NF: Cryo-EM

• sample preparation:

  • extraction of the protein/complex of interest

  • vitrification (plunge-freezing)

• challenges:

  • extraction might alter the structure of the complex

  • sample has to withstand high vacuum (à therefore, frozen)

  • sample has to withstand radiation

• Leidenfrost effect: When a room-temperature warm object is plunged into the liquid nitrogen, a gaseous layer forms and shields the object from the liquid nitrogen. This slows down the cooling rate which leads to crystallization of the water in the sample.

• Instead, liquid ethane is used which does not show the Leidenfrost effect and cools the sample faster.

• main advantage: no crystallization but vitrification

• low sample amount necessary (5 µL, 0.1 M)

• time-resolution

  • different conformations present

  • temperature-dependent studies possible

  • time-resolved studies possible

• negative stain sample preparation

  • carbon film and grid are applied on top of a solution of the sample

  • the sample molecules are adsorbed to the carbon

  • the solution is exchanged to a solution with heavy metal atoms

  • the grid is lifted out of the solution à grid with carbon film and heavy-metal-embedded sample molecules

  • resulting picture: shows “negative” of your sample (white spots instead of dark spots)

• advantages

  • high contrast

  • storability

  • radiation tolerance

  • fast and easy

• disadvantages

  • reduced resolution

  • flattening through stain

• 20 Å: domains

• 10 Å: α-helices

• 5 Å: β-strands

• 4 Å: big sidechains

• 2 Å: most sidechains

• diffraction limit in light microscopy is mainly determined by the wavelength of the light

• electrons have much lower wavelength, and their diffraction limit is below 1 Å (which is the size of an atom)

• the major resolution limitation in electron microscopy are the electron optical lenses

• the nucleus (NOT the electrons as in x-ray crystallography)

• 2 “modes”

  • inelastic scattering (electrons “bounce” off the nucleus and other electrons)

    • amplitude contrast

    • creates noise

  • elastic scattering (electrons are diverted by the nucleus, no loss of energy)

    • phase contrast

    • creates signal

• problem: we get the noise as amplitude contrast (easy to detect), but we get our signal as phase contrast (difficult to detect)

  —> we need to convert phase contrast to amplitude contrast

• this is achieved by:

  • defocus

  • spherical aberration

• defocus approach causes phase shift that is described by the Phase Contrast Transfer Function (CTF)

• somehow, that leads to higher resolution

• CFT can also be fitted and corrected to achieve even higher resolution

• aberrations of the lenses

  • spherical aberration

  • coma (no focus point when electron beam does not hit perpendicular to the lense)

  • astigmatism (elliptical instead of circular beam because of differences in x-/y-direction)

• radiation damage of the sample, even though you can avoid some of it by

  • cooling the sample

  • low-dose exposure on every single particle

• photographic film (outdated)

  • direct detection, excellent signal transfer

  • needs to be developed and digitalized, no direct feedback on image quality

• CCD devices

  • fast, direct feedback, relatively cheap

  • no direct detection (electrons need to be converted to photons), noisy

• direct electron detectors

  • very fast and precise

  • expensive, short life-time

• development of direct electron detectors

  • less noise than CCD

  • much faster than photographic film

  • possible to compensate for small beam-induced movements within the sample! (Incredibly huge difference to before)

• better maximum likelihood image processing routines

  • better averaging

  • see below

• https://doi.org/10.1126/science.1251652

• problems

  • variations in protein composition

  • damaged particles (purification and sample preparation)

• solutions

  • optimize purification protocols

  • optimize chemical conditions/buffers (Proteoplex: systematic screening for the best conditions)

  • optimize solubility

  • chemical modifications to improve stability (GraFix: macromolecules are centrifuged into a glycerol gradient)

• shows amplitude of all waves that contribute to the image depending on their frequency, neglecting the phase

• center has low resolution, edges have high resolution

• bright: signal, dark: no signal

  —> tells you how the distribution of your signal quality

• high-pass filter: filters out low-resolution information

• low-pass filter: filters out high-resolution information

  • helpful for initial image alignment

• band-pass filter: combination of high- and low-pass filter

• Imaging

• Particle Selection

  • single particles are selected manually or automatically and cut out

• Preprocessing

  • application of filters (removing noise and background) and round mask (facilitating rotation)

• Alignment

  • orientation of the particle

    • initially: all particles are averaged to obtain an “average disk” you can use to align your particles to the center

    • with reference: particles are added/aligned to a class average or 2D projection of a 3D reconstruction

• Classification

  • distinction between different specimens (e.g., conformations) and orientations via Principal Component Analysis (PCA)

• Angle determination

  • find out how the images of the different classes are located relative to each other (tooth paste example)

    • Random Conical Tilt (take another image of the sample from a different angle)

    • Angular Reconstruction (mathematically)

      • you construct a sinogram (1D projection of the image from different angles and stack them)

      • compare the sinograms of 2 images to find out the angle from which they are similar

    • Projection Matching (compare with 2D projections from 3D model)

• Reconstruction

  • 3D image is reconstructed from the classification and angle determination information

The steps alignment – reconstruction are repeated iteratively to improve image quality and resolution. From the reconstructed 3D image, you re-obtain 2D images and use them in the next round.

• Fourier Shell Correlation (FSC)

• the particles are separated into 2 halves and from each, a 3D image is reconstructed

• FFT is performed on both 3D reconstructions, results are compared via FSC

• difficult because you have to distinguish between different orientations AND different conformations

• approach: 3D MSA classification

  • unbiased approach!

  • you randomly overlap two particles from your dataset

  • randomly sample into 3D images

  • most do not make sense, but some do

  • you can use these for refinement

• approach: 3D maximum likelihood classification

  • you need a first model with very low resolution (80 Å)

  • you create copies of the model and add random noise to each

  • you use these for classification – because the models differ in noise, similar particles will be assigned to the same object

  • computationally very challenging

• problem: if you have a reference, you can find it everywhere (“Einstein from noise”)

  • avoid this by splitting your dataset in two

• problem: overfitting

  • if you overfit, you seem to get higher resolution, but you actually do not have the information in your dataset