Abstract
Polyphagous insect herbivores experience different selection pressures on their various host plant species. How this affects population divergence and speciation may be influenced by the bacterial endosymbionts that many harbor. Here, we study the population structure and symbiont community of the pea aphid (Acyrthosiphon pisum), which feeds on a range of legume species and is known to form genetically differentiated host-adapted populations. Aphids were collected from eight legume genera in England and Germany. Extensive host plant associated differentiation was observed with this collection of pea aphids comprising nine genetic clusters, each of which could be associated with a specific food plant. Compared to host plant, geography contributed little to genetic differentiation. The genetic clusters were differentiated to varying degrees, but this did not correlate with their degree of divergence in host use. We surveyed the pea aphid clones for the presence of six facultative (secondary) bacterial endosymbionts and found they were nonrandomly distributed across the aphid genetic clusters and this distribution was similar in the two countries. Aphid clones on average carried 1.4 species of secondary symbiont with those associated with Lathyrus having significantly fewer. The results are interpreted in the light of the evolution of specialization and ecological speciation.
A major aim of contemporary evolutionary biology is to understand the relative roles of natural selection and other processes in the generation of biodiversity (Schluter 2000; Coyne and Orr 2004). The last few years have seen rapid progress in the study of ecological speciation, the process by which natural selection in different environments can promote the initial divergence that may eventually lead to complete reproductive isolation (Rundle and Nosil 2005; Schluter 2009). A combination of new theory and experimental studies has demonstrated the potency of ecological forces in driving differentiation, especially where gene flow is reduced because selection and mating occur in the same sites (Via 2001; Bolnick and Fitzpatrick 2007). However, although divergent selection often leads to significant genetic and phenotypic differentiation of populations, it does not always lead to complete reproductive isolation and thus speciation (Nosil et al. 2009).
Studies of herbivorous insects have provided valuable model systems for understanding ecological specialization and speciation (Via 2001; Drès and Mallet 2002; Funk et al. 2002). Several ecological factors can promote adaptation to host plants, with their importance varying across systems. Plant species often differ considerably in their structural and chemical composition and this leads to selection on the feeding mechanisms and digestive physiology of the insects that attack them (Price et al. 1980; Strong et al. 1984). Host plants may also provide dissimilar physical microhabitats, or lead to exposure to different spectra of natural enemies, both of which will again select for plant-species specific adaptations (Bernays and Graham 1988; Denno et al. 1990; Feder 1995; Nosil and Crespi 2006; Rull et al. 2009). Many insect herbivores mate on the plant on which they have developed, leading to reduced gene flow between populations on different hosts. This process is enhanced by the evolution of host preferences. Finally, heritable symbiotic bacteria that are commonly associated with herbivorous insects may also influence their hosts’ ability to use different food plants and so could affect the course of ecological speciation (Tsuchida et al. 2004; Hosokawa et al. 2007).
The majority of studies that have looked at host use by phytophagous insects have concentrated on pairs or small groups of plant species (for exceptions see Futuyma and Philippi 1987; Carroll and Boyd 1992; Roininen et al. 1993; Nosil and Sandoval 2008; Xie et al. 2008). The pea aphid is a taxon that is composed of multiple host-adapted populations (Müller 1962; Via 1991, 1999; Ferrari et al. 2006, 2008; Peccoud et al. 2009a). It is therefore an ideal system for exploring how a species copes with a large array of potential hosts. Pea aphid populations are typically specialized on the plant from which they are collected (Via 1991; Ferrari et al. 2008; Peccoud et al. 2009a) and in choice tests preferentially colonize that plant species (Via 1999; Ferrari et al. 2006). The one exception is pea aphids collected from broad bean (Vicia faba), which does not support a specialized population (Ferrari et al. 2008; Peccoud et al. 2008, 2009a). Pea aphids from most specialized populations are known to perform well on broad bean and in the field this species tends to be colonized by aphid clones adapted to other host plants (Müller 1962; Ferrari et al. 2008; Peccoud et al. 2009a).
Host-associated pea aphid populations are genetically differentiated and are thought to have radiated onto different plant species within the last 16,000 years (Peccoud et al. 2009a, b). Quantitative trait loci (QTL) for host plant performance and acceptance have been mapped in North American pea aphids specialized on alfalfa (Medicago sativa) and red clover (Trifolium pratense). QTL for the two traits show close physical linkage that may facilitate the evolution of specialization (Hawthorne and Via 2001). Genomic regions that contain host-use QTL tend to be more differentiated among populations compared to other regions, as would be predicted if the extent of divergence is determined by a balance between selection and gene flow (Via and West 2008).
Bacterial symbionts may play a role in ecological speciation by directly affecting how their hosts use different plant species. Aphids harbor a variety of obligate and facultative bacterial endosymbionts. The obligate primary symbiont Buchnera aphidicola synthesizes amino acids that are deficient in the phloem on which aphids feed (Akman Gündüz and Douglas 2009). Primary symbionts are inherited purely vertically and their phylogeny is exactly concordant with that of the aphid (Moran et al. 1999; Clark et al. 2000). Aphids also carry secondary bacterial symbionts, which are predominantly vertically inherited, but horizontal transfer also occurs at unknown frequencies (Russell et al. 2003). The three best known aphid secondary symbionts are the γ-Proteobacteria Hamiltonella defensa, Regiella insecticola, and Serratia symbiotica, which are common and widespread across aphid species (Russell et al. 2003). A series of other taxa have also been recorded but are less well characterized. These include a Rickettsia and a Spiroplasma, and two further γ-Proteobacteria: a Rickettsiella and a bacterium referred to as PAXS or the X-type (Chen et al. 1996; Fukatsu et al. 2001; Sandström et al. 2001; Moran et al. 2005; Guay et al. 2009; Tsuchida et al. 2010). Secondary symbionts affect different aspects of aphid biology including defense against natural enemies (Oliver et al. 2003, 2005; Scarborough et al. 2005), resistance to heat shock (Montllor et al. 2002), and the propensity to form winged morphs (Leonardo and Mondor 2006). Most pea aphid clones carry a secondary symbiont and multiple infections with several species in the same individual are thought to be rare (Chen and Purcell 1997; Sandström et al. 2001; Tsuchida et al. 2002). However, these estimates are typically based on surveys of only a small subset of the symbionts that are known now. Some symbiont species are significantly overrepresented in pea aphid clones collected from particular host plants. For example pea aphids adapted to Trifolium nearly always carry R. insecticola (Tsuchida et al. 2002; Leonardo and Muiru 2003; Simon et al. 2003; Ferrari et al. 2004; Frantz et al. 2009). There is contradictory evidence as to whether secondary symbionts affect performance on different host plants (Leonardo 2004; Tsuchida et al. 2004; Ferrari et al. 2007; McLean et al. 2011) and it is therefore unclear whether the nonrandom distribution of secondary symbionts across host-adapted populations is a cause or consequence of divergence. Predominantly vertical transmission within host-adapted populations would lead to similar patterns as selection for certain symbiont/host plant combinations.
