Local co-adaptation leading to a geographical mosaic of coevolution in a social parasite system

Authors


Birgit Fischer, LS Biologie I, Universität Regensburg, Universitätsstr. 31, D – 93040 Regensburg, Germany.
Tel.: 49 941 943 2149; fax: 49 941 943 3304;
e-mail: birgit.fischer@biologie.uni-regensburg.de

Abstract

The geographical mosaic theory of coevolution predicts differences in the advance or trajectory of the coevolutionary process between local communities due to their composition and the strength of ecological selection pressures through competition and resource availability. In this study, we investigate local co-adaptation in different populations of a social parasite. We conducted cross-fostering experiments to test for interpopulational differences in raiding efficiency between various populations of a slave-making ant and the defence abilities of local hosts. Here, we demonstrate that the success of raids strongly depends on the combination of populations of the parasite Harpagoxenus sublaevis and its host Leptothorax acervorum, indicating very localized coevolution. We found no absolute differences between slave-maker populations; the outcome of an encounter depended more on whether the two opponents occur in sympatry or allopatry. Furthermore, this study supports the results of our earlier work, that the unparasitized English L. acervorum population is most aggressive against the parasite.

Introduction

Over the past 20 years, many convincing examples of coevolution between parasites and their hosts have accumulated, demonstrating the importance of coevolution as an evolutionary process shaping the organization of communities and influencing the diversification of life (Thompson, 1999a). These studies revealed that for the understanding of the coevolutionary dynamics, macro-evolutionary effects such as cospeciation can be less enlightening than micro-evolutionary processes (Thompson, 1994). According to the geographical mosaic theory of coevolution, the outcome of interactions and the degree of specialization vary among physical and biotic environments and depend on the community context in which the interaction takes place (Thompson, 1994, 1999b). Only very few pairs of interacting species have identical degrees of specialization or adaptation in different communities. Local populations may go extinct or occur outside the geographical range of the other species and may lose their adaptations for the interaction. The result is a mosaic of populations with different suites of adaptations. Apart from the different strength of selection pressures, coevolution can follow different trajectories (Thompson, 1994; Foitzik et al., 2001; Foitzik & Herbers, 2001). Strong geographical differences in host and parasite strategies can occur, with each pair of interacting species engaging in a different arms race from every other pair. Such a very localized scenario can be contrasted with a more universal pattern, where only the progression of the arms race differs in response to parasite pressure.

Avian brood parasites have been extensively studied as a model system for coevolution (Rothstein, 1975, 1990, 2001; Brooke & Davies, 1988; Davies & Brooke, 1989; Lotem et al., 1992; Soler et al., 1998; Winfree, 1999), but only very little is known about the effect of social parasitism in eusocial insects. Nevertheless, there are a few studies that uncovered coevolutionary interactions between social insect parasites and their hosts, and thus offer a new model system for the study of coevolution (Foitzik et al., 2001, 2003; Hare & Alloway, 2001). The phenomenon of social parasitism, where a eusocial species is exploited by another such species (Seifert, 1996), is common in ants, where more than 300 species exhibit this life style (Buschinger, 1993). In contrast to most avian parasite systems, host and parasite are, following Emery's rule, often closely related species (Emery, 1909; Wilson, 1971). This is because the more tightly related host and parasite are, the more likely they are to be compatible, in terms of communication, brood care, etc. This close phylogenetic relationship can be profitably explored for the study of coevolutionary interactions between parasite and host. Both opponents show similar generation times and population sizes and should therefore react with comparable evolutionary speed to selection pressures they impose on each other.

Social parasitism can be either temporary or permanent, and the slave-making ants studied here commonly belong to the permanent social parasites (Buschinger, 1970; Hölldobler & Wilson, 1990). They obligatorily depend on the work of slaves, which are workers of a different ant species, and replenish their labour force via regular slave raids. The obligate slave-making ant Harpagoxenus sublaevis (Nylander 1852) parasitizes three species of the genus Leptothorax, L. acervorum, L. muscorum and only rarely L. gredleri. The social parasite and its hosts are found in pine forests throughout Eurasia. The distribution pattern of these species, which is important for the present study, shows that only L. acervorum occurs on the British Isles, while the slave-maker H. sublaevis and its second main host species L. muscorum were never found there (Collingwood, 1971; Radchenko et al., 1999).

