Introduction

Bacteria of the genus Wolbachia are intracellular microorganisms that infect the reproductive tissues of arthropods and nematodes (O’Neill et al., 1997; Werren, 1997; Stouthamer et al., 1999). They are inherited cytoplasmically (i.e. passed from mother to daughter) and alter reproduction in their arthropod hosts in a number of ways, including cytoplasmic incompatibility, male killing, feminization and imposition of parthenogenesis (see O’Neill et al., 1997 and chapters therein). Wolbachia are extremely common, infecting 16–22% of insects (Werren et al., 1995; West et al., 1998; Werren & Windsor, 2000), with a recent study indicating that the percentage might be even higher (Jeyaprakash & Hoy, 2000). In addition to immediate reproductive modifications, Wolbachia infection has a range of longer term evolutionary impacts on host taxa (O’Neill et al., 1997; Werren, 1997; Stouthamer et al., 1999).

Most research on Wolbachia, particularly for insect hosts, has concentrated on understanding the phylogenetic distribution and extent of infection (Werren et al., 1995; West et al., 1998; Werren & Windsor, 2000), using a small number of individuals for each species. The few studies examining geographical variation in levels of Wolbachia infection of a single host have revealed spatial patterns in the presence/absence of the bacterium and the occurrence of multiple infections (Turelli et al., 1992; Solignac et al., 1994; Plantard et al., 1998; Malloch et al. 2000). Studies that combine analyses of spatial variation in Wolbachia infection, and of genetic diversity in the host, are of particular relevance because of the potential impact of Wolbachia on the host’s genetic structure (described below). In this paper, we address the potential impact of Wolbachia infection on large-scale genetic patterning in an insect host, the oak gallwasp Biorhiza pallida.

When an advantageous mutation is driven through a population to fixation (a process known as a selective sweep), much of the neutral variation at linked loci is eliminated during the process (Maynard Smith & Haigh, 1974). The genetic variants at the linked loci that are initially paired with the advantageous mutation, although neutral in themselves, ‘hitchhike’ to fixation. Spread of a specific Wolbachia strain through a host population can have a directly analogous effect on variation in other cytoplasmically inherited markers, such as mitochondrial DNA. Evidence from Drosophila suggests that a single mtDNA haplotype may become widespread in the host population through hitchhiking with a successful Wolbachia strain (Turelli et al., 1992; Ballard et al., 1996). Selective sweeps on mtDNA not only reduce haplotype diversity but also cause the remaining set of host haplotypes to deviate from predictions based on neutrality (Johnstone & Hurst, 1996).

These potential impacts of Wolbachia infection on mitochondrial markers are particularly important because the latter are often employed as phylogeographic markers under the assumption of neutral evolution (Johnstone & Hurst, 1996). Low variability in mtDNA, and departure from neutrality, can also be caused by demographic processes such as range expansions or population bottlenecks (Avise, 2000). Demographic and Wolbachia-related explanations for spatial patterning in mtDNA variation can be distinguished by comparing patterns seen for nuclear and mitochondrial markers. Unlike mitochondrial markers, nuclear markers are typically inherited in a Mendelian fashion and are not expected to show any change in nucleotide diversity or deviation from neutrality in response to Wolbachia infection, although this depends on the phenotypic effect that Wolbachia has on its host. More specifically, only if sexual reproduction is involved can this assumption hold. If the phenotypic effect of Wolbachia on its host is the induction of parthenogenesis (and assuming that genetic exchange with sexual relatives is rare), concordance between nuclear and mitochondrial markers of the host may be expected.

In contrast to a Wolbachia-induced selective sweep, demographic processes cause changes in the extent and nature of variation for both mitochondrial and nuclear markers, although mitochondrial markers are expected to show stronger responses due to their lower effective population size (Avise, 2000). This difference is the basis for demonstrating Wolbachia-associated selective sweeps using comparative studies of sequence variation in mitochondrial and nuclear markers (for example, Turelli et al., 1992; Ballard et al., 1996). For hosts infected with Wolbachia, concordance in spatial patterning of genetic diversity in nuclear and mitochondrial markers argues against a significant causative role for Wolbachia infection.

