I’ve been working on reconstructing amyloid fibers and noticed some reluctance in the community about using CryoSPARC for this purpose. Some reviewers/researcher in this field have even suggested that other software packages might be more suitable for solving these types of structures. I wanted to kindly ask if you might have any insights into the reasons behind this hesitation in using CryoSPARC for this purpose, or if it’s just a field-bias issue.
Additionally, if anyone you know has tips or tricks for handling amyloid fiber reconstructions, I would greatly appreciate any advice you can share. I’m particularly interested in improving the quality of my reconstructions and learning more about the nuances of different software options for these challenging datasets.
Thank you in advance for your time and assistance!
1 Like
It sounds like you are asking, “should I use cryosparc for helical reconstructions?”
If yes, then there are few approaches, but helical reconstruction is highly error prone if parameters are not correct - it is easy to get wrong.
The RELION lab https://www2.mrc-lmb.cam.ac.uk/group-leaders/n-to-s/sjors-scheres/ does much work on similar fibrils in addition to all of the software development. They have some systems were BLUSH regularization helps for similar systems.
Before getting there I would ensure you are using all of the tools for picking. The Filament Tracer https://guide.cryosparc.com/processing-data/all-job-types-in-cryosparc/helical-reconstruction-beta is a great start.
A big point is that the deep pickers (CrYOLO and TOPAZ) can do a great job picking repeating units on filaments if you know what to pick.
It could be worth to to explore beyond single particle. Sub-tomogram averaging has also been used with success in such systems. If you have a novel amyloid, AFM and ssNMR could also provide novel insights.
Hi Mark,
Thank you for your reply, but it wasn’t exactly my point. I was curious why does the Amyloidal-related community don’t use CS as an standard workflow for Helical reconstruction. I’m aware of the other packages and, indeed, Relion is responsible for almost 100% of EMPIAR/EMDB depositions.
So, I was just curious to know why is this community reluctant against CS as an useful tool for this type of data processing. For example, I was able to only find a single manuscript that uses only CS for it (https://www.pnas.org/doi/10.1073/pnas.2406775121).
It was more philosophical than technical question.
Thanks
1 Like
Understood. I think the whole thing with Sjors then. He has done excellent work on such systems and also develops RELION.
can also see https://github.com/asarnow/pyem/wiki/Export-from-cryoSPARC-v2-and-later. “Why do Relion and cryoSPARC results look different?” from @DanielAsarnow
It could be interesting to see if you can take an EMPIAR data sets and where it goes with CryoSparc vs. RELION or something else (SIMPLE 3.0, cisTEM, Thunder, …very long list). The general advice is to try as much as you can.
In fact we did it, and some of Relion-published EMDB entries are not reproducible. It’s interesting how some reviewers write in between lines pretty much saying "if you haven’t use Relion, it’s incorrect or unacceptable ".
Then, I was curious why amyloidal-field is “too biased” towards a single package and way of doing it.
1 Like
My understanding is that helical processing in CS is not yet optimized for dealing with amyloids, which have some very specific characteristics - small rise and small twist most particularly.
Sjors and others have put a lot of effort into developing protocols to use RELION to process such cases, and added a number of amyloid specific tools/scripts, whereas nothing comparable yet exists for CS.
Doesn’t mean you can’t use CS for amyloids though! Just that the tools are not as well-developed for that particular use case yet.
Re reviewers complaining, the density should speak for itself. If the density map (and model fit) is unambiguous, it should not matter what software package you used.
EDIT: Job: Helical Refinement | CryoSPARC Guide - note difficulties with amyloid fibrils are pointed out explicitly in the CS guide.
Hi @olibclarke,
I completely agree with your final statement, and this is something that we try to argue making the comment that it shouldn’t matter, but I think the common sense in the field is: if it’s not done using Relion, this is non-acceptable.
Anyway, thank you and @Mark-A-Nakasone for the insights.
