Social parasites exploit societies, rather than organisms, and rear their brood in social insect colonies at the expense of their hosts, triggering a coevolutionary process that may affect host social structure. The resulting coevolutionary trajectories may be further altered by selection imposed by predators, which exploit the abundant resources concentrated in these nests. Here, we show that geographic differences in selection imposed by predators affects the structure of selection on coevolving hosts and their social parasites. In a multiyear study, we monitored the fate of the annual breeding attempts of the solitary nesting foundresses of Polistes biglumis wasps in four geographically distinct populations that varied in levels of attack by the congeneric social parasite, P. atrimandibularis. Foundress fitness depended mostly on whether, during the long founding phase, a colony was invaded by social parasites or attacked by predators. Foundresses from each population differed in morphological traits and reproductive tactics that were consistent with selection imposed by their natural enemies and in ways that may affect host sociality. In turn, parasite traits were consistent with selection imposed locally by hosts, implying a geographic mosaic of coevolution in this brood parasitic interaction.

Recent studies have shown that interactions among coevolving species are much more variable than previously suspected, often forming geographic mosaics in the traits and outcomes of coevolving species (Thompson 1994, 2005; Benkman 1999, 2003; Brodie and Ridenhour 2003; Berenbaum and Zangerl 2006; Anderson and Johnson 2008; Bauer et al. 2009; Foitzik et al. 2009; Toju 2008, 2009). The geographic mosaic theory of coevolution argues that this coevolutionary variation can be partitioned into three components: variation among environments in the structure of selection that creates geographic selection mosaics, variation in the strength of reciprocal selection that creates coevolutionary hot spots and cold spots, and variation generated by trait remixing caused by gene flow, random genetic drift, and metapopulation dynamics (Thompson 1994, 2009; Gomulkiewicz et al. 2000; Nuismer et al. 2003). In effect, coevolving interactions may be viewed as a genotype by genotype by environment interaction in which geographic differences in selection on coevolving pairs of species are shaped by differences in the surrounding physical and biotic environments.

Under laboratory conditions, it is possible to evaluate selection mosaics very precisely by controlling for each genotype used in an experiment (Hoeksema and Thompson 2007; Piculell et al. 2008; Bryner and Rigling 2011). In natural populations, it is impossible to evaluate the effect of every genotype of one species on the fitness of every genotype of another species in each environment in which the interaction takes place. It is possible to ask the equally important more general question of whether the structure, rather than just the strength, of selection on two or more interacting species varies among biological communities due to differences in either the physical environment or to differences in selection imposed by yet other species. One of the current challenges in coevolutionary biology is therefore to understand the physical and biological causes of geographic variation in the structure of selection on coevolving interactions (Thompson 1994, 2005). Recent studies have shown, for example, that interactions between tephritid flies and goldenrods, as mediated by gall shape, have diverged among communities through selection pressures imposed by different natural enemies in different communities (Craig et al. 2007). Similarly, the interactions between crossbills and conifers differ geographically depending on the presence of squirrels that compete with the crossbills (Benkman 1999, 2003; Mezquida and Benkman 2005; Benkman et al. 2008), and the mutualistic outcome of interactions between pollinating seed-parasitic Greya moths and their host plants differ among communities depending on the abundance of a small number of co-pollinators (Thompson and Cunningham 2002; Thompson and Fernandez 2006).

Social parasites offer special opportunities for evaluating how the surrounding web of interactions in a community influences the coevolution of a pair of species, because the effects of interactions on lifetime fitness can often be readily determined. Social parasites are a special class of parasites and, as with avian brood parasites (Payne 2005; Krüger 2007), cause their hosts to misdirect parental care to parasite brood instead of their own brood (Wilson 1971). However, unlike brood parasites who lay their eggs in host nests and then have no subsequent interactions with the fostering parents, social parasites become members of the host colonies. They integrate into their host's social life and exploit the host workers for rearing their own brood, imposing strong and quantifiable costs on host fitness (in ants: Foitzik and Herbers 2001a; Brandt et al. 2005; Nash et al. 2008; in wasps: Lorenzi et al. 1992).

Social parasites often evolve by breaking into and often mimicking the communication system by which host individuals recognize each other (Lenoir et al. 2001; Lorenzi 2006; Bagnères and Lorenzi 2010). Subsequent or concomitant coevolution may involve multiple other parasite and host traits (e.g., Foitzik and Herbers 2001b; Foitzik et al. 2001, 2003, 2009; Ortolani and Cervo 2009, 2010). The interactions are often geographically complex, because social parasites are patchily distributed (Wilson 1971), and may involve different numbers of host and social parasite species in different localities (Johnson and Herbers 2006). This complex geographic structure makes them attractive models for studies of geographically structured coevolution. Indeed, a chemical coevolution mosaic has been demonstrated recently in a myrmecophyle, where geographically distinct populations of Maculinea butterflies differ in their surface chemistry depending on the surface chemistry of the prevailing local ant host (Nash et al. 2008).

The study of how selection diversifies among populations of coevolving social parasites and hosts may be particularly informative in social wasps, whose social stages may be transitional toward eusociality (Hines et al. 2007). Polistes wasps represent a step in the progression to a eusocial state through which other social taxa may have passed (Reeve 1991) and the genus includes three monophyletic species of obligate social parasites (Choudhary et al. 1994; Carpenter 1997). We already know that social parasitism may have shaped social wasp lifestyle (Brockmann 1993) and that predation pressure also may have shaped social insect life style, for example, by promoting foundress associations in wasps (Strassmann 1981; Reeve 1991). Nevertheless, predator impact on wasp colonies has not yet been thoroughly assessed (Yamane 1996), and we mostly do not know how social parasitism and predation at different stages of colony cycle affect host lifetime fitness and the structure of selection on host reproductive allocation. When a foundress founds her colony solitarily, she invests her reproductive resources into different tasks, such as building her nest, laying eggs, foraging, nursing each brood, and guarding the colony. The foundress’ decisions on how to partition her resources into these time- and energy-consuming tasks determine the pattern of colony growth, the numbers of brood, the ratio of sterile to fertile brood, and therefore the social structure of her colony. Consequently, knowledge of reproductive allocation is central to the analyses of the relationships among social parasitism, geographically variable coevolution, and eusociality.

