Protein Folding

Protein folding is the physical and thermodynamic self-assembly process by which a linear polypeptide chain—synthesized on the ribosome from a genetically encoded sequence of L-amino acids—spontaneously transitions from a disordered random coil into its biologically active, three-dimensional native conformation. Across the structural hierarchy codified by Kaj Linderstrøm-Lang (1952), the one-dimensional primary structure (amino acid sequence linked by planar peptide bonds) dictates the formation of regular hydrogen-bonded secondary structure elements (α-helices, β-sheets, and β-turns), which pack together into a compact three-dimensional tertiary structure (and, in multisubunit complexes, assemble into a quaternary structure). Because virtually all biological functions of proteins—including enzymatic catalysis, receptor signaling, ion channel gating, and cytoskeletal mechanics—depend on the sub-angstrom spatial positioning of functional side chains, understanding how amino acid sequence specifies three-dimensional fold (the protein folding problem) bridges molecular biology, statistical mechanics, and computational biophysics.

Prior to the late 1950s, it was widely debated whether nascent polypeptides require an external cellular template to sculpt their tertiary fold. Between 1957 and 1961, American biochemist Christian B. Anfinsen (awarded the 1972 Nobel Prize in Chemistry) resolved this question using bovine pancreatic ribonuclease A (RNase A), a single-chain 124-amino-acid enzyme stabilized by four specific intramolecular disulfide bonds (Cys26–Cys84, Cys40–Cys95, Cys58–Cys110, and Cys65–Cys72) among its eight cysteine residues:

1. Reductive Denaturation: Treating native RNase A with 8 M urea (a chaotropic agent that disrupts non-covalent hydrogen bonds and hydrophobic packing) and 2-mercaptoethanol (a thiol reducing agent that cleaves all four disulfide bonds into eight –SH groups) unfolded the protein into a random coil with 0% catalytic activity.
2. Spontaneous Refolding Upon Dialysis: When both the 8 M urea and the 2-mercaptoethanol were dialyzed away simultaneously in the presence of trace air oxygen at physiological pH, the unfolded polypeptide spontaneously refolded and oxidized the exact four native disulfide pairings—out of 7 × 5 × 3 × 1 = 105 statistically possible disulfide combinations—restoring >95%–100% of native enzymatic activity.
3. Scrambled Ribonuclease Control: Crucially, if the disulfide bonds were allowed to re-oxidize while 8 M urea was still present, the random coil formed a mixture of all 105 possible disulfide pairings ("scrambled RNase," ~1% active). Subsequent removal of urea and addition of a catalytic trace of 2-mercaptoethanol (enabling disulfide interchange) allowed non-covalent side-chain interactions to drive rearrangement into the single native four-disulfide state.

From these experiments, Anfinsen formulated Anfinsen's dogma (the thermodynamic hypothesis): all information required to specify the three-dimensional tertiary structure of a protein is contained entirely within its linear amino acid sequence, and the native state corresponds to the global minimum of Gibbs free energy (G) under physiological solvent conditions.

At physiological temperature, the net thermodynamic stability of a typical globular domain relative to its unfolded ensemble is surprisingly marginal—typically Δ*G*_fold = −20 to −65 kJ·mol⁻¹ (equivalent to only 2 to 5 hydrogen bonds), because folding represents a delicate cancellation between large opposing enthalpic and entropic terms:

• Conformational Entropy of the Polypeptide Chain (Δ*S*_chain < 0 — Strongly Opposes Folding): In the unfolded random coil, each amino acid residue samples multiple main-chain dihedral angle minima (φ, ψ on the Ramachandran plot) and side-chain rotamers (W_U ≈ 3 to 10 conformations per residue). Restricting a 100-residue polypeptide into a single unique native fold (W_N ≈ 1) incurs a massive conformational entropy penalty (−*T*Δ*S*_chain ≈ +300 to +500 kJ·mol⁻¹ at 298 K).
• The Hydrophobic Effect (Δ*S*_solvent > 0 — Primary Thermodynamic Driver of Collapse): Originally formulated by Walter Kauzmann (1959), nonpolar aliphatic and aromatic side chains (leucine, isoleucine, valine, phenylalanine, tryptophan, methionine) cannot form hydrogen bonds with water. In the unfolded state, water molecules surrounding exposed nonpolar side chains are forced into ordered, low-entropy hydrogen-bonded cages (clathrate-like solvation shells). Burying these nonpolar side chains inside a water-excluded hydrophobic core releases the ordered water molecules back into bulk solvent, generating a large positive increase in solvent entropy (Δ*S*_solvent > 0) accompanied by a large decrease in heat capacity (Δ*C*_p < 0).
• Intramolecular Hydrogen Bonding, London Dispersion Packing, and Salt Bridges (Δ*H*_fold < 0): Because polar backbone amide (N–H) and carbonyl (C=O) groups form strong hydrogen bonds with water in the unfolded state, burying an unsatisfied polar group inside the hydrophobic core incurs a severe desolvation penalty (+15 to +25 kJ·mol⁻¹). Secondary structures (α-helices with i to i + 4 C=O···H–N hydrogen bonds, and β-sheets) satisfy backbone hydrogen-bond donors and acceptors internally upon burial. Simultaneously, tight jigsaw-like steric packing inside the core (achieving a packing density of ~0.75, comparable to organic crystals) maximizes short-range London dispersion forces, while surface and buried ionic salt bridges (e.g., Lys⁺/Arg⁺ paired with Asp⁻/Glu⁻) and disulfide bonds (Cys–S–S–Cys) lock the native topology in place.

Levinthal's Paradox (1969)

In 1969, molecular biologist Cyrus Levinthal pointed out a fundamental kinetic contradiction if protein folding occurred via an unbiased random search of all possible conformations. If each amino acid residue in a modest 100-residue polypeptide samples just 3 backbone states, the polypeptide possesses 3¹⁰⁰ ≈ 5 × 10⁴⁷ distinct conformations. Even if the chain could interconvert between conformations at the picosecond bond-vibration limit (10¹² to 10¹³ conformations per second), an exhaustive random search would require ~10³⁵ seconds (~10²⁷ years)—orders of magnitude longer than the 13.8-billion-year age of the universe (~4 × 10¹⁷ seconds). Yet in reality, small single-domain proteins fold reproducibly in microseconds to seconds.

Resolution: Funnel Energy Landscapes and the Principle of Minimal Frustration

Developed in the late 1980s and 1990s by Joseph Bryngelson, Peter Wolynes, José Onuchic, and Ken Dill, statistical-mechanical energy landscape theory resolves Levinthal's paradox by demonstrating that proteins do not search conformational space at random. Instead, natural selection has evolved biological sequences that obey the principle of minimal frustration: native-like contacts are systematically more energetically favorable than non-native contacts, sculpting the multidimensional free-energy surface into a folding funnel:

• Width of the Funnel (Horizontal Axis): Represents the conformational entropy (S) of the polypeptide chain (broad at the top where myriads of unfolded conformations exist; narrow at the bottom where only the unique native state N exists).
• Depth of the Funnel (Vertical Axis): Represents the internal free energy (E or H − TS_solvent) (inclusive of hydrophobic solvation and intramolecular interactions).

Because any local native-like contact lowers the energy of the chain, a folding protein slides downhill along multiple parallel converging trajectories toward the native basin:

1. Hydrophobic Collapse and the Molten Globule (<1 ms): Within microseconds of dilution from denaturant, nonpolar side chains coalesce into a compact molten globule intermediate possessing substantial native-like secondary structure (α-helices and β-strands) and a hydrated hydrophobic core, but lacking fixed side-chain tertiary packing.
2. Nucleation–Condensation: A small subset of native secondary and tertiary contacts forms a folding nucleus (the transition state ensemble), around which the remainder of the domain rapidly condenses into the native state, occasionally pausing to isomerize cis-proline peptide bonds (catalyzed in vivo by peptidyl-prolyl cis–trans isomerases, PPIases) or rearrange disulfide bonds (catalyzed by protein disulfide isomerase, PDI, in the endoplasmic reticulum).

