The same tools that they use to build and maintain large complex features: version control, engineering process artifacts (like specs and architecture overviews), and bugs/task/issue tracking.
So let's say Alice does a STRIDE analysis and finds a set of spoofing issues. As part of doing that, she adds some trust boundaries to an architecture picture, and updates the version in the wiki.
She documents the set of threats and checks it in as a .txt file next to the code for the component. Some apply to her component and become issues to investigate, some apply to Bob's component, and so she opens bugs for him, and some are unclear, so she shoots off an email which leads to a discussion meeting.
(This is a pretty broad question, so if my answer isn't helpful, let me re-frame your question a little: What tools do your teams use to collaborate across their large projects, and in what ways do those not work for threat models?)