1 Like
I can add some helpful discussion to this topic. Recently, we published a preprint of an amyloid fibril at 2.9 Å resolution using cryoSPARC. We included detailed data processing methods in the manuscript. I can summarize them here. Cryo-EM of cardiac AL-224L amyloid reveals shared features in λ6 light chain fibril folds | bioRxiv
Relion is indeed better optimized for amyloids, so you will need to perform a few extra important steps and change several parameters to find success in cryoSPARC.
Particle cleaning:
Since the rise/twist symmetry of an amyloid is very small you need to get a high-quality clean fibril segment stack from thin ice to see the beta sheet separation well enough for cryoSPARC’s symmetry search algorithm to reliably find the correct rise/twist values. It was important to collect data from thin ice areas and collect a large amount of data (10k movies) to get a high quality homogeneous fibril segment stack. Particle cleaning was most effective using iterative 2D classification jobs. Clear contaminant (ice features), poor quality fibril segments (fuzzy fibril segments), and double filament segments were removed.
Initial reconstruction:
Fibril segments were first extracted using large boxes sizes to more accurately align the extracted segments along the fibril axis. An initial reconstruction was generated by selecting the best volume from a multi-structure ab-initio reconstruction then performing a standard helix refine job with the ab-initio volume and the entire fibril segment stack to obtain a more clean initial reconstruction and a particle stack with estimated 3D alignments of the fibril segments.
Helical structure refinement:
This is a tricky stage of the data processing workflow and required many adjustments to find success. It was necessary to re-extract the cleaned partially-aligned fibril segment stack reducing the box size to be very small (~2x the diameter of the fibril). A very small box size seems to help cryoSPARC see the beta sheet separation and reliably find the correct rise/twist symmetry values. A homogeneous reconstruction and helix refine were performed without applying any symmetry search/averaging to allow cryoSPARC to find beta sheet features in an unbiased way. The output volume needs to show bumps corresponding to beta sheet stacking of an amyloid (~4.6 Å separation). If you can’t see the bumps by eye then cryoSPARC likely won’t be able to see it during its symmetry search and will probably apply the wrong rise/twist symmetry values during the refinement. Collecting additional data or performing additional fibril segment cleaning is necessary if the bumps are not visible. The next step is to perform another helix refine job but with symmetry search/averaging enabled. It is important to use an input volume that has visible beta sheet bumps and to set the initial model lowpass filtering parameter to 4.5 Å (high enough resolution to retain beta sheet features “bumps” in the input volume). *The default parameter of 20 Å will blur the visible beta sheet features making it impossible to find the correct rise/twist values. Monitor the symmetry search output graphs throughout the refinement and make sure that cryoSPARC is finding a reasonable hotspot in the rise/twist coordinate space. The goal is to guide cryoSPARC to find the correct rise/twist values in an unbiased way, not to tell it what you think the rise/twist values should be, so set the minimum and maximum bounds of rise/twist to allow for some symmetry search leeway. The correct rise/twist values and a successful helical refinement should produce a structure with clear beta sheet separation, a normal FSC plot without large dips in the curve, clear backbone path through the fibril cross-section, and some clear side-chain density.
Post-processing:
We had some good results with local CTF-refinement and reference-based motion correction. CryoSPARC’s reference based motion correction is pretty awesome. Make sure to validate any post-processing steps by performing another refinement or reconstruction afterwards to make sure that the step is improving resolution and map quality.
Final note:
Heterogeneity in the fibril segment stack from double filament fibrils or from fibrils with heterogeneous rise/twist symmetry along the fibril axis may cause the helical refinement to fail. Additional particle cleaning or classification in 2D or 3D space should help improve the homogeneity of the fibril segment stack. However, it may be difficult to get a good enough fibril segment stack for a successful symmetry-applied helical refinement if the heterogeneity in the rise/twist is too large in the dataset.
I hope this helps!
-Chad Hicks
4 Likes