In this paper we describe the achievements and pitfalls encountered in
doing structure predictions of protein mutants using molecular dynami
cs simulation techniques in which properties of atoms are slowly chang
ed as a function of time. Basically the method consists of a thermodyn
amic integration (slow growth) calculation used for free energy determ
ination, but aimed at structure prediction; this allows for a fast det
ermination of the mutant structure. We compared the calculated structu
re of the mutants Met222Ala, Met222Phe and Met222Gln of subtilisin BPN
' with the respective X-ray structures and found good agreement betwee
n predicted and X-ray structure. The conformation of the residue subje
ct to the mutation is relatively easy to predict and is mainly determi
ned by packing criteria. When the side chain has polar groups its exac
t orientation may pose problems; long-range Coulomb interactions may g
enerate a polarization feedback involving system relaxation times beyo
nd the simulation time. Changes induced in the environment are harder
to predict using this method. In particular, rearrangement of the hydr
ation structure was difficult to predict correctly, probably because o
f the long relaxation times. In all conversions made the changes obser
ved in the environment were found to be history-dependent and in parti
cular the hydrogen bonding patterns provided evidence for metastable s
ubstates. In all cases the structure predicted was compared with avail
able kinetic data and the reduced activity could be explained in terms
of changes in the configuration of the active site.