The dynamics of knowledge transfer is an important topic for engineering managers. In this paper, we study knowledge boundaries - barriers to knowledge transfer - in groups of experts, using topic modeling, a natural language processing technique, applied to transcript data from the U.S. Food and Drug Administration's Circulatory Systems Advisory Panel. As predicted by prior theory, we find that knowledge boundaries emerge as the group faces increasingly challenging problems. Beyond this theory, we find that knowledge boundaries cease to structure communications between communities of practice when the group's expert ability is insufficient to solve its task, such as in the presence of high novelty. We conjecture that the amount of expert knowledge that the group can collectively bring to bear is a determining factor in boundary formation. This implies that some of the factors underlying knowledge boundary formation may aid - rather than hinder - knowledge aggregation. We briefly explore this conjecture using qualitative exploration of several relevant meetings. Finally, we discuss implications of these results for organizations attempting to leverage their expertise given the state of their collective knowledge.