This paper explores the affordances of virtual reality (VR) simulations for facilitated model-based reasoning. Thirty-four undergraduate students engaged with simulated scientific models in head-mounted displays and their facilitator in a co-located mixed-presence configuration. We coded the facilitator--participant interactions using the Structure-Behavior-Function framework to examine how the simulation and learning activity supported meaning-making and model-based reasoning. We inductively analyzed the exchanges for key themes relating to how the learner--facilitator pairs interrogated the model. Findings showed that the model's function was most frequently examined, followed by structure, then behavior. Productive engagement patterns included spontaneously observing the model's structural components, which at times led to more in-depth student-driven inquiries. Learners attended to components' behavior in the form of immediate situated feedback while demonstrating their conceptual understanding. A "compare and contrast" pattern through multiple simulated states allowed dyads to elaborate on the model's function. However, mental workload and prior knowledge were mediating factors. Further, pairs leveraged the simulation environment to access conceptual content. Lastly, predictions about the model's system dynamics led to shifts between different levels of reasoning through structure, behavior, and function. We discuss affordances that made the VR simulation an effective mediating representation for a facilitator and learner to interrogate complex scientific models together. This work extends the possibility space for embodied learning in VR, addressing a broader range of difficult-to-teach abstract concepts.