Although the diversity of bacterial endosymbionts in arthropods is well documented, whether and how such diversity is maintained remains an open question. We investigated the temporal changes occurring in the prevalence and composition of endosymbionts after transferring natural populations of Tetranychus spider mites from the field to the laboratory. These populations, belonging to three different Tetranychus species (T. urticae, T. ludeni and T. evansi) carried variable infection frequencies of Wolbachia, Cardinium, and Rickettsia. We report a rapid change of the infection status of these populations after only 6 months of laboratory rearing, with an apparent loss of Rickettsia and Cardinium, while Wolbachia apparently either reached fixation or was lost. We show that Wolbachia had variable effects on host longevity and fecundity, and induced variable levels of cytoplasmic incompatibility (CI) in each fully infected population, despite no sequence divergence in the markers used and full CI rescue between all populations. This suggests that such effects are largely dependent upon the host genotype. Subsequently, we used these data to parameterize a theoretical model for the invasion of CI-inducing symbionts in haplodiploids, which shows that symbiont effects are sufficient to explain their dynamics in the laboratory. This further suggests that symbiont diversity and prevalence in the field are likely maintained by environmental heterogeneity, which is reduced in the laboratory. Overall, this study highlights the lability of endosymbiont infections and draws attention to the limitations of laboratory studies to understand host–symbiont interactions in natural populations.