How can frozen molecules reveal the shape of a protein?

A protein's detailed shape can be inferred by combining electron measurements of many copies viewed from different directions. Single-particle cryo-electron microscopy holds those copies in rapidly frozen water. The water becomes vitreous, a glass-like solid without large ice crystals that could disrupt the sample, preserving particles for measurements an ordinary light microscope cannot provide.

Electrons passing through a particle provide a two-dimensional projection of its structure. One view compresses depth information: features at different positions along the viewing direction can overlap. Like the shadow of a complicated object, a projection can therefore fit more than one possible three-dimensional arrangement.

The sample contains copies lying in different orientations. A feature hidden behind another in one projection can be separated in another. Computational methods group similar views, estimate their orientations and seek a three-dimensional density map whose projections agree with the measurements. The views constrain one another rather than merely being stacked as unrelated flat pictures.

Each image is noisy because exposing a biological sample to too many electrons damages it. Combining many sufficiently similar particles improves the usable information. But differences between molecular states also matter: averaging unlike shapes indiscriminately can blur real features. Researchers may separate groups and assess which details each reconstruction supports before fitting a molecular model.

The resulting structure is an inference checked against measurements, with resolution and uncertainty varying across the map. Preferred particle orientations can leave some information weakly constrained. Different frozen states can offer evidence about protein changes, but they are not a movie of one living molecule; the reconstruction gains depth by bringing many partial views into agreement.