To each its own: differential response of specialist and generalist herbivores to plant defence in willows
Summary
- Plant–insect food webs tend to be dominated by interactions resulting from diffuse co-evolution between plants and multiple lineages of herbivores rather than by reciprocal co-evolution and co-cladogenesis. Plants therefore require defence strategies effective against a broad range of herbivore species. In one extreme, plants could develop a single universal defence effective against all herbivorous insects, or tailor-made strategies for each herbivore species. The evolution and ecology of plant defence has to be studied with entire insect assemblages, rather than small subsets of pairwise interactions.
- The present study examines whether specialists and generalists in three coexisting insect lineages, forming the leaf-chewing guild, respond uniformly to plant phylogeny, secondary metabolites, nutrient content and mechanical antiherbivore defences of their hosts, thus permitting universal plant defence strategies against specialized and generalist folivorous insects from various taxa.
- The extensive data on folivorous assemblages comprising three insect orders and 193 species are linked with plant phylogeny, secondary chemistry (salicylates, flavonoids and tannins), leaf morphological traits [specific leaf area (SLA) and trichome coverage], nutrient (C : N) content and growth form of eight willow (Salix) and one aspen (Populus) species growing in sympatry.
- Generalists responded to overall host plant chemistry and trichomes, whilst specialists responded to host plant phylogeny and secondary metabolites that are unique to willows and that are capable of being utilized as an antipredator protection. We did not find any significant impact of other plant traits, that is SLA, C : N ratio, flavonoids, tannins and growth form, on the composition of leaf-chewing communities.
- Our results show that the response to plant traits is differential among specialists and generalists. This finding constrains the ability of plants to develop defensive traits universally effective against herbivores and may lead to diversification of plant defensive mechanisms into several complementary syndromes, required for effective protection against generalists and specialists from multiple insect taxa comprising most leaf-chewing assemblages. These results point to the necessity of broad studies of plant–herbivore interactions, across multiple insect taxa and guilds.
Introduction
Ehrlich & Raven (1964) suggested that herbivorous insects and their host plants co-evolve, leading to the genesis of novel plant defences followed by the origin of specialized herbivores able to overcome the enhanced protection. Although there are some well-studied examples showing tight co-evolution and co-cladogenesis (Farrell & Mitter 1990; Cruaud et al. 2012), the majority of plant–insect interactions result from diffuse co-evolution between plants and insect assemblages (Janz 2011) where host switches are common even in the systems with high consumer specialization (Wilson et al. 2012). As a result, host plant defensive traits tend to be better predictors of insect community composition than host phylogeny per se (Becerra 1997), although plant traits governing insect food choice often differ among herbivores (Koricheva, Nykänen & Gianoli 2004).
Herbivores with different levels of specialization are frequently affected by different plant characteristics (Ali & Agrawal 2012). Whereas unspecialized insects are often excluded from plants defended by unique or highly toxic secondary metabolites (Becerra 1997; Agrawal 2005), insect specialists have repeatedly evolved mechanisms to overcome toxic or deterrent effects of such defences (Denno, Larsson & Olmstead 1990; Treutter 2006). Physical defensive traits, such as trichomes or leaf toughness, may be more effective against specialists as they reduce feeding efficiency (Dimarco, Nice & Fordyce 2012). Plant defences affecting generalists tend to increase mortality, whereas those affecting specialists prolong time needed for feeding, increasing the mortality indirectly through enhanced risk of predation or parasitism (Richards et al. 2010). The importance of nitrogen content also differs between specialist and generalist herbivores. In specialized insects that are able to cope with toxins, nitrogen is often an important determinant of host preference as its high content may help to overcome the impact of traits lowering feeding efficiency. In contrast, generalist insects may not be able to fully overcome toxic host chemistry and are thus prevented from benefiting from high nitrogen content (Coley, Bateman & Kursar 2006).
The effectiveness of defensive traits against specialist and generalist herbivores, which often use different mechanisms to overcome plant defence, in combination with plant tolerance to herbivore damage may shape host plant defensive patterns (Ali & Agrawal 2012). For instance, the defence of individual Piper species appears to be dependent on either secondary metabolites, physical traits or protection by ants (Fincher et al. 2008). Such a strategy can be effective only if various herbivore groups, such as specialists and generalists, respond uniformly to a particular plant defence. In contrast, defensive traits were uncorrelated or correlated positively among Asclepias species (Agrawal & Fishbein 2006), suggesting that effective protection can be maintained by a set of defensive traits forming a complex defence against multiple herbivores with different levels of specialization (Koricheva, Nykänen & Gianoli 2004).
