This paper investigates a distributed and adaptive approach to manage spectrum usage in dynamic spectrum access networks. While previous works focus on centralized provisioning, we propose a distributed low complexity approach where nodes selforganize into coordination groups and adapt their spectrum assignment to approximate the global optimal assignment. We design a distributed coordination protocol to regulate the coordination format and achieve fast system convergence. The proposed approach achieves similar performance in spectrum assignment compared to that of the conventional centralized approaches, but significantly reduces the number of computations and message exchanges required to adapt to topology changes. As a case study, we investigate the detailed coordination strategy to improve proportional fairness in spectrum assignment, and derive a theoretical lower bound on the minimum spectrum/througput each node can get from coordination. Such bound can be utilized to guide the ...