Determining Protein Structure

Introduction

A protein's atomic model is an interpretation of experimental measurements, not a direct photograph of each atom. X-ray crystallography, nuclear magnetic resonance (NMR) and cryogenic electron microscopy (cryo-EM) each probe structure in a different physical way. Their strengths depend on sample size, flexibility, crystallisability and the scientific question. A structure can reveal a binding pocket or an interface, but confidence in any claimed contact depends on the quality of the underlying data.

Core explanation

In X-ray crystallography, a purified protein is arranged in a crystal so many copies diffract X-rays coherently. Recorded diffraction intensities and estimated phases are used to calculate an electron-density map; an atomic model is built and refined against the data. Crystallography can give detailed coordinates for ordered regions, but producing suitable crystals can be difficult. Flexible loops may give weak or missing density, and crystal packing or sample conditions can favour a particular conformation. A crystal structure is not proof that the protein is rigid in solution.

Solution NMR uses the responses of certain atomic nuclei in a magnetic field. Chemical shifts, coupling and nuclear Overhauser effects provide information about local environment, connectivity and spatial proximity. Researchers use many restraints to calculate conformations consistent with the spectra, often reporting an ensemble of models. NMR can study dynamics and interactions in solution, but spectral complexity, sample concentration and molecular size may limit a particular project. It is inaccurate to say NMR simply measures every interatomic distance directly.

Single-particle cryo-EM vitrifies a purified sample in a thin film of amorphous ice and records many low-dose electron images of particles in different orientations. Computational classification and averaging reconstruct a three-dimensional density map. This approach has been especially powerful for large complexes and some membrane proteins that are difficult to crystallise. Resolution can vary within one map: a rigid core may be well resolved while a flexible domain remains blurred. Image processing helps recover a signal from noisy exposures, but a model still has to be fitted and validated against the map.

“Resolution” is useful but not a universal certificate of accuracy. A nominal global resolution may hide local variation, and the ability to place a side chain depends on local data quality. A model can also include assumptions about stereochemistry and known bond lengths. Inspecting maps, uncertainty, clashes, Ramachandran outliers and missing residues helps determine what conclusions are supported. A predicted protein structure can guide hypotheses, but it is not the same as an experimental measurement of a particular ligand-bound state.

Methods can complement one another. An X-ray or cryo-EM model may reveal an overall fold, while NMR or other biophysical measurements detect motions or binding changes in solution. Comparing structures with and without a ligand may suggest a conformational mechanism, but the difference could also reflect crystal packing, sample preparation or different experimental conditions. Functional tests and kinetic measurements are needed to establish the consequences of a structural change.

Step-by-step reasoning

When choosing a method, ask whether the target can form crystals, whether it is small enough and well behaved for solution NMR, and whether it is suitable for cryo-EM particle imaging. For any reported structure, identify the measured observable: diffraction intensities, NMR signals or electron images. Then examine the map or restraints, the local model quality and the sample conditions. Only after those checks connect an observed geometry to a proposed biochemical function.

Visual explanation

Draw three parallel pipelines. The X-ray path runs crystal → diffraction spots → electron-density map → model. The NMR path runs protein solution → spectra and restraints → conformational ensemble. The cryo-EM path runs vitrified particles → many two-dimensional images → classified three-dimensional map → model. At the end of each path, put a question mark beside flexible loops to show that missing or weak data limit local interpretation.

Real-world analogy

One can infer a sculpture's shape from its shadow patterns, sound reflections or many photographs taken at different angles. Each measurement constrains the same object differently and leaves some ambiguity. Protein methods likewise reconstruct structure from physical signals, although their mathematics and sample states are quite different from everyday imaging.

Real-world example

A large membrane-embedded enzyme complex might resist crystallisation and be too large for routine high-resolution solution NMR. Cryo-EM could reveal the arrangement of its major subunits and a bound inhibitor. If the inhibitor pocket is locally blurred, claiming the exact orientation of every ligand atom would exceed the data; biochemical binding tests and improved maps could strengthen the interpretation.

Why?

Why might a flexible loop be absent from a deposited crystallographic model? If it adopts many conformations, its averaged electron density may be too weak to place a unique set of coordinates confidently. The loop still exists in the protein sequence; “not modelled” does not mean chemically absent.

Common misconception

“A high-resolution structure shows the protein's one true shape.” Proteins fluctuate, and experiments sample particular conditions and ensembles. An atomic model can be accurate for ordered regions while still missing transient conformations or poorly resolved segments.

Worked example

A cryo-EM study reports a well-resolved catalytic core but a mobile regulatory domain with weak local density. A student proposes that one specific lysine in the domain forms a salt bridge to the core. The proposed bridge is plausible from chemistry but not yet established by the map: the lysine side-chain orientation may be unresolved. A careful report would describe the domain's approximate location, flag the uncertain contact and test it through mutagenesis or higher-quality structural data.

Quick check

1. Which method requires a crystal of the target in its conventional form? Answer: X-ray crystallography; solution NMR and single-particle cryo-EM use different sample preparations and observables.

Exam focus

Match each method to its measured signal and principal limitation. Avoid describing an atomic model as a direct photograph. Use local model quality and biological conditions to judge whether a claimed catalytic contact or conformational change is supported.

Advanced insight

Structural refinement is an inverse problem: many possible models can sometimes fit limited or noisy data. Stereochemical restraints reduce ambiguity, but they can also make a weakly supported region look deceptively plausible. Validation against information not used to fit the model, along with comparison across methods, makes structural conclusions more robust.

Summary

X-ray crystallography infers structure from diffraction by crystals, NMR uses solution-state nuclear signals, and cryo-EM reconstructs maps from vitrified particle images. Each produces a data-constrained model with method-specific strengths and uncertainties. Interpret biological mechanisms in light of local evidence, dynamics and sample conditions.

Practice questions

1. Why might cryo-EM be attractive for a large multiprotein assembly that does not crystallise? Answer: Single-particle cryo-EM can reconstruct a three-dimensional map from images of vitrified particles without requiring a crystal, and large ordered assemblies often give useful image information. 2. A crystal structure contains no coordinates for residues 120–135 although the sequence includes them. Give a likely explanation. Answer: Those residues may form a flexible or disordered segment whose electron density does not support one reliable conformation. Their absence from the coordinate model does not imply deletion from the protein. 3. What additional evidence would strengthen a claim that a modelled pocket is catalytically important? Answer: Functional measurements such as activity changes after targeted mutation, ligand-binding data, and clear local experimental density or restraints would connect the structural geometry to biochemical function.