The activated charcoal trick suggested by Paul should work.  It does a pretty good job with GroEL.  However, in a few cases you need to first saturate the solution with something GTPgS to displace it (assuming that this binds less well) then do the dialysis with charcoal to get rid of all of the dinucleotides.  It does not need to be GTPgS but any nucleotide that binds with lower affinity to the same site. 

Thanks, Paul and John! I will give a test.

I saw a protocol from Pierce GTPase enrichment kit with ActivX probes that suggests to pre-treat the GTPase with EDTA, with which the GDP/GTP can be removed from the GTPase active sites. Is there any people used it before and how does it work?

That is always a good idea.  In most cases, the metal ion plays a key role in the binding.  However, if it is very tight, the off rate/on rate of the GTD/Mg complex may  not be favorable even with EDTA.  Certainly, EDTA and activated charcoal (in the dialysis buffer is the most standard way these will trap the molecules that are released.  The addition of a competitor ligand that binds less tightly (but in excess) and is easier to remove just completes the cycle by competing with the GDP, which will be very dilute stoichometric to sub-stoichiometric with the protein.

Thanks, John. This is a very helpful information.

Hi Xiaobo, sorry, this is a late answer to your question, but you could incubate your sample with alkaline phosphatase, which will hydrolyze all phosphoesters.