A variable impact of particular defensive traits on specialist and generalist herbivores has been amply demonstrated (e.g. Richards et al. 2010; Reudler et al. 2011). However, it is important to quantify this impact at the level of the entire herbivore assemblage, as insect–plant food webs tend to be dominated by interactions resulting from diffuse co-evolution rather than from pairwise co-evolution with single plant and herbivore species (Janzen 1980; Futuyma 2000). There is an increasing body of literature focused on explaining host plant defences, but most studies relate these defences to a single herbivore species or lineage, representing a small subset of herbivore assemblages, or study only herbivory damage, lacking insect data completely (Becerra 1997; Agrawal, Lajeunesse & Fishbein 2008; Pearse & Hipp 2012; Schuldt et al. 2012; Peñuelas et al. 2013). Although studying model insect species and their impact on host plants provides valuable information on pairwise insect–plant interactions, it may be hard to apply these findings to species-rich communities dominated by diffuse interactions. To our knowledge, there is only one study analysing responses to defensive traits by a diverse insect assemblage within a phylogenetic context (Lavandero et al. 2009), but this study used regional host plant data collated from the literature rather than species locally coexisting on each plant species.
Salix, one of the few species-rich genera of woody plants in Europe, is an excellent model for the analysis of defensive traits and their impact on herbivorous assemblages. Willows are widely distributed trees and shrubs with diverse defensive mechanisms, hosting rich communities of herbivorous insects consisting of species from several lineages with different levels of specialization. Leaves of willows are defended by trichomes and common secondary metabolites, such as flavonoids and condensed tannins, but also by salicylates. Salicylates are phenolic glycosides that are characteristic of, and to a large extent unique to, the Salicaceae family (Julkunen-Tiitto 1989). Salicylates function mainly as a defence against herbivores. They have been found to be effective against unspecialized herbivores, deterring them from feeding and increasing larval mortality (Matsuki & Maclean 1994; Kolehmainen et al. 1995). Despite this defence, willows harbour numerous insect species ranging from generalists to specialists, including well-adapted herbivores that can even use salicylates as a source of energy or protection against invertebrate predators (Rowell-Rahier & Pasteels 1986; Denno, Larsson & Olmstead 1990).
Using species-rich communities of herbivores associated with Salix hosts, we analyse the effect of phylogenetic distance and plant traits on willow specialists and generalists from a leaf-chewing guild. We test whether assemblages of specialists and generalists from three coexisting insect lineages respond uniformly to plant phylogeny, secondary metabolites, mechanical antiherbivore defences, nutrient content and growth form of their hosts, thus permitting universal plant defence strategies against various folivorous insects with different levels of specialization and life history.
Material and methods
Host Plants and Study Sites
Willows are usually divided into species defended mainly quantitatively by tannins and those defended qualitatively by salicylates (Julkunen-Tiitto 1989). The eight species of trees and shrubs from the genus Salix (S. aurita, S. caprea, S. cinerea, S. fragilis, S. pentandra, S. purpurea, S. rosmarinifolia and S. viminalis) studied here were selected to represent both of these groups. Further, Populus tremula was studied as an outgroup (Table 1).
