Tegunov D, Xue L, Dienemann C, Cramer P, Mahamid J, 2021. Multi-particle cryo-EM refinement with M visualizes ribosome-antibiotic complex at 3.5 Å in cells. Nat. Methods 18, 186–193. 10.1038/s41592-020-01054-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Tocheva EI, Li Z, Jensen GJ, 2010. Electron Cryotomography. Cold Spring Harb. Perspect. Biol. 2 (6), a003442. [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Weis F, Hagen WJH, 2020. Combining high throughput and high quality for cryo-electron microscopy data collection. Acta Crystallogr. Sect. Struct. Biol. 76, 724–728. 10.1107/S2059798320008347. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Wolf SG, Mutsafi Y, Dadosh T, Ilani T, Lansky Z, Horowitz B, Rubin S, Elbaum M, Fass D, 2017. 3D visualization of mitochondrial solid-phase calcium stores in whole cells. eLife 6, e29929. 10.7554/eLife.29929. [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Xiong Q, Morphew MK, Schwartz CL, Hoenger AH, Mastronarde DN, 2009. CTF determination and correction for low dose tomographic tilt series. J. Struct. Biol. 168, 378–387. 10.1016/j.jsb.2009.08.016. [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Xu CS, Hayworth KJ, Lu Z, Grob P, Hassan AM, García-Cerdán JG, Niyogi KK, Nogales E, Weinberg RJ, Hess HF, 2017. Enhanced FIB-SEM systems for large-volume 3D imaging. eLife 6, e25916. 10.7554/eLife.25916. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Material
NIHMS1871136-supplement-Supplementary_Material.pdf (13.8MB, pdf)
MMC6
Download video file (13.8MB, mp4)
MMC7
Download video file (18.7MB, mp4)
MMC4
Download video file (22.1MB, mp4)
MMC3
Download video file (20.1MB, mp4)
MMC2
Download video file (16.8MB, mp4)
MMC1
NIHMS1871136-supplement-MMC1.pdf (13.8MB, pdf)
MMC5
Download video file (34.3MB, mp4)

Data Availability Statement

Raw data are available at the Caltech Data Repository (https://data.caltech.edu) under accession IDs 2096, 2099, and 2103. The processed tilt-series binned to 5.3 Å can be found in the Caltech Electron Tomography Database (https://etdb.caltech.edu/). The code developed for data collection and processing is available at https://github.com/apeck12/montage.

RESOURCES