Consequences for host–parasitoid interactions of grazing-dependent habitat heterogeneity


A. J. Vanbergen, Centre for Ecology and Hydrology Banchory, Hill of Brathens, Banchory AB31 4BW, UK. Fax: +44 1330826303; E-mail:


  • 1Environmental heterogeneity can produce effects that cascade up to higher trophic levels and affect species interactions. We hypothesized that grazing-dependent habitat heterogeneity and grazing-independent host plant heterogeneity would influence directly and indirectly a host–parasitoid interaction in a woodland habitat.
  • 2Thistles were planted randomly in 20 birch woodlands, half of which are grazed by cattle. The abundances of two species of seed herbivore and their shared parasitoid were measured, and related to habitat and host-plant heterogeneity.
  • 3The presence of cattle grazing created a structurally and compositionally distinct plant assemblage from the ungrazed seminatural situation. Grazing did not affect the number or dispersion of the host plant underpinning the host–parasitoid interaction.
  • 4The density of one insect herbivore, Tephritis conura, and its parasitoid Pteromalus elevatus was significantly increased by the presence of cattle; but another herbivore, Xyphosia miliaria, was unaffected. The percentage of parasitism of T. conura was increased in grazed habitat occurring at twice the rate found in ungrazed habitat.
  • 5The increase in T. conura abundance was correlated with increased species richness and cover of forbs in grazed sites. This effect of grazing-dependent habitat variation on host insect density cascaded up to parasitoid density and percentage of parasitism. Habitat heterogeneity had a further direct, positive effect on parasitoid density and percentage of parasitism after controlling for host-insect density.
  • 6Independent of grazing, heterogeneity in host-plant flowering, architecture and stature further affected T. conura and its parasitoid's densities. Parasitoid density was also affected by the dispersion of the host plant.
  • 7A combination of habitat and host-plant scale environmental heterogeneity influenced a host–parasitoid interaction indirectly and directly, providing a rare example of an anthropogenic disturbance positively affecting a tertiary trophic level. This finding highlights the need to consider not only the importance of bottom-up effects for top-down processes, but also the role of environmental heterogeneity arising from anthropogenic disturbance for trophic interactions such as parasitism.


Trophic cascades occur when one species (or group of species) indirectly affects a second species (or group of species) by influencing the abundance or biomass of a third, connected to the first two by trophic interactions (Rosenheim, Wilhoit & Armer 1993). The indirect effect of predators on plants mediated via impact on herbivore populations is one type of trophic cascade (Gomez & Zamora 1994; Moran, Rooney & Hurd 1996; Schmitz 2003). Typically, these studies focus on simple interaction chains (Matsumoto, Itioka & Nishida 2003) or webs (Finke & Denno 2004), and examine the effect of top predators on the density or behaviour of other predators (Rosenheim et al. 1993; Lang 2003) and herbivores (Krivan & Schmitz 2004). Fewer studies have investigated the extent to which such relatively simple food chains may be affected by perturbation of the wider community or habitat in which the interaction chain is embedded (Chase 1996; Jones, Lawton & Shachak 1997; Grabowski 2004; Preisser & Strong 2004).

Cattle can alter significantly plant communities and habitats (Hobbs 1996; Rambo & Faeth 1999). Cattle grazing can suppress competitive dominants and facilitate the emergence of a distinct, more diverse plant community (Rambo & Faeth 1999; Fowler 2002; Pykala 2003) and affect higher trophic level diversity and abundance (Gibson et al. 1992; Kruess & Tscharntke 2002; Woodcock et al. 2005). Studies examining the effect of cattle grazing on species interactions are less common (Kruess & Tscharntke 2002; Vazquez & Simberloff 2003, 2004). Cattle grazing in Argentine forests, for example, disrupted pollinator–plant mutualism webs via the loss of a number of ‘keystone interactions’ (Vazquez & Simberloff 2003), and for one plant species cattle-driven changes to plant population density affected plant reproduction by reducing pollen deposition by insects (Vazquez & Simberloff 2004).

Parasitism has the potential to be affected by the cascading effects of cattle ecosystem engineering (Jones et al. 1997; Kruess & Tscharntke 2002), as environmental heterogeneity at both coarse (e.g. habitat) and fine scales (e.g. host plant) can affect parasitism. Forest fragmentation is known to influence parasitism rates (Roland & Taylor 1997), and isolated ‘habitat fragments’ tend to support reduced parasitoid diversity and percentage of parasitism compared with contiguous habitat (Kruess & Tscharntke 1994). Host-plant patch size (Sheehan & Shelton 1989; Doak 2000), patch isolation (Doak 2000), plant species identity (Roland 1986; Englishloeb, Brody & Karban 1993) and plant architecture (Gingras & Boivin 2002; Gingras, Dutilleul & Boivin 2003) affect both the pattern and level of parasitism. In summary, the effect of spatial and structural heterogeneity at lower trophic levels, and at different scales, can affect parasitoid density and parasitism rates.

