Improved cryo-EM reconstruction of sub-50 kDa complexes using 2D template matching

Reviewer #1 (Public review):

Summary:

This paper describes an application of the high-resolution cryo-EM 2D template matching technique to sub-50kDa complexes. The paper describes how density for ligands can be reconstructed without having to process cryo-EM data through the conventional single particle analysis pipelines.

Strengths:

This paper contributes additional data (alongside other papers by the same authors) to convey the message that high-resolution 2D template matching is a powerful alternative for cryo-EM structure determination. The described application to ligand density reconstruction, without the need for extensive refinements, will be of interest to the pharmaceutical industry, where often multiple structures of the same protein in complex with different ligands are solved as part of their drug development pipelines. Improved insights into which particles contribute to the best ligand density are also highly valuable and transferable to other applications of the same technique.

Weaknesses:

Although the convenient visualisation of small molecules bound to protein targets of a known structure would be relevant for the pharmaceutical industry, the evidence described for the claim that this technique "significantly" improves alignment of reconstruction of small complexes is incomplete. The authors are encouraged to better evaluate the effects of model bias on the reconstructed densities in a revised paper.

Reviewer #2 (Public review):

In this manuscript, Zhang et al describe a method for cryo-EM reconstruction of small (sub-50kDa) complexes using 2D template matching. This presents an alternative, complementary path for high-resolution structure determination when there is a prior atomic model for alignment. Importantly, regions of the atomic model can be deleted to avoid bias in reconstructing the structure of these regions, serving as an important mechanism of validation.

The manuscript focuses its analysis on a recently published dataset of the 40kDa kinase complex deposited to EMPIAR. The original processing workflow produced a medium resolution structure of the kinase (GSFSC ~4.3A, though features of the map indicate ~6-7A resolution); at this resolution, the binding pocket and ligand were not resolved in the original published map. With 2DTM, the authors produce a much higher resolution structure, showing clear density for the ATP binding pocket and the bound ATP molecule. With careful curation of the particle images using statistically derived 2DTM p-values, a high-resolution 2DTM structure was reconstructed from just 8k particles (2.6A non-gold standard FSC; ligand Q-score of 0.6), in contrast to the 74k particles from the original publication. This aligns with recent trends that fewer, higher-quality particles can produce a higher-quality structure. The authors perform a detailed analysis of some of the design choices of the method (e.g., p-value cutoff for particle filtering; how large a region of the template to delete).

Overall, the workflow is a conceptually elegant alternative to the traditional bottom-up reconstruction pipeline. The authors demonstrate that the p-values from 2DTM correlations provide a principled way to filter/curate which particle images to extract, and the results are impressive. There are only a few minor recommendations that I could make for improvement.

Reviewer #3 (Public review):

Summary:

Due to the low SNR of cryo-EM micrographs necessitated by radiation damage, determining the structure of proteins smaller than 50 kDa is exceedingly challenging, such that only a handful have been solved to date. This work aims to improve the reconstruction of small proteins in single-particle cryo-EM by using high-resolution 2D template matching, an algorithm previously used to locate and align macromolecules in situ, to align and reconstruct small proteins. This approach uses an existing macromolecular structure, either experimentally determined or predicted by AlphaFold, to simulate a noise-free 3D reference and generates whitened projections, crucially including high-spatial-frequency information, to align particles by the orientation with maximal cross-correlation. They demonstrate the success of this approach by generating a 3D reconstruction from an existing dataset of a 41.3 kDa protein kinase that had previously evaded attempts at high-resolution structure determination. To alleviate concerns that this is purely from template bias, they demonstrate clear density at two regions that were not present in the template: 6 residues in an alpha helix and an ATP in the ligand binding pocket. The latter is particularly important for its implications in determining structures of ligand-bound proteins for drug discovery. Additionally, the authors provide an update to the classic calculation in Henderson 1995 to predict the minimum molecular mass of a protein that can be solved by single-particle cryo-EM.

Strengths:

I am in no doubt that this technique can be used to gain valuable insights into the structures of small proteins, and this is an important advancement for the field. The ability to determine the structure of ligands in a binding site is particularly important, and this paper provides a method of doing that which outperforms traditional single-particle cryo-EM processing workflows.

The claim that using high-spatial frequency information is essential for aligning small proteins is a valuable insight. A recent pre-print published at a similar time to this manuscript used high-resolution information in standard ab-initio reconstruction to generate a high-resolution reconstruction from the same dataset, supporting the claims made in the manuscript.

The theoretical section outlined in the appendix is also theoretically sound. It uses the same logic as Henderson, but applies more up-to-date knowledge, such as incorporating dose-weighting and altering the cross-correlation-based noise estimation. This update is valuable for understanding factors preventing us from reaching the theoretical limit.

Weaknesses:

Given that this technique creates template bias, only parts of the reconstruction not in the template can be trusted, unlike standard single-particle processing, where the independent half-maps from separate, ab initio templates are used to generate a 3D reconstruction. Although, in principle, one could perform the search many times such that every residue has been omitted in at least one search, this will be extremely computationally intensive and was not demonstrated in this manuscript. It is therefore currently only realistically applicable when only a small portion of the sub-50 kDa protein is of interest.

The applicability of this technique to more than a single target was also not demonstrated, and there are concerns that it may not work effectively in many cases. The authors note in the results that "the ATP density was consistently recovered more robustly than nearby residues" and speculate that this may be because misalignments disproportionately blur peripheral residues. Since the region of interest in a structure is not necessarily in the center, this may need further investigation. The implications of this statement may also be unclear to the reader. For example, can this issue be minimized by having the region of interest centered in the simulated volume?

In Figure 3, the authors demonstrate that it is not solely improved particle filtering and a noise-free reference that improves alignment, but that the high spatial frequency information is important. This information is very valuable since it can be applied to other, more standard methods. However, this key figure is not as clear or convincing as it could be. The FSC curves are possibly misleading, since the reduced resolution could be explained by reduced template bias when auto-refining with a map initially low-pass filtered to 10 Å. Moreover, although the helix reconstruction does look slightly better using the 2DTM angles, the improvement in density for ATP in the binding pocket is not clear. A qualitative argument only clear in one out of two cases is not as convincing as a quantitative metric across more examples.