Undetected selective sweeps on mtDNA variation due to Wolbachia infection can thus generate at least two types of artefact if interpreted in a phylogeographic context. Firstly, loss of mtDNA diversity in part of the host species’ range could be attributed to a demographic effect having a similar impact, such as a population bottleneck. Here, a process that in fact affects only cytoplasmically inherited markers could be mistaken for a demographic process affecting both mitochondrial and nuclear markers. Secondly, patterning in mtDNA variation generated by spatial patterning in Wolbachia infection could be interpreted mistakenly as indicative of the phylogeographic history of the host. These potential pitfalls are highly relevant to studies of insect phylogeography, because of the high proportion of insect species infected with Wolbachia (Werren et al., 1995; West et al., 1998; Jeyaprakash & Hoy, 2000; Werren & Windsor, 2000) and the widespread use of mitochondrial markers in phylogeographic reconstruction (Avise, 2000).

In this paper, we analyse spatial patterns of variation in Wolbachia infection, and mitochondrial and nuclear markers for a widespread European phytophagous insect, the oak-apple gallwasp Biorhiza pallida (Olivier 1791). B. pallida induces galls on oaks in the genus Quercus (Csóka, 1997). B. pallida is extremely widely distributed in the western Palaearctic, extending from Morocco in the west to Georgia in the east, and as far north as Sweden. Wolbachia infection was detected in central European populations of B. pallida as part of a broader phylogenctic survey of Wolbachia in oak gallwasps (Rokas, unpublished data). These two features of B. pallida make it a suitable taxon within which to examine large scale spatial patterns in Wolbachia infection, and any associated impacts on variation at both mitochondrial and nuclear markers.

Oak gallwasps are obligate parasites of oak trees, and their spatial patterns of genetic variation are expected to reflect to an extent the phylogeographic history of their oak hosts. During the Pleistocene (1.8 mya until 0.01 mya), cycles of glacials and interglacials led to repeated range contraction and expansion of many taxa across Europe, with many species surviving the glacials in refugial areas in the southern of Europe (Italy, the Balkans and Spain) (Huntley & Webb, 1989; Hewitt, 1999). The retreat of the last ice sheet across Europe at the end of the Pleistocene era was followed by northern range expansion by many organisms from one or more of these refugia, resulting in the distribution pattern we see today. A growing body of work on oak gallwasps, covering 12 species in the genera Andricus and Cynips, shows that their current distributions and geographical patterns of genetic variation are determined largely by two factors: (a) the number and location of regions that acted as glacial refugia, and (b) the extent to which alternative refugia have contributed colonists to postglacial range expansion (Stone & Sunnucks, 1993; Sunnucks & Stone, 1996; Atkinson, 2000; Stone et al. 2001). Consistent features of all species studied include (a) the existence of refuge-specific allozyme alleles and mtDNA haplotypes, and (b) a decline in genetic diversity with increasing latitude and distance from refugia. Where both have been studied, patterns of genetic variation north of glacial refugia are similar for mitochondrial and nuclear markers (Stone et al. 2001), and are consistent with genetic subsampling of neutral variation associated with the range expansion process (Stone & Sunnucks, 1993; Sunnucks & Stone, 1996; Atkinson, 2000). Preliminary sampling of the species whose phylogeography has been studied also suggests that, unlike B. pallida, they are free of Wolbachia infection. Although precise phylogeographic scenarios vary among species, these studies indicate that in the absence of any impact of Wolbachia infection, similar concordance in spatial patterning in nuclear and mitochondrial markers represents a qualitative null expectation for B. pallida. In contrast, discordant patterning in nuclear and mitochondrial markers, and concordance in patterning of Wolbachia infection and mtDNA diversity would suggest a significant impact of Wolbachia.

We established spatial patterns in Wolbachia infection by using a polymerase chain reaction (PCR)-based screening technique for 218 B. pallida individuals sampled from 46 localities in eight countries across Europe (Fig. 1). Strain diversity of Wolbachia was assessed by sequencing a fragment of the wsp gene. Host genetic diversity was analysed using a mitochondrial sequence (a fragment of the cytochrome b gene) and two nuclear sequences (the internal transcriber regions ITS1 and ITS2). We extended the diversity of nuclear markers sampled by screening 270 individuals for five polymorphic allozyme loci. We use these data to address the following questions: (a) how many strains of Wolbachia are present in this host, and how many infection events have occurred?; (b) is infection with Wolbachia associated with lower sequence diversity, and departure from neutrality, for the mitochondrial marker?; (c) do the mitochondrial and nuclear markers show similar or discordant patterns of variation?; and (d) can we discriminate between demographic processes and a Wolbachia-associated selective sweep as possible causes of observed variation in host mitochondrial DNA?