In this study, we explore differences in the geographic structure of phenotypic selection on host foundresses and their social parasites among natural populations that differ in intensity of predation on nests. Our goal is to evaluate geographic differences in the multispecific structure of selection on the pairwise interaction between host wasps and their social parasites. We evaluate population differences in morphological and life-history traits of host foundresses as mediated by social parasites and predators. Finally, we assess how multiple biotic interactions can affect the geographic diversification of the reproductive allocation of host foundress populations and contribute to a geographic mosaic of social evolution.



Polistes biglumis is a free-living social wasp that in southern Europe lives only in open mountain meadows at elevations above 1000 m. The geomorphology of these landscapes and the distribution of coniferous forests over vast areas result in a distribution of P. biglumis populations that is patchy and separated by large areas unsuitable for nesting. Geographic barriers may enhance separation of populations together with philopatry, the typical behavior of Polistes foundresses, who establish their nests near their natal-nest site (review in Reeve 1991; Hunt 2007). Additionally, matings tend to be local (Seppä et al. 2011). Establishment and growth of the small, annual colonies is roughly synchronous within a population. Each overwintered, inseminated female (foundress) builds her paper nest on a small stone in a south-facing meadow during May–June, soon after snow melting. Nests are built as vertical combs on well-lit, south-facing vertical sides of stones, usually less than 10 cm from the ground. Nests are obvious to human observers and are not concealed to predators; nests and their sideward-facing cells are exposed to sun and rain. Foundresses lay eggs in the central nest-cells. Peripheral cells remain empty throughout the nesting season and are likely to provide protection against physical and/or thermal stress (Lorenzi and Turillazzi 1986; Hozumy et al. 2008). The founding phase (i.e., the stage before the first brood emerge) lasts about 45 days and during this phase the foundress lays eggs and solitarily cares for her own brood, while enlarging her nest, foraging for nectar and prey, feeding her larvae, and defending her colony. The first brood emerges by late July and may be composed of both workers, which forage and care for the natal colony, and sexuals. Sexuals consists of both males and next-year foundresses (gynes). Gynes mate during summer, overwinter, and solitarily found their annual colonies the next year (Lorenzi and Turillazzi 1986). Workers are very rare in some populations (Fucini et al. 2009) and caste ratio (proportion of workers to gynes) varies significantly among populations (see Fig. S1). Brood emergence continues throughout the summer until adult wasps desert the natal colony by late August–September and 1-year-old foundresses die (Fig. 1). The whole colony cycle lasts three to four months during which each foundress is the main forager and the only egg-layer both before and after offspring emergence (Lorenzi and Turillazzi 1986).

Figure 1.

Life cycle of Polistes biglumis. Foundresses found their nests solitarily at the end of May (nest foundation). During the next 45-day-long founding phase, foundresses enlarge their nest, lay eggs, feed their brood, and defend their colony. During the founding phase, colonies can be killed by predators, usurped by conspecific females (intraspecific usurpation) or parasitized by Polistes atrimandibularis, which are obligate social parasites (interspecific parasitism). The brood of free-living foundresses, as well as those of usurped or parasitized foundresses, begin to emerge by mid July. The colony cycle (and potentially brood emergence) continues until September. Mating occurs in August–September, after which next-year foundresses hibernate under stones and crevices. They will found their own colony solitarily the next spring. Illustrations by Federica Rossi.

During the founding phase, nests in some populations are parasitized by a conspecific female (usurper) or by an obligate social parasite (Polistes atrimandibularis). Invasions by usurpers and parasites peak at the end of June. Conspecific usurpers are foundresses that have lost their nests (Lorenzi and Cervo 1995). Following violent fights, a conspecific usurper may chase away the resident foundress, destroy the foundress’ youngest immature brood, and take over the nest and the initial offspring of the displaced foundress (Cervo and Lorenzi 1996a). Then the usurper begins laying eggs and caring for the usurped colony. In contrast, a P. atrimandibularis obligate parasite peacefully enters a host colony, resists the defensive attack by the resident foundress but eventually dominates and enslaves her (Cervo et al. 1990a). The parasite and the enslaved foundress live together in the colony throughout the remainder of the nesting season, the parasite mimicking the host recognition code (Bagnères et al. 1996). Once enslaved, the host foundress stops laying eggs and allocates all her resources into working (Lorenzi et al. 1992). Like conspecific usurpers, obligate social parasites destroy the youngest host immature brood at invasion (but the oldest are reared to adulthood) then begin laying eggs. When the host-female brood emerge, they nurse the parasite brood, which is composed exclusively of sexuals. Newly emerged parasite females mate at the end of summer, overwinter, and usurp host colonies the next summer (Lorenzi et al. 1992).

In colonies without usurpers or obligate parasites, P. biglumis brood emerge continuously throughout the nesting season, but in usurped and parasitized colonies host brood emergence stops one month earlier and parasite (or conspecific-usurper) brood emerge next. Thus, both conspecific usurpation and interspecific parasitism dramatically reduce host fitness (for conspecific usurpation: Lorenzi and Cervo 1995; for parasitism by P. atrimandibularis: Lorenzi et al. 1992). Because the foundress is the only adult in the colony throughout the founding phase and often leaves the nest unattended to forage, nests are especially vulnerable: more than 50% of the nests fail during the founding phase, mostly as a result of predation by invertebrate predators (e.g., ants), birds, other small vertebrates, and humans (Lorenzi et al. 1992).


We assessed the fate of P. biglumis foundresses and their colonies in 2004, 2005, 2006, 2007, and 2008. We sampled two populations in the Western Alps separated by approximately 70 km and seven parallel valleys. Obligate parasites were present in these populations (Montgenèvre, French Hautes Alpes, 44°55′N, 6°43′E; and Ferrere: Valle Stura di Demonte, Cuneo, Italy, 44°22′N, 6°57′E), whereas they were absent in a third population in the Central Alps (Carì: Val Leventina, Canton Ticino, Switzerland, 46°29′N, 8°49′E). This population was separated from the closest Western Alps population by approximately 250 km. We also sampled a population in the Apennines where parasites were also present (Monte Mare: Catena delle Mainarde, central Italy, Isernia, 41°37′N, 14°00′E). This population was separated from the closest Alps population by approximately 650 km. All populations were located at a similar elevation (1600–1900 m. a.s.l., Table 1) and were dense, roughly averaging 50 colonies per ha. Polistes biglumis was the only free-living Polistes wasp nesting there. Therefore, where the social parasite P. atrimandibularis was present, they had no alternate hosts to P. biglumis.

