Mito-nuclear genetic comparison in a Wolbachia infected weevil: insights on reproductive mode, infection age and evolutionary forces shaping genetic variation

Background Maternally inherited endosymbionts like Wolbachia pipientis are in linkage disequilibrium with the mtDNA of their hosts. Therefore, they can induce selective sweeps, decreasing genetic diversity over many generations. This sex ratio distorter, that is involved in the origin of parthenogenesis and other reproductive alterations, infects the parthenogenetic weevil Naupactus cervinus, a serious pest of ornamental and fruit plants. Results Molecular evolution analyses of mitochondrial (COI) and nuclear (ITS1) sequences from 309 individuals of Naupactus cervinus sampled over a broad range of its geographical distribution were carried out. Our results demonstrate lack of recombination in the nuclear fragment, non-random association between nuclear and mitochondrial genomes and the consequent coevolution of both genomes, being an indirect evidence of apomixis. This weevil is infected by a single Wolbachia strain, which could have caused a moderate bottleneck in the invaded population which survived the initial infection. Conclusions Clonal reproduction and Wolbachia infection induce the coevolution of bacterial, mitochondrial and nuclear genomes. The time elapsed since the Wolbachia invasion would have erased the traces of the demographic crash in the mtDNA, being the nuclear genome the only one that retained the signal of the bottleneck. The amount of genetic change accumulated in the mtDNA and the high prevalence of Wolbachia in all populations of N. cervinus agree with the hypothesis of an ancient infection. Wolbachia probably had great influence in shaping the genetic diversity of N. cervinus. However, it would have not caused the extinction of males, since sexual and asexual infected lineages coexisted until recent times.


Background
Parthenogenetic reproduction is fairly common in Curculionidae [1,2]. Many cases have been reported in three different subfamilies: Scolytinae (bark beetles), Listroderinae and Entiminae (broad-nose weevils), especially in species from the Old World (e.g. [1,[3][4][5][6]). Although less studied, weevils from South America also concentrate a huge number of asexual species. For instance, Lanteri and Normark [7] provided a list of more than 30 parthenogenetic or presumptive parthenogenetic species within the tribe Naupactini (subfamily Entiminae). There is no direct evidence of parthenogenesis for the majority of these species. However, Floyd F. Smith and colleagues from the United States Department of Agriculture (USDA) proved this reproductive mode through rearing experiments for the species Naupactus cervinus and N. leucoloma respectively (see [8]), and Marvaldi [9] did the same with N. ruizi.
Parthenogenesis in Curculionidae is thelytokous and apomictic (i.e. ameiotic) [2]. Apomixis was confirmed for several Old World broad-nosed weevils [3,10], although nothing is known for South American species. As a consequence of this kind of reproduction, progeny would result in a group of females genetically identical to their mothers [11]. Nevertheless, the absence of recombination would originate heterozygote genotypes, as rare mutations accumulate [12,13]. The lack of meiosis could be detrimental in the short term [14][15][16], although heterozygosity could explain the higher dispersion and adaptation ability of some parthenogenetic species over the sexual ones in the long term [1].
Traditionally, hybridization and polyploidy were the main traits invoked to explain the origin of asexuality in weevils [2,17]. However, the report of the parthenogenesis inductor bacterium Wolbachia pipientis (hereafter "Wolbachia") in several species of the tribe Naupactini [18,19] and also in the genera Cathormiocerus (Entiminae, Trachyphloeini) [20] and Otiorhynchus (Entiminae, Otiorhynchini) [21], suggested another possible explanation for the origin of this reproductive mode in broadnosed weevils.
Wolbachia infection can produce drastic consequences on the evolution of its host species, such as extinction or sex role reversal [22,23]. However, it can also affect the population genetics and molecular evolution of the vertical transmitted genomes (mitochondrial, and for apomictic species also nuclear DNA), because of the linkage disequilibrium among these cotransmitted molecules [24][25][26].
When an arthropod population is invaded by a Wolbachia strain that rapidly spreads, the host mtDNA (and perhaps the nDNA) associated to the initial infection (i.e. in linkage disequilibrium with the bacterial genome) will hitchhike through the population to fixation [26,27]. Consequently, this selective sweep ("indirect selection on the mtDNA") will lead to a loss of mtDNA/nDNA diversity in part of the host species distribution. Since the expected coalescent time of mtDNA in an uninfected species is twice the effective population size of females (N f ), a decrease in mtDNA/ nDNA diversity will be evident for ca. 2N f generations after the Wolbachia invasion [25]. Then, current levels of genetic diversity in infected arthropod populations should be analyzed in the context of the infection age.
The spread and prevalence of the infection can also be related to other important factors. For instance, the success of the infected females in producing more daughters than uninfected females [28]. If infected females have a higher fitness than unifected ones, a perfect transmission of the endosymbiont will lead to a high prevalence or to its fixation [23].
From the considerations stated before, we formulate the following hypotheses: i) If parthenogenesis is apomictic, the lack of meiosis precludes the occurrence of recombination; ii) Then, asexuality can generate linkage disequilibrium among mitochondrial and nuclear genomes; iii) Vertical transmission of Wolbachia can generate linkage disequilibrium among mitochondrial, nuclear and bacterial genomes; iv) Infection with Wolbachia can sweep the genetic variation on both mitochondrial and nuclear DNAs of the hosts; v) Infection age correlates with recovery of genetic diversity of host genomes and also with the high prevalence of Wolbachia in its populations.
To test these hypotheses we carried out a study on the genetic diversity of the species Naupactus cervinus, as part of a broader research project on the evolution of parthenogenesis in South American weevils of the tribe Naupactini. This pest insect, commonly known as the Fuller's rose weevil, is very attractive for studying the evolution of asexuality, because it is worldwide distributed and it is very abundant in nature. Although probably native to Northeastern Argentina, Southern Brazil and Uruguay, in currently occurs in Australia, Azores Islands, Canary Islands, Chile, France, Italy, Japan, Mexico, Morocco, New Zealand, Spain, USA, etc. as a consequence of commercial trade of several crops, especially ornamental and fruit plants [29][30][31].
Naupactus cervinus reproduces by parthenogenesis, although some sexual lineages have been recorded from Northeastern Argentina and Southern Brazil in the 1940's (see [32]). Rodriguero et al. [19] and Rodriguero [33] reported that all asexual lineages of this species are infected with the Wolbachia strain wNau5, which belongs to the supergroup B.
The main goal of this contribution is to understand the conquences of Wolbachia infection and asexuality on the genomes of N. cervinus through a mito-nuclear genetic comparison. To test the hypotheses previously formulated, particularly the occurrence of apomixis, we will quantify the minimum number of recombination events and the linkage disequilibrium between both genomes. Additionally, the estimation of the number of Wolbachia strains, their prevalence and the infection age, will bring insights on the influence of this endosymbiont on both host genomes, and will contribute to explain the distribution of their genetic variation.

