Study area and studied species

The aphids and ants were collected on dunes on the Ishikari Coast. The dunes were covered by the dominant species Miscanthus sinensis (Hitchc.) Ohwi, Rosa rugosa Thunberg, Celastrus orbiculatus Thunberg, and Carex kobomugi Ohwi. Although many Q. dentata were distributed widely throughout the study area, a few oak trees on which the mutualism between T. quercicola and F. yessensis lasts until autumn have been observed over nine consecutive years from 1995 to 2003. Such trees are never found in the adjacent oak forest, perhaps because F. yessensis needs an open site for nest development. The nests near an oak tree are recognizable because their surface has mounds consisting of decayed stalks, blades and hulls of M. sinensis. Six trees (on average 3.1m tall), on which the mutualism occurred in every year were selected for the monitoring of the seasonal changes in the genetic structures of aphids and ants on each tree. With this monitoring, it is possible to determine whether the initial differences in the population structures between host trees are maintained across seasons. The location of each tree is shown in Figure 1.

Collection of aphids and ants

A preliminary study showed that the aphid population grew exponentially until August, followed by a steep decline by late September. Thereafter, the population increased in density again (Ito and Higashi 1991; Yao 2004). Aggregations of T. quercicola along the midrib of the upper or lower surface of a single leaf of Q. dentata were defined as a colony. The number of aphids and attending ants per aphid colony were counted before collection (N in Table 1). Collection was conducted three times: (1) 1st to 4th June 2004, hereafter referred to as spring samples and corresponding to population foundation by the first generation, (2) 31st July 2004, hereafter referred to as summer samples, and (3) 12th October 2004, hereafter referred to as autumn samples, which consisted of the oviparae and winged males. Four to eight aphids and two to five attending ants were collected from a single aphid colony. Two to eight colonies of each of the six trees were sampled three times through the seasons in 2004. The sample sizes of the three subsets were as follows: for the spring sample, a total of 33 colonies averaging 5.4 ± 0.7 SD aphids and 3.8 ± 1.0 SD ants (corresponding to 69% and 78% of the total number of aphids and ants per aphid colony, respectively); for the summer sample, a total of 43 colonies averaging 7.1 ± 1.0 SD aphids and 4.3 ± 0.6 SD ants (corresponding to 33.3% and 56.4% of the total number of aphids and ants per aphid colony, respectively); for the autumn sample, a total of 32 colonies averaging 7.7 ± 0.8 SD aphids and 2.6 ± 0.8 SD ants (corresponding to 43.9% and 78.7% of the total number of aphids and ants per aphid colony, respectively). No colonies were found in two trees (trees 5 and 6) in the autumn, probably due to the deterioration in host plant quality. All aphid colonies collected are listed in Table 1. The aphids and attending ants collected were stored in acetone (Fukatsu 1999) and held at -20 °C prior to the experiment.

Microsatellite analysis

To extract DNA, the whole aphid or hind femur of an ant was used. Genomic DNA was extracted following the Chelex procedure (Walsh et al. 1991), and resuspended in 15 µ″l of TE (10 mM Tris-HCl, 1 mM EDTA, pH 8). Five microsatellite loci were used to examine the genotypes of individual T. quercicola (Tq15, Tq17, Tq18, Tq23, and Tq26; Yao et al. 2003) and F. yessensis (Fy3, Fy4, Fy7, and Fy15; Hasegawa and Imai 2004, and FL12; Chapuisat 1996). Because many stutter bands emerged in some T. quercicola individuals amplified by Tq23, the design of the forward primer was changed to 5′-TCACACGCGCATACGATATT-3′. Forward primers were labeled with Beckman Dye fluorescence (Proligo). The polymerase chain reaction (PCR) amplifications and determination of allele sizes for both species were the same as described in Yao et al. (2003) and Hasegawa and Imai (2004).