Obligate slave-makers, presumably owing to the behavioural and morphological specializations for raiding, depend on slaves for all necessary tasks of colony maintenance, such as foraging or brood care (Wilson, 1971; Buschinger et al., 1980; Stuart & Alloway, 1983). Thus, their colonies typically contain a large proportion of slaves in the total worker force (Wilson, 1975). During the raiding season in late summer, slave-maker workers attack neighbouring host colonies and salvage the host brood, which will be enslaved after emergence from the pupae in the slave-maker nest (Viehmeyer, 1908; Buschinger, 1966a, b, 1968, 1974a, 1980). During slave raids, H. sublaevis uses a specially evolved chemical weapon from the Dufour's gland, the so-called propaganda substance. When smeared on host workers, this substance induces intra-colonial fights among host workers (Regnier & Wilson, 1971; Buschinger, 1974a, b; Allies et al., 1986), and previous behavioural observations demonstrated that many host workers are killed during these fights (Foitzik et al., 2003). Due to the destructiveness of the slave raids, the impact on a host population can be substantial, depending on the number and size of slave-maker colonies in the community, the virulence of the social parasite and its raiding frequency (Foitzik et al., 2001; Foitzik & Herbers, 2001; Herbers & Foitzik, 2002). Studies on the effect of slave raids on attacked host colonies revealed only a very low survival rate of attacked host colonies (Foitzik et al., 2001; B. Fischer, unpublished). This substantial impact of the parasite on a host population should provoke counter-adaptations in the host.

In this study, we investigate the degree of specialization on the local opponent in different populations of the small European slave-making ant H. sublaevis and its main host species L. acervorum, by focusing on the outcome of interactions in different communities. We used demographic data and behavioural experiments to compare two parasitized host populations and one parasite-free population to assess how the coevolutionary mosaic is influenced by parasite pressure and local co-adaptation. We included a demographic analysis, because previous studies demonstrated both interpopulational variation in social organization that is in the number of reproducing queens per colony (Heinze et al., 1995a, b), and an impact of social structure on the efficiency of anti-parasite defence (Foitzik et al., 2003). We conducted raiding experiments in the laboratory in a cross-fostering design to analyse parasite efficacy and destructiveness during slave raids and the effectiveness of nest defence of various host populations.

Material and methods

Study system

The small European slave-making ant H. sublaevis is distributed throughout the boreal region of Western Eurasia, but does not occur in Great Britain (Collingwood, 1971; Radchenko et al., 1999). The hosts L. acervorum and L. muscorum populate the boreal region of the Palaearctic, but L. muscorum did not colonize the British Isles (Collingwood, 1971; Radchenko et al., 1999). Because of these distribution patterns, our analysis focuses on interactions between the social parasite and its main host, L. acervorum.

Colonies of the slave-maker H. sublaevis were collected in Bavaria, Germany and South Tyrol, Italy. In Germany, ants were obtained from pine forests close to Abensberg (altitude: 400 m) in the summers of 2001 and 2002. The Italian study site near Innichen (1300 m) was sampled in June 2002. We used host colonies of L. acervorum from Germany, Italy and England. The German colonies were collected in the summers of 2001 and 2002 at the same site as the H. sublaevis colonies. Italian host colonies were obtained in 2002 from two populations, the one near Innichen and a second one near Kastelruth. Ants were collected in England in July 2000 and April 2002 at a site in the New Forest, close to Norleywood, Hampshire (1 m) and additionally in 2002 at a second site in Santon Warren, Norfolk (27 m). At all three study sites, the habitat was a pine forest with an undergrowth of heather, blueberries and ferns. For the behavioural experiments, we used only slave-maker and host colonies that were collected in 2002.

