![Page 1: Improving cryo-EM maps: focused 2D & 3D classifications ... › presentations › Bruno_Klaholz... · 3) maximum likelihood based classifications e.g. Scheres et al., JMB 2005; Meth.Enzymol](https://reader033.vdocument.in/reader033/viewer/2022052801/5f160eff46df23632247c2be/html5/thumbnails/1.jpg)
3rd International Symposium on Cryo-3D Image Analysis23.3.2018, Tahoe
Bruno KlaholzCentre for Integrative Biology, IGBMC, Illkirch, France
http://www.igbmc.fr/Klaholz
Improving cryo-EM maps: focused 2D & 3D classifications and focused refinements
Note: you are welcome to reuse slides elsewhere but please cite your source
![Page 2: Improving cryo-EM maps: focused 2D & 3D classifications ... › presentations › Bruno_Klaholz... · 3) maximum likelihood based classifications e.g. Scheres et al., JMB 2005; Meth.Enzymol](https://reader033.vdocument.in/reader033/viewer/2022052801/5f160eff46df23632247c2be/html5/thumbnails/2.jpg)
Conformational changes of cats?
B. Klaholz
![Page 3: Improving cryo-EM maps: focused 2D & 3D classifications ... › presentations › Bruno_Klaholz... · 3) maximum likelihood based classifications e.g. Scheres et al., JMB 2005; Meth.Enzymol](https://reader033.vdocument.in/reader033/viewer/2022052801/5f160eff46df23632247c2be/html5/thumbnails/3.jpg)
I. Particle sorting
![Page 4: Improving cryo-EM maps: focused 2D & 3D classifications ... › presentations › Bruno_Klaholz... · 3) maximum likelihood based classifications e.g. Scheres et al., JMB 2005; Meth.Enzymol](https://reader033.vdocument.in/reader033/viewer/2022052801/5f160eff46df23632247c2be/html5/thumbnails/4.jpg)
How to sort out heterogeneity (composition / conformation)?
particle sorting
4 different approaches in the cryo-EM field:
1) reference-based, i.e. cross correlation with forward-projections of known structures
2) multivariate statistical analysis (MSA): 2D classification or 3D classification, variance analysis + resampling, bootstrapping, 3D resampling
3) maximum likelihood based classifications
4) deep learning methods
![Page 5: Improving cryo-EM maps: focused 2D & 3D classifications ... › presentations › Bruno_Klaholz... · 3) maximum likelihood based classifications e.g. Scheres et al., JMB 2005; Meth.Enzymol](https://reader033.vdocument.in/reader033/viewer/2022052801/5f160eff46df23632247c2be/html5/thumbnails/5.jpg)
1) reference-based, i.e. cross correlation with forward-projections of known structures
Loerke et al., Meth. Enzymol. 2010
example:
adding a reference
Determining structures of multiple conformational states in a single sample
![Page 6: Improving cryo-EM maps: focused 2D & 3D classifications ... › presentations › Bruno_Klaholz... · 3) maximum likelihood based classifications e.g. Scheres et al., JMB 2005; Meth.Enzymol](https://reader033.vdocument.in/reader033/viewer/2022052801/5f160eff46df23632247c2be/html5/thumbnails/6.jpg)
side-view
Determining structures of multiple conformational states in a single sample
distinguish: orientational classification and conformational classification
2) multivariate statistical analysis (MSA): 2D classification, 3D classification
B. Klaholz
![Page 7: Improving cryo-EM maps: focused 2D & 3D classifications ... › presentations › Bruno_Klaholz... · 3) maximum likelihood based classifications e.g. Scheres et al., JMB 2005; Meth.Enzymol](https://reader033.vdocument.in/reader033/viewer/2022052801/5f160eff46df23632247c2be/html5/thumbnails/7.jpg)
Determining structures of multiple conformational states in a single sample
distinguish: orientational classification and conformational classification
2) multivariate statistical analysis (MSA): 2D classification, 3D classification
front-view, conformation 2B. Klaholz
![Page 8: Improving cryo-EM maps: focused 2D & 3D classifications ... › presentations › Bruno_Klaholz... · 3) maximum likelihood based classifications e.g. Scheres et al., JMB 2005; Meth.Enzymol](https://reader033.vdocument.in/reader033/viewer/2022052801/5f160eff46df23632247c2be/html5/thumbnails/8.jpg)
Determining structures of multiple conformational states in a single sample
local 2D MSA (focused classification)
50S view of the 70S or 50S particle?
nom. defocus -4
70S particle with or without factors?
Klaholz et al., Nature 2004; see Suppl. Mat.
30S 50S
local MSA
Perform 2 classifications:
(i) global MSA for classification according to particle orientations (i.e. classical class averages)
(ii) local MSA with a smaller mask for classification according to particle variability.
