Root hemiparasites and plant diversity in experimental grassland communities

Authors


Jasmin Joshi (fax +41 1 635 40 67; e-mail joshi@uwinst.unizh.ch).
*Present address: Fachbereich Biologie, Philipps-Universität Marburg, D-35032 Marburg, Germany.

Summary

1  We studied the relationship between the diversity of grassland communities and the effects of the generalist hemiparasitic plant Rhinanthus alectorolophus. We compared resistance against biomass loss as a consequence of infection, performance of the parasite and resistance of the parasitized communities to invasion by other plant species. Seeds of the parasite were sown into experimental plots containing 1, 2, 4, 8 or 32 plant species belonging to one or more of three functional groups (grasses, legumes and non-leguminous herbs).

2  We predicted that infection will reduce host biomass, total community biomass and resistance to invasion, particularly in host communities with low diversity, but that the performance of the parasite will be at its lower level in such communities.

3  The presence of the parasite caused an overall reduction in host biomass per plot, which was mainly due to a strong reduction in the biomass of grasses. As predicted, the effect was smaller in communities with greater functional diversity. However, total community biomass (including the parasite biomass) was increased by more than a third in infected communities of one or two host species, while the parasite had no effect on total biomass of species-rich communities.

4  Germination of the parasite was hardly influenced by the diversity of its host community, but early survival decreased with increasing number of functional groups and was lower in plots with legumes than without. However, our hypothesis that the performance of the surviving parasites would benefit from a high functional diversity of hosts was supported. Parasite biomass per individual and per m2 increased with the number of functional groups in the host community, as did reproductive potential.

5  Death of the parasite led to a higher proportion of bare ground in communities when the previously infected communities had low functional diversity, thus enabling subsequent colonization by weeds.

Introduction

There is increasing evidence from experimental studies that plant diversity can influence ecosystem processes and that increased diversity can enhance ecosystem stability (Tilman & Downing 1994; Tilman 1996; McGrady-Steed et al. 1997; Naeem & Li 1997; see Peterson et al. 1998 and Schläpfer & Schmid 1999 for a review). It can therefore be expected that a disturbance such as invasion by a parasitic species may have greater effects in low- than in high-diversity communities. Thus, we studied the effects of (i) adding a root hemiparasite to experimental grassland systems of different diversity levels, ranging from monocultures to 32-species assemblages, and (ii) the response of the parasite itself to this plant-diversity gradient.

Parasitic plants with a wide host range represent a particularly well defined functional group in grassland communities (Press 1998). Root hemiparasites of the genera Rhinanthus, Castilleja and Triphysaria are common components of temperate grasslands (Ter Borg 1985; Marvier 1998a), and some species used to be important agricultural weeds (e.g. Fu¨rst 1931). Root hemiparasites are at least partially autotrophic with respect to carbon, but via special contact organs they extract water, nutrients and organic solutes from the roots of other plants (Press & Graves 1995). Hence, it can be argued that they belong not only to the first trophic level in grassland ecosystems, but also to the second trophic level, making them comparable in function to insect herbivores (Atsatt 1977).

There is some evidence that parasitic plants may fundamentally affect ecosystem processes. For example, it has been shown that they can severely reduce the growth of their hosts in experimental systems (Matthies 1995, 1997) and also influence the structure of natural plant communities (Pennings & Callaway 1996; Marvier 1998b). Root hemiparasites are, to some degree, selective and can also alter the competitive balance between host plants (Gibson & Watkinson 1991; Matthies 1996) and thus the composition of plant communities (Ter Borg & Bastiaans 1973; Gibson & Watkinson 1992; Davies et al. 1997), with the most commonly observed effect being a reduction in the proportion of grasses (Davies et al. 1997; Marvier 1998b). High plant diversity might be expected to buffer the effects of over-exploitation of individual host species or functional groups by a plant parasite as is seen in some cases for its effect on herbivore damage (MacArthur 1955; Root 1973). Such an increased resistance of host communities to a parasite attack as diversity increases might be expressed in a smaller loss of total community biomass because there is a greater chance in high- than in low-diversity systems that less sensitive host species will in effect compensate for reductions in biomass of the more sensitive species.

