Structural Biology from Zero · 4: The Levels of Structure

Protein architecture is commonly described at four connected scales: the linear sequence, local backbone conformations, the fold of a polypeptide chain, and assemblies of multiple chains. This vocabulary provides a precise way to discuss enzymes, antibodies, and experimental structures. Each level emerges from the same underlying geometry and interactions developed in the previous articles.

Four levels, one chain

The vocabulary is old — it dates to Linderstrøm-Lang in the 1950s — and it has survived because it carves the problem at real joints.

Primary structure is the amino-acid sequence: the order of residues, read N-terminus to C-terminus, held together by the peptide bonds of Part 2. It is one-dimensional information, and by Anfinsen's hypothesis it contains everything needed to specify the rest.
Secondary structure is the local, repetitive backbone conformation stabilized by backbone hydrogen bonds: α-helices, β-sheets, and the turns that connect them. It is set by the $(\varphi,\psi)$ angles from Part 2 and lives over stretches of a few to a few dozen residues.
Tertiary structure is the full three-dimensional fold of a single polypeptide chain — how its helices and sheets pack against each other around a hydrophobic core into one compact domain.
Quaternary structure is the assembly of two or more folded chains (subunits) into a functional complex, held together by the very same non-covalent forces that drive folding, only now acting between chains.

Read that list again with Part 3 in mind and you'll see it is really one continuous process viewed at four magnifications. The hydrophobic effect and backbone hydrogen bonding don't switch off at some boundary; they operate from the scale of four adjacent residues up to the scale of a four-subunit machine. Let's take the levels in turn and make each one quantitative.

Secondary structure: the geometry of a satisfied backbone

Every residue carries a polar $\mathrm{N-H}$ donor and a $\mathrm{C{=}O}$ acceptor. Buried groups generally need suitable hydrogen-bond partners to avoid an energetic penalty. Two especially common repetitive arrangements satisfy many of these groups using backbone atoms: the α-helix and β-sheet. Linus Pauling and colleagues derived their geometries in 1951 from the planar peptide bond and hydrogen-bond constraints.

The α-helix

The α-helix satisfies the backbone hydrogen bonds within a single continuous stretch of chain. The chain coils into a right-handed spiral in which the $\mathrm{C{=}O}$ of residue $i$ points along the helix axis and hydrogen-bonds to the $\mathrm{N-H}$ of residue $i+4$ . Every backbone group is satisfied except the four donors at one end and four acceptors at the other. The geometry is remarkably precise and worth memorizing because it recurs everywhere:

3.6 residues per turn.
1.5 Å rise per residue along the helix axis.
Therefore a pitch (axial distance per full turn) of $3.6 \times 1.5 = 5.4\ \text{Å}$ .
Backbone angles clustered near $\varphi \approx -57°,\ \psi \approx -47°$ — the right-handed α-basin you dragged the marker into in Part 2.

That the numbers close consistently is a genuine check, not a coincidence. If you place residues on a cylinder at $100°$ of rotation each ( $360°/3.6 = 100°$ ) and $1.5\ \text{Å}$ of rise, the $i \to i+4$ pair comes out $2.8\text{–}3.0\ \text{Å}$ apart at close to the linear donor–acceptor geometry a hydrogen bond wants — which is precisely why $i{+}4$ , and not $i{+}3$ or $i{+}5$ , is the rule. The side chains all point outward from the axis, so a helix presents a cylinder with residues arrayed around it every $100°$ ; this is what lets one face of a helix be hydrophobic and the other polar, a fact we'll use for both membrane proteins and coiled coils below.

Two close relatives round out the family. The 3₁₀ helix hydrogen-bonds $i \to i+3$ , making a tighter coil with exactly 3 residues per turn; it shows up in short segments and at the ends of α-helices. The π-helix bonds $i \to i+5$ , is wider and slightly strained, and is rare. All three are just different registers of the same idea — bond the backbone to a partner a fixed number of residues down the chain.

The β-sheet

The β-sheet solves the same bookkeeping problem the opposite way: it satisfies backbone hydrogen bonds between separate stretches of chain lying side by side. An individual β-strand is nearly fully extended — $\varphi \approx -120°,\ \psi \approx +130°$ , the upper-left basin of the Ramachandran plot — with about 3.5 Å of rise per residue, more than twice the helix's. A single strand satisfies none of its own hydrogen bonds; it must pair with neighbors. When several strands align, their $\mathrm{N-H}$ and $\mathrm{C{=}O}$ groups zip across the gap between strands to form a pleated sheet.

Strands can align in two ways, and the distinction is real, not cosmetic:

In an antiparallel sheet, neighboring strands run in opposite N→C directions. Their hydrogen bonds are comparatively linear and evenly spaced.
In a parallel sheet, neighboring strands run in the same direction. The hydrogen bonds are angled, and parallel topology often appears as part of larger recurring folds such as α/β barrels and sheets.

