Cryo-EM & vitrification
Rapid freezing turns a biological specimen into vitreous ice, preserving native structure for electron imaging.
Cryogenic electron microscopy (Cryo-EM) images biological specimens that have been frozen so fast that the surrounding water solidifies without crystallizing. The defining step is vitrification: a thin aqueous sample is plunged into a cryogen such as liquid ethane (cooled by liquid nitrogen) at cooling rates exceeding K/s. Water passes directly into an amorphous, glass-like solid — vitreous ice — rather than forming hexagonal crystals.
Picture photographing the salt dissolved in a glass of water. Freeze the water slowly and the growing crystals shove the salt out to the grain boundaries as they expand; the photo shows salt that has been pushed around and deformed, not where it actually sat. Vitrification does the opposite: freeze fast enough and the water molecules have no time to line up into a lattice, so they seize in place — locking solutes and macromolecules at their liquid-state positions. The diagram below breaks the process into steps.
How cooling rate decides the ice state — watch the molecular arrangement and its diffraction:
Slow cooling lets water settle onto a hexagonal ice lattice — diffraction shows sharp Bragg spots and the crystals damage the specimen. Fast cooling (plunging into liquid ethane) traps the molecules in an amorphous glass — diffraction collapses to a diffuse halo. This vitreous ice is what Cryo-EM needs.
Liquid ethane conducts heat away from the specimen more effectively than liquid nitrogen, which boils on contact and wraps the sample in an insulating vapor film (the Leidenfrost effect) that slows heat extraction; ethane or an ethane–propane mixture is therefore the usual cryogen. Excess liquid is first blotted away with filter paper so the film is only tens to a few hundred nanometers thick, thin enough for the required cooling rate to be reached throughout.
Why that “tens to a few hundred nanometers” scale? Heat diffuses from the interior of the sample out to the cryogen at the surface, and the time this takes grows roughly with the square of the thickness — double the thickness and it takes about four times as long to cool through. To cool the entire thickness within the brief window before water can nucleate crystals, the film has to be thin. An ethane–propane mixture has a further advantage over pure ethane: its lower melting point keeps it from solidifying into a plug at preparation temperatures, widening the operating window.
Crystalline ice is destructive twice over: growing crystals push solutes aside and disrupt fine structure, and the ordered lattice produces strong diffraction that overwhelms the weak signal from the specimen. Vitreous ice avoids both problems. It embeds macromolecules in their hydrated, near-native conformations, with no stain, no fixative, and no dehydration. This near-native preservation is the central advantage of Cryo-EM and of electron tomography over conventional plastic-embedded methods.
The conventional plastic-embedding route first cross-links the sample with chemical fixatives, then dehydrates it, then stains it with heavy metals — and every one of those steps rewrites the structure: fixation can tug conformations around, dehydration strips away the hydration shell, and what staining shows is the outline of a metal cast rather than the molecule itself. Vitreous ice skips all of them. The price is a specimen that is extremely fragile to both temperature and the electron beam, and the entire workflow is designed around that fragility.
Cryo-EM and Cryo-ET
Cryo-electron microscopy (Cryo-EM) is the umbrella term for imaging vitrified specimens with electrons; cryo-electron tomography (Cryo-ET) is one branch of it, and the two share the same dose and signal-to-noise limits. What sets them apart is what is imaged:
- Single-particle analysis (SPA) images thousands of copies of one purified molecule, each frozen in a random orientation and recorded as a single projection. Aligning and averaging those differently-oriented projections fills 3-D Fourier space in every direction — so there is no missing wedge, and near-atomic resolution is reachable.
- Cryo-ET images one unique object (a slice of a cell, say) by tilting the stage to view it from many angles. But the tilt range is limited (typically ±60°–70°), so the angles never reached leave a missing wedge; and a fixed total dose is split across many tilt images, making each one noisier.
Another way to see the split: SPA uses “many identical samples, one angle each” to assemble all directions; Cryo-ET uses “one sample, many angles” to approximate the 3-D volume. The first buys complete angular coverage with the sample’s reproducibility; the second gives up some angles for the ability to look at a unique, in-place structure. That is exactly why the missing wedge is a problem specific to Cryo-ET and absent from SPA.
Subtomogram averaging is the bridge between the two: when a tomogram contains many copies of the same structure, aligning and averaging them likewise fills in information and lowers noise — pushing Cryo-ET toward SPA-like resolution.
A crystal is water that had time to organize. Vitrification wins a race against that organization: cool the sample faster than molecules can arrange into a lattice, and they freeze in place as a disordered glass — a snapshot of the liquid state.
Biological material is extraordinarily sensitive to the electron beam. Each electron that contributes to an image also breaks bonds, so imaging operates under a strict dose budget. The total exposure (electrons per Ų) must stay low enough to keep high-resolution information intact, which forces every image to be acquired at very low signal-to-noise ratio.