Growth form | Specific leaf area (cm2 g−1) | Trichome cover (%) | Flavonoids (mg g−1) | Salicylates (mg g−1) | Tannins (mg g−1) | Carbon (mg g−1) | Nitrogen (mg g−1) | |
---|---|---|---|---|---|---|---|---|
Salix aurita (4) | Shrub | 144·8 ± 27·0 | 19 ± 3·0 | 29·5 ± 4·4 | 0·0 | 194·8 ± 44·4 | 45·4 ± 1·1 | 3·05 ± 0·66 |
Salix caprea (6) | Tree | 146·3 ± 31·8 | 26 ± 3·5 | 10·6 ± 2·2 | 0·0 | 138·8 ± 43·0 | 43·8 ± 1·4 | 2·51 ± 0·62 |
Salix cinerea (7) | Shrub | 131·5 ± 38·9 | 21 ± 2·0 | 15·0 ± 2·5 | 0·0 | 159·1 ± 61·3 | 45·6 ± 1·9 | 2·93 ± 0·58 |
Salix fragilis (6) | Tree | 134·8 ± 32·7 | 0 | 25·5 ± 7·2 | 27·8 ± 10·0 | 51·9 ± 33·2 | 44·0 ± 1·0 | 2·98 ± 0·60 |
Salix pentandra (3) | Tree | 118·5 ± 39·6 | 0 | 60·6 ± 10·5 | 41·8 ± 20·3 | 190·7 ± 34·3 | 45·2 ± 2·0 | 2·56 ± 0·75 |
Salix purpurea (5) | Shrub | 141·2 ± 39·8 | 0 | 21·3 ± 2·6 | 164·8 ± 36·5 | 42·7 ± 56·5 | 44·1 ± 0·8 | 2·74 ± 0·89 |
Salix rosmarinifolia (2) | Shrub | 125·3 ± 29·9 | 14 ± 4·0 | 20·9 ± 1·2 | 169·0 ± 32·2 | 133·4 ± 82·1 | 45·5 ± 0·6 | 2·47 ± 0·91 |
Salix viminalis (4) | Shrub | 165·8 ± 29·8 | 36 ± 7·3 | 16·0 ± 3·4 | 0·0 | 137·4 ± 35·6 | 46·0 ± 3·6 | 3·57 ± 0·62 |
Populus tremula (5) | Tree | 144·5 ± 76·9 | 0 | 33·8 ± 8·3 | 19·4 ± 14·9 | 38·2 ± 35·6 | 43·3 ± 2·8 | 2·51 ± 0·70 |
Our study was carried out within a 10 × 10 km area comprising lowland wetlands and wet meadows, situated in South Bohemia, Czech Republic (48°51′58′′–48°59′45′′N, 14°26′20′′–14°35′48′′E). All host plants studied represented mature trees and mature shrubs. Shaded plants were excluded since their life-history traits (including leaf chemistry) could be significantly different from plants exposed to sunlight. We also avoided plants which had obviously experienced browsing by herbivores or damage from other sources prior to the sampling as these factors can cause significant changes in plant traits (Nakamura et al. 2005; Ohgushi 2005). All host plant traits (described below) were measured for two to seven plant individuals per species (a total of 44 plants), and means were used to characterize each species (Table 1).
Physical Traits
We measured trichome density and specific leaf area (SLA), a surrogate for leaf thickness and toughness (Groom & Lamont 1999), as measures of leaf morphology with a possible impact on leaf-chewing insects. Leaves from the central part of shoot were used for the measurement, avoiding apical leaves. Trichome density was estimated as the average percentage of surface area (0·5 cm square) covered by trichomes for mature leaves; values for dorsal and ventral sides were averaged. SLA was calculated as the area per unit mass of a dried leaf disc of known diameter. Leaf discs were cut, avoiding the central vein, and air-dried to constant weight. Three leaf discs were obtained every 2 weeks throughout the 2010 vegetation season (10 sampling occasions) for each of the 44 plant individuals examined. The obtained values were used to estimate the mean SLA.
Chemical Analysis
For chemical analyses, samples of leaf lamina were cut (avoiding primary and secondary leaf veins) and air-dried at room temperature immediately after collection. We used the central parts of the leaf blade, avoiding both apical and basal part. Total carbon and nitrogen content was determined by dry combustion with a Carlo Erba NC 2500 element analyser (Carlo Erba, Milano, Italy) using 30 mg of dried and homogenized leaf material. Tissue sampling was repeated six times throughout the 2010 vegetation season. For all nine study species, we sampled three individuals to estimate total N and C content over the entire vegetation season. The obtained N and C contents were used to estimate the mean C : N ratio.
