Abstract
Phytophagous insects frequently use multiple host-plant species leading to the evolution of specialized host-adapted populations and sometimes eventually to speciation. Some insects are confronted with a large number of host-plant species, which may provide complex routes of gene flow between host-adapted populations. The pea aphid (Acyrthosiphon pisum) attacks a broad range of plants in the Fabaceae and it is known that populations on Trifolium pratense and Medicago sativa can be highly specialized at exploiting these species. To find out whether adaptation to a broad range of co-occurring hosts has occurred, we tested the performance of pea aphid clones collected from eight host-plant genera on all of these plants in a reciprocal transfer experiment. We provide evidence for pervasive host-plant specialization. The high performance of all aphid clones on Vicia faba suggests that this host plant could be a site of gene flow between different populations that could limit further host-associated divergence. The genetic variance in host-plant usage was partitioned into within- and among-population components, which represent different levels of host adaptation. Little evidence of within-population trade-offs in performance on different plant species was found.
Many organisms live in heterogeneous environments and are exposed to divergent selection on traits associated with resource use. Depending on the balance between selection and gene flow, microhabitat-specific genetic structuring may occur, leading to the formation of genetically distinct and specialized populations (Slatkin 1973; Felsenstein 1976; Drès and Mallet 2002; Hendry et al. 2002). The conditions under which these specialized populations continue to differentiate genetically to become full species are topics of much current interest in evolutionary ecology (Schluter 2001; Via 2001; Rundle and Nosil 2005). Here, we report on patterns of fitness within and among eight populations of pea aphids (Acyrthosiphon pisum) collected from different genera of host plant (all Fabaceae) and then allowed to feed in a factorial experiment on all eight food plants.
Phytophagous insects very frequently feed on more than one species of host plant in which they experience different selection pressures. Plants vary in their nutritional composition, chemical defenses, and in the substances that can be used as attractants in plant location and insect populations feeding on different species will be under selection to adapt to these differences (Ehrlich and Raven 1964; Krischik et al. 1991; Mopper et al. 1995; Egan and Ott 2007). Insects living on different plants may also experience different microenvironmental conditions, or be exposed to different suites of natural enemies (Bernays and Graham 1988; Denno et al. 1990; Feder 1995; Nosil 2004; Nosil and Crespi 2006), causing additional divergent selection. In phytophagous insects mating frequently occurs on the host plant and this de facto assortative mating facilitates population differentiation (Bush 1969; Feder et al. 1994; Via 2001). A recent survey of a community of herbivores on two species of goldenrod (Solidago altissima and S. gigantea) found that four of nine species showed some host-associated genetic differentiation, suggesting that such differentiation is common and not restricted to well studied but idiosyncratically chosen model systems (Stireman et al. 2005). Plant-feeding insects are thus excellent model systems for exploring population differentiation, specialization, and ecological speciation in the field (Via 2001; Berlocher and Feder 2002; Drès and Mallet 2002; Funk et al. 2002).
Many herbivorous insects face and use a complex environment in which many more than two different hosts may be found close together (Futuyma and Gould 1979; Thompson 1994), but the degree of genetic differentiation across multiple hosts has rarely been examined (see Futuyma and Philippi 1987; Carroll and Boyd 1992 and Roininen et al. 1993 for exceptions). From studies on two-host systems, there is ample evidence for genetic variation in host-plant preference and performance both within and among populations (e.g., Tabashnik et al. 1981; Futuyma et al. 1984; Rausher 1984; Futuyma and Philippi 1987; Prokopy et al. 1988; Via 1991a; Craig et al. 1993; Mackenzie 1996; Bossart 2003; Emelianov et al. 2003; Ueno et al. 2003). Surprisingly, only a few studies have estimated the relative magnitude of the genetic variation in host-use parameters both within and between host-associated populations (but see Via 1991a; Ferrari et al. 2006), which provides evidence on the balance between selection and gene flow. The conditions under which host-associated differentiation and ecological speciation may occur are of central importance to understanding the evolution of specialization. The presence of genetic trade-offs in performance is a straightforward way to explain the evolution of specialization (Futuyma and Moreno 1988; Jaenike 1990; Via 1990). However, in experimental studies, negative genetic correlations in fitness across host plants have rarely been observed within populations (e.g., Rausher 1984; Hare and Kennedy 1986; Karowe 1990; reviewed in Scheirs et al. 2005). There is some evidence from pea aphids, A. pisum, that loci associated with performance on two plant species colocate on the genome, which indicates close genetic linkage or negative pleiotropy (Hawthorne and Via 2001). Specialization can also evolve in the absence of genetic trade-offs, for example if the rank order of genotypes' fitness differ across host plants (Fry 1996; Kawecki 1996).
The same questions that can be asked about population differentiation and specialization on two host plants can of course be asked of more complex systems. For example, Futuyma and Philippi (1987) explored the performance of a series of parthenogenetic clones of the polyphagous geometrid moth Alsophila pometaria on four different host plants and found substantial statistical effects of genotype and host plant, as well as a significant interaction, although little evidence of trade-offs in performance across host plants. In addition to the questions highlighted above, it is possible to explore whether the level of genetic differentiation in the herbivore population is related to the phenotypic or genetic distinctiveness of the host. We can also look for more complex patterns of host usage, for example differentiated populations that overlap in their spectrum of suitable host plants. If two host-associated populations share a subset of food plants then this might provide a channel for gene flow that could prevent or retard future adaptation.
Our model organism, the pea aphid (A. pisum (Harris)), feeds on a broad range of host plants, almost exclusively in the family Fabaceae (see Ferrari et al. 2006 for a summary). In North America, populations feeding on two agriculturally important legumes, alfalfa (Medicago sativa) and red clover (Trifolium pratense) show strong local adaptation (Via 1991a) and preferentially choose their home plant in choice trials (Caillaud and Via 2000; Via et al. 2000). The two host-associated populations can interbreed and there is approximately 10% interhost migration in the field (Via 1999). However, F1 hybrids are at a severe ecological disadvantage with lower fitness on both hosts than the specialized parent (Via et al. 2000). Quantitative trait loci associated with performance on, and acceptance of, the two host plants have been genetically mapped (Hawthorne and Via 2001).
Pea aphid was introduced into North America but the same host-associated populations occur in the Old World (Sandström 1996; Simon et al. 2003; Ferrari et al. 2006; Frantz et al. 2006). There is also evidence for differentiation on other plant species and some host-associated populations have been accorded subspecies status based on minor, quantitative morphological distinctions. The two most distinct subspecies are A. p. ononis (Koch) from restharrow (Ononis spp.) and A. p. spartii (Koch) from broom (Cytisus scoparius), whereas aphids from pea (Pisum sativum) have been described, more controversially, as the subspecies A. p. destructor (Johnson) (Eastop 1971; Heie 1994; Blackman and Eastop 2007). There has been relatively little molecular study of Old World aphid populations but in a microsatellite survey of 127 aphid genotypes collected in France from the perennial crops T. pratense and M. sativa and the annual crops broad bean (Vicia faba) and pea (P. sativum), Frantz et al. (2006) found aphids collected from the annual and the perennial crops to be distinct, with a further clustering of aphid clones collected from Trifolium and Medicago but not from Vicia and Pisum.