In this study, we investigate the genetic structure of pea aphid populations, collected from eight plant genera in England and Germany, and survey their secondary symbionts. Specifically we ask whether pea aphid populations are genetically structured by host plant and geography and explore the relative importance of the two factors. We also ask whether there are any instances of the independent evolution of specialization on the same plant species. For the populations from England, data on host preference and performance are available from our earlier work (Ferrari et al. 2006, 2008). This has shown that some host plant associated populations are more specialized than others and we ask whether there is a correlation between phenotypic and genetic differentiation across populations. We also explore whether adaptive divergence in the performance on the different plant species exceeds the neutral molecular genetic differentiation. Our study extends recent genetic analyses by Peccoud et al. (2009a, b) of host-associated pea aphid populations and a detailed comparison is provided in the Discussion. We survey all the aphid clones for six secondary symbionts and ask whether the presence of different symbionts was influenced by collection plant or aphid genetic structure. We test whether infections with multiple species or specific combinations of symbionts occur less often than expected by chance. Finally, where secondary symbionts showed sufficient genetic variation and occurred in multiple specialized populations, we explore whether the genetic structure of the symbiont reflected that of their aphid hosts.
Methods
APHID COLLECTIONS
We collected pea aphids, Acyrthosiphon pisum (Harris), from plant species in seven genera of legume on which this insect feeds commonly in Northwest Europe: Lotus pedunculatus Cav., M. sativa L., T. pratense L., V. faba L., Lathyrus pratensis L., Pisum sativum L. and Cytisus scoparius (L.) Link. Pea aphids specialized on the latter two plant species are sometimes considered different subspecies, A. p. destructor (Johnson) and A. p. spartii (Koch), respectively. We also collected aphids from Ononis spinosa L. and O. repens L. which were much rarer in our study areas, but are host to the described subspecies A. p. ononis (Koch). With the exception of V. faba, all of these plant species are known to host a specialized population of pea aphids (Ferrari et al. 2008; Peccoud et al. 2009a). Aphids were sampled in Southern England and Germany from the same plant species (which we refer to as “collection plants” using just their generic names). In England, all collections were made from sites within a circle of 25-km radius centered at Silwood Park, Berkshire (51°9′30″N, 0°38′15″W), with the exception of five clones from Ononis that were collected from sites up to 55 km away from this area. The German aphid clones were collected in Lower Saxony, most from sites between Nordstemmen (52°9′30″N, 9°47′0″E) and Friedland (51°25′10″N, 9°55′0″E) within a 30-km-wide corridor along the river Leine. In Germany, all the aphids from Cytisus were sampled further north between Soltau (52°59′0″N, 9°50′30″E) and Hanover/Langenhagen (52°27′0″N, 9°44′30″E) due to the rarity of C. scoparius in the main study area. Three of the aphid clones from Pisum and two clones from Lotus also came from this area. In both countries, we collected 20–24 aphid genotypes from each plant genus, except for Ononis in England (nine clones: six from O. spinosa, three from O. repens) and Germany (four clones, all from O. spinosa), Lotus in Germany (nine clones) and Trifolium in Germany (30 clones). For each host plant/country combination aphids were collected from at least three different sites and, to reduce the probability of sampling the same aphid genotype repeatedly, individual clones were collected from plants separated by at least 30 m. In all, 297 clones of pea aphids were studied. Information about the performance of most of the English clones on different host plants was available from Ferrari et al. (2006, 2008). A clone was considered to be specialized on a particular plant species if it performed well on that species and relatively poorly on others (clones from each specialized population display a characteristic spectrum of performances on the panel of eight host plant species; Ferrari et al. 2008). However, it should be noted that some of these aphids might perform better on plant species not used in these studies.
MOLECULAR ANALYSIS
DNA was extracted from single aphid adults using the Qiagen DNeasy Blood and Tissue kit (Qiagen Ltd., Crawley, UK), following the manufacturer's instructions. We sequenced six nuclear markers (referred to as 380, 598, 650, 870, 901, 940) which were chosen because they were widely spaced on a QTL map of genes involved in host plant use by aphids feeding on T. pratense and M. sativa in the United States (Hawthorne and Via 2001). The primer sequences are available online (Table S1). The sequences were aligned using CodonCode Aligner version 1.6 (CodonCode Corporation). All markers contained heterozygous indels, which were also analyzed using CodonCode Aligner and afterwards checked manually.
Pea aphids are diploid and hence the gametic phase of multiple-site heterozygote sequences has to be estimated. We used a maximum likelihood method implemented in the program PHASE version 2.1 to infer the most likely alleles (Stephens et al. 2001; Stephens and Scheet 2005). Whenever heterozygous indels were present, the gametic phase could also be inferred from the electropherograms because peak shifts occur in a predictable manner. The alleles scored in this way agreed with those predicted by PHASE.
We also scored three microsatellite markers, S23, S24, S30, described in Wilson et al. (2004), with some modifications to their protocol (see Supporting information). Four pairs of clones (out of the 297 clones) had identical multilocus genotypes and were retained in the analysis.
SECONDARY SYMBIONTS
All 297 pea aphid clones were tested for the presence of facultative secondary endosymbionts. We initially screened the aphid clones by amplifying and partially sequencing the 16S rRNA gene (using the universal primers 10F and 35R; Sandström et al. 2001; Russell and Moran 2005). The sequences are available on GenBank (accession numbers JN398657–JN398666). This method is likely to miss multiple infections when several symbiont species occur in the same aphid individual. We therefore designed specific reverse primers for each of the symbiont species detected based on the sequences that we had obtained (details in the Supporting information, Table S2). Diagnostic PCRs for all symbionts were then performed using DNA from all clones with the amplicon run on a 2% agarose gel stained with ethidium bromide and visualized under a UV light. In addition, we also screened our samples for Rickettsia and Spiroplasma which have recently been shown to be common in pea aphids (Frantz et al. 2009) and which would not have been detected using the initial screen described above. All diagnostic PCRs were repeated twice to avoid the risk of false negatives. Because the DNA samples were the same samples as those used to score the aphid nuclear markers it is unlikely that absences occurred due to poor sample quality.
ANALYSIS
Population structure
Our basic dataset was generated from a collection of 297 pea aphid clones from 16 populations obtained from eight host plant genera at two locations. For each of the typically 20 aphid clones within each population, we have the two haplotype identities at the six sequenced loci and the size of the alleles at the three microsatellite loci.
Because there might be ongoing gene flow between the populations and because aphids may have been collected feeding on a plant that they were not adapted to, clustering methods were employed to look at population structure in more detail. We used a Bayesian model-clustering algorithm implemented in the program STRUCTURE version 2.2 (Pritchard et al. 2000; Falush et al. 2003). The model assumes that the overall population is divided into a number of clusters (K), which vary in allele frequency. Taking the multi-locus genotype of each individual into account, the model probabilistically assigns each individual to one or more clusters. To do this, it uses only allele frequency data and not information about collection plant or geography. The program can allow parts of an individual's genome to originate from more than one cluster (admixture) and we used this option which is recommended when the extent of gene flow and recombination is uncertain (Falush et al. 2003). Running the model without admixture gave very similar results (data not shown). Assigning individuals to clusters is a two-phase iterative process and 1,000,000 iterations were performed after a burn-in period of 100,000 iterations. This was repeated seven times for each K ranging from K= 1 to K= 20. We used a heuristic method based on a priori hypotheses about the structure of the population (see Results) to determine the number of clusters, but also employed the method of Evanno et al. (2005) combined with the common sense approach advocated in the STRUCTURE manual (Pritchard et al. 2007) which uses no a priori information. For some of the subsequent analyses, we assigned each aphid clone to the cluster to which STRUCTURE assigned the largest part of its genome assuming K= 9 (see the Results for justification of this partition) and we will refer to these groups as “genetic clusters.”
