Escrever um comentário

This innovative study from Mäeots et al. introduces Chronobot, a sample preparation platform for time-resolved cryo-EM. The device and accompanying AI-based algorithms that comprise Chronobot enable assessment of a grid’s quality mid-plunge, as well as during low-magnification TEM screening, thus increasing reproducibility and efficient grid preparation. The authors demonstrate the capabilities of the Chronobot platform by resolving previously unknown millisecond intermediates (250 ms bins) of RecA as it carries out homology search functionality, demonstrating the utility of time-resolved cryo-EM for future studies. The authors further enhance data collection efficiency by integrating a deep learning-based cryo-EM ice filter. This combination of feedback-enabled time-resolved grid preparation and smart data collection schemes represents a valuable addition to the field of cryo-EM. To strengthen the manuscript, we encourage the authors to revise the following aspects before publication:

Major points:

1. A large portion of the results section (Line 127-174) is better suited for the introduction or discussion sections, as it doesn’t adhere to the traditional structure of rationalizing selected methods to show actual results. Generally, the manuscript would benefit from restructuring to make it more concise.

2. The authors should introduce the mode of operation earlier than Line 339 and add a panel to Fig. 2 to pictorially describe the sequence of automated operations. 

3. What is the resolution limit for the device compared to previously reported approaches? This should be validated and discussed for future developments.    

4. The first section emphasizes improvements of the new Chronobot and comments on future commitment to making the workflow available to the cryo-EM community. The authors should disclose that this may only be a commercial product. Otherwise, the authors are encouraged to support open-source, community-driven improvements on the current system (e.g. via SerialEM and Leginon), by sharing the component details for academic purposes. Can the ice filter be adopted for use in SerialEM or Leginon, or through an API in EPU?

5. The authors should validate the ice thickness filter in TEM by back-tracking particles from curated and uncurated data to quantify the actual benefit of cutoffs and their effect on reconstruction. Regarding the speed of data collection (line 321), what is the benefit of the new method to curate contaminants over smartEPU (https://assets.thermofisher.com/TFS-Assets/MSD/Datasheets/smart-epu-software-ds0490-en.pdf)

6. The validation of cryo-EM data is missing and needs to be added as Table 1, as described at: https://www.nature.com/documents/nr-tables-cryo-em.doc. The authors should also include local map quality and Q-score.

Minor points:

1.     Title: Missing hyphenation: “Deep learning-guided.”

2.     Line 79: The authors should cite the Vitrocam work (https://doi.org/10.1101/2022.06.16.496351) from the Carragher Lab.

3.     Line 81: The authors should reference the use of Energy filter-based Plasmon Imaging from Hagen (https://doi.org/10.3389/fmolb.2022.912363) when talking about medium magnification ice thickness screening in EPU.

4.     Line 146: The authors should comment on the accuracy of predicted data quality using their ALS model for ice thickness estimation in comparison to the quality obtained from standard EPU workflows and the use of the built-in “Filter ice quality” during EPU data collection.

5.     Line 165: Add full name to PDMS abbreviation

6.     Line 183: Methods for characterizing and distinguishing the flow velocity profiles of circular vs rectangular channel cross-sections not described.

7.     Line 252: Add frame rate, mode of operation of optical camera (maybe share a video of operation in suppl.)

8.     Line 269: This clause does not read clearly, may be missing “that” or “which” after 2D convolutions.

9.     Fig. 3e: Color gradient hard to make out differences. Select a suitable color gradient.

10.  Line 330 / Fig3g: The authors should show Chronobot-prepared grids using carbon grids and speculate on the reason for improved ice distribution on gold foil.

11.  Line 344: The authors should speculate on why variable rpm plunging increases collectable areas over constant rpm plunging. Variable acceleration could drive increased solvent evaporation and thinning.

12.  Line 350: The plasma cleaning optimization was done on gold grids?

1. The authors should add suitable significance tests (e.g., fig. 3)

14.   Line 356: The authors should mention that evaporation thinning comes with a tradeoff when combining with time-resolved studies, effectively increasing the shortest observable time-point.

15.  Line 359: 20K holes of which the ice-thickness range. Authors should add the range they deem collectable and check for empty holes (no ice).

16.  Line 372: What is the physiological relevance of the short intermediates? The authors should add a limitation of this approach, as filament-dynamics are constrained, potentially altering the native dynamics of the process compared to the native system.

17.  Line 383: The authors should rationalize their design and reference previous work using designed nucleotides for structural studies. They should also comment on the choice of microhomology sequence length and if additional lengths that varied in length by more than a single nucleotide were tested to assess weaker or stronger binders.

18.  Line 389: Global resolution is a poor metric to assess the local resolution in areas of interest (DNA-protein interaction). Referring to the local resolution plot would be beneficial.

19.  Line 391/430 and others: Particle statistics for occupancy are limited to methodological errors. This limitation needs to be addressed. Additionally, the authors should test local differences in compositional heterogeneity with tools like OccuPy (https://doi.org/10.1038/s41467-023-41478-1).  

20.  Fig 4e-i: DNA-biding sites I and II of RecA should be labeled.

21.  Line 404: The authors speculate about the state of DNA interacting with RecA to be either ds or ssDNA. Supplementary prediction using structure-based constraints would aid the line of reasoning (Chai Discovery supports such constraint predictions).

22.  Line 425: The authors should test whether longer nucleotide probes would stabilize the proposed intermediates. 

23.  Line 426: Sentence incomplete. Please add more details to the selected approach.

24.  Line 472: Missing punctuation: … added Notably …

25.  Line 640: Authors mention that hole estimation is effective for gold foils. Can this approach also to be applied for carbon grids and if so, is there a reduction in efficacy?

26.  Ext Fig 6a, 7a: Authors should provide an explanation for the high number of movies collected (~50-60k) and why does this not correspond to higher particle counts (reported ~600-700k). 

27.  Ext Fig 6e, 7e: Authors should include FSCs for masked maps as well. 

28.  Ext Fig 8b-d: Insets should be included on the model in panel b to clarify where the density in panels c and d correspond to. It would also be beneficial to dock the protein model into the density to further orient the viewer.

Competing interests

The authors declare that they have no competing interests.