Discussion
Understanding the historical eco-evolutionary processes that determine the structure of biodiversity are primary goals in ecology and evolutionary biology. Therefore, the current understanding that a large proportion of arthropod diversity (Zug & Hammerstein 2015) harbours a non-systematically distributed agent of speciation constitutes a major academic challenge with respect to identifying predictable processes that generate biodiversity. Here we introduce the ‘contact contingency’ predictive model for Wolbachia strain distributions based on phylogenetic relationships, ecological contact, and host abilities to determine their own infection status, that shows remarkable accuracy on an empirical fig wasp dataset sampled from elevational transects in Papua New Guinea. We also present the ‘oviposition trade-off’ model to account for the influence of post-zygotic fitness losses imposed by Wolbachia which could potentially invalidate our proposal. We show that post-zygotic fecundity losses may be tolerated under multiple-mating scenarios when considering the ecological characteristics of the fig syconia microcosm. We hope our work stimulates further debate around these phenomena in order to either accept and refine our proposed mechanism, or to prompt a dismissal accompanied by a more parsimonious theoretical framework that accounts for distributions of Wolbachia endosymbionts at the host taxonomic scales we are querying. Our models may be particularly suited to the unusual ecology of fig wasps, but their underlying dynamics may contribute to differing degrees among other ecological systems (Box 1).
An inspection of Wolbachia strain distributions among our fig wasp phylogenies reveals suggests that systemic processes may be in operation. Wasp populations at different elevations display different infection patterns. For example, 77% of Ceratosolen armipes are infected but only one out of 34 individuals of its clade that pollinates Ficus umbrae carries Wolbachia . Further, there are Wolbachia strains that are restricted to lowland and highland elevations, even among the same host (e.g., F. arfakensis ). As such, parapatric populations of pollinating wasps often have contrasting infection status that may represent RI, mediated by either uni- or bi-directional CI. Our data repudiate vertical transmission co-divergence hypotheses. However, we consider it important to note that this also repudiates a horizontal exchange hypothesis, as Wolbachia strains invariably do not occupy all or multiple wasp clades infecting the same host species/complex, despite potential for ecological contact. These results are consistent with the idea that Wolbachia only infects certain insect groups under particular host-adaptive circumstances.
Systemically generated patterns in host Wolbachia statuses could arise if infection is largely controlled by endosymbiont abilities to manipulate their own infection status as is often posited (Werrenet al. 2008); for example, because particular clades of host lineages’ immune responses are unable to repel the infection. However, our observed patterns do not obviously support manipulation by the endosymbiont given the correlation with abiotic drivers such as altitude, and no known a priori reason to assume that host immunity is principally determined by abiotic circumstances. Relative fitness costs in host mating potential, fecundity, or inappropriate sex ratio distortion associated with Wolbachia invasion may also generate the conclusion that Wolbachia is the chief architect of its own success. However, it is known that Wolbachia offers host fitness benefits that may have ecological contingencies (Zug & Hammerstein 2015; Correa & Ballard 2016). We may consider a host insect species that exhibits a broad phenotypic range, say for ovipositor length that has distinct optima at different altitudinal ranges according to host plant morphological divergence. Any mechanism that prevented reproductive events between extreme phenotypes (yielding intermediate morphs) at inappropriate altitudes would be favoured providing the benefits (increased local fitness) outweighed any deleterious costs.