To determine whether any aphids did not belong to any of the genetic clusters detected in the STRUCTURE analysis, we conducted a factorial correspondence analysis implemented in the package Genetix 4.05 (Belkhir et al. 1996–2004), which is less sensitive to the number of individuals in a cluster. The results were visualized in three-dimensional space, using all possible combinations of the six main factors (an individual belonging to a cluster that was not detected in the STRUCTURE analysis should appear as an outlier). In addition, we estimated the likelihood that a genotype belonged to the cluster to which it was assigned to in the STRUCTURE analysis using Baudouin and Lebrun's criterion (Baudouin and Lebrun 2001) as implemented in the program GeneClass 2 (Piry et al. 2004).
Pairwise FST values between all 16 populations were estimated using FDIST2 (Beaumont and Nichols 1996). Pairwise FST values were also estimated between the genetic clusters identified in the STRUCTURE analysis described above. The contribution of collection plant and geography to the overall genetic variance was analyzed using hierarchical analysis of molecular variance (AMOVA), using the Arlequin version 3.11 package (Excoffier et al. 2005).
Comparison of molecular genetic and phenotypic divergence
Divergence of quantitative traits can be estimated by QST (Spitze 1993), the fraction of the additive genetic variance of a phenotypic trait accounted for by between-population effects. This can be approximated by the fraction of the variance among aphid clones in a trait that can be accounted for by the between-population term. There are statistical issues involving bias and accuracy in estimating QST from relatively small samples, as well as with the use of phenotypic surrogates for additive genetic variation (O’Hara and Merila 2005; Leinonen et al. 2008; Whitlock 2008). In the present case, we are estimating QST using clonal data and hence there is the possibility of the inclusion of nonadditive heritable components. We therefore exercise caution in interpreting our results.
Phenotypic data on several measures of the performance of most of the English clones on the full range of eight plant species (henceforth “test plants”) are available from Ferrari et al. (2008). We analyzed differences in the probability that a first instar aphid survived for six days on a particular test plant, the measure that gave the greatest resolution in distinguishing performance on different plant species. Aphids of about 18 clones from each of the eight collection plant genera were tested on all eight plant species to obtain a host plant-use profile. The proportion of the variance explained by the interaction between the collection plant and test plant terms provides an estimate of population differentiation in the performance profile that is interpreted against the background of the total residual variation among aphid clones (see Ferrari et al. 2008 for further details). A similar analysis can be performed to estimate the heritable differentiation in the pattern of host plant use among all possible pairs of genetic clusters. We carried out these analyses using a generalized linear modeling framework (assuming a quasibinomial error distribution) and estimated QST as DB/ (DB+ 2DW) where DB is the deviance explained by the genetic cluster × test plant interaction and DW is the clone × test plant deviance. For each pair of genetic clusters, we calculated three measures of QST which differed in the set of test plant species on which performance measures were included: (1) the complete set of the eight test plant species (QST all plants), (2) the two home plant species of the pair of genetic clusters under consideration (QST home plants), and (3) the six away plant species used by neither member of the pair of genetic clusters (QST away plants). Divergent selection and differentiation in performance is likely to be strongest for QST home plants and absent in QST away plants, with QST all plants intermediate.
Divergence at neutral markers (estimated by FST) can be compared with divergence in performance (estimated by QST) to provide an indication as to whether divergent selection is occurring (see discussion in McKay and Latta 2002; Leinonen et al. 2008). Because we expect divergent selection and hence phenotypic differentiation to be greatest when performance on the two home plants of a pair of genetic clusters are used to calculate QST, we predicted that QST home plants is greater than FST. If only away plants are used, differentiation in performance should be equal to neutral divergence (QST away plants=FST). The comparison involving QST all plants is predicted to be intermediate. QST and FST values were compared with paired t-tests. We then tested whether there was a significant correlation between genotypic differentiation (FST) and the measure of differentiation that best describes the overall divergence in performance between genetic clusters (QST all plants) using a Spearman rank correlation.
Population genetic structure of secondary symbionts
Where the same species of secondary symbiont occurred frequently in several host-specialized populations and where there was sufficient sequence variation in the symbiont, we asked whether the genetic structure of the symbiont reflected that of their aphid hosts. Such parallel differentiation might reflect limits to the horizontal transmission of symbionts between aphids on different host plants. We employed the TCS version 1.21 software, which uses parsimony (Clement et al. 2000), to construct a gene genealogy of the partially sequenced 16S rRNA gene (using the sequences obtained in the initial screening). We then tested whether these symbiont haplotypes were distributed nonrandomly across the aphid genetic clusters using multinomial regression. This analysis assumes that the infection status is independent between individuals.
Results
PEA APHID POPULATION STRUCTURE
We sampled aphids from eight plant genera in two different geographic localities and hence a priori one might expect the assemblage of aphids to be structured into two (geography only), eight (collection plant species only), or 16 (both) clusters. We explored this heuristically by using the Bayesian cluster-fitting algorithm assuming K= 2, 8, or 16 clusters. When only two clusters were allowed, these aligned with groups of collection plants rather than geography, while assuming 16 clusters gave rise to a complex picture with a few clusters representing particular geography/host plant combinations but with many individuals assigned to multiple clusters. Assuming eight clusters led to a series of groupings, most but not all of which correspond to collections from particular host plants that span the two sampling sites. Eight clusters thus provided the most meaningful partition of the three a priori hypotheses.
However, the precise identity of the eight clusters varied among replicate runs, suggesting that a greater number of clusters might be more appropriate. Increasing the number of clusters to nine, gave a stable partition that was consistent across replicates (Fig. 1) and increased the posterior probability of the partition (see Fig. S1A for posterior probabilities of all K). Increasing the number of clusters to more than nine led to the identification of further substructure in some clusters. For K= 10 or 11, these could not be interpreted in terms of host plant or geography. Assuming 12 clusters (which had the highest unpenalized likelihood, although only just higher than K= 9), the pea aphid clones associated with Trifolium showed some geographic substructure (light and dark orange bars in Fig. S2A).
Genetic structure of pea aphids in England and Germany on eight plant genera. Nine genetic clusters were found using Bayesian cluster analysis. Each aphid clone is denoted by a vertical bar, whose colors represent the partition of the clone's ancestry to each of the inferred genetic clusters. Genotypes are sorted by the plant genus from which they have been collected and by location (GB: England, DE: Germany). Populations are separated by a thin black line.
These analyses demonstrate that host plant is the main and geography is a secondary structuring factor in these populations. Our sampling of aphids did not aim at detecting any further small-scale geographic or other substructure and therefore we chose the partition with nine clusters as the most informative. All nine clusters were associated with a specific plant species. In two cases, a pair of genetic clusters was associated with the same species of plant: on Pisum, where one was largely found in Germany (gray bars in Fig. 1) and the other mostly in England (purple bars) and on Lathyrus, where the more common cluster was found in both countries (henceforth Lathyrus 1, light green bars) and the other was restricted to England (Lathyrus 2, yellow bars). Vicia was associated with no distinct cluster and appeared to be largely fed upon by aphids from the Pisum and Trifolium (orange bars) clusters. Vicia is a special case in that it is a permissive host for most pea aphid clones, and we explore this further in the Discussion. Some individuals belonged to clusters that were not typical for their collection plant, which we discuss further below.
