Then, as I said, if the interaction is strong enough, it should be sufficient to have them both in a bound state at the beginning of the simulation, without any additional bonds specified in the topology.
Of course, interactions with ions are never perfectly modeled in additive (non-polarizable) force fields so it’s an open question whether a naive approach will work for you; if it’s somehow experimentally known that the two should stay together and in the simulation they don’t, one can try out different sodium parameter sets, or simply add a restraint. This very much depends on the ultimate purpose of your simulation.