Reading the symbol term by term: is the total number of electrons landing on a unit of area, in units of e⁻/Ų; the Ų in the denominator (square ångström, m) is an area at molecular scale. A larger means less statistical noise and a clearer image, but also more broken bonds and more structural damage — and high-resolution detail is what gets destroyed first. So the dose budget is fundamentally a trade-off: a compromise between “so noisy you can’t see anything” and “so much dose that the structure is wrecked.”
Tie the dose budget to signal-to-noise. The electrons recorded in each image follow counting statistics, so the number landing in a pixel fluctuates by about the square root of its expected value; the signal-to-noise ratio of a single projection therefore grows roughly as . But in tomography the total dose is fixed and must be split across tilt images, leaving only each. That is the dilemma: raising the number of tilts samples Fourier space more densely and narrows the angular gaps, but makes every image noisier (SNR drops toward ); lowering makes each image cleaner but samples the angles more sparsely. Acquisition schemes like the dose-symmetric scheme are precisely a choice of how to spend angles along this fixed dose-budget line. And because individual images are pushed to such low SNR, downstream reconstruction and restoration must model the noise explicitly rather than pretend the data is clean.
Managing this trade-off shapes the entire workflow, from acquisition through reconstruction and downstream restoration.
Vitrification only works for thin specimens, because the cryogen must remove heat from the full thickness fast enough to outrun crystallization, and electrons penetrate only a few hundred nanometers of ice. Isolated molecules and thin cellular peripheries can be frozen directly on a grid. Thicker cells and tissue must be thinned after freezing, most commonly by cryo-focused-ion-beam (cryo-FIB) milling, which carves a thin lamella — a window typically 100–300 nm thick — out of the frozen cell. The lamella then becomes the specimen for tomographic imaging, opening the door to in situ structural biology.
Cryo-FIB uses a focused beam of gallium ions like a plane to shave material away from the top and bottom faces of the frozen cell layer by layer, leaving only a thin slab in the middle. The whole operation has to stay below the vitrification temperature throughout; otherwise local heating from the ion beam, or warming during transfer, would devitrify the lamella and ruin it.
Plunge freezing and high-pressure freezing
Two freezing routes cover different thickness regimes. Plunge freezing drops a blotted grid into liquid ethane and vitrifies a film of a few hundred nanometers; it is the standard route for purified molecules and thin samples, prepared on a perforated support grid whose holes hold spans of free-standing ice. Beyond roughly a few micrometers, surface cooling alone cannot outrun crystallization in the interior. High-pressure freezing raises the sample to about 2,000 bar before cooling, which suppresses ice nucleation and growth and extends vitrification to specimens up to a couple of hundred micrometers thick, such as tissue and multicellular samples. The blotting step that sets film thickness is a frequent source of variability, since too much water prevents vitrification and too little dries or distorts the specimen.
Why pressure helps is clearest from the phase diagram: raising the pressure lowers the melting point of water and increases the supercooling needed for homogeneous crystal nucleation, which effectively narrows the temperature window in which water has a chance to crystallize. The narrower that window, the easier it is to drop through the whole of it during cooling without giving crystals a chance to nucleate — so even a sample hundreds of micrometers thick, whose interior cools slowly, can still vitrify. The cost is more complex equipment and the need to load the sample into a dedicated carrier first.
Vitreous versus crystalline ice
Vitreous ice is a metastable amorphous solid, and warming it allows the water to find its lower-energy ordered states — a transition called devitrification. On warming, vitreous ice first converts to cubic ice and then to the stable hexagonal form, both of which diffract and damage the embedded structure.
“Metastable” is the key word here: vitreous ice is not water’s lowest-energy state, only a higher-energy disordered configuration that the sample was frozen into too fast to escape — like a ball resting in a small dip partway down a hillside. Give it a little energy (that is, warm it) and it rolls toward the lower-energy ordered crystal at the bottom. That is why vitreous ice must be held down by sustained low temperature rather than trusted to stay put on its own.
That is why grids, lamellae, and every transfer step must stay below the devitrification temperature throughout preparation and imaging; a brief excursion is enough to crystallize a sample that froze correctly. The same constraint bounds lamella thickness: a thicker slab is harder to keep vitreous during milling and scatters more electrons, while too thin a slab risks removing the target, so the 100–300 nm window is a compromise between vitrification, beam penetration, and preserving the structure of interest. Because the resulting images are dose-limited and noisy, subtomogram averaging recovers high-resolution detail by combining many low-dose copies of the same complex.
Putting this page back in the context of the whole site: vitrification sets what kind of sample you can image and how thin and clean it is; the dose budget sets how clear each image can be; and the tilt range sets how large a missing wedge is left behind. Together these three physical constraints define what the data going into Cryo-ET reconstruction looks like — and the statistical machine-learning methods covered later on this site are aimed at exactly this kind of data: thin, noisy, and missing a slab of angles.