Although root hemiparasites are capable of using a wide range of species, their performance often varies significantly if they are grown with different individual hosts (e.g. Seel & Press 1993; Matthies 1998; Matthies & Egli 1999). We therefore predicted that parasite performance, as reflected in demographic traits such as growth and fecundity, should be higher in species-rich than in species-poor communities (despite the fact that the overall effects of the parasite should be less in species-rich communities). Two different possible mechanisms may cause such a pattern. First, the probability that a parasite will be able to find a single ‘best’ host should be higher in a high- than a low-diversity community. Second, in the field an individual parasite may attach to several different hosts simultaneously (Musselman & Mann 1977; Gibson & Watkinson 1989), enabling it to obtain different compounds from particular host species (Govier et al. 1967) or to extract the same compound from different regions of source space. It has been shown that a ‘mixed diet’ may indeed improve performance of both parasites (Marvier 1998a; but see Matthies 1996 and Marvier 1998b) and herbivores (Pennings et al. 1993). If parasite performance is considered an ecosystem process, the first explanation corresponds to the sampling hypothesis (Aarssen 1997) and the second to the complementarity hypothesis for diversity effects (Naeem et al. 1994; Tilman et al. 1997b; Loreau 1998).

The death of annual root hemiparasites will probably leave gaps in the vegetation and if gaps are subsequently colonized by weedy species, decreased invasion resistance and higher plant species diversity may be seen in the long term. In experimental studies, both a decrease and an increase of diversity due to parasitic plants have been found (Gibson & Watkinson 1992; Pennings & Callaway 1996).

We used the annual root hemiparasite Rhinanthus alectorolophus (Scop.) Pollich and experimental plots containing 1, 2, 4, 8 or 32 plant species belonging to the three functional groups grasses, legumes and non-leguminous herbs (Diemer et al. 1997; Spehn et al. 2000) to model the effects of parasite infection. Species (n = 48) were randomly selected from those which commonly occur with R. alectorolophus in semi-natural species-rich grassland communities in Switzerland. Although not all of these species may be good hosts for the parasite, we use the term ‘host community’ to distinguish this group from the total community, which also include the parasite and potential weeds. The number of functional groups in the host communities (1–3) was partly confounded with the number of species (see Diemer et al. 1997; Spehn et al. 2000). Control and parasitized quadrats within the experimental diversity plots allowed us to address three hypotheses: (i) the parasite will reduce both host and total community biomass and this effect will decrease with increasing numbers of species or functional groups present in the host community; whereas (ii) the establishment, survival and growth of the parasite will increase, and (iii) the parasite will reduce the resistance of communities against invasion by other species and the magnitude of this effect will depend on the diversity of the host community.

Materials and methods

The study species

The annual Rhinanthus alectorolophus (European yellowrattle; here after referred to as Rhinanthus) is a typical root hemiparasite in that it can complete its life cycle without a host although growth and seed production are greatly increased by attachment to a host (Matthies & Egli 1999). Field observations indicate that this parasite attacks a wide range of grasses and herbs (Weber 1976), and experiments with closely related species suggest that Rhinanthus can use many different species, including conspecifics, simultaneously as hosts (Gibson & Watkinson 1989; Matthies 1995, 1997; Prati et al. 1997), albeit with preferences for some species (Matthies & Egli 1999). Rhinanthus is widely distributed in Central Europe and before modern agricultural techniques were available it was one of the few annual weeds found in pastures and hayfields and was also a serious weed of cereal crops (Fürst 1931).

Experimental design

The study was carried out at the Swiss site of the European-wide BIODEPTH project (Biodiversity and Ecological Processes in Terrestrial Herbaceous Ecosystems) in which the impact of plant diversity on ecosystem processes is being investigated. The Swiss site is situated at Lupsingen (439 m a.s.l.) in the Jura Mountains near Basel. In April 1995, two replicate blocks, each consisting of 32 plots, were set up on a former arable field overlaying calcareous nutrient-rich soil (Diemer et al. 1997; Spehn et al. 2000). The plots measured 8 × 2 m and were separated by at least 1 m from each other. In May 1995, 32 different assemblages of species were created from a pool of 48 local grassland species belonging to 13 plant families and were sown into the plots at a density of 2000 viable seeds m−2 (equally divided between the number of species in each plant assemblage) to produce five different levels of diversity (for details see Appendix 1 in the Journal of Ecology archive on the World Wide Web; see a cover of a recent issue of the journal for the WWW address). Each plot within a block contained a different assemblage and to avoid confounding the effects of diversity level and species identity each diversity level was represented by a number of assemblages (10 monocultures, 7 plots with 2 species per block, 8 plots with 4 species per block, 5 plots with 8 species per block and 2 plots with 32 species per block). We used constrained random selection from the pool of 48 species (see also Diemer et al. 1997) to form the experimental plant assemblages, such that all polycultures contained at least one grass, the 2, 4 and 8 species levels included grass only, grass plus legume and grass plus other herb mixtures, respectively, and the 4, 8 and 32 species levels included assemblages of all three life forms (see Appendix 1).