Our wolPredictor simulation of the ‘contact contingency’ hypothesis attempts to model such a scenario that is founded on: (i) host abilities to modulate/purge Wolbachia when evolutionarily apposite; and (ii) Wolbachia causing a circumstantial host fitness advantage. Our hypothesis is that recently evolved sister-species diversifying within the same host community are at a selective advantage when they harbour alternate strains of Wolbachia. Divergent strains facilitate the initial stages of speciation; this is because we expect that closely related, co-diverging Wolbachia strains would not confer RI between hosts. Given that Wolbachia infection does appear to contain some fitness cost component (Zug & Hammerstein 2015), we predict that these patterns should not be evident among diverging lineages that are not in regular ecological contact, thus leaving Wolbachia subject to host immunosuppression due to the circumstantial absence of any RI-enforcing benefit. Finally, as Wolbachia typically drops out of host lineages after approximately 7 million years (Bailly-Bechet et al. 2017), and because alternative mechanisms of RI that require cytogenetic or morphological modification may take longer to evolve (Bordenstein et al. 2001), we predict that closely-related species that have not recently diverged should also be purged of Wolbachia infection. This would reflect a hypothesis that observed lineage dropout (Bailly-Bechet et al. 2017) results from temporal changes in the adaptive benefits of Wolbachia , which may subsequently become redundant (the adaptive decay hypothesis).
Simulation results among our fig wasp data are particularly impressive (ca. 88% accuracy for positive strains) when the species delimitation algorithm matches the empirical understanding of ecological diversity patterns (Souto-Vilarós et al. 2019). In particular, when multiple congeners within fig host communities are algorithmically predicted among F. arfakensis , F. trichocerasa subsp. pleioclada and F. wassa , wolPredictor ascribes multiple strains within these communities that reflect the data. Prediction accuracy when also considering negative (uninfected) Wolbachia strains is less precise. This may be due to variation in the empirical data where negative and positive strains co-occur within clades or the tendency for non-infected individuals to appear among lineages that comprise multiple/incipient species within a single host fig (e.g., F. wassa - infection rate = ca. 10%), where wolPredictor ascribes positive Wolbachia associations (such patterns may actually represent unidirectional RI whilst we model the bidirectional mechanism).
To model the ‘adaptive decay’ hypothesis, our wolPurger function operates to remove Wolbachia from lineages after extended evolutionary timescales. However, as it uniformly removes Wolbachia across all samples exceeding the distance threshold it constitutes a crude method that may not always be appropriately applied across lineages. Inconsistencies in strain associations could result from other drivers we have not accounted for, and alternative mechanisms of RI (hypothetically rendering Wolbachia redundant) may appear at different rates across lineages possibly due to serendipitous genomic architecture or unconsidered ecological contingencies. For example, among fig wasps, syconia access is partially controlled by relative syconia-wasp size (Bronstein 1987), which mechanically prevents hybridisation opportunity among some species.
Furthermore, under a simple model of panmixis and infinite population size, CI is predicted to sweep to fixation, contrary to the population level polymorphism observed in our data. Theoretical expectations on how CI spreads through populations are largely determined by population structure (Engelstädter & Telschow 2009). However, this depends on perfect transmission and infection rates may still eventually decay even if fixation is achieved. Fig wasps are both haplodiploid and inbred. Haplodiploidy can facilitate the survival of infected haploid males (Breeuwer & Werren 1990), which like inbreeding can result in both a higher invasion threshold and a reduced stable equilibrium frequency (Engelstädter & Hurst 2006). These considerations deserve further attention, but they may, at least in part, explain infection frequencies below 100% as observed in our study system.
Overall, our wolPredictor simulation of the ‘contact contingency’ hypothesis is a rule-based algorithm that manages to capture much of the embedded structure in a dataset that presents a superficially stochastic appearance. Thus, it suggests that some environmentally contingent symbiotic benefits (Correa & Ballard 2016) may systematically sum to yield predictable Wolbachia distributions. Our methods cannot test for precise mechanisms determining Wolbachia distributions across our study systems, and the algorithm underpinning wolPredictor may inadvertently represent some other set of real-world contingencies.