In this article we explore the extent of host adaptation within pea aphids collected from eight host-plant species, largely from within the same geographic area in the south of England. Previously we showed that aphid clones tend to have a strong, genetically based behavioral preference for the plant species from which they were collected. Most pea aphid clones also showed a strong secondary preference for V. faba (Ferrari et al. 2006). Here we report on the fitness of pea aphid clones when forced to feed on each of the eight host plants in a reciprocal transfer experiment. We ask to what degree the pea aphid population is structured into host-plant adapted populations—in other words, is the ecological specialization observed in pea aphids feeding on T. pratense and M. sativa, particularly in North America, also seen on other cultivated and wild hosts. We then partition the genetic variation in host use into components representing adaptation to the “home” plant, idiosyncratic patterns of host use by different host-associated populations, and within-population clonal variation. To assess the degree of local adaptation, we analyze both whether aphid populations perform better on the plant species from which they were collected compared to the average on other plant species (home vs. away) and whether each population outperforms aphids from other host-associated populations on their home plant (local vs. foreign; Kawecki and Ebert 2004). Finally we ask whether there is evidence of negative genetic correlations between performance on different plants that could facilitate population differentiation.
Methods
ORIGIN AND CULTURE
Pea aphids were sampled from species belonging to eight genera of Fabaceae: V. faba L., Lathyrus pratensis L., P. sativum L., T. pratense L., M. sativa L., Lotus pedunculatus Cav. (=uliginosus Schkuhr), C. (=Sarothamnus) scoparius (L.) Link, and Ononis spp. (Ononis spinosa L. and O. repens L.). In the remainder of the article host plants will usually be referred to simply by the genus name. Pea aphids reproduce asexually throughout the spring and summer and produce a single sexual generation in the fall without changing host plant. Clonal cultures were established from individuals collected in the field and maintained under summer conditions in which only asexuals are produced. The main advantage of working with cyclical parthenogens is that single genotypes can be maintained for long periods of time. This enabled us to partition statistically the genetic variance in host-plant performance of a large collection of aphid clones into components with different biological interpretations. Throughout this article we will refer to the descendents of one field collected aphid as an “aphid clone” and to a group of aphid clones collected from the same plant species as a “population.”
Between 17 and 19 clonal lines were set up for seven of the eight host-plant genera. Insects were collected from the field within a circle of 25-km radius centered at Silwood Park, Berkshire, UK. The eighth host plant, Ononis, was uncommon in this area and in total we were only able to obtain 10 aphid clones, three from the study area, four from sites located up to 80 km from Silwood Park, and three from near Alfeld (Leine) in Lower Saxony, Germany. Aphid clones from each plant species were collected from at least three different sites and within a site from spatially separate host plants to ensure representative genetic diversity from each host-associated population. To check we had not multiply sampled the same aphid clone, all aphids were genotyped (see below). In southern England Lathyrus, Trifolium, Lotus, Cytisus, and less commonly Ononis grow, often together, in species-rich grassland, whereas Pisum, M. sativa, and V. faba are grown as crops although with alternative host plants common in the field margins. Wild species of Medicago and Vicia are also common in this region.
Pea aphid clones from any host plant can be maintained on V. faba (Müller 1962; Sandström and Pettersson 1994; Ferrari and Godfray 2006) and all cultures were kept on this species (cultivar “The Sutton”) at 15°C, 70% r.h., and a 16:8 h light:dark cycle within the same controlled environment room. Experiments were performed on aphid clones that had been kept on Vicia for at least four, but in most cases approximately 30, generations that will have minimized any transgenerational maternal effects arising from the plant species on which they were collected.
GENOTYPING
All aphid clones were genotyped using four microsatellite markers and by sequencing six loci. Details of these markers are given in online Supplementary Table S1 and a full analysis will be presented elsewhere as part of a study of a much larger collection of pea aphids sampled from different host plants at different geographic sites. Only three pairs of aphid clones could not be distinguished using these markers, one collected on Cytisus, another on Pisum, and the third on Trifolium. Depending on the precise population genetic assumptions we expect a small number of distinct aphid clones to be indistinguishable with this amount of information, and hence we decided to retain all aphid clones in the analysis. This may lead to a small overestimate of population divergence, although it will have no qualitative effect on any of our conclusions. At the end of the experiment each aphid clone was tested again with the microsatellite markers and this showed that no contamination had occurred during culturing.
EXPERIMENTAL DESIGN
The performance of each aphid clone was tested in a full factorial design on all eight host-plant genera. The test plant species were those from which we had collected the aphids except that O. spinosa alone was used for this genus. Three separate measures of performance on the different host plants were recorded, which we call acceptance, survival, and fecundity. The protocols for assessing these measures are based on Ferrari and Godfray (2003, 2006) and Ferrari et al. (2007).
We define acceptance as the number of offspring an aphid produces on a test plant in the first 24 h after transfer from Vicia. This measure will be influenced by the physiological state of the aphid and its behavioral response to host-plant transfer, as well as its preference for the new host plant. Five 13- to 14-day-old adult pea aphids, which had been reared at low densities on Vicia, were transferred to a Petridish (9 cm diameter) containing leaves of the test plant with their stems, secured with cotton wool, placed in a 0.5 mL Eppendorf tube containing water. The number of nymphs produced over a 24 h period was recorded. In a minority of cases fewer than five aphids from a particular aphid clone of the right age were available for testing and account of this was taken in the statistical analysis.
From the aphid nymphs produced in this acceptance assay, 10 were taken and allowed to develop on a potted plant of the same species, which was enclosed in a clear cylindrical plastic cage (diameter 10 cm, height 25 cm) with two mesh windows. Survival was defined as the fraction alive after six days (approximately two days before they became adults). When fewer than ten nymphs were produced, they were supplemented with aphids produced under the same conditions but on Vicia, taken from dishes that were prepared for this purpose. This was done to ensure a survival assay could be conducted for all aphid clones on all test plants, even if they were to produce no offspring during the acceptance phase. There is a small risk that aphid nymphs that have spent the first 24 h of their life on Vicia as opposed to a poorer host plant will have a higher probability of subsequent survival. We have no evidence of this effect occurring but were it to happen it would tend to reduce rather than increase our estimate of the extent of local adaptation. In 26 of 3625 replicates fewer than 10 aphids were used in this assay.
To measure fecundity, one individual aphid from the survival assay was taken and transferred to another potted individual of the same plant species and the number of offspring produced during the following eight days was recorded. In 35% of the replicates no aphid survived on a specific test plant and thus the fecundity assay could not be performed. The index of fecundity should thus be interpreted as reproductive output conditional on survival.
All plants for this experiment were grown in a greenhouse with supplementary lighting. It proved very difficult to grow Cytisus under these conditions and we were therefore unable to perform the same number of replicates for each aphid clone on Cytisus as on the other test plant species.
The experiments were performed in controlled temperature rooms at 20 ± 1°C, 70% r.h. and a 16:8 h light:dark cycle. Replicates were stratified across eight temporal blocks. Each block included half of the aphid clones from each collection plant, and these were tested on all test plant species (it was logistically not possible to test all aphid clones in a single block). We thus obtained four replicates for each combination of aphid clone and test plant. Different replicates of the same aphid clone on a particular test plant were separated by at least three generations.
