Protein folding: Anfinsen, Levinthal, and molecular chaperones

Intuition Beginner

A protein is built as a linear chain of amino-acid units, yet it works only once it has collapsed into a precise three-dimensional shape. That shape — the native state — is what gives an enzyme its active site, an antibody its binding pocket, and a muscle fibre its spring. An unfolded chain is a floppy rope with almost no function. Understanding how the chain finds its shape is the protein-folding problem.

In the 1960s, Christian Anfinsen unfolded a small protein, ribonuclease A, using harsh chemicals, and snipped the chemical staples (disulfide bonds) holding its shape. When he washed the chemicals away, the chain refolded on its own into the exact original structure and recovered full activity, with no other machinery helping. This experiment suggested the amino-acid sequence alone carries enough information to specify the fold — the native shape is the most stable arrangement the chain can reach.

Yet Cyrus Levinthal spotted a paradox. A chain of 100 amino acids can adopt a staggering number of shapes. Trying them at random would take far longer than the age of the universe, but real proteins fold in milliseconds. The resolution is that a protein does not search blindly. It folds down an energy landscape shaped like a funnel, sliding through a few stable intermediates toward the native state, and for the hardest cases it gets help from molecular machines called chaperones.

Visual Beginner

A folding funnel is the mental picture for how a protein reaches its native shape. The top is wide: a huge number of unfolded, high-energy conformations. The funnel narrows as the chain forms local contacts and loses freedom, passing through partially folded intermediates (the molten globule). The bottom is a single, low-energy native state.

Free energy
    |
    |   \ . . . . /        unfolded ensemble (many states, high energy)
    |    \  . .  /
    |     \  *  /          molten globule (compact, fluctuating core)
    |      \ | /
    |       \|/
    |        V             native state (one state, low energy)
    +---------------------  reaction coordinate / conformation

Worked example Beginner

The Levinthal counting argument. Take a small protein of 100 amino-acid residues. Each residue has two backbone bonds that can rotate (the angles $\varphi$ and $\psi$ introduced in 15.12.02 pending), and each can sit in roughly three stable positions. The number of possible backbone shapes is therefore about 3 multiplied by itself 100 times — roughly 5 followed by 47 zeros.

If the chain tried one shape every nanosecond (a billionth of a second), visiting every shape would take about 10 to the power 38 seconds. That is roughly a thousand billion billion times the age of the universe. A real 100-residue protein folds in about a millisecond to a second. Blind trial-and-error cannot be the mechanism.

What this tells us. The native shape cannot be found by searching the whole space. Folding must follow guided routes that skip almost all conformations and slide directly down toward the native basin. The thermodynamic hypothesis tells us the destination; the funnel and the chaperones tell us how the chain gets there quickly.

Check your understanding Beginner

Formal definition Intermediate+

Let the conformation space $\Omega$ be the set of all backbone and side-chain torsion-angle combinations accessible to a polypeptide chain (each residue contributing the two backbone angles $\varphi$, $\psi$ plus side-chain $\chi$ angles defined in 15.12.02 pending). A conformation is a point $x\in \Omega$. The free-energy landscape is the function $G:\Omega \to \mathbb{R}$ assigning a Gibbs free energy to each conformation, averaging over solvent and microscopic degrees of freedom ^{[Bryngelson et al. 1995]}.

The native state $N$ is the biologically active, structured conformation at physiological conditions. The denatured (unfolded) state $D$ denotes the large ensemble of high-free-energy conformations. The folding free energy is

$$\Delta G_{\text{fold}}=G_N-G_D<0,$$

typically $-5$ to $-15\text{kcal/mol}$ ($-20$ to $-60\text{kJ/mol}$). Note the magnitudes: native proteins are stabilised by hundreds of $\text{kcal/mol}$ of favourable interactions yet are only a few $\text{kcal/mol}$ more stable than the unfolded chain, because the loss of conformational entropy and the cost of unsatisfied hydrogen bonds nearly cancel the stabilising terms.

Anfinsen's thermodynamic hypothesis ^{[Anfinsen 1973]}. Under native conditions, for a single-domain globular protein, the native state is the global minimum of $G$ over $\Omega$, and the location of this minimum is determined by the amino-acid sequence alone.

