Massive Open Online Courses (MOOCs) are designed on the assumption that good students will help poor students thus offloading the individual support tasks from the instructor to the class. However prior research has shown that this is not always true. Students in MOOCs tend to form distinct sub-communities and their grades are closely correlated with those of their closest peers. That work, however, was only based on analyzing the final social network in a MOOC. In this paper, we study the evolution of these co-performing clusters over time. We explore a longitudinal approach to detect how students form their social connections on the discussion forum and we show that students form close coequal communities early in the course and maintain them over the duration of the course. [For the full proceedings, see ED593090.]