Physical infrastructures are commonly composed of interconnected and intra- and interdependent subsystems, which in their essence constitute systems-of-systems (S-o-S) with multiple functions, operations, and stakeholders. Their complexity is characterized by highly interconnected and interdependent physical, cyber, organizational, and economic subsystems through shared resources, decisions, and states. This paper explores potential sources of systemic risks in complex S-o-S by analyzing unique failure modes of such systems in a nonlinear dynamic multiobjective sequential decision-making process. We posit that certain preferences of decision-makers when choosing among multiple competing objectives may reduce the safety margin of a specific subsystem within the S-o-S to withstand unexpected external perturbations due to the interdependencies manifested by shared states among the subsystems. This can contribute to the failure of an entire S-o-S even though all decisions are made based on Pareto-optimality. By quantifying the level of subsystem interdependency caused by shared states, this paper develops a method to mitigate such systemic risks through decomposing and coordinating the interconnected subsystems of an S-o-S in a decentralized way. The implication of the proposed theory is demonstrated using an illustrative example of a highway bridge S-o-S.