Table 1.  The relative impact of obligate parasites, conspecific usurpers, and predators on foundress breeding attempts across populations and years at the end of the founding phase. Sample size in parenthesis.
Population (elevation in m)Sample size (years of study)Parasite prevalence (measured as percentage of parasitized breeding attempts) (n)Conspecific usurper prevalence (measured as percentage of usurped breeding attempts) (n)Mean percentage of breeding attempts killed by predators during the founding phase (n)
Carì170 colonies010.667.6
(1650 m)(3 years)(0)(18)(115)
Ferrere464 colonies5.63.763.8
(1900 m)(5 years)(26)(17)(296)
Monte Mare516 colonies9.31.234.1
(1740 m)(5 years)(48)(6)(176)
Montgenèvre498 colonies23.50.871.5
(1860 m)(5 years)(117)(4)(356)
Pearson χ2 χ2= 110.022χ2= 52.780χ2= 169.739
df = 3 P < 0.0001P < 0.0001P < 0.0001

From the early founding phase (May-June) until the end of the nesting cycle, we monitored the breeding attempts of 1648 individually marked foundresses (Table 1). Colony checks were made every 2–10 days to tabulate the numbers of cells and immature brood (eggs, larvae, and pupae), the presence/absence of the marked foundress, unmarked conspecific females (usurpers), or P. atrimandibularis parasite females (parasites). When we found usurpers and parasites on nests, we individually marked them. Nest predation was recorded as loss due to colony killing by ants (or occasionally by P. atrimandibularis, see Cervo et al. 1990b), which depredated immature brood leaving the nest intact, by vertebrates (mostly birds), which destroyed nests partially or completely, or by humans.

Each colony was defined as free-living (i.e., without parasites) when a foundress controlled her colony until the end of colony cycle, as usurped when a usurper chased the original foundress away and took over the colony, or as parasitized when a P. atrimandibularis female invaded the colony and cohabited with the host foundress until the colony declined. In P. biglumis, nest size does not increase significantly after the end of the founding phase and almost all colony productivity and reproductive success result from the eggs laid during the founding phase (or soon after its end) (Lorenzi and Turillazzi 1986). Therefore, we investigated how foundresses from four populations partitioned their reproductive resources into the different tasks of building and rearing during the founding phase. Using data collected near the end of founding phase and before the peak of parasite or usurper nest-invasion occurred (end of June), we counted for each foundress (1) the number of nest cells she built, as an index of her building effort and (2) the number of immature brood (eggs, larvae and pupae) in the nest, as an index of brood investment, and we calculated (3) the percentage of empty cells (number of cells with no brood × 100 divided by the total number of cells) as an index of protection effort (ranging from 100 = only protection, to 0 = no protection). Our data thus report the reproductive decisions made by foundresses before their nests were usurped or parasitized. Consequently, trait variations among populations cannot be ascribed to foundress adjustments that depend on the later status of the nest.

In 2007 and 2008, we estimated head width of 361 foundresses, 40 conspecific usurpers, and 48 obligate parasites, using calipers accurate to 0.05 mm. Head width is a reliable indicator of body size and relative fighting ability in Polistes (Eickwort 1969; Ortolani and Cervo 2009).

Because colonies are annual and monogynous and each foundress is the only egg-layer in the colony, we estimated the true lifetime fitness of the foundresses as the number of pupae produced during the nesting cycle. In “normal”, free-living P. biglumis colonies, most of the offspring are fertile (i.e., next-year foundresses or males) (Lorenzi and Turillazzi 1986; Fucini et al. 2009) and newborn wasps may disperse soon after emerging (Lorenzi and Turillazzi 1986). Therefore, counting the adults emerged from colonies as a measure of foundress fitness would have been inaccurate because of the early dispersal of newly emerged wasps. The pupal phase lasts approximately 3 weeks. We avoided pseudoreplication (i.e., counting the same pupa repeatedly during a colony cycle) by counting pupae in each colony in late June, late July, and late August, that is, at sufficiently long time intervals to avoid overlap. In parasitized and usurped nests, the brood of the displaced foundresses emerges first, followed by the brood of usurpers or parasites. In usurped nests, we calculated foundress fitness as the number of pupae produced within one month from usurpation; after that time interval pupae belong to usurpers and were assigned to usurper fitness. In parasitized nests, no confusion is possible, because host and parasite pupae are easily identified by their cocoon colors, which are black for hosts and white for parasites (Lorenzi et al. 1992). White pupal cocoons were counted to calculate parasite fitness. If not otherwise stated, foundresses (or parasites) which failed to produce any pupa were included in the fitness calculations, because the mean of the nonzero class would have been the mean after selection (Brodie and Janzen 1996). For each year, sample sizes varied according to the number of colonies founded at each site. The data collected from each nesting season varied over time, because we discovered nests throughout the seasons, nest predation occurred at different stages during colony cycle, and a few nests were missed on some surveys.


For each population and year, an estimate of the opportunity for selection was quantified by using Crow's index I, which scores the intensity of selection as the ratio of variance in fitness to the squared mean fitness. This measure provides the upper bound for the strength of selection (Arnold and Wade 1984; Brodie et al. 1995).

In our analysis, we aimed to produce the simplest statistical model that fitted the data. We followed backward stepwise procedures and removed nonsignificant factors or interaction terms to obtain the most parsimonious minimal model (Crawley 1993; Wilson and Hardy 2002). We used general linear models (GLMs) for normally distributed response variables and generalized linear models (GZLMs) for response variables that were not normally distributed.

For data with binomial distributions (e.g., nest mortality; protection effort), we tested differences among populations or years using a GZLM for binomial distributions with a logit link function.

Because many foundresses and parasites had 0 fitness, fitness data had a nonnormal negative binomial distribution and were analyzed using a GZLM for negative binomially distributed dependent (response) variables with a log link function. Two different models of foundress reproductive success were developed: one that included all foundresses and any potential effects of population and year on their fitness, and a second that classified foundresses as a function of their colony-status by mid July (free-living, usurped, parasitized, or killed by predation). In this second set of analyses, the effect of year on fitness was not included, because a preliminary analysis revealed it was not a significant factor.