The three functional groups of plants, based on similar morphological and physiological characteristics (grasses, legumes and non-leguminous herbs) provided an additional measure of diversity, assuming a greater niche overlap within than among functional groups (functional redundancy). Two of the three functional groups were also taxonomic groups (grasses = Poaceae, legumes = Fabaceae). The number of functional groups in a plot was used as a measure of resource-use diversity in the experimental communities, but was inevitably partly confounded with the number of species. Thus, plots with one functional group varied in species number from 1–8, plots with two functional groups from 2–8 and plots with three functional groups from 4–32.

From 1995–97 all plots were mown in June and September, following the traditional management of meadows in this area. Diversity treatments were maintained by regular removal of any species that had not been sown with the original species assemblage up until the beginning of the parasite experiment.

Seeds of Rhinanthus were collected in a large population in a species-rich meadow in central Switzerland in June 1996 and stored under dry conditions. In October 1996 the seeds were sown at a density of 800 seeds m−2 (well within the range of natural seed densities) into a 50 × 50 cm quadrat established in each plot of the diversity experiment. A second quadrat of the same size within each plot served as parasite-free control. Seedlings of Rhinanthus started to appear above ground at the end of February 1997. From mid-March onwards the number of parasites in each quadrat was counted at monthly intervals. The maximum number of parasite seedlings present in March to May censuses was used to calculate the rate of establishment of the parasite. Ten parasites were selected at random in each treatment quadrat in mid-May. The height of each plant was recorded, as was the total length of inflorescences (a measure of reproductive potential). In addition, the number of capsules per cm of inflorescence length was counted in a subset of 41 plants from different diversity levels. The leaf area index (LAI) was measured at the May census in both treatment and control quadrats (five replicate readings per quadrat) with an electronic fisheye sensor (LAI 2000; Li-Cor, Lincoln, NE, USA).

The survival probability of the parasite was calculated as the ratio between the number of surviving plants in June and the number of seedlings established. An integrative measure of parasite fitness on a quadrat basis was obtained by calculating the number of seeds produced per seed germinated (survival × total inflorescence length × mean number of capsules per cm of inflorescence × mean number of seeds per capsule). A mean value of 6.3 seeds per capsule (D. Matthies, personal observation) was assumed.

At the end of June, the total cover of the vegetation, the proportion of bare ground, the cover of litter and the cover of each species (sown and weed species) were visually estimated in all quadrats. The Shannon-Wiener diversity index (H′) and the evenness index (E) were calculated from the cover data after exclusion of Rhinanthus and weed cover. The evenness index was calculated as E = H′/ln (number of species). In the statistical analysis of the evenness data, monocultures were excluded because obviously they showed no variation in this measure.

At the beginning of July, the vegetation was clipped to 5 cm within each quadrat and the above-ground biomass (including standing dead matter) was separated out for both the host and parasite plants and dried for 24 h at 80 °C before weighing. In mid-September 1997, the total cover of the vegetation, the proportion of bare ground, the percentage of area covered with litter, the composition of the vegetation and the above-ground biomass were again determined in each quadrat. In this autumn harvest, the plant material was sorted into sown-community biomass and weed biomass. Because the parasites are annual and had been harvested earlier, they did not contribute any biomass to the autumn harvest. For the autumn cover estimates, non-sown species within each plot were classed as weeds, with those that did not belong to the pool of 48 species from which the experimental communities were assembled being defined as general weeds. Species from the pool that did not belong to the assemblage originally sown into the plot in which they were observed were classed as plot weeds.

Statistical analysis

According to the hierarchical experimental design, the data were analysed with analysis of variance and deviance using multiple linear and logistic regression approaches (McCullagh & Nelder 1989). All significance tests were carried out using software products especially designed to handle complex experimental designs (Genstat 5, General Statistical Program, release 3, and SPSS, release 7.5).

