Can protein unfolding simulate protein folding?

Since the discovery that the native structure of a globular protein can fold spontaneously (Anfinsen et al., 1961), numerous experimental and theoretical efforts have been made to elucidate folding pathways, intermediates, transition states, etc. (for recent reviews, see Bryngelson et al., 1995; Fersht, 1995, 1997; Karplus and Šali, 1995; Ptitsyn, 1995; Miranker and Dobson, 1996; Shakhnovich, 1997). Protein folding pathways are of great interest not only in themselves, but also because understanding them is important for both protein structure predictions (one has to know what to search for: the most stable chain fold or the one resulting from a rapid folding pathway) and for de novo protein design (one has to know what to design: a stable fold only or a pathway to this fold also). Molecular dynamics (MD) simulations aim to provide the most detailed description of the behavior of ‘realistic’ atomic models of proteins during their folding or unfolding processes. In practice, however, these simulations are used mostly to study the unfolding from the native state (Caflisch and Karplus, 1994, 1995; Daggett and Levitt, 1994; Hunenberger et al., 1995; Daggett et al., 1996; Li and Daggett, 1996; Williams et al., 1997). The reason is simple. Current MD simulations, even with the most powerful computational techniques, can barely explore more than a few nanoseconds for an all-atom model of a protein and its solvent environment. On the other hand, the folding or unfolding experiment takes not less than hundreds of microseconds (Huang and Oas, 1995; Schindler et al., 1995; Eaton et al., 1997), and usually milliseconds, seconds or minutes (Segawa and Sugihara, 1984; Radford et al., 1992; Itzhaki et al., 1994; Eaton et al., 1997; Roder and Colón, 1997). Thus, one has to accelerate the events in simulations. This can be done with high temperature, but a high temperature corresponds to unfolding conditions. Therefore, unfolding is much more tractable than folding for molecular dynamics. To observe anything interesting on the nanosecond simulation time-scale, MD has to operate under extreme unfolding conditions where the protein virtually explodes: usually, a temperature of 500–600 K is used (Caflisch and Karplus, 1994, 1995; Daggett and Levitt, 1994; Hunenberger et al., 1995; Daggett et al., 1996; Li and Daggett, 1996) to surmount the unfolding activation barrier over 1 ns. Such a high temperature must accelerate the processes by about six orders of magnitude (Karplus and Šali, 1995) compared with room temperature but, certainly, it allows an investigation of the protein unfolding process. So far, to the author’s knowledge, no-one has suggested an effective way to accelerate the folding process, while unfolding can be accelerated in different ways. To avoid an unrealistically high temperature of 500–600 K, some investigators have used low pH (strong electrostatic repulsion) combined with a