We analysed 71 P. imperialis wasps from 26 different F. rubiginosa trees, representing 18 sites in Australia and one introduced population in the USA (Table 1, Fig. 1). We found 44 unique cytb haplotypes, 444 bp in length, of which 139 (31.1%) nucleotide sites were polymorphic and 107 (24.1%) parsimony-informative.

Forty-four P. imperialis sequences were used in phylogenetic analyses, together with sequences from 14 other Pleistodontes species 33 and four Ceratosolen fig-pollinating wasp species 34 as outgroups. The MP (not shown) and Bayesian phylogenies (Fig. 2) had very similar topologies, both supporting the monophyly of P. imperialis, but dividing the species into four deep mitochondrial clades. The (GTR) pairwise genetic distances between taxa fell into two distinct groups: 1) distances within each of the four P. imperialis clades (0–7%); and 2) distances between P. imperialis clades (9–17%) and between morphologically distinguishable Pleistodontes species (10–26%). Consequently, the large genetic distances between the four P. imperialis clades are similar to those between morphologically distinct Pleistodontes and indicate cryptic species. There was one divergent P. imperialis sample, from which three wasps grouped only weakly with the rest of clade 1 (Figure 2). Based upon the commonly used mitochondrial DNA (mtDNA) clock rate of 2.3% pairwise divergence/Myr 35, clades 1 and 2 diverged from clades 3 and 4 at least 7.1 million years ago (MYA), clades 3 and 4 split 5.6 MYA and clades 1 and 2 split 4.4 MYA

The four cytb clades show different geographic distributions (Fig. 1). Clade 1 was found throughout the natural range (Queensland (except Townsville) and New South Wales (NSW)) of P. imperialis and was also the only clade found outside the native range (in S. Australia, Victoria and USA). Clade 2 was found only in the Townsville area of Northern Queensland and contained all the yellow wasps that are also restricted to this area. Clade 3 wasps were found only in N. Queensland, while clade 4 was found in N. and S. Queensland, but not in NSW. Overall, clade diversity appears to decrease from North to South (Fig. 1).

There is not, however, simple geographic replacement of clades, as most sites in Queensland have two or more clades present. For example, wasps collected from just five trees near Townsville represent three of the four clades (Table 1). In most cases, all individuals from one fig tree belonged to the same clade. However, in three cases individuals from the same tree belonged to two different clades (Table 1). Given that only 3–5 wasps were sampled per fig tree, this suggests the frequent occurrence of wasps from two or more clades in a single crop of fig fruits.

All 71 wasps harboured Wolbachia, but some carried one strain and others two. We obtained wsp sequences from 27 insects, revealing 6 with single and 21 with double infections (Table 1). The wsp sequences all belong to the Wolbachia A-clade, and revealed three strains (W1, W2 and W3). W1 and W2 strains differed by a single synonymous A/G substitution at position 268, while there was 8.5% nucleotide divergence, including a 3 bp and a 21 bp indel between W2 and W3. There was a highly consistent pattern between Wolbachia infection status and cytb clades. Wasps in clade 3 had a single infection with W1, while wasps from the other three clades all had both the W2 and W3 infections. The single exception was one clade 2 wasp that harboured only strain W2.

We sequenced nuclear 28S (1033 bp) and wg (433 bp) DNA fragments for 22 wasps representing the four cytb clades (Table 1). The 28S data revealed two weakly supported clades (28S.I and 28S.II) that differed from each other by 4–8 nucleotide substitutions, while there was a maximum of 2 base differences between wasps within the two clades (Fig. 3). 28S.II contains individuals from cytb clade 4 only, while 28S.I individuals fall into cytb clades 1–3 (Table 2). Wg also divided the wasps into two clades (wgA and wgB) and these differed from each other by a single nucleotide substitution at a synonymous site (position 175) (0.23% nucleotide divergence). Group wgA contained individuals from cytochrome b clade 3 only, while wgB individuals belonged to clades 1, 2 and 4 (Table 2). To place this in context, we also sequenced wg from eight outgroup taxa: four other Pleistodontes species and four Ceratosolen species. Nucleotide divergence ranged from 1.16% to 3.23% between morphologically distinguishable Pleistodontes species and up to 16% between genera. While there is no clear association between 28S variation and Wolbachia infection, there is an association of Wolbachia with genetic divergence of wg between clade 3 (W1 infection; wgA) and clades 1, 2 and 4 (W2 and W3 infections; wgB). The overall correspondence of different markers is summarised in Table 2.

