Because there is only one allowed RANDOM statement with one SUBJECT option, many people think that NLMIXED is only for 2-level hierarchies. But if you understand variance-covariance matrices, one can force NLMIXED to fit multiple-hierarchy models to data. This is tricky, requiring a lot of coding. It is not the type of thing that can be simply written out here (at least for me). Example 15.4 in Littell et al. (2006; SAS for Mixed Modell, 2nd edition) deals with this explicitly. You can buy this for SAS Press or Amazon (etc.).