ANALYSIS
The aim of the analysis was to determine the degree to which the pea aphids we sampled were differentiated into host-adapted populations. To assess the relative importance of different biological processes we took a statistical modeling approach and asked how the variability in the data could be partitioned into different components. Components of variation (or more technically deviance) were estimated using generalized linear modeling (GLM) techniques implemented in the R statistical program (http://www.r-project.org/). We began by fitting a factor with eight levels for each of the temporal blocks, followed by a factor with eight levels for each test plant (environment main effects) and then a further factor with 135 levels for each aphid clone (main effect of genotype). In doing this we are fitting the main effects of aphid genotype and environment and we will call this our minimal model. Adding the aphid clone by test plant interaction to this model includes all the genotype by environment interactions, both those within and across populations. We will call this our maximal model, and below explain how we partitioned the difference between the variance explained by the minimal and maximal models into different biologically relevant components. The residual variation, after fitting the maximal model, arises from nongenetic, within-clone sources and from measurement error.
To partition the genotype by environment interaction, we sequentially added four components (which we will refer to as A, B, C, and D) to the minimal model, the final model including all four components being identical to the maximal model. First, we asked how much variation could be explained by fitting the single factor “home/away” to the minimal model (A). We define an aphid clone's home plant to be the species from which it was collected with any of the other seven species being an away plant. Fitting the home/away factor thus tested whether on average aphids performed better on their collection plant species compared to the alternatives. Second, we fitted the interaction between home/away and collection plant (B), which allows the strength of the performance advantage on the home plant to vary across populations. Third, we added the interaction between the test plant and collection plant (C). This component represents differences among populations in performance on the away plant species. Finally, we fitted the interaction of test plant and aphid clone (D) to give the maximal model. Adding this component reveals within-population clonal variation in performance on the various host plants. This includes both strict within-population variation but also variation due to any of the aphid clones that had been collected on host plants to which they are not adapted. All aphid clones were reared on Vicia prior to the start of the experiment and this maternal environment might have affected the aphids' performance. We therefore paid particular attention to the contribution of terms involving Vicia to the variation explained by the different models.
For the acceptance and fecundity assays the raw data are counts suggesting a log-linear analysis with a Poisson error distribution in the GLM, whereas for the survival assay the data are ratios suggesting logistic analysis. However, in both cases there was significant overdispersion and hence the components of deviance were calculated using quasi-likelihood techniques (the quasipoisson and quasibinomial options in the R glm function), which estimate the relationship between the variance and mean from the data. To illustrate some of the results we plot the product of survival and fecundity as a composite measure of fitness. This is easier to interpret biologically—it is the average number of offspring an aphid is expected to produce within the first two weeks of its life—but as it is a composite variable with a complex error distribution it cannot be easily analyzed using GLM techniques.
We also explored whether within each host-associated population there were negative correlations between performance on the home plant species and on any of the away plant species. We restricted the analyses to clones collected from a single host plant because within-population correlations can reveal the contemporary genetic architecture of a trait, whereas among-population correlations may have arisen through a range of processes and are harder to interpret (Futuyma and Moreno 1988; Schluter 2000). As described below a number of aphid clones had performance profiles that suggested they were migrants that had wandered from other plant species to which they were adapted. The inclusion of these migrants, which have poor performance on the home plant species and high performance on away plant species, might lead to the spurious identification of trade-offs within a population adapted to a particular host plant. In effect they would tend to convert a pure within-population analysis into a partial among-population analysis. To avoid this, we analyzed the data with and without the putative migrants. We identified these migrants formally by using a hierarchical clustering method implemented in the R program (hclust, using the complete linkage option). As raw data we used the mean normalized product of survival and fecundity on each test plant except Vicia. Aphid clones that were found within clusters of specialists from another collection plant were classed as migrants, whereas aphid clones that had no clear host-plant association were considered to belong to the aphid population from their collection plant. Then, to test for within-population trade-offs, we calculated the Pearson's product moment correlation between mean survival of a clone on its home plant species and on each of the alternative away plant species. We used Bonferroni corrections to control for the effects of multiple significance tests.
Results
PATTERNS OF PERFORMANCE ON DIFFERENT TEST PLANTS
The different test plants varied in their average suitability for pea aphids (significance of host plant in the analysis of all clones on all test plants: acceptance, F7,2613= 246.9, P < 0.001; survival, F7,2581= 479.2, P < 0.001; fecundity, F7,1463= 320.2, P < 0.001). As has been previously noted, pea aphids perform well on Vicia and this accounts for a large fraction of the test plant effect (acceptance: 43%, survival: 42%, fecundity: 88%). Performance on away plants other than Vicia is generally poor. This is illustrated in Figure 1 in which we plot the average fitness of aphids on test plants other than that from which they were collected (i.e., solely on away plants). Here, we use as the composite measure of fitness, the product of survival and fecundity.
Average performance of “foreign” pea aphids on eight away test plant species (i.e., aphids other than those collected on this host plant). The bars show means ± SE, predicted by the main analysis. Performance is calculated as the product of survival and fecundity. On average, pea aphids perform much better on Vicia than on any other species.
PARTITIONING THE GENOTYPE × ENVIRONMENT INTERACTION
As described in the Methods section, the genotype by environment interaction can be partitioned into four components. These represent average and population-specific host adaptation, and then idiosyncratic population and clonal performance. For each measure of performance the partition is shown as a pie chart in Figure 2 and an overview of the analysis is provided in Table 1.
Partitioning of the genotype by environment interaction in performance among eight host-associated populations of pea aphids. The pie charts show the fraction of the genotype by environment interaction that can be explained by (A) simple constant mean performances on home and away plants, (B) variation among populations in the degree of adaptation to the home plant, (C) idiosyncratic patterns of performance on different host plants of different populations, and (D) within population variation in performance on the eight different plant species. This partitioning is shown for three measures of fitness: acceptance, survival, and fecundity and also for preference (data from Ferrari et al., 2006).
Analysis of deviance of performance of pea aphids on eight different plant species. Three measures of fitness were assayed: acceptance, survival, and fecundity. The minimal model includes the main effects of environment (block, test plant, and for acceptance, the number of adults used) and the main effects of genotype (collection plant and aphid clone, added sequentially). The four terms labeled A–D were then added sequentially to fit different levels of the genotype × environment (G×E) interaction and create the maximal model. All factors were highly significant (P<0.001).
