Wolbachia screening of field collected samples
From 253 screened wasps, 41.1% (104 individuals) displayed Wolbachia infection. We found no evidence for multiple infections as no chromatograms contained multiple peaks. Individual wsp and MLST phylogenies confirm the monophyly of identified strains (Table S1; Fig. S1); the wsp tree contained five major clades while the MLST tree contained six. wsp clade assignment mostly matched with MLST markers, with the exception of the MLST clades associated with F. trichocerasa subsp. pleioclada and F. microdyctia , which were contained within the same wspclade (Table S1; Fig. S1). We therefore followed wsp clade designation for all wasps apart from wsp clade six which was split into two (wspC6_1 and wspC6_2) giving six identified Wolbachia clades.
While 77% of all Ceratosolen armipes (the pollinator of lowlandF. itoana ) were infected with Wolbachia, only one (out of 34) of the mid-elevation C. sp (ex. mid-elevationFicus umbrae ) was infected. Similarly, ca. 63% of C. “kaironkensis” (ex. highland F. microdyctia ) were infected. These wasps and their fig hosts form monophyletic clades with species replacing each other in parapatry (Fig. 3), mid-elevation F. umbrae (largely Wolbachia free) and highland F. microdyctia being sister species and lowland F. itoana the outgroup (Souto-Vilaros et al., 2018). Alternative populations of the four Ceratosolen pollinator species associated with the single species F. arfakensis showed disjunct infection statuses, with 26% infected overall but with different infection frequencies and strains at different elevations (e.g., wspC3 in the lowlands and wsp C2 in the highlands). For F. trichocerasa (a single species comprising two distinct subspecies) the proportion of infected wasp pollinators differed between host fig subspecies with lowland subsp. trichocerasa and highland subsp. pleioclada having 84% and 54% infection frequency, respectively. Strain identity was also largely distinct to a given subspecies. In the case of F. wassa (a genetically homogenous entity across the gradient hosting a two major pollinator clades), only 10% of pollinator wasps (all individuals from highland populations) were infected. Overall, sister species/populations of wasps usually had different Wolbachia infection status or strain type (Fig. 3 & Fig. S1). These sister species of wasps were not infected by monophyletic MLST or wsp (except wspC6 ) clades of Wolbachia.
wsp strains appear restricted to lowlands or highlands. For instance, wsp clades 1, 6_1 and 6_2 are present in wasps from elevations above 2,200m while the rest occur in the lowlands (below 1,200m). An exception is for wasps originating from the mid-elevation site (here considered as 1,700m) “Degenumbu” where both lowland and highland strains occur. For instance, wsp clade 1 (a highland strain) occurs in F. wassa wasps from this location; similarly, wsp clade 2, a lowland strain occurs in F. arfakensis wasps from 1,700m. Overall, bar a few exceptions, strain type segregates by (sub)species while infection status seems to be influenced by elevation.
Simulation of Wolbachia distribution among host species under the ‘contact contingency’ hypothesis
Our wolPredictor simulation was able to predict positive strain associations at up to 88.46% (92/104 individuals at species clustering levels of 10-13; SI runs ‘pleio ’ 4, 5, 12 & 18) accuracy against the empirically observed infection statuses across our fig wasp phylogeny (Fig. 4). Predictive accuracy of greater than 80% was found in 16 of 20 runs at species clustering levels of 10-19. Investigation at these species delimitation assessments show high congruence with species diversity patterns in Souto-Vilarós et al. (2019), notably with wasps from F. arfakensis , F. pleioclada and F. wassa split into two or three putative species featuring alternate Wolbachia strain statuses. The highest accurate overall strain prediction, 65.61%, (30 positive and 136 negative predictions) regularly occurred at species clustering levels of 5-7 – with the wolPurger function removing around 30 positive predictions and adding >100 negative predictions. In general, improved negative strain accuracy often trades-off with losses in correct positive strain predictions. High non-infection prediction accuracy occurs at lower species clustering levels where large singleton clades within communities are ascribed negative Wolbachia associations.
One-sample T-tests of these best-scoring results against 1000 randomly generated predictions (mean = 14.16% accuracy for positive strains only; mean = 37.86% accuracy including negative strains) indicates that our model simulation predicts Wolbachia infection status with significantly higher accuracy (t = -698.98, d.f. = 999, p < 2.2e-16 for positive strains only; t = -335.61, d.f. = 999, p < 2.2e-16 including negative strains). As a further control, we also ran 100 wolPredictor simulations (see SI files:pleio_shuff* ) with assayed wsp clades randomly shuffled – the best predictive power for positive strains fell to around 46.15% (mean = 40.19%) significantly less than our best prediction for positive strains (t = -33542, d.f. = 99, p < 2.2e-16).