We used GLMs after transforming foundress and parasite traits, except for building effort, to accommodate the assumptions of normality (Shapiro-Wilk W test) and homogeneity of variance (Levene test or F-max test). Head width and number of immature brood were ln-transformed, and data on proportion of empty cells were arcsine-transformed. Repeated measure analyses of variance (GLM repeated measure procedures) with two between-subject factors (population, year) and one within-subject factor (head width) were performed to analyze differences between matched parasite (or usurper)—foundress pairs (each nest had one foundress and one parasite).

Finally, we used a principal components analysis (PCA) of foundress traits to further characterize the variation in foundress morphology and reproductive decisions among populations with different compositions of parasites and predators. For these analyses, we entered the foundress traits into the analyses to extract four noncorrelated components. PCA was based on correlations and varimax rotation. We retained the two components with eigenvalues larger than 1 (i.e., we dropped the two components that accounted for less variability than did a single variable). The first principal component, which captured much of the variance, and the second component, which captured much of the remaining variance, were used to test the similarity among P. biglumis populations by means of two univariate GLMs.


We used selection gradients to measure the link between fitness and a particular trait (Lande and Arnold 1983; Brodie et al. 1995). We estimated selection on phenotypic traits by examining the covariance of a trait and the relative fitness of individuals with that trait or by analyzing selection imposed by each species on phenotypic traits of the other species. We derived the selection function describing this relationship by regressing data on individual phenotype and fitness (Lande and Arnold 1983). Selection can change not only the mean value of a trait within a population (directional or linear selection) but also the variance of a trait (nonlinear or quadratic selection). We therefore measured the modes of selection by analyzing different components of the selection function. The standardized regression coefficient or selection gradient, β, estimated the strength of directional selection in a linear model. Logistic regression estimated the strength of curvilinear selection and the statistical significance of parameter estimates when the response was categorical (Brodie et al. 1995) such as the likelihood of reproductive success of a foundress or that of invasion of her colony by parasites. In logistic regression, the quadratic selection coefficient cannot be calculated. Therefore, we determined the sign of directional selection by means of the odds ratio. This ratio indicated the change in the probability of occurrence of an event, such as invasion or predation of a foundress breeding attempt, due to each of the factors with an increase of one unit. An odds ratio greater than 1 indicates that, as the value of the character increases, the probability that a breeding attempt escapes invasion by parasites or destruction by predators increases, whereas an odds ratio less than 1 indicates a decrease in the probability that a breeding attempt escapes invasion by parasites or destruction by predators as the trait value increases (Craig et al. 2007).

We estimated the strength of curvilinear selection by multiplying the quadratic coefficient by two to estimate the selection gradient γ (Lande and Arnold 1983; Stinchcombe et al. 2008). We then visually inspected the regression plots of relative fitness as a function of trait variation and used the equation to estimate the maximum or minimum values (i.e., –β/γ) to identify whether quadratic selection on a trait was stabilizing or disruptive, that is, whether variance decreased or increased (Mitchell-Olds and Shaw 1987). We inferred that selection on a trait was stabilizing when the quadratic coefficient was negative and exhibited a maximum value within the range of data, and that selection on a trait was disruptive when the quadratic coefficient was positive and showed a minimum value within the range of data. Interpretation of selection gradients requires knowledge of the correlation among traits because selection can act on traits correlated with the measured trait, rather than on the trait itself (Mitchell-Olds and Shaw 1987). To interpret selection gradients correctly, we calculated a matrix of Pearson correlations among traits for each population.

Data analyses were performed with SPSS 17.00 (Garson 2009, for procedures).



Parasites, conspecific usurpers, and predators had significantly different impacts on the breeding attempts of foundresses in the four populations during the founding phase (Table 1). Throughout the study period, parasite prevalence and predation impact were highest in the Montgenèvre population. Parasites were moderately common in the Monte Mare and Ferrere populations, which had low and the high levels predation, respectively. Predation and usurpers both had a relatively high impact on the Carì population. The observed differences among populations in parasite prevalence and predation levels were generally consistent among years (Fig. 2).

Figure 2.

The state of the nests across four populations and years. (A) Percentages of free-living, parasitized or usurped colonies. (B) Percentages of nests of free-living, parasitized or usurped nests killed by predators.

The opportunity for selection imposed by enemies on foundresses differed significantly among populations (Kruskall Wallis test, χ2= 11.37, df = 3, P= 0.010), and was particularly high in the Montgenèvre population and weak in the Monte Mare population (Table 2). Overall, the values for the opportunity for selection, which reflect variances in individual foundress fitness, suggest that the potential for phenotypic selection imposed by enemies is high in most populations.

Table 2.  The opportunity for selection, I, on host foundresses per population and year (I = variance of foundress reproductive success divided by the squared mean foundress reproductive success).
CarìFerrereMonte MareMontgenèvre
  1. 1Largely caused by increased disruption that year due to construction of a building.

2004 1.681.42 3.04
2005 3.581.36 4.91
20061.943.491.59 4.18
2008 1.92  5.45

The relative effects of obligate parasites, usurpers, and predators affected foundress fitness (estimated as lifetime offspring number in n= 1578 foundresses) differently across populations (Fig. 3). Free-living foundresses had the highest fitness in Ferrere, but not in the other populations, where parasitized or usurped foundresses had fitness gains similar to free-living foundresses. These calculations included foundresses that did not produce any descendants, often due to predation. Thus, high fitness in parasitized or usurped foundresses suggests that parasites or usurpers may help defend host nests from predators, for example, by reducing the time nests are unattended.

Figure 3.

Lifetime foundress fitness as a function of the state of the colony at the end of the founding phase (mid July) (mean ± 1 SE). Foundresses that failed to produce brood were included in the calculation. Foundress fitness varied significantly depending on population (Wald χ2 = 22.699, df = 3, P < 0.0001) and state of the nest (Wald χ2 = 511.637, df = 3, P < 0.0001), and the effect of the state of the nest on fitness differed among populations (interaction term population × state of nest: Wald χ2 = 28.471, df = 8, P < 0.0001). Year, entered as a predictor in a preliminary GZLM, was removed because it had no significant impact on the comparison, reinforcing the indication that fitness is site-specific. In the figure, data were averaged across years.


