You should be able to use the normal purification methods as for any protein. FAD is usually covalently or tightly bound, so the enzyme should already have FAD in it and will be yellow in color. If the enzyme is not fully loaded with FAD, then you can add it at the beginning of the purification, which will help stabilize the protein.
Dear Vincent and Adam, Many thanks for your answer. The target protein concentration is very low, might be less than 20ug/ml and the solution (from plant) is almost colorless. Unfortunately, I only have several ml of the protein solution. 2DE showed proteins in the solution are not complex, with less than 20 proteins but the target protein is a minor component, less than 10% of the total protein. I do hope a FAD affinity column chromatography method to isolate the protein but Adam’s words make me worried that the method might not work if the protein already contains FAD. I am confused. Can you give some details about the FAD affinity column chromatography or which company might supply such column or any more suggestions?
If you know the size of your protein, you can do a size-exclusion column chromatography using a stationary phase whose pores closely match the size of your protein. Other proteins will quickly pass through the column while your protein will be slowed down considerably ("retained"). You can recover your protein simply by eluting further. HTH.
The substrate is L-gulono-1,4-lactone. I don't think affinity chromatography with an immobilized version of this substrate is feasible. I saw a remark in BRENDA that the FAD is covalently bound, so the protein probably already has FAD in it. Therefore, FAD affinity chromatography won't work. I would try a small-scale ion exchange column. If you have access to an FPLC with a mono-Q or mono-S column and run a shallow salt gradient, you will probably be able to substantially increase the purity. It appears to be a pretty large protein (oligomer of 50 kDa monomers), so the gel filtration chromatography approach suggested by Vicente would probably also work.