We have demonstrated that P. imperialis wasps fall into four major mtDNA clades. These differ from each other by 9–17% nucleotides and are supported strongly by both MP and Bayesian phylogenetic analyses (Fig. 2). These deep divergences are very similar to those (10–26%) found between morphologically distinct members of the same genus 33. The mtDNA genetic distances between our clades are also slightly higher than those reported between cryptic species of Pegoscapus fig-pollinating wasps in Panama 14. These deep divergences suggest strongly the existence of four cryptic species within the morphologically defined P. imperialis.

We also showed that all wasps harboured Wolbachia bacteria and that clade 3 wasps had a different infection to all other wasps. Wolbachia can influence both the diversity and evolution of mtDNA and can maintain mtDNA divergences within or between populations of a species 36. The best-studied example is Drosophila simulans which has three distinct haplotypes 37 and is infected by at least five strains of Wolbachia 383940414243. However, in this and other cases where populations are polymorphic for Wolbachia infection (e.g. the gall wasp Biorhiza pallida 44) the genetic distances between different haplotypes (clades) are very much lower than we report here. Nevertheless, given the very high incidence of Wolbachia infections in fig-pollinating wasps 2627, these endosymbionts have probably played a role in host mtDNA evolution and reduce confidence in the applicability of a general mtDNA clock.

Two wasp species may co-occur on a single fig species if one of them has shifted from another host fig. Indeed, some mismatches between figs and wasps at deep phylogenetic levels suggest that host shifts have occurred at times during their coevolutionary history 915. In addition, there are cases of extant co-pollinators that are not closely related species, suggesting more recent host shifts (e.g. 1434. However, despite high genetic distances between the four cytb clades, all three wasp genes sequenced support the monophyly of the P. imperialis complex relative to other Pleistodontes species. Consequently, it seems unlikely that any of our four clades has shifted to F. rubiginosa from another Ficus species. In addition, we have not recorded P. imperialis from another fig species, nor detected mitochondrial haplotype groups shared by Pleistodontes wasps from different fig species (33 and further unpublished data).

Alternatively, the F. rubiginosa/P. imperialis species pair could be in the process of cospeciation, with the partner species diverging together. However, our existing data do not support this notion and it is also difficult to envisage how the fig species might split simultaneously into four genetic units. We found that in most areas, and even some individual trees, two or more wasp clades were present (Table 1), despite limited within-site sampling. Furthermore, as part of a long-term survey in Townsville, we have found that most F. rubiginosa syconia are entered by more than one foundress and that yellow (clade 2) and black wasps (clades 3 or 4 at this site) co-enter about 20% of syconia (53/198 figs from 10 different trees over 2 years). These results suggest that there is no simple segregation of wasp clades between different host trees and that they co-occur regularly in the same trees and even fruits, as reported in Panama for cryptic species of Pegoscapus fig-pollinating wasps 1417.

One possibility is that the different wasp clades predominate in different habitats and that the fig species is diverging into habitat-specific races. For example, in West Africa Ficus ottonifolia occurs in a mosaic of forest and open habitats and has two pollinators. These co-occur locally, but one predominates in open habitat patches and the other in forest patches 1245. F. rubiginosa is also found in both forest and open habitats and further sampling might reveal a similar pattern. However, we believe that this is unlikely as, for logistical reasons, almost all of our samples are from open habitats.

A further possibility is that the fig is diverging into two or more races that are not habitat-specific. Indeed, Dixon et al. (2001) recognised two forms of F. rubiginosa (see background). Most of our sampling was performed before the description of these two forms, so we cannot yet compare pollinators between the different forms. However, most of our samples came from form rubiginosa, arguing against a simple split by form. In addition, the only taxonomic character that separates the two forms is the presence of hairs on the leaves, which may well be a simple polymorphism that has no connection with the pollinators. Finally, even if there is segregation by fig form, this can only provide a partial answer since there are four wasp clades, but only two fig forms. Clearly, genetic studies of the figs are needed to test ideas further. However, we note that recent genetic studies of wasps have revealed cryptic species 14, but genetic studies of the corresponding figs have not 815.

We argue above that there is no good evidence for either parallel divergence of F. rubiginosa and P. imperialis, or for recent host shifts by wasps. Fig wasp divergence might instead occur following the development of spatial or temporal barriers within a single wasp species. For example, temporary geographic isolation of wasp (and fig) populations could occur for periods of time that allow the evolution of reproductive isolation in the wasps, but not the figs. Selection and/or genetic drift may be involved and the approximately 100 times faster generation time of the insects may facilitate their population divergences 11. Many scenarios are possible, but we suggest that the role of Wolbachia deserves further study, since we have detected infection differences and the acquisition of different Wolbachia infections in isolated populations can facilitate or even cause speciation 232425.