Fig. 1
figure 1

Map of collection sites of populations of Biorhiza pallida. Spain: (1) Prado del Rey (2–8) Guadalix de la Sierra; Los Molinos; Zarzalejo; Avila-El Escorial; Cercedilla; Soto del Real; Villaviciosa. Northern Spain: (9) Puerto de Velate. Hungary: (10–16) Tiszaigar; Mátrafüred, Szentendre; Karcag; Bajna; Szeghalom; Visegrad. France: (17) Rennes; (18) St Jean pied de Porte; (19) Clermont-l’Hérault. U.K. (20–39) Cambridge; Hertford; Birnwood Forest; Oxford; Isle of Wight; Hampstead Heath; Fakenham; Broughton; SW Lincoln; Thetford; Elsfield; London; Ascot; Cawood; Chatham; West Shropshire; Grace Dieu Wood; Shorne; Hertingfordbury; Stoughton. Sweden: (40) Uppsala. Italy: (41) Chianti; (42) Volterra; (43) Casina. Ireland: (44) Dublin. Switzerland: (45) Luin. Populations in bold italics in central Spain (2–8) indicate absence of Wolbachia infection.

Full size image

Materials and methods

Collection and DNA extraction

Sexual generation galls of B. pallida were collected from 46 localities in eight European countries (Fig. 1, Table 1). B. pallida galls are multilocular (more than one offspring emerges from a single gall). To minimize screening of siblings we used one female from each gall, except where there was just a single gall from a particular location, when two females were screened. DNA was extracted from 218 female wasps as described by Stone & Cook (1998). To avoid contamination, each female wasp was soaked in 5% bleach and then serially rinsed in drops of sterile water prior to DNA extraction. With each DNA extraction three control extractions were performed using a Nasonia Wolbachia-positive strain, a Nasonia Wolbachia-negative strain and a no-DNA sample.

Table 1 Sampling regions of Biorhiza pallida. + indicates infection with Wolbachia − indicates absence of infection
Full size table

Wolbachia screening

Screening for Wolbachia was performed by PCR using Wolbachia-specific primers for the ftsZ cell-cycle gene (Werren et al., 1995). These and all other PCRs were performed in a PTC-200 DNA engine (MJ Research Waltham, MA, USA). The forward primer was ftsZF1 (Werren et al., 1995) and a new reverse primer WOLG-R was designed based on sequences available in GenBank. The sequence of WOLG-R (26 nucleotides) is 5′-GCA GVA TCA ACY TCA AAY ARA GTC AT-3′ (V=G/A/C, Y=C/T, R=A/G). The ftsZF1–WOLG-R pair amplifies the A, B and C groups of Wolbachia. Screening PCRs were attempted for sample DNA extractions at dilutions ranging from 1/10 to 1/100. Control PCRs were always performed. The PCR cycle for ftsZ was: one cycle of 94°C for 3 min, 55°C for 90 s and 72°C for 5 min, followed by 35 cycles of 94°C for 30 s, 55°C for 90 s, 72°C for 5 min and a final extension step at 72°C for 5 min. The PCRs were performed in 25 μL volumes, consisting of 1 μL DNA sample, 2.5 μL 10 × PARR buffer (Hybaid Ashford, UK), 1 μL MgCl2 (25 mM), 0.5 μL dNTPs (10 mM), 0.35 μL of each primer (20 mM), 0.25 μL Taq (Promega Madison, WI, USA) and 19.05 μL of distilled, deionized H2O. A 1% ethidium bromide-stained agarose gel was used for electrophoresis, and loaded with 15 μL of each reaction. To check that any Wolbachia-negative samples were not artefactual because of (a) failed DNA extraction; (b) presence of PCR inhibitors; or (c) incorrect DNA concentration, control PCRs with the general eukaryotic 28S rDNA primers 28Sf and 28Sr were performed as described in Werren et al. (1995). Of the 218 individuals screened for Wolbachia infection, 12 did not amplify for 28S rDNA and were discarded.