We used Evanno et al.'s (2005) method to determine the number of clusters in STRUCTURE without any a priori assumptions. This identified either two or four clusters (Figs. S1B, S2B), but where two clusters were found their composition varied between runs. With four clusters, the partition was stable and consisted of: (1) the Lathyrus 1 cluster, (2) the German and English Pisum clusters, (3) the Cytisus, Ononis and Lathyrus 2 clusters, and finally (4) the Lotus, Medicago, and Trifolium clusters. Applying Evanno et al.'s method to these four groups of aphid clones separately produced a pattern similar to Figure 1, suggesting hierarchical structuring (Fig. S3). In addition, this analysis revealed geographic structure in the Cytisus and Trifolium clusters (Fig. S4A, B), but the division of the German and English Pisum clusters was less well supported (Fig. S4C).
With our chosen partition of K= 9 (Fig. 1) an average of 82% of an individual's “ancestry” as estimated by STRUCTURE could be associated with one particular cluster. For 42% of the individuals this figure was more than 90% and for 93% of the individuals it was more than 50%. These figures not only provide evidence for considerable structuring in the population, but also show that the reproductive isolation between clusters is incomplete, which suggests ongoing gene flow.
It is possible that some of the aphid clones belong to genetic clusters that are associated with plant species from which we did not sample and such clusters might not have been detected in the analysis above. To test this, we conducted a factorial correspondence analysis and found that all the host-associated clusters detected in the STRUCTURE analysis could be distinguished and there was no suggestion that that any of the pea aphid clones did not belong to the nine STRUCTURE clusters.
GENETIC DIFFERENTIATION
A hierarchical AMOVA was carried out with the host plant from which the aphids were collected and collection site (Germany/England) as the two explanatory factors. The STRUCTURE analysis described above suggests that collection plant is the main determinant of population structure and we therefore chose a hierarchical model with collection site nested within collection plant. The effect of collection plant was significant (P < 0.001) and explained 12.2% of the molecular genetic variance at these loci while collection site explained a further 3.7% of this variance and was also significant (P < 0.001). Geographic structure is thus detectable within host-associated groups.
The average FST between collections from different host plants (combining data from all nine loci and both countries) was 0.15 and ranged from 0.02 to 0.35. Figure 2A shows the pattern of pairwise differentiation among the eight host-associated populations as measured by FST (see Table S3 for a matrix of FST values). The population from Cytisus was on average the most distinct whereas the populations from Lotus, Trifolium, and Vicia were the least differentiated. Generally, FST for populations collected on the same host plant in different countries was low (ranging from 0.03 to 0.11), with the highest value that between the German and English populations on Cytisus.
Genetic differentiation (measured by FST) (A) of pea aphid populations collected from eight plant genera in England and Germany and (B) of nine genetic clusters inferred by the Bayesian clustering analysis. The thickness of the lines connecting the populations represents the FST value.
These values of FST may be influenced by the presence of vagrants, clones found on a plant species to which they are not adapted. We explored the possible effects of vagrants by calculating the equivalent FST network for clones placed within the genetic clusters identified by STRUCTURE with K= 9 (Fig. 2B). Overall the values of FST were higher, ranging from 0.08 to 0.42 with a mean of 0.23, a simple consequence of the clusters being defined by the same data used to calculate FST. The Lathyrus 1 cluster was on average the most differentiated, followed by Cytisus. Interestingly, the higher-level clusters found by STRUCTURE with K= 4 (Lathyrus 1; German and English Pisum; Cytisus, Ononis and Lathyrus 2; Lotus, Medicago and Trifolium) were also apparent as groups with relatively low internal FST. Pairwise FST values were lower between the two Pisum clusters and within the Lotus/Medicago/Trifolium group than they were within the Cytisus/Ononis/Lathyrus 2 group. There was little differentiation between aphids from England and Germany that were assigned to the same genetic cluster (FST from 0.01 to 0.11). Similar results were obtained when using a smaller dataset with only aphid clones where the phenotype matched the collection plant (Table S3).
We estimated the degree of differentiation in aphid performance (juvenile survival) on all eight test plant species (QST all plants) between each pair of the nine genetic clusters (Table 1). We then tested whether there was a correlation between this measure of the differentiation in performance and genetic differentiation (FST) and found it was not significant (Spearman rank correlation coefficient = 0.23, P= 0.24).
Differentiation in juvenile survival on all eight test plant species (QST all plants) between aphids assigned to nine different genetic clusters.
| Lathyrus 1 | Cytisus | Ononis | Lathyrus 2 | Lotus | Medicago | Trifolium | |
|---|---|---|---|---|---|---|---|
| Lathyrus 1 | |||||||
| Cytisus | 0.17 | ||||||
| Ononis | 0.33 | 0.31 | |||||
| Lathyrus 2 | 0.20 | 0.21 | 0.28 | ||||
| Lotus | 0.25 | 0.28 | 0.32 | 0.16 | |||
| Medicago | 0.23 | 0.27 | 0.31 | 0.04 | 0.16 | ||
| Trifolium | 0.17 | 0.23 | 0.25 | 0.09 | 0.14 | 0.10 | |
| Pisum GB | 0.18 | 0.28 | 0.31 | 0.07 | 0.22 | 0.11 | 0.07 |
| Lathyrus 1 | Cytisus | Ononis | Lathyrus 2 | Lotus | Medicago | Trifolium | |
|---|---|---|---|---|---|---|---|
| Lathyrus 1 | |||||||
| Cytisus | 0.17 | ||||||
| Ononis | 0.33 | 0.31 | |||||
| Lathyrus 2 | 0.20 | 0.21 | 0.28 | ||||
| Lotus | 0.25 | 0.28 | 0.32 | 0.16 | |||
| Medicago | 0.23 | 0.27 | 0.31 | 0.04 | 0.16 | ||
| Trifolium | 0.17 | 0.23 | 0.25 | 0.09 | 0.14 | 0.10 | |
| Pisum GB | 0.18 | 0.28 | 0.31 | 0.07 | 0.22 | 0.11 | 0.07 |
Differentiation in juvenile survival on all eight test plant species (QST all plants) between aphids assigned to nine different genetic clusters.