The treatment model consisted of the effects of the number of species (species richness), the number of functional groups, the presence of legumes and the effect of the Rhinanthus treatment and its interactions with the other factors. Appendix 2 (WWW site) provides a complete skeleton analysis of variance/deviance and shows how the variance ratios were calculated for every factor (see also Hector et al. 1999). The final model was found by backward elimination of factors of minor interest (defined as factors that did not reach an F-value > 2; Green & Tukey 1960).

Significance tests were based on F-tests (anova) or quasi-F-tests (see, e.g. Meyer & Schmid 1999). If necessary, dependent variables were transformed prior to the analysis to meet the assumptions of homoscedasticity and normality. Least-square means (LSQ means) adjusted for block, sub-block and plot are shown with their standard errors in the figures. In the case of transformed data, means and standard errors were back-transformed to the original scale for presentation. The main effects of the diversity treatments on the host community were of no direct interest in this study and are presented elsewhere in conjunction with results from the other BIODEPTH experimental sites (see, for example, Hector et al. 1999; Spehn et al. 2000).

Results

Effects of the host community on the parasite

On average, about 20% of the Rhinanthus seeds sown into the experimental host communities produced a seedling, resulting in a maximum mean density of 161 seedlings m−2 between March and May. Seedling establishment was slightly influenced by the number of functional groups (linear contrast F1,25 = 2.93, P = 0.1) but not by the number of species present in the host communities (linear contrast F1,25 < 1). Instead, establishment varied among assemblages within diversity treatments (F25,14 = 2.32, P = 0.051). Highest seedling establishment (30–50% germination) was observed in grass monocultures (e.g. Arrhenatherum elatius and Poa pratensis), in two- and four-species assemblages containing grass species only, and in a grass/non-legume assemblage containing Arrhenatherum elatius, Cynosurus cristatus, Poa pratense and Achillea millefolium.

The proportion of parasites surviving to the flowering stage decreased with the number of functional groups in the host community (linear contrast F1,25 = 5.01, P < 0.05; analysis of deviance). More than 70% of the parasites survived in host communities with one or two functional groups, but only 55% survived in host communities with three functional groups. Differences in parasite survival among assemblages within diversity treatments were marginally significant (F25,14 = 1.98, P < 0.1). However, if host communities with or without legumes were compared using a ‘legumes’ contrast within the ‘assemblages’ term, this contrast was highly significant (F1,25 = 13.4, P = 0.001). Averaged over all diversity treatments, 83% of the established parasites survived to the fruiting stage in plots without legumes whereas only 52% survived in plots with legumes. An obvious difference between plots with and without legumes was the higher leaf area index (LAI) in plots with legumes (Spehn et al. 2000). When the LAI was fitted in the model as a covariate it explained a considerable part of the variation in parasite survival (F1,13 = 22.4, P < 0.001). As a consequence, the effect of the number of functional groups was no longer significant if it was fitted after the LAI covariate (linear contrast F1,25 < 1). The legume contrast, however, remained significant even after fitting the LAI as a covariate (F1,24 = 8.87, P < 0.01).

Mean parasite density decreased to 130 individuals m−2 at the time of flowering in June. As a consequence of higher seedling establishment and higher survival in plots with few functional groups, the number of parasites per quadrat tended to decrease with the number of functional groups present (linear contrast F1,25 = 4.03, P < 0.06). Among assemblages within diversity levels there were still significant differences in parasite number (F25,14 = 4.61, P < 0.01), but the legume contrast within assemblages was only marginally significant (F1,24 = 3.73, P = 0.065), at the time of flowering.

Mature plants of the parasite had an average biomass of 1.555 g (end of June). In contrast to parasite number, mean parasite size increased linearly with increasing number of functional groups in the host communities (linear contrast F1,24 = 4.69, P < 0.05) and was also positively influenced by the presence of legumes (legumes contrast F1,23 = 4.40, P < 0.05). The total parasite biomass per quadrat also increased with the number of functional groups in the host communities (Fig. 1a, linear contrast F1,25 = 7.85, P < 0.01). It was almost three times as high in host communities with three functional groups (262.8 g m−2) as in host communities with one functional group only (90.5 g m−2). Even if host biomass was fitted as a covariate (F1,13 = 5.39, P < 0.05), it did not completely explain the variation in parasite biomass as the effect of functional group number still remained significant (linear contrast F1,25 = 5.85, P < 0.05).

Figure 1.