Genetic variation estimates
Three hundred and nine individuals from 38 different locations were screened for genetic variation in a 748 bp fragment of the COI gene. All of them were females. Seventeen mitochondrial haplotypes (arbitrarily named A-R [GenBank: GQ406827 -GQ406843]) and 25 segregating sites were identified, of which only three were singletons. Some haplotypes were found in multiple locations (e.g. haplotype B, Table 1), while others occurred at a single site (e.g. haplotypes A, D, E, G, H, J, L, N, P, Table 1).
Alignment of the translated COI sequences showed a distribution of the genetic variation similar to that reported by Lunt et al. [34]. The absence of stop codons and mutations that alter the reading frame excluded numt amplifications. The total proportions of nucleotides were 31.8% A, 17.0% C, 15.7% G and 35.6% T, with a strong A + T bias (67.4%), which is higher in the third codon position. Estimates of genetic variation for the total sample were θ π = 0.007 ± 0.001, θ W = 0.007 ± 0.002 and Hd = 0.844 ± 0.011. Maximum parsimony search yielded 13 most parsimonious trees 113 steps long. A strict consensus tree is shown, where two main clades have been recovered (Figure 1). One of these clades includes mitochondrial haplotypes from forests ("forest clade"), and the other one mitochondrial haplotypes from open vegetation areas ("grassland clade"), except "F", "G" and "H" that come from a transition zone along the Uruguay River ( Figure 1, Table 1). The most parsimonious trees differentiate in minor changes in the relationships among haplotypes within the "grassland clade", yielding a highly unresolved consensus tree with several branches of zero length (i.e. a large polytomy). Sequencing of ITS1 in 282 individuals yielded a fragment of ca. 1,100 bp, 155 bp of it belonging to the highly conserved 18 S rDNA region, and the remaining to the ITS1 region. Most specimens analyzed were a subsample of those sequenced for COI. Twenty three of these individuals showed chromatograms with double peaks ("double string sequences") ( Table 1). This finding was indicative of at least two simultaneously amplified sequences from the same individual.
Forty-four segregating sites were identified for ITS1 (only four singletons) yielding eight different nuclear haplotypes (arbitrarily named I-VIII [GenBank: GQ406818 -GQ406825]). Unlike COI, insertion/deletion events were frequent in this dataset. However, based on a gap insertion:substitution cost ratio 10:1, primary homologies could be unambiguously established. Nuclear haplotype distribution depicted a pattern similar to COI (e.g. haplotypes VI and VII are widely distributed, and haplotypes I, II and III are restricted to a single site, Table 1). The following values of genetic variability were obtained: θ π = 0.026 ± 0.001, θ W = 0.013 ± 0.002 and Hd = 0.574 ± 0.028. The Hd value was remarkably lower than that obtained for COI, in agreement with the minor number of ITS1 haplotypes. Maximum parsimony search yielded one most parsimonious tree 153 steps long ( Figure 2). In agreement with the mitochondrial dataset, two divergent clades were recovered: one from open vegetation areas ("grassland clade") and the other from forests ("forest clade"), although the later includes a single haplotype (VI) from the transition zone previously mentioned.