Inside the living cell, macromolecular concentration reaches 300 to 400 mg·mL⁻¹ (macromolecular crowding). As nascent polypeptide chains emerge vectorially from the ribosomal exit tunnel (at ~5–20 amino acids per second), exposed hydrophobic segments on partially folded intermediates risk intermolecular aggregation with neighboring chains before their C-terminal domain has even been synthesized.

To prevent off-pathway aggregation without violating Anfinsen's principle (since they do not specify the final tertiary structure), cells deploy ATP-driven molecular chaperones (many designated heat shock proteins, Hsps):

• Hsp70 Chaperone System (Bacterial DnaK/DnaJ/GrpE; Eukaryotic Hsp70/Hsp40/BAG): Monomeric ~70-kDa chaperones bind short extended hydrophobic heptapeptide segments exposed on nascent or stress-unfolded chains. Driven by Hsp40-stimulated ATP hydrolysis and nucleotide exchange factors, Hsp70 cycles between an ATP-bound open state and an ADP-bound closed state, shielding sticky patches and unfolding kinetically trapped misfolded intermediates so they can retry folding down the native funnel.
• Hsp60 Cage Chaperonins (Bacterial GroEL/GroES; Eukaryotic Cytosolic TRiC/CCT): Resolved crystallographically by Arthur Horwich and Franz-Ulrich Hartl, bacterial GroEL (~800 kDa) consists of two back-to-back heptameric rings forming two cylindrical Anfinsen cages, capped by the dome-shaped heptameric co-chaperonin GroES. A non-native substrate polypeptide (~20–60 kDa) is captured via hydrophobic lining residues inside the open cis GroEL ring. Cooperative binding of 7 ATP and the GroES cap triggers a large domain twist that buries GroEL's hydrophobic wall and exposes a hydrophilic, negatively charged chamber interior, encapsulating the single polypeptide in total isolation for ~10 seconds (the duration of ATP hydrolysis) so it can fold without any possibility of intermolecular aggregation.

Because the cross-β amyloid fibril state—in which β-strands from multiple polypeptide chains stack perpendicular to the fibril axis into insoluble, protease-resistant cross-β sheets spaced 4.7 Å apart—represents an even deeper thermodynamic free-energy minimum at high protein concentration than the monomeric native state, failures of cellular proteostasis (or clearance by the ubiquitin–proteasome system and autophagy) cause protein misfolding diseases (proteopathies):

• Neurodegenerative Amyloidoses: Extracellular plaques of amyloid-β (Aβ₄₂) and intracellular neurofibrillary tangles of hyperphosphorylated tau in Alzheimer's disease; intracellular Lewy bodies of α-synuclein in Parkinson's disease; polyglutamine (polyQ) expansion tracts in Huntington's disease; and SOD1 / TDP-43 aggregates in amyotrophic lateral sclerosis (ALS).
• Infectious Prions (PrP^C → PrP^Sc): Formulated by Stanley B. Prusiner, mammalian prion protein (PrP^C) is a normal neuronal GPI-anchored glycoprotein rich in α-helices (~42% α-helix, ~3% β-sheet). If an exogenous or spontaneous misfolded scrapie isoform (PrP^Sc) (~43% β-sheet) contacts native PrP^C, PrP^Sc acts as a conformational template that catalyzes the autocatalytic conversion of α-helical PrP^C into infectious β-sheet-rich PrP^Sc amyloid aggregates—transmitting Creutzfeldt–Jakob disease (CJD), kuru, ovine scrapie, and bovine spongiform encephalopathy without any nucleic acid genome.
• Deep-Learning Structure Prediction (AlphaFold): In 2020–2024, Demis Hassabis, John Jumper, and David Baker (awarded the 2024 Nobel Prize in Chemistry) demonstrated that deep-learning neural networks (AlphaFold2/3 and RoseTTAFold) trained on evolutionary multiple-sequence alignments (co-evolving residue pairs in contact) and the Protein Data Bank can predict atomic-resolution 3D coordinates directly from primary amino acid sequences across more than 200 million proteins, while de novo computational design constructs entirely new protein folds never sampled by biological evolution.

