Fog computing adds decentralized computing, storage, and networking capabilities with dedicated nodes as an intermediate layer between cloud data centers and edge devices to solve latency, bandwidth, and resilience issues. However, introducing a fog layer imposes new system design challenges. Fog systems not only exhibit a multitude of key system characteristics (e.g., security, resilience, interoperability) but are also beset with various interdependencies among their key characteristics that require developers' attention. Such interdependencies can either be trade-offs with improving the fog system on one characteristic impairing it on another, or synergies with improving the system on one characteristic also improving it on another. As system developers face a multifaceted and complex set of potential system design measures, it is challenging for them to oversee all potentially resulting interdependencies, mitigate trade-offs, and foster synergies. Until now, existing literature on fog system architecture has only analyzed such interdependencies in isolation for specific characteristics, thereby limiting the applicability and generalizability of their proposed system designs if other than the considered characteristics are critical. We aim to fill this gap by conducting a literature review to (1) synthesize the most relevant characteristics of fog systems and design measures to achieve them, and (2) derive interdependences among all key characteristics. From reviewing 147 articles on fog system architectures, we reveal 11 key characteristics and 39 interdependencies. We supplement the key characteristics with a description, reason for their relevance, and related design measures derived from literature to deepen the understanding of a fog system´s potential and clarify semantic ambiguities. For the interdependencies, we explain and differentiate each one as positive (synergies) or negative (trade-offs), guiding practitioners and researchers in future design choices to avoid pitfalls and unleash the full potential of fog computing.