Linkage Disequilibrium Analysis
Direct analysis of multilocus genotypes (COI and ITS1) in 282 individuals shows that in most cases, nuclear and mitochondrial genotypes cosegregate (Table 1). In fact, ten out of 17 mitochondrial haplotypes (A, D, G, H, I, J, K, M, N and P) from 104 individuals are exclusively associated with a single nuclear haplotype. For instance, 20 individuals bearing the mitochondrial haplotype "K" always carry the nuclear haplotype "VII", 11 individuals with "P" always carry "IV", 22 individuals with "M" always carry "VII", and so on (see Table 1). Additionally, the most frequent haplotypes "B" and "F" are linked with nuclear haplotypes "VII" and "VI" respectively, in 118 out of 126 individuals (Table 1).
Moreover, the analysis of congruence between COI and ITS1 gene trees reveals signatures of coevolution between both genomes. For instance, eight mitochondrial haplotypes belonging to the "grassland clade" ("A", "B", "D", "I", "J", "K", "M" and "N") share the grassland nuclear haplotype "VII" (Figure 3). The remaining haplotypes from this same clade "F", "G" and "H, which occur in a transition zone between forests and grasslands (Table 1), are linked to the nuclear haplotype "VI". More interestingly, within the mitochondrial "forest clade" the sister haplotypes "P" and "Q" share the forest nuclear haplotype "IV", and the "R" haplotype bears the nuclear genotype "I", phylogenetically related to "IV" (Figure 3). The inverse association was also observed, since some related nuclear haplotypes are linked to the same mitochondrial haplotype. For example "II", "III" and "IV" are linked with "Q", whereas "V" and "VII" are associated with "B" (Figure 3).
Only four individuals show traces of recombination. They are those carrying the haplotype combinations "B-VI" (N = 2), "C-II" (N = 1) and "F-VII" (N = 1). Moroever, ITS1 double strings sequences were found in three mitotypes ("C", "E" and "R"), a fact that could be the consequence of hybridization events.
The hypothesis of mito-nuclear coevolution is also supported by the result of the statistical assessment of the genetic linkage between both genomes. In fact, the 1,403 comparisons of nuclear vs. mitochondrial sites yielded 772 significant values of linkage disequilibrium (see Additional File 1, Table S1). After Bonferroni's correction, the number of significant associations decreased to 606. This led to the rejection of the null hypothesis of linkage equilibrium between the nuclear and mitochondrial fragments herein sequenced.

Analysis of recombination in nuclear sequences
The estimation of the recombination parameter according to R M indicates lack of recombination in the nuclear sequence under study (R M = 0, p = 0). Although ITS1 is a small fragment in comparison with the whole nuclear  genome, this result may be regarded as an indirect evidence of apomixis. Then, the null hypothesis of apomixis cannot be rejected.