The contents of salicylates, flavonoids and condensed tannins were analysed from 5 to 9 mg samples from leaves sampled at the beginning of June when insect density in the study area reaches its peak. Phenolic compounds were extracted with methanol as described in Nybakken, Horkka & Julkunen-Tiitto (2012). Extracts were dried and kept in a −20°C freezer. Before the analysis, dried samples were redissolved in 600 μl of 1 : 1 methanol–water solution. We used 20 μl of redissolved samples for HPLC analysis of salicylates and flavonoids following Nybakken, Horkka & Julkunen-Tiitto (2012). Compounds were separated using a Zorbax SBC18 (4·6 × 60 mm) HPLC column (Agilent Technologies, Santa Clara, CA, USA) employing a water/methanol gradient (Julkunen-Tiitto & Sorsa 2001). Salicylate and flavonoid contents were measured based on the absorbance at 270 nm and 320 nm, respectively. Retention times and spectra compared to standards were used to identify the compounds.
Soluble condensed tannins were determined by the acid–butanol assay according to the method of Hagerman (2002) from an aliquot of the HPLC sample and insoluble condensed tannins from room-dried tissue residues. After hydrolyses, absorbance values at 550 nm were measured (Spectronic 20 Genesys TM spectrophotometer, Thermo Fisher Scientific, Waltham, MA, USA). The condensed tannin content was calculated based on a calibration curve obtained for S. purpurea leaf tannins.
Host Plant Phylogeny Reconstruction
We chose four loci that are usually variable at genus level for host plant phylogeny reconstruction: matK, ITS, trnT-trnL and ADH. We used standard procedures, reaction conditions and primer sequences for DNA extraction and PCR amplification, which were the same as those used in the original studies employing these markers (White et al. 1990; Taberlet et al. 1991; Cronn et al. 2002; Selosse, Bauer & Moyersoen 2002; Savage & Cavender-Bares 2012). Since multiple copies of ADH were present in each individual except S. viminalis, the ADH PCR products were cloned to separate potential paralogs and hybrid sequences. Some S. cinerea and S. fragilis individuals exhibited a hybrid origin for some of their ADH sequences. This trend was pronounced in individuals sharing a site with their sibling species. Their sequences therefore did not form monophyletic lineages and the position of a proportion of the sequences was reconstructed with high support as an internal group within the sibling species. Since willow species are known to frequently hybridize with their sibling species (Skvortsov 1968), these sequences were considered of hybrid origin and removed from analysis.
Sequences were assembled and edited using geneious 5.4 (Drummond et al. 2011). Trees for individual genes were not in conflict, allowing us to reconstruct the host plant phylogeny based on a matrix with all four loci combined. Host plant phylogeny was reconstructed using Bayesian inference in MrBayes 3.1.2 (Ronquist & Huelsenbeck 2003). A GTR substitution model selected using Akaike Information Criterion (AIC) was used for Bayesian analysis with a flat Dirichlet prior probability density for the distribution of substitution rates and stationary nucleotide frequencies. Sampling was carried out every 103 generations for 107 generations. The first 25% of generations were discarded as burn-in, and the results were summarized with a 50% majority-rule consensus tree.
Insect Sampling
We focused on externally feeding and semi-concealed leaf-chewing insects as this guild often inflicts the most damage among insect herbivores (Schoonhoven, van Loon & Dicke 2005). This guild also includes various distantly related insect lineages (e.g. Coleoptera, Lepidoptera and Hymenoptera). All leaf-chewing insects were sampled during the 2008–2011 vegetation seasons (April–September) at 1-week intervals by sweeping the foliage and by manual searching, and our samples also included leaf-tying and leaf-rolling herbivores. We kept the time of sampling events constant, with 3 min of sweeping and 3 min of manual searching per inspection. Due to variation in willow population densities, the sampling effort was not completely balanced. For most tree species, the total sampling effort was 200–400 min; however, for S. pentandra, which is locally rare, total sampling effort was only 100 min. Insect larvae were reared to adults for identification. Dead larvae were morphotyped based on photographs or discarded when morphotyping proved impossible.
Statistical Analyses
Insect abundance was standardized as the number of insects obtained per sampling time (in minutes). Singletons and doubletons were excluded from analyses. The number of host plants used by individual insect species was analysed by anova to compare host specificity among insect lineages. We compared the host plant specificity of the examined herbivores by anova using Kullback–Leibler distances to remove the bias due to total observation frequencies (Blüthgen, Menzel & Blüthgen 2006). The impact of chemical composition and host plant phylogeny on insect communities was quantified by two analyses – a Mantel test and a multivariate analysis. We used a Mantel test to examine how insect community similarity reflects willow phylogeny and secondary metabolite dissimilarity (representing overall willow chemical defence) and multivariate ordination to analyse the role of individual plant traits in forming insect communities.