Real sheets are usually slightly twisted rather than flat — a right-handed twist of a few degrees per strand that comes out of the chirality of L-amino acids — and this twist is what curls extended sheets into the β-barrels and saddle-shaped surfaces you meet in real folds.

Turns and loops

Helices and sheets are the ordered elements, but a chain also has to change direction, and it does so in turns and loops. The tightest is the β-turn: four residues that reverse the chain's direction, often with a hydrogen bond from the $\mathrm{C{=}O}$ of the first to the $\mathrm{N-H}$ of the fourth. Turns are where glycine (flexible, able to access the forbidden Ramachandran regions) and proline (which kinks the backbone) earn their keep, and it is no accident they are enriched there. Loops are longer, irregular connectors, usually on the protein's surface — and, as we'll see with antibodies in Part 6, the irregular surface loops are exactly where biology puts the sequence variation that creates specificity. Order does the structural work; the disordered connectors do much of the recognition.

Supersecondary structure: the recurring motifs

Between "a helix" and "a whole fold" sits an intermediate scale that is easy to miss but pays off enormously in pattern recognition. Secondary-structure elements don't combine arbitrarily; a small number of motifs (supersecondary structures) recur across otherwise unrelated proteins because they are the geometrically and energetically easy ways to connect elements. Jane Richardson catalogued these in 1981 in a paper that still repays reading. A few you will see constantly:

The β-hairpin: two adjacent antiparallel strands joined by a tight turn.
The β-α-β motif: two parallel strands connected by an α-helix that packs against them — the building block of the most common α/β folds.
The Greek key: four antiparallel strands connected so the topology traces the pattern from Greek pottery; the core of many β-sandwich domains, including the immunoglobulin fold.
The coiled coil: two or more α-helices wound gently around each other. This one is worth a sentence of physics because it shows the geometry paying off directly. Coiled-coil helices carry a seven-residue (heptad) repeat, conventionally lettered $a$ through $g$ , with hydrophobic residues at the $a$ and $d$ positions. Because the helix turns every 3.6 residues, those two positions fall on the same face of the cylinder, forming a continuous hydrophobic stripe that lets two helices interlock in a "knobs-into-holes" packing that Francis Crick worked out in 1953. It is a lovely case of the $100°$ -per-residue helix geometry directly dictating a quaternary interaction.

Learning to see these motifs is most of what "reading a structure" feels like in practice: a fold that looks like an incomprehensible tangle resolves, once you know the motifs, into a short sentence like "a Greek-key β-sandwich" or "a Rossmann-fold α/β domain."

Tertiary structure: the domain

Zoom out once more and the motifs pack into a domain — a compact, independently folding unit, typically 100–200 residues, with its own hydrophobic core. The domain is arguably the most important single concept in structural biology, for three reasons that all point the same way. It is the unit of folding (many multidomain proteins fold one domain at a time); the unit of function (a domain often carries one binding site or one catalytic activity); and the unit of evolution (domains are shuffled between proteins as modular cassettes, which is why the same domain turns up in wildly different proteins). A large protein is best read not as one enormous fold but as a few domains on a string.

Because domains are the recurring vocabulary, biologists have catalogued them. The two classic databases, SCOP (Murzin and colleagues, 1995) and CATH (Orengo and colleagues, 1997), classify domains hierarchically, and both converge on the same top-level split by secondary-structure content:

all-α — domains built essentially only from helices (the globin fold of hemoglobin is the textbook case).
all-β — domains built essentially only from sheets (the immunoglobulin β-sandwich of Part 6).
α/β — alternating helices and parallel strands, usually as repeated β-α-β units (the TIM barrel and the Rossmann fold; these are the most common enzyme scaffolds).
α+β — helices and antiparallel sheets in separate regions of the domain.

These catalogues show extensive reuse of domain architectures. The number of folds depends on the classification scheme and continues to change as structures are deposited, but a comparatively modest set of recurring architectures accounts for much of the known structural database. This reuse helps structure prediction and remote-homology detection because new sequences often adopt variations of previously observed folds.

Every domain, whatever its class, obeys the Part 3 rules internally: hydrophobic residues buried in a tightly packed core, polar and charged residues on the water-facing surface, and the packing so complete that a folded domain's interior has about the density of an organic crystal. When we get to enzymes in Part 5, the active site will turn out to be a deliberate local violation of this rule — a pocket where the protein pays to place reactive polar groups exactly where the chemistry needs them.

Quaternary structure: assembling the machine

Many proteins are not functional as a single chain. Quaternary structure is the assembly of multiple folded subunits into one complex — hemoglobin's four chains, a viral capsid's hundreds, the ribosome's dozens of proteins and RNAs. The interfaces holding subunits together are made of exactly the forces from Part 3 (the hydrophobic effect, hydrogen bonds, salt bridges, van der Waals packing), now acting between chains instead of within one. A subunit interface is, physically, just a patch of "core" that happens to fall between two molecules.