Analysis of genetic data

Individuals with a failure in PCR amplifications, four for T. quercicola and 10 for F. yessensis, were excluded from all analyses. Because clonal reproduction in aphid populations probably leads to deviations from the Hardy-Weinberg equilibrium and linkage disequilibrium, all analyses except for clonal mixing analysis were carried out using only one aphid per genotype (Sunnucks et al. 1997). Only a few individuals with an identical genotype were detected in the ant populations; this allowed us to use the complete dataset of ants for genetic analyses.

The number of alleles, allele frequencies, and heterozygosity for the populations of the aphids and ants were calculated using the Microsatellite analyzer (MSA) 4.00 (Dieringer and Schötterer 2002). The gene diversity (h) of each microsatellite locus was calculated using the equation,

Linkage disequilibrium between loci and departure from the Hardy-Weinberg equilibrium at each locus were tested using the Fisher's exact probability test with a sequential Bonferroni correction for multiple comparisons (Rice 1989) in Genepop ( http://genepop.curtin.edu.au/) (Raymond and Rousset 1996).

Differences in genetic structure between trees and between seasons

An analysis of molecular variance (AMOVA; Excoffier et al. 1992) was conducted using Arlequin version 2.0 (Schneider et al. 2000) to partition the genetic variation into components attributable to differences among the specified hierarchical groups (FCT), among populations within hierarchical groups (FSC), and within populations (FST). Three criteria were used to calculate the genetic variation. First, the genetic variation was calculated for among trees, among colonies within trees, and within aphid colonies. Second, the genetic variation was calculated among seasons, among trees within seasons, and within trees. Third, the genetic variation was calculated among seasons, among aphid colonies with seasons, and within aphid colonies. In addition to AMOVA, pairwise F_ST with a sequential Bonferroni correction for multiple comparisons was calculated using MSA 4.00 to examine the genetic differences among trees divided by seasons. Relationships between genetic differentiation and geographical distance separating trees (isolation by distance) were examined in each season using Isolde in Genepop. Isolation by distance was tested with a Mantel procedure (10 000 permutations) by correlating pairwise F_ST with the natural logarithm of the straight-line distance (m) between pairs of trees.

In order to examine whether the genetic differences among host-associated populations, if any, vary seasonally for both the aphids and ants, F_ST values were analyzed with two-way ANOVA model in which F_ST between two consecutive seasons on each tree was treated as the dependent variable, while seasons (spring-summer and summer-autumn) and trees were treated as independent variables. The interaction between trees and seasons failed to be included in the ANOVA because of the lack of data from Trees 5 and 6 in the autumn. The F_ST value was transformed to arcsine square-root in order to satisfy the requirement of normality. Furthermore, to examine the microgeographical and seasonal genetic similarities among populations on the six trees, principal coordinate analysis (PCO) (Gower 1966) with PCO 2.0 (Iwata 2005) was used.

Although the information that samples of both species was collected from six different trees and their genotypes were distinguished individually, which gives a measure of genetic diversity per tree, the actual extent of assemblage of individuals characterized by a set of allele frequencies at each locus across all trees or seasons is not clear. Thus, dividing the total sample into clusters of individuals, each of which fits some genetic criterion that defines it as a group, would provide an estimate of the number of biologically realistic subpopulations (Pearse and Crandall 2004). The potential existence of K subpopulations across the study area was addressed using the Bayesian clustering approach implemented in the software Structure 2.0 ( http://pritch.bsd.uchicago.edu/software/structure2_beta.html) (Pritchard et al. 2000; Falush et al. 2003). We conducted two analyses to determine the effect of seasons on the subpopulations. First, the entire data set with the information on seasons not considered in the calculation was analyzed with assumed values of K ranging from one to 11 for T. quercicola and from one to six for F. yessensis. Second, the data set was analyzed each season separately with assumed values of K ranging from one to 10 for T. quercicola and from one to five for F. yessensis. As recommended in the manual for the software, an admixture model was chosen with the assumption that allele frequencies were independent in each population. For the selected model, the Bayesian posterior probabilities were calculated as an estimated Pr(X|K), where individual genotypes are assigned to a predefined number of clusters (K) in a given genotype (X). The program was run setting the length of the burn-in period and the number of MCMC replications after burn-in at 30,000 and 10⁶, respectively, and a maximum value of lnP(X|K) was obtained.