The similarity of herbivore communities between willow species employed in the Mantel test and partial Mantel test (with phylogenetic distance used as covariate) was estimated using the Bray–Curtis index, computed in estimates 8.2 (Colwell 2006). A chemical dissimilarity matrix was obtained using UPGMA with Euclidean distances based on log-transformed mean concentrations of individual salicylates and flavonoids, and log-transformed total concentration of condensed tannins. For visualization, the similarities of host plant chemistry and insect communities were also analysed by PCA employing the same data. The phylogenetic distances (in substitutions per base) between host plants were based on the mean branch lengths derived from the Bayesian phylogeny. We chose to use a one-sided Mantel test, as we expected that both chemically similar host plants and closely related host plants will support similar insect communities. Following the whole community analysis, separate similarity matrices for Coleoptera, Lepidoptera, Hymenoptera, generalist herbivores and Salicaceae specialists (i.e. species feeding only on Salicaceae; based on Smreczyński 1966, 1972, 1974; Lacourt 1999; Warchalowski 2003; Macek et al. 2007; Kopelke 2007a,b; Macek et al. 2008; Macek, Prochazka & Traxler 2012) were computed and analysed as above.
The effect of individual defensive traits on insect communities was analysed using multivariate ordination analyses conducted in canoco for Windows 4.56 (ter Braak & Smilauer 2002). The impact of total salicylate, flavonoid and condensed tannin concentrations, trichomes, SLA, C : N ratio and plant growth form on herbivore community composition harboured by different host plant species was analysed by redundancy analysis (RDA) with host plant phylogeny used as a covariate. Phylogenetic distances between plant species were transformed from the ultrametric tree into coordinates using PCoA. Since the values of host plant traits were not available for all plant individuals from which insects were sampled or for all periods of the season, we used mean values for plant species as individual data points. In this analysis, the insect data obtained from conspecific plant individuals were combined into one data point and variables best explaining their variability were selected by forward selection under 9999 permutations.
Results
Host Plant Traits and Phylogeny
Host plant phylogenetic distance and chemical dissimilarity were not correlated (r = 0·15, d.f. = 43, P = 0·334) (Fig. S1, Supporting information). The plant species studied here differed widely with respect to their morphological and chemical traits (Tables 1 and S1, Supporting information). In particular, large differences were found in trichome density and salicylate content and composition. Most of the salicylate and flavonoid compounds found were unique to a single plant species (Fig. S2, Supporting information), which resulted in a high level of reconstructed variability in chemical composition. PCA analysis revealed four distinctive groups of willows with assorted chemical profiles: (I) S. ix purpurea and S. rosmarinifolia with very high salicylate content; (II) S. aurita, S. caprea, S. cinerea and S. viminalis containing no salicylates; (III) S. fragilis and P. tremula with moderate salicylate content; and (IV) S. pentandra with chemical profile distinctively different from other studied host plants (Fig. 1b).

Willow sequences were rather conservative, with a low proportion of informative sites. The proportion of informative sites in matK was extremely low, and we excluded this marker from further analyses. ADH was the most informative marker, although its use was limited by the presence of hybrid sequences requiring cloning. Bayesian inference provided a topology supporting the traditional willow taxonomy and suggesting monophyly of both examined willow subgenera, Salix and Vetrix (Skvortsov 1968) (Fig. 1a). However, support for some of the clades was quite low. The most ambiguous grouping is that of S. viminalis as it is often reconstructed as a sister species to S. caprea, S. cinerea and S. autita group or forms a monophyletic group with S. purpurea and S. rosmarinifolia.