2) multivariate statistical analysis (MSA): 2D classification, 3D classification
correlated movements
2D
![Page 9: Improving cryo-EM maps: focused 2D & 3D classifications ... › presentations › Bruno_Klaholz... · 3) maximum likelihood based classifications e.g. Scheres et al., JMB 2005; Meth.Enzymol](https://reader033.vdocument.in/reader033/viewer/2022052801/5f160eff46df23632247c2be/html5/thumbnails/9.jpg)
Simonetti et al., Nature, 2008.
Sorting out heterogeneity of complexes:
3D statistical analysis and 3D classification:3D sampling and classification (3D-SC)
2D 3D 4D
200kV FEG data;total 80 000 particlesresolution of 3D's: 9Å
used by Fischer et al., Nature, 2010; Papai et al., Nature 2010.
40 % 28 % 10 % 8 % 14 %
3D reconstruction of single particles:
unique particle type in different orientations
does both re-sampling and 3D classification;see also work by P. Penczek (bootstrapping (re-sampling), used primarily to find region of variance)see also S. Scheres/J-M Carazo (maximum likelihood parameter refinement and classification)
Klaholz, Open J. of Stat., 2015.3D
![Page 10: Improving cryo-EM maps: focused 2D & 3D classifications ... › presentations › Bruno_Klaholz... · 3) maximum likelihood based classifications e.g. Scheres et al., JMB 2005; Meth.Enzymol](https://reader033.vdocument.in/reader033/viewer/2022052801/5f160eff46df23632247c2be/html5/thumbnails/10.jpg)
3) maximum likelihood based classifications
e.g. Scheres et al., JMB 2005; Meth. Enzymol. 2010;
Lyumkis et al., JSB 2013
Introduction of the ML concept in cryo-EM: F. Sigworth, JSB 1998;
in X-ray crystallography: G. Bricogne, Acta Cryst A, 1991
assign particles to different 3D classes based on maximum likelihood
(probability distribution; uses randomly selected references + ML-weighting)
Practically:
random subsets are optimized and a low-resolution average structure is used as reference,
i.e. resampling is used in combination with likelihood optimization
Determining structures of multiple conformational states in a single sample
![Page 11: Improving cryo-EM maps: focused 2D & 3D classifications ... › presentations › Bruno_Klaholz... · 3) maximum likelihood based classifications e.g. Scheres et al., JMB 2005; Meth.Enzymol](https://reader033.vdocument.in/reader033/viewer/2022052801/5f160eff46df23632247c2be/html5/thumbnails/11.jpg)
Strong heterogeneity of a reconstituted eukaryotic translation initiation (eIF5B) complex:
sorting 5143 particles, representing 3% of the population in the sample, 6.6 Å reconstruction.
Fernández et al., Science 2013; V. Ramakrishnan & S. Scheres.
5 000 40 000 particles
Examples of ML-based 3D classification
![Page 12: Improving cryo-EM maps: focused 2D & 3D classifications ... › presentations › Bruno_Klaholz... · 3) maximum likelihood based classifications e.g. Scheres et al., JMB 2005; Meth.Enzymol](https://reader033.vdocument.in/reader033/viewer/2022052801/5f160eff46df23632247c2be/html5/thumbnails/12.jpg)
Abeyrathne et al., eLife 2016
e.g. ML-based focused classificationof 80S / TSV IRES complex with eEF2/GDP/sordarin
See also: von Loeffelholz et al., Curr. Opin. Struct. Biol. 2017.
![Page 13: Improving cryo-EM maps: focused 2D & 3D classifications ... › presentations › Bruno_Klaholz... · 3) maximum likelihood based classifications e.g. Scheres et al., JMB 2005; Meth.Enzymol](https://reader033.vdocument.in/reader033/viewer/2022052801/5f160eff46df23632247c2be/html5/thumbnails/13.jpg)
Myasnikov et al., Nat. Comm. 2016.
e.g. ML-based focused classification
sorting scheme for human 80S/antibiotic complex
![Page 14: Improving cryo-EM maps: focused 2D & 3D classifications ... › presentations › Bruno_Klaholz... · 3) maximum likelihood based classifications e.g. Scheres et al., JMB 2005; Meth.Enzymol](https://reader033.vdocument.in/reader033/viewer/2022052801/5f160eff46df23632247c2be/html5/thumbnails/14.jpg)
II. Focused refinement
![Page 15: Improving cryo-EM maps: focused 2D & 3D classifications ... › presentations › Bruno_Klaholz... · 3) maximum likelihood based classifications e.g. Scheres et al., JMB 2005; Meth.Enzymol](https://reader033.vdocument.in/reader033/viewer/2022052801/5f160eff46df23632247c2be/html5/thumbnails/15.jpg)
Klaholz et al., 2004.
