Tertiary Structure and the Fold

A protein is a string until it folds; then it becomes a machine. Myoglobin's 153 amino acids, written out, are a monotonous sequence of one-letter codes. There is no obvious logic to this sequence. Fold that same chain in water, at 37 °C, in under a second. It becomes a heme-cradling oxygen sponge with a defined pocket. It gains a defined entrance channel, and a defined affinity constant. Tertiary structure is the name for this three-dimensional arrangement. It is the specific way a single polypeptide chain packs its helices, sheets, and loops into one compact, functional object. Nothing about the primary sequence looks like an answer. Yet somehow the sequence is the answer: this is Anfinsen's founding observation. It is the reason tertiary structure prediction was, for fifty years, biology's most famous unsolved problem. Why does this matter beyond an exam question? Because function lives in the tertiary fold, not in the sequence or even in the secondary structure alone. Two helices sitting near each other in a folded protein can form a ligand-binding cleft. Neither helix has this cleft on its own. An active-site aspartate and a histidine sit 80 residues apart in the primary sequence. They become adjacent once the chain folds. Their spatial adjacency — not their sequence proximity — is what makes catalysis possible. Misfolding, by the same logic, is not a cosmetic defect. It is the direct molecular cause of diseases. These range from cystic fibrosis (ΔF508 CFTR misfolds and is degraded) to Alzheimer's disease (amyloid-β aggregates through corrupted tertiary/quaternary packing). The three-dimensional fold is where sequence information becomes biological information. The big picture is a hierarchy of stabilization. Backbone hydrogen bonds fix local secondary structure. But it is the burial of nonpolar side chains away from water that matters most. This burial — the hydrophobic effect — drives collapse into a compact globule. It is a dense, precisely shaped hydrophobic core that locks the tertiary architecture in place. This core is reinforced by van der Waals packing, hydrogen bonds, ion pairs, and sometimes disulfide bonds. The fold is a free-energy minimum — actually a shallow, marginal one. It is typically only 20–65 kJ/mol more stable than the unfolded state. It is reached not by exhaustive search; Levinthal's paradox forbids that. Instead, it is reached by a funneled, cooperative folding process. This article builds the vocabulary and physical logic of the tertiary fold. It covers how domains, motifs, and cores are organized in space. It covers the energetic bookkeeping that makes folding thermodynamically favorable despite a large entropic cost. It covers the kinetic funnel that makes folding fast. It also covers the classification schemes (all-α, all-β, α/β, α+β). These schemes let structural biologists organize the thousands of folds now known. It sets up the transition from a single folded chain to multidomain architecture. It also sets up the assembly of folded subunits into oligomers, using myoglobin as the running structural example throughout.

Key takeaways

Tertiary structure is the complete 3D arrangement of one polypeptide chain, packing secondary-structure elements into a compact, often globular, fold.
Anfinsen's thermodynamic hypothesis: sequence alone specifies the native fold, which is the global (or a deep local) free-energy minimum under physiological conditions.
The hydrophobic effect drives collapse; van der Waals packing, hydrogen bonds, and ion pairs then lock in a specific fold rather than any compact blob.
Net folding stability (ΔG°_fold) is small, 20–65 kJ/mol, because large favorable enthalpy/hydrophobic terms nearly cancel a large unfavorable conformational entropy loss.
Levinthal's paradox is resolved by a funneled energy landscape, not exhaustive conformational search — folding is fast because it is thermodynamically biased at every step.
Folds are classified by secondary-structure content and topology (all-α, all-β, α/β, α+β) into families like the globin fold, TIM barrel, and immunoglobulin fold.
Key idea: tertiary structure is a marginally stable, kinetically accessible free-energy minimum in which a hydrophobic core, built from packed secondary-structure elements, converts sequence information into a specific, functional three-dimensional shape.

Where this lives today

Tertiary structure prediction, once biology's grand unsolved problem, was substantially transformed by deep-learning methods (AlphaFold2, RoseTTAFold, and successors), which predict native-like tertiary folds directly from sequence (plus evolutionary co-variation information) with accuracy rivaling experimental crystallography for a large fraction of single-domain proteins — a direct, technological vindication of Anfinsen's thermodynamic hypothesis, since these methods implicitly learn the sequence-to-structure mapping that the hypothesis says must exist. This has reshaped structural biology workflows: structures for previously "unsolved" proteins across entire proteomes are now predicted computationally as a starting hypothesis, to be tested and refined by experiment (cryo-EM, crystallography, NMR). Rational and computational protein design (Rosetta, and newer deep generative methods) works in the reverse direction — specifying a target tertiary fold (a binding pocket, a catalytic scaffold, a self-assembling nanostructure) and then designing a sequence predicted to fold into it — producing de novo enzymes, vaccine immunogen scaffolds, and engineered binding proteins that did not exist in nature. Disease-relevant misfolding is a direct clinical application of the fold-stability concepts developed here: cystic fibrosis's most common mutation (ΔF508) destabilizes the tertiary fold of CFTR's nucleotide-binding domain enough that quality-control machinery degrades it before it reaches the membrane, and correctors like lumacaftor/tezacaftor work by chemically stabilizing the mutant fold's marginal $\Delta G^{\circ}_{\text{fold}}$ . Amyloid diseases (Alzheimer's, Parkinson's, transmissible spongiform encephalopathies) represent the pathological alternative outcome of the funneled landscape picture — sequences or conditions in which a competing, highly stable cross-β aggregate state lies at lower free energy than, or is kinetically more accessible than, the correctly folded native monomer. Structure-based drug design routinely exploits tertiary pockets (like myoglobin's heme crevice, but for kinases, proteases, GPCRs) revealed only by the folded three-dimensional structure, not by sequence alone — the entire discipline of "structure-based" (as opposed to purely sequence- or ligand-based) drug discovery rests on the fact that binding sites are tertiary-structure features.