If the native state is the thermodynamic minimum (Anfinsen's dogma), why do molecular chaperones like GroEL/GroES require ATP hydrolysis?

Molecular chaperones do not alter the thermodynamic equilibrium between the monomeric unfolded state and the monomeric native state, nor do they provide steric blueprints that dictate which fold a protein adopts. However, inside the crowded cytosol, partially folded intermediates face a kinetic competition between first-order intramolecular folding (U → N) and higher-order intermolecular aggregation (I + I → Aggregate), and multi-chain aggregates sit in deep kinetic traps. Chaperones expend the free energy of ATP hydrolysis to drive cyclical conformational changes that pull apart misfolded kinetic traps (iterative annealing) and transiently isolate individual chains inside hydrophilic cages (GroEL/GroES), resetting the chain to the top of the folding funnel so it can fold spontaneously toward its Anfinsen native state.

How can prions transmit inherited and infectious disease without DNA or RNA?

Under the central dogma of molecular biology, sequence information flows from nucleic acid to protein. Prions do not violate the rule that amino acid sequence is encoded by DNA; rather, the host's own chromosomal PRNP gene encodes the normal PrP^C polypeptide sequence, which possesses two distinct local minima on its free-energy landscape separated by a high kinetic activation barrier: the monomeric α-helical PrP^C state and the oligomeric β-sheet PrP^Sc amyloid state. On its own, monomeric PrP^C almost never crosses the barrier to PrP^Sc; however, when a pre-formed PrP^Sc aggregate enters the tissue, its exposed β-sheet edges bind host PrP^C and lower the activation barrier (seeded nucleation–polymerization), converting host PrP^C into additional PrP^Sc fibrils via autocatalytic fibril fragmentation and end-seeding.

What are Intrinsically Disordered Proteins (IDPs), and do they violate the structure–function paradigm?

Roughly 30% to 40% of eukaryotic proteins contain long segments (intrinsically disordered regions, IDRs) or entire polypeptide chains (intrinsically disordered proteins, IDPs, such as the transactivation domain of tumor suppressor p53, α-synuclein in solution, and nuclear pore FG-nucleoporins) that lack a fixed three-dimensional tertiary structure under physiological conditions. Rather than folding into a single funnel minimum, IDPs are depleted in bulky hydrophobic core residues (Ile, Leu, Val, Phe, Trp) and enriched in charged and polar residues (Pro, Glu, Lys, Ser, Gln), giving them a flat, multi-minima energy landscape. This conformational plasticity enables IDPs to undergo coupled folding and binding upon contacting specific partner proteins, expose dense arrays of post-translational modification sites, and drive liquid–liquid phase separation (LLPS) of membrane-less biomolecular condensates such as the nucleolus and stress granules.

1. Anfinsen, C. B. (1973). "Principles that govern the folding of protein chains." Science, 181(4096), 223–230. https://doi.org/10.1126/science.181.4096.223
2. Dill, K. A., & MacCallum, J. L. (2012). "The protein-folding problem, 50 years on." Science, 338(6110), 1042–1046. https://doi.org/10.1126/science.1219021
3. Onuchic, J. N., Luthey-Schulten, Z., & Wolynes, P. G. (1997). "Theory of protein folding: The energy landscape perspective." Annual Review of Physical Chemistry, 48, 545–600. https://doi.org/10.1146/annurev.physchem.48.1.545
4. Hartl, F. U., Bracher, A., & Hayer-Hartl, M. (2011). "Molecular chaperones in protein folding and proteostasis." Nature, 475(7356), 324–332. https://doi.org/10.1038/nature10317
5. Jumper, J., Evans, R., Pritzel, A., et al. (2021). "Highly accurate protein structure prediction with AlphaFold." Nature, 596(7873), 583–589. https://doi.org/10.1038/s41586-021-03819-2