Selective sweeps
The MK test, based on polymorphism and divergence, showed a significant deficit of fixed non-synonymous differences in COI (G = 8.199; p < 0.01). Then, purifying or negative selection penalizes the non-synonymous substitutions between N. cervinus and N. dissimulator. The NI and the α parameter support this conclusion (NI = 9.474; α = -8.474). Tajima's D T and Fu and Li's F were nonsignificant (D T = 0.183, p > 0.05 and F = 1.151, p > 0.10), suggesting that genetic variation of COI is under selective neutrality or in drift-mutation equilibrium.
The models selected by MrModelTest were GTR + G [35,36] for the mitochondrial dataset, and HKY85 + I [37,38] for the nuclear dataset. Bayesian phylogenetic analyses converged after 100,000 generations, based on the inspection of the burn-in plot of log-likelihood scores, tree lengths, all model parameters, and the analysis of cumulative posterior probabilities and the standard deviation of split frequencies. Therefore, the first 250 samples from each analysis were discarded, resulting in two posterior distributions containing 750 samples each. A plot of the posterior probabilities of all splits from the two separate MCMCMC runs demonstrated a linear relationship, suggesting that these analyses were not restricted to local optima. The ITS1 sites were optimized onto this phylogram. The likelihood ratio tests obtained were LRT 1 = 1.001 (p = 0.606) for test 1, and LRT 2 = 2.001 (p = 0.317) for test 2. Therefore, the null hypothesis of no positive selection could not be rejected for the nuclear dataset. In fact, out of the 838 nuclear sites examined, 837 were under negative selection (0.99 >p > 0.50) and only one was under positive selection (p > 0.99) (data not shown). On the other hand, Tajima's D T and Fu and Li's F were significantly positive for ITS1 (D T = 2.979, p < 0.01 and F = 2.193, p < 0.02), not accounting for a selective sweep.

Demographic analysis
ModelTest selected the GTR model for ITS1 [35]. The values of the parameters and confidence intervals (α = 0.05) were: g = -1.748 [(-4.950)-(-0.289)] and θ = 0.004 [0.002-0.008]. The sign of g indicates that N. cervinus most likely passed through a bottleneck (i.e. a decline in population size). In turn, the equation of g implies that θ was higher in the past.

Prevalence and diversity of Wolbachia in Naupactus cervinus
A total of 247 individuals of N. cervinus from different locations were screened by PCR assay using Wolbachia 16 S rDNA gene-specific primers. All of them were positive for Wolbachia infection, as it was demonstrated by the amplification of a product of approximately 800 bp. This fragment size was similar to that of the positive control. No PCR band was obtained from the negative control.