Geographic variations in traits


Foundress head-width, building effort, brood investment, and protection effort differed significantly among populations and years (Fig. 4 and Table S1). Head width (as a proxy for body size) differed among populations and the differences were consistent across years, but building effort, protection effort, and brood investment (i.e., the number of brood) all showed a significant interaction between population and year (Fig. 4 and Table S1). Variation among populations, however, was larger than variation among years for all traits, except for building effort (Table S1, see F values of population and year). Foundresses from Carì, where obligate parasites were absent and usurpers were common, had the largest head widths, but were intermediate in other traits (Fig. 4). Foundresses from Monte Mare, where parasites were uncommon, consistently had the smallest head widths, the highest brood investment, and the lowest protection effort (i.e., relatively few peripheral cells were left empty). Foundresses from Montgenèvre and Ferrere, where obligate parasites were common, had intermediate head widths and varied considerably in other traits. Among these three potential predictors of reproductive investment, head width was not correlated with any other trait and protection effort and building effort were not mutually correlated (Table S2).

Figure 4.

Variation in foundress traits across populations and years. Populations are represented as different lines. (A) The variation in foundress head-width. (B) The variation in foundress building effort. (C) The variation in brood investment. (D) The variation in protection effort. Statistical analyses are shown in Table S1.

The difference in foundress traits among populations was further supported by combining traits as new, uncorrelated components using a principal components analysis (Fig. 5). The PCA yielded a first component mostly associated with the investment in brood and the effort in building protective, empty cells (i.e., the kind of foundress reproduction), which ranged from producing small numbers of well-protected brood to producing large numbers of unprotected brood. Foundresses from the Monte Mare population had larger numbers and less-protected brood than the three populations in the Alps, which had fewer but more-protected brood (Fig. 5). Within the Alps populations, Montgenèvre produced the smallest numbers of brood with the highest protection, and it differed significantly from Carì and marginally from Ferrere in that respect (Fig. 5). By combining foundress head-width and nest size in component 2, the PCA showed that the very large foundresses from Carì differed from the other populations, which had the smallest foundresses and nests (Fig. 5). Unlike the other populations, Carì lacked parasites, but had many conspecific usurpers. Among the other three populations, Montgenèvre, where parasites were very common, had the smallest values in nest and foundress size and differed significantly from Monte Mare but not from Ferrere in that respect.

Figure 5.

Principal Component Analysis showing the variation in foundress traits among the four populations (data pooled by year) in relation to the first two principal component. The empty squares indicate the centroids of the four populations. The first and second components captured together more than three-fourth of the total variance. The traits with the highest loadings (rotated) on component 1 (PC1) were brood investment (+ 0.94) and protection effort (–0.88), so that PC1 accounted for reproduction; the traits with the highest loadings on the second component (PC2) were building effort (+ 0.86) and head width (+ 0.61) so that PC2 mainly accounted for size. There were consistent differences in both principal components among the four populations (GLM on PC1: F3,260= 36.357, P < 0.0001; GLM on PC2: F3,260= 28.593, P < 0.0001), showing clear divergence between these isolated populations. Post-hoc LSD tests showed that the Monte Mare population significantly differed from any other population when PC1 was taken into account (P < 0.0001, all comparisons) and that Montgenèvre significantly differed from Carì for the same component (P= 0.010) and marginally from Ferrere (P < 0.066). Finally, PC2 significantly separated Carì from any other population (P < 0.0001, all comparisons) and separated Montgenèvre from Monte Mare (P= 0.001). All other comparisons were not significant.

 Obligate parasites.—

Parasites invaded nests nonrandomly in Montgenèvre and Monte Mare, choosing the largest nests in each population: the size of the nests targeted by parasites was significantly larger than the mean nest size in the population (Fig. 6A). Parasites also preferentially targeted the smallest host foundresses in Montgenèvre, whereas in the other parasitized populations parasites did not target hosts on the basis of body size (Fig. 6B).

Figure 6.

The variation in parasite preference for the size of the host nest and foundress and the variation in nest-mortality ratio among the three populations where obligate parasite were present (white bars: nests not targeted by parasites; gray bars: nests targeted by parasites). (A) Preferential attack of large nests by parasites, as shown by mean size of nests targeted or not by parasites at the end of June, when the peak of nest invasions occurred. In Montgenèvre and Monte Mare parasites targeted the largest nests (GLM on the size of nests with or without parasites in Montgenèvre: state of nest: F1,303= 18.496, P < 0.0001; year: F4,303= 13.336, P < 0.0001; interaction state of nest × year: F4,303= 1.670, P= 0.157; in Monte Mare: state of nest: F1,306= 5.090, P= 0.025; year: F4,306= 8.441, P < 0.0001; interaction state of nest × year: F2,306= 0.022, P= 0.978). In contrast, in Ferrere, parasites invaded nests without regard to their size (state of nest: F1,325= 1.380, P= 0.421; year: F4,325= 15.603, P < 0.0001; interaction state of nest × year: F3,325= 0.430, P= 0.732). (B) In Montgenèvre parasites targeted the smallest hosts in the population (GLM on head width of Montgenèvre foundresses with or without parasites: F1,58= 7.336, P= 0.009). In Ferrere and Monte Mare, however, parasites did not exhibit any preference for large or small hosts (Ferrere: F1,59= 1.095, P= 0.300; Monte Mare: F1,88= 0.163, P= 0.688). C: The ratio of nest-mortality (proportion of nests escaped to killed by parasites) varied depending on whether nests were with or without parasites (mean ratio ± 1 ES among years). Nests with parasites were proportionally less predated than those without parasites (GZLM on predated to survived nests: Montgenèvre: state of nest (with or without parasite): Wald χ2= 76.855, df = 1, P < 0.0001; year: Wald χ2= 31.382, df = 4, P < 0.0001; Ferrere: state of nest: Wald χ2= 4.095, df = 1, P= 0.043; year: Wald χ2= 17.169, df = 4, P < 0.002; Monte Mare: state of nest: Wald χ2= 3.995, df = 1, P < 0.046; year: Wald χ2= 65.809, df = 4, P < 0.0001; the interaction term state of nest × year was removed because it had no significant impact on the comparisons).