There is very little variation in the slower-evolving nuclear genes studied. However, 28S sequences split the wasps into two clades that correspond to cytb clades 1–3 and clade 4, while wg also splits them into two groups, but corresponding to cytb clade 3 and clades 1, 2 and 4. Cytb clade 3 stands out further by differing in its Wolbachia infection status. Consequently, three markers support isolation of clade 3 from the others, and clade 4 differs from all others on the basis of variation in 28S. Clades 1 and 2 differ strongly in mtDNA, but not in the other genes studied. However, clade 2 contains only (and all) the yellow wasps sampled, while all other wasps are black and this effectively provides a nuclear marker supporting isolation of clade 2.

Further resolution of gene flow in the "P. imperialis complex" now requires data from from substantial numbers of wasps representing the different clades. Given the limited variability of the nuclear sequences studied here, a population genetic approach, using microsatellites, may be most appropriate.

Between 1998 and 2003 ripe syconia were collected from F. rubiginosa trees growing at many disparate locations throughout its natural range. Wasps were allowed to emerge naturally from their syconia and were then stored in 95% ethanol at -20C. A single collection tube comprised wasps from syconia from the same tree on the same day. We later conducted genetic analyses on 1–5 female wasps from each sample tube.

The head from each insect was removed and retained as a voucher specimen, and DNA was extracted from the remaining body parts using a simple Chelex extraction procedure 46. We amplified a 444 bp fragment of mitochondrial cytochrome b (cytb) for all wasps using the primers CB1 and CB2 47. For selected individuals (see results), we also amplified a region of the nuclear 28S rRNA gene (28S), using primers D1F and D3R 334849 as well as the wingless (wg) gene, using primers LepWG1 and LepWG2 50. All wasps were also screened for Wolbachia infection by PCR, employing the primers wsp81F and wsp 691R, which amplify part of the Wolbachia surface protein gene (wsp) 51.

Amplification of cytb was performed using a GeneAmp 2400 machine (Perkin-Elmer Cetus) with 3 min at 95°C, followed by 35 cycles of 30 s at 95°C, 1 min at 45°C, 1 min 30 s at 72°C, and a final elongation step of 7 min at 72°C. We increased the annealing temperature for the other gene fragments as follows: 55°C for wsp, 50°C for 28S and 58°C for wg. The sizes of PCR products were 444 bp (cytb), 564–588 bp (wsp), ~1,060 bp (28S), and 433 bp (wg). Ten microlitres of each PCR product was electrophoresed through a 1% agarose gel to determine amplicon size, and the gel band was excised for purification using a GFX DNA Purification Kit (Amersham Pharmacia Biotech Inc). We then sequenced fig wasp genes (cytb, 28S and wg) directly using the same primers employed in PCR.

For wsp, PCR products were first sequenced directly and, if direct sequencing failed three times (or repeatedly generated sequences with multiple peaks), the PCR product was cloned and 6–10 different clones were sequenced to test for the presence of multiple Wolbachia strains. We ligated each PCR product into a T-tailed vector (pGEM-T Easy Vector system, Promega Ltd.) and transformed into E. coli JM109. Positive colonies were selected and plasmid DNA was purified using a GFX Micro Plasmid Prep Kit (Amersham Pharmacia Biotech Inc) and the wsp inserts were sequenced using M13 vector primers. In all cases, we sequenced using the ABI PRISM BigDye Terminator Cycle Sequencing Kit (Perkin Elmer Inc.) and an ABI PRISM 3700 DNA Analyzer (Perkin Elmer Inc). All isolates were sequenced fully in both directions, and sequences have been deposited in GenBank: cytb [GenBank: AY567594–AY567638], 28S [GenBank: AY567639–AY567660] and wg [GenBank: DQ539361–DQ539391].

Sequences were edited and aligned using Sequencher™ (Gene Codes Corporation), and final adjustments to 28S and wsp alignments were made by eye, following previous alignments (wsp: Shoemaker et al. 2002; 28S: Lopez-Vaamonde et al. 2001).

Cytb phylogenies were estimated using Maximum Parsimony (MP) in PAUP* version 4.0b10 52, and Bayesian methods in MRBAYES version 3.0 53. For MP analyses, we conducted an initial heuristic search with 10,000 random additions and TBR branch swapping, holding one tree per replicate. Trees generated by the initial search were then used as starting trees for a second heuristic search, in which multiple trees were saved. We assessed clade support using 1000 bootstrap replicates. For Bayesian analyses, the most appropriate model of nucleotide substitution was determined, using MrModeltest v2.2 54, to be the general time reversible model (nst = 6). The analyses were run for 10⁶ generations, with one tree retained every 100 generations. Likelihood stationarity occurred after 2.5 × 10⁵ generations, and this "burn-in" period was excluded before creating a 50% majority-rule consensus tree in PAUP*.