| Total deviance explained (%) | Acceptance | Survival | Fecundity | |||||||
| 65 | 72 | 77 | ||||||||
| G × E contribution to total explained deviance (%) | 21 | 37 | 38 | |||||||
| Summary of statistical model | Factor | Deviance | df | Percent G × E explained | Deviance | df | Percent G × E explained | Deviance | df | Percent G × E explained |
| Minimal model | No. of adults used | 2109 | 1 | |||||||
| Block | 121 | 7 | 40 | 7 | 127 | 7 | ||||
| Test plant | 7111 | 7 | 3766 | 7 | 13,728 | 7 | ||||
| Collection plant | 2038 | 7 | 212 | 7 | 1862 | 7 | ||||
| Clone | 3458 | 127 | 330 | 127 | 2209 | 127 | ||||
| G × E interaction | ||||||||||
| A. Average performance on home vs. away plants | Home/away | 769 | 1 | 11 | 907 | 1 | 20 | 4025 | 1 | 23 |
| B. Population specific degree of adaptation to the home plant | Home/away × collection plant | 943 | 7 | 13 | 1043 | 7 | 23 | 3604 | 7 | 21 |
| C. Population differentiation in performance on away plants | Collection plant × test plant | 915 | 41 | 13 | 443 | 41 | 10 | 1504 | 41 | 9 |
| D. Within population variation | Clone × test plant | 4530 | 837 | 63 | 2142 | 836 | 47 | 8007 | 666 | 47 |
| Total deviance explained (%) | Acceptance | Survival | Fecundity | |||||||
| 65 | 72 | 77 | ||||||||
| G × E contribution to total explained deviance (%) | 21 | 37 | 38 | |||||||
| Summary of statistical model | Factor | Deviance | df | Percent G × E explained | Deviance | df | Percent G × E explained | Deviance | df | Percent G × E explained |
| Minimal model | No. of adults used | 2109 | 1 | |||||||
| Block | 121 | 7 | 40 | 7 | 127 | 7 | ||||
| Test plant | 7111 | 7 | 3766 | 7 | 13,728 | 7 | ||||
| Collection plant | 2038 | 7 | 212 | 7 | 1862 | 7 | ||||
| Clone | 3458 | 127 | 330 | 127 | 2209 | 127 | ||||
| G × E interaction | ||||||||||
| A. Average performance on home vs. away plants | Home/away | 769 | 1 | 11 | 907 | 1 | 20 | 4025 | 1 | 23 |
| B. Population specific degree of adaptation to the home plant | Home/away × collection plant | 943 | 7 | 13 | 1043 | 7 | 23 | 3604 | 7 | 21 |
| C. Population differentiation in performance on away plants | Collection plant × test plant | 915 | 41 | 13 | 443 | 41 | 10 | 1504 | 41 | 9 |
| D. Within population variation | Clone × test plant | 4530 | 837 | 63 | 2142 | 836 | 47 | 8007 | 666 | 47 |
Analysis of deviance of performance of pea aphids on eight different plant species. Three measures of fitness were assayed: acceptance, survival, and fecundity. The minimal model includes the main effects of environment (block, test plant, and for acceptance, the number of adults used) and the main effects of genotype (collection plant and aphid clone, added sequentially). The four terms labeled A–D were then added sequentially to fit different levels of the genotype × environment (G×E) interaction and create the maximal model. All factors were highly significant (P<0.001).
| Total deviance explained (%) | Acceptance | Survival | Fecundity | |||||||
| 65 | 72 | 77 | ||||||||
| G × E contribution to total explained deviance (%) | 21 | 37 | 38 | |||||||
| Summary of statistical model | Factor | Deviance | df | Percent G × E explained | Deviance | df | Percent G × E explained | Deviance | df | Percent G × E explained |
| Minimal model | No. of adults used | 2109 | 1 | |||||||
| Block | 121 | 7 | 40 | 7 | 127 | 7 | ||||
| Test plant | 7111 | 7 | 3766 | 7 | 13,728 | 7 | ||||
| Collection plant | 2038 | 7 | 212 | 7 | 1862 | 7 | ||||
| Clone | 3458 | 127 | 330 | 127 | 2209 | 127 | ||||
| G × E interaction | ||||||||||
| A. Average performance on home vs. away plants | Home/away | 769 | 1 | 11 | 907 | 1 | 20 | 4025 | 1 | 23 |
| B. Population specific degree of adaptation to the home plant | Home/away × collection plant | 943 | 7 | 13 | 1043 | 7 | 23 | 3604 | 7 | 21 |
| C. Population differentiation in performance on away plants | Collection plant × test plant | 915 | 41 | 13 | 443 | 41 | 10 | 1504 | 41 | 9 |
| D. Within population variation | Clone × test plant | 4530 | 837 | 63 | 2142 | 836 | 47 | 8007 | 666 | 47 |
| Total deviance explained (%) | Acceptance | Survival | Fecundity | |||||||
| 65 | 72 | 77 | ||||||||
| G × E contribution to total explained deviance (%) | 21 | 37 | 38 | |||||||
| Summary of statistical model | Factor | Deviance | df | Percent G × E explained | Deviance | df | Percent G × E explained | Deviance | df | Percent G × E explained |
| Minimal model | No. of adults used | 2109 | 1 | |||||||
| Block | 121 | 7 | 40 | 7 | 127 | 7 | ||||
| Test plant | 7111 | 7 | 3766 | 7 | 13,728 | 7 | ||||
| Collection plant | 2038 | 7 | 212 | 7 | 1862 | 7 | ||||
| Clone | 3458 | 127 | 330 | 127 | 2209 | 127 | ||||
| G × E interaction | ||||||||||
| A. Average performance on home vs. away plants | Home/away | 769 | 1 | 11 | 907 | 1 | 20 | 4025 | 1 | 23 |
| B. Population specific degree of adaptation to the home plant | Home/away × collection plant | 943 | 7 | 13 | 1043 | 7 | 23 | 3604 | 7 | 21 |
| C. Population differentiation in performance on away plants | Collection plant × test plant | 915 | 41 | 13 | 443 | 41 | 10 | 1504 | 41 | 9 |
| D. Within population variation | Clone × test plant | 4530 | 837 | 63 | 2142 | 836 | 47 | 8007 | 666 | 47 |
Average performance on home and away plants
There was clear evidence for host adaptation (Fig. 3). We asked how much genetic variation on different test plants can be explained by simply assuming that aphid performance can be summarized by two numbers: average performance on home and away plants. This home/away factor explained 11%, 20% and 23% of the genotype by environment interaction in acceptance, survival, and fecundity, respectively (Fig. 2). We can also estimate the difference in the average survival or number of nymphs produced on a home or away plant. The statistical technique we used fits a linear model to either the log of the response variable (acceptance and fecundity) or in the case of survival to its logit (log-odds). In the acceptance assay, aphids produced 1.31 (±0.04) times more aphids on their home plant compared to an average away plant although their fecundity was 2.61 (±0.13) times higher on the home plant. In the survival assay, the odds ratio of remaining alive was 2.32 (±0.10) times higher on the home plant. All these multipliers are highly significantly different from 1 (P≪ 10−3) indicating a strong effect of host-plant adaptation (Table 1).
Performance of pea aphids collected from eight plant genera on their home plant, the average away plant and Vicia. Performance is measured as (A) acceptance, (B) survival, and (C) fecundity. The host-associated populations perform best on their home plant and Vicia, but vary in the degree of host adaptation. Bars show means ± SE. There are only two bars for the population from Vicia as “home” and “Vicia” are identical.
Population-specific degree of adaptation to the home plants
The populations collected from different plant species varied in their degree of host adaptation (Fig. 3). The additional variation explained by adding the interaction between the home/away and collection plant factors (for seven degrees of freedom) quantifies the importance of this effect. Adding this term explained an additional 13%, 23%, and 21% of the genotype by environment interaction in acceptance, survival, and fecundity, respectively (Fig. 2).
To explore this in more detail, we analyzed the home/away factor separately for the eight host-associated populations. As described above, Vicia is an exceptionally good away plant for all host-associated populations (Fig. 1) and this thus inflates the mean performance on the average away plant. We therefore excluded Vicia as a test plant from this analysis. All host-associated populations performed significantly better on their home plant than on the average away plant for all three measures of fitness (Table 2). The Cytisus population in particular showed high survival and fecundity on the home plant compared to away plants, whereas the populations from Medicago and Pisum also had markedly higher fecundity on their home species relative to the away plants.