A folding funnel is the gross geometry of $G$ near the native basin: a broad, high-enthalpy ensemble of unfolded conformations at the top that narrows toward a single low-energy native basin, so that downhill flow on $G$ is biased toward $N$. A molten globule is a compact, partially folded intermediate with native-like secondary structure but a fluctuating, loosely packed hydrophobic core ^{[Fersht 1999]}. A molecular chaperone is an ATP-using protein machine that binds exposed hydrophobic surfaces on non-native chains, preventing aggregation and providing an environment conducive to folding, without supplying the folding information itself ^{[Hartl & Hayer-Hartl 2009]}.

Counterexamples to common slips

• "The native state is always the global free-energy minimum." Prions and amyloid fibrils are alternative conformations that can be thermodynamically more stable than the functional native state yet biologically catastrophic. The Anfinsen hypothesis holds for many single-domain globular proteins under native conditions, not as a universal law over all of $\Omega$.

• "Chaperones instruct the chain how to fold." They do not encode the fold. The folding information resides in the sequence; chaperones reshape the effective landscape by suppressing off-pathway aggregation traps and by isolating chains in a protected cavity.

• "Thermodynamic stability equals folding rate." The destination (global minimum) and the speed of reaching it are separate questions. This distinction is the content of the Levinthal paradox and motivates the kinetic theory below.

Key result Intermediate+

Anfinsen's thermodynamic hypothesis (operational form). If a single-domain globular protein is reversibly denatured and then returned to native conditions, the chain refolds spontaneously to the native state, recovering biological activity, because $N$ is the global free-energy minimum.

The ribonuclease A experiment. Bovine pancreatic ribonuclease A has 124 residues and four disulfide bonds among eight cysteines. Anfinsen denatured the protein in concentrated urea with a reducing agent ($\beta$-mercaptoethanol), which broke the disulfide bonds and destroyed the tertiary structure, abolishing activity. On removal of the denaturant and slow re-oxidation in air, the chain refolded and regenerated the correct four native disulfide pairings, recovering nearly full enzymatic activity ^{[Anfinsen 1973]}.

The disulfide-pairing count sharpens why this is decisive. Eight cysteines can be paired into four disulfide bonds in

$$\frac{8!}{2^{4}4!}=7\times 5\times 3\times 1=105$$

distinct ways. Only one of the 105 patterns is native. Random re-oxidation would therefore recover the native disulfide set with probability $1/105$. The experiment nonetheless recovers near-full activity, so the protein does not pair cysteines at random: it folds first to the free-energy minimum, which brings the correct cysteines into proximity, and the disulfides form on the already-folded scaffold.

The Levinthal paradox. Even setting chemistry aside, $\Omega$ is vast. A 100-residue chain with three backbone conformations per residue has roughly $3^{100}\approx 5\times 10^{47}$ states. Sampling one state per bond vibration ($\sim 10^{-13}\text{s}$) gives an exhaustive search time of $\sim 10^{35}\text{s}$, vastly exceeding the age of the universe ($\sim 4\times 10^{17}\text{s}$) and the observed folding time ($10^{-3}$ to $1\text{s}$). An unbiased exhaustive search cannot locate $N$ on any biological timescale ^{[Levinthal 1969]}.

Resolution. The paradox is resolved by separating thermodynamics from kinetics. The thermodynamic hypothesis fixes the destination (the global minimum of $G$). The folding funnel fixes the route: because $G$ is funnel-shaped, the chain descends through a restricted ensemble of downhill pathways, gaining enthalpy as it loses entropy, so the effective search explores only a vanishing fraction of $\Omega$. The chaperone machinery handles the cases where the funnel is frustrated or where the cellular concentration makes the off-pathway aggregation trap competitive.

Bridge. The thermodynamic hypothesis fixes the destination of folding as the global free-energy minimum, and this is exactly the foundational reason the amino-acid sequence determines the three-dimensional structure; the question of how the chain reaches that minimum quickly appears again in the Master-tier energy-landscape analysis, where the folding funnel generalises the Anfinsen picture into a statistical-mechanical ensemble of downhill pathways that builds toward the native basin. The central insight is that thermodynamics locates the bottom of the free-energy surface while kinetics describes the routes down it.

