What is Subtomogram Averaging?

Turning crowded cryo-ET data into structural insight with subtomogram averaging

Cryo-electron tomography (cryo-ET) is capable of capturing detailed, three-dimensional views of vitrified biological samples, supporting a broad range of life sciences research. Subtomogram averaging helps refine that crowded data for clearer structural insight by finding repeated particles, extracting them as small 3D volumes, and then aligning and averaging them together so that the shared signals becomes easier to see (Figure 1).

Subtomogram averaging is especially useful for recurring cellular assemblies such as ribosomes, proteasomes, microtubules, and other macromolecular structures, such as nuclear pore complexes, all of which may appear many times in a single dataset. By grouping those copies, aligning them to a common reference frame, and averaging them, subtomogram averaging is able to strengthen shared structural features while reducing random noise (Figure 1).

Schematic overview of a subtomogram averaging workflow in cryo-electron tomography (cryo-ET), illustrating the identification, extraction, alignment, and averaging of repeating molecular structures from tomographic data to generate a high-resolution 3D reconstruction. — *Figure 1. Overview of a typical subtomogram averaging workflow, showing the processing of repeat structures in a cryo-ET dataset. Figure reproduced under CC BY 4.0 from Pyle and Zanetti (2021).*

How does a tomogram become an averaged structure?

Subtomogram averaging begins with the identification of particles and structures of interest within the crowded cellular tomograms. This can be done with segmentation, which can help identify components such as membranes or microtubules, as well as 3D template matching, which can find features that match a selected reference structure. During the template matching process, the reference is low-pass filtered and rescaled to the pixel size, then searched exhaustively across positions and orientations to generate a score map. Peaks in that map point to candidate particle positions and orientations.

Once particles are located using either method, they are separated out as a subtomogram. Those subtomograms are then rotated and translated into the same orientation and averaged. If these particles are the same structure, common features become stronger while noise is reduced.

How can template matching help answer spatial questions?

Template matching is useful not only as a part of subtomogram averaging but also as a general way to localize a structure within a tomogram. It makes it possible to ask where a particle is located and how it is oriented, even when the target is sparse or when there is not enough data to run a full subtomogram averaging pipeline.

Schematic illustration of template matching for visual proteomics, showing (A) a cryo-electron tomogram, (B) cross-correlation with a library of reference structures to identify molecular targets, and (C) 3D localization maps generated from z-score data to visualize protein distributions and spatial interactions. — **Figure 2.** Template matching for visual proteomics. A tomogram (A) is cross-correlated independently with each template in a library of reference structures (B) to identify points with high constrained-cross-correlation values. From the z-score maps, 3D localization maps (C) are generated for visualization and analysis of spatial interactions. Figure reproduced under CC BY 4.0 from Cruz-León et al. (2024).

What is the importance of particle orientation in subtomogram averaging?

Cryo-ET reconstruction is limited by the so-called missing wedge: the range of angles at which information is sampled from the specimen is limited, meaning that some structural information is not collected. It can result in structures that are less well resolved or visible in certain directions. As each particle is usually oriented differently within the sample, their orientation can play an important role in these variations and produce a more complex result. The effect of the missing wedge is illustrated in Figure 3A, where progressively reducing the tilt range leads to increasing reconstruction artifacts and loss of horizontally oriented features; Figure 3B shows this effect in a biological sample.

Illustration of the missing-wedge effect in cryo-electron tomography (cryo-ET), showing (A) simulated reconstructions with decreasing tilt ranges from ±90° to ±45°, resulting in increased artifacts and loss of horizontally oriented features, and (B) an experimental reconstruction of a frozen E. coli cell demonstrating reduced visibility of cellular structures due to limited angular sampling. — **Figure 3.** The missing-wedge effect in cryo-ET reconstruction. (A) Simulation showing the effect of progressively reducing the angular range used for reconstruction from ±90° to ±45°. As the tilt range decreases, reconstruction artifacts increase and horizontally oriented features are increasingly lost. (B) Experimental example from a frozen E. coli cell showing how limited angular sampling can affect the visibility of cellular structures. Reproduced and adapted from Palmer and Löwe (2014) under CC BY 3.0.

What can subtomogram averaging reveal that a single tomogram may not?

The value of subtomogram averaging is not just sharper density, it is the ability to connect structure with context. Once particles are averaged, the result can be placed back into the original tomogram to probe spatial questions: Where is the complex located? How is it oriented? How does it sit relative to nearby cellular structures? Template matching can also be used when these questions do not require a full subtomogram averaging pipeline.

For virology, cell and membrane biology, as well as other in situ applications, this contextualization is especially valuable. These approaches can help reveal recurring molecular assemblies inside cells, show how they are positioned relative to one another, and support a more spatial view of biology. This can also serve as a useful route into visual proteomics, where molecular machines are studied in 3D rather than as isolated particles.

Recent publications show the practical benefit of subtomogram averaging for in situ studies, helping to resolve structures across viral replication organelles, mitochondrial machinery, and membrane-associated coat assemblies. This includes a SARS-CoV-2 double-membrane vesicle pore at 4.2 Å (Huang et al. (2024)), mitochondrial ATP synthase in whole Polytomella cells at 4.2 Å (Dietrich et al. (2024)), and COPII coat assemblies on cargo-containing membranes at 4.1 Å (Pyle et al. (2024)).

How can you optimize a subtomogram averaging workflow?

Subtomogram averaging depends on the quality of the full cryo-ET workflow, and it works best when sample preparation, acquisition, reconstruction, particle detection, and alignment are all matched to the question being asked.

To start, thick samples may need thinning before imaging, using methods such as cryo-focused ion beam milling or cryo-ultramicrotomy. The alignment of the tilt series after data collection is also critical. This can be performed manually or automatically using either fiducial markers or marker-free approaches such as patch tracking or machine learning.

The reconstruction method is no less important. Weighted back projection is commonly used when downstream averaging is the goal, while iterative methods such as simultaneous iterative reconstruction technique (SIRT) and simultaneous algebraic reconstruction technique (SART) can produce sharper, higher-contrast volumes that are useful for visualization, annotation, and segmentation. Denoising can also be used to make particle finding easier, but weighted back-projection is generally preferred because it preserves the original signal more directly.

When is subtomogram averaging the right next step?

Subtomogram averaging is most useful when a sample contains repeated copies of the same structure, and those copies appear in enough orientations to be well aligned and averaged. It has particular value when the biological question is not just “What is here?” but also “How is it organized?” and “What does that organization suggest about function?” With subtomogram averaging, crowded cryo-ET data can turn into a clear view of molecular structure in its cellular setting.

Learn more about cryo-electron tomography and subtomogram averaging >>

Header image shows Chikungunya virus assembly and budding, visualized in situ using subtomogram averaging. Recreated from EMD-26448. Original study by Chmielewski et al. (2022).