this depends on the answer you need. I would rather use a dedicated storage methodology as multiple freeze and thaw cycles otherwise might change the answer (especially if you look for compositions of bacterial communities). RNA later is an option that does work and prevent these kind of bias.