Parasites did not differ significantly in body size among populations, but varied in size among years in different ways depending on the population (GLM on head width: population: F2,42= 1.015, P= 0.371; year: F2,42= 1.083, P= 0.348; population × year: F1,42= 5.706, P= 0.021). Across populations, however, each parasite was significantly larger than her host foundress (repeated measures GLM on the difference between parasite and host-foundress head-width: F1,34= 22.109, P < 0.0001; interaction terms by population and year were not significant).

Obligate parasites also mediated the effect of predators on nests. The proportion of nests that were killed by predators or that escaped predation depended on whether the nest was parasitized or free-living. Across populations, predators were less likely to kill parasitized nests than nests without parasites, but this effect was much stronger in Montgenèvre than in Ferrere and Monte Mare. In Ferrere and Monte Mare, nest-mortality ratio varied between years much more than between nests with parasites and without parasites (Fig. 6C).


The head width of conspecific usurpers was consistently larger than that of the ousted foundresses (repeated measures GLM on the difference between usurper and foundress head-width: Wilks’ lambda F1,34= 8.051, P= 0.008; population: F3,34= 1.441, P= 0.184). Surprisingly, usurpers preferentially targeted the largest foundresses, but not the largest nests, in Carì, which was the only population in which usurpers were common enough to allow statistical comparison (GLM on head width of foundresses targeted or not by usurpers: F1,130= 4.639, P= 0.033; GLM on the size of nests targeted or not by usurpers: F1,153= 0.115, P= 0.735).


Selection on foundress traits

The structure of phenotypic selection on foundresses differed markedly among populations, with parasites and predators differing among populations in how they shaped selection on the traits of foundresses (Table 3 and Table S3).

Table 3.  Probability values of selection by predators, usurpers, and parasites on host and parasite phenotypic traits in the four populations.
PopulationSource of variationSelectionFoundress phenotypic traitHost-foundress trait acting on parasite phenotypic traitParasite phenotypic trait
Head width PBuilding effort PBrood investment PProtective effort PHost reproductive success PHost brood investment PHost protection effort PHost building effort PDate of host nest invasion PHead width P
  1. 1Adding usurper fitness data to this calculation yielded substantially similar results.

  2. 2Selection gradients due to conspecific usurpers were not calculated due to small sample sizes.

 Conspecific usurpersDirectional0.034nsnsns      
Ferrere2FoundressDirectionalns< 0.0001 0.023nsnsnsnsns  
  Nonlinearns0.004<0.0001 nsnsnsnsns  
 Obligate parasitesDirectionalnsnsnsns    ns 
  Nonlinearnsnsnsns    ns 
Monte Mare2FoundressDirectionalns0.0090.0010.0160.014nsnsns  
  Nonlinearns< 0.0001 0.004nsnsnsnsns  
 Obligate parasitesDirectionalns0.0040.020ns    nsns
  Nonlinearns0.0130.039ns    nsns
Montgenèvre2FoundressDirectionalns< 0.0001 <0.0001 ns0.053nsnsns  
  Nonlinearns< 0.0001 0.002ns0.025nsnsns  
  Nonlinearnsns (0.068)0.030ns      
 Obligate parasitesDirectional0.014< 0.0001 <0.0001 ns    nsns
  Nonlinear0.033< 0.0001 <0.0001 ns    nsns
 Head width.—

Obligate parasites exerted strong positive directional selection for large foundresses in Montgenèvre, because they preferentially attacked the smallest foundresses (Table 3 and Table S3). Montgenèvre parasites also had the potential to impose disruptive selection, but the curve reached a minimum so far below the range of data that it would not mitigate the effects of positive directional selection. In Carì, conspecific usurpers imposed mild but significant directional selection for small foundresses, but there was also significant but moderate stabilizing selection on foundress head-width (with the maximum at medium trait-value: approx. 3.20 mm). The summed effect of these selection pressures in Carì suggested relaxed selection on small- and medium-sized foundresses but reinforced selection on large foundresses for small head width.

 Building effort.—

In all four populations, there was significant directional selection for increased building effort, which acted especially on foundresses that built small nests (Table 3 and Table S3). Although there was no significant nonlinear selection on building effort at Carì, there was disruptive selection at Montgenèvre and stabilizing selection at Ferrere and Monte Mare.

The combination of parasitism and predation resulted in a complex overall pattern of phenotypic selection on foundress building effort (Table 3 and Table S3). Predators, most likely small predators such as mice or ants, exerted a moderate but significant directional selection for large nests in Monte Mare and Montgenèvre, but did not impose selection on building effort in Ferrere and Carì. In Monte Mare, a quadratic selection term resulted in relaxed selection on foundresses that built small nests and in strong selection favoring foundresses that built large nests. Parasites imposed significant directional selection on Monte Mare and Montgenèvre foundresses for a decrease in building effort. In both populations, such selection was stronger on foundresses that built large nests, because significant stabilizing selection reached a fitness maximum at moderately low building-effort values. Overall, fitness was higher for foundresses that built large nests, although parasites acted against that. Consistent with that result, among the three alpine populations (Carì, Ferrere, and Montgenèvre), nests were smaller in Montgenèvre, which was the population with more parasites.

 Brood investment.—

The harsh climatic conditions of the mountain environment constrained foundresses to build their nests after snow has melted. As a consequence, any selective pressure on foundresses for beginning the nest even sooner cannot operate, which could therefore increase pressure for increased egg-laying rate at the beginning of the nesting cycle. We found such selective pressures in our populations, but the pattern was complex (Table 3 and Table S3). Mild but significant directional selection for increasing early brood investment (i.e., for increasing egg production rate at the beginning of nesting cycle) operated on all foundresses, except those in Carì. In Ferrere and Monte Mare, foundresses that laid the most eggs were subject to a greater than linear increase in selection pressure for that trait, because quadratic selection reached a fitness minimum at low values of brood investment. In Montgenèvre, the quadratic term was also positive but the fitness minimum was below zero, resulting in weakening the selection on foundresses with many eggs.

