No, all the processing is done in real/scanner space. You can define ROIs on the high-res anatomical or whatever other image you want.
mrthreshold is the simplest, and if needed you can use mrcalc for more complex operations.
There’s different ways depending on what you want to do. One option is to map the streamlines to an track density image (TDI) using tckmap, then warp all of these images using whatever registration program you normally use. Another option is to warp the tractograms themselves to some common standard space, as discussed here. Yet another option is to use tcksample and tckresample – some discussion on that here. There’s probably other ways to combining the streamlines too, it really depends on precisely what your research question is…