Seasonal changes in clonal mixing in T. quercicola colonies

The numbers of genotypes of aphids per season per tree per colony and per tree were calculated using the Groups/Summary command in the statistical software package JMP 5.0.1 J (SAS 2002). The clonal diversity of T. quercicola in an aphid colony and within a tree were calculated using s = G/n (Llewellyn et al. 2003), where G is the number of different genotypes in an aphid colony and n is the number of individuals examined. Additionally, the Shannon-Wiener (S-W) diversity index (H) with the binary logarithm was used to determine clonal diversity as follows:

Figure 1.

Distribution of trees used in the present study. Tree height (m) is given in parentheses. Distances among trees (m) are given as an actual measurement (solid line) and estimated measurement (broken line) of aerial photos.

Table 1.

Collection data for the aphid Tuberculatus quercicola and the ant Formica yessensis from the oak tree Quercus dentata.

cont

Allelic variation and genotyping

The average number of allelic variants detected per locus for tree-associated populations of T. quercicola and F. yessensis ranged from two (Tq17 and Tq18) to 15 (Tq15), with a mean of 5.8 ± 4.79 SD and ranged from two (Fy4) to nine (Fy3 and Fy7), with a mean of 6.2 ± 2.79 SD, respectively. The gene diversity estimates for populations of T. quercicola and F. yessensis ranged from 0.40 (Tq18) to 0.73 (Tq23), with an average value of 0.55 ± 0.13 SD and ranged from 0.43 (Fy4) to 0.74 (Fy7), with an average value of 0.61 ± 0.12 SD, respectively. Only two loci (Tq15 and Tq23) had private alleles (an allele found in only one subpoplation) (Hartl and Clark 1997) with a frequency of more than 1% in each locus (Supplemental Data 1). Heterozygosity deviations from the Hardy-Weinberg equilibrium to heterozygosity excess for each locus across populations were significant at two loci of T. quercicola (Tq15 and Tq23) and F. yessensis (Fy7 and Fy15) (Fisher' s exact probability test; P < 0.0083 for both T. quercicola and F. yessensis) (Supplemental Data 2). Linkage disequilibrium was not detected for any pair of loci across all T. quercicola populations and F. yessensis populations with the Bonferroni correction for multiple comparisons.

The combined five microsatellite loci of T. quercicola and F. yessensis had a high resolution for discriminating different genotypes within a colony (Table 1). All but six colonies (94%) contained multiple clones of T. quercicola across seasons, and almost all F. yessensis workers visiting T. quercicola colonies had different genotypes (Table 1). 317 genotypes within 726 aphids and 356 genotypes within 383 ants were detected across seasons. Although common genotypes across the six trees were not found in populations of both species, a single genotype of T. quercicola and 19 genotypes of F. yessensis were found to be common in four trees and in several pairs of trees, respectively.

Differences in genetic structure between trees and between seasons

For both of the populations of T. quercicola and F. yessensis, AMOVA indicated that significant levels of genetic partitioning was explained by trees (Table 2). Percentage of variation in analysis with partitioning the data among trees was higher in T. quercicola than in F. yessensis, which explained 10.5% and 4.1% of total genetic variation, respectively. Irrespective of setting the hierarchical subpopulation to trees or aphid colonies, partitioning the data among seasons provided a significant effect for F. yessensis but T. quercicola, indicating that season is the best predictor of genetic discontinuity for F. yessensis.

Pairwise comparisons of F_ST values between the tree-associated population in each season indicated that genetic differentiation among populations was greater in T. quercicola than in F. yessensis (the averages were as follows: for T. quercicola, for spring, 0.219 ± 0.127 SD, range 0.023–0.514, for summer, 0.159 ± 0.081 SD, range 0.020–0.329, for autumn, 0.151 ± 0.062 SD, range 0.067–0.201; for F. yessensis, for spring, 0.096 ± 0.067 SD, range 0.018–0.233, for summer, 0.053 ± 0.040 SD, range 0.004–0.129, for autumn, 0.045 ± 0.039 SD, range 0.002–0.108) (Table 3). Isolation by distance analysis showed that a significant positive relationship between F_ST and geographical distance (ln-transformed) was found for F. yessensis but not for T. quercicola (Figure 2).