Laboratory maintenance

All colonies were censused in the laboratory and housed in artificial nests inside small plastic boxes with a moistened plaster floor (Buschinger, 1974a; Heinze & Ortius, 1991). During the test period from August to September 2002, the ants were kept in an incubator at 20 °C for 12 h light, and 10 °C for 12 h dark. Twice weekly all colonies were fed with diluted honey and pieces of cockroaches.

Raiding experiments

We observed slave raids in the laboratory in a cross-fostering design using slave-maker colonies from Germany and Italy and L. acervorum host colonies from Germany, Italy and England (in the two latter localities both subpopulations were tested). We abbreviated the different host and parasite populations as follows: Germany = G, Italy = I and England = E. We chose 37 host colonies from each population with a similar demographic composition for the raiding experiments, because size differences could influence the efficiency of nest defence. Therefore, host colonies from the three populations did not differ in the number of queens, workers or brood items they contained [Kruskal–Wallis test N = 111, P > 0.10]. Each of the 111 host colonies was used only once. In contrast, each of the 21 and 16 slave-maker colonies from Germany and Italy, respectively, was tested three times, once against a host colony from each population in a randomized order. As a result, we staged a total of 111 slave raids.

To ensure that there was no influence of testing period, which was slightly later than the natural raiding season in the field (Buschinger, 1968), we tested for differences in the amount of brood in host colonies and the most important behavioural parameters between the first two and the last 2 weeks of the testing period. These tests show that there was no influence of the time when the raids took place [Mann–Whitney U-tests, P > 0.05].

All behavioural experiments were conducted in arenas of 0.43 × 0.27 × 0.16 m with a moistened plaster floor to regulate humidity. The slave-maker colony was placed in the arena 12 h before the start to allow slave-makers to get accustomed to the arena. We started the experiment by introducing a host colony to the arena at a distance of 40 cm from the slave-maker colony. As an escape option for the hosts, we additionally provided an empty nest site in the arena. We observed the reaction of host ants and the behaviour of the slave-maker workers by direct observation or videotaping. We terminated each trial 24 h after all host brood was captured or had been evacuated by host workers. In the final analysis, we recorded the number of dead and injured individuals of each species and caste (i.e. lost legs or antennae in fight), as well as the number of raided larvae and pupae. We counted a raid as successful if the slave-maker managed to obtain at least one piece of brood. We focussed on five parameters characterizing the behaviour during the raids and the outcome of the slave raids: percentage of brood successfully raided by the slave-maker, percentage of brood saved by the host colony to a secure site, percentage of host workers that escaped to this secure site, percentage of slave-maker workers killed or severely injured and number of attacks (i.e. biting) by host workers against the first five intruding slave-maker workers. Although the first two parameters were clearly associated with each other due to abandonment of some brood items in the raiding arena, they do not add up entirely.

Statistics

Our demographic data were not normally distributed and thus were analysed with nonparametric tests. In behavioural experiments, we controlled for population differences in demographic composition of host colonies, but we were unable to adjust all parameters for the two slave-maker populations. In the experiments, H. sublaevis colonies from Italy contained a higher slave-maker/host worker ratio (MWU test, U = 1090.5, P < 0.05). We corrected for this difference by using the number of slave-maker workers per host worker (SM/host) as a cofactor in the statistical analysis. We used a general linear model to perform a repeated measures anova using the cofactor above, although most variables did not conform to a normal distribution. Deviation from normality is only problematic for analysis of variance when caused by kurtosis, while the effects of skew on F can generally be ignored (Box & Andersen, 1955; Mardia, 1971; Lindman, 1974; Tabachnick & Fidell, 2000). For variables with significant deviation in kurtosis, we corrected all P-values according to Box & Andersen (1955) by calculating the degrees of freedom as 1 + Ku/N, where Ku is the kurtosis of the variable. In the following, only the corrected P-values are given. Because we tested each slave-making colony against a host colony from all three populations, the within-subject factor of our repeated measures anova was ‘host population’. All statistical tests were performed with the program Statistica© (StatSoft, Inc., Tulsa, OK, USA). We abbreviate the post hoc Sheffé test (PS test) and the post hoc Fisher LSD test (LSD test).

