ECOLOGICAL VERSUS PHYLOGENETIC DETERMINANTS OF TROPHIC ASSOCIATIONS IN A PLANT–LEAFMINER–PARASITOID FOOD WEB

Authors


Abstract

Specialized trophic interactions in plant–herbivore–parasitoid food webs can spur “bottom–up” diversification if speciation in plants leads to host-shift driven divergence in insect herbivores, and if the effect then cascades up to the third trophic level. Conversely, parasitoids that search for victims on certain plant taxa may trigger “top–down” diversification by pushing herbivores into “enemy-free space” on novel hosts. We used phylogenetic regression methods to compare the relative importance of ecology versus phylogeny on associations between Heterarthrinae leafmining sawflies and their parasitoids. We found that: (1) the origin of leafmining led to escape from most parasitoids attacking external-feeding sawflies; (2) the current enemies mainly consist of generalists that are shared with other leafmining taxa, and of more specialized lineages that may have diversified by shifting among heterarthrines; and (3) parasitoid–leafminer associations are influenced more by the phylogeny of the miners’ host plants than by relationships among miner species. Our results suggest that vertical diversifying forces have a significant—but not ubiquitous—role in speciation: many of the parasitoids have remained polyphagous despite niche diversification in the miners, and heterarthrine host shifts also seem to be strongly affected by host availability.

Plants, plant-feeding insects, and insect parasitoids make up well over half of the currently known species on the Earth, and plant–herbivore–parasitoid food webs therefore form a central component of practically all terrestrial ecosystems (Price 2002; Smith et al. 2008; Novotny et al. 2010). Understanding the mechanisms that have led to the origin of these hyperdiverse networks has been a long-standing challenge in biological research (Singer and Stireman 2005; Kitching 2006). However, it has not been possible to tackle the question effectively until recently, when advances in molecular-phylogenetic methods made it feasible to reconstruct the evolutionary history of many species-rich taxa interacting across multiple trophic levels.

A characteristic of trophic associations in plant–herbivore–parasitoid networks is resource specialization, such that most herbivores feed on only a subset of co-occurring plant species (Novotny et al. 2010), whereas individual parasitoid species tend to attack only a small proportion of the available herbivores (Hawkins 1994; Cagnolo et al. 2011). However, in both herbivores and parasitoids, species-level diet breadth ranges from extremely specialized to broadly generalized (Godfray 1994; Nosil and Mooers 2005; Stireman 2005). Over long-time periods, such variation in the degree of specialization constitutes the raw material that permits evolutionary shifts in ecological associations (Janz 2011). Niche shifts may be an important driver of speciation in plant-feeding insects as well as in their enemies (Nyman et al. 2007; Fordyce 2010; Segar et al. 2012), which raises the possibility of “bottom–up” speciation cascades: speciation of plants imposes a diversifying pressure on herbivores, which, in turn, paves the way for additional diversification in their parasitoids (Stireman et al. 2006; Forbes et al. 2009; see also Eubanks et al. 2003).

A steadily increasing number of phylogeny-based studies has led to a good understanding of speciation and niche divergence in many herbivore groups (Winkler and Mitter 2008; Janz 2011). By contrast, the ecology and evolution of parasitoid host use remains far less clear, because parasitoid–herbivore associations have rarely been studied in a phylogenetic framework (Zaldívar-Riverón et al. 2008), and even fewer studies have dealt with tritrophic systems (Lopez-Vaamonde et al. 2005; Ives and Godfray 2006; Nyman et al. 2007; Bailey et al. 2009; Wilson et al. 2012). This is unfortunate, because it is becoming increasingly evident that, like ecological, morphological, and chemical differences among plants, natural enemies can also influence the probability and direction of herbivore host-plant shifts (Bernays and Graham 1988). Parasitoid-driven “top–down” diversification of herbivores could result if parasitoid species attack herbivores preferentially on specific plants (Rott and Godfray 2000; Murphy 2004); in such cases, colonization of novel plant species may provide a herbivore lineage with “enemy-free space”, which may then facilitate shift-associated speciation (Lill et al. 2002; Singer and Stireman 2005). Conversely, examples of parasitoid species in which host preference and performance is linked to herbivore phylogeny instead of ecology (e.g., Desneux et al. 2012), suggest that resistance against parasitism is incurred by novel physiological defenses rather than by niche shifts. Identification of potential diversifying forces generated by parasitoids is further complicated by the typical association with each herbivore species of numerous enemies that differ in their diet breadth and specificity (Godfray 1994; Hawkins 1994). The most efficient drivers of ecological divergence should be relatively specialized parasitoids that are not lineage-specific, but select their victims mainly on the basis of their niches.

