Coarse-grain (CG) techniques allow considerable extension of the accessible size and time scales in simulations of biological systems. Although many CG representations are available for the most common biomacromolecules, very few have been reported for nucleic acids. Here, we present a CG model for molecular dynamics simulations of DNA on the multi-microsecond time scale. Our model maps the complexity of each nucleotide onto six effective superatoms keeping the "chemical sense" of specific Watson-Crick recognition. Molecular interactions are evaluated using a classical Hamiltonian with explicit electrostatics calculated under the framework of the generalized Born approach. This CG representation is able to accurately reproduce experimental structures, breathing dynamics, and conformational transitions from the A to the B form in double helical fragments. The model achieves a good qualitative reproduction of temperature-driven melting and its dependence on size, ionic strength, and sequence specificity. Reconstruction of atomistic models from CG trajectories give remarkable agreement with structural, dynamic, and energetic features obtained from fully atomistic simulation, opening the possibility to acquire nearly atomic detail data from CG trajectories.