Results

Demography of slave-maker nests and parasite prevalence

In the summer 2002, we collected a total of 67 and 21 H. sublaevis colonies in Germany and Italy, respectively. Slave-maker colonies from both sites contained at equal rates either L. acervorum or L. muscorum slaves or slave workers of both species and therefore there was no interpopulational difference in terms of host species usage (inline image = 0.29, n.s.).

At our study sites, we found slave-maker/host nest ratios of 1 : 10 in Germany and ratios between 0 (plot A near Kastelruth) and 1 : 3.3 (plots B, C near Innichen) in Italy. When we include both Italian subpopulations, relative slave-maker frequencies did not differ between Germany and Italy (inline image = 0.29, n.s.). However, when we compared only the parasitized communities from Germany and Italy, we found higher social parasite prevalence in Italy (inline image = 7.54, P < 0.01). Slave-maker colonies from the Italian population were larger (number of slave-maker workers and number of slaves; MWU test, U = 499.5, P < 0.05), but less productive (number of larvae per slave; MWU test, U = 226.0, P < 0.001). In both social parasite populations slave number increased significantly with the number of slave-maker workers (regression on log-transformed data, Germany: r = 0.78, t63 = −3.42, P < 0.01; Italy: r = 0.76, t19 = −2.29, P < 0.05). Interestingly, we found a significant difference in the Y-intercept between Germany and Italy (t-test on log-transformed data, t30 = 3.2, P < 0.05), indicating that Italian slave-maker queens manage to usurp host colonies with four times more host workers than German parasite queens.

Colony demography of host populations

Between 2000 and 2002 we collected a total of 562 free-living L. acervorum colonies; 250 colonies in Germany, 60 colonies in Italy and 252 colonies in England. Colonies from the three different populations varied in worker number (mean: G, 54.1 ± 3.4; I, 63.0 ± 6.8; E, 68.5 ± 3.4; KW test, Hadj,2 = 15.7, P < 0.001); Germany host colonies were smaller than those from England (PS tests: G–I, P > 0.05; G–E, P < 0.0; I–E, P > 0.05). While the two Italian subpopulations did not differ in worker number (MWU test: U = 362.5, n.s.), slightly more workers were found in the less dense English subpopulation Santon Warren (MWU test: U = 2009.5, P < 0.05). In Germany, we collected smaller colonies in 2001 than in 2002 (MWU test: U = 5524.5, P < 0.05). We also uncovered interpopulational variation in social structure. The percentage of monogynous colonies (with only one functional queen in the colony; Hölldobler & Wilson, 1990) decreased from Italy to Germany and England (Italy 80%, Germany 54%, England 43%; inline image = 29.5, P < 0.001; PS test: G–I, P < 0.001; G–E, P < 0.05; I–E, P < 0.001). However, the social organization did not vary between subpopulations or years (MWU tests: n.s.). Besides, we noted strong interpopulational differences in nest density. While the New Forest population in England contained more than four nests per m2, colony densities were much lower at the second English site and both in Germany and Italy (<0.5/m2).

Raiding experiments

Influence of demography and social structure

Larger slave-maker colonies were more destructive and more successful during raids. We found a positive correlation between the number of slave-maker workers in the colony and the number of host workers killed and injured during a raid (Spearman Rank correlation: rs = 0.33, t101 = 3.51, P < 0.01). Larger parasite nests also obtained more host brood (rs = 0.39, t101 = 4.31, P < 0.001). As shown above, due to the large size difference between Italian (mean N of slave-maker workers = 29.4 ± SE 6.1) and German (15.6 ± SE 2.5) social parasite colonies, we could not control for this demographic parameter in our raiding experiments. Because we were interested in interpopulational variation in slave-maker raiding efficiency in absence of confounding demographic differences, as described above, we statistically adjusted for interpopulational size differences by using the cofactor ‘slave-maker workers per host worker’. Social structure of the host colony, i.e. the number of queens in a colony, did not influence the effectiveness of anti-parasite defence (MWU tests, n.s.).