The purpose of our study was to elucidate phylogenetic and ecological determinants of parasitoid resource use by using the tri-trophic network centered on leafmining sawflies of the subfamily Heterarthrinae (Hymenoptera: Tenthredinidae) as a model system. Heterarthrine larvae feed on a wide variety of plant taxa (Altenhofer 2003), and are attacked by a diverse community of parasitoids exhibiting highly divergent preferences and diet breadths (Pschorn-Walcher and Altenhofer 1989). We reconstructed phylogenetic relationships among participant species on each trophic level, and then applied tree comparisons and phylogenetic regression methods to identify and to evaluate the relative contributions of factors that can influence parasitoid–leafminer associations. In particular, we wanted to know: (1) Is the current parasitoid community a result of a long coevolutionary history between the miners and their enemies, or is it a haphazard collection of co-occurring parasitoid taxa; (2) Has speciation and/or niche divergence of heterarthrine leafminers led to diversification in parasitoids; and (3) Are parasitoid–leafminer associations influenced more by phylogenetic relationships among the leafminer species, or by the miners’ ecological niches? Our results provide novel insights into the ecology and evolutionary assembly of tritrophic food webs, and into vertical selective pressures that have the potential to generate diversity in plant–herbivore–parasitoid networks.

Materials and Methods

STUDY SYSTEM AND ASSOCIATION DATA

The leafmining subfamily Heterarthrinae constitutes a monophyletic group within the highly diverse symphytan family Tenthredinidae (Hymenoptera) (Leppänen et al. 2012). Representatives of this subfamily, which includes over 150 species, can be found on all continents except Africa, Australia, and Antarctica (Taeger et al. 2010). Individual heterarthrine species typically are restricted to utilizing plants belonging to a single plant genus (Altenhofer 2003), but their combined host range encompasses at least 22 plant genera in 10 families (Leppänen et al. 2012). As we have recently shown, heterarthrine leafminers are generally millions of years younger than their host lineages (Fig. 1), and the present host repertoire therefore has been assembled by an alternation of niche conservatism and occasional shifts among evolving plant groups (Leppänen et al. 2012).

Figure 1.

Relaxed molecular-clock phylogeny of the leafmining sawfly subfamily Heterarthrinae (A) and selected host and non–host-plant genera (B) (modified from Leppänen et al. 2012). Plant genera used by heterarthrine leafminers are given in parentheses after species names in (A), and are indicated by different branch colors in (B). Black lines in (B) denote plant genera not used by any heterarthrine species. Corresponding colors in (A) show current and ancestral host-plant associations according to a maximum-likelihood reconstruction, black internal branches denote equivocal reconstructions. Numbers above branches are posterior probabilities (%), grey bars show 95% confidence intervals for node ages.

Numerous hymenopteran parasitoids inflict heavy mortality on heterarthrine larvae. For this study, we constructed a binary association matrix (Fig. 2) for 25 heterarthrine species and their parasitoids on the basis of Pschorn-Walcher and Altenhofer's (1989) extensive dataset, which lists the enemies of European heterarthrines on the basis of long-term mass rearings conducted in Austria, Switzerland, and Germany. Additional data on the parasitoids of Profenusa thomsoni (Konow) were taken from Digweed et al. (2009). The parasitoid community of the focal heterarthrines includes 59 species from three hymenopteran families (Ichneumonidae, Braconidae, and Eulophidae). We note that Pschorn-Walcher and Altenhofer (1989) recorded Shawiana as Phanomeris, Pseudichneutes as Ichneutes, Grypocentrus albipes as Grypocentrus sp. C1, and Minotetrastichus frontalis as M. ecus. The leafminer Scolioneura vicina (Konow) was listed as Scolioneura sp. cf. betulae.

