Density-functional geometry optimization of the 150,000-atom photosystem-I trimer

We present a linear-scaling method based on the use of density-functional theory (DFT) for the system-wide optimization of x-ray structural coordinates and apply it to optimize the 150,000 atoms of the photosystem-I (PS-I) trimer. The method is based on repetitive applications of a multilevel ONIOM procedure using the PW916-31G(d) DFT calculations for the high level and PM3 for the lower level; this method treats all atoms in the structure equivalently, a structure in which the majority of the atoms can be considered as part of some internal "active site." To obtain a realistic single structure, some changes to the original protein model were necessary but these are kept to a minimum in order that the optimized structure most closely resembles the original x-ray one. Optimization has profound effects on the perceived electronic properties of the cofactors, with, e.g., optimization lowering the internal energy of the chlorophylls by on average 53 kcal mol(-1) and eliminates the enormous 115 kcal mol(-1) energy spread depicted by the original x-ray heavy-atom coordinates. A highly precise structure for PS-I results that is suitable for analysis of device function. Significant qualitative features of the structure are also improved such as correction of an error in the stereochemistry of one of the chlorophylls in the "special pair" of the reaction center, as well as the replacement of a water molecule with a metal cation in a critical region on the C3 axis. The method also reveals other unusual features of the structure, leading both to suggestions concerning device functionality and possible mutations between gene sequencing and x-ray structure determination. The optimization scheme is thus shown to augment the molecular modeling schemes that are currently used to add medium-resolution structural information to the raw scattering data in order to obtain atomically resolved structures. System-wide optimization is now a feasible process and its use within protein x-ray data refinement should be considered.