PCR amplification and sequencing

All sequencing reactions were carried out at least twice (either with the forward and reverse primers for wsp and cytochrome b or twice with the forward primers for the ITS fragments) to minimize PCR artefacts, ambiguities and base-calling errors. Sequencing was carried out using the a sequencing kit and an automated sequencer (BigDye Terminator kit and ABI 377 sequencer; Perkin-Elmer Foster City, CA, USA). Wolbachia diversity in infected populations was assessed by PCR and sequencing of a fragment of the wsp gene (nine specimens were analysed: two from Hungary; one from southern Spain; two from France, southern and central; one from Switzerland; one from the U.K.; one from Ireland; and one from Italy). This is the most polymorphic gene so far isolated from Wolbachia (Zhou et al., 1998) and hence, the most likely to distinguish between two closely related Wolbachia strains. Wolbachia-infected individuals were sequenced for wsp using the 81F and 691R primers following the methods described by Zhou et al. (1998). The total volume of three PCR reactions for each individual wasp was electrophoresed on a 1% agarose gel. The expected bands were cut from the gel and cleaned with a DNA extraction kit (QIAQuick gel extraction kit, no. 28704; Qiagen, Crawley, UK), and the clean DNA fragment was quantified and sequenced. Twenty-nine individuals (10 from central Spain; five from southern Spain; three from the U.K.; three from southern France; four from Hungary; one from northern Spain; one from Switzerland; one from Ireland; and one from Italy) were sequenced for a 433-base pair (bp) fragment of cytochrome b showing 13 distinct haplotypes (Table 2). The cytochrome b fragment was amplified using the primers CB1 and CB2 as described by Stone & Cook (1998), and purified and sequenced as described above for wsp. Internal transcriber regions were amplified using the universal primers ITS4 and ITS5 (White et al., 1990). The amplified fragment consisted of the internal transcriber regions, ITS1 and ITS2, and the 5.8S rDNA region of the rDNA array. Fifty-four clones of a 474-bp ITS2 fragment were sequenced for 13 individuals (three from central Spain, one from southern Spain, one from northern Spain, two from France, two from Italy, one from Hungary, one from Switzerland, one from Ireland and one from the U.K.) and 28 clones of a 635-bp fragment of ITSl for six (of the 13) individuals (see below). The PCR cycle consisted of an initial denaturation step of 94°C for 2 min, followed by 35 cycles of 94°C for 15 s, 55°C for 60 s and 72°C for 2 min, and a final extension step of 15 min at 72°C. Reaction conditions were as for Wolbachia screening, except that 1.5 μL MgCl2 and 18.55 μL of deionized, distilled H2O were used. The rDNA array is present in multiple copies in the typical eukaryote genome and undergoes concerted evolution (Hillis & Dixon, 1991). However, for fast-evolving regions of the array, concerted evolution is not always perfect, resulting in intraindividual variation. To cheek for this variation, PCR products were cloned using a cloning kit (TOPO TA cloning kit no. 4500–01; Invitrogen Groningen, Netherlands) and 2–6 clones from each specimen were subsequently sequenced. Plasmids containing the fragment of interest were isolated using a commercial kit (QIAprep Spin miniprep kit, no. 27104; Qiagen). Plasmid DNA was subsequently quantified and sequenced.

Table 2 List of haplotypes for the 433-bp fragment of cytochrome b from Biorhiza pallida. Parsimony-informative sites are indicated by an asterisk (*) and singletons by a full stop (.)
Full size table

Allozyme screening

Two hundred and seventy individuals were screened at five variable allozyme loci using cellulose–acetate gel electrophoresis (Zip Zone equipment; Helena Laboratories Gatshead, UK) as described in Stone & Sunnucks (1993) and Stone et al. (2001). Of an original set of 13 loci (see Stone et al. 2001) screened in B. pallida, five systems were found to be polymorphic: peptidase b (PEPb) (EC 3.4.11); aspartate aminotransferase (GOTm) (EC 2.6.1.1); 6-phosphogluconate dehydrogenase (6PGD) (EC 1. 1. 1.44); malate dehydrogenase (MDHs) (EC 1.1.1.37); and glucose-phosphate isomerase (GPI) (EC 5.3.1.9). PEPb, GOTm, 6PGD and MDHs were run on a sodium phosphate buffered gel (pH 6.3), and GPI was run using a Tris-EDTA–maleate–MgCl2 buffered gel (pH=7.6).