Figure 2.

Association matrix and phylogenies of heterarthrine sawflies (to the left of the matrix) and the hymenopteran parasitoids (above matrix) that attack the leafminers’ larvae. Black squares in the matrix indicate that a given parasitoid species (columns) uses a given leafminer species (rows) as a host. The inset shows the plant-based miner pseudophylogeny used for testing the effect of miner host-plant use on parasitoid–miner associations (see also Fig. S1).

PHYLOGENY RECONSTRUCTION

Evaluating the effects of plant, miner, and parasitoid phylogenies on parasitoid–miner associations requires that phylogenetic trees are available for each of these three trophic levels. In the case of Heterarthrinae, we took advantage of our recently published Bayesian relaxed molecular-clock phylogeny of this subfamily (Leppänen et al. 2012), which was based on 3308 bp of DNA sequence data from four genes. For plants, we used the time-calibrated genus-level phylogeny of heterarthrine hosts and representative non-host taxa reconstructed in the same study on the basis of 9606 bp of sequence data from five genes. For the purposes of the present study, the leafminer tree (Fig. 1A) was pruned in Mesquite v. 2.71 (Maddison and Maddison 2009) to include only the 25 species for which detailed data on parasitoid communities were available (Fig. 2). To enable tests of the effect of host plants on parasitoid–miner associations, we also created a plant-based “pseudophylogeny” for the miners by first removing genera not used by any of the included miners from the plant phylogeny (Fig. 1B), and then grafting each miner species onto the tree according to their host genera; this creates a miner polytomy in place of each plant genus used by multiple miner species (Figs. 2, S1). Branch lengths within the polytomies were set to equal the shortest terminal branch on the original pruned plant tree (cf. Ives and Godfray 2006). Because the pseudophylogeny groups miner species according to the relationships among their hosts, a significant effect of this tree on parasitoid–miner associations can be interpreted as a sign of herbivore niche dependence in parasitoid resource use (Ives and Godfray 2006).

Phylogenetic relationships among hymenopteran parasitoids are poorly known especially at the species level, so we constructed a phylogenetic tree for the focal parasitoids by combining published molecular and morphological higher-level phylogenies with traditional classifications. Parasitoids of heterarthrines belong to the hymenopteran superfamilies Chalcidoidea and Ichneumonoidea (Fig. 2). All of the chalcidoids belong to the family Eulophidae, so we based the genus-level structure of this part of the tree on the molecular study by Gauthier et al. (2000) (see also Munro et al. 2011). Within Ichneumonoidea, the phylogeny of the four braconid genera follows the molecular and morphological trees of Pitz et al. (2007) and Zaldívar-Riverón et al. (2006), whereas interrelationships among the nine Ichneumonidae genera are based on the molecular and morphological phylogeny of Quicke et al. (2009). After constructing a genus-level topology for all parasitoids, we grafted congeneric species to the tips of each genus to form a polytomy, and then used Grafen's (1989) method to obtain branch lengths for the whole tree using the PDAP package v. 1.16 (Midford et al. 2011) in Mesquite.

