Can someone shed a light on B-factor refinement in PHENIX.refine.

I am refining the structure with 2.5 A resolution, with individual B-factors and automatically defined TLS groups. Despite isotropic refinement of B-factors I still see ANISOU entries for atoms.

I figured out that these ANISOU lines appear only after TLS refinement.

So the question is how these ANISOU values are calculated?

--