The influence of functional diversity of the host community on the performance of the parasite R. alectorolophus. (a) Total parasite biomass per unit area, and (b) total length of inflorescences per parasite as a measure of reproductive potential. Vertical bars denote ± 1 SE.

Total inflorescence length, and thus reproductive potential per parasite, increased with the number of functional groups in the host communities (Fig. 1b, linear contrast F1,25 = 9.69, P < 0.01). No other factors influenced inflorescence length. In contrast with its effects on survival, the presence of legumes in the host communities, even when fitted before the number of functional groups and species (i.e. legumes contrast as first term of the diversity treatments), was not significant (F1,24 < 1).

The number of new seeds produced per Rhinanthus seed germinated, used as an estimate of the fitness of the parasite, was only marginally influenced by the number of functional groups in the host communities (linear contrast F1,25 = 3.72, P = 0.07), but did increase with the number of species present in the host communities (linear contrast F1,25 = 5.10, P < 0.05 if fitted before the number of functional groups; Fig. 2).

Figure 2.

Number of Rhinanthus seeds produced per seed germinated (an integrative measure of parasite fitness) as influenced by species richness of the host communities. Vertical bars denote ± 1 SE.

Effects of the parasite on the host community

Effects on host-community biomass and cover

The parasite strongly reduced the biomass of its host communities. In June 1997, host-community biomass in plots with the parasite was 62% lower than in control plots without the parasite (92.2 g vs. 245.4 g; F1,29 = 159.2, P < 0.001). However, the negative effect of the parasite on host-community biomass decreased with the number of functional groups in the host communities. While the parasite caused a reduction in biomass of 74.5% in plots with one functional group present, it only caused a reduction of 38.4% in plots with three functional groups present (Fig. 3; significant interaction term between parasite and linear contrast of the number of functional groups; F1,25 = 12.0, P < 0.01). The reduction in host-community biomass by the parasite also varied significantly between assemblages within the diversity treatments (significant interaction term between parasite and assemblages; F25,29 = 1.90, P < 0.05).

Figure 3.

Host community biomass per m per m per m2 (without parasite biomass) in parasitized (●) and control (○) quadrats as influenced by the number of functional groups present in June 1997. Vertical bars denote ± 1 SE. Note log scale for biomass.

The main effect of Rhinanthus on total above-ground biomass in the plots (i.e. including both host community and parasite biomass) was only marginally significant (F1,29 = 2.99, P = 0.1). However, there was a significant interaction between the parasite treatment and species richness (F1,25 = 5.15, P < 0.05). While the parasite had no effect on total biomass in species-rich assemblages, it increased total biomass by 36% and 48% in monocultures and two-species assemblages, respectively. Hence, at low diversities the reduction in host-community biomass caused by the parasite was more than compensated for by the parasite's own biomass (Fig. 4a).

Figure 4.

Total community above-ground biomass per unit area (including parasite biomass) as influenced by species richness in parasitized (●) and control (○) quadrats in June 1997. The same data were plotted against two different measures of diversity: (a) species richness of the host and control community and (b) total species richness (including the parasite species). Vertical bars denote ± 1 SE. Note log scales.

Total vegetation cover (including the cover of Rhinanthus) in June was reduced by 8% in parasitized plots (72.3% cover compared with 78.5%; F1,29 = 13.3, P = 0.01). The interaction between parasite treatment and the number of functional groups in the host communities was marginally significant, indicating that the decrease of total cover was greatest in plots with one functional group (interaction term between parasite and the linear contrast of the number of functional groups; F1,25 = 2.96, P = 0.1). On the other hand, the cover of litter increased from 6.3% to 9.7% in parasitized quadrats (F1,29 = 14.9, P < 0.001). There was also a trend for litter cover to be greatest in host communities with one functional group (F1,25 = 3.05, P = 0.09).

At the time of the second harvest in September 1997, two months after the Rhinanthus plants had died, the biomass of the host community in parasite-treatment quadrats was still 26% lower than in control quadrats (F1,29 = 20.0, P < 0.001). This carry-over effect of the presence of the parasite did not depend on the number of functional groups or species in the host community (interaction term between parasite and the linear contrast of the number of functional groups or species; F1,25 = 1.50, P > 0.2 and F1,25 < 1, respectively).