Influence of slave-maker population

Our first main result is that the behavioural experiments uncovered no general differences in the raiding efficiency of slave-maker colonies from the two different populations in Germany and Italy (Tables 1 and 2; repeated measures anova, F1,28 = 0.49, n.s.). All post hoc analyses are summarized in Table 3.

Table 1.  Summary of the results of the repeated measures anova analysing behavioural differences in slave raids depending on social parasite and host origin.
EffectFd.f.1,d.f.2P-value
Cofactor (=slave-maker workers per host worker) 8.631,280.007
Categorical predictor (slave-maker population) 0.491,280.490
Within-subject factor 1 (host population) 3.302,560.044
Interaction effect (slave-maker population × host population) 3.442,560.039
Table 2.  Summary of values for variables measured during the slave raids, considering the parasite and host populations separately and without regard to the source of the interacting species.
 Slave-maker populationHost population
GermanyItalyGermanyItalyEngland
  1. Mean values and standard errors were corrected with covariate ‘slave-maker worker per host worker’.

Brood raided (%)66.4 ± 5.167.7 ± 5.872.3 ± 6.174.3 ± 6.663.3 ± 6.8
Brood saved (%)31.0 ± 5.231.1 ± 5.927.4 ± 6.227.3 ± 6.930.6 ± 6.8
Host workers escaped (%)26.3 ± 3.918.8 ± 4.421.7 ± 4.716.5 ± 5.220.9 ± 5.2
Slave-maker killed + injured (%)12.2 ± 2.015.3 ± 2.311.0 ± 2.08.6 ± 1.817.2 ± 2.3
Attacks on slave-makers1.2 ± 0.11.2 ± 0.11.1 ± 0.11.0 ± 0.11.3 ± 0.1
Table 3.  Summary of the results of the Fisher LSD post hoc tests on the outcome of slave raids demonstrating the interaction between slave-maker and host origin.
Host populationSlave-maker populationBrood raided (%)Brood saved (%)Host workers escaped (%)Slave-maker killed + inuredAttacks on slave-maker
  1. P-values of the LSD tests are given for all six pairs of social parasite and host populations. The first column gives the factor held constant in the analysis, while the second column shows the source populations tested against each other (G = Germany, I = Italy, E = England), i.e. in the upper part we tested within each host population for differences between the two slave-maker populations, while in the lower part differences within each slave-maker population were analysed in respect to the three host populations.

GermanyG–I0.580.540.0040.810.0001
ItalyG–I0.0020.0010.0230.610.0002
EnglandG–I0.510.950.100.430.06
Slave-maker populationHost population     
GermanyG–I0.480.330.540.930.06
I–E0.980.340.930.850.09
G–E0.490.980.610.910.0003
ItalyG–I0.370.360.910.820.002
I–E0.370.190.630.370.005
G–E0.570.690.710.500.81

Influence of host population

During raids, host workers from all three populations were observed to attack the social parasites with no obvious differences in behaviour. Moreover, slave raids against host colonies from the two subpopulations each in England and Italy did not differ and we consequently pooled over subpopulations (MWU tests: n.s.). However, we found a significant influence of the host population (England, Germany and Italy) on the outcome of slave raids in our anova analysis (Table 1; repeated measures anova, F2,56 = 3.30, P = 0.04). When investigating the behavioural parameters involving the host in more detail, we found no differences between the host populations in their ability to save brood, to escape or to kill and injure slave-makers (LSD tests: n.s.; see also Table 2). In contrast, host populations differed in how often they attacked intruding slave-makers with the English host colonies being most aggressive against slave-makers from both sites (Fig. 1 and Table 2; LSD tests: G–I, n.s.; G–E, P < 0.05; I–E, P < 0.05).