DETERMINANTS OF TRITROPHIC ASSOCIATIONS

We used Ives and Godfray's (2006) phylogenetic bipartite linear model (pblm) as implemented in the Picante v. 1.2 package (Kembel et al. 2010) in R v. 2.10 (R Development Core Team 2009) to estimate how strongly ecological versus phylogenetic factors influence the structure of parasitoid–leafminer associations (Figs. 2, S1). The pblm function was used to estimate the signal strength arising from the parasitoid phylogeny (dp) and either the “real” miner phylogeny or the plant-based pseudophylogeny (dm) on parasitoid–miner associations. The signal-strength parameter d equals 0 if no phylogenetic effect is present (which corresponds to a star-like phylogeny), whereas a value of 1 corresponds to a model of Brownian motion evolution described in Felsenstein (1985), and values above 1 indicate signal exceeding the Brownian motion assumption (Ives and Godfray 2006). In essence, dp is high and significantly non-zero if related parasitoid species are more likely to attack the same miner species, whereas dm increases whenever (pseudo)phylogenetically close miner species are more likely to be attacked by the same parasitoid species. We obtained 95% confidence bounds for signal-strength estimates by a bootstrapping procedure using 500 replications. Overall signal strengths were evaluated by comparing mean squared errors calculated for the full model in which dp and dm are estimated from the data (MSEd), for a model assuming no phylogenetic covariances (MSEstar), and for a model incorporating a Brownian motion assumption of evolution (MSEb). Because MSE measures model error, lower values indicate a better fit of the model to the data. We also estimated phylogenetic signal strengths separately for the two parasitoid superfamilies Ichneumonoidea and Chalcidoidea. Nonparametric Kruskal–Wallis one-way analysis of variance, testing for differences in mean diet breadth among parasitoid families, and associated pairwise contrasts, were performed using SPSS v. 19.0.0.1 (SPSS, Inc., Chicago, IL).

Results

Parasitoid–miner associations appear to be highly dynamic, so that the identity and number of heterarthrines attacked can differ markedly even among congeneric parasitoids (Fig. 2). However, the main divisions in the level of specificity seem to lie among parasitoid families. Individual eulophid species attack significantly more leafminer species (mean = 7.42, SE = 1.48) than do ichneumonids (mean = 1.25, SE = 0.13) and braconids (mean = 3.63, SE = 0.73; Kruskal–Wallis test, H’ = 30.66, df = 2, P < 0.001). However, in subsequent pairwise comparisons, differences were statistically significant only between Ichneumonidae and the two other families (T* = −3.62 and −5.02 for contrasts with Braconidae and Eulophidae, respectively, both P ≤ 0.001), whereas the average host ranges of Braconidae and Eulophidae are statistically indistinguishable (T* = −0.05, P = 0.96).

Phylogenetic versus ecological effects on parasitoid–miner associations were estimated by contrasting the association matrix and the parasitoid phylogeny either with the true miner phylogeny (Fig. 2), or with a pseudophylogeny in which miner species were attached onto the plant phylogeny in accordance with their host-plant associations (Fig. S1). When the real miner tree was used, associations showed a clear non-zero signal resulting from the parasitoid phylogeny (dp = 0.120), whereas the independent effect of the miner phylogeny was essentially nonexistent (dm = 2.32 × 10−13; Table 1). However, the situation changed when the miner phylogeny was substituted with the miner–plant pseudophylogeny: The signal from the pseudophylogeny (dm = 0.333) exceeded the one from the parasitoid phylogeny (dp = 0.079), and the lower bounds of the 95% confidence intervals of both estimates were above 0 (Table 1). Furthermore, overall model error was lowered when using the plant-based pseudophylogeny (MSEd = 0.097) instead of the true miner phylogeny (MSEd = 0.113), which indicates that incorporating an effect of plant phylogeny improves the fit of the model. In both cases, however, the error from the full estimated model (MSEd) was closer to the one obtained by assuming no phylogenetic covariances (MSEstar) than to the error based on a Brownian motion model (MSEb), suggesting that overall phylogenetic effects are relatively weak or noisy (Table 1).

Table 1. Phylogenetic signal in parasitoid–miner associations when the real miner phylogeny (Fig. 2) and a miner pseudophylogeny constructed from the host-plant phylogeny (Fig. S1) are contrasted with the parasitoid phylogeny, and dm and dp measure the strength of signal arising from the miner and parasitoid phylogenies, respectively. Numbers in parentheses show approximate 95% confidence intervals for signal strengths estimated by bootstrapping. Mean squared errors are given for the full model in which dm and dp are estimated from the data (MSEd), for a model assuming no phylogenetic signal (MSEstar), and for a model assuming Brownian motion evolution (MSEb). Corresponding values are also shown for analyses based on the real miner phylogeny and parasitoids from either the superfamily Ichneumonoidea or the Chalcidoidea
Miner PhylogenydmdpMSEdMSEstarMSEb
Real phylogeny2.32×10−130.1200.1130.1220.611
 (4.08×10−37–2.56×10−07)(0.056–0.198)   
Plant-based pseudophylogeny0.3330.0790.0970.1220.219
 (0.258–0.414)(0.033–0.142)   
Real phylogeny (vs. Ichneumonoidea)5.59×10−1040.0050.0640.0640.458
 (1.65×10−175–1.86×10−13)(4.64×10−21–0.043)   
Real phylogeny (vs. Chalcidoidea)3.56×10−100.2560.1870.2090.738
 (7.83×10−31–1.47×10−4)(0.132–0.374)   

