Read on Twitter
The first part of my work as a postdoc is available as a pre-print! https://www.biorxiv.org/content/10.1101/2021.04.20.440648v1
Tweetorial">https://www.biorxiv.org/content/1... below:
Tweetorial">https://www.biorxiv.org/content/1... below:
With cryo-EM, we use high-resolution images of molecules to build models of their structures. Unlike X-tallography or NMR, it shouldn& #39;t matter how pure the sample is. Why can& #39;t we just take pictures of cells and determine the structure, location and interactions at the same time?
The problem is that cells are crowded and locating a specific molecule is not trivial. There are currently two solutions to this problem. The most common is to collect a tilt series, this makes low resolution features stand out and molecules of distinctive shapes can be ID& #39;d.
This approach, termed 3DTM, is limited to complexes with distinctive shapes in regions of the cell where molecular crowding allows them to be distinguished. It currently suffers from a relatively high false-positive rate, needing extensive validation.
Another approach developed by @opticalSieve locates molecules in 2D cryo-EM images using their internal structure, which reduces false positives and allows detection even in dense environments (check out his 2017 eLife paper and bioRxiv pre-print for more information)
In this pre-print we describe implementation of this 2DTM method into cisTEM (by #twitterless Tim Grant) allowing parallelization and a streamlined workflow. This was made speedy with GPU acceleration by @cryo2go
I used 2D images of M. pneumoniae prepared by @Liang_Xue_ and Julia Mahamid @embl to locate ribosome LSUs. We averaged ~6,500 LSUs from 2DTM and determine a 3D reconstruction, which showed features consistent with translating ribosomes. Note: this is without any refinement!
Side note: we filtered the reconstructions to 20Å because 1: averaging particles ID& #39;d using a high-res template can introduce template bias (2B quantified), and 2: to compare the two reconstructions. Will the threshold in 2DTM take care of high-res bias? (we predict it will) TBD
Since this method relies on existing high-res structures, I tested whether we can locate LSUs in M.pneumoniae with a B.subtilis LSU. YES, but fewer. We were able to ID species-specific structural features. Useful jumping off point for non-model organisms (which is most)
How do 3DTM and 2DTM compare on the same sample? @Liang_Xue_ and Julia Mahamid collected 2D images and overlapping 3D tomos of the same area and we each ID& #39;d LSUs. 2DTM found fewer overall, but in thin (~100nm) ice LSU detection was ~ the same.
In 3DTM one of the major steps is IDing false positives. 2DTM and 3DTM each provide independent info, we can assign high-confidence to targets detected by both. 2DTM brings the specificity, 3DTM brings the sensitivity in thick ice (in thin ice 2DTM is fine), more Z & context info
Finally, we show that, just as predicted by @opticalSieve, it is possible to distinguish overlapping ribosomes from 2D images in cells without needing to tilt. This addresses one of the main concerns with 2D images relative to tomography for in situ visual proteomics.
