I agree that the jumps in RMSD are nothing more than PBC effects.
Most SO4 found in PDB files comes from co-solvents or salts used for crystallization. Usually these are discarded for simulations in water because they are not biologically relevant.
In general, the solution to your error is found easily online:
Note that using trjconv is not an "either/or" situation - you almost always need a few steps to properly re-image the trajectory. This is all spelled out on http://manual.gromacs.org/current/user-guide/terminology.html#suggested-workflow