However, at the same time a significant interaction between parasite treatment and the number of functional groups was found for the cover of host-community species and for the proportion of bare ground in the plots. In host communities containing only one functional group the cover of the host-community species was reduced by 17% by the parasite, whereas in host communities with two or three functional groups, the reduction was negligible (interaction term between parasite and the linear contrast of the number of functional groups; F1,25 = 6.5, P < 0.05). Conversely, the amount of bare ground was increased by more than 25% by the parasite in host communities containing only one functional group but hardly changed in host communities with two or three functional groups (interaction term between parasite and the linear contrast of the number of functional groups; F1,25 = 7.10, P < 0.05, Fig. 5).

Figure 5.

Proportion of bare ground per plot as influenced by the number of functional groups present in formerly parasitized (●) and control (○) quadrats in September 1997, after the death of the annual parasites.

Effects on host-community structure on weed invasion

In June 1997, the evenness of host communities was lower in quadrats with Rhinanthus than in control quadrats (F1,16 = 6.12, P < 0.05). The evenness was not affected differentially between host communities varying in the number of functional groups or species present (interaction term between parasite and the linear contrast of the number of functional groups or species; F1,25 < 1 or F1,25 = 2.46, P > 0.14, respectively). Two months later, in September, differences in community structure between parasite treatment and control quadrats were more pronounced. Again, the evenness of the host communities was lower in Rhinanthus than in control quadrats (F1,16 = 18.94, P < 0.001).

The different functional groups were affected differentially by the parasite. In June 1997, the cover of grasses was 18% lower in quadrats with the parasite than without (23% vs. 28% grass cover), whereas the cover of legumes and herbs did not significantly differ between treatment and control quadrats (F1,29 = 7.98, P < 0.01 for grass cover; F1,29 < 1 for legume and herb cover). The variation in cover among functional groups in treatment quadrats was even larger in autumn 1997 when grass cover in the parasite treatment was reduced by 22% (Fig. 6; F1,29 = 41.36, P < 0.001). The difference in legume and herb cover increased as well but was still not significant between treatment and control plots (F1,29 < 1 and F1,29 = 3.14, P < 0.09, respectively; Fig. 6).

Figure 6.

Proportion of area covered by the three different functional groups in autumn 1997 in formerly parasitized and in control quadrats. Vertical bars denote ± 1 SE. Significant differences in cover of functional groups between parasite and control quadrats are indicated with asterisks (*** P < 0.001, (*) P < 0.1).

At the time of the second harvest in September 1997, the parasite treatment significantly increased invasion of the host communities by weed species. The effect for general weeds (i.e. those not sown into any plot) was strongest in host communities with only one functional group (interaction term between parasite and the linear contrast of the number of functional groups; F1,25 = 6.83, P < 0.05; Fig. 7). Cover of these species was also significantly lower in control than in parasite treatment quadrats (0.77 vs. 1.84%; F1,29 = 8.87, P < 0.01). The cover of plot weed species also was significantly increased by the parasite treatment (from 17.9% in control quadrats to 24.7% in treatment quadrats; F1,29 = 10.96, P < 0.01). This increase was stronger in the host communities containing only one or two functional groups than in those containing three where no difference was observed (30.56% vs. 40.10% and 11.12% vs. 17.42%, respectively; interaction term between parasite and the linear contrast of the number of functional groups tested after the species number effect F1,25 = 4.45, P < 0.05). Not surprisingly the amount of above-ground biomass of weed species was also higher (an increase of 58%) in treatment than in control quadrats (F1,29 = 4.47, P < 0.05).

Figure 7.

Mean number of general weed species (species that did not belong to the pool of 48 species from which the experimental communities were assembled) in formerly parasitized (●) and control (○) quadrats (50 × 50 cm) as influenced by the number of functional groups in September 1997. Vertical bars denote ± 1 SE.

Discussion

Several experimental studies have reported that overall ecosystem processes such as biomass production and nutrient retention can be positively related to increasing plant diversity (e.g. Naeem et al. 1994, 1996; Tilman et al. 1997a; Hector et al. 1999). The question of whether single components of ecosystems (i.e. certain functional groups of plants or higher trophic levels) also are influenced by the diversity of plant communities is less frequently addressed but is nonetheless important (Mulder et al. 1999). Root hemiparasites represent both a plant functional group (Press 1998) and a higher trophic level (Atsatt 1977) and their performance in relation to the diversity of the host community can be regarded as an ecosystem process. On the other hand, the presence of a parasite can itself affect the performance of the community. Bottom-up effects of plant diversity as a resource for the parasite and top-down effects of the parasite on plant diversity as a feature of the host community are intrinsically linked with each other.

