Endosymbiont-conferred resistance to parasitoids is common in aphids, but comes at a cost to the host in the absence of parasitoids. In black bean aphids (Aphis fabae), costs in terms of reduced lifespan and lifetime reproduction were demonstrated by introducing eleven isolates of the protective symbiont Hamiltonella defensa into previously uninfected aphid clones. Transfection of H. defensa isolates into a common genetic background allows to compare the costs of different endosymbiont isolates unconfounded by host genetic variation, but has been suggested to overestimate the realised costs of the endosymbiont in natural populations, because transfection creates new and potentially maladapted host-symbiont combinations that would be eliminated by natural selection in the field. In this experiment, we show that removing H. defensa isolates from their natural host clones with antibiotics results in a fitness gain that is comparable to the fitness loss from their introduction into two new clones. This suggests that cost estimated from transfecting endosymbiont isolates into a shared host genotype does not lead to gross overestimates of their realised costs, at least not in the two recipient genotypes used here. By comparing our data with data reported in previous publications using the same lines, we show that symbiont-induced costs may fluctuate over time. Thus, costs estimated after extended culture in the laboratory may not always be representative of the costs at the time of collection in the field. Finally, we report the accidental observation that two isolates from a distinct haplotype of H. defensa could not be removed by cefotaxime treatment, while all isolates from two other haplotypes were readily eliminated, which is suggestive of variation in susceptibility to this antibiotic in H. defensa.