| Lathyrus 1 | Cytisus | Ononis | Lathyrus 2 | Lotus | Medicago | Trifolium | |
|---|---|---|---|---|---|---|---|
| Lathyrus 1 | |||||||
| Cytisus | 0.17 | ||||||
| Ononis | 0.33 | 0.31 | |||||
| Lathyrus 2 | 0.20 | 0.21 | 0.28 | ||||
| Lotus | 0.25 | 0.28 | 0.32 | 0.16 | |||
| Medicago | 0.23 | 0.27 | 0.31 | 0.04 | 0.16 | ||
| Trifolium | 0.17 | 0.23 | 0.25 | 0.09 | 0.14 | 0.10 | |
| Pisum GB | 0.18 | 0.28 | 0.31 | 0.07 | 0.22 | 0.11 | 0.07 |
| Lathyrus 1 | Cytisus | Ononis | Lathyrus 2 | Lotus | Medicago | Trifolium | |
|---|---|---|---|---|---|---|---|
| Lathyrus 1 | |||||||
| Cytisus | 0.17 | ||||||
| Ononis | 0.33 | 0.31 | |||||
| Lathyrus 2 | 0.20 | 0.21 | 0.28 | ||||
| Lotus | 0.25 | 0.28 | 0.32 | 0.16 | |||
| Medicago | 0.23 | 0.27 | 0.31 | 0.04 | 0.16 | ||
| Trifolium | 0.17 | 0.23 | 0.25 | 0.09 | 0.14 | 0.10 | |
| Pisum GB | 0.18 | 0.28 | 0.31 | 0.07 | 0.22 | 0.11 | 0.07 |
SELECTION ON DIFFERENT PLANT SPECIES
As described in the methods QST can be compared to FST to infer whether divergent selection on certain traits may have occurred. We compared the pairwise differentiation in performance on different sets of test plant species (measured by QST, see Table S4 for a matrix of QST home plants and QST away plants values) with the molecular genetic differentiation (FST). As predicted, QST home plants was on average higher than FST (average QST home plants= 0.48, FST= 0.25, t26= 6.35, P < 0.001). Both, QST all plants (average 0.20) and QST away plants (0.07) were on average lower than FST (t27= 2.55, P= 0.02; t27= 12.30, P < 0.001).
DO THE GENETIC CLUSTERS PREDICT SPECIALIZATION?
For most of the English aphid clones, we have three pieces of information: the plant species from which the clone was collected, the plant species on which it is specialized, and the host-associated genetic cluster to which it was assigned above. We attempted to obtain full data for the 135 English clones but excluded seven because of specific technical problems obtaining one of the pieces of information, leaving 128. For the vast majority of these aphid clones (112; 88%), the plant that they are specialized on matched the genetic cluster they were assigned to in the STRUCTURE analysis. In most cases (88 clones; 69%), this was also the plant species on which they had been collected, giving clear evidence of specialization. Of the 24 clones (19%), where host plant performance and genetic cluster matched but the aphid was collected from a different plant species, 14 were collected from Vicia, which has no associated genetic cluster and appears to be an exceptional plant on which most pea aphid clones do well. This leaves 10 clones that may be true vagrants collected on a plant to which they are not adapted: five were assigned to one of the two Pisum clusters and four were collected from Lathyrus (Table S5). None of these clones was assigned to the genetic clusters that were identified as the most genetically distinct in the previous section (Cytisus or Lathyrus 1), suggesting higher host fidelity of these aphids.
For the remaining 16 clones (13%), the plant species on which the clone was specialized did not match the genetic cluster to which it was assigned (see also Table S5). This category of clones was most common in certain genetic clusters: Lathyrus 2 (three out of 10 clones in this cluster, all of which appeared to be Pisum specialists), Lotus (five out of 21 clones in this cluster), and Trifolium (four out of 21 clones in this cluster, all specialized on Medicago). In several of these cases, the cluster analysis showed substantial mixed ancestry, which might indicate recent introgression or reflect limits to our power to resolve population structure with the genetic markers we employed. We analyzed whether any of these 16 clones had an unusually low likelihood of belonging to the cluster to which they were assigned using an assignment test implemented in the program GeneClass 2 (Fig. S5). This was the case for most of the aphid clones with an atypical phenotype in the Ononis cluster (two clones that were collected from and performed best on Trifolium) and in the Lotus cluster, but not for the clones in the Trifolium, Medicago or Lathyrus 2 clusters. It is possible that some of these aphid clones may belong to genetic clusters that are associated with plant species we did not sample.
SECONDARY SYMBIONTS
Distribution of secondary symbionts
We detected the facultative aphid symbiotic bacteria H. defensa, R. insecticola, S. symbiotica, a Rickettsia and a Spiroplasma, as well as a sixth bacterium that has recently been referred to as the X-type or PAXS (for Pea Aphid X-type Symbiont; Guay et al. 2009). Overall 26%, 12%, 41%, 23%, 27%, and 16% of aphid clones were infected by these six bacteria, respectively, whereas 9% of aphid clones carried none of these facultative symbionts (these frequencies add up to more than 100% because of multiple infections of many aphid clones, see below). The bacteria were distributed nonrandomly across populations from different collection plants, and the pattern was largely consistent across the two countries (Fig. 3A, Table 2). The most common associations were: H. defensa with aphids collected from Lotus, Ononis and Medicago; R. insecticola with aphids collected from Trifolium; S. symbiotica with aphids from Cytisus, Pisum and Vicia; and the X-type with aphids from Vicia, Medicago, and Cytisus. The association of the Rickettsia and the Spiroplasma with aphids collected from different plant species was less pronounced, but the Rickettsia was most common in aphids collected from Lotus, whereas the Spiroplasma occurred most frequently in aphids from Medicago, Trifolium, or Vicia. Pea aphid clones from Lathyrus had low frequencies of these secondary symbionts and 33% of clones collected from this species harbored none.
The frequency of the six facultative symbiont species in different host-associated pea aphid populations. Frequencies in (A) all 297 aphid clones collected from eight plant genera in England and Germany, (B) in clones collected from Lathyrus pratensis belonging to the two Lathyrus or the Pisum clusters, (C) in clones collected from Trifolium pratense and assigned to the Trifolium or Pisum clusters, (D) in clones collected from Vicia faba and also assigned to the Trifolium or Pisum clusters, and (E) in clones collected from Pisum sativum and assigned to each Pisum cluster. The number of aphid clones in a group (N) is given at the top of each panel.
The relationship between collection plant, pea aphid genetic cluster and collection site (Germany, England) and the distribution of secondary symbionts. The table shows the percentage of the variation that could be explained by the three factors entered in the order shown in a generalized linear model binary regression on symbiont presence or absence (significance levels:***P < 0.001, **P < 0.01, *P < 0.05). (A) Collection plant was fitted first. (B) Genetic cluster was fitted first.
| Explanatory variable | Distribution of | Symbionts/clone | ||||||
| No symbiont | H. defensa | R. insecticola | S. symbiotica | X-type | Rickettsia | Spiroplasma | ||
| A | ||||||||
| Collection Plant | 14*** | 32*** | 34*** | 44*** | 20*** | 5* | 3 | 14*** |
| Genetic Cluster | 14** | 16*** | 15*** | 19*** | 17*** | 4 | 6** | 7*** |
| Country | 0 | 0 | 1 | 0 | 2* | 6*** | 0 | 2** |
| B | ||||||||
| Genetic Cluster | 25*** | 41*** | 29*** | 56*** | 30*** | 8** | 5* | 19*** |
| Collection Plant | 3 | 7** | 20*** | 7*** | 8** | 1 | 4* | 3 |
| Country | 0 | 0 | 1 | 0 | 2* | 6*** | 0 | 2** |
| Explanatory variable | Distribution of | Symbionts/clone | ||||||
| No symbiont | H. defensa | R. insecticola | S. symbiotica | X-type | Rickettsia | Spiroplasma | ||
| A | ||||||||
| Collection Plant | 14*** | 32*** | 34*** | 44*** | 20*** | 5* | 3 | 14*** |
| Genetic Cluster | 14** | 16*** | 15*** | 19*** | 17*** | 4 | 6** | 7*** |
| Country | 0 | 0 | 1 | 0 | 2* | 6*** | 0 | 2** |
| B | ||||||||
| Genetic Cluster | 25*** | 41*** | 29*** | 56*** | 30*** | 8** | 5* | 19*** |
| Collection Plant | 3 | 7** | 20*** | 7*** | 8** | 1 | 4* | 3 |
| Country | 0 | 0 | 1 | 0 | 2* | 6*** | 0 | 2** |
The relationship between collection plant, pea aphid genetic cluster and collection site (Germany, England) and the distribution of secondary symbionts. The table shows the percentage of the variation that could be explained by the three factors entered in the order shown in a generalized linear model binary regression on symbiont presence or absence (significance levels:***P < 0.001, **P < 0.01, *P < 0.05). (A) Collection plant was fitted first. (B) Genetic cluster was fitted first.
