Structural biology and cryo-electron microscopy

Single-particle cryo-electron microscopy determines a three-dimensional structure without a crystal. A thin aqueous film containing the purified molecule is frozen, imaged by transmission electron microscopy, and hundreds of thousands of individual particle images — each a noisy projection from an unknown direction — are computationally assigned orientations and combined into a density map into which an atomic model is built.

Vitrification, not freezing

The specimen must be frozen fast enough that water solidifies without crystallising. Crystalline ice diffracts the electron beam and destroys the image, and its formation also excludes solute, concentrating salts and damaging the specimen. Plunging a sub-micrometre film into liquid ethane achieves cooling rates high enough to trap water in an amorphous, glass-like state. The specimen is then imaged at cryogenic temperature under a strictly limited electron dose, because the beam damages the sample: radiation damage, not detector performance, sets how much signal can be collected per particle, which is why thousands of particle images are needed to average the noise down.

The air-water interface is the practical problem

In the seconds before freezing, the film is only tens of nanometres thick and the molecules diffuse across it repeatedly. Most encounter the air-water interface many times, and that interface is denaturing: it can unfold protein, dissociate complexes and, for those that survive, impose a preferred orientation because a particle adsorbs in a particular attitude. This is the origin of most failed grids, and of the field’s characteristic remedies — surfactants, support films of graphene or amorphous carbon, affinity supports that tether the particle away from the interface, and faster freezing devices that shorten the exposure time.

Preferred orientation matters because reconstruction needs views distributed over all directions. If particles adopt only a narrow set of attitudes, the map is anisotropic: sharp in some directions and smeared in others, with the average resolution figure concealing the defect. Tilting the stage during collection is the standard mitigation, at the cost of thicker effective ice and worse images.

Heterogeneity, not the microscope

Modern instruments and direct-electron detectors are, for most projects, no longer the constraint. What limits resolution is that averaging assumes the particles are identical, and biological molecules move. Flexible domains, partially occupied ligands, compositional variability and continuous conformational motion all blur the average. Classification separates discrete states and can recover higher resolution for each at the cost of fewer particles per class, and methods that model continuous motion are an active area, but a mobile region will remain poorly resolved no matter how many images are collected.

Reading a resolution number

Reported resolution is normally a global estimate from Fourier shell correlation between two independently reconstructed half-maps, conventionally quoted at a threshold of 0.143. It is an average over the whole map. A structure quoted at high global resolution routinely contains a well-ordered core near that value and peripheral regions several times worse, which is why local resolution estimates and the deposited half-maps matter more than the headline figure. The correct question about a published structure is not its number but whether the density supports the specific feature being claimed.