Ask the tutor: How does a molten globule intermediate differ energetically and structurally from both the fully unfolded chain and the final native tertiary fold?

The physical picture

Start from a folded secondary-structure toolkit — α-helices, β-strands connected by turns, the βαβ and helix-hairpin-helix motifs already assembled locally. Tertiary structure is the next level up: it specifies how these elements pack against one another in three dimensions to bury nonpolar surface and build a single coherent domain. Two organizing facts dominate the picture. First, globular proteins are compact: a folded chain occupies a volume only modestly larger than the sum of its atomic volumes, with a packing density (~0.74) rivaling close-packed solids, far higher than the loose, solvent-swollen unfolded state. Second, this compaction is not random — it produces a specific, reproducible arrangement in which nearly all nonpolar side chains point inward and nearly all charged/polar side chains point outward toward solvent, with fewer exceptions than a naive hydrophobic-collapse picture would predict. Myoglobin is the classic illustration. Its chain folds into eight α-helices (named A through H) with almost no β-structure, arranged into a roughly globular shape enclosing a nonpolar crevice that clamps a heme (iron-protoporphyrin IX) prosthetic group. The proximal histidine (His93, helix F) coordinates the heme iron directly; the distal histidine (His64, helix E) sits across the pocket, hydrogen-bonding to and sterically gating the bound O₂ without touching the metal. Neither histidine's role is visible from the sequence — His93 and His64 are 29 residues apart — but both are essential once you look at the folded, three-dimensional pocket. This is the general lesson: tertiary structure creates active sites and binding pockets by bringing sequence-distant residues into spatial contact. A second organizing idea is topology: the order and handedness with which secondary-structure elements connect. Two proteins can have identical secondary-structure content (say, eight parallel β-strands and eight α-helices) but differ completely in how the strands are connected — this is the difference between, say, a TIM barrel and a Rossmann-fold sandwich. Topology is what higher-order fold classification (SCOP, CATH) actually tracks, because sequence identity can be near zero between two proteins sharing the same fold.

Figure

Myoglobin's tertiary fold: eight helices packing around a heme pocket

The energetic accounting behind this picture separates into four noncovalent contributions layered on top of the hydrophobic effect: van der Waals packing in the tightly filled core, an extensive network of backbone and side-chain hydrogen bonds (many already used to build helices and sheets, but many more formed only upon tertiary packing), electrostatic ion pairs (salt bridges) typically on the surface, and, in extracellular or secreted proteins, disulfide bonds that covalently staple the fold. None of these forces alone is large enough to fold a protein; their simultaneous, cooperative satisfaction is what produces a deep, narrow free-energy minimum instead of a shallow one.

Figure

Energetics of folding: enthalpy–entropy compensation

The core ideas

Folding stability is governed by a standard free-energy difference between the folded (native, N) and unfolded (denatured, D) states:

\Delta G^{\circ}_{\text{fold}} = G^{\circ}_{N} - G^{\circ}_{D} = \Delta H^{\circ}_{\text{fold}} - T\Delta S^{\circ}_{\text{fold}}

where:

$\Delta H^{\circ}_{\text{fold}}$ — enthalpy change on folding (favorable/negative: new H-bonds, van der Waals contacts, ion pairs)
$\Delta S^{\circ}_{\text{fold}}$ — entropy change on folding (net unfavorable/negative: loss of backbone/side-chain conformational freedom dominates over the positive entropy gained by releasing ordered water from nonpolar surfaces)
$T$ — absolute temperature (K)
$\Delta G^{\circ}_{\text{fold}}$ — net stability; typically −20 to −65 kJ/mol at 25 °C for a well-behaved single-domain globular protein

Reading: the fold is stable only because a large negative enthalpy term (favorable) narrowly outweighs a large positive $-T\Delta S^{\circ}_{\text{fold}}$ penalty. This is why proteins are only marginally stable — comparable to a handful of hydrogen bonds' worth of net energy — and why point mutations, pH shifts, or a few degrees of heating can unfold them. The equilibrium constant for folding follows directly:

K_{\text{fold}} = \frac{[N]}{[D]} = e^{-\Delta G^{\circ}_{\text{fold}}/RT}

where:

$R = 8.314\ \text{J mol}^{-1}\text{K}^{-1}$
$T$ in kelvin
$\Delta G^{\circ}_{\text{fold}}$ as above (negative for a stable fold, so $K_{\text{fold}} > 1$ )

Reading: because $\Delta G^{\circ}_{\text{fold}}$ is only a few tens of kJ/mol, $K_{\text{fold}}$ is large but not astronomically so — most native proteins exist in a small dynamic equilibrium with a tiny unfolded population, which is one reason proteolysis, chaperone action, and thermal denaturation are all physiologically tunable processes rather than all-or-nothing events. The kinetic side of the problem is addressed by Levinthal's estimate of the size of conformational space. For a chain of $N$ residues, if each residue's backbone dihedral pair $(\phi,\psi)$ can adopt roughly $c$ distinguishable conformations, the total number of chain conformations is:

\Omega \approx c^{N}

where:

$c$ — number of accessible backbone conformations per residue (commonly estimated as $c \approx 3$ –10)
$N$ — number of residues
$\Omega$ — total conformational space size

Reading: even with a conservative $c = 3$ and $N = 100$ , $\Omega \approx 3^{100} \approx 5\times10^{47}$ . Sampling these at a generous $10^{13}\,\text{s}^{-1}$ (a bond-vibration timescale) would take far longer than the age of the universe — yet real proteins of this size fold in milliseconds to seconds. This mismatch is Levinthal's paradox, and its resolution (below) is the funneled energy landscape, not a faster search.

Derivation AWhy the fold is only marginally stable: the enthalpy–entropy ledger

Goal: show quantitatively why $\Delta G^{\circ}_{\text{fold}}$ comes out as a small residual of two much larger, opposing terms, using order-of-magnitude values for a 100-residue protein.

Step 1 — Enthalpic gain from packing and hydrogen bonding. Folding buries roughly 100–150 backbone and side-chain hydrogen bonds and forms extensive van der Waals contacts in the core. Each hydrogen bond contributes roughly 2–8 kJ/mol of favorable enthalpy (weakened by the fact that unfolded-state water already hydrogen-bonds to many of the same donors/acceptors, so the net per-bond enthalpic gain on folding is smaller than the bond's intrinsic strength). Taking a representative net value of ~6 kJ/mol per bond over ~130 bonds:

\Delta H_{\text{H-bond}} \approx -6 \times 130 \approx -780\ \text{kJ/mol}

Van der Waals packing in a well-filled core adds a further favorable increment, often estimated at −1 to −2 kJ/mol per buried CH₂-equivalent group summed over ~100 groups, contributing roughly another −100 to −200 kJ/mol. Take $\Delta H_{\text{vdW}} \approx -150$ kJ/mol.

Step 2 — Hydrophobic transfer free energy. Burying nonpolar side-chain surface area (ASA) out of water releases ordered ("clathrate-like") water and is favorable. Using an empirical transfer coefficient of about $-0.1$ kJ/mol per Å² of buried nonpolar ASA, and a typical buried nonpolar area for a 100-residue globular domain of order 6000–8000 Å²:

\Delta G_{\text{hyd}} \approx -0.1\ \text{kJ/mol}\,\cdot\,\text{Å}^{-2} \times 7000\ \text{Å}^2 \approx -700\ \text{kJ/mol}

This term itself splits into an enthalpy part and an entropy part (the classic signature: hydrophobic transfer has a large positive $\Delta S$ near room temperature from water release, but this article keeps the hydrophobic term as one lumped $\Delta G$ contribution, consistent with the empirical ASA-based formula used throughout The Hydrophobic Effect and the Protein Core).

Step 3 — Conformational entropy loss. Each residue in the unfolded state samples multiple backbone (and side-chain) conformations; folding restricts this to essentially one. A standard estimate per residue is:

\Delta S_{\text{conf}} \approx -R\ln(c)\ \text{per residue}, \qquad c \approx 3\text{–}10

Taking $c \approx 8$ (a moderate estimate including side-chain rotamers) and $N = 100$ :

\Delta S_{\text{conf}} \approx -100 \times R\ln(8) \approx -100 \times 8.314 \times 2.08 \approx -1730\ \text{J/(mol}\,\cdot\,\text{K)}

At $T = 298$ K:

-T\Delta S_{\text{conf}} \approx -298 \times (-1730\ \text{J/(mol}\,\cdot\,\text{K)}) \approx +5.16\times10^{5}\ \text{J/mol} \approx +516\ \text{kJ/mol}

This is the dominant unfavorable term (in this article's convention, "unfavorable" here means positive, opposing folding).