| Explanatory variable | Distribution of | Symbionts/clone | ||||||
| No symbiont | H. defensa | R. insecticola | S. symbiotica | X-type | Rickettsia | Spiroplasma | ||
| A | ||||||||
| Collection Plant | 14*** | 32*** | 34*** | 44*** | 20*** | 5* | 3 | 14*** |
| Genetic Cluster | 14** | 16*** | 15*** | 19*** | 17*** | 4 | 6** | 7*** |
| Country | 0 | 0 | 1 | 0 | 2* | 6*** | 0 | 2** |
| B | ||||||||
| Genetic Cluster | 25*** | 41*** | 29*** | 56*** | 30*** | 8** | 5* | 19*** |
| Collection Plant | 3 | 7** | 20*** | 7*** | 8** | 1 | 4* | 3 |
| Country | 0 | 0 | 1 | 0 | 2* | 6*** | 0 | 2** |
| Explanatory variable | Distribution of | Symbionts/clone | ||||||
| No symbiont | H. defensa | R. insecticola | S. symbiotica | X-type | Rickettsia | Spiroplasma | ||
| A | ||||||||
| Collection Plant | 14*** | 32*** | 34*** | 44*** | 20*** | 5* | 3 | 14*** |
| Genetic Cluster | 14** | 16*** | 15*** | 19*** | 17*** | 4 | 6** | 7*** |
| Country | 0 | 0 | 1 | 0 | 2* | 6*** | 0 | 2** |
| B | ||||||||
| Genetic Cluster | 25*** | 41*** | 29*** | 56*** | 30*** | 8** | 5* | 19*** |
| Collection Plant | 3 | 7** | 20*** | 7*** | 8** | 1 | 4* | 3 |
| Country | 0 | 0 | 1 | 0 | 2* | 6*** | 0 | 2** |
We analyzed the distribution of each secondary symbiont (and the absence of these six symbionts) across aphid clones using individual binary logistic regressions (we did not use multinomial analysis because of multiple infections). Collection plant was a highly significant explanatory variable in six of the seven cases (Table 2A; not significant for the Spiroplasma), and except for the Rickettsia the explanatory power of the model was significantly increased by adding aphid genetic cluster to the analyses. The further addition of collection site (country) was significant in only two cases (the X-type and Rickettsia) and here it only explained a small additional amount of variation in the data. Collection plant and genetic cluster are highly correlated although the latter explains more variation in the data (except for Regiella; Table 2B), indicating that it is the more reliable predictor of symbiont identity.
Inspection of the data revealed why consideration of both collection plant and genetic cluster was informative. On Lathyrus, there are two genetic clusters that differ in symbiont frequency with Lathyrus 1 having particularly low rates of infection (50%; Fig. 3B). Both Pisum genetic clusters had high frequencies of S. symbiotica, but this bacterium occurred less frequently in the clones assigned to these clusters that were collected from plants other than Pisum (Fig. 3B–E; χ2= 12.77, P < 0.001). As mentioned above, most of the aphid clones collected from Vicia were genetically assigned to the Trifolium or Pisum clusters and based on this they would be expected to carry either R. insecticola or S. symbiotica as the most frequent symbiont. In fact, the Pisum genotypes on Vicia nearly all harbored S. symbiotica (20/25; Fig. 3D) whereas those assigned to the Trifolium cluster only rarely hosted R. insecticola (3/12) and instead often had the X-type (8/12) and/or Spiroplasma (8/12). A further exploration of these patterns can be found in the Supporting information.
Aphid clones on average carried 1.4 species of secondary symbionts though with significant variation among collection plants, genetic clusters and location (Fig. 4, Table 2). To analyze this further we treated the distribution of individual symbiont species across the different host-associated populations as fixed and looked for patterns in co-occurrence that are not explained by the marginal frequencies. The average number of double infections was as expected but overall infections with three or more species were less common than predicted (χ23= 53.0, P < 0.001). A few combinations of symbionts occurred less frequently than expected (analyses were restricted to cases where the expected numbers of double infection was at least 4.5 and a Bonferroni correction was applied; see Tables S6, S7): H. defensa and R. insecticola, and R. insecticola and the X-type occurred rarely together in aphids in the Trifolium cluster, and S. insecticola and the X-type occurred less often than predicted in aphids collected from Vicia. The deficit of the latter two double infections was due to the different frequencies of R. insecticola and the X-type in aphids assigned to the Trifolium cluster, but collected from Vicia or Trifolium, as described above (Table S7, see also Fig. 3C–D).
The number of secondary symbiont species per aphid clone (mean ± SE) in aphids assigned to different genetic clusters. Striped bars show the observed number and solid bars the number expected based on the frequencies of different symbiont species in the different aphid populations.
Genetic population structure of secondary symbionts
We asked whether the genetic structure of symbiont species that occurred in multiple host-associated populations was similar to that of their aphid hosts. This analysis was only possible for H. defensa because there was no 16S rRNA sequence variation among isolates of S. symbiotica or the X-type. We identified three haplotypes of this gene for R. insecticola, but the low frequency of this species in aphids not associated with Trifolium did not allow a more detailed analysis. Sequence data were not available for the Rickettsia or the Spiroplasma.
We found five haplotypes of the 16S rRNA gene in H. defensa which were distributed nonrandomly across aphid genetic clusters (χ2= 90.1, df = 20, P < 0.001). The most divergent haplotype (number 5 in Fig. 5) occurred only in pea aphids in the Ononis genetic cluster (which is the most divergent of the aphid clusters in which H. defensa occurred), whereas the remaining four haplotypes had clear, but nonexclusive, associations with aphids in the Medicago and Trifolium clusters or in the Lotus cluster.
Genealogy of the 16S rRNA gene in Hamiltonella defensa. Each circle represents a haplotype of this gene (labeled 1–5). The diameter of the circle corresponds to the frequency of this haplotype in all samples. Samples from different aphid genetic clusters are represented by different colors, following the same color scheme as in Figure 1.
