Cryo-ET

The core data collection strategy where the vitrified specimen is incrementally rotated (typically ±60°) around a single axis in the electron microscope. A 2D projection image is recorded at each tilt angle, generating a tilt-series. Due to the slab geometry of the sample, the effective path length increases at higher tilt angles, limiting the maximum achievable tilt and creating the missing wedge of information in Fourier space. Modern workflows use dose-symmetric tilt schemes that start at low tilts and move outward to preserve high-resolution information in the most critical views before radiation damage accumulates.

The computational process of converting a tilt-series into a 3D volume called a tomogram. The standard method is weighted back-projection (WBP) , which smears each 2D projection back into a 3D volume at its corresponding tilt angle. More sophisticated iterative methods like SIRT (Simultaneous Iterative Reconstruction Technique) refine the volume by comparing re-projections of the current estimate with the original data. The resulting tomogram represents the 3D Coulomb potential density of the specimen, with a typical resolution of 2-10 nm, sufficient to visualize macromolecular complexes in their cellular context.

A computational method that extracts 3D sub-volumes (subtomograms) containing copies of the same macromolecular complex from multiple tomograms, aligns them in 3D, and averages them to achieve near-atomic resolution. This overcomes the low signal-to-noise ratio of individual tomograms. Key steps include: 3D particle picking to identify complexes, 3D alignment to correct for orientation and position, and missing wedge compensation to handle anisotropic resolution. This technique has resolved ribosome structures inside cells to better than 4 Å, revealing drug-binding interactions in situ.

Cryo-ET's defining advantage is the ability to study macromolecules in their native cellular environment without purification or lysis. This preserves transient interactions, spatial relationships, and the structural consequences of crowding. Researchers can directly visualize: the nuclear pore complex embedded in the nuclear envelope, actin filaments and their branching networks, ribosomes translating on mRNA, and viral glycoproteins arrayed on an enveloped virion. This bridges the gap between high-resolution in vitro structures and lower-resolution light microscopy of live cells.

A sample preparation technique essential for imaging thick cellular specimens. A gallium ion beam is used to ablate material from the top and bottom of a vitrified cell, creating a thin lamella (100-300 nm thick) that is transparent to the electron beam. This is performed under cryogenic conditions to maintain the vitreous state. The workflow involves: fluorescence microscopy to locate target cells, cryo-FIB/SEM to precisely mill lamellae, and cryo-ET on the resulting thin sections. This enables structural analysis of previously inaccessible regions deep inside cells.

A fundamental limitation of single-axis tilt tomography. Because the specimen cannot be tilted to 90°, a wedge-shaped region of Fourier space remains unsampled. This results in anisotropic resolution: the tomogram has better resolution in the direction perpendicular to the tilt axis and poorer resolution along the beam direction (Z-axis). Artifacts include elongation of features along Z and fan-shaped distortions. Computational methods like constrained cross-correlation during subtomogram averaging and deep learning-based denoising (e.g., IsoNet) are used to mitigate these effects and restore isotropic information.

A computational method analogous to single-particle analysis but applied to 3D sub-volumes (subtomograms) extracted from cryo-ET tomograms. By aligning and averaging thousands of subtomograms containing identical macromolecular complexes, the signal-to-noise ratio is dramatically improved, enabling in situ structure determination at resolutions approaching 3-4 Å. This technique resolves structures within their native cellular context without purification.

Computational methods to compensate for the wedge-shaped region of missing Fourier space information inherent in tomographic tilt-series data. Due to physical constraints limiting the tilt range (typically ±60°), a wedge of information remains unsampled, causing anisotropic resolution and elongation artifacts in the reconstruction direction. Correction algorithms use iterative reconstruction, prior information, or deep learning to mitigate these distortions.

The critical preprocessing step where 2D projection images acquired at different tilt angles are mutually aligned to sub-pixel accuracy before tomographic reconstruction. Fiducial marker-based alignment uses gold nanoparticles as reference points, while patch-based tracking corrects for local, non-linear deformations in the specimen caused by beam-induced motion during acquisition.

A neural network architecture trained to reconstruct clean images from noisy inputs, applied in cryo-ET for tomogram restoration. Techniques like Noise2Noise train on pairs of independently noisy images without requiring clean ground truth. These models suppress noise while preserving high-resolution structural features, dramatically improving particle picking accuracy and subtomogram alignment in low-dose imaging conditions.

An acquisition strategy that starts at low tilt angles and alternates to higher positive and negative tilts, ensuring that the most critical low-frequency information is collected before significant radiation damage accumulates. This scheme, implemented in SerialEM, optimizes the use of the limited electron dose budget (typically 60-120 e⁻/Ų) across the tilt series to preserve high-resolution features.

A computational search method that cross-correlates a 3D reference template against a tomogram to locate macromolecular complexes. GPU-accelerated implementations like PyTOM enable rapid scanning of large cellular volumes. While powerful for identifying known structures, template matching is computationally intensive and prone to false positives in crowded cellular environments, increasingly being augmented or replaced by deep learning-based particle picking.