Difference between performance of the host-associated pea aphid populations on their home plant and the average away plant (±SE). The differences are expressed as ln(number of offspring) for acceptance and fecundity and as logits for survival. The values are estimates from separate analyses for each of the eight populations. All estimates are significantly different from 0 (P<0.01). To avoid possible maternal effects Vicia was not included as a test plant in the analysis (except for the population from Vicia, where Vicia is the home plant).
| Population | Acceptance | Survival | Fecundity |
| Cytisus | 0.51±0.09 | 1.36±0.16 | 1.60±0.31 |
| Lotus | 0.30±0.07 | 1.05±0.11 | 1.34±0.15 |
| Ononis | 0.45±0.10 | 1.13±0.16 | 1.04±0.23 |
| Medicago | 0.30±0.06 | 0.81±0.10 | 1.63±0.13 |
| Pisum | 0.60±0.06 | 1.04±0.11 | 1.69±0.13 |
| Lathyrus | 0.15±0.07 | 1.10±0.12 | 1.09±0.18 |
| Trifolium | 0.38±0.07 | 0.70±0.13 | 1.06±0.19 |
| Vicia | 0.48±0.05 | 0.97±0.10 | 1.27±0.12 |
| Population | Acceptance | Survival | Fecundity |
| Cytisus | 0.51±0.09 | 1.36±0.16 | 1.60±0.31 |
| Lotus | 0.30±0.07 | 1.05±0.11 | 1.34±0.15 |
| Ononis | 0.45±0.10 | 1.13±0.16 | 1.04±0.23 |
| Medicago | 0.30±0.06 | 0.81±0.10 | 1.63±0.13 |
| Pisum | 0.60±0.06 | 1.04±0.11 | 1.69±0.13 |
| Lathyrus | 0.15±0.07 | 1.10±0.12 | 1.09±0.18 |
| Trifolium | 0.38±0.07 | 0.70±0.13 | 1.06±0.19 |
| Vicia | 0.48±0.05 | 0.97±0.10 | 1.27±0.12 |
Difference between performance of the host-associated pea aphid populations on their home plant and the average away plant (±SE). The differences are expressed as ln(number of offspring) for acceptance and fecundity and as logits for survival. The values are estimates from separate analyses for each of the eight populations. All estimates are significantly different from 0 (P<0.01). To avoid possible maternal effects Vicia was not included as a test plant in the analysis (except for the population from Vicia, where Vicia is the home plant).
| Population | Acceptance | Survival | Fecundity |
| Cytisus | 0.51±0.09 | 1.36±0.16 | 1.60±0.31 |
| Lotus | 0.30±0.07 | 1.05±0.11 | 1.34±0.15 |
| Ononis | 0.45±0.10 | 1.13±0.16 | 1.04±0.23 |
| Medicago | 0.30±0.06 | 0.81±0.10 | 1.63±0.13 |
| Pisum | 0.60±0.06 | 1.04±0.11 | 1.69±0.13 |
| Lathyrus | 0.15±0.07 | 1.10±0.12 | 1.09±0.18 |
| Trifolium | 0.38±0.07 | 0.70±0.13 | 1.06±0.19 |
| Vicia | 0.48±0.05 | 0.97±0.10 | 1.27±0.12 |
| Population | Acceptance | Survival | Fecundity |
| Cytisus | 0.51±0.09 | 1.36±0.16 | 1.60±0.31 |
| Lotus | 0.30±0.07 | 1.05±0.11 | 1.34±0.15 |
| Ononis | 0.45±0.10 | 1.13±0.16 | 1.04±0.23 |
| Medicago | 0.30±0.06 | 0.81±0.10 | 1.63±0.13 |
| Pisum | 0.60±0.06 | 1.04±0.11 | 1.69±0.13 |
| Lathyrus | 0.15±0.07 | 1.10±0.12 | 1.09±0.18 |
| Trifolium | 0.38±0.07 | 0.70±0.13 | 1.06±0.19 |
| Vicia | 0.48±0.05 | 0.97±0.10 | 1.27±0.12 |
Specialization is not only characterized by the fact that a population performs well on its home plant compared to its performance on other plants (home vs. away), but also by it outperforming other populations on its home plant (local vs. foreign). The difference between the fitted values of the performance of local and foreign populations on the different test plants provides a measure of this second aspect of specialization (plotted in Fig. 4 on the appropriate scale for the statistical model). The populations from Cytisus and Ononis had the greatest local advantage. The population from Vicia was by this measure the least specialized as it did no better than populations from other collection plants on Vicia.
Performance of “local” pea aphid populations compared to “foreign” aphid clones (i.e., those collected from other plant species) on each test plant species. The bars show the difference between the performance of local and foreign populations on the different test plants (±SE) on the appropriate scale for the statistical model ((A) acceptance, (B) survival, (C) fecundity).
Population differentiation in performance on different away plant
We explored whether there was any evidence that groups of host-associated populations had adapted to clusters of host-plant species that might share physiological or morphological characteristics (Fig. 5). To test this, we added the interaction between collection plant and host plant to the model by fitting a separate term for every aphid population-test plant combination for an additional 41 degrees of freedom. In doing so, we explained an extra 13%, 10%, and 9% of the genotype by environment interaction in acceptance, survival, and fecundity, respectively (Fig. 2). This is a relatively small increase considering the extra number of parameters fitted, and suggests that most variation at the population level can be explained by the more general home/away comparisons. For survival and fecundity much of this variation was explained by differences in performance on Vicia among the host-associated populations (acceptance: 13%, survival: 27%, fecundity: 35%).
Performance of pea aphids collected from eight host-plant genera when tested on all eight species. Three measures of performance are shown (A) acceptance, (B) survival, and (C) fecundity. Bars with different patterns represent different test plant species (means ± SE) and are labeled with the first letter of the test plant name (with Lotus being placed before Lathyrus). The home plant of each population is highlighted with an arrow.
The populations from Vicia and Trifolium were the most similar to each other and showed the highest correlations in the spectra of performance on the eight test plants. The population from Pisum was also quite similar to these two, especially in survival. Correlations between populations collected from other host plants were weaker.
Within-population variation
Finally, by adding the aphid clone × test plant interaction we explain the rest of the genotype by environment interaction, that is, all of the variation among aphid clones in their relative performance across test plants. This term accounts for 63%, 47%, and 47% of the genotype by environment interaction in acceptance, survival, and fecundity, respectively (Fig. 2) although for 837 degrees of freedom (666 for fecundity where missing values occur when aphids were unable to reach the adult stage on a test plant). This among-clone variation is illustrated in Figure 6 for the composite measure of survival and fecundity. Part of the variation may be due to migrants that are not adapted to the host plant on which they are collected (see below). The rest will be due to within-population genotype-specific effects. There was no more variation among aphid clones in their performance on Vicia than there was in their performance on other away plants (the response to Vicia explained 11%, 5%, and 12% of the total variation explained by this term in the analysis of acceptance, survival, and fecundity, respectively).
Performance of pea aphid clones collected from eight host-plant genera when tested on all eight test plant species. Performance is calculated as the product of survival and fecundity. Bars with different patterns represent different test plant species (means ± SE). The different panels show aphids collected from different plant genera: (A) from Cytisus, (B) Lotus, (C) Ononis, (D) Medicago, (E) Pisum, (F) Lathyrus, (G) Trifolium, and (H) from Vicia.