Predators did not exert selection on brood investment except for Montgenèvre, where they imposed significant directional selection for an increase in brood investment on foundresses with small investments and imposed stabilizing selection on foundresses with large investments (Table 3 and Table S3). In contrast, parasites imposed selection on brood investment, but did so in different ways in each population (Table 3 and Table S3). Although parasites did not have any significant selective effect in Ferrere, they selected for small brood investment in Monte Mare and Montgenèvre. Such selection was even stronger on foundresses with large brood investments, as indicated by a significant stabilizing selection with a close-to-zero fitness maximum.

 Protection effort.—

Selection on foundress protective effort was evident only in Monte Mare, where it favored less protective foundresses (Table 3 and Table S3). That result was particularly striking, because Monte Mare foundresses already made the smallest protective effort among the four populations.

The net effect of the pressures exerted by natural enemies was to favor well-protected nests with few offspring over less-protected nests with many offspring.

Selection on parasite traits

We detected little evidence of direct selection on parasite morphological traits, but we found significant and geographically variable selection on parasite behavior, as mediated by their effects on host reproduction (Table 3 and Tables S4 and S5). In Monte Mare, there was significant directional selection on parasites that favored the reproduction of their host foundresses (Table 3 and Table S4). Indeed, the higher the fitness of the host foundress, the higher the fitness of the parasite. In Montgenèvre, marginally significant, positive directional selection for parasites that favored host fitness was counteracted by significant stabilizing selection on parasites that limited the reproduction of their hosts (Table 3 and Table S4).


The results suggest that natural populations of P. biglumis wasps have diverged geographically through coevolutionary pressures exerted by their obligate congeneric social parasite, P. atrimandibularis and by nest predators. These geographic differences are greater than the temporal differences found among years within a site. Polistes atrimandibularis populations also appear to have diverged geographically through selection mediated by effects on host reproduction. The overall geographic mosaic of coevolution between these two Polistes species therefore appears to be driven by geographic differences in the relative importance of social parasitism and predation in shaping the life-history traits of P. biglumis.


We found that each locality was unique in the relative impact of natural enemies on this coevolving interaction. Foundress populations differed not only in morphological traits but also in strategies of resource allocation to reproduction, providing evidence that natural enemies can influence reproductive decisions and ultimately the expression of sociality. Among these populations, the opportunity for selection, which estimates the highest possible bound of selective change within a population, was within the range of those reported for natural populations (e.g., see data in Rossiter et al. 2006). It differed, though, among the studied populations and was consistently highest across years in Montgenèvre.

This high opportunity for selection at Montgenèvre was compatible with the observation that natural enemies of P. biglumis wasps were very common at that site and phenotypic selection was strong on multiple traits. Montgenèvre foundresses had intermediate size, but built the smallest nests and produced the smallest numbers of most protected brood. Indeed, Montgenèvre foundresses devoted a very high proportion of their resources to building marginal nest-cells that were not used for rearing brood. These cells might function as an envelope around the brood-rearing cells that might help in brood protection and thermoregulation (Hozumi et al. 2008), or some combination of benefits in this complex environment.

At the other extreme, selection on Monte Mare foundresses by predators and parasites was relatively relaxed. Monte Mare foundresses were very small, but built the largest nests and produced the largest numbers of brood with the least protection among the studied populations. In between these two extremes, the Carì and Ferrere populations differed in the relative importance of predation and parasitism. In Carì, predators were common, parasites were lacking but conspecific usurpers, which were rare in the other populations, were abundant. Carì foundresses had the largest head width, built relatively large nests and reared intermediate numbers of brood, with limited protection. The Ferrere population had relatively high predator impact and low parasite prevalence, and was intermediate in foundress traits. In general, the pattern of directional selection on foundress traits differed markedly among populations and was further complicated by disruptive or stabilizing selection which modulated the shape and strength of directional selection in each population.

Site-specific geographic conditions and social-parasite pressure did not work in harmony on wasp populations. Foundresses were generally under selection to start the nesting season with large nests and large broods, but parasites operated against that in Monte Mare and Montgenèvre—simply because they preferentially targeted the largest nests. Foundresses were under phenotypic selection to diminish their already small protection effort in Monte Mare, but not in the other populations. Predators also may have played a role in the evolution of foundress traits, because they curtailed the period available for host reproduction. Where natural enemies were more common, we expected that foundresses would concentrate their reproductive efforts in the early brood. Instead, we found that in those populations foundresses exhibited a reduced reproductive effort. Rearing smaller numbers of brood may diminish reproduction, but foundresses can trade-off between numbers and quality of the brood: rearing small numbers of brood can increase their quality because foundresses share food supplies among fewer larvae. These larvae may therefore survive the winter and behave as next-year foundresses rather than workers in the current year (see also Fucini et al. 2009). In wasps, the caste determination largely depends on larval provisioning levels and larvae that develop into next-year foundresses receive more food than those that develop into workers (O’Donnell 1998; Hunt and Amdam 2005; Hunt et al. 2007).


Selection on parasites was primarily mediated by effects on host reproduction. Social parasites obtain their fitness by exploiting the nest and the work of their hosts as an “extended phenotype” (Dawkins 1982; Hughes 2008). Therefore, it was not surprising that we detected large differences in the way selection operated on how parasites impacted host reproduction. In Monte Mare, parasites were favored that sustained host foundress reproduction, whereas in Montgenèvre, parasites were favored who constrained it. This difference is consistent with differences between these populations in the timing of worker/gyne production and in caste ratio. Polistes biglumis foundresses produced almost exclusively sexuals in Montgenèvre, whereas a few workers emerged before sexuals in Monte Mare (Fucini et al. 2009). Host sexuals are wasted resources in the parasite economy, whereas parasites cannot reproduce without worker-like hosts.

The choice of host nests by parasites was not random, as they invaded the largest nests in Montgenèvre and Monte Mare, but parasites also differed geographically in other traits. In Montgenèvre, they parasitized the smallest foundresses, possibly as a counteradaptation to enhanced fighting abilities in local hosts, whereas there was no preference based on host size in the other populations.

Overall, our sample sizes are unlikely to be large enough to correctly identify all selection pressures in hosts and parasites (Kingsolver et al. 2001), but P. biglumis wasp populations are usually small and their social parasites very rare (Cervo 2006). Hence, these numbers reflect the typical situation in natural populations.