Insect Communities
We collected 7786 individuals of leaf-chewing insects from 192 species, representing three insect orders – Coleoptera, Hymenoptera and Lepidoptera. Salicaceae specialists included 28 Coleoptera, 49 Hymenoptera and 29 Lepidoptera species. Generalists included 30 Coleoptera, 4 Hymenoptera and 52 Lepidoptera species (Table S2, Supporting information). We found significant difference in host breadth with examined orders (F2,109 = 8·46, P < 0·001) with Hymenoptera being the most host specific and Coleoptera and Lepidoptera being moderately and the least host specific, respectively. Hymenoptera also included the highest proportion of Salicaceae specialists, that is species feeding only on Salicaceae (96%), whereas the specialist proportion was moderate in Coleoptera (49%) and lowest in Lepidoptera (32%). The PCA analysis pointed to relatively smaller variability in herbivorous insect community structure on low salicylate willows compared to large interspecific variability in herbivore community structure on high salicylate willows (Fig. 1c).
Impact of Host Plant Traits on Insect Communities
Both total chemical dissimilarity (r = 0·37, d.f. = 43, P = 0·027) and phylogenetic distance (r = 0·38, d.f. = 43, P = 0·028) exhibited a significant impact on insect community similarity (Table 2). When specialists and generalists were analysed separately, generalist community structure was significantly affected by chemical dissimilarity, whereas specialist community structure was affected by phylogenetic distance (Table 2, Fig. 2).
P/r | |||
---|---|---|---|
Whole community | Specialists | Generalists | |
Phylogenetic distance | 0·028/0·38 | 0·023/0·45 | 0·103/0·24 |
Chemical dissimilarity | 0·027/0·38 | 0·078/0·23 | 0·004/0·54 |
Chemical diss. (par. Mantel) | 0·049/0·34 | 0·107/0·18 | 0·007/0·52 |
- Significant (P < 0·05) results are highlighted in bold, d.f. = 43 for Mantel tests and d.f. = 42 for partial Mantel tests.

Redundancy analysis revealed significant impact of host plant phylogeny on community structure of leaf-chewing herbivores on willows (F1 = 1·3, P = 0·044). However, when other plant traits were added into analysis, phylogeny had no effect on insect communities (F3 = 1·3, P = 0·178) as chemical and physical plant traits explained insect community structure better. RDA analysing the impact of individual physical and chemical plant traits, including host plant phylogeny as a covariate, showed significant effects of total salicylate content and trichome cover on herbivore communities (Fig. 3, Table 3). Salicylates exhibited a significant effect on the whole leaf-chewer communities (F1 = 1·5, P = 0·031) and Salicaceae specialists (F1 = 1·5, P = 0·043). Some Salicaceae specialists exhibited a strong positive response to salicylates, whereas the majority of specialists showed a weak negative or nearly no response to these secondary metabolites (Fig. S3, Supporting information). Trichomes played a significant role in structuring assemblages of generalists (F1 = 1·80, P = 0·035).

P/F | |||
---|---|---|---|
Whole community | Specialists | Generalists | |
Salicylates | 0·032/1·50 | 0·040/1·27 | 0·059/1·37 |
Flavonoids | 0·523/0·96 | 0·517/0·90 | 0·497/1·01 |
Tannins | 0·305/1·16 | 0·265/1·16 | 0·406/1·08 |
Trichomes | 0·224/1·18 | 0·367/1·11 | 0·035/1·80 |
Specific leaf area | 0·489/0·98 | 0·498/0·98 | 0·429/1·07 |
C : N | 0·507/0·98 | 0·500/0·99 | 0·455/1·02 |
Growth form | 0·180/1·31 | 0·501/1·00 | 0·332/1·30 |
- Significant results are in bold.
In the analysis comparing the response of insect orders to defensive traits, Coleoptera responded to host plant chemical dissimilarity and salicylate content and Hymenoptera to trichomes (Tables S3 and S4, Supporting information).
Discussion
In this study, we analysed whether assemblages of specialists and generalists from three coexisting insect orders respond uniformly to host plant traits. Our results suggest that specialists and generalists exhibit a specific response to host plant traits, which may play a major role in forming host plant defences.