Separate analyses of the two parasitoid superfamilies likewise indicated a stronger signal from the parasitoid phylogeny than from the miner tree, but the effect was significantly non-zero only for Chalcidoidea (dp = 0.256). In the analysis restricted to Ichneumonoidea, confidence intervals of both parasitoid and miner signal strengths included zero (Table 1), and star phylogenies provided an equally good overall fit to the data (MSEd = MSEstar = 0.064).

Discussion

Research on the ecology and evolution of plant–herbivore–parasitoid food webs has been hampered by the immense number of species involved, as well as by difficulties in parasitoid identification and in establishing trophic relationships (Smith et al. 2008; Wilson et al. 2012). In the case of Heterarthrinae, combining the rearing data of Pschorn-Walcher and Altenhofer (1989) with new phylogenetic information allows for many insights in this field, especially when it comes to the assembly of parasitoid communities, factors that govern species-level host use, and the potential for bottom–up and top–down diversification effects.

ORIGIN OF THE HETERARTHRINE PARASITOID COMMUNITY

Heterarthrine leafminers most likely constitute a monophyletic group that evolved from external-feeding lineages within the tenthredinid subfamily Blennocampinae 100 to 80 million years ago (Leppänen et al. 2012). Heterarthrine larvae are plagued by a rich assemblage of hymenopteran parasitoids (Fig. 2), but few of these appear to have been inherited from the heterarthrines’ exophagous ancestors: Heterarthrines lack most of the hymenopteran, and all dipteran, parasitoids that ravage external-feeding sawfly larvae (cf. Pschorn-Walcher 1982; Price and Pschorn-Walcher 1988), whereas many of the current enemies are polyphages that attack multiple miner taxa but are not found on exophagous sawflies. The most generalized Heterarthrinae parasitoids, such as Colastes braconius Haliday and several Pnigalio and Chrysocharis species, also attack leafmining (or otherwise endophagous) herbivores in Lepidoptera, Diptera, and Coleoptera (Godfray 1994; Noyes 2012). Although the enemies of extant external-feeding tenthredinids may differ from the ones present at the time when heterarthrines originated, the minor overlap in parasitoid assemblages strongly supports Pschorn-Walcher and Altenhofer's (1989) postulate that the transition to leafmining was associated with a massive shift in parasitoid-community structure (cf. Sugiura 2007). It should be emphasized that the enemy complex on Heterarthrinae evidently has been accumulated in a stepwise fashion, because the community is not monophyletic as a whole; and even within the three parasitoid families, the involved genera are phylogenetically scattered among groups associated with other herbivore taxa (see Gauthier et al. 2000; Pitz et al. 2007; Zaldívar-Riverón et al. 2006; Quicke et al. 2009). For example, within the more specialized Ichneumonidae, the three genera including species that generally inflict the heaviest mortality on heterarthrine larvae (Lathrolestes, Dolophron, and Grypocentrus; Pschorn-Walcher and Altenhofer 1989) are from different subfamilies and, hence, are widely separated on the ichneumonid phylogeny of Quicke et al. (2009).

