Figure 1. Schematic (‘ribbon’) diagram of a typical, well-studied protein, bovine pancreatic ribonuclease A. The helices and strands are indicated. The thin connections between secondary structure elements do not indicate unstructured segments, but rather segments of irregular structure.

Figure 2. The three fundamental processes of nucleation scenarios: growth, rearrangements and coalescence steps. The thin lines represent unstructured segments of the protein, while the rectangular regions represent structured segments. The grey interior and the dark grey interior of these rectangles (seen here in the rearrangements) represent native and nonnative structure, respectively. In the growth step, the structured segment becomes larger at the expense of the unstructured segment. In the rearrangement step, native and nonnative structure interconvert. Finally, in the coalescence step, two structured segments coalesce and mutually stabilize one another. The entropy of the unstructured segments equals that of the statistical-coil state, minus a correction if a loop is formed at its two ends, e.g. in the coalescence step.

Figure 3. The chain-folding initiation site (CFIS) scenario (Lewis et al., (1970) (1971); Scheraga, (1971); Anfinsen, (1973); Tanaka and Scheraga, (1977); Matheson and Scheraga, 1978). In this model, elements of flickering local structure such as nascent helices and hairpins are present in the denatured state, with a probability roughly equal to that of the statistical-coil state. These elements coalesce with a probability roughly proportional to the simple loop entropy penalty, although the association of certain elements may be especially favoured by the propensities of the intervening segment. At the transition state, sufficient structure has coalesced and become stabilized that the forward rate of folding becomes faster than the backward rate of unfolding; the fully native structure then forms quickly, presumably by further growth and/or coalescence steps. This CFIS model does not feature an initial hydrophobic collapse. However, it does assume that the transient structures are stabilized primarily by hydrophobic interactions and by the indirect stabilization of hydrogen bonds by nonpolar residues (see main text). This model naturally allows for multiple pathways, with probabilities corresponding to those of the nucleation and coalescence steps outlined above; for example, in the diagram, the second and third helices could associate first for some molecules, instead of the first and second helices as depicted. However, contrary to the funnel models and other delocalized nucleation mechanisms, this model supposes that certain segments of the protein are much more likely to become structured than are others.

Figure 4. Other conformational folding scenarios illustrating alternative ways of trading entropy for energy during folding. In the Levitt–Warshel (1975) scenario, the unfolded protein rapidly adopts the global native topology without forming any structure; a small section of native structure can then form in one region and grow to encompass the whole protein. In the G (1984) scenario, the protein collapses hydrophobically, forming secondary structure in the process; however, the initial globule does not have a defined topology. Rearrangements of the globule lead to the molten-globule state predicted by Ptitsyn (1973), which has the native secondary structure and native topology but lacks the specific side-chain packing of the native state. In the Dill (1985) scenario, the protein collapses hydrophobically to a condensed, disordered state with neither structure nor topology; random searching in this condensed, entropically favoured state leads to the formation of a section of native structure that grows to encompass the whole protein.