POSSIBLE MIGRANTS
Inspection of the detailed data on clonal performance suggested that some aphid clones had performance profiles that match those of specialists from other collection plants (Fig. 6). These might represent migrants that had permanently or temporarily colonized a less suitable food plant. As described in the Methods we used hierarchical clustering analysis to identify these potential migrants. No aphid clones of this type were identified in the populations from Medicago and Pisum whereas the populations from Ononis and Cytisus had two migrants each. In the population from Lotus three aphid clones appeared to be migrants, one each with profiles suggesting them to be Trifolium, Lathyrus, or Pisum specialists. Potential migrants were most common in the Lathyrus population; two-thirds of the aphid clones were Lathyrus specialists, but we also found aphid clones with Pisum, Lotus, or Medicago performance profiles. In the population from Trifolium, there were six aphid clones with Pisum or Medicago profiles. There were no aphid clones that were solely able to use Vicia and all aphid clones from this collection plant clustered within populations from other plant species. Half of the aphid clones from Vicia had Trifolium profiles, seven had Pisum profiles, and the remaining two appeared to be adapted to Medicago. Data from our earlier study of host-plant behavioral preferences using a subset of these aphid clones (Ferrari et al. 2006) support this identification of migrant aphid clones, as did our molecular analysis (J. Ferrari, J. West, M. Harrison, J. Trakalo, D. Hawthorne, S. Via, and C. Gadfray, unpubl. data).
WITHIN-POPULATION TRADE-OFFS
We tested whether there were any within-population negative correlations among aphid clones in survival on their home plant and on each of the away plant species. As described in the Methods we excluded the potential migrants identified above from our main analysis. Of 56 comparisons, only 19 showed negative trends (Table 3) and none of these were significant after Bonferroni correction. Even when the migrants were included, no negative correlation was significant after Bonferroni correction (online Supplementary Table S2). Although this analysis has limited statistical power there is no evidence for pervasive within-population negative correlations in performance on different host plants that might indicate negative pleiotropy or close genetic linkage and therefore trade-offs.
Within population, among-aphid clone correlations between survival on the home plant and each away plant species. Pearson's product moment correlation coefficients are shown. Putative migrants that were identified with multivariate methods were excluded from this analysis. None of the correlations were significant after Bonferroni correction.
 | Cytisus | Lotus | Ononis | Medicago | Pisum | Lathyrus | Trifolium | Vicia |
| Cytisus | 0.30 | −0.06 | 0.68 | −0.40 | 0.72 | −0.04 | 0.31 | |
| Lotus | 0.27 | 0.69 | 0.42 | −0.42 | 0.23 | −0.19 | 0.31 | |
| Ononis | 0.13 | −0.03 | −0.09 | 0.28 | 0.26 | −0.04 | 0.24 | |
| Medicago | 0.30 | −0.05 | 0.19 | 0.12 | 0.00 | 0.01 | −0.21 | |
| Pisum | 0.34 | 0.19 | −0.12 | 0.31 | 0.05 | 0.44 | −0.23 | |
| Lathyrus | −0.23 | −0.17 | −0.09 | −0.36 | 0.49 | −0.24 | 0.04 | |
| Trifolium | −0.14 | 0.66 | 0.15 | 0.11 | 0.45 | 0.03 | 0.26 | |
| Vicia | 0.52 | 0.53 | 0.33 | 0.49 | 0.05 | −0.17 | 0.19 |
| Cytisus | Lotus | Ononis | Medicago | Pisum | Lathyrus | Trifolium | Vicia |
| Cytisus | 0.30 | −0.06 | 0.68 | −0.40 | 0.72 | −0.04 | 0.31 | |
| Lotus | 0.27 | 0.69 | 0.42 | −0.42 | 0.23 | −0.19 | 0.31 | |
| Ononis | 0.13 | −0.03 | −0.09 | 0.28 | 0.26 | −0.04 | 0.24 | |
| Medicago | 0.30 | −0.05 | 0.19 | 0.12 | 0.00 | 0.01 | −0.21 | |
| Pisum | 0.34 | 0.19 | −0.12 | 0.31 | 0.05 | 0.44 | −0.23 | |
| Lathyrus | −0.23 | −0.17 | −0.09 | −0.36 | 0.49 | −0.24 | 0.04 | |
| Trifolium | −0.14 | 0.66 | 0.15 | 0.11 | 0.45 | 0.03 | 0.26 | |
| Vicia | 0.52 | 0.53 | 0.33 | 0.49 | 0.05 | −0.17 | 0.19 |
Within population, among-aphid clone correlations between survival on the home plant and each away plant species. Pearson's product moment correlation coefficients are shown. Putative migrants that were identified with multivariate methods were excluded from this analysis. None of the correlations were significant after Bonferroni correction.
| Cytisus | Lotus | Ononis | Medicago | Pisum | Lathyrus | Trifolium | Vicia |
| Cytisus | 0.30 | −0.06 | 0.68 | −0.40 | 0.72 | −0.04 | 0.31 | |
| Lotus | 0.27 | 0.69 | 0.42 | −0.42 | 0.23 | −0.19 | 0.31 | |
| Ononis | 0.13 | −0.03 | −0.09 | 0.28 | 0.26 | −0.04 | 0.24 | |
| Medicago | 0.30 | −0.05 | 0.19 | 0.12 | 0.00 | 0.01 | −0.21 | |
| Pisum | 0.34 | 0.19 | −0.12 | 0.31 | 0.05 | 0.44 | −0.23 | |
| Lathyrus | −0.23 | −0.17 | −0.09 | −0.36 | 0.49 | −0.24 | 0.04 | |
| Trifolium | −0.14 | 0.66 | 0.15 | 0.11 | 0.45 | 0.03 | 0.26 | |
| Vicia | 0.52 | 0.53 | 0.33 | 0.49 | 0.05 | −0.17 | 0.19 |
| Cytisus | Lotus | Ononis | Medicago | Pisum | Lathyrus | Trifolium | Vicia |
| Cytisus | 0.30 | −0.06 | 0.68 | −0.40 | 0.72 | −0.04 | 0.31 | |
| Lotus | 0.27 | 0.69 | 0.42 | −0.42 | 0.23 | −0.19 | 0.31 | |
| Ononis | 0.13 | −0.03 | −0.09 | 0.28 | 0.26 | −0.04 | 0.24 | |
| Medicago | 0.30 | −0.05 | 0.19 | 0.12 | 0.00 | 0.01 | −0.21 | |
| Pisum | 0.34 | 0.19 | −0.12 | 0.31 | 0.05 | 0.44 | −0.23 | |
| Lathyrus | −0.23 | −0.17 | −0.09 | −0.36 | 0.49 | −0.24 | 0.04 | |
| Trifolium | −0.14 | 0.66 | 0.15 | 0.11 | 0.45 | 0.03 | 0.26 | |
| Vicia | 0.52 | 0.53 | 0.33 | 0.49 | 0.05 | −0.17 | 0.19 |
Discussion
We found strong evidence that pea aphid populations are made up of a complex of host-associated populations each adapted, although to varying degrees, to different plant species. We chose to work with species from the seven genera of Fabaceae that were most common in our study area, plus an eighth (Ononis) that was much rarer but which is host to a named subspecies of A. pisum (Eastop 1971; Heie 1994). There may be further host-plant related structure in our pea aphid population: possibly other aphid clones are specialized on less common genera of Fabaceae or on different species belonging to the same genus. We also found evidence of two processes that may act against increasing host specialization. First, virtually all aphid clones, from whatever host plant, had high fitness on Vicia. There is thus the possibility that this host plant may be a site of gene exchange between different host-associated populations. Second, we identified some aphid clones with host-plant performance profiles that suggested that they were better adapted to species other than those from which they were collected. We provisionally categorized these as migrants.
