The lecture opens deliberately outside biology. Structure is defined as the way the parts of a system are arranged or organised. Why it matters is the same in engineering, geology, science and social studies: once you know the arrangement, you can decode the system, predict outcomes, and carry knowledge across disciplines. The deck's own example is a tectonic framework over Cambrian salt basins.

Structural biology then applies that idea to living material — formed, composed, maintained and refined by living cells — at every organisational level. The reason it matters clinically is the oldest idea in the field: function follows form. A structure tells you which residues touch, which pocket a ligand occupies, and therefore what a mutation or a drug would do.

Optical microscopy stops at roughly 0.2 µm — enough for mitochondria and bacteria, not for molecules. Two walls: the diffraction limit (visible light is 400–750 nm, so nothing smaller than ~200 nm resolves) and photon damage (higher resolution needs more intense light, which destroys the sample).

The lecture then redefines seeing. A model is what lets you understand and interpret a structure; a good model (i) captures all key elements and relationships, (ii) is verifiable and predictive, and (iii) stays simple enough to communicate. This is scientific constructivism: a scientist “sees” a protein through a model built from experimental data, not by looking at it.

Three ideas, in order. Order: a crystal is a solid whose molecules sit in a long-range, orderly arrangement — that repeating lattice is what turns scattering into sharp spots. Diffraction: high-energy X-rays passing through the crystal are scattered, following Bragg's Law. Inversion: the diffraction pattern is mathematically transformed into an electron density map, and a molecular model is fitted into it.

Resolution is the only currency here, and the deck gives it as a ladder:

At high resolution the method answers the question industry actually pays for: how does a drug bind? The deck's two examples are databases rather than single papers — a collection of known enzyme mechanisms (M-CSA) and coordinate/ligand records in the PDB.

Note how much of this is now automated: data processing and model building are largely software-driven, so the human decisions concentrate at step 0–1 (can I make the protein, can I crystallise it) and at step 6 (is the answer real).

Small molecules crystallise easily — evaporation, cooling, organic solvents. Macromolecules do not: they need high purity, controlled conditions and specialised methods. The classic recipe is buffer (pH for stability) + salt (ammonium sulfate to lower solubility) + precipitant (PEG, to dehydrate gradually).

Four standard techniques: vapour diffusion (hanging drop over a reservoir, or sitting drop beside it — both rely on gradual evaporation), microbatch (mixed under oil, no evaporation), and microdialysis (dialyse against precipitant). Modern practice adds commercial screening kits, robotic dispensing (down to ~100 nL per condition), automated imaging, fine screening of pH/concentration/temperature, and seeding — introducing crushed microcrystals to nucleate better ones.

Complexes are the interesting cases. For protein–ligand work you either co-crystallise (ligand added before) or soak (ligand diffused into existing crystals); the constraints are binding affinity, the protein:ligand ratio, and stabilisation of the complex. For protein–protein or protein–nucleic-acid complexes the constraints shift to complex stability, flexibility of large assemblies, and stoichiometry — sometimes solved by fusing partners.

Cryo-EM was developed in the 1980s but spent decades as “blobology” — low-resolution shapes. Three advances made atomic resolution routine after ~2013: direct electron detectors (better sensitivity and contrast), better image-processing algorithms (averaging thousands of images to beat noise and radiation damage), and cryo-stage stability (less movement, sharper images). The 2017 Nobel Prize in Chemistry marked the arrival.

The instrument: a beam of electrons as the “light”, magnetic coils as lenses, high vacuum throughout, and image formation from elastically scattered electrons — those that change direction but keep their energy, carrying the structural information.

Why a single molecule is not enough: biological samples are weak scatterers (C, N, S, O, H against water) so contrast is intrinsically low, and even frozen samples suffer radiation damage. The answer is statistics — assume a homogeneous population, collect a large dataset, classify and align, then average. Fine detail emerges from the average, not from any one image.

The processing pipeline: micrographs → particle picking → 2D classification and averaging → CTF correction → initial 3D model → 3D classification and refinement → post-processing and validation. Published benchmarks in the deck: β-galactosidase at 3.2 Å (Bartesaghi 2014), GroEL at 3.5 Å (Roh 2017).

NMR works on the magnetic behaviour of nuclei (¹H, ¹³C, ¹⁵N) in a strong field: they absorb and emit radiofrequency energy, and the signal is exquisitely sensitive to the local chemical environment, reporting distances up to ~5 Å and bond angles. Its niche is solution structure of small proteins (< 30 kDa), real-time dynamics and interactions, and being non-destructive under near-physiological conditions. Its costs are size, spectral complexity, mM concentrations of very pure protein, and expensive high-field instruments — which is why cryo-EM and crystallography took most of its territory.

SAXS measures X-ray scattering at small angles in solution to report size, shape and conformation. It is explicitly a low-resolution technique, needs very little protein (0.1–0.5 mg), and is extremely sensitive to aggregation — hence the standard practice of running it inline with SEC, so only the clean monomer reaches the beam.

Two case studies worth the detail: a de novo designed transmembrane protein where the NMR structure (PDB 9FDG) and the crystal structure (8UZL) both match the computational design (Berhanu et al., Science 2024) — two independent techniques validating a design; and a JACS 2018 study using SAXS to distinguish different conformations of the same molecule in solution.

The lecture is explicit that this block is about technological leadership, not bench craft. Protein expression, purification and crystallisation are mature enough for CRO outsourcing; specialised companies can deliver high-purity, active protein in roughly two weeks, often at lower cost, which frees the research team to do the structural analysis.

1. Install PyMOL (pymol.org) and work through the tutorial — it is used in class next week.
2. Skim paper 1 for how the authors expressed and crystallised NDM-1.
3. Read paper 2 for how the co-crystal structures of benzylpenicillin, oxacillin, methicillin and meropenem were obtained, then load them in PyMOL and look at the bound small molecules.
4. Read one of the earliest reports of the SARS-CoV-2 spike cryo-EM structure.

Reference shelf from the deck: RCSB PDB (rcsb.org) · M-CSA enzyme mechanisms (EBI) · UniProt · Caltech cryo-EM course · cryoem101.org · SSRL SAXS workshop material.