In this present study we assess the impact of grazing by cattle in birch woodlands, and the environmental heterogeneity it generates, on a simple thistle-seed herbivore–parasitoid interaction chain. Independent of grazing effects we also evaluate the influence of host-plant heterogeneity (e.g. architecture) to this same interaction chain. The only previous study, to our knowledge, to consider the effects of grazing for host–parasitoid interactions was carried out in anthropogenic grassland (Kruess & Tscharntke 2002). The present study considers the importance to a host–parasitoid interaction of habitat heterogeneity arising from anthropogenic disturbance of seminatural woodland.

We hypothesized, first, that cattle act as ecosystem engineers in birch woodland fragments with indirect, cascading and direct effects on the host–parasitoid interaction. Indirect, cascading effects on host insect and parasitoid populations would arise as a result of cattle grazing and trampling reducing plant competitive exclusion, creating niche space and thus increasing the density and evenness in the thistle distribution. We predicted that: (i) the presence of cattle would increase the density, and create a more even dispersion, of thistles compared to ungrazed woodlands; (ii) parasitoid density would be positively correlated to host-insect density at the individual host-plant and habitat scale; and (iii) an increase in host-insect and parasitoid densities, and the amount of parasitism, will result from the greater density and more even dispersion of host-plants in grazed woods. We also predicted that cattle would modify the woodland habitat producing a more diverse, compositionally and structurally different plant assemblage compared to the ungrazed woodlands. This grazing-dependent habitat heterogeneity would directly affect the abundance of the two species of seed herbivores, their shared parasitoid and the amount of parasitism.

Secondly, we hypothesized that grazing-independent host-plant heterogeneity (stature, architecture and proportion of open inflorescences) would affect the density of both host-insect and parasitoid populations with the prediction that taller, bushier plants with more flowers will support greater densities of herbivores and parasitoids, and increased parasitism.


tri-trophic system

The marsh thistle Cirsium palustre L. is a multistemmed biennial ranging from 50 to 120 cm in height with composite flowerheads clustered at the end of each stem. Cirsium palustre tends to have an aggregated dispersion where it occurs naturally (Williams, Jones & Hartley 2001). Flowering occurs from early June to mid-September and seed herbivores attack the flowerheads − the most common of which are tephritid flies (Diptera: Tephritidae). The adult tephritid fly inserts its eggs among the florets of recently opened flowerheads and the larva burrows into the flowerhead where it feeds on the receptacle and seeds (Janzon 1984; Jones, Godfray & Hassell 1996). Two tephritid species were considered. Tephritis conura Loew is univoltine with gregarious larvae (up to seven conspecific larvae per capitulum) feeding during June and July, and the adult emerging in August to overwinter (Janzon 1984; White 1988; Romstöck-Völkl 1990). Xyphosia miliaria Schrank, on the other hand, is generally solitary (occasionally two conspecific larvae per capitulum) and bivoltine: the first generation completes larval development and emerges as an adult between July and September, the second completes larval feeding by the autumn overwintering in the final instar to pupate in spring, and emerge as the adult fly between May and July (White 1988). In Britain, T. conura has also been recorded feeding on Cirsium heterophyllum L. Hill (White 1988), a species not seen in the sites studied here. The alternative host plant of X. miliaria, which occurred in some of the grazed study sites, is Cirsium arvense L. (White 1988). The tephritids share a parasitoid Pteromalus (syn: Habrocytus) elevatus (Walker) (Hymenoptera: Pteromalidae). This endoparasitoid probes the thistle capitula with its ovipositor and deposits an egg externally on the tephritid larvae. The peak in parasitoid oviposition occurs during early to mid-August (Jones et al. 1996). The parasitoid overwinters within the larval host (yielded from X. miliaria and late-season T. conura) or possibly as an adult (when yielded from early-season T. conura) (A. J. Vanbergen personal observation).

sites and experimental design

A total of 20 birch (Betula pendula/pubescens) woodland sites were selected in the Deeside region of Aberdeenshire (57°3 0′N, 2°30 2′E−57°3 3′N, 2°57 9′E) according to whether they supported existing populations of C. palustre and to the presence (n = 10) or absence (n = 10) of livestock grazing. Care was taken to ensure that sites selected were not geographically confounded (Fig. 1). Livestock in the grazed sites were predominantly cattle (n = 10), with occasional and additional grazing by sheep (n = 3) and horses (n = 1), and this management had been in place for at least the preceding 10 years. The low incidence of grazing by wild ungulates (i.e. roe deer) was not measured because it was assumed to be of minimal importance when compared with the intensive livestock management. No other systematic management (e.g. logging) occurred at either grazed or ungrazed sites. At the end of April (coincident with the date of birch budburst) eight second-year C. palustre thistles were transplanted, from two nearby sites, into each of the woodland locations. Within each site replicate plants were assigned randomly to a position within a 4 × 4 grid (15 m2) with each point on that grid separated by 5 m. The centre of the grid was situated at least 50 m from the woodland edge and, at most, 50 m from the pre-existing population of thistles. Prior to flowering, cattle trampling destroyed some replicate thistles (18%) and neighbouring thistles of local origin acted as replacements. No evidence of vertebrate grazing was seen on experimental thistles during the course of the experiment.

Figure 1.