Step 4 — Net balance. Summing the favorable terms (Steps 1–2) and the unfavorable term (Step 3):

\Delta G^{\circ}_{\text{fold}} \approx (-780 - 150 - 700) + 516\ \text{kJ/mol}

\Delta G^{\circ}_{\text{fold}} \approx -1630 + 516 = -1114\ \text{kJ/mol}

This crude order-of-magnitude sum overshoots real measured stabilities by roughly an order of magnitude, and that overshoot is itself instructive: real hydrogen bonds and hydrophobic terms are heavily compensated by residual structure and residual solvation in the unfolded state (the unfolded chain is not a fully extended, fully solvated random coil; it retains significant local structure and hydrogen bonding to water that partially cancels the naive per-bond/per-Å² increments). When these compensations are folded in with realistic experimental parameters (calibrated against measured $\Delta H_{\text{cal}}$ and $\Delta S_{\text{cal}}$ from differential scanning calorimetry), the two multi-hundred-kJ/mol terms are found to cancel to within a residual of order:

\Delta G^{\circ}_{\text{fold}} \approx -20\ \text{to}\ -65\ \text{kJ/mol}

which is the experimentally observed range for single-domain globular proteins. The lesson of the derivation is not the exact numeric cancellation (which requires calorimetric calibration) but the structural one: folding stability is a small difference of much larger opposing enthalpic and entropic terms, and this is precisely why point mutations of only a few kJ/mol (a single lost hydrogen bond, a single buried charge) can destabilize a protein by 30–60% of its total stability, and why moderate heating or urea/GdnHCl concentrations can unfold proteins entirely.

Limiting check: as $N \to 1$ (a single residue, no real "fold"), $\Delta S_{\text{conf}} \to -R\ln c$ , a tiny term, and there is essentially no hydrophobic core to bury — consistent with the fact that single amino acids and very short peptides do not fold into stable tertiary structures at all; the physics only produces a net-favorable, cooperative fold once $N$ is large enough to build a buried core with a large ASA-to-entropy-cost ratio.

Go deeper · Derivation B · 3 minResolving Levinthal's paradox: the funneled free-energy landscape

Derivation BResolving Levinthal's paradox: the funneled free-energy landscape

Goal: show why real proteins fold on millisecond–second timescales despite $\Omega \approx c^{N}$ possible conformations, by replacing "random search" with a biased, funnel-shaped free-energy surface.

Step 1 — The false premise of exhaustive search. Levinthal's calculation assumes every conformation is equally likely to be sampled and that the protein must find one specific "correct" conformation among all $\Omega$ of them by unbiased diffusion in conformation space. If folding truly worked this way, the search time would scale as:

t_{\text{search}} \sim \frac{\Omega}{\nu_0}

where $\nu_0 \sim 10^{13}\,\text{s}^{-1}$ is a bond-rotation attempt frequency. For $N=100$ , $c=3$ , this gives $t_{\text{search}} \sim 5\times10^{47}/10^{13} \approx 5\times10^{34}\,\text{s}$ — absurdly longer than the age of the universe ( $\sim4\times10^{17}$ s). Real folding times for 100-residue proteins are $10^{-3}$ – $10^2$ s. The premise, not the arithmetic, must be wrong.

Step 2 — Energy landscape theory: the funnel. Folding is not a flat, high-dimensional maze; it is a landscape with a systematic downhill gradient toward the native state, described by the "New View" (Bryngelson, Wolynes, Onuchic, and coworkers, early 1990s). Native-like local contacts are, on average, more stabilizing than non-native ones (this is the principle of minimal frustration — evolution has selected sequences whose energetically favorable contacts largely coincide with their native-state contacts). Consequently the free-energy surface, plotted against a reaction coordinate $Q$ (fraction of native contacts formed, $0 \le Q \le 1$ ), looks like a funnel: broad and rugged at $Q \to 0$ (many unfolded microstates, similar energies), narrowing and deepening toward $Q \to 1$ (few native-like microstates, much lower energy).

Step 3 — Why a funnel resolves the paradox. On a funneled landscape, the protein does not need to sample all $\Omega$ states with equal probability; each successful local folding event (forming a native-like helix turn, a native-like hydrophobic contact) is thermodynamically favored over competing non-native alternatives, so the population flows preferentially toward $Q=1$ . The effective number of conformations that matter kinetically is not $\Omega$ but a much smaller number restricted to the funnel's "productive" pathway region. Formally, the mean first-passage time to the native state can be written in Kramers/diffusion form:

\tau_{\text{fold}} \sim \tau_0\, e^{\Delta G^{\ddagger}/RT}

where:

$\tau_0$ — a molecular prefactor (roughly $10^{-6}$ – $10^{-7}$ s, reflecting the reconfiguration time of a partially collapsed chain, far slower than a single bond vibration but far faster than $\Omega/\nu_0$ )
$\Delta G^{\ddagger}$ — the free-energy barrier at the transition state ensemble (funnel "neck"), typically 40–85 kJ/mol for two-state folders

Step 4 — Numerical check against real folding rates. Take a small, fast-folding two-state protein with $\Delta G^{\ddagger} \approx 54$ kJ/mol at $T = 298$ K:

e^{-\Delta G^{\ddagger}/RT} = e^{-54000/(8.314\times298)} = e^{-21.8} \approx 3.4\times10^{-10}