Parasite performance increased with host-community diversity

Seedling establishment of the root hemiparasite Rhinanthus hardly differed in host-communities of different diversity levels and varied mainly among the assemblages of species. At this stage of their life cycle, Rhinanthus species may be relatively independent of both external conditions and attachment to suitable host plants because their large seeds, containing nutrients and reserves (Hartl 1974), allow a fast start that is typical of the life cycle of some plant parasites (Van Hulst et al. 1987). In contrast, the survival of established parasite seedlings was negatively affected by plant functional diversity. The leaf area index (LAI) of the communities, fitted as a covariate in the analysis of parasite survival, completely explained the effect of functional group number, indicating that this effect was mediated by increased shading in the high-diversity plant communities. Similar negative effects of increased plant community biomass on hemiparasite survival have also been found in another study (Van Hulst et al. 1987). However, mechanisms other than direct shading must have also affected parasite survival, since differences in LAI did not entirely explain the negative effect of legume presence. In general, parasite individuals grew bigger in plots with legumes, suggesting that rapid growth and fast attachment might have allowed some individuals to benefit from an enriched nitrogen supply and to outcompete smaller parasites, thus leading to increased self-thinning of the parasite population (see also Van Hulst et al. 1987). However, mean size of mature parasites in June increased with the functional diversity of the host community, and estimated lifetime fitness of the parasite (i.e. number of seeds produced per seed germinated, Fig. 2) increased with species richness. Thus, our hypothesis that establishment, survival and growth of the parasite increases with increasing diversity of its host community is partly supported. Earlier glasshouse studies investigating the effect of a mixed diet on parasite performance have given ambiguous data. While higher parasite performance in assemblages of two host species compared with host monocultures has been found in Castilleja wightii (Marvier 1998a), no such effect was observed in Melampyrum arvense (Matthies 1996).

If considered an ecosystem process, the higher biomass production of surviving parasite individuals in plots of higher diversity could be an effect of either sampling or complementarity. Host species may differ in the compounds they provide for a hemiparasite and thus complement each other (Govier et al. 1967). Therefore, the higher performance of parasites in our high-diversity plots may be explained by the uptake of a more balanced mixture of compounds when several host species are simultaneously parasitized in communities of higher functional diversity (see also Pennings et al. 1993; Marvier 1998a). However, increased growth of host individuals and therefore a higher host quantity in plots containing a higher number of plant functional groups might also have had a beneficial effect on parasite performance in our experiment. The increased total above-ground host biomass with increasing host diversity (see Spehn et al. 2000) supports this notion. However, as the increase in parasite biomass still remained significant in multiple regression models containing host biomass as a covariate, it can be concluded that the complementarity of resources did indeed have a positive influence on parasite growth. Compared with monocultures, host communities of higher functional diversity also had a higher total root length (E. Spehn, unpublished data).

Sampling effects by which surviving parasite individuals grow bigger because they have a greater chance to attack a single ‘best’ host species in plots of higher diversity cannot, however, provide a sufficient explanation for the increased parasite performance with increased diversity. The sampling hypothesis assumes that assemblages become dominated by the species that performs best in monoculture (Aarssen 1997). However, high-diversity communities were not dominated by any host species, as suggested by the evenness index, which only varied from 0.61 in host communities of two species to 0.51 in host communities of 32 species (the evenness index has a maximum of one and approaches zero if a single species becomes dominant in a community).

Negative effects of the parasite on host community biomass decreased with diversity

It has often been suggested that the stability of plant communities (see Pimm 1984) in terms of withstanding biomass losses caused by a parasite or herbivore attack, or caused by extreme events such as drought, will increase with increasing species diversity and thus with increasing functional redundancy among species (e.g. Brown & Ewel 1987; Frank & McNaughton 1991; Tilman 1996). The results of our study are partially consistent with this hypothesis. The presence of the parasite severely reduced the biomass of the host community, but at the time of peak biomass in June the functionally more diverse communities lost a smaller proportion of their biomass due to parasitism than did the less diverse communities (Fig. 3). Grasses were the group of species most vulnerable to parasite attack (see also Davies et al. 1997; Marvier 1998b; Matthies & Egli 1999), and in plots with several functional groups present, it has been found that legumes and non-leguminous herbs partly compensated for the reduced biomass of grasses. Similar buffering effects of plant diversity have been reported in response to herbivory. In four tropical successional ecosystems differing in plant-species number and plant-species composition, communities of higher diversity lost a lower proportion of available leaf area to herbivory (Brown & Ewel 1987).