Schematic map showing the location of the study area in Scotland and the name, distribution (West–East) and area (ha) of grazed and ungrazed birch woodlands used in the study.

insect abundance

Colonization of the thistles by both the tephritid herbivores (X. miliaria and T. conura) and their shared parasitoid (P. elevatus) was assessed by bagging and excising the thistle stem (14–15 August 2003). Twelve seedheads were selected randomly from each bagged thistle (160), a total of 2112 flowerheads, and dissected for tephritid prepupae or pupae. These were removed and placed in labelled Eppendorf™ tubes to rear-on the adult tephritid or parasitoid within the host. Identification of X. miliaria and T. conura was based on White (1988) and P. elevatus on (Hoebeke & Wheeler 1996).

habitat variables

Six 1-m2 quadrats per site were placed randomly on points on the aforementioned grid not occupied by experimental thistle replicates, and an assessment was made in each quadrat of a range of field layer variables (28 July−12 August 2003). The species richness of functional groups (forbs, grasses) and total vascular plant richness was determined, and the percentage cover for each vascular plant species and functional group (forbs, grasses, bryophyte) was estimated visually. The proportion of bare, disturbed soil surface arising from cattle trampling was also recorded in each quadrat. The percentage cover of the tree canopy above each quadrat was estimated using a canopy densiometer (Forestry Suppliers Inc., Jackson, Mississippi, USA) and at the four corners of each quadrat the maximum sward height (mm) was recorded. The area of each birch woodland fragment was determined from digitized maps (Ordnance Survey, Southampton, UK) using a geographic information system (GIS) (ArcGIS) (Fig. 1).

host-plant variables

Towards the end of the adult tephritid activity period (20–25 July 2003) a number of host-plant level parameters were measured to quantify the effect of heterogeneity between individual plants on insect abundance and parasitism rates. The structure of each thistle replicate was quantified by measuring: the height (cm) of the stem (or tallest stem in case of multistemmed individuals), the architectural complexity (number of stems plus branches) and the proportion of open inflorescences. The degree to which the density and distribution of the pre-existing C. palustre populations in each birch wood affecting colonization of experimental thistles was assessed. C. palustre density was mapped in a 100-m2 quadrat centred on the experimental grid. The quadrat was subdivided into 5-m2 cells, giving a total of 400 cells per 100 m2, and the number of pre-existing thistles in each cell was recorded. The density and aggregation (s/x̄) (Hassell et al. 1991) of C. palustre host plants was derived at two spatial scales (100 m2, 25 m2) around the experimental thistles, set by what is known about tephritid and pteromalid maximum dispersal distance (Jones et al. 1996), and these data were used in subsequent tests.

statistical analysis

Effect of grazing on the habitat and plant assemblage

Whether mean habitat variables between the grazed and ungrazed sites were significantly different was determined with Wilcoxon's signed-rank tests. Plant assemblage structure was assessed using redundancy analysis (RDA; canoco version 4·5); a constrained form of principal components analysis (PCA) that identifies trends in the scatter of species data that are related linearly to a set of constraining, explanatory variables (ter Braak & Šmilauer 1988; Jongman, ter Braak & van Tongren 1995). Vascular plant species with < 10% cover (constituting only 1% of the total vegetation) when summed across all sampled sites were removed from the data set to avoid the RDA being biased by species that occurred only in a limited number of sites; data were log-transformed prior to analysis (ter Braak & Šmilauer 1988; Jongman et al. 1995). RDA was used to relate the percentage cover of vascular plant species to measured explanatory variables (canopy density, sward height, proportion of bare and trampled soil, bryophyte and leaf-litter cover). Variables derived from the plant species data (forb and grass percentage cover, vascular plant species richness) were fitted as supplementary variables only in order to illustrate further trends in the plant assemblage and do not affect the variance explained by the RDA of the vascular plant assemblage. The significance of the explanatory variables in structuring the plant assemblage was determined with a forward selection procedure using Monte-Carlo tests constrained within sites (9999 permutations). Thus, the model presented is a partial redundancy analysis (pRDA) controlling for site-level variance when assessing the impact of grazing-dependent habitat variation on the plant assemblage.

Effect of grazing on thistles, herbivore and parasitoid populations

Data were analysed using generalized linear mixed models (GLMM) (sas version 8·01, SAS Institute) at two scales: (i) the site (n = 20) to test for treatment (grazed or ungrazed) and cattle-dependent habitat effects and (ii) individual plants (n = 160) nested within site to test for host-plant effects. Dependent variables were host-plant (C. palustre) density and aggregation (s/x̄), counts per plant and mean counts per site of tephritid (X. miliaria and T. conura) and parasitoid (P. elevatus) populations. Parasitism was modelled with the count of P. elevatus as the dependent variable offset against the log-transformed count of host-insect pupae. Poisson error distributions were fitted to all count data except mean parasitoid counts, as these data had residuals that were normally distributed. Solution of fixed explanatory (treatment, habitat and host-plant variables, host insect density) and random (categories: ‘site’ and ‘thistle nested within site’) effects was estimated by residual maximum likelihood (REML) (Schall 1991; Elston et al. 2001). Denominator degrees of freedom were estimated using Satterthwaite's approximation (Littell et al. 1996). Details of the models run to test our hypotheses are summarized in Table 1. Model simplification was by stepwise elimination of the least significant term until the most parsimonious model was found and F-ratios of fixed effects using adjusted sums of squares (Type 3 tests) are reported.