\tau_{\text{fold}} \sim \tau_0 \times e^{\Delta G^{\ddagger}/RT} \approx 10^{-6}\,\text{s} \times \frac{1}{3.4\times10^{-10}} \approx 3\times10^{3}\ \text{s}^{-1\ (\text{rate})}

i.e. a folding rate constant of order $10^{2}$ – $10^{4}\,\text{s}^{-1}$ , corresponding to folding times of roughly 0.1–10 ms — squarely in the experimentally observed range for small, fast-folding domains (e.g., villin headpiece folds in ~4 μs to ~10s of μs depending on construct; many small two-state proteins fold in 1–100 ms). Crucially, this timescale emerges from a biased barrier-crossing process over a modest free-energy barrier, not from an unbiased search over $\Omega \approx 10^{47}$ states — the funnel converts an astronomically improbable search into an ordinary activated process.

Limiting check: in the limit of zero landscape bias (a perfectly flat, "golf-course" landscape with no energetic preference for native-like contacts, i.e., a "maximally frustrated" sequence), the funnel disappears, $\Delta G^{\ddagger}$ effectively becomes comparable to the full conformational entropy $T\Delta S_{\text{conf}}$ , and the folding time reverts toward the Levinthal estimate — consistent with the observation that poorly designed or highly frustrated sequences (and many mutant or aggregation-prone proteins) fold slowly, populate misfolded traps, or fail to fold at all, exactly as the funnel picture predicts.

Worked examples

Example 1 (foundational) — Estimating buried hydrophobic surface area and its energetic payoff. Suppose a small 80-residue globular domain buries 5400 Å² of nonpolar accessible surface area upon folding (an asymmetric, protein-specific number, not a round textbook figure), using the standard empirical transfer coefficient $\Delta g_{\text{tr}} \approx -0.10\ \text{kJ mol}^{-1}\,\text{Å}^{-2}$ for nonpolar ASA burial:

\Delta G_{\text{hyd}} = \Delta g_{\text{tr}} \times \text{ASA}_{\text{buried}} = (-0.10\ \text{kJ mol}^{-1}\text{Å}^{-2})(5400\ \text{Å}^2) = -540\ \text{kJ/mol}

This is the hydrophobic driving force alone, before subtracting the conformational entropy cost. If this domain's total conformational entropy penalty is estimated (via $\Delta S_{\text{conf}} \approx -R\ln(c)\,N$ with $c=6$ , $N=80$ ) as:

\Delta S_{\text{conf}} = -80 \times 8.314 \times \ln(6) = -80\times8.314\times1.792 \approx -1191\ \text{J/(mol}\,\cdot\,\text{K)}

-T\Delta S_{\text{conf}} \ \text{at } 310\ \text{K (body temperature)} = -310\times(-1191) = +369{,}210\ \text{J/mol} \approx +369\ \text{kJ/mol}

Even before adding hydrogen bonding and van der Waals contributions, we already see the near-cancellation pattern: $-540$ kJ/mol (hydrophobic) vs. $+369$ kJ/mol (conformational entropy) leaves a raw residual of about $-171$ kJ/mol — of the same order of magnitude as, but still larger in size than, real net stabilities, again reflecting the need for calorimetric calibration of the full enthalpy/entropy budget (Step 4 of the first derivation). The qualitative point survives untouched: the two terms are comparable in size, and their difference — not either term alone — sets the stability. Limiting check: if ASA buried were 0 Å² (a chain that never buries any nonpolar surface, i.e., stays fully extended), $\Delta G_{\text{hyd}} \to 0$ while $-T\Delta S_{\text{conf}}$ remains strongly positive (unfavorable) — correctly predicting that such a chain cannot fold into a stable compact structure, matching the physical fact that a hydrophobic core is a prerequisite for tertiary stability, not an optional decoration.

Example 2 (applied) — Myoglobin's distal histidine and O₂ affinity. Myoglobin's oxygen affinity is governed by the distal histidine (His64, helix E), which hydrogen-bonds to bound O₂ (donating an N–H···O hydrogen bond of roughly 8–15 kJ/mol) but plays no role in this stabilization when the pocket is empty or when CO (rather than O₂) is bound, since CO is a weaker hydrogen-bond acceptor and binds end-on/tilted in the pocket, losing much of this stabilization. This differential stabilization is why myoglobin's CO/O₂ discrimination ratio in the protein is only ~25, whereas free heme in solution (no distal histidine, no steric gating) prefers CO over O₂ by a ratio of roughly 20,000:1 — a physiologically essential discrimination, since endogenous CO (from heme catabolism) would otherwise poison oxygen transport. Quantitatively, if the distal histidine H-bond contributes an extra $\Delta\Delta G_{\text{H-bond}} \approx 10$ kJ/mol of stabilization specifically for bound O₂ relative to bound CO, the relative equilibrium constant for O₂ vs. CO binding shifts by:

\frac{K_{\text{O}_2}/K_{\text{O}_2,\text{no-His}}}{K_{\text{CO}}/K_{\text{CO,no-His}}} = e^{\Delta\Delta G/RT} = e^{10000/(8.314\times298)} = e^{4.04} \approx 57

This factor of ~57 (order-of-magnitude consistent with the measured ~700-fold reduction in the discrimination ratio between free heme, ~20,000:1, and myoglobin, ~25:1, once combined with additional steric contributions from the distal pocket geometry not captured by hydrogen bonding alone) illustrates the central lesson of tertiary structure directly: a single residue, positioned correctly in three-dimensional space by the fold, converts an indiscriminate cofactor (free heme) into a selective, tunable biological sensor. Limiting check: remove the distal histidine entirely (as in the H64A/H64G mutants studied experimentally) and both the O₂ affinity and the O₂/CO discrimination collapse toward the free-heme-like regime, exactly as $\Delta\Delta G_{\text{H-bond}} \to 0$ predicts in the equation above.

Example 3 (exam/competition-level) — Estimating a folding rate constant from a measured $\Delta G^{\ddagger}$ and comparing two mutants. A two-state, 64-residue protein domain has a measured folding free-energy barrier $\Delta G^{\ddagger}_{\text{WT}} = 47.0$ kJ/mol at $T=298$ K, with prefactor $\tau_0 = 2\times10^{-7}$ s (a plausible chain-reconfiguration time, chosen off any "nice" value). Find the folding rate constant, then find the new folding rate for a destabilizing mutation that raises the barrier by 6.3 kJ/mol (e.g., a core-packing defect from a Leu→Ala substitution that creates a cavity). Using the Kramers-type relation $k_f = \tau_0^{-1} e^{-\Delta G^{\ddagger}/RT}$ :

RT = 8.314 \times 298 = 2477.6\ \text{J/mol}

e^{-\Delta G^{\ddagger}_{\text{WT}}/RT} = e^{-47000/2477.6} = e^{-18.97} \approx 5.79\times10^{-9}

k_{f,\text{WT}} = \frac{1}{2\times10^{-7}\,\text{s}} \times 5.79\times10^{-9} = 5\times10^{6}\,\text{s}^{-1}\times 5.79\times10^{-9} \approx 2.9\times10^{-2}\ \text{s}^{-1}

This corresponds to a folding time $\tau_{f,\text{WT}} = 1/k_{f,\text{WT}} \approx 34$ s — a plausible timescale for a moderate-sized, non-optimized folder. Now for the Leu→Ala mutant, $\Delta G^{\ddagger}_{\text{mut}} = 47.0 + 6.3 = 53.3$ kJ/mol:

e^{-\Delta G^{\ddagger}_{\text{mut}}/RT} = e^{-53300/2477.6} = e^{-21.51} \approx 4.55\times10^{-10}

k_{f,\text{mut}} = 5\times10^{6}\,\text{s}^{-1} \times 4.55\times10^{-10} \approx 2.3\times10^{-3}\ \text{s}^{-1}

\frac{k_{f,\text{WT}}}{k_{f,\text{mut}}} = \frac{2.9\times10^{-2}}{2.3\times10^{-3}} \approx 12.6

A modest 6.3 kJ/mol destabilization of the transition state — roughly the free energy of a single lost van der Waals packing interaction in a cavity-creating mutation — slows folding by a factor of ~13. This is the quantitative signature of $\phi$ -value analysis in protein folding kinetics: mutations at residues that are already well-ordered in the transition state (high $\phi$ ) produce exactly this kind of large, exponential sensitivity of $k_f$ to $\Delta\Delta G^{\ddagger}$ , while mutations at residues still disordered in the transition state (low $\phi$ ) barely change $k_f$ at all, because they mainly affect the native-state stability rather than the barrier itself. Limiting check: setting $\Delta\Delta G^{\ddagger} \to 0$ recovers $k_{f,\text{mut}} \to k_{f,\text{WT}}$ identically, as required — a mutation with zero effect on the transition-state free energy must leave the folding rate unchanged, regardless of its effect on native-state stability.

Limiting cases, scope & common traps

Regime / case	What happens	Why
Very short peptides ( $N \lesssim 20$ –30)	No stable, unique tertiary fold; only local secondary structure or none at all	Insufficient buried hydrophobic core; $-T\Delta S_{\text{conf}}$ penalty is not offset by enough favorable burial
Small single-domain proteins (this article's default)	Two-state folding, funneled landscape, $\Delta G^{\circ}_{\text{fold}} \approx -20$ to $-65$ kJ/mol	Core large enough for net-favorable hydrophobic collapse; small enough to avoid stable folding intermediates
Large multidomain proteins	Fold domain-by-domain, often co-translationally, with populated intermediates	Full-chain Levinthal search is even less tractable; domains act as independent, quasi-autonomous folding units — foreshadowing Protein Domains and Structural Classification
Denaturing conditions (high T, urea, GdnHCl)	$\Delta G^{\circ}_{\text{fold}} \to 0$ then positive; native state depopulated	Denaturants weaken hydrophobic effect / compete for H-bonds; heat increases $-T\Delta S_{\text{conf}}$ penalty and disrupts H-bonds
Membrane proteins	Hydrophobic residues face outward (lipid-facing helices), polar residues line internal pores	Hydrophobic effect is inverted relative to the aqueous case — the "inside-out" rule
Frustrated / poorly designed sequences	Slow folding, kinetic traps, aggregation	Landscape funnel is shallow or has competing local minima (non-native contacts nearly as stable as native ones)
Multi-subunit assembly	Individually folded tertiary structures further associate	Sets up Quaternary Structure and Subunit Assembly; tertiary fold is a prerequisite, not a replacement, for oligomerization