Table 2.

AMOVA results for populations of (a) T. quercicola and (b) F. yessensis partitioned by trees and seasons nested with trees or colonies.

A significant difference in the level of genetic differentiation between two consecutive seasons was found in F. yessensis but not in T. quercicola (Table 4). For F. yessensis, pairwise F_ST between spring and summer samples was greater than between summer and autumn samples (averaging 0.135 ± 0.006 SE between spring and summer, averaging 0.021 ± 0.008 SE between summer and autumn).

PCO analysis showed that in spite of a drastic change of genetic difference in F. yessensis between spring and summer, the relative position among the six populations was maintained from spring to summer (Figure 3b). In contrast, the pattern of genetic differentiation in T. quercicola changed greatly between spring and summer (Figure 3a).

The results of Structure analysis for the entire data set including the label of seasons revealed that the number of cryptic populations (K) were 10 for T. quercicola and three for F. yessensis (Figure 4). Clear segmentation between the spring and summer populations was found in F. yessensis, but not in T. quercicola (Figure 4). Analyses for the sample belonging to each season showed that five to eight clustering and one to two clustering existed in the population of T. quercicola and F. yessensis, respectively.

Table 3.

Pairwise F_ST between host trees in each season.

Seasonal changes in clonal mixing in T. quercicola colonies

Two-way ANOVA indicated that both the G/n ratio and S-W index of T. quercicola in each colony decreased with the advance in season (G/n ratio, F_{2, 100} = 26.92, P < 0.0001; S-W index, F_{2, 100} = 4.97, P = 0.0087). Both the G/n ratio and S-W index were high in spring colonies (G/n ratio averaging 0.89 ± 0.04 SE; S-W index averaging 2.17 ± 0.11 SE) and low in autumn colonies (G/n ratio averaging 0.55 ± 0.04 SE; S-W index averaging 1.68 ± 0.12 SE) (Figures 5a and 5b). Significant effect on both the G/n ratio and S-W index was found in trees (G/n ratio, F_{5, 100} = 6.34, P < 0.0001; S-W index, F_{5, 100} = 6.23, P < 0.0001). The colony in which all T. quercicola individuals were assigned to different clones (i.e. a G/n ratio = 1) accounted for 67% (22 out of 33 colonies) in spring, 21% (seven of 43 colonies) in summer, and 13% (four of 32 colonies) in autumn (Table 1). Common genotypes that were found through all three seasons were detected only in T. quercicola and accounted for 4% (13 of the 317 genotypes).

Table 4.

Two-way ANOVA for the effects between seasons on the genetic differentiation in T. quercicola and F. yessensis. ^**P < 0.01

Differences in genetic structure between trees and between seasons

Analysis of microsatellite loci revealed that a significant genetic difference was maintained on the microgeographic scale in both T. quercicola and F. yessensis populations, and most of the genotypes were found only once on respective host trees throughout the seasons. Microgeographic genetic differences have been reported in a number of studies (De Barro et al. 1995; Sunnucks et al. 1997; Haack et al. 2000; Vorburger 2006), where the authors postulate that stochastic genetic drift or founder effects were mainly responsible for changes in the frequency of genotypes. In the present example, a marked genetic differentiation was evident among the host-associated populations separated by an average of 240m, despite the fact that all T. quercicola adults have wings. No isolation by distance, however, was found in T. quercicola populations (Figure 2). Moreover, the Structure analysis for T. quercicola in each season revealed a close match between genetic partitions and actual trees (Figure 4). Private alleles with a frequency of more than 1% at each locus were found in two loci (Tq15 and Tq23) (Supplemental Data 1). These results suggest that T. quercicola has such a low migratory ability that they cannot disperse among the host trees. Attending ants may also inhibit the dispersal of aphids because the aphids are always surrounded by the ants. Besides the limitation of gene flow, selective pressures under heterogeneous host environments would serve to maintain a high level of genetic differences. A number of studies have documented that the performance of herbivores with low migratory ability, such as gall-formers, leafminers, and scale insects, are particularly susceptible to the heterogeneity of their host plants, which sometimes leads to a large amount of genetic variance in insect hatch dates in relation to synchrony with host budburst (Feeny 1970; Akimoto and Yamaguchi 1994; Karban 1989; Komatsu and Akimoto 1995). In the present system, significant differences in honeydew excretion behavior were reported between T. quercicola populations on the respective host trees (Yao 2004). This fact supports that T. quercicola is affected by heterogeneous environmental conditions.