There were major differences in the importance of host plant phylogeny for community structure of examined insect lineages. These differences appeared to be linked to the level of specialization of respective herbivores. Host plant phylogeny had a significant effect on the specialist component of the leaf-chewing community, that is those species feeding exclusively on Salicaceae. However, although some lineages of insect specialists are known to have diversified on Salicaceae, their phylogenetic conservatism probably does not result from co-speciation. For instance, Chrysomela leaf beetles, ancestrally associated with Salicaceae, show multiple reversals back to Salicaceae from other hosts, suggesting there is a higher chance that host shifts will occur to plants that had been used by the lineage in the past than shifts to novel host plants (Termonia et al. 2001). This pattern of host shifting back and forth between the same plant lineages may result in phylogenetic conservatism, as was observed here (Janz, Nyblom & Nylin 2001). However, correlations between plant phylogeny and insect community structure may be observed only in specialist lineages which do not include numerous monophages since such species cannot contribute to the relationships between plant phylogenetic distance and community similarity, as demonstrated in this study by the absence of significant effects of plant phylogeny on Hymenoptera in this study. The congruence of host plant phylogeny and insect community structure is also frequently an outcome of conservative evolution of plant traits important for insect host preference (Becerra 1997). Although sometimes conserved on higher phylogenetic level, these traits are often variable within plant genera (Julkunen-Tiitto 1989; Fincher et al. 2008). Host plant phylogeny is thus expected to play a minor role as a determinant of community assembly of herbivores on large plant genera, particularly for generalists that respond predominately to host plant defences.
Previous studies showed that secondary metabolite profile may differ among willow genotypes (Hochwender & Fritz 2004). Nevertheless, this variability among willow genotypes is generally considered to be smaller than differences among species (Nyman & Julkunen-Tiitto 2005). Although our results show that differences in secondary metabolite profile can be rather small, especially among salicylate poor species, willows examined in this study formed several chemically well-defined groups that have significant impacts on insect communities.
Specialists and generalists differed in their responses to host plant chemical profiles. Generalists were affected by total host plant chemistry, whereas specialists were affected only by secondary metabolites unique to willows. The impact of total chemical dissimilarity (based on salicylates, flavonoids and tannins) on generalists shows that willow secondary metabolites have a strong impact on less adapted groups and suggests that a degree of adaptation is required in order to overcome chemical defences of willows. On the other hand, the lack of significant effect of individual groups of secondary metabolites revealed by RDA suggests that none of these secondary metabolite groups might be effective enough to affect generalist communities when employed alone.
Multivariate analysis highlights the importance of salicylates in host plant preference by specialists, indicated by their significant impact on specialist community structure. Although high salicylate content had a slightly negative impact on some Salicaceae specialists, certain specialists showed strong positive response to the secondary metabolites. The Phratora leaf beetles, which utilize salicylates and use them for protection against invertebrate predators (Rowell-Rahier & Pasteels 1986), were among the species with strongest positive response. Other specialists showing positive response to salicylates may use them as an extra source of energy which explains their faster growth on willows with high salicylate content (Matsuki & Maclean 1994). There is also a phagostimulating effect of salicylates on some specialists (Kolehmainen et al. 1995), which may interfere with preference for high nitrogen content of host plants, possibly explaining why the C : N ratio had no effect on specialist assemblages on willows. On the other hand, willow secondary metabolites are known to be effective against generalists by increasing their mortality (Matsuki & Maclean 1994). For plants with toxic secondary metabolites, generalist food choice is thus likely to be governed by secondary metabolite content rather than C : N ratio, as observed here. In summary, although secondary metabolites of willows influence both generalist and specialist community structure, the response of these two herbivore groups to willow chemistry differs.
Growth form and plant architecture may play an important role in forming herbivorous insect communities, perhaps related to predation and parasitism risk (Lavandero et al. 2009; Sipos & Kindlmann 2013). However, we did not find any significant effect of plant growth form on leaf-chewing insect communities associated with willows. Since many studies reporting significant impact of host plant architecture focused on plants with less pronounced chemical defence (Marquis, Lill & Piccinni 2002; Sipos & Kindlmann 2013), our findings may indicate a lesser role of plant architecture in structuring insect communities on chemically well-defended plants. Large interspecific differences in defensive traits may be more important for structuring insect communities in such cases. However, it would require further analysis incorporating the third trophic level to confirm this.