Table 1.  Structure of GLMM models run to test the effect of treatment (cattle presence or absence), grazing-dependent heterogeneity, host-plant heterogeneity, host-insect density on host-plant, tephritid seed herbivore, parasitoid abundance and the amount of parasitism in birch woodlands
Effect of grazing on host plantHost-plant densityHabitat (25 and 100 m2)Treatment
Host-plant aggregationHabitat (25 and 100 m2)Treatment
Effect of host-plant distribution on insectsHost-insect countHabitat (25 and 100 m2)Treatment, host-plant density or aggregation
Parasitoid countHabitat (25 and 100 m2)Treatment, host-plant density or aggregation
ParasitismHabitat (25 and 100 m2)Treatment, host-plant density or aggregation
Effect of cattle presence on insect densityHost-insect countHabitatTreatment
Parasitoid countHabitatHost-insect count, treatment
Host–parasitoid correlationMean parasitoid countHabitatMean host-insect count, treatment × mean host-insect count
Parasitoid countHost plantHost-insect count
Effect of cattle-dependent habitat heterogeneityHost-insect countHabitatPCA1, PCA2
Parasitoid countHabitatHost-insect count PCA1, PCA2
ParasitismHabitatPCA1, PCA2
Effect of host-plant heterogeneityHost-insect countHost plantPCA1, PCA2
Parasitoid countHost-plantHost-insect count PCA1, PCA2
ParasitismHost plantPCA1, PCA2

To assess the effect of cattle-dependent habitat heterogeneity on the abundance of tephritids and parasitoid and percentage parasitism, the term ‘treatment’ was replaced with specific habitat variation (e.g. plant species richness) between grazed and ungrazed sites. Owing to intercorrelation between habitat variables (Table 2), the separation of their effects by stepwise multivariate regression was confounded. We therefore performed PCA on these habitat data. The orthogonal PCA axes derived (which are uncorrelated with each other, but principally correlated with specific habitat variables) were then used as the fixed effects in the models. This was conducted to separate the effect of these intercorrelated habitat variables on the insect abundance and reduce the number of explanatory variables in an ecologically meaningful manner. Similarly, variables describing heterogeneity between host plants (e.g. architecture, height) were intercorrelated (Table 2) and we treated these in the same way.

Table 2.  Inter-correlation between birch habitat and host-plant variables. Values are Pearson's correlation coefficients (SAS version 8·01) *P < 0·05, **P < 0·01 and ***P < 0·001. Habitat variables are the percentage cover of bare soil, forbs, grasses and bryophytes, tree canopy density (%), height of the sward (mm) and the number of vascular plant species (Plant S), forb species (Forb S) and grass species (Grass S). Host-plant variables are: architecture − the total number of stems and branches per plant, height − the height (cm) of the tallest stem of each plant and inflorescence – the proportion of flowers open by the end of the adult tephritid activity period
 SwardPlant SForb SGrass SBare soilForbGrassBryophyte
 Sward −0·36***−0·54***0·16−0·48***−0·49***0·31***−0·10
 Plant S  0·86***0·56***0·47***0·43***−0·20**−0·33***
 Forb S   0·27**0·52***0·69***−0·30***−0·39***
 Grass S    0·27**0·130·42***−0·52***
 Bare soil     0·47***−0·09−0·38***
 Forb      −0·16*−0·44***
 Grass       −0·57***


effect of cattle on the plant assemblage and host plant

The presence of cattle in birch woodlands altered the plant assemblage in this habitat. The grazed sites had a greater species richness of vascular plants, and forbs in particular, compared to ungrazed locations (Table 3). The percentage cover of bryophytes (principally Hylocomium splendens (Hedw.) Br. Eur. − an undercanopy species) was significantly decreased in the grazed locations with a concomitant increase in the overall percentage cover of forbs (particularly genera such as Ranunculus L. and Trifolium L.) (Fig. 2, Table 3). In addition, grazed sites had a less dense tree canopy, a greater proportion of bare earth on the soil surface (a consequence of trampling by cattle) and a lower mean sward height compared to the ungrazed situation (Fig. 2, Table 3).

Table 3.  Wilcoxon signed-ranks comparison of mean habitat variables from grazed (n = 10) and ungrazed (n = 10) sites
Plant species richness 26·3 16·8  6·3  5·12·880·002
Grass species richness  7·4  6·0  1·8  1·91·340·09
Forb species richness 15·5  6·4  4·1  3·93·410·0003
Bare ground (%) 18·0  1·2  0·1  0·02·820·002
Sward height (mm)347·1503·8161·7176·1−1·850·03
Grasses (%) 76·9 76·0 14·1 30·7−0·570·285
Forbs (%) 42·6 13·7 20·1  8·63·210·0007
Bryophytes (%)  7·3 26·0  7·3  6·6−2·830·002
Canopy density (%) 63·0 87·4  0·3  0·1−1·700·04
Figure 2.

