Cryo-Electron Microscopy

Cryo-electron microscopy (cryo-EM) determines three-dimensional structures of biological macromolecules by recording two-dimensional projection images of vitrified single particles in random orientations and reconstructing the underlying density map computationally. The “resolution revolution” of the 2010s closed the gap between cryo-EM and X-ray crystallography for many targets; the methodological frontier has since shifted to four axes that determine which problems are now tractable: sample preparation (the air-water interface and grid-deposition pathologies that destroy or denature particles before they reach the microscope), signal extraction at low dose (squeezing more information from each electron, since beam damage caps the dose any specimen can absorb), particle identification and map interpretation (turning megapixel micrographs into reliable particle sets and turning density maps into atomic models), and modality extension (recording information beyond elastic scattering — chemical identity, lipid environment, conformational ensembles).

Surviving the grid: sample preparation as the new bottleneck

For many biologically interesting complexes, the limiting step is no longer microscopy but biochemistry on the grid: particles preferentially adsorb to the air-water interface where they denature, orient anisotropically, or aggregate during the milliseconds between blot and plunge. Zheng et al. (2024) attack this directly with a self-assembled superstructure that protects particles from the air-water interface during vitrification, producing isotropic particle distributions for targets that previously yielded preferred-orientation reconstructions only. The methodological lesson — engineer the grid environment rather than the protein — has reshaped how challenging targets are approached: many “cryo-EM failures” are now reframed as sample-preparation failures with grid-chemistry solutions.

Squeezing the dose budget

Beam damage sets a hard ceiling on how many electrons can be delivered to a single particle, which in turn sets a floor on the noise of each image and a ceiling on the resolution recoverable from any number of particles. Küçükoğlu et al. (2024) demonstrate low-dose cryo-electron ptychography of proteins at sub-nanometer resolution: rather than collecting conventional bright-field images, they record diffraction patterns from a focused probe and reconstruct phase information that conventional imaging discards. The result is a higher information-per-electron yield, which translates into reachable resolutions at doses that classical single-particle workflows cannot exploit. The methodology imports techniques from materials-science electron microscopy into the biological pipeline and reopens the question of what the optimal imaging mode is once dose, not optics, is the binding constraint.

From micrographs to atomic models

Even with the microscope solved, the computational pipeline has its own bottlenecks. Gyawali et al. (2024) attack the first one — particle picking — with CryoSegNet, a hybrid system that combines a foundation-model image segmenter with an attention-gated U-Net specialised for cryo-EM micrographs; the result is robust particle picking even on noisy or contaminated grids where classical template-based pickers fail. Maddhuri Venkata Subramaniya et al. (2023) attack the downstream end of the pipeline with 3D deep generative networks that enhance cryo-EM density maps to assist atomic-model building: the network learns the local correspondence between observed density features and the underlying atomic motifs, and effectively post-processes a map into a form that downstream model-building tools fit more accurately. Together the two papers map a pattern across modern cryo-EM software: replace bespoke filters and hand-tuned heuristics with learned models that absorb the long tail of failure modes the previous generation accumulated workarounds for.

Reading more than density

Cryo-EM has classically delivered electrostatic-potential density and little else. Pfeil-Gardiner et al. (2024) extend the readout with elemental mapping in single-particle reconstructions by reconstructed electron energy-loss analysis: by recording EELS spectra alongside elastic scattering and back-projecting them into the same reconstructed volume, they distinguish chemical species (notably metals) at specific atomic positions within the macromolecule. The methodology adds a chemical axis to a previously chemistry-agnostic technique. Ansell et al. (2023) extend the readout in a different direction with LipIDens, a simulation-assisted framework that interprets the lipid-shaped densities surrounding membrane proteins in cryo-EM maps: rather than ignoring or hand-modelling these blobs, LipIDens uses molecular-dynamics simulations to assign specific lipid species and orientations to the observed density, turning a routinely-discarded feature into an annotated lipid-protein interaction map. Open methodological questions span the four axes: how far can ptychography and other phase-recovery modalities push the resolution-versus-dose frontier, can foundation-model approaches generalise across both single-particle and tomography pipelines, what is the right computational treatment of conformational ensembles versus discrete states, and how reliably can chemical-identity and lipid-environment readouts be made quantitative rather than qualitative?