A low-resolution model is used together with recently developed knowledge-based potentials for exploring the dynamics of proteins. Configurations are generated using a Monte Carlo/Metropolis scheme combined with a singular value decomposition technique (SVD). The approach is shown to characterize the cooperative motions in good detail, at least 1 order of magnitude faster than atomic simulations. Trajectories are partitioned into modes, and the slowest ones are analyzed to elucidate the dominant mechanism of collective motions. Calculations performed for bacteriophage T4 lysozyme, a two-domain enzyme, demonstrate that the structural elements within each domain are subject to strongly coupled motions, whereas the motions of the two domains with respect to each other are strongly anticorrelated. This type of motion, evidenced by the synchronous fluctuations of the domain centroids by up to (4.0 Å in opposite directions, is comparable to the movements observed by recent spin-labeling experiments in solution. The potential of mean force governing these fluctuations is shown to be anharmonic. The -sheet region at the N-terminal domain and the helix E in the C-terminal domain are identified as regions important for mediating cooperative motions and, in particular, for the opening and closing of the active-site cleft between the domains. Residues Leu66-Phe67 in the central helix C stop the propagation of correlated motions between the domains. There is a correlation between the groups involved in highly cooperative motions revealed by simulations and the highly protected regions during unfolding measured by pulsed H/D exchange and 2-D NMR.A multitude of conformational substates, each of them constituting a local minimum on the highly structured energy landscape, exist in the neighborhood of the native state. These deviate only slightly from the X-ray structure. Fluctuations between such substates are commonly observed in molecular dynamics (MD) 1 simulations (1), while larger-scale conformational changes, including those that eventually lead to unfolding, are less commonly accessible to simulations. Atomic models used in such simulations necessitate the adoption of time steps of the order of femtoseconds, which do not permit attaining time scales longer than nanoseconds. A typical protein MD simulation samples only a limited portion of the overall conformational space, as evidenced by the projection of the trajectory onto the subspace spanned by the three dominant eigenvectors of the displacement covariance matrix (2). Another difficulty with MD simulations is that cross-correlations between the displacements of different atoms in a given protein cannot be precisely captured (2). Such limitations motivate the quest for simplified models and more efficient computational tools.Numerous recent studies of protein conformations and interactions have aimed at less detailed coarse-grained models. Some of these efforts have been directed toward clarifying the principles governing the structural preferences of protein...