Partial redundancy analysis (pRDA) of plant assemblage composition between 10 grazed and 10 ungrazed birch woodlands. Significant environmental variables structuring the plant assemblage are represented by solid line vectors: tree canopy density (Canopy), plant litter (Litter) and bryophyte (Bryo) percentage cover, mean height (mm) of ground vegetation (Sward), and the percentage cover of bare, disturbed soil (Bare). Supplementary variables are derived from the plant assemblages and are shown by dashed line vectors: total vascular plant species richness (Plant S), percentage cover of plant functional groups (Forb and Grass). The species scores of the 10 most abundant (percentage cover) plant species are shown (in rank order: Holcus lanatus (, Agrostis capillaris (, Pteridium aquilinum (, Holcus mollis (, Poa trivialis (, Agrostis stolonifera (, Molinia caerulea (, Deschampsia flexuosa (De.fl.), Ranunculus repens (, Trifolium repens (Tr. Re.). Open and closed symbols denote the mean sample scores per site ± SE. Significance of environmental variables was determined using a forward selection procedure with Monte-Carlo permutation (9999) tests.

Plant assemblage structure was affected by the presence of grazing livestock; pRDA revealed a clear separation between plant assemblages depending on whether or not they were grazed (Fig. 2). The axes of the pRDA were significant (Monte-Carlo global permutation tests: first canonical axis: P = 0·008, all canonical axes: P = 0·0006) and it was the first axis of the pRDA that explained most of the variation (eigenvalue = 0·08) in the plant species assemblage (Fig. 2). This primary axis of variation was positively correlated with increasing bryophyte percentage cover (correlation coefficient = 0·74), tree canopy density (0·56), plant litter cover (0·56) and average sward height (0·50), and negatively correlated with the amount of disturbed and trampled soil (−0·56). These fitted environmental variables explained 14% of the total variance in the plant species percentage-cover data. Plant assemblages in grazed sites were positively related to greater amounts of soil disturbance (Monte-Carlo P = 0·04, Fig. 2) and negatively associated with lower levels of bryophyte cover (P = 0·004, Fig. 2), tree density (P = 0·003, Fig. 2), sward height (P = 0·04, Fig. 2) and reduced amounts of plant litter on the soil surface (P = 0·02, Fig. 2). Grazed plant assemblages were characterized by high species diversity and forb percentage cover (Fig. 2). There was no statistical support for the prediction that the density or dispersion of thistles was affected by the presence of cattle at any of the spatial scales examined (Table 4).

Table 4.  The effect of treatment (grazed or ungrazed woodland) on the density and aggregation of C. palustre host plants at two spatial scales (100 m2, 25 m2) centred on the experimental grid and effect of host-plant density and aggregation on the abundance of two seed herbivores (X. miliaria, T. conura) their parasitoid P. elevatus and the proportion of parasitized T. conura. F-ratios and P-values are from GLMM with Poisson errors and log link, †P = 0·05, **P < 0·01
ThistleGrazedUngrazedTreatmentF (d.f.)X. miliaria F (d.f.)T. conura F (d.f.)P. elevatus F (d.f.)Parasitism F (d.f.)
Mean (SD)
25 m2
 Density 4·90 (5·28) 3·60 (5·99)0·01 (1,14)3·94 (1,18)0·25 (1,14) 0·00 (1,13)0·87 (1,10)
 Aggregation 1·28 (1·16) 2·19 (1·71)0·55 (1,13)1·25 (1,11)2·47 (1,15)10·82 (1,18)**9·98 (1,18)**
100 m2
 Density58·0 (34·8)73·8 (85·6)0·16 (1,17)3·89 (1,11)0·02 (1,15) 0·00 (1,15)0·92 (1,15)
 Aggregation 3·82 (2·23) 5·03 (3·16)0·55 (1,16)2·80 (1,17)1·57 (1,17) 4·50 (1,13)3·48 (1,16)

effect of cattle presence on tephritid and parasitoid populations

The two tephritid species dominated the endophagous capitulum fauna; only rarely were microlepidopteran and hymenopteran (Palloptera spp.) larvae found in samples. The response of the tephritid herbivores to the presence of cattle in birch woodlands contrasted strongly: Xyphosia miliaria was not affected by the presence of grazing livestock (Table 5, Fig. 3a), while Tephritis conura increased in number in grazed compared to ungrazed birch woodlands (Table 5, Fig. 3a). Parasitoid abundance was also significantly affected by cattle presence with greater numbers found in grazed woodlands compared with the ungrazed situation (Table 5, Fig. 3a). Furthermore, the percentage of T. conura pupae parasitized was significantly higher in grazed woodlands compared with ungrazed woodlands (Fig. 3b, F1,18 = 9·69, P = 0·006), suggesting this increase in parasitism could not be explained solely by increasing host-insect abundance.