Divergence time estimation
The age of the demographic event (i.e. moderate bottleneck) was calculated through a molecular clock, using both the nuclear and the mitochondrial datasets. Table 2 shows a matrix of uncorrected genetic distances among all the nuclear haplotypes (I-VIII). The relative rate tests indicate that all these nDNA lineages may have split at a constant substitution rate (p > 0.10 for all comparisons). The average pairwise divergence among them was D = 0.03198, so the divergence time for this dataset would be ca. 1.9 Myrs. Table 3 shows a matrix of uncorrected genetic distances among all the mitochondrial haplotypes (A-R). As in the former case, the relative rate test indicate that all these mtDNA lineages may have splitted at a constant substitution rate (p > 0.10 for all comparisons). Divergence time among all mitochondrial haplotypes is more recent than that obtained for the nuclear dataset. Based on an average distance D = 0.00962, it was 400,000 years. The difference between both markers will be discussed later.
The wsp is the most rapid evolving gene in Wolbachia [39], but strickingly, it remains invariant in all the individuals assayed. Assuming 0.002 changes per site per Myrs, the wsp sequence should have accumulated 0.0008-0.0036 changes per site in 0.4-1.9 Myrs. In a sequence ca. 580 bp in lenght, we should have found 0.46-2 changes. Therefore, invariancy is not an unexpected result.
The comparison of the haplotype diversity between the two markers herein investigated suggests that the substitution rate of the nuclear genome would be slower than that of the mitochondrial genome, as it has been reported for several animal species (see [50] for a review), including N. cervinus [51].
Apomixis implies clonal transmission of all genomes and lack of recombination. As a consequence, this kind    of asexual reproduction will induce cosegregation of mitochondrial and nuclear genotypes and the consequent linkage disequilibrium among all genetic markers [52,53]. The cosegregation of nuclear and mitochondrial haplotypes seen in most individuals herein analyzed provides evidence of the clonal transmission of both genomes in N. cervinus. Furthermore, the congruence between ITS1 and COI phylogenetic trees indicates that coevolution between these genomes have occurred, since many phylogenetically related mitotypes share the same nuclear genotype or viceversa. This phylogenetic congruence suggests that an ancestral mitochondrial haplotype and all its derived lineages evolved along with the same nuclear haplotype across evolutionary times, possibly as a consequence of the higher substitution rate for the mitochondrial gene. Therefore, the hypothesis of ancient linkage disequilibrium and long history of mitonuclear genomic association is strongly supported.
Only a few number of individuals (1.4%) showed signals of recombination. Besides, traces of hybridization were inferred from individuals carrying at least two nuclear haplotypes or "double strings" ITS1 sequences (8.2%) (a more detailed study of this exciting finding is currently being undertaken). These individuals could be either remnants of historical gene flow, or the consequence of occasional crosses between parthenogenetic females of N. cervinus and males of the same or other related species (as it was reported by Saura et al. [17] and Stenberg et al. [54] for other weevils with similar reproductive behavior). However, these sporadic sexual events would not invalidate the hypothesis of apomixis for this weevil, since there is clonal transmission of nuclear and mitochondrial genomes, lack of recombination (R M = 0) and non-random association for several mitochondrial and nuclear sites.
The cytoplasmic location of the Wolbachia endosymbiont and the mitochondrial organelle will naturally induce linkage disequilibrium among all their genetic markers. Furthermore, the non-random association among mitochondrial and nuclear markers would indicate that the nuclear genome is also linked with the other two. Then, the selective sweep on mtDNA induced by Wolbachia should have affected also the nuclear genome.
If the initial Wolbachia infection have sweeped the mitochondrial diversity, a significant and negative Tajima's D T , and a significant and negative NI should be expected. However, the first statistics suggests selective neutrality and the second, negative selection. The neutral result from Tajima's D T test could be explained by the antiquity of the infection. Most probably, the time elapsed since this event had been long enough to recover the genetic variation to levels prior to the initial selective sweep, increasing Hd and θ W values. A similar result was reported by Shoemaker et al. [55], Keller et al. [56] and Marshall [57], who proposed ancient infections for Solenopsis invicta (Hymenoptera, Formicidae), Chelymorpha alternans (Coleoptera, Chrysomelidae) and Allonemobius fasciatus-socius species complex (Orthoptera, Gryllidae). On the other hand, the apparent contrasting result of the MK test (i.e. negative selection) might be interpreted under the Nearly Neutral Theory [27,[58][59][60][61][62]. In fact, the value for this statistic would indicate an excess of deleterous mutations, which according to the mentioned theory, would be fixed by chance only if the effective population size is small. In other words, the selective pressure coefficients that operate on the COI gene would be so small (i.e. nearly neutral), that elimination cannot occur.