Figure 2.

Analysis for isolation by distance for seasonal populations of T. quercicola and F. yessensis.

Figure 3.

Principal coordinate analysis (PCO) showing the similarities among seasonal populations of (a) T. quercicola and (b) F. yessensis. PCO axis 1 explains 95.7% and 83.2% of the contribution and PCO axis 2 explains 4.2% and 12.7% for T. quercicola and F. yessensis, respectively. The numbers in the graphs indicate individual trees. Convex hulls were drawn to compare the relative similarities among six populations between spring and summer.

In contrast to T. quercicola, the average of pairwise F_ST for F. yessensis was much lower (Table 3). These results suggest that a weak genetic differentiation exists between trees, and populations of F. yessensis have a high relatedness to neighbors, as seen in Formica paralugubris type B, which is highly polygynous and has supercolonies (Chapuisat et al. 1997). In F. paralugubris B, the number of migrants entering a population per generation was 2.5, which implies that about 99.5 % of the queens are recruited from within the same nest. Chapuisat et al. (1997) suggests that the continuous isolation by distance in the supercolony may be due to nest budding or dispersal of sexual individuals to nearby nests. In the present study, all Structure analyses for F. yessensis indicated a partition with the most likely K below the actual value of trees (Figure 4). The discrepancy between the numbers of clusters and sampled trees is attributed to differentiation between populations corresponding to the two- dimensional model. Genetic differentiation between F. yessensis populations occurred at distance smaller than half of the width of the habitat (see Table 3 and Figure 1), indicating that isolation by distance follows the two-dimensional model (Rousset 1997). Isolation by distance refers to the idea that individuals with slow dispersal may be spatially distributed across a region. In this situation, allele frequencies vary gradually across the region. The underlying Structure model is not well suited for data from this kind of scenario, except for one-dimensional isolation by distance (Pritchard et al. 2000). The inseminated new F. yessensis queens have been observed returning to the natal nests after mating (Higashi 1983), supporting that genetic differentiation occurred within close nests and can be maintained by limited gene flow. Thus, newly founded nests will therefore tend to be genetically similar to the nest from which they originated.

Figure 4.

Summary plots of estimates of Q for T. quercicolo and F. yessensis calculated with the entire data set and with the data divided into each season. Q indicates the estimated membership coefficients for each individual in each cluster. Each individual is represented by a single vertical line broken into K colored segments with lengths proportional to each of the K inferred clusters. The graph on the right indicates Bayesian posterior probabilities estimated for K. Higher values represent a higher probability of the assumed partition.

A striking, significant genetic difference was found between spring and summer samples in F. yessensis but not in T. quercicola (Tables 2 and 4, Figures 3 and 4), indicating that changes of genetic composition occurred in F. yessensis nests. The demographic pattern of F. yessensis was characterized by a sharp increase from late July to late August. Ito (1973) also observed many pupae and new workers in F. yessensis nests from mid-June to late September. A large genetic difference in attending workers over summer may therefore be due to eclosion of the first workers and addition of newly-eclosed workers to the foraging force.

