Science is increasingly produced in collaborative teams, but collaborative teams in science are self-assembled and fluid. Such characteristics call for a network approach to account for external activities responsible for team product but taking place beyond closed team boundaries in the open network. Given such characteristics of collaborative teams in science, we empirically test the interdependence between collaborative teams in the same network. Specifically, using fixed effects Poisson models and panel data of 1310 American scientists' life-time publication histories, we demonstrate knowledge spillovers from new collaborators to other teams not involving these new collaborators. Our findings have important implications for studying the organization of science.