Low nucleotide diversity in the mitochondrial genome is expected in recently infected species [25,26]. However, a comparison with the related sexual species Naupactus xanthographus, which is unifected [19], would indicate that both species exhibit comparable levels of mitochondrial genetic diversity [63]. This result is interpreted as another evidence of the ancient Wolbachia infection in N. cervinus.
Regarding the nuclear gene, even though ITS1 is a non-functional region, it does not evolve freely in insects [64]. In sexual species, ca. 40% of nucleotide sites are implied in complex secondary structures, and hence under natural selection [65]. Among the 838 sites analyzed herein, 837 showed evidence of negative selection, which is overwhelming in comparison with the result of Schlötterer et al. [65]. This bias in the sites under negative selection could be explained by the genetic linkage among the three genomes, in agreement with Schön et al. [52]. Then, the negative selection operating on bacterial, mitochondrial or nuclear genes could also drive other linked nuclear sites to a similar fate.
The positive Tajima's D T obtained for the nuclear gene may suggest a balancing selection, a spatial population structure of the genetic variation or a moderate bottleneck (after Depaulis et al. [66]). The fact that directional negative selection was detected for ITS1 rules out the first possibility and points to a demographic phenomenon as the most likely explanation for the D T result. Indeed, the LAMARC analysis supported this conclusion, suggesting that N. cervinus could have suffered a bottleneck, probably as a consequence of the Wolbachia invasion [67]. The moderate perturbation in the population size of the initial uninfected population could have allowed some lineages to survive this invasion.
Why this demographic phenomenon was not detected for the mitochondrial gene? The mitochondrial genome usually undergoes faster evolution than the nuclear genome (e.g. [68,69]). Thus, the signatures of the bottleneck could have been erased in N. cervinus mtDNA. Then, the significant Tajima's D T suggests that the bottleneck was probably ancient, since the traces of this demographic event are still recovered from the nuclear dataset, but not from the mitochondrial one.
The fact that all the individuals tested were infected with the same Wolbachia strain (identified by three genes of rapid evolution) favors the hypothesis that the infection of N. cervinus occurred once, and that all extant haplotypes of this weevil species are descended from a single ancestral infected female. However, the positive Tajima's D T for nDNA suggests a different scenario, according to which at least two genetic lineages had been infected with the same strain.
The divergence time estimated from mitochondrial and nuclear datasets allowed us to infer that the bottleneck in N. cervinus population occurred between 400,000-1,900,000 years ago, i.e. during the Plio-Pleistocene, a geological period of enormous cyclical changes in the South American forests and grasslands [70,71]. The older age obtained for nDNA can be attributed to those lineages that survived the demographic event, which could produce an overestimation of the calculated age. On the other side, the younger age inferred for the mitochondrial dataset can be attributed to the faster accumulation of genetic variation, which yielded younger lineages that decreases the mean distance, leading to an underestimation of the age. Therefore, nDNA and mtDNA estimates can be considered as the upper and the lower age limits of the demographic crash.
The amount of evolutionary change accumulated in mtDNA is very high in contrast to the lack of change in wsp, which is the most rapidly evolving gene known in Wolbachia [39]. However, according to the age estimated for the infection in N. cervinus, wsp might have accumulated ca. 0.46-2 substitutions in 0.4-1.9 Myrs. So, lack of variation is not an unexpected result. A similar wsp invariant pattern was found in other organisms like sandflies, leaf beetles and fruitflies (e.g. [25,56,72]. An alternative explanation could be that substitution rates for mitochondrial and nuclear genomes are different from those assumed herein, due to asexual reproduction. In fact, acceleration [73] and deceleration [52] of substitution rates have been reported for species with clonal reproduction. The first alternative implies a more recent infection, in agreement with the lack of variation of wsp Wolbachia gene. However, a decelerated substitution rate would indicate an age of infection much older than that herein estimated.
Although our present data do not allow an accurate inference of the infection age, the high prevalence of Wolbachia in the whole species distribution contradict the hypothesis of a recent invasion. In fact, although prevalence of Wolbachia is usually related to different phenotypes [74], it can also be associated to the infection age [23]. As Wolbachia was detected in 100% of the individuals of N. cervinus assayed, its high prevalence is compatible with the hypothesis of an ancient infection.
Our results suggest that Wolbachia "lived together" with its host N. cervinus at least for a third of its lifespan [see [33]]. If the origin of the apomictic lineages were related to this infection, could Wolbachia be responsible for the extinction of all the bisexual populations of N. cervinus? Given that some males were collected in 1945 and 1947 in the forests of Northern Argentina and Southern Brazil [32], it is unlikely that this bacterium caused their extinction. Rather, the intense deforestation of the Paranaense forest that occurred during the last 40 years [75] is the most important factor to explain the extinction of the bisexual populations of this weevil.

