ated, representing an uncoiled helical dimer. Hu- 
man bias in the placing of the other atoms is reduced 
by an automatic building procedure that employs 
simulated annealing with simple geometric re- 
straints. The resulting all-atom model is then al- 
lowed to relax during a short molecular dynamics 
run in vacuo, using an empirical energy function 
and weak restraints that reflect the helical supercoil 
assumption. Several models can be obtained by us- 
ing diff'erent initial conditions for the simulated an- 
nealing procedure. These models can be further re- 
fined through use of unrestrained molecular 
dynamics in the appropriate environment. 
As a first step toward understanding the specific- 
ity of protein-protein interactions in leucine zip- 
pers, Dr. Briinger's group applied the simulated an- 
nealing procedure to the dimerization region of the 
transcriptional activator protein GCN4. The pre- 
dicted models were obtained prior to the publica- 
tion of the x-ray structure of GCN4 by Dr. Thomas 
Alber's group (University of Utah) . The predicted 
models turned out to be fairly close to the x-ray 
structure. 
Both the predicted structure and the x-ray crystal 
structure are close to a classical left-handed coiled 
coil conformation. The local helix-helix crossing 
angle of the x-ray structure falls within the range 
predicted by the models; a slight unwinding of the 
coiled coil toward the amino-terminal DNA-binding 
end of the dimerization region has been correctly 
predicted. Distance maps between the helices are 
largely identical. The region around asparagine 20 
is asymmetric in the x-ray structure and in the mod- 
els. Surface side-chain dihedrals showed a large vari- 
ation in the models as expected, because of the 
highly solvent-exposed surface area of the leucine 
zipper. 
Recent studies suggest specific roles for trans- 
membrane helix interactions in a range of functions, 
but understanding of the conformation and ener- 
getics of such interactions has been elusive. Dr. 
Briinger's group has studied the dimerization of the 
transmembrane helix of glycophorin-A by the simu- 
lated annealing procedure and has tested the models 
against mutational analysis data obtained by Dr. 
Donald Engelman's group (Yale University). 
It was found that the dimer is a right-handed su- 
percoil, that an extensive region of close packing 
lies along the dimer interface, and that there is good 
agreement with the mutagenesis data. Furthermore, 
a sequence-specific propensity for a right-handed 
supercoil was observed when the simulated anneal- 
ing modeling was started from a dimer of uncoiled 
helices. Upon replacement of the simulated glyco- 
phorin-A sequence by an 1 8-residue fragment of the 
dimerization region of the transcription factor 
GCN4, the system shows a high propensity for a left- 
handed supercoil. This result supports the hope of 
predicting folded transmembrane protein structures 
in the hydrophobic region of the lipid bilayer. 
Thermodynamics of Protein-Peptide 
Interactions 
Internal packing effects are thought to be impor- 
tant in determining protein structure and stability. 
Packing effects also participate in recognition and 
binding, and their analysis is a step toward rational 
methods of drug design. Several authors have stud- 
ied the effect that substitutions of buried hydropho- 
bic residues have on protein stability. Mutation ef- 
fects can be very complex, with small changes in 
stability resulting from large enthalpy-entropy com- 
pensation, and from many large, mutually cancel- 
ing, atomic contributions, involving both protein 
and solvent and both folded and unfolded states. 
Theoretical work can help in understanding the de- 
tails of these microscopic effects. Free-energy cal- 
culations are one such approach. 
The hydrophobic interactions between the S pep- 
tide and S protein in the ribonuclease-S complex 
were investigated using molecular dynamics simula- 
tions and free-energy calculations, in a collabora- 
tion with Dr. Frederic Richards (Yale University). 
Three mutations at the buried position Met 13 were 
simulated — M ^ L, L I, and I V — for which 
x-ray structures and experimental thermodynamic 
data were available. The calculations gave theoreti- 
cal estimates of the changes in binding free energies 
associated with these mutations. The calculated 
free-energy differences were small (0-1.6 kcal/ 
mol), in agreement with experiment. However, the 
simulated structures deviated significantly from the 
experimental ones, and a large uncertainty in the 
calculated free energies (~2 kcal/mol) arose from 
the multiple minima problem. Indeed, multiple 
conformations are available to the side chains 
around the mutation site, and the sampling of dihe- 
dral rotamer transitions is limited, despite long 
simulations. 
The uncertainty due to multiple conformations is 
much greater than that due to random statistical 
errors. Thus the general criterion for a precise simu- 
lation is not merely to reduce the random statistical 
error, as has been suggested, but rather to sample all 
the important local minima along the mutation 
pathway and to reduce the statistical error for each 
one. The calculations suggest that the packing 
changes associated with the mutations are energeti- 
cally small and localized, and largely cancel when 
the complex and the S peptide are compared. 
STRUCTURAL BIOLOGY 465 