Discussion
We begin by briefly summarizing our main findings. We found strong evidence that most pea aphid populations collected from eight different legume genera in the United Kingdom and Germany are genetically differentiated to varying degrees. Geography was a less important factor in structuring these populations, but aphids associated with Pisum, Trifolium, Cytisus, and possibly Lathyrus showed some differentiation. The best description of the data using cluster analysis placed the aphid clones in nine groups that corresponded to the eight host plants with Pisum and Lathyrus each having two associated clusters, and the absence of a cluster restricted to the near-universally permissible host V. faba. We had previously collected performance data on aphids from the English clones on all host plants and found most aphid clones were both collected from the host plant on which they were specialized and assigned to that host plant's genetic cluster, but there were exceptions due to vagrancy and other reasons that we discuss. Molecular genetic differentiation was not significantly correlated with the heritable differentiation in the performance across all eight test plant species. Using comparisons of molecular differentiation and differentiation in performance, there was evidence for divergent selection on performance on the home plants, but not on any other plant species. Our analysis of the secondary symbiont communities in the different aphid populations confirms previous associations between certain bacteria and host plant specialized aphid populations and reveals new ones. It also shows that the genetic structure of H. defensa is similar but not identical to that of their aphid hosts. Unexpectedly, we found examples of aphids from the same genetic cluster that tend to have different symbiont species when they were collected from different plant species. Furthermore, these symbionts were not necessarily the species most common in aphids associated with these plant species.
Recently, Peccoud et al. (2009a) published a genetic study of host-associated pea aphid populations in Germany and France, which included all of the plant species we studied as well as some additional ones. In a companion study, Peccoud et al. (2009b) used the purely vertically transmitted primary symbiont B. aphidicola to construct a phylogeny of pea aphid matrilines from different host-associated populations. For several of the host-associated populations distinct monophyletic matrilines were found, but for others the separation was less clear. Like us they found clear evidence for host-associated genetic differentiation with little geographical variation. The main differences were that Peccoud et al. (2009a) detected additional genetic clusters on plant species that we did not survey, whereas we found a second host-associated genetic cluster on Lathyrus as well as some geographic structure not shown in their study. Considering just the host plants included in both studies there was a significant correlation between the two sets of FST estimates (r19= 0.69; P < 0.001). The phylogeny of matrilines (Peccoud et al. 2009b) is in broad agreement with the pairwise degree of differentiation between host-adapted genetic clusters. In particular, the distinctiveness of the Lathyrus 1 and Ononis clusters is well supported.
Taken together these studies demonstrate extensive genetic differentiation associated with host plant specialization in Northwest European populations of pea aphids. In some cases genetic clusters are associated with all the species in a genus that have been examined whereas in other cases they are restricted to a subset of species (Peccoud et al. 2009a). Separate and highly differentiated genetic clusters on the same plant species in the two countries would have indicated independent adaptation to the same plant species. However, for many adapted populations, the degree of geographical variation was small and where it was found, FST values between these clusters were low. This suggests that adaptation to a specific host has occurred only once. The correlation between the FST values reported here and in Peccoud et al.'s study (2009a) suggests the same thing, although it is also possible that specialization evolved more frequently with homogenization occurring through gene flow. It is unclear why geographic genetic structuring is stronger on some plant species because the same barriers to gene flow would be expected to act on all populations. Other factors might have affected geographic structuring, such as different divergence times between geographic populations or different population sizes.
An exception to the pattern of one cluster per host plant species were the two clusters we call Lathyrus 1 and Lathyrus 2 that occurred together on L. pratensis. In the United Kingdom, these often were found at the same locality, whereas Lathyrus 1 alone occurred in Germany (and we believe that this is also the cluster found on this host plant by Peccoud et al. (2009a)). The two clusters are genetically very distinct (and indeed Lathyrus 1 diverges markedly from all other clusters) although both perform well on this plant species. Lathyrus 1 is also unusual in its poor performance on V. faba (data not shown) as well as in having low frequencies of the six secondary symbionts (Lathyrus 2 also has a relatively low incidence but not as extreme). There must be a nongeographical barrier to gene flow; it would therefore be interesting to conduct mating experiments to see whether there are behavioral isolating mechanisms separating the two clusters that co-occur and presumably breed on the same plant species. It is possible that the Lathyrus 2 cluster is actually a genetic cluster that is more strongly associated with a plant we did not sample. Possible candidates would be the genetic clusters associated with Vicia cracca or Medicago lupulina, which contain aphids that perform relatively well on Lathyrus (Peccoud et al. 2009a).
Vicia faba has repeatedly been shown to be a suitable host plant for almost all pea aphid clones (Sandström and Pettersson 1994; Ferrari et al. 2008) and there appears to be no population uniquely specialized on this species and unable to feed on other plant species (Ferrari et al. 2008). Both, we (sampling V. faba) and Peccoud et al. (2008, 2009a; sampling V. faba, V. hirsuta and V. sativa) found no genetic cluster associated with these vetches, with most collections being dominated by aphid clones from the cluster associated with Pisum whereas clones in the Trifolium cluster were most common on Vicia in England. It has been suggested that V. faba and related species may act as a bridge for gene flow between different host-associated pea aphid populations that would otherwise not encounter each other (Ferrari et al. 2008). The genetic data confirm this is a possibility and also suggests that the bridge might only operate between a subset of the host plant associated populations.
Approximately 7% of the aphid clones we sampled were collected feeding on host plants to which they were not well adapted. Pisum specialists in particular were frequently recorded as vagrants, a pattern also found by Peccoud et al. (2009a). Laboratory data suggest that vagrants usually perform poorly on these alternative plants (Ferrari et al. 2008) and it is unlikely that most will form clonal lineages that persist long enough to produce sexuals and reproduce with locally adapted genotypes in the autumn. Here, we also found cases in which a clone was clearly adapted to a host plant yet the genetic analysis assigned it to a cluster associated with a different species (e.g., we recorded several clones placed in the Trifolium cluster that were specialized on Medicago). This occurred most often among clusters that were relatively weakly differentiated. It is possible that these individuals are the result of ongoing gene flow. Alternatively, our panel of genetic markers may not always be sufficient to assign all clones to the correct cluster. We explored this further by testing the robustness of assignments in the STRUCTURE analysis by removing one or two markers (analysis not shown). These analyses were generally consistent with the initial analysis, but the removal of certain combinations of markers occasionally merged closely related clusters or changed the assignment of clones between these clusters.
QST is a measure of differentiation based on quantitative genetic traits that we estimated using phenotypic data on aphid clone performance across three different sets of the test plant species: the full range of test plant species, only the home plants for each pair of genetic clusters, and only the away plants for each pair of clusters. As mentioned in the Methods, there are possible statistical biases in calculating QST from small samples or from clonal data and hence the results have to be considered with some caution. However, as predicted, the differentiation in performance on the home plant was on average greater than the molecular genetic differentiation detected using probably neutral markers, indicating divergent selection for performance on these plants. In contrast, the molecular genetic variation was higher than the differentiation in performance on the away plants, indicating the absence of divergent selection on these traits. The genetic clusters showed differing degrees of pairwise differentiation in QST all plants but this did not correlate significantly with FST. Similarly, Peccoud et al. (2009a) did not find a correlation between genotypic differentiation and differentiation in performance, using a slightly different measure of trait variation that assessed performance only on home plant species. This lack of a correlation and the higher differentiation in home plant use compared to neutral-marker genetic differentiation indicate that gene flow between the populations has not prevented a degree of adaptive divergence.