Exercises Intermediate+

Energy landscapes, folding pathways, and chaperone machinery Master

The energy-landscape theory of Bryngelson, Wolynes, Onuchic, and coworkers recasts folding as motion on the free-energy surface $G(x)$ over conformation space $\Omega$ ^{[Bryngelson et al. 1995]}. For a random heteropolymer the landscape is rugged and pocked with deep kinetic traps, and folding is glassy. A natural protein sequence is minimally frustrated: evolutionary selection has pruned conflicting interactions so that native contacts are mutually reinforcing, and the landscape collapses into a funnel — a broad, high-enthalpy ensemble of unfolded conformations at the top that narrows toward the single native basin at the bottom. The funnel resolves the Levinthal paradox not by denying the vastness of $\Omega$ but by making the search biased: at each level the chain gains enthalpy as it loses entropy, the number of productive conformations contracts, and the chain descends through a restricted set of pathways rather than exploring $\Omega$ uniformly. This is a model of the folding mechanism as a statistical-mechanical ensemble of routes, not a single pathway.

Folding kinetics separates proteins broadly into two-state folders, which convert denatured $D$ to native $N$ without detectable intermediates, and folders that populate discrete intermediates. The molten globule — compact, with native-like secondary structure but a fluctuating, loosely packed core — is the canonical intermediate, characterised by Ptitsyn and coworkers ^{[Fersht 1999]}. Whether a given intermediate lies on-pathway (productive) or off-pathway (a kinetic trap) is a central experimental question. Fersht's $\Phi$-value analysis is the tool that answers it: by measuring the effect of conservative point mutations on folding and unfolding rates and on equilibrium stability, one assigns each residue a $\Phi \in [0,1]$ that reports how native-like its local environment is in the transition state. The resulting maps show that folding transition states are partially structured ensembles, consistent with the nucleation-condensation mechanism, in which weak local structure nucleates and then condenses the surrounding chain around it — a description that subsumes the older framework-and-side-chain and hydrophobic-collapse models as limiting cases.

In the cell the effective landscape differs from Anfinsen's dilute in vitro experiment in three respects: high protein concentration, which makes the off-pathway aggregation reaction competitive; the obligation to fold co-translationally as the chain emerges from the ribosome; and the presence of oxidative and thermal stress. Molecular chaperones manage these perturbations without changing the thermodynamic destination ^{[Hartl & Hayer-Hartl 2009]}. The Hsp70 system (DnaK/DnaJ/GrpE in E. coli) is a monomeric, ATP-driven holdase: the co-chaperone DnaJ delivers a short exposed hydrophobic segment of a nascent chain to the substrate-binding pocket of ATP-bound DnaK; ATP hydrolysis, stimulated by DnaJ, clamps the lid closed and isolates the sticky segment; the nucleotide-exchange factor GrpE then releases ADP, ATP rebinds, the lid opens, and the substrate is freed. Because a nontrivial fraction of newly synthesised chains would otherwise aggregate before reaching the native state, Hsp70 acts iteratively and co-translationally, buying time for productive folding.

The Hsp60 chaperonins (GroEL/GroES) handle a complementary, post-translational problem: small single-domain chains that have misfolded but remain compact enough to be rescued. GroEL is a double-ring tetradecamer whose open ring presents hydrophobic apical residues that capture a non-native substrate. ATP binding to the cis ring and docking of the GroES heptamer as a lid trigger a large apical-domain rotation that buries the hydrophobic residues and exposes polar surfaces, encapsulating the substrate in a now-hydrophilic cavity — the Anfinsen cage — where it can fold shielded from aggregation. ATP hydrolysis in the cis ring times the residence; ATP binding to the trans ring then allosterically ejects GroES and the substrate, which may rebind for further cycles if it has not yet folded.