Moreover, we acknowledge that our model may be considered problematic since Wolbachia -mediated CI is a post-zygotic mechanism that elicits an immediate fitness cost in host fecundity. However, it is feasible that the unique life-histories and ecological conditions of fig wasps means they may be tolerant of CI: oviposition sites are at especially high premium (Dunn et al. 2015), fig wasps are known to produce surplus eggs (Dunn et al. 2011), and co-evolved species are renowned for precise tolerances in interacting traits that may render hybridisation particularly costly. This critique prompts our ‘oviposition trade-off’ hypothesis and second simulation model. Here we investigated the impact of CI in multiple-mating scenarios when considering the oviposition limiting constraints of fig syconia. We show that inclusive fitness of multiple mated females can be higher despite fecundity losses providing that egg load reduction (or selective ovipositioning) facilitates oviposition into higher-quality fig ovules that are less vulnerable to parasitoid attack (Dunn et al. 2008). Notably, the parameter values (i.e., conspecific mating levels and relative fitnesses) that yielded CI favouring outcomes are realistically achievable among natural fig wasp populations. The results imply that bi-directional CI may adaptively evolve in fig wasps without accompanying mechanisms.
The interaction of CI-inducing Wolbachia on multiple-mating in fig wasps has not been studied but we found a single study in Drosphila demonstrating Wolbachia associated reductions in sperm competition abilities (Champion de Crespigny & Wedell 2006). Given the unusual reproductive manipulations of haplodiploid Hymenoptera such as selective fertilisation, adjustment of sex-ratios, and control over oviposition order according to ploidy, and given the dearth of research of CI under multiple-mating conditions, it is entirely possible that such dynamics are at play among fig wasps at least. Thus, future work examining whether incompatible matings result in differential pre-oviposition embryo mortality, and whether selective oviposition of conspecific versus heterospecific eggs occurs would be of great value. It may be that RI-inducing Wolbachia constitutes a mutualist symbiont preventing the inefficient use of highly valuable oviposition sites with intermediate hybrid form, lower fitness offspring.
We further note that our models diverge from some conventionally held opinions regarding Wolbachia and host manipulation/purging (e.g., Werren et al. 2008) capabilities and a general view thatWolbachia is in conflict with its hosts in many respects (e.g., Charlat et al. 2007). While it has been shown that Wolbachia may employ microRNAs to alter host gene expression (Hussain et al. 2011), it has not been investigated whether any Red Queen-type arms race dynamics facilitate host resistance – as our model implies. We argue that despite potential pitfalls it is difficult to propose an alternative systemic model or explain observed structural patterns as random, within our or other published datasets. For example, an overview of malvantheran fig wasps shows that communities featuring singleton congeners invariably display negative Wolbachia associations while the reverse is true for multi-congener communities (Haine & Cook 2005). Finally, we also contend that the most parsimonious interpretation of reported empirical patterns (i.e., sister-species hosting paraphyletic infections) leads to a view that the interests of Wolbachia may be aligned with their hosts under conditions of ecological speciation. Wolbachia is known to impart some host benefits (Zug & Hammerstein 2015) and we are yet to fully understand the nuances, trade-offs and ecological contingencies that determine whether it is rendered circumstantially advantageous.
Our proposed models are particularly suited to testing in fig wasps due to the high degree of easily-collatable host-specificity relationships exhibited when wasps co-occur within the fig microcosm. Such processes may also be subtly at play among other taxa where ecological contact is not easily assessed – such data have not been previously incorporated into a predictive phylogenetic model. Variables such as ecological contact may act along a continuum among species rather than the easily assessable binary states observed in fig wasps. Indeed, a suite of variables that comes into play in fig wasp communities, e.g., degrees of inbreeding or variation in oviposition site quality, may differentially contribute to different systems and may interact with other factors regarding Wolbachia cost-benefit trade-offs that determine infection status. Thus, fig wasps may offer an ideal window into understanding not only the determinant ecological contingencies at play but also offer insight into the nature of the types of variables that may be significant or even overridden by other factors (for example, we note that our ‘contact contingency’ hypothesis is pertinent to all fig wasps whereas the ‘oviposition trade-off’ hypothesis is relevant to pollinators and non-pollinators that enter the fig). Even with these system-specific limitations in mind, the vital issue then may be that for most of global diversity we simply do not have the kind of detailed ecological information that can reliably inform us about the processes underpinning community assembly. Our theoretical reasoning may therefore be extremely generalizable and we urge that model communities representing different ecosystem states are identified and investigated (Box 1) under a proposed methodological framework (Box 2).