In Polistes wasps, 40–97% of singly founded colonies fail to reach the worker stage largely due to predation (Strassmann 1981; Strassmann et al. 1988; Queller 1996; Hunt 2007), as we found in P. biglumis. Renesting, a viable option for some Polistes foundresses (Strassmann et al. 1988; Queller 1996), is rare in these populations because the breeding season is short (Lorenzi and Turillazzi 1986). However, an unexpected side effect of parasite invasion was a lowered risk of predation of the host nest that was particularly clear in Montgenèvre, probably as a simple result of the presence of two adult wasps, rather than one, on the nest during the founding phase. In colonies without parasites, foundresses leave their nests when foraging, making the nests unattended. In colonies with parasites, at least one adult wasp is on the nest most of the time. It is not surprising that parasites guard host nests, because their brood is inside the host nests and parasite fitness is driven by the success of the host colony. Indeed, in Montgenèvre, where predators were common and parasitized nests particularly successful in resisting predators, parasites actively defended host nests (MCL, unpubl. data). In primitively eusocial wasps, foundresses often associate with conspecific females at nest foundation (Stenogastrinae: Turillazzi 1991; Polistinae: Reeve 1991) and associated foundresses may protect their nests from predators better than solitary ones (Reeve 1991). The advantage of joint nest defense is debated because it can be counteracted by reduced per-capita reproduction of cofoundresses, who may or may not be related to each other (Queller et al. 2000; Leadbeater et al. 2010). Our data indicated that ecological factors, such as predation, can promote joining even irrespective of any relatedness among associated females: P. biglumis foundresses had higher fitness, on average, when they shared reproduction with a female of another species than when they nested alone, as long as nest predation was common. This fortuitous “alliance” between host foundresses and parasites was, of course, far from being mutualistic. Indeed, with respect to those free-living foundresses who were successful, parasites drastically reduced foundress fitness by causing brood losses (see also Lorenzi et al. 1992) and by “sterilizing” foundresses (Cervo and Lorenzi 1996). In fact social parasites are perceived as enemies, as indicated by foundresses actively fighting parasites when they attempt host-nest invasion (Cervo et al. 1990a).

Because parasites curtail the period open for host reproduction, they place a selective pressure on host populations to skip worker production and target host resources toward few early brood. This effect is likely greater in the northern, alpine, environments of these populations than in populations at similar altitudes—but at lower latitudes. Local climatic conditions themselves certainly play a role in P. biglumis geographic differentiation (Fucini et al. 2009), but differences in climate alone cannot explain the observed differences found in this study. The three northern populations (Montgenèvre, Ferrere, and Carì) do not differ in any major way in climatic regimes (Fucini et al. 2009), yet they differ in foundress life-history traits, supporting the observation that local ecological factors other than climate are more likely to drive the differentiation.


The foundress life-history traits that we examined are complex, as they express both reproductive strategies and social strategies of foundresses. Overall, parasite presence promoted a reduction in nest size and brood number and the early production of gynes, which might ensure at least some reproductive output in case of parasite invasion. Hence, parasites appear to be an important selective agent on host populations, but their effect depends in part on the co-occurrence of predators. In this respect, predation, which is traditionally considered as a factor promoting sociality, may instead operate against social life. At these high elevations, foundresses were often selected to reduce one of the distinctive traits of sociality, that is, reproductive division of labor. This implies that geographic differences in the relative impacts of parasites and predators may be driving social evolution along different trajectories, creating a geographic mosaic of populations with distinct levels of sociality. Variation in social structure is common in nature (Yamane 1996, Wcislo 1997; Wcislo and Danforth 1997; Schwarz et al. 2007; Danforth 2002; Danforth et al. 2003), but has not received much emphasis in Polistes. In facultatively eusocial taxa, social evolution, primed by lifetime monogamy, occurs at the brink of a transition to obligate eusociality on one side, and a reversal to solitarily living on the other (Boomsma 2007, 2009). Social parasites are rare and patchy distributed, which makes them candidates for testing the geographic mosaic theory of coevolution (e.g., Nash et al. 2008, Ruano et al. 2011) but their role in shaping the evolution of host social-traits is largely unexplored. In our system, social parasites were forcing the evolution of their hosts toward the reduction of the worker caste. The lack of workers is typical of solitary species, but also of social parasites. The evolution of workerless parasitism in bumblebees has been associated to the short favorable season that occurs in cold climates, where free-living queens produce few workers and can usurp nests (Hines and Cameron, 2007), two features they share with parasitic Polistes. Similarly, hosts in the most parasitized P. biglumis populations produce few workers and can usurp nests, which raises the question as to whether they are “preparasites” themselves. Up till now, it is unknown whether social parasitism in wasps arose at the northern edge of distributions—the only three obligate parasitic species are monophyletic, have a circummediterranean distribution, and are philogenetically close to a lowland species (Choudhary et al. 1994). More studies are needed on other systems to understand whether social parasites, in the long term, may trigger the origin of other social parasites, when their high prevalence is combined with short favorable season and high nest mortality.

Geographic differences undoubtedly explain some aspects of social evolution in Polistes wasps (Yamane 1996), but we know from studies of other taxa that organisms often experience their strongest selection pressures through interactions with other species (Brodie and Ridenhour 2003), fueling ongoing coevolutionary change (Thompson 1994, 2005). In P. biglumis, natural enemies can disfavor sociality in some populations by raising the cost to foundresses of postponing the production of sexuals. Our results therefore suggest that natural enemies may contribute to the diversification of the traits of coevolving species not only through change in morphological traits (e.g., Craig et al. 2007) but also through change in their social structure.

Associate Editor: T. Craig


We are grateful to the many students who helped with field data collections and especially to P. Cocco, V. Di Bona, S. Fucini, G. Marras, M. Mignini, F. Mola, C. Piccaluga. We thank A. Vigna-Taglianti for discovering the P. biglumis population located in Valle Stura di Demonte and the personnel of the Parco Nazionale d’Abruzzo, Lazio e Molise (Vittorio Ducoli and Cinzia Sulli) for permission to work in the park and for logistic help. We appreciate the helpful comments of K. Boomsma, R. Cervo, T. Craig, J. Strassmann, and two anonymous referees on earlier versions of the manuscript. Funding for this work was obtained from the MURST ex 60% and World-Wide-Style Project WWS of the University of Turin for researcher mobility to M. C. L.