A recurring and initially surprising feature is symmetry. Most oligomeric proteins are symmetric — their subunits are related by rotation axes — and there is a clean argument for why. If two copies of the same subunit stick together using a specific patch, the lowest-energy, self-limiting way to do it is a symmetric arrangement, which is why identical subunits so often form cyclic ( $C_n$ , a single $n$ -fold rotation axis) or dihedral ( $D_n$ , an $n$ -fold axis plus perpendicular 2-fold axes) point-group symmetries. Hemoglobin is an $\alpha_2\beta_2$ tetramer with approximate $D_2$ symmetry (a dimer of αβ dimers); many enzymes are $C_2$ or $D_2$ ; and viral capsids exploit icosahedral symmetry to build a huge shell from one gene's worth of protein.

Why assemble at all, rather than encode one bigger chain? Several real advantages, each worth stating:

Genomic economy and error control. Building a large machine from many copies of a small subunit means one short gene, and it means a mistranslated subunit can be discarded rather than ruining an enormous single chain.
Allostery. This is the big one. When subunits touch, an event at one site can be transmitted through the interface to change another site's behavior — the structural basis of regulation. Hemoglobin is the canonical example: binding oxygen at one heme shifts the whole tetramer from a low-affinity T (tense) state toward a high-affinity R (relaxed) state, so the four subunits bind oxygen cooperatively rather than independently. Max Perutz spent much of his career working out the stereochemistry of this transition from crystal structures, and Monod, Wyman, and Changeux captured it in 1965 in the elegant two-state MWC model. Cooperativity is why hemoglobin's oxygen-binding curve is a sigmoid rather than a hyperbola, and that sigmoid shape is what lets it load oxygen efficiently in the lungs and dump it in the tissues.
Avidity and new geometry. Multiple copies of a binding site (an antibody has two identical arms) bind targets far more tightly than one, and assembly builds channels, pores, and enclosed reaction chambers that no single chain could form.

Quaternary structure, in short, is where a folded protein stops being a molecule and starts being a machine.

The interactive: one structure, four levels

I built a viewer that shows all four levels on real, deposited structures — because the hierarchy is far more convincing when you can step up it yourself than when you read a list. Start with hemoglobin, the molecule Perutz solved: view it as a bare backbone trace (primary), color it by secondary-structure type to see it is almost entirely α-helix, isolate a single β-subunit with its heme and buried core (tertiary), then assemble all four chains and see the $\alpha_2\beta_2$ symmetry that makes cooperativity possible (quaternary). Switch to triosephosphate isomerase to see the other great architecture — an α/β "TIM barrel" of eight parallel strands wrapped by eight helices, which also happens to be a dimer, so it too has all four levels.

Interactive. Step through primary → secondary → tertiary → quaternary structure on two real proteins from the PDB: hemoglobin (2HHB) and triosephosphate isomerase (1TIM). Secondary structure is colored by type (helix / sheet / loop); the tertiary view isolates one subunit and its core; the quaternary view colors each chain to show the symmetric assembly. Opens full-screen in a new tab.

Intrinsic disorder

The four-level description is most direct for proteins with well-defined folds. Many intrinsically disordered proteins and regions (IDPs/IDRs) instead populate fluctuating ensembles while remaining functional. Estimates put significant disorder in roughly a third of eukaryotic proteins, concentrated in signaling and regulatory roles.

This is not a contradiction of Anfinsen so much as a boundary condition on him. Disordered regions have a characteristic sequence signature — low overall hydrophobicity and high net charge — so they cannot bury a hydrophobic core and therefore never collapse into a stable fold; the physics of Part 3 predicts their disorder directly. Many of them fold upon binding a partner, paying the chain-entropy cost only when the enthalpic reward of a specific interface is there to cover it. They matter here for two reasons: they are why sequence-based structure predictors must learn to say "this region has no single structure" rather than hallucinate one, and they are a reminder that the four-level hierarchy describes the folded majority, not a law without exceptions.

Where this leaves us

The four levels are not four separate subjects; they are one physical process — the burial of hydrophobic surface and the satisfaction of backbone hydrogen bonds — resolved at four magnifications. Local $(\varphi,\psi)$ preferences give helices and sheets; the economical ways to connect those give the recurring motifs; motifs pack around a hydrophobic core into domains drawn from a finite catalogue of folds; and domains assemble, usually symmetrically, into the allosteric machines that actually run a cell. Every later article in this series leans on this vocabulary. In Part 5 we use it to understand catalysis — how a specific local arrangement of a few residues in a domain's active site can accelerate a reaction by a factor of $10^{17}$ — and we'll watch the general rule "bury the polar groups" get deliberately broken to make chemistry happen.