Figure 1.

Number of attacks (i.e. biting) by host workers against the first five intruding slave-maker workers. Least square mean ± SE are given and were computed for covariates at their mean values.

Interaction effects

In contrast to the few general interpopulational differences, the outcome of slave raids strongly depended on interaction effects between parasite and host populations (Table 1; repeated measures anova, F2,56 = 3.44, P < 0.04). In Table 3 we first summarized all effects from the viewpoint of the three host populations, followed by the perspective of the two slave-maker populations.

German host colonies did not save more or less brood in raids against sympatric or allopatric slave-maker colonies (Fig. 2; LSD test: n.s.). However, more host workers escaped in raids by the sympatric than by the allopatric social parasite (Fig. 3; LSD test, P < 0.01) and the hosts attacked sympatric slave-makers less often (Fig. 1; LSD test, P < 0.001). In the Italian host population we found a different pattern in that the amount of brood the slave-maker obtained or the host salvaged strongly depended on sympatry or allopatry of the opponents. Confronted with a sympatric slave-maker the Italian host colonies managed to save significantly less brood but the social parasite obtained more (Fig. 2; LSD tests, P < 0.001, P < 0.01). As in the German community, Italian hosts attacked intruding sympatric slave-makers less severely (Fig. 1 and Table 3; LSD test, P < 0.001). Furthermore, fewer Italian host workers escaped in raids by the sympatric slave-maker (Fig. 3; LSD tests, P < 0.05). For experiments with colonies from the unparasitized English host population all trials represented inevitably allopatric combinations and we found that the outcome and behaviour during raids did not vary with slave-maker origin (Fig. 2; LSD tests, P > 0.05). There was a tendency for English hosts to attack intruding German slave-makers more severely than slave-makers from Italy (Fig. 1; LSD test, P = 0.06), and in raids against this parasite population slightly more host workers escaped (Fig. 3; LSD test, P = 0.10).

Figure 2.

Effectiveness of slave-maker colonies measured as the percentage of successfully raided host brood contrasted by the percentage of brood salvaged by the host for all six pairs of host and slave-maker populations. Least square mean ± SE are given and were computed for covariates at their mean values.

Figure 3.

Percentage of host workers that escaped during a slave raid for all six pairs of host and slave-maker populations. Least square mean ± SE are given and were computed for covariates at their mean values.

German slave-maker colonies fared equally well in raids against host colonies from all three populations; they managed to raid a similar fraction of the host brood and also the number of killed and injured slave-makers did not vary with host population (Table 3; LSD tests, P > 0.30), although the intensity of defence differed with host origin. Intruding parasite workers were extremely rarely attacked by the sympatric German host colonies, more often by the Italian hosts and even slightly more by English hosts (Fig. 1 and Table 3; LSD tests: G–I, P = 0.06; G–E, P < 0.001; I–E, P = 0.09). For the Italian slave-maker we found the same pattern as there were no differences in the outcome of the raids in respect to all behavioural categories we included (Table 3; LSD tests, n.s.), except for the frequency of attack against intruding parasites. Once more, Italian slave-makers trying to initiate a raid were confronted with significantly less aggression by sympatric host colonies compared with both allopatric populations (Fig. 1 and Table 3; LSD tests: G–I, P < 0.01; G–E, n.s.; I–E, P < 0.005).

Discussion

This study demonstrates that the interactions during slave raids by the social parasite H. sublaevis against its main host L. acervorum critically depend on the origin of slave-maker and host. Our raiding experiments are the first to reveal adaptations of brood parasites to their local hosts. A cross-fostering design allowed us to explicitly test whether the outcome of slave raids, which is the crucial interaction between slave-making ants and their hosts, depends on sympatry or allopatry of the two opponents. This possibility to investigate local adaptation is lacking for many host–parasite systems. For example, cross-fostering laboratory experiments on the highly analogous systems of avian brood parasites are often strongly constrained by the fact that these birds often cannot be reared in captivity (N.B. Davies, pers. comm.). Our study further shows geographical variation in the coevolutionary interactions between parasite and host, supporting the geographical mosaic theory of coevolution, for which empirical evidence is only slowly accumulating. Investigations on the dynamics of species interactions will not only be crucial to our understanding of the coevolutionary process itself, but have profound implications for host–pathogen dynamics, which also influence human health and welfare in a multitude of ways (Thompson, 1999a).