Common trapMisconception: "hydrophobic collapse" and "the native fold" are the same thing

Burying nonpolar surface area is necessary but not sufficient to explain tertiary structure. A polypeptide can collapse into many different compact, hydrophobically buried "molten globule" states with roughly similar overall burial but very different, non-native internal packing and topology. What selects the unique native fold from among these compact alternatives is the additional, more geometrically demanding requirement of close-packed, complementary side-chain packing (van der Waals contacts with essentially no internal voids) together with a specific, mutually consistent hydrogen-bonding network. This is why molten-globule intermediates are real, experimentally observed states (compact, some secondary structure, but fluctuating tertiary packing) distinct from the fully native state — and why the last, slowest step of folding is often the "fine-tuning" of exact side-chain packing rather than the initial hydrophobic collapse.

Common trapMisconception: tertiary structure is rigid and static

The word "structure" suggests a fixed, crystal-like object, but the folded state is a thermally fluctuating ensemble, not a single conformation. Side chains rotate among rotamers on picosecond timescales; loops breathe on nanosecond–microsecond timescales; even whole domains can hinge relative to one another. Myoglobin's own oxygen-binding pocket, as resolved crystallographically, has no direct static channel from bulk solvent to the buried heme — O₂ enters via transient, thermally driven cavity fluctuations of the packed helices. Treating the tertiary fold as a single rigid PDB coordinate file, rather than as the dominant member of a dynamic conformational ensemble, is a common and consequential oversimplification, especially when reasoning about ligand binding, allostery, or enzyme catalysis.

The chemist behind it

Origins · 1 minChristian B. Anfinsen and how this was discovered

Christian B. Anfinsen

1916–1995 · American biochemist, NIH; Nobel Prize in Chemistry 1972

Anfinsen's defining experiment used bovine pancreatic ribonuclease A (RNase A), a 124-residue enzyme with four disulfide bonds. He denatured it completely with 8 M urea plus a reducing agent (β-mercaptoethanol), scrambling both its noncovalent structure and its disulfide connectivity, and destroyed essentially all enzymatic activity. When he then removed the urea and reducing agent by dialysis, allowing the protein to reoxidize its cysteines in air, the enzyme spontaneously regained the great majority of its original catalytic activity — meaning it had refolded into (very nearly) its correct native tertiary structure, re-forming the correct set of disulfide bonds out of a combinatorially much larger number of possible incorrect pairings, with no other information present beyond the amino acid sequence itself and the surrounding solvent conditions. This was the direct experimental basis of the thermodynamic hypothesis: for many single-domain, soluble proteins, the amino acid sequence completely specifies the native three-dimensional structure, because that structure represents (at least a very deep, kinetically accessible) minimum of the free energy under physiological solution conditions. Anfinsen shared the 1972 Nobel Prize in Chemistry (with Stanford Moore and William H. Stein) explicitly "for his work on ribonuclease, especially concerning the connection between the amino acid sequence and the biologically active conformation." The finding reframed protein folding from a mysterious biological assembly process into a tractable physical chemistry problem — one whose kinetic half (how the search is actually accomplished so quickly) took another two decades of biophysics, culminating in energy landscape theory, to resolve satisfactorily.

History

- 1958 — Kendrew solves the first protein tertiary structure, myoglobin, by X-ray crystallography.

- 1960 — Perutz solves hemoglobin's structure, revealing quaternary assembly of four myoglobin-like subunits.

- 1961 — Anfinsen's RNase A refolding experiments establish the thermodynamic hypothesis of protein folding.

- 1968 — Levinthal formulates the conformational-search paradox, arguing folding cannot proceed by random search.

- 1972 — Anfinsen awarded the Nobel Prize in Chemistry for the sequence–structure–function connection.

- 1990s — Bryngelson, Wolynes, Onuchic, and Dill develop energy landscape theory and the folding funnel, resolving Levinthal's paradox conceptually.

What is the difference between secondary and tertiary structure?

Secondary structure describes local, repeating backbone conformations — α-helices and β-strands — stabilized mainly by backbone hydrogen bonds between residues close together in sequence. Tertiary structure describes the complete three-dimensional arrangement of an entire single polypeptide chain, including how all its secondary-structure elements and connecting loops pack together in space, stabilized mainly by side-chain interactions (hydrophobic burial, van der Waals packing, ion pairs, disulfides) between residues that can be very far apart in sequence. A useful test: secondary structure can be described using only $(\phi,\psi)$ angles of nearby residues, while tertiary structure requires knowing the full 3D coordinates of the whole chain, including long-range contacts.