When the folding and chaperone network fails, the consequence is aggregation. Misfolded chains expose hydrophobic surfaces that nucleate beta-sheet-rich aggregates: amyloid fibrils, the histological hallmark of diseases including Alzheimer's (amyloid-beta and tau), Parkinson's (alpha-synuclein), and Huntington's (polyglutamine). Prion disease is a qualitatively distinct failure, in which a misfolded conformer (${\text{PrP}}^{\text{Sc}}$) is itself a stable, self-templating alternative free-energy minimum that catalyses conversion of the native ${\text{PrP}}^{\text{C}}$ — a counterexample to the naive reading of the Anfinsen hypothesis as "the native state is always the global minimum." Loss-of-function folding diseases, such as the $\Delta$F508 cystic-fibrosis variant of CFTR, show the complementary failure mode: the cellular quality-control machinery (endoplasmic-reticulum-associated degradation) efficiently destroys even slightly misfolded but partially functional proteins.

Synthesis. Putting these together, the Anfinsen thermodynamic hypothesis and the Levinthal kinetic paradox are the two complementary faces of a single free-energy surface: the hypothesis fixes the global minimum as the native state, while the funnel generalises the search for that minimum into a guided, statistical-mechanical descent through molten-globule intermediates and nucleation-condensation transition states; the chaperone machinery (Hsp70 co-translational holdase, GroEL/GroES Anfinsen cage) is dual to the funnel in that it reshapes the effective landscape by suppressing aggregation traps rather than encoding the fold. The central insight is that sequence encodes the destination, landscape geometry encodes the route, and chaperones police the route against dead ends — this is exactly the foundational reason protein folding is both thermodynamically determined and kinetically feasible in the cell.

Connections Master

• Amino-acid and protein chemistry 15.12.01. The side-chain properties — hydrophobic, charged, polar, and disulfide-forming — set the interaction terms in the folding free energy $G(x)$. The disulfide-bond chemistry that anchors Anfinsen's ribonuclease A experiment is precisely the chemistry of cysteine oxidation introduced in the preceding unit.

• Peptide-bond geometry and the Ramachandran plot 15.12.02 pending. The backbone torsion angles $\varphi$ and $\psi$ define the conformation space $\Omega$ over which the folding landscape is erected, and the allowed regions of the Ramachandran plot constrain the funnel: folding pathways run through sterically permitted $(\varphi ,\psi )$ combinations, and the secondary structures (alpha helix, beta sheet) that nucleate early in folding occupy characteristic Ramachandran basins.

• Conformational analysis and steric energetics 15.01.02. The folding free energy inherits the steric and torsional analysis of small-molecule conformational chemistry, generalised from a single rotatable bond to the coupled torsion space of a biopolymer. The "allowed versus forbidden" reasoning of the Ramachandran plot is the biopolymer-scale descendant of Newman-projection conformational analysis.

• Enzyme mechanism 15.14.01. Catalytic activity presupposes the correct native fold: the active-site geometry, the precise positioning of catalytic residues, and the substrate-binding pocket all exist only once folding is complete. Protein folding is the upstream gate through which the chemistry of enzyme catalysis must pass.

• Protein structure and function in biology 17.05.01. The orgchem-side analysis of folding as a free-energy problem is completed on the cellular side by the biology of co-translational folding, the secretory pathway, and endoplasmic-reticulum quality control, where chaperone action and degradation machinery decide the fate of each nascent chain in vivo.

Historical & philosophical context Master

Christian Anfinsen's work at the U.S. National Institutes of Health through the 1950s and early 1960s established, through the ribonuclease A refolding experiments, that a denatured protein chain could refold to its native, active conformation without any external template. He articulated the consequence as the thermodynamic hypothesis: the native structure is the thermodynamically most stable form accessible to the sequence under native conditions ^{[Anfinsen 1973]}. The 1973 Science review that bears this argument was the distilled statement of the work cited in his share of the 1972 Nobel Prize in Chemistry (shared with Stanford Moore and William Stein, for work on ribonuclease).

Cyrus Levinthal's contribution came as an aside in a 1969 discussion at the Allerton House meeting on Mossbauer spectroscopy in biological systems, published as "How to fold graciously" ^{[Levinthal 1969]}. Levinthal observed that the number of conformations available even to a modest-sized protein is so large that an exhaustive search would exceed cosmological timescales, and concluded that folding must follow specific, biased pathways rather than sample conformation space freely. The remark, brief as it was, defined the kinetic folding problem and has anchored the field's motivation ever since.

