3DEM methods Electron tomography Tibor Füzik Lecture 8 Electron tomography Koning et al. 2018 https://doi.org/10.1016/j.aanat.2018.02.004 0o -60o 60o Tomography vs Single particle analysis • Single particle analysis • Taking single projection image of the sample • Exposing the acquisition area only once (“high” SNR) • Assuming that the object of interest is in random uniform orientations • Cannot make 3D reconstruction form a single 2D projection • Not suitable for samples where 3D information from single acquisition area is needed (cells, non-uniformly organized structures) • Sample needs to be electron transparent • Tomography analysis • Taking multiple projection images of the same area under different tilts • Exposing the acquisition area multiple times (low dose, low SNR) • Tilt angles can be assigned to the projection images • From series of tilts, we can reconstruct a 3D volume • Suitable for studying large samples, macromolecules in situ, poorly organized structures • No ambiguity in handedness determination • Sample needs to be electron transparent Belnap et al., 1997 https://doi.org/10.1006/JSBI.1997.3896 Purified particles Cellular samples Thin Thick Plunge-freezing in liquid ethane High-pressure freezing Cryo-FIB milling Cryo-ultramicrotomy CryoEM (SPA) Cryo-electron tomography Sample preparation for tomography Sample preparation for tomography Klumpe et al. 2022 https://doi.org/10.1017/S1551929521001528 Acquisition • Acquisition of tilt-series • Single acquisition area exposed multiple times • Radiation damage • Electron dose per tilt ~2-3 e-/Å2 • Total electron dose = number of tilts * dose per tilt (100-150 e-/Å2) • Dose symmetric tilt scheme • Inclusion of fiducials in the sample • Small (5-10 nm) gold beads allowing precise tracking of the tilts • Defocus during tilt-series • Usually kept constant • Proper alignment of tilt axis • Set of eucentric height (center of rotation) • Corrections Correct eucentric height – no tilt axis offset Zero tilt-axis offset Tilt-axis Non-zero tilt-axis offset Tilt-axis Tilt-axis offsetSample lateral movement of the point of interest changeindefocus Nafari et al. 2008 http://dx.doi.org/10.1109/JMEMS.2007.912714 • After each tilt change a “tracking area” is imaged and by cross correlation with the previous image correction to image shift is done • After each tilt change autofocus is done on focusing area • Tracking and focusing area must lay on the tilt- axis Compensation for tilt-axis offset Stage tilt Tracking Focusing Tilt-image acquisition Tilt schemes • Order in which the tilts are collected From most positive to most negative tilt 60o -> -60o From zero to most positive From zero to most negative tilt 0o -> 60o ; 0o -> -60o Positive, positive, negative, negative, positive, positive, negative, negative, 0, 3; -3, -6; 6, 9; -9, -12; 12, 15; -15…. Hagen et al., 2017 https://doi.org/10.1016/j.jsb.2016.06.007 Dose symmetric tilt-scheme • First the small angle tilts are collected • Sample thickness is minimal => most transparent part • High contrast • Lower radiation damage • Contain the most useful high freq. information • Last the high angle tilts are collected • Tilt-induced grow in sample thickness => decreased transparency • Lower contrast • Higher radiation damage 0o 60o Radiation damage of cellular samples Koning et al. 2018 https://doi.org/10.1016/j.aanat.2018.02.004 Acquisition setup • Stage setup • Choice of angular increment / max tilt angle (e.g. +-60o, 3o step) • Single axis/dual axis • Camera setup • Counting mode (ideally CDS mode on K3) • Short exposure times (per tilt dose 2-3 e-/A2 => on K3 ~0.5 sec) • Fractionation into few dose fractions (~4 fractions; <1e-/A2/fraction) • Energy filter setup • High-tilts => thick sample (more inelastically scattered electrons) • Zero loss mode – slit set to 10-20 eV • Increased contrast • Phase-plate setup • Low dose low contrast => compensated by high defocus • Volta phase plate – comparable contrast at lower defocus (by applying phase shift on CTF) • Combining SPA and tomography • Zero tilt acquired at higher dose (10-20 e-/A2) – serves as micrograph for SPA • Other tilts acquired at standard dose Energy filter, Volta phase plate (VPP) Koning et al. 2018 https://doi.org/10.1016/j.aanat.2018.02.004 -5µm defocus, without VPP In focus, width VPP https://doi.org/10.1073/pnas.1418377111Danev et al. 2014 Tilt range limitation Missing wedge • Missing wedge – missing in Fourier space (therefore affects all the point in real-space) • The information is missing there is no possibility to add it or recover it (it was not recorded at all) Ideal (+- 90o) Real (+- 60o) FFT XY plane of a tomogram slice FFT XZ plane of a tomogram slice Koning et al. 2018 https://doi.org/10.1016/j.aanat.2018.02.004 Dual (multi) tilt tomography • Dual-axis sample holders Koning et al. 2018 https://doi.org/10.1016/j.aanat.2018.02.004 DOI: 10.1186/s40679-016-0021-2Phan et al. 2016 Tomography - Data processing • Motion correction of raw movie data of single tilts • Possibility of dose weighting of single tilts • Alignment of tilt series • Coarse alignment – cross correlation between neighboring tilts • Fine alignment – fiducial model that describe the tilting transformations • Aligned tilt series postprocessing • Fiducial removal, CTF correction • Reconstruction of the tomogram • Back-projection algorithms • Tomogram filtering • Segmentation of tomogram • Sub-tomogram averaging • Refining a high-resolution structure from the subparts of the tomogram Tilt alignment – Coarse align After acquisition