Dynamically adaptive software systems are inherently distributed. These systems enable the dynamic adaptation of systems' behavior according to information gathered from their environment, through sensors. As the system behavior is associated with different sensing devices, the complete system is not contained in a single device, but the complete system behavior can come from cooperation and communication between multiple nodes, as the environment evolves. However, existing context-oriented programming languages underlying such systems work under a closed world assumption, in which the system is completely contained and managed in a single node. This assumption hinders the applicability of dynamic adaptations in environments as those proposed in cyber physical systems. To develop the full potential of dynamically adaptive systems, we propose a run-time execution model, distributed context Petri nets, to manage the interactions of behavioral adaptations between remote nodes. Each node consists of a context Petri net extended with the capability to remotely interact with context Petri nets in other nodes. To validate the ability of our model to manage behavioral adaptations in distributed settings, we demonstrate the semantics of the different context dependency relations in face of different situations yielding inconsistencies.