Conclusion
We provide first genetic evidence of apomixis for the weevil Naupactus cervinus. This kind of parthenogenesis, in addition to Wolbachia infection, induces linkage disequilibrium among three genomes: the nuclear and mitochondrial weevil genomes and the bacterial genome. Hence, the mito-nuclear genetic variation of the host would have been shaped by apomictic reproduction, the moderate bottleneck probably caused by the initial Wolbachia infection, the long time passed since then, and the high prevalence of the unique Wolbachia strain infecting this weevil.
Even if Wolbachia accounts for the origin of parthenogenesis in N. cervinus, this infection would have not caused the extinction of the bisexual population. The current absence of males and the lack of sexual reproduction during the last 50 years would be a consequence of the destruction of the native forest where these populations occurred.

Sampling and specimens examined
Adults of N. cervinus (Entiminae: Naupactini) were collected during the summers of 2004-2007, on wild and cultivated plants from several geographic locations of Argentina, Southern Brazil and Uruguay. Samples from the countries where the species has been introduced were also included (see Figure 4 and Table 1) (N = 309). Specimens were collected using a beating sheet (55 cm × 55 cm) and stored at -80°C or in 100% ethanol at 4°C for molecular analyses.

PCR assay and sequencing
Total genomic DNA was extracted following the protocol of Reiss et al. [76]. The negative controls were samples lacking DNA template.
Search of multiple Wolbachia lineages within N. cervinus was accomplished through amplification and sequencing of the wsp gene fragment using the primers brought by Braig et al. Prevalence of Wolbachia infection was studied through amplification of 16 S rDNA gene in ten weevils (whenever it was possible) from every sampling site, using the primers designed by O'Neill et al. [82]: forward (5'-TTG TAG CCT GCT ATG GTA TAA CT-3') and reverse (5'-GAA TAG GTA TGA TTT TCA TGT-3').
Total genomic DNA from Drosophila melanogaster naturally infected with Wolbachia was used as a positive control. Negative controls consisted of samples lacking DNA template from insects and D. melanogaster treated with tetracyline. D. melanogaster DNA was kindly provided by Dr. Scott O'Neill (Queensland University, Australia). All experiments were repeated at least twice.
Amplification was carried out in a 50 μl volume reaction with 50-100 ng of DNA used as template, 0.5 μM of each primer, 0.1 mM of each dNTP, 3.0 mM MgCl 2 , 0.05 units of Taq polymerase and reaction buffer 1× (Invitrogen). The reactions were performed in a Gen-eAmp® PCR System 2700 thermal cycler (Applied Biosystems) under the conditions described by Scataglini et al. [48] for COI and Szalanski and Owens [83] for ITS1 fragments.
Double-stranded PCR products were separated by electrophoresis on a 1% agarose gel with TAE buffer containing 0.5 mg⁄ml of ethidium bromide. The bands were excised from the gel and the DNA was purified with a QIAquick Gel Extraction Kit (Qiagen Inc.). DNA was sequenced using a 3130-XL Automatic Sequencer (Applied Biosystems).

Sequence analysis
Standard chromatographic curves of forward and reverse sequences were edited using the Bioedit program [84]. Sequences were translated with the program MEGA v. 4.0.1 [85] to check for the presence of stop codons or frame shifts which might indicate the amplification of pseudogenes [86][87][88][89]. Aminoacid sequences were inferred according to the invertebrate mitochondrial code [34]. Alignment was done using CLUSTAL W [90] and adjusted by eye.
Phylogenetic analyses of mitochondrial and nuclear haplotypes were performed by maximum parsimony using haplotypes as "terminal taxa" with the program NONA v. 2.0 [95], executed through the interface Win-Clada v. 1.00.08 [96]. The implicit enumeration search option was used to get the most parsimonious trees (number of taxa < 25). All characters were regarded as unordered and unweighted. For the nuclear dataset, gaps were treated as fifth state. Clade stability was assessed by 10,000 parsimony bootstrap replications [97]. Naupactus dissimulator was sequenced to use it as outgroup [Gen-Bank: GQ406844 for COI and GQ406826 for ITS1].

Detecting Linkage disequilibrium
Significant associations between mtDNA and nDNA sites were tested using linkage disequilibrium analysis based on the D parameter [98]. This parameter was obtained with DnaSP v. 5.10.00 [94]. Bonferroni's correction for multiple comparisons [99,100] was used for avoiding spurious rejections of the null hypothesis (i.e. linkage equilibrium).

Detecting Recombination
The population recombination parameter was estimated with the method of Hudson and Kaplan [101] based on the minimum number of recombination events in a sample (R M ) using DnaSP v. 5.10.00 [94]. The significance of the test was calculated by performing 10,000 coalescent simulations based on a Monte Carlo process with no recombination [102].