After coarse alignment Fiducial based fine alignment • Need fiducials • Finding the same fiducial on all of the tilts and fitting a model on their trajectory • Allows subpixel precision alignment of tilt series https://www.renafobis.fr/...../leforestier-tomography-renafobis_2021.pdf Patch tracking – fine alignment • Tilt images split into patches • Correlating the patches between the tilts • Fitting trajectory on the movement of the patches • Every patch area needs enough signal to be traceable • “Fiducial model” is created from the data itself without any additional fiducials included • Less precise than fiducial based alignment models Fine alignment using the fiducial model Reconstruction • Back-projection • Low frequencies are over-represented • Weighted back-projection • Low frequencies are down weighted • SIRT • Simultaneous Iterative Reconstruction Technique Nováček J.https://www.int.kit.edu/1731.php SIRT (Simultaneous Iterative Reconstruction Technique) • Start with a tomogram reconstruction from the backprojection of the tilts • Reprojecting from the tomogram in the original tilt orientations • Taking the difference between the original projection data and this reprojection at each pixel (this difference represents the amount of error in the current estimate) • Adding the error to the tomogram by a back-projection operation https://bio3d.colorado.edu/imod/doc/SIRTexample.html Reconstruction Weighted Back-projection SIRT Missing wedge XY Plane XZ Plane FTT XY Plane FFT XZ Plane Tomogram interpretation • Visual inspection • Annotation of features • Segmentation • Annotating/extracting segments from the tomogram • Sub-tomogram averaging • Filtering – postprocessing (denoising) • Lowpass filters – increase contrast loose details • High pass filters – edge detection • Median filters • NAD – Nonlinear Anisotropic diffusion • Need to be tuned for balance between contrast and preserved details • Neural network based denoising • Noise model trained on data Bepler et al., 2020 https://doi.org/10.1038/s41467-020-18952-1 NAD – nonlinear anisotropic difusion Before After Segmentation Šiborová, M. et al. 2022 Segmentation • Annotating parts of the tomogram by connecting continuous segments • Manual • Drawing • Thresholding • “Missing wedge” limits the interpretability of the features • Semiautomatic • Simultaneous annotation on multiple slices at once • Correlation based annotation • Convolutional neural networks • Training CNN on small part of annotated tomogram • Long training process • General models not working well Sub-tomogram averaging • Increasing signal to noise ratio • Increasing resolution • By averaging uniformly orientationally distributed sub-tomograms we fill the missing wedge in the Fourier space http://dx.doi.org/10.1016/bs.mie.2016.04.014Wan & Briggs 2016 Tilt-dose weighting • High angle tilts heavily radiation damaged • Not containing high resolution information • Essential for high resolution sub-tomogram averaging • Dose weighting during movie motion correction process • Implemented in many sub-tomogram averaging software • Beware not to double dose weight the data exposure-dependent amplitude attenuation accumulated exposure of the frame Spatial frequency Dose-weighted Fourier transform of Image i-th fraction / i-th tilt From lecture 7 Grant & Grigorieff 2015 https://doi.org/10.7554/eLife.06980.001 CTF estimation – defocus gradient CTF Estimation Unconstrained CTF fit Constrained search range, constraining resolution Gradually depending on the tilt angle CTF correction during tomogram reconstruction Turonova et al., 2017 https://doi.org/10.1016/j.jsb.2017.07.007 Template matching • Hard to locate particles in 3D tomograms with “missing wedge” • Manual picking • When no template is available • Manual curation of template matching results • Template matching • Need template for matching • Looking for a 3D volume in a 3D volume • Cross-correlation methods • Orientational search of the template • Computationally expensive • Produce not only coordinates (X, Y, Z) but also the best fitting orientation (φ, θ ,ψ) • Many false positive matches 3D correlation map template Template Matching Sub-tomogram Averaging process • Iterative process as SPA refinement • Initial model can be generated from the data (need to be lowpass filtered) • Computationally expensive • Aligning 3D volume on 3D volume • 3 Euler angles, 3 shifts • Cross correlation/Maximum Likelihood methods • Need to compensate for the “Missing wedge” • Speeding up calculations by gradual unbinning of the sub-tomograms • Coarse search on heavily binned particles • Fine (local) search on unbinned particles Sub-tomogram Averaging process http://dx.doi.org/10.1016/bs.mie.2016.04.014Wan & Briggs 2016 Sub-tomogram averaging – gradual refinement Placeback of averaged sub-tomograms • Positions of the sub-tomograms int the tomogram is known • After averaging the sub-tomograms have refined orientations and shifts • Placing back the refined volume into the original tomogram • Describes better the data in tomogram than segmentation Placeback of averaged sub-tomograms Selected placeback particles 200 290Å Conclusion • Electron tomography • Advantages / disadvantages • Sample preparation • Data acquisition (limitations) • Data processing • Electron tomography future • Sub-tomogram averaging in situ • Speeding up data acquisition • Automated data processing • Combination of tomography and SPA • Correlative microscopy (combination of fluorescent and electron microscopy) Thanks for your attention! #teamtomo