Table 5.  The effect of grazing, cattle-driven environmental and host-plant heterogenity on the abundance (counts) of two tephritid herbivores Xyphosia miliaria, Tephritis conura and their parasitoid Pteromalus elevatus. Summary of GLMM models (SAS version 8·02) with Poisson error structure and log link function, F-ratios of significant fixed effects adjusted to account for other significant variables, non-significant terms were eliminated in stepwise manner. Due to intercorrelation between habitat and host-plant variables PCA axes scores of covariates were obtained and used as explanatory terms in the model. Principal components correlated with the orthogonal axes scores are given in parentheses (+ vascSR − plant species richness, + forbSR − forb species richness, + forb − % cover of forbs, + Arch − host-plant architecture (number of branches and stems per thistle), + flower − % of open inflorescences during the adult flight period, + height − height (cm) of the thistle flower stalk)
Fixed effectX. miliariaT. conuraP. elevatus
Treatment (grazed or ungrazed)−0·89−0·211,140·110·74−1·641·611,17 7·750·012−2·981·971,1812·370·003
 T. conura density          −2·720·241,3749·15< 0·0001
 PCA1 (+ vascSR, + forbSR, + forb)−1·030·011,130·000·97−0·850·451,17 9·180·007 0·381,15 6·600·02
 PCA2 (+ grassSR, + grass, + sward) 0·231,160·920·35 0·161,24 0·490·49 0·011,19 0·000·97
Host plant
 T. conura density          −2·890·211,3139·19< 0·0001
 PCA1 (+ Arch, + height)−0·96−0·161,1312·180·14−0·880·441,144 6·680·01 0·791,13810·590·001
 PCA2 (+ flower) 0·021,1410·010·92 0·801,13810·250·002 0·841,80 5·300·024
Figure 3.

The effect of the presence of cattle grazing on (a) the abundance of two tephritid herbivores (Tephritis conura F1,17 = 7·75 P = 0·012 and Xyphosia miliaria F1,14 = 0·11, NS) and their shared parasitoid (Pteromalus elevatus F1,21 = 8·55, P = 0·008) and (b) the proportion of parasitized T. conura (F1,20 = 8·98, P = 0·007). Mean values ± SE derived from 10 grazed and 10 ungrazed birch woodland sites.

tephritid–parasitoid interactions

There was a strong, positive association between T. conura and parasitoid number at the scale of host plant (Fig. 4a, F1,118 = 82·73, P < 0·0001) and woodland site (Fig. 4b, F1,18 = 69·87, P < 0·0001). Parasitoid abundance was significantly affected by the interaction between host-insect number (T. conura) and cattle presence or absence at the woodland site scale (T. conura × treatment: F1,18 = 9·91 P = 0·006), with parasitism of T. conura at grazed sites occurring at more than twice the rate found in ungrazed sites (Fig. 4b). Parasitoid abundance was not correlated with X. miliaria abundance either at the host-plant (F1,158 = 0·82, NS) or woodland site (F1,18 = 0·03, NS) scales. There was no correlation between the abundance of the tephritid herbivores at the host-plant scale (F1,158 = 1·83, NS) or at the scale of the woodland site (F1,18 = 0·49, NS).

Figure 4.

Relationship between the abundance of the parasitoid P. elevatus and its host T. conura at the scale of (a) the individual host plant (F1,118 = 82·73, P < 0·0001) and (b) the woodland site (T. conura F1,18 = 69·87, P < 0·0001, T. conura × treatment F1,18 = 9·91 P = 0·006). Fitted lines from GLM with normally distributed error distribution.

cattle habitat engineering effect on tephritid populations

The compositional changes to the plant assemblage driven by cattle grazing directly affected the T. conura population. When T. conura abundance was modelled against the orthogonal PCA axes of the habitat variables it was the first ordination axis − positively correlated with forb species richness (eigenvector = 0·46), plant species richness (eigenvector = 0·45) and forb cover (eigenvector = 0·38) − that was significantly and positively correlated with the numbers of this herbivore (Table 5, Fig. 5). The second PCA axis − correlated with the percentage of cover (eigenvector = 0·56) and species richness of grasses (eigenvector = 0·43) and average sward height (eigenvector = 0·43) − did not correlate with T. conura density. X. miliaria was not significantly affected by either of the PCA axes (Table 5). The woodland area had no significant effect on the abundance of either herbivore (T. conura F1,17 = 0·26, NS, X. miliaria F1,12 = 3·50, NS).

Figure 5.

Relationship of T. conura and P. elevatus abundance to the first principal component of grazing-dependent habitat heterogeneity (correlated with plant (vascSR) and forb (forbSR) species richness and forb percentage cover (forb%)). T. conura F1,17 = 9·18, P = 0·007; P. elevatus F1,15 = 6·60, P = 0·02. Fitted line derived from GLMM with Poisson error structure and log link function.

cattle habitat engineering effect on the parasitoid population

Grazing-dependent changes to the composition of the plant assemblage both directly and indirectly (via changes in host-insect density) affected the highest trophic level. The abundance of the parasitoid, P. elevatus, was positively and significantly correlated with the first PCA axis (positively correlated with vascular plant species richness, forb species richness and percentage of cover) (Fig. 5). This is explained partially by the observed increase in floral richness cascading up to the parasitoid population via changes to host-insect density (Table 5, Fig. 4b). However, after accounting for host density, there remained a direct and positive effect of this grazing-dependent habitat variation on parasitoid abundance (Table 5, Fig. 5). Furthermore, the percentage parasitism (proportion of T. conura pupae parasitized) in these woodlands was also positively related (F1,17 = 6·85, P = 0·02) to this grazing-dependent habitat variation represented by the first axis of the PCA. The woodland area had no significant effect on parasitoid abundance (F1,17 = 0·09, NS).