Some studies reported pronounced impacts of physical defences on specialists (Dimarco, Nice & Fordyce 2012), but our study indicates in contrast that trichomes affected the community structure of generalists. Moreover, some of our results suggest that insect response is not directly connected with the level of specialization as trichomes also affected Hymenoptera, which included almost only specialized insects. We suggest that life-history traits other than specialization may be more important determinants of insect response to trichomes. Since trichomes influence mainly small insects (Agrawal 2005), body size and traits correlated with it (e.g. size of mandibles or ovipositor length) could be such factors. The second examined physical trait, SLA, had no impact on insect assemblages, probably because of low variability in this trait among examined willow species.
Differential response by herbivores to defensive traits may restrain plants from developing a universal antiherbivore defence. This may lead to defensive trait diversification. In the case of willows, specialized insects were able to adapt to salicylates and reach high densities on salicylate-rich hosts. Although salicylates play a significant role in structuring insect communities, their protective value against specialized herbivores appears to be low. Maintaining an effective defence thus probably requires several defensive mechanisms, such as chemical defence and trichomes which affect both generalists and some specialists. These findings suggest plant defensive traits to be mutually independent or positively correlated, as observed by Agrawal & Fishbein (2006) or Hattas et al. (2011). Trade-offs between individual defensive traits may be expected only given specific conditions, such as in low nutrient environments or in the case of negative dependence in metabolic pathways (Agrawal, Salminen & Fishbein 2009; Sampedro, Moreira & Zas 2011).
Our analysis revealed a significant response to host plant traits by assemblages of insect species sharing similar levels of specialization. With a systematic response by specialists and generalists, an effective plant defence may be based on a relatively small number of defensive traits since each defensive trait is likely to affect multiple herbivore species with similar levels of specialization. In the case of willows, a combination of physical defence and secondary metabolites probably provides good protection against a large proportion of both specialists and generalists.
It appears that defensive syndromes are not phylogenetically conserved. Related species thus often exhibit diverging strategies relying on different traits in their protection against insects (Agrawal & Fishbein 2006; Fincher et al. 2008). This pattern could be tentatively identified also in willows characterized by large interspecific variation in salicylate concentration and trichome density – two defensive traits with the most pronounced impact on herbivores. It may be advantageous for sibling species to use different mechanisms for their defence, as this may lower the number of host shifts by insect herbivores between related plant species. These shifts would otherwise be likely due to phylogenetic conservatism. Colonization from phylogenetically more distant host plants remains less likely, as many plant traits important for insect preference, such as plant phenology, are phylogenetically rather conservative (Davies et al. 2013). This suggests that herbivore pressure may lead towards divergence in defensive syndromes between closely related plant species, or bias community assembly towards chemical heterogeneity as reported by Becerra (2007), which may help plants to escape herbivory.
In summary, our results show that the response to plant traits by herbivores differs systematically among insects with different levels of specialization. This constrains the ability of plants to develop defensive traits that are universally effective against a broad range of herbivores and may lead to diversification of plant defensive mechanisms into several complementary syndromes, required for maintaining effective protection against diverse insect communities. These findings suggest that plant defences should be considered from the perspective of diffuse impact from a broad range of insect species rather than as a result of reciprocal co-evolution with a particular insect lineage. Further, the impact of the next trophic level, predators and parasitoids, on herbivores, as well as induced plant defences, is among other important factors potentially modifying plant–herbivore interactions (Ohgushi 2005; Wilson et al. 2012). Studying these factors is required for comprehensive understanding of the insect–plant associations and the resulting host plant defensive patterns.
Acknowledgements
We thank Jan Macek, Lukas Sekerka and Richard Ctvrtecka for assistance with insect identification; Anna Odvarkova for help with insect sampling; Michaela Borovanska and Jessica Anne Savage for assistance with molecular analysis; Sinikka Sorsa for help with chemical analyses; and Conor Redmont, Simon Segar and anonymous reviewers for helpful criticism of the manuscript. This work was supported by Grant Agency of the Czech Republic (14-04258S), Czech Ministry of Education and European Commission (CZ.1.07/2.3.00/20.0064), the Grant Agency of the University of South Bohemia (GAJU 156/2013/P), US National Science Foundation (DEB-0841885) and the Academy of Finland (128652).
Data accessibility
Host plant chemistry and insect community data are included in the supporting information (Tables S1 and S2, Supporting information). Host plant nucleotide sequences are available from the European Nucleotide Archive: www.ebi.ac.uk/ena/data/view/LN734767-LN734821.