Analysis of sequence data

Sequences were aligned by CLUSTALW (Thompson et al., 1994) using the default settings. ITS1 and ITS2 alignments were manually checked to verify that there were no ambiguities. Departures from neutrality were tested using Tajima’s D (Tajima, 1989) and Fu and Li’s D* and F* (Fu & Li, 1993) statistics incorporated into the software program (DnaSP, version 3.0) (Rozas & Rozas, 1999). For ITS1 and ITS2, sites with alignment gaps were excluded from the neutrality tests. All generated sequences were used (including the multiple clones from each individual).

Phylogenies for cytochrome b were generated using parsimony (MP) and maximum likelihood (ML) in PAUP* (version 4.0b3) (Swofford, 1999). MP was performed for 1000 bootstraps replicated with all sites equally weighted, using the tree bisection-rooting (TBR) option in a heuristic search. All ML analyses were heuristic searches using the TBR option. To reduce computational time, only 100 bootstrap replications were performed for each ML analysis. A hierarchical series of increasingly complex models of sequence evolution was employed to identify the model that made the data most likely using likelihood ratio test (LRT) statistics (reviewed by Huelsenbeck & Rannala, 1997). We tested, singly and together, the effects of unequal base frequencies, different rates between transitions and transversions (ti/tv) and rate variation over nucleotide sites (Hasegawa et al., 1985). The assumption of among-site rate heterogeneity (Yang, 1993) and the enforcement of a molecular clock were also tested for the best-fitting model. The shape parameter α of the gamma distribution and the ti/tv ratio were calculated from a 50% majority rule consensus tree using unweighted parsimony. Whenever unequal base frequencies were employed, we used the empirical frequencies of the nucleotides (Yang et al., 1994).

For ITS, phylogenies were constructed using combined data for ITS1 and ITS2. Each gap was coded as missing and as a fifth nucleotide character. Trees were constructed using MP (as above) and ML, using PUZZLE (Strimmer & von Haeseler, 1996). MP bootstraps were carried out as above for 100 replicates. ML was performed using the quartet puzzling algorithm with 10 000 puzzling steps, the HKY85 model of DNA substitution (Hasegawa et al., 1985) and a gamma-shaped distribution to account for rate heterogeneity. All parameters were estimated from the dataset. A parsimony analysis was also performed using the gap insertions as unique indel (insertion/deletion) events. They were treated separately from the sequence dataset because no data exist concerning their frequency of substitution.

Results

Wolbachia screening and diversity

Of the 206 B. pallida successfully screened, 85% were infected with Wolbachia. Over the entire sampled range only populations in central Spain were not infected (Fig. 1, Table 1). All sequences were 564-bp long (GenBank accession number AF339629), with only two nucleotide positions polymorphic. Position 218 was polymorphic for T/C and position 230 was polymorphic for A/G in all the individual sequences. As the sequencing was done from a PCR fragment, it could not be determined whether these sites are genuinely polymorphic (not fixed yet, four alleles) or represent two different strains (two alleles). The low wsp diversity suggests that all infected European populations carry the same strain of Wolbachia.

Mitochondrial phylogeography

The 29 individuals sequenced for the cytochrome b fragment were polymorphic for 25 positions, 18 of which were parsimony informative (4.18%). This variation yielded a total of 13 discrete haplotypes (see Table 2). Likelihood ratio tests showed that the best-fitting model for the B. pallida data used empirical base frequencies, a ti/tv ratio equal to 10.44 and variation in rate among sites (using a gamma distribution with α=0.0057 and four rate categories). The estimated value α=0.0057 denotes a very strong rate variation (see Yang et al., 1994). The assumption of a molecular clock could not be rejected at the 0.05% probability level (unconstrained model versus model with molecular clock enforced, −ΔL=44.42, d.f=27, 0.01 < P < 0.05).