Local MSA / focused 2D/3D classification & focused refinement:
von Loeffelholz et al., Curr. Opin. Struct. Biol. 2017.
see also: Klaholz et al., Nature 2004;White et al., JSB 2004;Penczek et al., JSB 2006;Wong et al., Elife 2014;…
Helps: use a slightly larger region than the region of interest, e.g. 30-50 Å in diameter
![Page 16: Improving cryo-EM maps: focused 2D & 3D classifications ... › presentations › Bruno_Klaholz... · 3) maximum likelihood based classifications e.g. Scheres et al., JMB 2005; Meth.Enzymol](https://reader033.vdocument.in/reader033/viewer/2022052801/5f160eff46df23632247c2be/html5/thumbnails/16.jpg)
Natchiar et al., Nature 2017.
Focused refinement:
See also: von Loeffelholz et al., Curr. Opin. Struct. Biol. 2017.
![Page 17: Improving cryo-EM maps: focused 2D & 3D classifications ... › presentations › Bruno_Klaholz... · 3) maximum likelihood based classifications e.g. Scheres et al., JMB 2005; Meth.Enzymol](https://reader033.vdocument.in/reader033/viewer/2022052801/5f160eff46df23632247c2be/html5/thumbnails/17.jpg)
Natchiar et al., Nature 2017.
![Page 18: Improving cryo-EM maps: focused 2D & 3D classifications ... › presentations › Bruno_Klaholz... · 3) maximum likelihood based classifications e.g. Scheres et al., JMB 2005; Meth.Enzymol](https://reader033.vdocument.in/reader033/viewer/2022052801/5f160eff46df23632247c2be/html5/thumbnails/18.jpg)
Allows visualization of chemical modifications in rRNA:
Natchiar et al., Nature 2017.
![Page 19: Improving cryo-EM maps: focused 2D & 3D classifications ... › presentations › Bruno_Klaholz... · 3) maximum likelihood based classifications e.g. Scheres et al., JMB 2005; Meth.Enzymol](https://reader033.vdocument.in/reader033/viewer/2022052801/5f160eff46df23632247c2be/html5/thumbnails/19.jpg)
III. Improving data collection quality
![Page 20: Improving cryo-EM maps: focused 2D & 3D classifications ... › presentations › Bruno_Klaholz... · 3) maximum likelihood based classifications e.g. Scheres et al., JMB 2005; Meth.Enzymol](https://reader033.vdocument.in/reader033/viewer/2022052801/5f160eff46df23632247c2be/html5/thumbnails/20.jpg)
Detectors, dose weighting, movie alignment
Orlov et al., Biology of the Cell, 2017.
![Page 21: Improving cryo-EM maps: focused 2D & 3D classifications ... › presentations › Bruno_Klaholz... · 3) maximum likelihood based classifications e.g. Scheres et al., JMB 2005; Meth.Enzymol](https://reader033.vdocument.in/reader033/viewer/2022052801/5f160eff46df23632247c2be/html5/thumbnails/21.jpg)
Volta phase plate data collection facilitates image processing and cryo-EM structure determination
von Loeffelholz et al., JSB 2018.
dfVPP data behave more robustly during
image processing:
particle selection, accuracy in alignments,
2D & 3D classifications, map interpretation
MSA-based classification (3 particles)
![Page 22: Improving cryo-EM maps: focused 2D & 3D classifications ... › presentations › Bruno_Klaholz... · 3) maximum likelihood based classifications e.g. Scheres et al., JMB 2005; Meth.Enzymol](https://reader033.vdocument.in/reader033/viewer/2022052801/5f160eff46df23632247c2be/html5/thumbnails/22.jpg)
Single- / dual-tilt cryo electron tomography
single tilt
Titan Krios, GIF/K2, VPP, SerialEM, Tom toolbox, IMOD
dual tilt
see also Myasnikov et al., Ultramicroscopy 2013.
T. Frosio & J. Ortiz / Klaholz lab.
Less artefacts(missing pyramid)
Stripes(missing wedge)
![Page 23: Improving cryo-EM maps: focused 2D & 3D classifications ... › presentations › Bruno_Klaholz... · 3) maximum likelihood based classifications e.g. Scheres et al., JMB 2005; Meth.Enzymol](https://reader033.vdocument.in/reader033/viewer/2022052801/5f160eff46df23632247c2be/html5/thumbnails/23.jpg)
Summary
Improving 3D reconstructions and cryo-EM maps:
1) particle sorting (to address heterogeneity)
2) focused refinement ( composite maps; for a given conformation)
3) improve data quality (CMOS detectors, VPP, dual-axis tomography)
4) to help interpretation: map sharpening
(B-factor, bp-filtering, LocScale, phenix.autosharpen etc.)
Keep in mind:
- cryo-EM maps are electrostatic potential maps ( Glu's, Asp's etc.)
- initial 3D reconstruction (not "model")
atomic model building & refinement, proper geometry & B-factors