I agree with Jorge. On the other hand, I suppose that such problem could be attenuated since my specific case deal only with protein structures (no small ligands, nor nucleic acids, etc.). Moreover, in my GROMACS' simulations I have also adopted the amber forcefield.
You can use CatDCD (either the standalone program or the VMD plugin, http://www.ks.uiuc.edu/Research/vmd/plugins/catdcd/ ) to manipulate trajectory files. It offers the option to convert your trajectory to and from various formats, among them AMBER and GROMACS.
As for the topology, as far as I know there is no option that will work without errors. However, since, as you say, your system consists of proteins alone, you could simply generate the topology anew, with AmberTools.