The last 20 years have witnessed an explosion of interest in reproductive parasites: maternally inherited microbial infections of arthropods that manipulate the reproduction of their host species towards the production or survival of infected female hosts. This interest has come largely from the study of one bacterium, Wolbachia. This bacterium was found to be associated with a variety of different manipulations: forcing asexuality on its host, feminising hosts, inducing incompatibility between infected males and uninfected (or differently infected) females, killing males and, most recently, being required for oogenesis [1–4]. Each of these manipulations has been fascinating in its own right, and is also of ecological and evolutionary importance to the particular host species that is infected, potentially inducing reproductive isolation, driving changes in sexuality, or alteration in reproductive ecology [5–7]. They also engender strong selection against their action, driving some of the fastest natural selection observed in the field [8]. Wolbachia was also found to be common [9, 10]. Thus, the effects described above in particular species were likely to be widely important.

The incidence of Wolbachia [9, 10], and more recently Cardinium [11, 12], have received much attention in studies of arthropod biology. However, the incidence of the broader spectrum of inherited bacteria associated with reproductive parasitism is not known, apart from focussed studies of spiders [22, 24] and Drosophila [25]. In addition, a generally low sampling intensity per species for which screens have taken place in previous surveys may have resulted in the underscoring of Wolbachia and Cardinium infections, as some reproductive parasite infections are rare within a host species [22, 26, 27]. The global diversity of reproductive parasites in their natural contexts and the nature of the interactions with their hosts thus remain to be characterised in depth. Here we test a range of arthropods for the presence of the seven known reproductive parasite clades within the eubacteria, estimating incidence, prevalence and geographical variation between natural populations of arthropods. We sample each species more intensively than in past surveys; that is, primarily more than 10 individuals per species. Lastly, we estimate the relatedness of bacterial strains from our collection to strains recorded previously. We conclude that two other inherited bacteria, members of the clade Arsenophonus and members of the Spiroplasma ixodetis clade, are important associates of insects and demand further study.

We sampled a medium number of male and female individuals (median 20 total) of 136 arthropods species for Wolbachia, a reproductive parasite well known to be common in various arthropod taxa [9, 10], for Cardinium, a reproductive parasite known to exist widely but in fewer species [11, 12, 22], and for five other unrelated reproductive parasites whose incidence and prevalence were largely unknown. The observed incidence of infection for arthropods was 32.4% for inherited bacteria putatively acting as reproductive parasites.

Our results reveal that Wolbachia is, as expected, the most common reproductive parasite clade associated with arthropods, being recorded in 22.8% of species in our sample. Three other clades, Cardinium, S. ixodetis and Arsenophonus bacteria were present widely, with each occurring in 4% to 7% of the arthropod species. Whilst there has been some work on the former of these bacteria, S. ixodetis and Arsenophonus clearly represent 'understudied groups' that merit careful investigation. We would also note that the frequency of these four bacteria means that around one fifth of infections co-occurred with other species of inherited bacteria (8 of 44 infected species carry two different bacteria), an estimate of course made conservative by our own restrictive sampling for 'known' symbiont groups. This observation reinforces the call of Weeks et al. [37] to adequately sample the inherited flora of a species before interpreting data in terms of particular inherited bacteria.

There were three clades of bacteria that were either found in just one species or not found: Rickettsia, S. poulsonii and flavobacteria relatives. With respect to flavobacteria, the PCR assay detected allied clades of beneficial bacteria and was thus broad spectrum. We can thus be clear that this bacterial clade does not commonly reach mid or high prevalence outside ladybird beetles, where infection was initially established [38]. It is possible that this bacterial group is phylogenetically restricted to being 'ladybird MK'. The low incidence of Rickettsia was more surprising. These have been conjectured as being common, understudied bacterial symbionts of arthropods [39], but we can be confident that spotted fever group Rickettsia does not commonly exist at medium to high prevalence. This comment is made with the caveat that our screen for Rickettsia was relatively narrow, designed to find bacteria allied to a clade of ladybird MK, and would exclude some Rickettsia. Finally, there is S. poulsonii, otherwise known as group I spiroplasmas [40]. Other spiroplasmas were detected in the assay, indicating the PCR assay had a relatively broad catch. S. poulsonii itself is a MK associate of Drosophila species [41], with related strains in leafhoppers. Clearly, whilst spiroplasmas may be common, this particular clade either has low incidence, or always has low prevalence (as is the case for S. poulsonii in Drosophila, for example [42]).

As there is an intuitive link between the frequency of symbiotic infection and the overall importance of symbionts in arthropod biology, some comment on the true incidence level of these bacteria is appropriate. Given that we have tested against the presence of false positives in the PCR assay by obtaining product sequence in all cases, we would emphasise that our estimate underscores true incidence. Two main sources of underestimation of incidence are inherent in our survey. First, there is the possibility of false negatives in the PCR assay. These would be individuals (and, hence, species) in our sample that are in fact infected, but where infection has not been detected. Second, there are species where the PCR assay was accurate and our samples were truly uninfected, but other members of the species, not sampled and tested, are in fact infected. We would argue that both of these factors lead to a considerable underestimation of the frequency of infection.