In our experiment, both the resistance of the community against biomass loss caused by the parasite and the performance of the parasite increased at high levels of diversity. The apparently contradictory situation where both components ‘win’ is due to the fact that host-community biomass increased considerably with the number of functional groups present. Host communities containing three functional groups produced 5–8 times as much above-ground biomass as those containing only one functional group (Hector et al. 1999; Spehn et al. 2000). Thus, in our study ecosystem stability and parasite performance were more closely related to the number of functional groups than to the number of species in the host communities. Many diversity effects in plant communities seem to be the result of niche complementarity among species (Hutchinson 1959; Tilman et al. 1997b). However, niches overlap to some degree between species. Grouping of species according to the functions they fulfil in the ecosystem is likely to be associated with considerable overlap within these functional groups. The effects of species number on ecosystem functioning in such cases should thus be fully explained by the number of functional groups (Grime 1997; Tilman et al. 1997a; Peterson et al. 1998).

The parasite can also be viewed as part of the whole ecosystem and processes can then be studied at the level of the total community. The presence of the parasite then increases both total species diversity and functional diversity by one. Analyses on this basis showed that the total above-ground biomass in low-diversity plots in June was higher if they contained the parasite than if they did not, indicating that even a hemiparasitic plant can contribute to resource-use complementarity and thus increases total community biomass (Fig. 4a, b). This is surprising because root hemiparasites have been found to be inefficient resource users and, in most glasshouse and field studies, led to reduced total productivity (Ter Borg & Bastiaans 1973; Matthies 1995, 1997), although Rhinanthus in northern Italian meadows was found not to suppress total community biomass significantly (Davies et al. 1997). Although total above-ground biomass was not negatively affected by Rhinanthus in our study, total biomass including roots could well have been reduced because, compared with their hosts, hemiparasites invest very little biomass into roots (Matthies 1995, 1997). In September, after the death of the annual parasites, total above-ground biomass was lower in formerly parasitized plots than in control plots at all levels of diversity.

Resistance to parasite-mediated weed invasion increased with host-community diversity

Hemiparasites like Rhinanthus are among the few annual plants that can successfully establish themselves in closed grassland communities (Ter Borg 1985; Van Hulst et al. 1987), where, as in this study, they can reduce the cover of vegetation (De Hullu 1985). Moreover, the death of an annual hemiparasite may leave gaps in the vegetation and may therefore facilitate the invasion of other plants. Our hypothesis was that the resistance of host communities against weed establishment would be lowered by the parasite, particularly in plots with low diversity where reduced resource competition or an increased number of potential invaders may already have increased susceptibility (e.g. Crawley 1987; Tilman 1997; Levine & D'Antonio 1999). Our hypothesis was supported by weed colonization of gaps left by the death of the parasite. In low-diversity plots the proportion of open ground, the number of weedy species and their cover were all higher when they had been formerly parasitized. Rhinanthus can thus be regarded as a driver species in the terminology of Walker (1992) or as a short-term allogenic physical engineer in the terminology of Jones et al. (1997).

By decreasing the resistance of low-diversity communities to invasion, the parasite may actually increase the diversity of its host communities in the medium term, because more species have a chance to become established. One could even speculate that this facilitation of diversity should eventually lead to a stable equilibrium of parasitized ecosystems where the increased stress resistance of the diversifying host community counterbalances its reduced invasion resistance. This positive effect of hemiparasites on diversity could be of practical use for the re-establishment of species-rich permanent grassland.

Acknowledgements

This project was supported by a grant from the Swiss Federal Office for Education and Science (Project EU-1311 to B.S.) to join the EU-funded BIODEPTH project. D. Matthies was supported by grant no. 5001–44626 of the Swiss National Science Foundation. We would like to thank E. Spehn, M. Diemer and various assistants for help and company in the field and M. Fischer, L. Haddon, A. Hector, M. Press, E. Spehn and two anonymous referees for improvements on earlier versions of this manuscript. Hospitality of the Centre of Population Biology at Silwood Park during the writing of this manuscript is also gratefully acknowledged.

Received 22 June 1999revisionaccepted 7 February 2000

Ancillary