VARIATION AMONG PARASITOID TAXA

Parasitoid attack patterns in the focal food web are evolutionarily unstable: As in many other leafminer parasitoid communities (Morris et al. 2004; Cagnolo et al. 2011), diet breadths as well as host preferences differ widely even among congeneric species (Fig. 2). Nevertheless, our phylogenetic regressions revealed that closely related parasitoids tend to resemble each other with respect to their preferred victims (Table 1). This contrasts with the results of Ives and Godfray (2006), who found that parasitoid phylogeny did not explain associations in a food web centered on leafmining Phyllonorycter moths. The presumable reasons for the discrepancy are that our analysis included more parasitoid (59) and miner (25) species than did the Phyllonorycter web (27 and 12, respectively), and that many Phyllonorycter enemies are effectively polyphagous across the congeneric moths (Rott and Godfray 2000). Discrepant results across different miner–parasitoid systems were in fact anticipated by Ives and Godfray (2006), and detection of phylogenetic signal arising from hosts and/or parasitoids may depend on the taxonomic scale of analyses (Cagnolo et al. 2011; Desneux et al. 2012). Indeed, much of the parasitoid-tree signal in our analysis likely follows from the difference in average diet breadth between the mostly polyphagous eulophids and the more specialized ichneumonids and braconids (Fig. 2). We also found phylogenetic conservatism in host use within Chalcidoidea, whereas the signal from the parasitoid tree was negligible in the analysis restricted to the Ichneumonoidea, apparently due to the narrow and nonoverlapping host ranges of most species in this superfamily. However, some conservatism is evidenced, for example, by the fact that species in the Dolophron + Olesicampe clade are exclusively restricted to attacking Heterarthrus species (Fig. 2).

TRITROPHIC PATTERNS AND VERTICAL DIVERSIFICATION EFFECTS

Parasitoid species numbers and attack rates are generally higher in leafminers than in other herbivorous insects (Hawkins 1994). This makes plant–miner–parasitoid networks especially suited for evaluating the hypothesis that host specialization and antagonistic coevolution in multitrophic food webs accelerates diversification in both bottom–up and top–down directions (Bernays and Graham 1988; Forbes et al. 2009).

The existence of differentially specialized species within the ichneumonid subfamily Campopleginae (= Dolophron + Olesicampe) and in the genera Lathrolestes and Grypocentrus (Fig. 2) suggests that at least some bottom–up diversification of parasitoids has occurred in situ by evolutionary shifting among heterarthrine species. Both of the latter genera are, however, rich in species and include species attacking other leafmining insect taxa (Jordan 1998; Reshchikov et al. 2010), so broader species-level phylogenetic analyses of these taxa are needed to exclude the alternative explanation that multiple colonizations from other herbivore groups have occurred. Nevertheless, the fact that 4 of 8 braconid species and 11 of 19 eulophids—that attack the same set of heterarthrine victims—have remained polyphagous (= feed on at least three heterarthrine genera and, in many cases, other leafminer taxa) instead of speciating, attests to the complexity of bottom–up forces. Although a few studies have found that diversification can cascade up across trophic levels (Stireman et al. 2006; Forbes et al. 2009), others have discovered that herbivore host-plant shifts have not triggered speciation in parasitoids (Cronin and Abrahamson 2001; Althoff 2008). Naturally, we cannot completely rule out the possibility that some of the perceived generalist enemies on Heterarthrinae are in fact complexes of cryptic specialists. Recent molecular-genetic studies have uncovered numerous morphologically indistinguishable parasitic species and, hence, have indicated that hymenopteran parasitoids may be even more specialized than is currently thought (Smith et al. 2008; Nicholls et al. 2010; Kaartinen et al. 2010).

More research has focused on parasitoid-driven top–down diversification, which could occur if herbivores can escape parasitoids into enemy-free space provided by alternative host plants (Bernays and Graham 1988; Singer and Stireman 2005). The key requirement for enemy-driven host divergence is that enemies are specialists on herbivore niches rather than on herbivore lineages. Examples of both parasitoid types can be found in the Heterarthrinae–parasitoid food web (Figs. 2, S1) but, overall, host-plant phylogeny constitutes a far better explainer of parasitoid attack probability than do relationships among miner species (Table 1). A potential caveat is that the plant-phylogeny effect could be amplified by unequal sample sizes in rearings of miner species collected from different plant hosts (Rott and Godfray 2000; Nyman et al. 2007). For example, the highest average numbers of enemies were found on miners on Alnus and Betula (16.0 and 11.8 parasitoid species per miner species, respectively; Fig. S1), both of which host miners that were reared in large amounts by Pschorn-Walcher and Altenhofer (1989). However, the authors noted that sample sizes of the heterarthrines included in their detailed lists should be adequate to detect all but the rarest enemy species (Pschorn-Walcher and Altenhofer 1989, p. 44), and these betulaceous taxa also support a comparatively rich community of non-hymenopteran leafminers (Kennedy and Southwood 1984), which could act as a reservoir for the generalized eulophids.