It should be noted that even after an alternation of antattended workers, a significant microgeographic genetic difference and isolation by distance were consistently detected among F. yessensis populations on the host trees through the seasons (the averages of pairwise F_ST: 0.096 for spring, 0.053 for summer, 0.045 for autumn). Moreover, PCO analysis of F. yessensis populations showed that after the addition of newly-eclosed workers, the pattern of genetic similarity between ant populations on the host trees was kept almost constant (Figure 3b). These findings imply that new nests are founded close to natal nests near trees. A number of studies have documented that ant nests often move in response to seasonal shifts in weather (Gordon et al. 2001; Heller and Gordon 2006), disturbances, and changes in food availability (Holway and Case 2000). Furthermore, polydomous colonies that live in more than one nest, bud into many new nests, which may enable colonies to maintain advantageous nutritive, reproductive, or micro-environmental conditions (Banschbach and Herbers 1996, 1999). Cherix (1987) examined the diet composition of F. yessensis in our study region and found that the ants depend mainly on aphid honeydew as a food resource. In ant nests experiencing difficulty in foraging new protein resources, inheritance of honeydew resources in the same nest would be advantageous and favored by natural selection. Given that some of the mounds near the study trees where F. yessensis workers are actively entering and exiting the nest have been observed not to dislocate their position for nine consecutive years, new nests appear to be founded close to natal nests near trees where honeydew could be available during all seasons. Experimental studies have shown that in F. paralugubris B, new queens can mate and stay in their natal nest or seek adoption in a foreign nest after a mating flight (Cherix et al. 1991). If F. yessensis queens have a similar philopatric behavior, genetic similarities may be kept in a nest that exploits an aphid population on a host tree as a honeydew resource. It is suggested that the limited distribution of susceptible Q. dentata trees enables aphid colonies to persist on a limited number of trees for a long time, resulting in long-lasting mutualistic interactions between T. quercicola aphids and the attending ant F. yessensis on such trees.

Seasonal changes in clonal mixing in T. quercicola

The present study also showed that a high level of clonal mixing occurred in T. quercicola colonies during summer. Both the G/n ratio and S-W index of T. quercicola within a colony were high in the spring and slowly decreased during summer and into autumn, indicating that aphids of the same genotype gradually increased within a single colony. This is due to an increase by parthenogenesis or selection among clones. Several phenomena, such as high temperatures during summer, a decline in nutritional quality of leaves, and an increase in natural enemies, have been considered as seasonal factors affecting the dynamics of the aphid population (Hales et al. 1997; Karley et at. 2004). In our previous studies, T. quercicola populations increased exponentially until early August (Yao 2004), and a following significant reduction was found in both total free amino acid concentration in phloem sap of Q. dentata (Yao and Akimoto 2002) and fecundity of aphids (Yao et al. 2000; Yao 2004). If there is interclonal variation, selection may favor some types of clones equipped with a tolerance to high temperatures or high fecundity through interclonal competition. A number of studies demonstrated that under the seasonal selective pressures, a few clone types predominantly increased as the season progressed (Carvalho 1987; Carvalho and Crisp 1987; De Barro et al. 1995; Fuller et al. 1999; Vorburger 2006). In the present study, 13 common clones were found throughout the seasons, but their frequency in autumn (12 %) was lower than reported for other aphid species (Fuller et al. 1999; Vorburger 2006). This may be because of the differences in selective pressures acting on maintenance of polymorphisms between monocultural host plants and heterogeneous long-lived trees.

Figure 5.

Seasonal changes in the clonal diversity based on (a) G/n ratio and (b) S-W index within a T. quercicola colony. Mean ± SE. Data points with different letters indicate that a significant difference between seasons was found by multiple comparisons (Tukey's HSD test).

This study showed that a substantial microgeographic genetic differentiation was found in the aphid T. quercicola, while the attending symbiotic ant, F. yessensis had a lower genetic variation among the studied trees. As a consequence of limited suitable sites for T. quercicola, trees where honeydew was available all year round were occupied by F. yessensis over the same period, and the tree could be heritable at the ant colony level. Tracking the seasonal movement of F. yessensis workers with a mark-recapture experiment would contribute to a comprehensive understanding of the evolutionary ecology of aphid-ant mutualisms.