ASSOCIATIONS BETWEEN SECONDARY SYMBIONTS AND SPECIALIZATION
We found a series of new associations between symbiont prevalence and host plant (Fig. 3) and observed an anomalously low frequency of any of the surveyed secondary symbionts in aphids specialized on Lathyrus. Further examples of nonrandom associations that have been recorded previously in pea aphids are confirmed here: aphids from Trifolium tend to carry R. insecticola, aphids from Pisum tend to harbor S. symbiotica and those from Medicago often have H. defensa (Tsuchida et al. 2002; Leonardo and Muiru 2003; Simon et al. 2003; Ferrari et al. 2004; Frantz et al. 2009). More surprisingly we found that a further secondary symbiont, provisionally referred to as the X-type (a member of the Enterobacteriaceae, unpublished data), is widespread and common. Separately we will report more fully on this bacterium but we have found that it is stably maintained in laboratory pea aphid clones for several years suggesting that it has a similar biology to the better known species. The same bacterium (with 98–100% similarity in the 16S rRNA gene) has recently been found in a small number of pea aphid clones in North America and in two other aphid species, Cinara juniperi and Maculolachnus submacula (Lamelas et al. 2008; Guay et al. 2009).
We found parallels between the genetic structure of H. defensa and that of their pea aphid hosts. Thus H. defensa in aphids from the highly differentiated Ononis cluster were relatively distinct compared with other isolates, whereas aphids from populations adapted to Lotus and Medicago tended to carry distinct strains of H. defensa. It is unlikely that this differentiation has evolved wholly within the pea aphid complex, because 16S rRNA is a slowly evolving gene and the host-associated populations have evolved relatively recently (Peccoud et al. 2009b). The different strains of H. defensa have probably therefore been acquired independently from other aphid species by different pea aphid matrilines (Russell et al. 2003). The imperfect association between H. defensa strains and different host-associated population is nevertheless consistent with predominantly vertical transmission of the symbiont within host-adapted populations, but suggests some transfer between them, which might occur, for example, through matings between aphids adapted to different plants (Moran and Dunbar 2006).
A vagrant aphid found on a host plant that is not typical for the genetic cluster to which it is assigned is nonetheless expected to carry a symbiont associated with its particular genetic cluster. Curiously, we found exceptions to this. Aphids collected from Trifolium and assigned to the Trifolium cluster almost always carry R. insecticola but when genetically very similar aphid clones are collected on Vicia they seldom bore this species but were more likely to have the X-type. A related but weaker pattern was found among aphids in the Pisum clusters that appeared to lose S. symbiotica when feeding on other plant species. These groups of aphids did not differ in their performance on the relevant plant species, nor were the aphids with atypical symbionts genetically differentiated from aphids with typical symbionts in the same cluster. We therefore do not know whether these differences are due to selection on symbiont composition on different host plants, or whether the symbionts are indicating more subtle patterns of genetic differentiation than those revealed by our nuclear genetic markers. Horizontal transfer is thought to be very rare and is unlikely to explain this pattern because other aphids found on Vicia do not carry high frequencies of the X-type.
The nonrandom distribution of symbiont species across host-associated populations may be due to historical accidents if colonization of a new host plant involved a small number of individuals that just happen to harbor particular species and if certain symbionts were by chance associated with nuclear genotypes under positive selection and hence hitchhiked to high frequency. This hypothesis is consistent with the relatively low maternal diversity found between host-adapted populations in Peccoud et al.'s study (2009b). Alternatively, there may be selection acting directly on certain symbiont/host plant combinations. Several studies have explored experimentally whether the presence of a symbiont may increase aphid performance on certain host plants. R. insecticola is the species that is found very frequently on Trifolium and one study suggested that it improved aphid performance on this host plant (Tsuchida et al. 2004). However, it involved a single aphid clone and subsequent studies found no clear pattern, as well as significant variation among clones (Leonardo 2004; Tsuchida et al. 2004; Ferrari et al. 2007; McLean et al. 2011). Experimental studies of other symbionts on different host plants also do not support a major role for these bacteria in host plant adaptation (Leonardo 2004; McLean et al. 2011). Symbionts are known to affect other aspects of aphid biology including defense against natural enemies (Oliver et al. 2003, 2005, 2009; Ferrari et al. 2004; Scarborough et al. 2005; Vorburger et al. 2009) and response to environmental extremes (Montllor et al. 2002) and it is possible that the value of these traits is influenced by host plant species. To decide among these different hypotheses, we need more data on the eco-evolutionary dynamics of the symbionts, especially rates of horizontal transfer and lineage extinction, as well as a better understanding of the selection pressures experienced by pea aphids on different plant species. However, with the available data, it appears that the distribution of secondary symbionts in the pea aphid complex is more likely a consequence of the divergence process than a cause of it.
Infections with multiple secondary symbionts have been thought to be relatively rare in pea aphids (Chen and Purcell 1997; Sandström et al. 2001; Tsuchida et al. 2002, but see Frantz et al. 2009). An explanation for this lack of multiple infections has been that these infections may be costly: in one experimental study it was shown that a double infection with H. defensa and S. symbiotica caused a severe reduction in fecundity compared to singly infected aphids, but that the same double infection increased resistance to a parasitoid (Oliver et al. 2006). The observed lack of multiple infections is partly due to the fact that usually only a smaller number of symbiont species has been surveyed. Here we show that most of the frequencies of multiple infections can be predicted given the different frequencies of the six symbiont species in the pea aphid genetic clusters. A challenge remains to investigate which factors allow the stable coexistence of these bacteria.
CONCLUSIONS
There is great current interest in exploring the ecological processes that may lead to population differentiation, specialization, and eventually speciation (Rundle and Nosil 2005; Butlin et al. 2008; Hodges and Derieg 2009; Schluter and Conte 2009; Via 2009). Herbivorous insects have proven to be important experimental models because of the strong selection pressure on host plant use and because mating often occurs on the host plant facilitating population differentiation (Via 2001; Funk et al. 2002). Most studies have looked at adaptations to pairs of host plant species but there is increasing evidence that population structure may be more complex and involve greater number of plant species (Nosil and Sandoval 2008; Xie et al. 2008). The work reported here and by Peccoud et al. (2009a,b) clearly shows that Old World pea aphid populations have a very complex population structure involving a large number of genetically differentiated clusters adapted to a wide range of host plants. The picture is complicated by certain plant species that are permissive hosts for many pea aphid clones adapted to other host plants, and by secondary symbiotic bacteria that are distributed nonrandomly with respect to host plant. Outstanding challenges for work on this system include obtaining a better understanding of the selection regimes experienced by aphids on different host plants and dissecting the genetic architecture of host plant adaptation. The complete pea aphid genome sequence (The International Aphid Genomics Consortium 2010) will greatly facilitate molecular study of this system.
Associate Editor: T. Craig
ACKNOWLEDGMENTS
We thank A. Faulconbridge, M. Harrison, and J. Trakalo for technical assistance. We would like to thank K. Lehmann and O. Volling for help with locating suitable collecting sites in Germany, W. Bertram, M. and D. Füllgrabe, C. Müller, A. Oelbke, and K. Scheibner for access to their land and D. Ferrari, D. Ferrari and L. Ferrari for help with the aphid collections. We also thank J. Peccoud, T. Craig, and an anonymous reviewer for constructive comments on an earlier version of this manuscript. This work was funded by the Biotechnology and Biological Sciences Research Council grant D19263.
LITERATURE CITED
Author notes
Data Archived: Dryad doi:10.5061/dryad.8qb00