Does every protein have a single, unique tertiary structure?

Most small, single-domain, soluble globular proteins under physiological conditions do adopt one dominant native tertiary fold, consistent with Anfinsen's thermodynamic hypothesis. But this is not universal: many proteins contain intrinsically disordered regions with no single stable tertiary structure at all (functioning instead as flexible linkers or coupled-folding-and-binding modules); some proteins are "metamorphic," populating two distinct folds in dynamic equilibrium; and prions represent an extreme case where an alternative, self-templating tertiary/quaternary conformation is thermodynamically or kinetically stable and infectious. The single-fold picture is the dominant and pedagogically central case, not an absolute law.

How do disulfide bonds relate to tertiary structure?

Disulfide bonds (Cys–S–S–Cys) are covalent cross-links formed by oxidation of two cysteine thiols that are brought into proximity by the folded tertiary structure — they do not usually drive folding themselves but instead lock in and kinetically stabilize a tertiary fold that noncovalent forces have already largely specified. They are common in secreted and extracellular proteins (antibodies, insulin, ribonuclease) where the oxidizing extracellular/ER environment permits their formation, and rare in cytosolic proteins, where the reducing intracellular environment (maintained by glutathione and thioredoxin systems) keeps cysteines reduced. Anfinsen's RNase A experiment specifically showed that even with all four disulfides reduced and scrambled, the correct native disulfide pattern re-forms upon correct tertiary refolding — evidence that tertiary structure information resides in the sequence, not primarily in the disulfides.

Why do hydrophobic core mutations often destabilize proteins so much more than surface mutations?

Core mutations frequently disrupt the exceptionally tight, complementary van der Waals packing of the hydrophobic interior — for example, mutating a large residue (Leu, Phe) to a smaller one (Ala, Gly) leaves an internal cavity that is energetically costly (lost van der Waals contacts, sometimes an unfilled void with no compensating structural rearrangement), while mutating a small residue to a much larger one can cause steric clashes that force local unfolding or repacking. Surface mutations, by contrast, exchange one solvent-exposed side chain for another without disturbing the tightly packed core, so their energetic effect on stability is usually much smaller (often just a few kJ/mol from altered surface electrostatics or entropy) unless they remove a specific salt bridge or surface hydrogen bond important for function.

How does the "inside-out" rule work for membrane protein tertiary structure?

Soluble globular proteins bury nonpolar residues inward, away from water, and display polar/charged residues on the solvent-facing surface. Integral membrane proteins embedded in the lipid bilayer largely invert this pattern for the membrane-spanning region: the lipid-facing surface of a transmembrane helix is dominated by nonpolar residues (matching the nonpolar acyl-chain environment of the bilayer core), while polar or charged residues are found lining any internal aqueous pore or channel, or clustered at the membrane–water interface (often tryptophan and tyrosine, which favor this interfacial zone). The underlying physical principle — bury nonpolar surface against the locally nonpolar environment, expose polar/charged groups to the locally polar environment — is the same hydrophobic-matching logic as in soluble proteins, just applied to a different solvent (lipid instead of water).

What experimental methods reveal tertiary structure?

X-ray crystallography (used for myoglobin by Kendrew in 1958 and for the great majority of solved protein structures historically) requires a well-ordered crystal and gives atomic-resolution coordinates. Nuclear magnetic resonance (NMR) spectroscopy works in solution (better reflecting dynamic, physiological conditions) but is generally limited to smaller proteins (historically under ~30–40 kDa, though methods have extended this). Cryo-electron microscopy (cryo-EM), especially after the "resolution revolution" of the 2010s (direct electron detectors, better image processing), now routinely achieves near-atomic resolution for large complexes without crystallization, and has become the dominant method for large or heterogeneous assemblies. Computational prediction (AlphaFold2 and successors) now supplements all three experimentally validated methods with rapid, often highly accurate, structural hypotheses directly from sequence.

Is tertiary structure classification (SCOP, CATH) based on sequence similarity?

No — this is a common point of confusion. Fold classification systems like SCOP (Structural Classification of Proteins) and CATH (Class, Architecture, Topology, Homology) classify proteins primarily by their three-dimensional topology (the spatial arrangement and connectivity of secondary-structure elements), not by sequence identity. Two proteins can share a fold — the same arrangement of helices/strands in the same topological order — with less than 15–20% pairwise sequence identity, well below the threshold usually used to infer common ancestry from sequence alone. This is strong structural evidence that tertiary fold is far more evolutionarily conserved than sequence, and it is the basis for template-based structural classification of proteins whose sequences alone would give no clue to their relationship.

Tertiary Structure and the Fold

Where this lives today

The physical picture

The core ideas

Worked examples

Limiting cases, scope & common traps

The chemist behind it

Before this

Where this leads