All MP and ML analyses reveal a deep and well-supported split (based on 10 changes) between populations in central and southern Spain versus those in northern Spain and the rest of Europe (Fig. 2). Sequence diversity was higher in Wolbachia-free populations in central and southern Spain (P=0.0082) than in the Wolbachia-infected populations throughout northern Spain and the rest of Europe (P=0.0029). The 14 sequences from the populations in northern Spain and the rest of Europe showed significant deviation from predictions under neutrality (Tajima’s D=−2.09, P < 0.05; Fu and Li’s D*=−2.73, P < 0.05 and F*=−2.93, P < 0.05). In contrast, central and southern Spanish sequences, from either the 10 uninfected central Spanish individuals (Tajima’s D=0.52, P > 10; Fu and Li’s D*=0.62, P > 0.10 and F*=0.67, P > 10), or the central and southern Spanish individuals combined (15 sequences) (Tajima’s D=1.20, P > 0.10; Fu and Li’s D*=0.89, P > 0.10 and F*=1.13, P > 0.10), showed no significant deviation from neutrality.

Fig. 2
figure 2

Consensus phylogram (50% majority rule) for cytochrome b haplotypes using an ML model that accounts for unequal base frequencies, a different ti/tv ratio and rate variation among sites. See Table 2 for haplotype information. Values above the branches denote bootstrap support under ML and MP, respectively. The tree is 35 steps long with a consistency index (CI) of 0.714 and a retention index (RI) of 0.931. GenBank accession numbers: AF339616–AF339628.

Full size image

ITS phylogeography

The 13 individuals sequenced for ITS2 and the six individuals sequenced for ITS1 and ITS2 revealed 43 parsimony-informative positions (Table 3). The length of the fragments varied between 456 and 467 bp for ITS2 and between 626 and 632 bp for ITS1. Length differences from clones from the same individual were usually much smaller or absent. The ML tree for ITS constructed with the quartet puzzling algorithm, and the MP tree coding alignment gaps as missing information, were both poorly resolved. However, MP analysis with gaps coded as a fifth base strongly supports the split of the population from central and southern Spain versus that from northern Spain and the rest of Europe revealed by the mitochondrial cytochrome b data (Fig. 3). MP analysis of the ITS indel dataset (with 10 of 12 indels being parsimony informative) revealed the same topology, although with lower bootstrap support.

Table 3 Parsimony-informative sites for ITS1 and ITS2. A dash (–) denotes a gap and a full stop (.) denotes an identical site. Individuals with the same parsimony-informative sites were grouped together (see below.) Numbers indicate number of individual and number of clone, respectively. All numbers refer to the individuals in Fig. 3. An asterisk (*) indicates individuals that were sequenced for both ITS1 and ITS2. Individuals without an asterisk were sequenced only for ITS2. Groups – 1: 123.5; 2: 140.10*; 3: 117.3*, 123.11, 123.13, 48.4, 48.11, 53.4, 212.11, 35.3, 48.12, 53.11, 96.20*, 53.2, 48.7, 35.1, 48.3; 4: 212.3, 212.9, 212.7, 212.8, 96.13*; 5: 96.10*, 96.18*, 96.16*; 6: 117.15*; 7: 117.2*; 8: 123.1; 9: 141.6*, 140.5*, 141.10*, 141.3*, 48.5; 10: 96.19*; 11: 140.13*; 12: 141.5*; 13: 141.12*; 14: 147.5*; 15: 147.14*; 16: 147.15*; 17: 157.6, 157.16; 18: 157.13, 157.7, 172.5; 19: 147.19*; 20: 147.13*; 21: 32.5*; 22: 32.7*, 32.3*, 32.8*; 23: 32.10*; 24: 147.3*; 25: 123.3
Full size table
Fig. 3
figure 3

Consensus tree (50% majority rule) for ITS using MP, with gaps coded as the fifth base. Numbers in branches denote bootstrap support under MP (alignment gaps as fifth base). For southern and central Spain versus northern Spain and the rest of Europe split, the bootstrap support given by MP analysis of the indel dataset is shown in italics. Numbers next to countries’ names indicate number of individual and number of clone, respectively. The tree is 141 steps long with a consistency index (CI) of 0.794 and a retention index (RI) of 0.925. GenBank accession numbers: ITS1, AF340069–AF340096; ITS2, AF340097–AF340150.