Our models imply a genomic mechanism among insects making Wolbachia tolerance/purging a highly labile and evolutionarily unstable trait that would concomitantly render RI events a relatively trivial occurrence. Other potentially similar mechanisms have been documented in other taxa (see also Box 1). In the hymenopteran Nasonia, Wolbachia -induced CI has been shown to precede other incompatibility mechanisms (Bordenstein et al. 2001). In Drosophila , readily occurring genomic inversions (Noor & Bennett 2009) often serve to maintain RI among closely-related species/subspecies in sympatry (although CI is also found in fruit-flies; Merçot & Charlat 2004), while such phenomena are less prominent in allopatry (Noor et al. 2001). And among certain gastropods, a single gene mutation coding for shell chirality can cause RI between sister-species (Hoso et al. 2010). Under the islands of speciation paradigm (Noor & Bennett 2009), regions of genomic divergence incrementally build until differences between lineages yield distinct evolutionary trajectories. A conundrum exists in explaining how this might generate entirely separate lineages under ecologically driven divergence in sympatry and in the face of gene flow. Future work may reveal that it is highly labile, binary decision-making reproductive switches such as CI or chromosomal inversions that resolve this puzzle by providing tipping point mechanisms that promote evolutionary schisms when net selective pressures favour speciation.
The stark disparity between recent advances in genomic data accrual relative to the laborious efforts required to record phenotypic/ecological data is well acknowledged – our work highlights the need to ameliorate this. Biodiversity cannot be simplistically evaluated as a metric indicating number of species nor considered solely as the outcome of interactions between closely-related species within the same trophic level. There is a growing consensus that we need to consider interactions both within and between all trophic levels whilst also identifying what constitutes significant versus trivial dynamics (i.e., intensity of interaction; Segar et al. 2020), or, more generally, ecological contingency. Thus, we face a massive challenge to document community dynamics of not only obviously tractable relationships (e.g., competition between focal species) but also of both mutualistic (e.g., pollination, seed dispersal, nutrient sequestration, mycorrhizal) and antagonistic factors (e.g., parasitic, disease) that may sometimes be bacterial, fungal or viral in origin. Failure to account for such agents may mean we never fully disentangle the myriad determinants of ecosystem dynamics nor quantify the relative contributions of stochastic (viz . neutral; Hubbel 2001) processes.
CONCLUSIONS Our results indicate that Wolbachia distributions are predictably structured among an arthropod dataset based on a predictive model invoking adaptive responses in host fig wasps. A parsimonious interpretation of these findings suggests that ecologically contingent co-evolutionary benefits of Wolbachia induced CI, particularly with respect to opportunity for lineage diversification, may systematically sum to yield predictable distributions despite initial appearances that the endosymbiont is stochastically distributed at some taxonomic resolutions. In particular, our data suggests that future work assessing biodiversity patterns among arthropods should incorporate Wolbachia infection data (alongside other microorganisms) as an added modelling dimension in order to account for a potentially confounding variable. Our aim is to stimulate debate and subsequent research in unravelling a rather puzzling phenomenon within arthropod biodiversity.
Acknowledgments : We thank villagers from all collecting sites for both providing local assistants and offering us accommodation during our stay along the transect. We also would like to think all staff of the New Guinea Binatang Research Centre in Papua New Guinea and the Papua New Guinea Department of Environment and Conservation for help granting export permits. We thank Sylvain Charlat for constructive criticism and for the improvement of this manuscript. We acknowledge funding from the Grant Agency of the Czech Republic (grant number 15-24571S). STS acknowledges departmental support from Harper Adams University.