Investigating selective sweeps
Tajima's D T [103] and Fu and Li's F [104] test statistics were applied to evaluate whether the frequency spectrum of segregating sites departs from neutral expectations [105] for both datasets. In addition, the MK test [106] was applied to evaluate a possible correlation between polymorphism and divergence for the synonymous to non-synonymous variation ratio predicted by the Neutral Theory [105] only for the mtDNA dataset (coding sequences). The neutrality index (NI) [107] and the α parameter [108] were used as indicators of the degree and direction of departures from neutrality; NI = 1 indicates strict neutrality and NI < 1 suggests an excess of non-synonymous fixation (i.e. adaptive evolution). The α parameter reflects the proportion of fixed mutations by positive selection. Unlike D T , MK is highly recommended when demographic factors can obscure the results of neutrality tests [27,109,110]. These estimations were performed with DnaSP v. 5.10.00 [94], using the same outgroup species as mentioned above.
For the non-conding nDNA dataset, the approach developed by Wong and Nielsen [111] was used. It is a maximum likelihood method for inferring positive selection in non-coding regions, by comparing non-coding divergence vs. synonymous divergence. This approach assumes that synonymous mutations are neutral and occur at a constant rate within a coding fragment. The ratio of the estimated nucleotide substitution rate in the non-coding region to the estimated synonymous substitution rate from the coding region (ξ), provides a numerical approach to assess whether a non-coding region has evolved by positive selection. The unit of evolution is one nucleotide, and ξ is the nucleotide substitution rate in the non-coding region normalized by the synonymous nucleotide substitution rate in the coding region. Therefore, when a site is subject to neutral selection, ξ = 1. Similarly, ξ > 1 indicates positive selection, while ξ < 1 suggests the occurrence of negative selection.
Three models are implemented in the EvoNC software [111]. The neutral, the two-category, and the threecategory models were used in two different tests to determine whether a non-coding region is under positive selection. The Test 1 compares the neutral vs. the three category model, and the Test 2, less conservative, compares the neutral vs. the two-category model [111]. The two mentioned tests were applied to identify noncoding sites under positive selection, with the coding sites of the mtDNA being considered as linked to the ITS1 sites. Therefore, mitochondrial sites were used to estimate the synonymous evolution rate because nuclear and mitochondrial genomes were assumed to be linked to each other and also to Wolbachia genome.
For obtaining ξ, the nucleotide positions must be optimized on a phylogenetic tree which includes the sequences under analysis (i.e. the mitochondrial and nuclear haplotypes concatenated). A matrix of concatenated sequences was built whenever the COI and ITS1 fragments came from the same individual. This matrix had 21 terminals (including N. dissimulator as outgroup) of 1,521 bp in length (683 bp COI + 838 bp ITS1).
The phylogenetic tree was estimated through Bayesian inference. MrModeltest software v. 2.2 [112] was used to infer the most appropriate model of molecular evolution for each dataset based on the Akaike Information Criterion (AIC) [113,114], as suggested by Posada and Buckley [115].
Bayesian phylogenetic analysis was performed using the "Metropolis-coupled Markov chain Monte Carlo" (MCMCMC) algorithm implemented in MrBayes ver. 3.1.2 [116,117]. A partitioned algorithm was used to account for heterogeneity between the two datasets. Program defaults were used for estimation of priors. Two independent analyses were run using a random starting tree with three heated chains and one cold chain over 400,000 generations, with sampling every 100 generations.
The tree space was explored using four chains: one cold chain and three incrementally heated ones, heat being set as 1/(1 + (i -1) T), where i is the chain number (i.e. 1-4) and T is used as default.
Stationarity of the Bayesian analysis was evaluated with the methodologies and statistics implemented in Tracer [118] and AWTY [119,120] and with the standard deviation of the split frequencies. All posterior samples of a run prior to this point were discarded as burn-in. Remaining trees were used to construct a 50% majority-rule consensus tree with mean branch length estimates. The frequency of all observed bipartitions was used to assess the level of support for each node [116,117].

Analyzing demographic structure
Tajima's test D T [103] results can reveal either selective or demographic events. To disentangle the demographic factors for the interpretation of this test, it was necessary to demonstrate that a population bottleneck had occurred. For this purpose we use the program LAMARC [121] to estimate the relative effective population size parameter (θ) and the exponential growth rate (g) by maximum likelihood, as well as the calculation of 95% confidence intervals. Since the maximum likelihood estimation procedure is computationally intensive, 60 individuals were selected randomly from each dataset because results do not change significantly at this sample size [122]. Positive values of g indicate population growth, while negative values indicate shrinkage. The estimation of the parameters was done using the MCMCMC sampling algorithm for ITS1 implemented in LAMARC v. 2.0 [121]. The analysis consisted of two simultaneous searches with heating temperature adjusted automatically with 10 initial chains of 5,000 steps sampled every 20 steps, followed by two final chains of 50,000 steps sampled every 20 steps. Final most likely estimates (MLEs) were calculated using parameter estimates from two replicated analyses.
The evolution model was selected using the program ModelTest [123] as previously described. According to Kuhner et al. [122], the inclusion of unlinked loci reduces the bias in estimating g with a small amount of individuals. As the reproductive mode of N. cervinus makes this recommendation meaningless, the sign of g was analyzed without taking into account its absolute value.

Estimating divergence times
Molecular clock estimates were calculated through the mean divergence time among all nuclear and mitochondrial haplotypes found for N. cervinus. Relative-rate tests were performed to test the equality of evolutionary rates between lineages [124] using MEGA v. 4.0.1 [85]. The divergence times were calculated from the average of the uncorrected pairwise genetic distances among haplotypes with MEGA v. 4.0.1, using the equation T = D/2k (where T is the mean divergence time, D is the mean number of pairwise differences per site, and k is the estimated rate of nucleotide substitution). We decided not to apply substitution models to correct the pairwise distances, because we considered that the levels of divergence among haplotypes within the same species is low enough to deserve a correction. In order to estimate these divergence times, we used a nucleotide substitution rate of 0.85% per site per Myrs for ITS1 [68] and 1.2% per site per Myrs for COI [125].
Additionally, we calculated the expected variation for Wolbachia wsp gene assuming a nucleotide substitution rate of 0.2% [126].