Full size image

Deviation from neutrality was tested separately for ITS1 and ITS2. Given that these are nuclear loci, within which recombination might be occurring, the estimates for deviation from neutrality are conservative. Significantly, regional variation in departures from neutral expectations parallel those seen in cytochrome b. For both loci, Fu and Li’s D* and F* and Tajima’s D showed significant departure from neutrality for populations outside Spain (ITS1: Tajima’s D=−1.92, P < 0.05; Fu and Li’s D*=−2.77, P < 0.05 and F*=−2.92, P < 0.05. ITS2: Tajima’s D=−1.98, P < 0.05; Fu and Li’s D*=4.15, P < 0.05 and F*=4.05, P < 0.05) but were nonsignificant for central Spanish specimens (ITS1: Tajima’s D=−1.01, P > 0.10; Fu and Li’s D*=−0.99, P > 0.10 and F*=−1.07, P > 0.10. ITS2: Tajima’s D=−0.77, P > 0.10, Fu and Li’s D*=0.86, P > 0.10 and F*=−0.95, P > 0.10).

Allozyme variability

The sample sizes obtained from each location (maximum 12 galls, and so 12 females screened) were too small to allow meaningful analysis of allele frequencies. We thus limit our interpretation of the data to the presence/absence of specific alleles, as summarized in Table 4. Four alleles were found only in Spain. Of these, two were present only in central Spain; allele 1 at 6PGD was present only in one population, whereas allele 2 at 6PGD was present in four populations. The third allele was present in the only southern Spanish population (allele 4 at GPI) and one was present in all 10 Spanish populations (allele 1 at MDHs). Populations in the rest of Europe possessed two alleles absent from Spain. One was from a single individual from Stoughton, U.K. (allele 2 at GPI) and the other was very common in individuals from Phoenix Park, Ireland (allele 4 at MDHs). These regional differences in locally restricted alleles support the substantial genetic divergence between central and southern Spain versus northern Spain and the rest of Europe implied by the cytochrome b and ITS sequence analyses.

Table 4 Biorhiza pallida allozyme alleles from 5 allozyme loci, from 4 areas of Europe. Locally restricted alleles are written in bold. Numbers in parentheses denote the number of individuals and number of populations screened, respectively
Full size table

Discussion

Geographic variation in Wolbachia infection

Extensive screening revealed that, with the exception of central Spain, all European populations of B. pallida sampled are infected with Wolbachia. The apparent single strain infecting B. pallida appears to be relatively cosmopolitan; it is shared with three other European gallwasp species (Rokas, unpublished data), and database searches suggest that it also infects tse-tse flies. There are three possible explanations for existence of infected populations of B. pallida both north and south of a Wolbachia-free region in central Spain: (a) there was a single infection event and central Spanish populations have lost their infection; (b) there was a single infection in a common ancestor of the southern Spanish, northern Spanish and the rest of Europe B. pallida populations not shared with populations in central Spain; or (c) infection has occurred independently in southern Spain and in Europe from the Pyrenees northwards. Too little is currently known about the mechanisms of Wolbachia transfer between lineages for these possibilities to be distinguished, although evidence for horizontal transfer of Wolbachia between and within species is accumulating (Vavre et al., 1999; Huigens et al. 2000). We also know very little about the phenotypic consequences of Wolbachia infection in B. pallida. Wolbachia has been shown to induce parthenogenesis in some non-oak cynipids (e.g. Plantard et al., 1999), but the presence of males in all our samples suggests that this does not occur in B. pallida. Additionally, the slow lifecycle of this host, and obligate development of the larva within oak tissues make the mating and curing experiments needed to examine Wolbachia-associated effects (such as for example, cytoplasmic incompatibility) extremely difficult (see also Plantard et al., 1998; Plantard et al., 1999).

Patterns of genetic diversity in B. pallida

Variability at the mitochondrial locus was higher in populations free of Wolbachia than in the northern Spanish and the rest of Europe populations infected with the symbiont. Similarly, infected northern Spanish and rest of Europe populations showed significant departures from neutrality, while central and southern Spanish populations showed no such signature. Such covariation between infection and sequence diversity is compatible with a Wolbachia-induced selective sweep.