The conceptual machinery to resolve the paradox matured over the following decades. Ptitsyn and coworkers formulated the molten globule as a physical intermediate state in the 1970s and 1980s. Bryngelson and Wolynes, and later Onuchic, Socci, and others, developed the energy-landscape theory and the folding-funnel picture, with the principle of minimal frustration as the evolutionary explanation for why natural-protein landscapes are funnel-shaped rather than glassy ^{[Bryngelson et al. 1995]}. Fersht and colleagues introduced $\Phi$-value analysis in the late 1980s and 1990s, giving the first experimental method for mapping the structure of a folding transition state residue by residue, and codified the nucleation-condensation mechanism.

The chaperone story reshaped the in vivo picture. The term "molecular chaperone" was coined by Ron Laskey in 1978 for nucleoplasmin, which assists nucleosome assembly without becoming part of the final structure. R. John Ellis extended the concept to chloroplast and ribosome-associated proteins. The decisive discovery that the Hsp60 class (GroEL/GroES in bacteria; the TRiC/CCT complex in the eukaryotic cytosol) actively assists protein folding came from Franz-Ulrich Hartl, Arthur Horwich, and coworkers in 1989, through genetic and biochemical analysis of mitochondrial import. Paul Sigler and colleagues solved the crystal structure of the asymmetric GroEL–GroES–(ADP)${}_7$ complex in 1996, revealing the Anfinsen cage directly. Hartl and Hayer-Hartl's 2009 review consolidated the convergence of the in vitro thermodynamic picture and the in vivo, chaperone-assisted picture ^{[Hartl & Hayer-Hartl 2009]}.

The philosophical content of the field is the relationship between thermodynamic and kinetic determinism. The Anfinsen hypothesis is a thermodynamic statement about a destination; the Levinthal paradox and the funnel theory are kinetic statements about routes. The modern synthesis holds that both are needed, and that in the living cell the relevant landscape is the effective one — reshaped at non-equilibrium steady state by ATP-driven chaperones that suppress dead ends without specifying the destination.

Bibliography Master

1. Anfinsen, C. B. (1973). Principles that govern the folding of protein chains. Science 181, 223–230.

2. Levinthal, C. (1969). How to fold graciously. In Mossbauer Spectroscopy in Biological Systems (P. Debrunner, J. C. M. Tsibris & E. Munck, eds.), Allerton House, University of Illinois Press, pp. 22–24.

3. Fersht, A. R. (1999). Structure and Mechanism in Protein Science: A Guide to Enzyme Catalysis and Protein Folding. W. H. Freeman, New York. (See Ch. 2 on protein stability, Ch. 19 on kinetics and folding pathways.)

4. Hartl, F. U. & Hayer-Hartl, M. (2009). Converging concepts of protein folding in vitro and in vivo. Nature Structural & Molecular Biology 16, 574–581.

5. Creighton, T. E. (1993). Proteins: Structures and Molecular Properties, 2nd ed. W. H. Freeman, New York. (Ch. 3–4 on denaturation, folding, and intermediates.)

6. Bryngelson, J. D., Onuchic, J. N., Socci, N. D. & Wolynes, P. G. (1995). Funnels, pathways, and the energy landscape of protein folding: a synthesis. Proteins: Structure, Function, and Bioinformatics 21, 167–195.

7. Fersht, A. R., Matouschek, A. & Serrano, L. (1992). The folding of an enzyme. I. Theory of subunit interactions in the transition state of barnase. Journal of Molecular Biology 224, 771–782. (Foundational paper on $\Phi$-value analysis.)

8. Ptitsyn, O. B. (1995). Structures of folding intermediates. Current Opinion in Structural Biology 5, 74–79. (The molten globule as a physical folding intermediate.)

9. Xu, Z., Horwich, A. L. & Sigler, P. B. (1997). The crystal structure of the asymmetric GroEL–GroES–(ADP)${}_7$ chaperonin complex. Nature 388, 741–750. (The Anfinsen cage at atomic resolution.)

10. Laskey, R. A., Honda, B. M., Mills, A. D. & Finch, J. T. (1978). Nucleosomes are assembled by an acidic protein which binds histones and transfers them to DNA. Nature 275, 416–420. (Origin of the term "molecular chaperone.")