Protein Folding: How Chains Find Their Shape

A newly made protein leaves the ribosome as a floppy chain of amino acids. Within milliseconds to seconds, most chains fold into one precise three-dimensional shape, the only shape in which the protein works. The chain does this without a template, without instructions beyond its own sequence, and with astonishing reliability. How it manages this is one of the great questions of biochemistry, and when folding goes wrong, the results include some of the most feared diseases in medicine.

The sequence contains the shape: Anfinsen’s experiment

In the late 1950s and early 1960s, Christian Anfinsen studied ribonuclease A, a small enzyme of 124 amino acids containing four disulfide bridges.

1. He treated the enzyme with urea (which disrupts hydrogen bonds and hydrophobic interactions) and a reducing agent (which breaks disulfide bridges). The protein unfolded completely and lost all activity.
2. He then removed the urea and reducing agent slowly, letting the protein re-oxidise in air.
3. The protein refolded spontaneously and recovered almost all of its activity, with all four disulfide bridges back in the right places, out of 105 possible ways to pair its eight cysteines.

His conclusion, often called Anfinsen’s dogma: the native (folded) structure is determined by the amino acid sequence and represents the most stable arrangement (the lowest free energy) under physiological conditions. He shared the 1972 Nobel Prize in Chemistry for this work.

What drives folding?

Folding is a thermodynamic balance. A folded protein has lower free energy (ΔG) than the unfolded chain, but not by much. Typical folded proteins are only about 20 to 60 kJ mol⁻¹ more stable than their unfolded forms, roughly the energy of a few hydrogen bonds. Several forces contribute:

The hydrophobic effect (the main driving force)

Non-polar side chains in an unfolded chain force surrounding water molecules into ordered cages, which lowers the entropy of the water. When the chain folds and buries these side chains in a hydrophobic core, the water is released and its entropy rises. This gain in water’s entropy is the biggest single driving force for folding. It’s the same effect that makes oil droplets merge (see intermolecular forces).

Hydrogen bonds and electrostatic interactions

Hydrogen bonds within the backbone form α-helices and β-sheets, and hydrogen bonds and ionic bonds between side chains help stabilise the fold (see protein structure levels). Because unfolded chains also form hydrogen bonds with water, these interactions contribute less to the difference in stability than you might expect, but they’re essential for making the correct fold more stable than wrong ones.

Van der Waals packing

The protein core is packed as tightly as a crystal, with many weak London dispersion contacts.

Opposing folding: conformational entropy

The unfolded chain can take a huge number of shapes; the folded one is essentially a single shape. Losing that freedom costs a lot of entropy, which works against folding. The final stability is the small difference between large opposing terms, which is why proteins are only marginally stable and why modest heating or pH changes can unfold them (denaturation).

Levinthal’s paradox and folding funnels

In 1969, Cyrus Levinthal pointed out a problem. Suppose each amino acid in a 100-residue chain could adopt just three backbone conformations. That gives 3¹⁰⁰, about 5 × 10⁴⁷, possible shapes. Even sampling a shape every 10⁻¹³ seconds, a random search would take vastly longer than the age of the universe. Yet real proteins fold in milliseconds to seconds.

The resolution is that folding is not a random search. Local interactions form quickly: helices can form in microseconds. Partially correct structures are more stable than random ones, so the chain is guided downhill. Modern theory describes this with an energy landscape shaped like a funnel:

• The wide top represents the many high-energy unfolded shapes.
• As the chain forms more native contacts, its energy falls and the number of possible shapes narrows.
• The narrow bottom is the single native structure.
• The funnel’s walls can be rough, with small traps where the chain may pause in partly folded intermediates.

Small proteins often fold in a single cooperative step; larger ones may pass through intermediates, such as a “molten globule” that has most of its secondary structure but a loosely packed core.

Chaperones: help inside the crowded cell

The inside of a cell is extremely crowded, with protein concentrations of around 300 grams per litre. Partly folded chains expose sticky hydrophobic patches that can clump together with other chains. Cells use molecular chaperones to prevent this:

• Hsp70 proteins bind to exposed hydrophobic stretches of newly made chains, preventing them from sticking together.
• Chaperonins such as the bacterial GroEL–GroES complex form a barrel-shaped chamber in which a single protein chain can fold in isolation, using energy from ATP.

Chaperones don’t contain information about the final shape; they just give the chain a better chance to follow its own sequence to the right fold. Many are “heat shock proteins”, made in larger amounts when cells are stressed by heat, which tends to unfold proteins.

When folding goes wrong

Misfolded proteins are usually recognised and destroyed by the cell’s quality-control systems. When these systems fail, or when a mutation makes misfolding more likely, the consequences can be severe.

Aggregation and amyloid

Some misfolded proteins stick together into long, insoluble fibres called amyloid, rich in β-sheet structure. Amyloid deposits are associated with:

• Alzheimer’s disease, involving amyloid-β peptides and the protein tau;
• Parkinson’s disease, involving α-synuclein;
• Huntington’s disease, involving huntingtin with an abnormally long run of glutamine residues;
• type 2 diabetes, where amyloid forms in the pancreas.

Prions

Prion diseases, such as Creutzfeldt–Jakob disease in humans and BSE (“mad cow disease”) in cattle, involve a normal brain protein (PrP) that can adopt a misfolded, β-sheet-rich shape. The misfolded form can convert normal PrP into more misfolded copies, a chain reaction that makes the disease transmissible without any DNA or RNA. Stanley Prusiner received the 1997 Nobel Prize in Physiology or Medicine for discovering prions.

Loss of function

In cystic fibrosis, the most common mutation removes a single amino acid from a chloride channel protein. The protein misfolds, is recognised as faulty and is destroyed before it reaches the cell membrane. Some modern drugs, called “correctors”, help the faulty protein fold well enough to reach the membrane and work.

Predicting folding: from physics to AI

For decades, predicting a protein’s structure from its sequence was a famous unsolved problem. Physics-based simulations can model folding for small proteins but need enormous computing power. In 2020, the AI system AlphaFold (from Google DeepMind) predicted protein structures at an accuracy close to experimental methods in an international blind test. Its developers, together with David Baker for computational protein design, received the 2024 Nobel Prize in Chemistry. Predicted structures for hundreds of millions of proteins are now freely available, transforming research in biology and drug discovery.

AI prediction doesn’t fully explain how proteins fold, and it struggles with some cases (such as proteins that change shape, or that are naturally disordered), but it has shown that the sequence does indeed contain the information for the fold, just as Anfinsen concluded.

Key takeaways

• Anfinsen’s experiment showed that the amino acid sequence contains all the information needed for a protein to fold.
• The hydrophobic effect is the main driving force; hydrogen bonds, ionic bonds and tight packing make the correct fold favoured.
• Folded proteins are only marginally stable, typically by 20 to 60 kJ mol⁻¹.
• Levinthal’s paradox is resolved by a funnel-shaped energy landscape: folding is guided, not random.
• Chaperones help chains fold in crowded cells; misfolding underlies amyloid diseases, prion diseases and cystic fibrosis.
• AI tools such as AlphaFold now predict most structures from sequence. Start with proteins: chains of amino acids for the basics.