host-plant level effects on tephritid and parasitoid populations

Plant structural variables, namely the height of the stem, architecture (number of branches and stalks per thistle) and the proportion of open inflorescences per thistle, were significantly intercorrelated (Table 2). None were significantly affected by the presence of cattle (architecture F1,18 = 0·14 NS, height F1,18 = 0·46 NS, inflorescences F1,18 = 0·17 NS). Both the first (correlated with plant architecture − eigenvector = 0·61, and stem height − eigenvector = 0·62) and second (correlated to the proportion of open inflorescences − eigenvector = 0·86) PCA axes were significantly and positively related of the abundance of T. conura and P. elevatus (Table 5). Taller, bushier thistles with a larger proportion of open flowers supported greater numbers of T. conura and, after controlling for variance due to host-tephritid density, the parasitoid (Table 5). Percentage parasitism was positively correlated with the first (F1,67 = 12·08, P = 0·0009) and the second PCA axes (F1,49 = 5·72, P = 0·02). There was no significant effect of host-plant structure on X. miliaria abundance (Table 5). Furthermore, independent of grazing, neither tephritid herbivore was affected by the density or degree of aggregation of their host plants at any spatial scale measured (Table 4). Parasitoid density, however, showed a highly significant negative relationship with the aggregation of thistles. Parasitoids occurred in greater numbers where the thistle distribution was more clumped at 25 m2 and marginally at the 100 m2 scale (Table 4). The level of parasitism also increased where thistles had a more clumped distribution at the smallest spatial scale measured (Table 4).


cascading and direct effects of cattle on a tri-trophic system

There was no effect of cattle engineering on the density or dispersion of thistles, and therefore no grazing-driven cascade via the host plant to the abundance of the higher trophic levels or the amount of parasitism in the tri-trophic system. The presence of cattle did, however, have consequences for one of the two host–parasitoid interactions studied. An increase was seen in the abundance of the herbivore, T. conura, and its parasitoid in the grazed compared to the ungrazed situation. No change was observed in the abundance of the second herbivore X. miliaria; the reason for the lack of a response from this herbivore remains unclear. This provides partial support for our prediction that the effects of cattle grazing would influence the primary consumers. While cattle grazing did not initiate a bottom-up cascade from host-plant to herbivores, grazing did lead to an increase in T. conura abundance that cascaded up to the tertiary trophic level of the interaction chain. The number of T. conura was highly correlated with the increase in parasitoid abundance in grazed woodland and the presence of cattle doubled the parasitism rate, contributing to the higher levels of parasitism in grazed sites.

Despite not affecting the tephritids’ host-plant cattle did act as ecosystem engineers (Jones et al. 1997) facilitating (via grazing and trampling) compositional (e.g. increased plant species richness) and structural (e.g. decrease in tree density) changes to the wider plant assemblage (Hobbs 1996). The grazed woodlands had a more diverse field layer and supported a plant assemblage compositionally different to that in ungrazed woods. In ungrazed sites the plant assemblage was dominated by competitively superior plant species (e.g. Molinia caerula L. (Moench), Pteridium aquilinum L. Kuhn, Holcus L. spp.). Grazing reduced the cover of these competitive dominants (Grant et al. 1996; Humphrey & Swaine 1997; Pakeman 2004) and facilitated a structurally simpler but compositionally diverse sward including grazing tolerant species (e.g. Ranunculus repens L., Trifolium repens L., Poa trivialis L.) more typical of open grassland.

Changes in plant assemblage structure driven by cattle appear responsible for the observed changes in abundance of the higher trophic levels of the interaction chain. The abundance of T. conura, P. elevatus and percentage of parasitism were positively correlated with the first principal component of the habitat variables (increased plant and forb species richness, and forb cover). This suggests that the greater floral diversity of the grazed sward may provide nectar resources throughout the flight period benefiting the adult insect; for example, improved fecundity, longevity and survival (Romstöck-Völkl 1990; Jervis 1998; Heimpel & Jervis 2005). Cattle therefore affected the host insect in an indirect manner, not mediated by the host plant but by wider plant diversity, and this effect cascaded up to the highest trophic level, the parasitoid population. In addition to this trophic cascade there was evidence of a further, direct effect of habitat engineering by cattle on the parasitoid populations: after controlling for host-insect density there remained a positive correlation between the increased floral richness in grazed woods, parasitoid number and percent parasitism. In contrast to studies that have demonstrated specific host-plant effects on parasitoids (Englishloeb et al. 1993; Van Nouhuys & Hanski 1999), this study revealed that parasitism rates occurring within a host plant were dependent in part on the wider plant community diversity. This is analogous to studies that have found at coarser spatial scales the amount of parasitism occurring in a habitat patch is influenced by the composition or heterogeneity of the surrounding landscape (Kruess 2003; Thies, Steffan-Dewenter & Tscharntke 2003). Parasitism is therefore sensitive to environmental heterogeneity from sources beyond the immediate host or habitat patch highlighting the potential for trophic interactions to be altered by anthropogenic disturbance.