Our finding of a clear host-plant effect on parasitoid–miner associations conforms fully with results from the Phyllonorycter food web (Ives and Godfray 2006), as well as with inferences from a few other herbivore guilds (Lill et al. 2002; Murphy 2004; Nyman et al. 2007). Nevertheless, the “noisiness” of the association matrix means that the possible diversifying force generated by parasitoids will depend on the relative abundance of different enemies, and especially on the combined mortality inflicted by generalists versus specialists. In the majority of heterarthrine miners, the most prevalent parasitoids belong to the “specialist” genera Lathrostizus, Grypocentrus, Dolophron, or Olesicampe, but some of the generalists (such as Chrysocharis nitetis [Walker]) are also common (Pschorn-Walcher and Altenhofer 1989). More importantly, the large number of generalist species means that the summed-up mortality caused by them may be high. Measuring the mean abundance of individual parasitoids is needed to assess the overall diversifying force imposed by the heterarthrine enemy complex, but this may be difficult as rates of parasitism and parasitoid community structure are known to fluctuate widely in time (Karban and de Valpine 2010; Stone et al. 2012) as well as in space (Heard et al. 2006; Klapwijk and Lewis 2011).

Conclusions

The tritrophic food web centered on heterarthrine leafminers evidently has been assembled through a complex history of shifts by miners and parasitoids among hosts and victims on lower trophic levels, which is likely to be the case also in other species-rich plant–herbivore–parasitoid networks (Lopez-Vaamonde et al. 2005; Zaldívar-Riverón et al. 2008). Our results support the view that cascading bottom–up speciation has a significant—but not ubiquitous—role in herbivore and parasitoid diversification. Probably less than half of the speciation events in the miners have involved shifts in host-plant use (Leppänen et al. 2012; cf. Winkler and Mitter 2008; Imada et al. 2011), and at least a quarter of the associated parasitoid species have remained polyphagous despite speciation and niche diversification in heterarthrines (cf. Cronin and Abrahamson 2001; Althoff 2008).

The significant effect of leafminer host plants on parasitoid attack patterns fulfills a key theoretical requirement for top–down herbivore diversification driven by enemy-free space in novel niches (Singer and Stireman 2005; Nyman et al. 2007). However, this does not as such prove that heterarthrine host-plant shifts were driven by parasitoid pressure, and the fact that the level of specialization varies dramatically among parasitoid taxa suggests that conditions favoring enemy-driven niche diversification occur only transiently or in geographically restricted settings (Thompson 2005; Heard et al. 2006). Large-scale studies on temporal and spatial variation in plant–herbivore–parasitoid metacommunities would evidently be a fruitful avenue for further research, especially if phylogenetic and newly developed coalescent approaches (Stone et al. 2012) can be applied to infer past community structures.

Interestingly, host shifts in many heterarthrine lineages have occurred onto plant genera that already harbor other heterarthrines (Leppänen et al. 2012; Fig. 1A). Such repeated convergent shifts, which have also been found in other insect herbivore groups (Mardulyn et al. 1997; Nyman et al. 2010; Slove Davidson 2012), are somewhat unexpected considering that the likelihood of finding enemy-free space should be highest on plants that are not used by related herbivores. We suggest that the oft-observed evolutionary accumulation of herbivorous insects on dominant plant taxa (Kennedy and Southwood 1984; Lewinsohn et al. 2005) results from a situation in which the long-term benefits of living on a common host outweigh the adverse effects of a potentially species-rich enemy community.

ACKNOWLEDGMENTS

We would like to thank all colleagues who provided sawfly specimens for the phylogenetic reconstructions, as well as A. R. Ives, A. Valtonen, and M. Helmus for their kind help with the tritrophic analyses. We are also grateful for T. P. Craig and two anonymous referees for helpful comments on previous versions of the manuscript. Funding for this project was provided by the Academy of Finland (project 14868 for T. Nyman).

Ancillary