Large-scale demographic processes such as range expansion can have a similar impact on mtDNA variation to that caused by a selective sweep in a population (Donnelly & Tavaré, 1995). Studies on other European oak gallwasps predict a significant impact of range expansion on spatial patterns in B. pallida, and should be regarded as a more parsimonious underlying cause if compatible with the data. As discussed above, demographic processes tend to generate qualitatively similar patterns at both mitochondrial and nuclear markers. This is exactly what we see in B. pallida, indicating that lower genetic diversity in central and northern Europe is far more likely to result from either historically low genetic diversity in nearby refugia, or loss of genetic diversity associated with range expansion, rather than a Wolbachia-associated selective sweep. Wolbachia infection may also generate similar patterns of genetic diversity in nuclear and mitochondrial markers, provided that the effect of Wolbachia on its host is to induce parthenogenesis. However, the presence of males in every population of B. pallida that we collected argues against a scenario of Wolbachia-induced parthenogenesis.

The dominant feature of genetic variation in B. pallida is the deep split between the central and southern Spanish population versus those of northern Spain and the rest of Europe, supported by all the B. pallida datasets. The division of these two groups is robustly supported by the cytochrome b (Fig. 2), and less robustly by the ITS (Fig. 3) analyses. The lower support observed in ITS is due to the fact that the informative sites are in the indel regions. When gaps are encoded as the fifth base or when they are treated as unique indel insertions, the split between central and southern Spain versus northern Spain and the rest of Europe is more strongly supported. Genetic discontinuities between areas north and south of the Pyrenees are known for many plants and animals, and many show hybrid zones at the Pyrenees (see reviews by Taberlet et al., 1998; Hewitt, 1999). For many species, including gallwasps (Stone & Sunnucks, 1993; Stone et al., 2001), expansion following retreat of the ice was principally from Italy and the Balkans, while Spanish populations failed to expand far into France. A similar scenario could explain the patterns seen in B. pallida. Region-specific haplotypes and nuclear alleles imply the existence of discrete refuge populations in Spain and in regions to the east. Some oak gallwasps show further differentiation between distinct refuges in Italy and the Balkans (Atkinson, 2000), but there is inadequate resolution in the B. pallida data to see if the same is true for this species. Range expansion from central or eastern Europe and associated rapid population growth would generate both the similarity among sites and the departure from expectations of neutrality seen outside Spain. Inability of Spanish populations to expand and spread beyond the Pyrenees would lead to absence of Spanish haplotypes or nuclear alleles from more northerly populations, and the absence of any deviation from neutrality. Inability to escape from the Spanish refuge has been clearly demonstrated in one other oak gallwasp, resulting from the evolution of oak-specific ecotypes whose hosts are essentially limited to Spain (Stone et al., 2001). It is interesting to note that patterns of post-Pleistocene range expansion for gallwasps and for their host oaks do not match. Molecular evidence for oaks suggests that individuals from all three refugia (Spain, Italy and the Balkans) contributed to the colonization of central Europe (Ferris et al., 1993; Dumolin-Lapègue et al., 1997). In contrast, gallwasps expanded principally from Italy and the Balkans (see above; Atkinson, 2000; Stone et al., 2001). This discordance may be explained by adaptation of gallwasp populations in the Spanish refuge to local endemic oak species that failed to expand further north (for more discussion see Atkinson, 2000; Stone et al., 2001).

This phylogeographic hypothesis is supported by dating estimates based upon a B. pallida mitochondrial molecular clock, assuming a 2.3% divergence of mtDNA sequences per million years (Brower, 1994). There are two issues for which timing estimates are meaningful. The first concerns the timing of the event leading to the current levels of genetic diversity among northern Spanish and rest of Europe populations. Assuming that the 0–0.9% divergence that is observed among these populations is the result of substitutions that have occurred just after that event, we get an estimation between 0 and 390 000 years ago. This estimation roughly encompasses the range expansion by the oak hosts of B. pallida following the end of the Pleistocene (Ferris et al., 1993; Dumolin-Lapègue et al., 1997). The second issue is the split between central and southern Spain versus northern Spain and the rest of Europe. Levels of divergence between the groups of around 2.5% suggest that this split is ancient, pointing to a separation long before the end of the Pleistocene. Such a division into long-standing eastern and western refugia is supported by similar data for other oak gallwasp species, such as Andricus kollari and Andricus quercustozae (Atkinson, 2000; Stone et al., 2001). Our study shows that while patterns of variation in mitochondrial sequence diversity in this system do not allow a Wolbachia-induced selective sweep to be discounted, consideration of nuclear marker diversity points to a demographic cause.