False negatives with respect to undetected presence in infected samples are an issue with PCR assays, their sensitivity to titre and their ability to detect diverse infections. For Wolbachia, for instance, our use of a PCR assay based on 16S rDNA is likely to include most of the Wolbachia diversity. For Rickettsia, we would note that whereas we found no Rickettsia infection among spiders, Goodacre et al. [24] used PCR primers with a broader 'catch' and reported 19.7% of species to be infected by Rickettsia strains related only distantly to Rickettsia MK strains. These are likely to be inherited bacteria, and possibly reproductive parasites, but they would not be detected in our screen. Furthermore, we probably also underestimated the number of bacterial strains present in species we found to be infected. Given that our survey is in most cases based on PCR amplification of fragments of the slow evolving 16S rDNA sequence, we had limited power to detect multiple infections of closely related bacterial strains. For instance, the 16S rDNA sequences of Wolbachia infecting the mosquito Culex pipiens are well known to be strictly identical, suggesting that only one Wolbachia strain could occur, but sequencing of faster evolving genes has demonstrated the occurrence of more than 60 Wolbachia strains in this host species [43–45].

The second issue that causes an underestimation of the frequency of these agents is failing to sample an infected individual in an infected species. There are two aspects to this, the first being cases where infections are rare within a population. For example, MK infections tend to exist at low prevalence and examples of infection at between 1% and 10% frequency in females represent about one-half of all known infections [46]. Within our survey, we sampled two species in which just 1 of 20 individuals was infected, which clearly indicates the potential for false negatives arising from low prevalence infection. The second cause of failure to sample an infected individual arises from infection presence varying from population to population. We sampled seven species from several localities and revealed three examples of the presence or absence variation between populations within Western Europe. Such geographical variation is not uncommon [47], and is likely to be much greater when the scale of sampling is increased beyond Western Europe (which because of the recent ice age represents one recently recolonised region in ecological terms). Insufficient geographical sampling will lead to serious underestimation of infection incidence.

Overall, therefore, our point estimate of 32.4% of species infected with bacteria allied to reproductive parasites is likely to seriously underestimate the true figure. Our data improve on past surveys by increasing the intensity of sampling within species. For instance, for the case of Wolbachia, resampling our data (taking one individual from each species that is infected with a probability of infection given by the overall prevalence found) indicates that the move from single individual sampling to our modest multiple individual sampling reveals around one-third more cases of infection: the median Wolbachia incidence on sampling one individual per species is 17.6%, rather than the observed 22.8%. However, the sample size within each species is still limited and, most importantly, the restriction in geographical sampling may produce a very serious underestimation of symbiont presence or absence.

We obtained some insight into the degree to which host taxa vary in the frequency of their interaction with inherited bacteria. The depth of conclusion is very limited by virtue of the intensity of sampling within particular clades in a broad survey. Nevertheless, spiders (the Araneae) harboured a higher richness of inherited bacteria than others and represented diversity hotspots for these bacteria with 61.5% of species infected, contrasting to the relatively low incidence observed within Diptera and Coleoptera. The Araneae hotspot was present for Wolbachia, Cardinium and S. ixodetis, although not for Arsenophonus. Why some arthropod taxa are hotspots for inherited bacteria remains one of the most challenging questions with regard to the ecology of reproductive parasites. Certain taxa are clearly more susceptible to acquiring bacteria via horizontal transfer and/or to stably maintaining infection. Given that cocladogenesis is rare, it is the establishment of new infections that appears important in dictating incidence [48]. Intimate contact with other species is likely to predispose to transfer, be this parasitism, predation, physical damage or becoming prey [49–53]. Indeed, species which include other living arthropods in their diet through predation or parasitism are more frequently infected (43.8%) than other species (26.1%; P = 0.05, Fisher's exact test). As a result, predation and/or parasitism could be an efficient transfer mechanism to elucidate the origin of hotspots, as illustrated by Araneae, which is exceptional because all species of this group depend completely on predation of other invertebrates which are largely polyphagous [54].

An alternative hypothesis to account for heterogeneity of endosymbiont incidence among host clades is based on differences in host phylogenies [55], but this does not seem to be a likely explanation for the Araneae.

We hope that this paper has established that Wolbachia and Cardinium are not alone: S. ixodetis and Arsenophonus relatives also exist in appreciable numbers of arthropod species. Perhaps the most important consideration for the future is not the incidence of infection, but phenotype. Whilst there have been detailed studies of Wolbachia and Cardinium in a variety of arthropods, there has been little work on the other pair of 'common' symbionts, S. ixodetis relatives and Arsenophonus. S. ixodetis relatives are known to be MKs in ladybirds [56, 57] and butterflies [58, 59], but have also been isolated from ticks [60] and aphids [61], where they do not kill males. A. nasoniae was established as the MK of the wasp N. vitripennis [62] but other strains of Arsenophonus isolated from a divergent range of arthropods show no evidence of sex-ratio distortion activity [21, 23]. The S. ixodetis and Arsenophonus strains isolated in our study do not show evidence of MK behaviour as they are present in both males and females. It is very likely that these symbionts represent important understudied components of arthropod biology. We would advocate an open approach to their biology; they may cause cytoplasmic incompatibility, but may also be secondary symbionts that could be conditionally beneficial, and their presence in bacterial clades containing reproductive parasites would be of great interest. It is also possible that horizontal (infectious) transmission plays a larger role in the population biology of these infections than is usually perceived. Sexual transmission of aphid secondary symbionts [63] and transmission on coparasitisation for Arsenophonus [31] make this avenue worthy of investigation. If biologists are to understand the role that inherited bacteria play in arthropod evolution, a thorough examination with respect to documenting their effects is now required.