We characterized performance on a test plant using three different metrics. The first is the number of nymphs produced by an adult aphid that has just been transferred to that host plant from V. faba, the species on which all aphid clones were maintained (“acceptance”). As we expected, this index showed the least degree of discrimination. The number of nymphs produced by these aphids will be determined by the general condition of the female prior to transfer, the quality of the sustenance she gets on the new host, and the behavioral response to the host plant. As Vicia is a universally good host we would expect all aphids to have the potential to produce nymphs in the first day after transfer to a new plant species. The dispersal stage of the pea aphid is the winged asexual morph and so wingless asexuals, the type used in the experiments, are less likely to encounter new host plants under natural conditions and might thus be expected to show low host discrimination. However, both morphs show clear preferences for their home plant (and Vicia) in choice experiments (Ferrari et al. 2006). Differences in this measure of performance are thus probably caused by a combination of insect behavior and the relative suitability of the plant to support the continuing production of nymphs by the adult.
The second measure of fitness we used is the probability of a nymph surviving for six days from birth. Because the insect must get all its food from the host plant, its nutritional suitability will have a major influence on survival and indeed this index shows strong discrimination across the plant species tested. The third measure is the number of nymphs produced by an adult aphid that has grown to maturity on the test plant. Again plant nutritional suitability will have a strong effect on this measure of fitness, although its interpretation is slightly complicated by the fact that when a host plant is completely unsuitable for an aphid clone, so that no nymphs survive until the adult stage, then no measure of fecundity is possible. Measures two and three can be combined to give a joint estimate of fitness, which we use to illustrate patterns of performance, but we chose to study them separately in the formal analysis as the statistical analysis of composite measures is less straightforward.
The willingness of nearly all aphid clones to feed on Vicia is a striking feature of pea aphid biology. The ability to culture all aphid clones on a single host plant is convenient, but does lead to the complication that maternal effects may influence the results of the fitness assay. If maternal effects were pervasive they could potentially explain the high fitness observed on this test plant. We cannot rule out some role for maternal effects in influencing pea aphid performance on Vicia, but for three reasons seems unlikely that maternal effects alone explain the high suitability of this host plant for all aphid clones. First, survival and fecundity is low for most aphid clones on most test plants, suggesting that it would not have been possible to maintain the aphids on any other plant. Second, we did not observe low initial fitness on Vicia when transferring aphid clones collected from the field onto this host plant. Third, Via (1991b) found no evidence of maternal effects influencing pea aphid performance on Trifolium and Medicago, although this experiment did not involve aphid clones reared on Vicia. Further preliminary data from our laboratory suggest that this lack of maternal effects is typical for a wider range of aphid genotypes. Were maternal effects to be present then this would uniformly increase performance on Vicia; however, such an effect is controlled for in our analysis which the minimal statistical model includes test plant as a main factor. More subtle effects involving interactions might also occur, which is why we separately discuss the contributions of interaction terms including Vicia. At the aphid clone level, there is variation in the precise degree to which different aphid clones are able to use this host plant. This may directly affect their fecundity after transfer to the test plant and hence the fitness we measure in the acceptance assay. If such effects were transmitted to offspring then they could also influence the survival and fecundity measures of fitness. Were this phenomenon to occur it would increase the between-aphid clone variance in fitness. However, its chief consequences would be accounted for by the inclusion of the factor “clone” in the minimal model.
In our analysis we defined a minimal model that contained the main statistical effects of test plant and aphid clone, and a maximal model that contained the test plant by aphid clone interaction. The difference between the two is the genotype by environment interaction that we sought to partition into different components that we could interpret biologically. Simply categorizing host plants as home or away (A) explained a little under a quarter of the deviance of the genotype by environment interaction in survival and fecundity, while allowing the strength of the home or away effect to vary across host-associated populations (B) raised this percentage to a little under a half. Acceptance was influenced to a lesser degree by these factors with the effect being approximately only half as strong. Previously we carried out a similar analysis on the behavioral preferences of a subset of these aphid clones for the same spectrum of host plants (Ferrari et al. 2006 and Fig. 2). Although not precisely comparable, categorizing chosen plants as home or away explained approximately 60% of the genetic variance in host-plant choice. Thus host-plant fidelity and host adaptation are pervasive features of this pea aphid population.
Not all host-plant associated populations show the same degree of host adaptation, which is the reason that allowing the home/away factor to vary across populations collected from different plants explains additional deviance. Most of this effect is due to differences among populations in performance on their home plants rather than variation in performance on away plants. There are at least three possible reasons for this. (1) Different plant species may be intrinsically more favorable for aphid reproduction, perhaps having greater nutritional qualities. This appears to be the case for Pisum, which is a relatively good away plant (Fig. 1) and whose host-associated population do much better on home compared to away plant species (Table 2). (2) The different host-associated populations may vary in the time that they have been associated with their home plant and some may not have become fully adapted or in some populations adaptation may be retarded by greater gene flow from populations on other host plants. (3) Finally, some aphid clones collected from a particular plant species may have been migrants adapted to other host plants. If more migrants were found on certain collection plants then this would reduce the average measure of home performance of the population from that species. We identified too few migrant aphid clones to account for the majority of differences although the performance of aphid populations from Trifolium and Lathyrus on their home plants was reduced through this effect. Overall, the populations from Cytisus and Pisum showed the greatest difference in performance on home compared to the average of away.
A different aspect of specialization can be addressed by comparing the performance of an adapted population with that of all other aphid populations on the same host plant: the “local vs. foreign” comparison (Kawecki and Ebert 2004). This reveals that the populations from Cytisus and Ononis have the highest performance on their local plant relative to aphids from other collection plants and the Vicia population the smallest (Fig. 4). Pea aphids from Cytisus and Ononis are often placed in different subspecies, as, less frequently, are those from Pisum. Our analyses to some extent support the functional distinctiveness of the Cytisus and Ononis populations.
Adding the interaction between collection plant and test plant to the statistical model already containing the home/away factor and its interaction with collection plant explained relatively little extra deviance, especially considering the number of degrees of freedom involved (Fig. 2). This means that there are relatively few idiosyncratic host preferences, cases in which a population from plant species A does much better or worse on plant species B in comparison with other populations for which this is also an away host. Interesting exceptions are the populations from Vicia and Trifolium, which exhibit very similar average performance profiles. The Vicia population appears to be composed of mostly Trifolium- and Pisum-adapted aphid clones and this can explain the similarity in average performance of these populations across all test plants. This contrasts with Frantz et al.'s (2006) study of French populations from Trifolium and Vicia which found them more genetically distinct. The interaction between collection plant and test plant also indicates that there is relatively little variation among populations in the breadth of the host-plant spectrum that the aphids are able to use. The population from Cytisus appears to have the narrowest range of plant species it can use followed by the population from Lotus (Fig. 5).