Recent coevolutionary models and empirical studies (e.g. Thompson, 1994; Gandon et al., 1996; Nuismer et al., 1999; Parker, 1999) demonstrated the importance of analysing species interactions over broad geographical scales. However, only in few host–parasite systems, as in the current study, adaptations to a local opponent have been found. For example, the outcome of slave raids of the North American slave-maker Protomognathus americanus depends on the origin of the parasite and its host L. longispinosus (Foitzik et al., 2001), but not on interaction effects between them. In this North American system only the progression of the coevolutionary arms race differs between communities. We suggest that this discrepancy between the otherwise very comparable systems is due to differences in gene flow and genetic structuring of host and parasite populations. Protomognathus americanus queens are generally fully winged, while 99% of H. sublaevis colonies contain primarily wingless queens with very low dispersal capabilities (Buschinger & Winter, 1975). Indeed unpublished data show high levels of gene flow among P. americanus populations, but strong genetic differentiation between H. sublaevis populations (M. Brandt & B. Fischer, unpubl.). In a few other host–parasite systems, however, co-adaptation was demonstrated. Lively & Dybdahl (2000) revealed local adaptation in a prosobranch snail and its parasite, where the parasite exhibits significantly higher infection rates in sympatric compared with allopatric host populations. Similarly, cross-inoculation trials on the plant host–pathogen interaction between Linum marginale and Melampsora lini indicated adaptation of the parasite to its local host population (Thrall et al., 2002).

Our behavioural experiments revealed local parasite adaptation with respect to avoidance of host attack at our two study sites. Host aggression towards intruding slave-makers critically depended on whether the two opponents occurred in sympatry or allopatry, irrespective of whether we focus on the German or Italian community. We interpret the lower aggression against sympatric slave-makers with selection on the parasite to mimic the colony odour of its local opponent, as was suggested for other social parasites (Kaib et al., 1993; Elmes et al., 1999). We currently investigate the colony odour of various slave-maker and host populations by gas chromatography to study potential interpopulational differences in cuticular hydrocarbon profiles.

In contrast to our findings on host aggression, other biologically relevant parameters showed different patterns at the two sites where the host co-occurs with its social parasite. Only the Italian host colonies suffered more severely in raids by their sympatric slave-maker, in which they salvaged less brood and fewer host workers escaped. Our demographic data further show a much more efficient colony-founding tactic by Italian slave-maker queens, which manage to obtain four times more slaves during host colony usurpation than German parasite queens, although mean host colony size does not differ between sites. Thus in the Italian community, the slave-maker appears to win the arms race against its local host, which is also mirrored in significantly larger slave-maker colonies and higher parasite prevalence in the field.

In contrast, although German host colonies were less aggressive towards the sympatric social parasites, they fared much better against them in that about three times as many host workers escaped from a slave-maker attack. This result is corroborated by earlier findings that German host colonies react less strongly to chemical manipulation by a propaganda substance from the Dufour's gland of their local opponent (Foitzik et al., 2003). Compared with other slave-maker populations, German social parasites appear to be less virulent in that slightly more host workers managed to escape during slave raids and slave-making queens overwhelm smaller host colonies than their counterparts from Italy. Furthermore, compared with Russian slave-makers, their propaganda substance is less effective, as it elicits fewer fights among host ants (Foitzik et al., 2003).

Interestingly, the outcome of slave raids against colonies from the English host population did not vary with slave-maker origin. Because all combinations including the unparasitized English host have to be all allopatric per definition this result supports our findings on local adaptation at the two other sites in Germany and Italy.

