Hello Emily,
This is a tricky question. The basic answer is that the exact tree you build will depend on the microbes included in the MSA, so it probably best to split up your samples, remove any OTUs that don’t appear in a cohort, then built one tree for each cohort.
However, the same is true of OTU clustering; the mixing cohorts will effect the exact OTUs that appear in each one, so your cohorts a little bit mixed-up no matter how you build the tree.
Short answer: either way would work OK… but 100% the best option 
would be to start from the very beginning and process each cohort separately.
I know this is not the question you asked, but I wanted to provide as much detail as possible. Let me know if you have any other questions.
Colin
P.S. Did someone give you an old version of Qiime? You should try the new version! 