By default, all systems are modeled in the gas phase. This means that any phenomenon that depends on solvent effects, such as the stabilization of ions on the surface of a protein, will not be modeled. When modeling proteins, use solvation (add keyword EPS=78.4). It lowers the geometric error by about 15%. The jobs will take a longer time to run, but the model is more realistic, and it's normally worth the extra CPU time.
The exception is when the positions of the hydrogen atoms are being optimized and everything else is frozen. This operation does not involve the non-hydrogen atoms moving, so not using solvation will not affect the results significantly. Of course, in order to allow heats of formation to be compared, for example in testing for salt bridges, the set of keywords used should be the same, in particular if EPS=78.4 is used in one job it should be used in the other.