The role of grazing in influencing host–parasitoid interactions had been considered previously in only one other study (Kruess & Tscharntke 2002), which demonstrated a decline in parasitoid number and parasitism rates associated with intense grazing. In contrast, we showed a positive consequence of grazing (and the environmental heterogeneity produced) for parasitoid abundance and parasitism rates. This distinction may have arisen as a result of differences between the study systems. Kruess & Tscharntke (2002) considered how intense grazing in an anthropogenic grassland habitat reduced the height of the vegetation and led to a decline in the numbers of insect hosts and their parasitoids. In this case the introduction of cattle to birch woodlands is a perturbation of a seminatural habitat, leading to a shift in the plant assemblage structure towards one more typical of open grassland. A consequence of this habitat modification was an increase in floral resources with unforeseen, positive consequences (direct and indirect) for parasitoid density and parasitism rates.

direct and indirect host-plant effects on higher trophic levels

Given the intimate relation (reproduction and larval feeding) between both these seed herbivores and their host plant, heterogeneity in host-plant attractiveness or larval resources would be expected to affect strongly on the probability of colonization by the herbivores (Romstöck-Völkl 1990; Williams et al. 2001). T. conura density was determined by the number of inflorescences and overall plant size, either because the larger individual plant with many inflorescences is more apparent in the habitat (Prokopy 1968; Aluja & Prokopy 1993) or because it offers more resources in terms of mating, oviposition and larval growth and survival (Romstöck-Völkl 1990; Williams et al. 2001). Parasitoid population density and parasitism rate were also greater on larger plants with many inflorescences, explained partly by the correlation between host (T. conura) density and these host-plant parameters, but also by a direct influence of this plant level heterogeneity on the parasitoid. The failure to detect any relationship between X. miliaria and the various host-plant structural variables is unexpected, particularly given the low occurrence in grazed and absence in ungrazed sites of the alternative congeneric host (C. arvense). One possibility is that the variables measured here are not those employed by X. miliaria for locating and colonizing hosts.

In direct contrast to previous studies (Jones et al. 1996; Williams et al. 2001), there was no effect of the number or aggregation of the pre-existing thistle population on the abundance of either tephritid. It may be that the lack of a response to host-plant density or aggregation reflects the overall commonness of the plants in both grazed and ungrazed birch woods. There may be little selective pressure for these insects, despite gradients in thistle distribution within and between sites, to distribute themselves non-randomly as the source of their food is abundant overall (Jakobsen & Jjohnsen 1987). In contrast, parasitoid density and percentage parasitism of T. conura increased in areas where thistles were aggregated. It appears likely, therefore, that the parasitoid uses the thistles as an environmental signal to initiate searching for the host insect and may find it simpler to locate host-plant patches rather than the more cryptic host insect (Cappuccino 1992). The difference between the response of the herbivore T. conura and the parasitoid to thistle aggregation may be explained as one strategy by the herbivore to avoid its enemy: in ovipositing randomly the herbivore may spread the risk of parasitism by decoupling the larval distribution from the host-plant distribution in a predictable manner (Cappuccino 1992; Williams et al. 2001).


Grazing-dependent habitat heterogeneity and grazing-independent host-plant heterogeneity influence the studied host–parasitoid interactions in both a direct and indirect cascading manner. Grazing cattle in forests can erode the strength and extent of mutualistic (Vazquez & Simberloff 2003) and antagonistic insect interactions (Kruess & Tscharntke 2002); however, in this study grazing cattle more than doubled the rate of parasitism compared with the ungrazed scenario. This provides a rare example of a tertiary trophic level being positively rather than negatively affected by anthropogenic disturbance (Didham et al. 1998; Davies, Margules & Lawrence 2000; Kruess & Tscharntke 2002). This was not a conventional bottom-up cascade because grazing did not alter the density, dispersion or architecture of the host plants underpinning this host–parasitoid interaction. Instead, it was grazing-driven changes to the diversity of the wider plant assemblage, cascading up via the insect host and acting directly on the parasitoid populations, which led to the observed increase in parasitism. Host-plant heterogeneity also affected parasitoid density and percentage parasitism both indirectly (via host-insect density) and directly. Along with scale (habitat and host-plant) and the relative role of vertical effects (bottom-up vs. top-down) the importance of wider environmental heterogeneity, and not only those features intrinsic to a single food chain, must be taken into consideration when determining how anthropogenic environmental perturbation affects trophic interactions (Hunter & Price 1992; Jones et al. 1997).


The authors thank D. Elston of Biomathematics anf Statistics Scotland (BioSS) for advice on statistical design and analysis, P. Lambdon for advice and assistance in identifying botanical specimens, J. Deeming for confirmation of identification of P. elevatus and D. Edwards, C. Beaudoin and J. Young for general assistance. Thanks to D. Vázquez for comments on an earlier version and to two anonymous referees. A NERC CEH Science budget supported this work as part of AJV's PhD training.