To obtain the maximal model we added the aphid clone by test plant interaction to this last model. For the survival and fecundity measures of fitness, just under half of the gene by environment interaction is accounted for by this term, and approximately two-thirds for the acceptance measure. We would expect this component to be large considering the number of degrees of freedom involved and our results underline the importance of individual clonal variation beyond that attributable to host-associated population. One process contributing to this term is the presence of aphid clones that we term migrants that appear to be on the wrong host plant. These aphid clones perform poorly on the plant from which they were collected, but have high fitness on one of the alternative test plants (other than Vicia). The population from Lathyrus is the most extreme case; only two-thirds of the aphid clones were adapted to this host. Clonal genetic variation includes both additive and nonadditive components and, provided the latter does not dominate, this variation may allow further evolution of host use. The within-population component in host use is greater than expected from a similar analysis by Via (1991a) for pea aphids from Trifolium and Medicago in eastern North America. She found that the within-population variation explained only 4% to 14% of the genotype by environment interaction for three measures of fitness.
Previously we studied the host-plant preference behavior of a subset of the aphid clones examined here (Ferrari et al. 2006). Individuals from eight aphid clones of each of the eight host-associated populations were placed in an arena and allowed to choose between all eight host-plants. Aphid clones showed strong preferences for their home plant and also for Vicia. At the time, we did not know whether this host-plant choice was adaptive, but the results here show a strong correlation between preference and performance across populations and a generally similar pattern in the breakdown of genetic variation in the two types of trait (Fig. 2). Caillaud and Via (2000) have previously demonstrated this correlation in North American populations from Medicago and Trifolium grown as agricultural forages. Our results show that adaptive specialization is not restricted to host plants grown extensively in agriculture, nor restricted to simple pairs of alternative hosts.
It is possible that not all adaptation to different host plants is due to genetic changes in the aphid. Virtually all aphids carry an obligate endosymbiont, Buchnera aphidicola, which provides its host with amino acids absent in its phloem diet. Buchnera is maternally transmitted with no evidence of horizontal transfer and functionally can be treated as an aphid organelle. However, aphids also carry secondary symbionts that are far more variable in their distribution. There are three main types found in pea aphid, all belonging to the (γ-Proteobacteria: candidatus Regiella insecticola (formerly known as U-type or PAUS), candidatus Hamiltonella defensa (T-type, PABS), and candidatus Serratia symbiotica (R-type, PASS) (Moran et al. 2005). These symbionts are mainly vertically transmitted, but sequence evidence shows that horizontal transfer also occurs (Russell et al. 2003). There is correlative and experimental evidence implicating these bacteria in affecting performance on different host plants (Tsuchida et al. 2002; Leonardo and Muiru 2003; Simon et al. 2003; Ferrari et al. 2004; Leonardo 2004; Tsuchida et al. 2004; Ferrari et al. 2007) in addition to effects on other traits (Montllor et al. 2002; Oliver et al. 2003; Scarborough et al. 2005). For example, aphid clones collected worldwide from Trifolium have a high likelihood of harboring R. insecticola (Tsuchida et al. 2002; Leonardo and Muiru 2003; Simon et al. 2003; Ferrari et al. 2004). A series of studies have experimentally manipulated the presence of R. insecticola to see if it influenced performance on Trifolium, but the results are variable and appear genotype specific (Leonardo 2004; Tsuchida et al. 2004; Ferrari et al. 2007). Further research is needed to understand the role of secondary symbionts in host-plant adaptation.
Is what we observe here the equilibrium population structure of a species consisting of a series of host-plant adapted races, or a snapshot of a taxon in the process of speciation? Pea aphid is native to the old world but the distribution and relative abundances of its host plants will have changed radically in the last 6000 years since the advent of agriculture, both because several of its host plants are major crops, but also because of anthropogenic landscape change affecting its wild hosts. It is possible that this changed environment has increased selection for population differentiation and in studying populations from a variety of host plants we are observing different stages in ecological speciation.
Alternatively, the pea aphid population structure may be at or near equilibrium. It is possible that the differentiation we observe represents a balance between gene flow and selection and that the presence of a universal host plant, Vicia, for which all host-plant adapted populations show a secondary preference and on which they perform well, may act as a channel for gene flow preventing further differentiation. The availability of Vicia will have increased through its domestication and agriculture might therefore have increased gene flow rather than decreasing it as suggested above. Many aphid clones collected on Vicia have a similar performance profile to those from Trifolium, whereas apparent Pisum and Medicago specialists were also collected on this host plant. It would be interesting to conduct a larger survey of aphids on Vicia to establish the degree to which aphid clones adapted to other host plants occur on this species. A similar case of a class of host plants acting as a genetic bridge between host-adapted populations has recently been reported by Craig et al. (2007). The sawfly Eurosta solidaginis forms differentiated populations on S. altissima and S. gigantea with hybrids having reduced fitness on most individuals of both parental species. However, certain genotypes, belonging to both Solidago species, are good hosts for hybrids, and hence may foster gene flow.
Host-plant adaptation can be facilitated by trade-offs between performances on different plant species (Futuyma and Moreno 1988; Jaenike 1990; Via 1990), although trade-offs per se are not a prerequisite for specialization (Fry 1996; Kawecki 1996). Across host-associated populations we see frequent negative genetic correlation between performances on different plant species. These may have arisen due to negative pleiotropy but also through other processes and such a pattern therefore contains little information on the underlying genetic architecture of host-plant use (Futuyma and Moreno 1988; Schluter 2000). Some insight into the latter can be obtained from the genetic correlations within host-associated populations (although with limitations discussed in Rausher 1988; Charlesworth 1990; Houle 1991; Fry 1993; Joshi and Thompson 1995). Using the mean survival of aphid clones on different host plants as our unit of replication we looked for evidence of trade-offs within populations. For some host-plant combinations we found negative correlations, but these were far less common than positive correlations and none reached significance when correcting for multiple comparisons. We thus find little evidence for genetic trade-offs that might affect the early stages of host-plant specialization, although we point out the relatively low statistical power of our analysis.
To conclude, this and our previous study of behavioral preferences (Ferrari et al. 2006) show that pea aphids in the south of England consist of a complex of populations adapted to different degrees to different species of host plant. We know from earlier work, particularly from North America, that host-plant adaptation has occurred on Trifolium and Medicago and our results show that it is more widespread, although complicated by the presence of a host plant on which all aphid clones do well and which may act as a bridge for gene flow between host-adapted populations. Future work, particular involving the molecular study of different populations, should enable us to say whether pea aphid population structure reflects a transient stage on the road to speciation, or an equilibrium between differentiation due to selection for host-plant specialization and homogenization caused by gene flow.
Associate Editor: J. Feder
ACKNOWLEDGMENTS
We are grateful to T. Evans, A. Faulconbridge, M. Harrison, K. Prior, J. Trakalo for technical assistance during the experiment. We would also like to thank R. Fuller and P. Vamvatsikos for growing thousands of plants. The Clarke family, P. Shaw and J. Whitby allowed access to their land during the aphid collections and D. Ferrari, L. Ferrari and J. West helped in collecting the insects. This work was funded by the Biotechnology and Biological Sciences Research Council grant D19263.