While in the North American slave-maker P. americanus and its hosts, the effectiveness of parasite and host populations co-vary (Foitzik et al., 2001), we find within our European study sites no association between parasite virulence and host resistance. The geographical mosaic theory of coevolution suggests that species interactions differ between locales due to the age of the interaction and varying species composition or ecological pressures (Thompson, 1994, 1997, 1999b; Benkman et al., 2003; Choo et al., 2003; Nuismer et al., 2003). Hosts that co-occur with their parasites only for short evolutionary times should exhibit lower defences compared with populations with ancient parasite sympatry, as was demonstrated for avian brood parasites (Soler & Moller, 1990). In accordance with this prediction, the Italian host population, which exhibits the lowest defence is situated in a high alpine valley in South Tyrol at 1300 m a.s.l., a location that was certainly post-glacially colonized by the wingless parasites much later than the German lowland site. Life history theory suggests trade-offs between different costly traits (Roff, 2002) and parasite defence characters can theoretically only be maintained under favourable ecological conditions for the host. The evolutionary equilibrium hypothesis suggests that, for hosts of avian brood parasites, selection should only favour the evolution of parasite egg rejection behaviours if their benefits outweigh the costs (Rothstein, 2001). In line with this view, we argue, that L. acervorum host populations are only able to sustain efficient defences against attacking slave-makers if the environmental conditions are benign as in the German or English lowland sites. Empirical studies supporting ecological variation in selection pressures as a cause for a coevolutionary mosaic are rare, but the existing studies focus on the importance of community context rather than the influence of habitat conditions. It has been shown that close species interactions can be disrupted at sites, where other competitors, predators or mutualists occur (Thompson & Cunningham, 2002; Benkman et al., 2003). However, this cannot account for the differences between the Italian and German community in our study, as in both locations the two main host species occur at similar densities with the same species of social parasite.

Leptothorax acervorum colonies from different populations varied in their ability to defend their nest against the social parasite. In contrast to several studies on avian brood parasites, which demonstrated less or no egg rejection in unparasitized host populations (Davies & Brooke, 1989; Soler & Moller, 1990), the ‘naïve’ colonies from the unparasitized English population were highly aggressive, corroborating earlier findings of well-developed parasite defences in this population (Foitzik et al., 2003). The high aggression level of British host colonies can be explained by geographical isolation of the populations from the British Isles: the cuticular hydrocarbon profiles of the English population might have become distinct from mainland populations and because H. sublaevis slave-makers exhibit a very similar cuticular hydrocarbon profile to that of their hosts (Kaib et al., 1993; Heinze et al., 1994), English L. acervorum should be able to recognize the social parasite more easily due to its deviant odour.

Earlier findings showed a more effective nest defence of monogynous L. acervorum host colonies against single intruding slave-makers (Foitzik et al., 2003). In contrast, the social structure of the host colonies did not influence the outcome of slave raids, which represents a more complex behavioural interaction. Monogynous colonies may be able to recognize intruding slave-makers faster because of their more uniform colony odour, and thus may be better to kill or drive away single parasite queens. However, a faster recognition ability does not necessarily benefit a host if an entire slave-making colony attacks. This might also explain why the more aggressive English host colonies do not salvage more brood in slave raids compared with other host populations.

In summary, our experiments demonstrate that the outcome of slave raids strongly depends on an interaction between the slave-maker and host population, revealing local adaptations to the sympatric opponent. Furthermore, the coevolutionary trajectories differ between communities, possibly caused by a different history, community context and ecology at each site. Thus the coevolutionary interactions between the parasite H. sublaevis and its L. acervorum host are strongly structured geographically, as predicted by the geographical mosaic theory of coevolution (Thompson, 1994, 1999b).

Acknowledgments

We thank Miriam Brandt, Andreas Trindl and Katja Pusch for help with ant collecting. We are grateful to Jürgen Heinze and Miriam Brandt for intensive discussions and helpful comments on the manuscript. This work was supported by the DFG (Fo 298/4).

Ancillary