1. Although carnivores indirectly improve plant fitness by decreasing herbivory, they may also decrease plant reproduction by disrupting plant–pollinator mutualism. The overall magnitude of the resulting net effect of carnivores on plant fitness and the factors responsible for the variations in strength and direction of this effect have not been explored quantitatively to date.
2. We performed a meta-analysis of 67 studies containing 163 estimates of the effects of carnivores on plant fitness and examined the relative importance of several potential sources of variation in carnivore effects.
3. Carnivores significantly increased plant fitness via suppression of herbivores and decreased fitness by consuming pollinators. The overall net effect of carnivores on plant fitness was positive (32% increase), indicating that effects via herbivores were stronger than effects via pollinators.
4. Parasitoids had stronger positive effect on plant fitness than predators. Active hunters increased plant fitness, whereas stationary predators had no significant effect, presumably because they were more prone to disrupt plant–pollinator mutualism. Carnivores with broader habitat domain had negative effects on plant fitness, whereas those with narrow habitat domain had positive effects.
5. Predator effects were positive for plants which offered rewards (e.g. extrafloral nectaries) and negative for plants which lacked any attractors.
6. This study adds new knowledge on the factors that determine the strength of terrestrial trophic cascades and highlights the importance of considering simultaneous contrasting interactions in the same study system.
Several characteristics of carnivores may affect the strength and direction of their effects on plant fitness. For instance, predator’s hunting mode has been shown to affect predator’s impact on prey behaviour, plant biomass and ecosystem functioning (Schmitz & Suttle 2001; Preisser, Orrock & Schmitz 2007; Schmitz 2007, 2008). Active hunters (e.g. birds and ants) forage on both vegetative and reproductive plant structures and are more likely to come into contact with (and hence feed more on) relatively sedentary herbivores than with very mobile and transient pollinators. In contrast, sit-and-wait predators like crab spiders and lizards usually forage upon or close to sites of predictably high prey densities like flowers (Morse 2007) and hence are likely to capture more pollinators than herbivores. Also, carnivores having broad habitat domain, i.e. large range of microhabitat use and high ability to pursue prey, may influence the whole pollinator community and hence have stronger cascade effects on plants than carnivores having narrow habitat domain (Preisser, Orrock & Schmitz 2007; Schmitz 2007). Furthermore, we predict that parasitoids have stronger positive effects on plant fitness than predators; this is because parasitoids are more specialized than other carnivores in their herbivore host requirements and also because they are less likely to disrupt plant–pollinator interactions.
Some plants have mutualistic relationships with carnivores and use them as indirect biotic defences against herbivores (Heil 2008; Rosumek et al. 2009). In this case, plants usually provide carnivores with various kinds of rewards (e.g. extrafloral nectaries and domatia), which greatly affect carnivore densities and activity and, hence, their effect on herbivores (Rosumek et al. 2009). Therefore, we predict that carnivores have stronger positive effects on fitness of plants that have mutualistic interactions with carnivores when compared to plants that lack any biotic defence.
The majority of studies on carnivore effects on plant fitness have been conducted in either temperate or tropical climatic zones. Tropical plants are more prone to be pollen limited (Larson & Barrett 2000), and the longevity of tropical flowers is shorter than that of temperate ones (Primack 1985). Therefore, disruption of pollination by carnivores may have more negative consequences for tropical plants than for temperate ones. On the other hand, tropical plants suffer more damage from herbivores than temperate plants (Coley & Barone 1996; Schemske et al. 2009) and hence may be expected to benefit more from carnivore effects on herbivores. Therefore, we predict that both the negative and positive cascading effects of predators on plant fitness are stronger in the tropics.
In this review, we explore the above sources of variation by conducting a meta-analysis of 67 individual studies on carnivore effects on plant fitness. The main questions addressed were: (i) What is the overall net magnitude of the carnivore effect on plant fitness components? (ii) Do predator hunting mode and habitat domain affect the magnitude or direction of carnivore effects on plant fitness? (iii) Does the presence of indirect plant defences and, hence, attractors for carnivores affect the magnitude and direction of the carnivore effects on plant fitness? and (iv) Does the strength of the trophic cascades vary between temperate and tropical regions?
Materials and methods
Data base, Inclusion Criteria and Sources of Variation
The search strategy and the complete data base are presented in the Appendix S3 (Supporting Information). To be included in our meta-analysis, the papers had to meet the following criteria. First, the study should have provided a comparison of plant fitness in the presence of carnivores and under conditions where carnivores were absent or their density was reduced. To evaluate the effects of carnivores on plant fitness, we included in our analyses only the estimates from experiments in which the treatments (presence/absence of carnivores) were applied to separate plants rather than to individual flowers, inflorescences or flowering stems within the same plant. We made an exception for ants and included studies comparing effects of ant presence and absence on different stems within the same plant for the following reasons: (i) in contrast to other predators that typically forage on a single flower (e.g. crab spiders), ants forage over the whole plant they have access to; (ii) there was no significant difference between effects of ants on whole-plant fitness and reproductive output of individual stems within a single plant (Qb = 2·6, d.f. = 1, P = 0·305). We included experimental studies that have manipulated carnivore presence or density [either directly, by introducing or removing carnivores, or indirectly, by removing plant rewards such as extra-floral nectaries (EFNs)] as well as observational studies conducted in sites where carnivores were naturally present or absent or present at different densities (gradient studies). In the latter cases, we only included data from sites with min and max carnivore densities, as carried out in previous meta-analyses (e.g. Knight et al. 2006). In the case of observational studies, we designated as control the condition in which predators were absent or occurred at very low density and as experimental treatment the condition in which predators were present or occurred at higher density. Second, to be included in our meta-analysis, studies in the control and experimental groups had to be carried out simultaneously and using the same plant species. Third, we considered papers that explicitly investigated the influence of predators or parasitoids on plant traits that are directly related to plant fitness. The plant fitness components included were number of flowers, fruits or seeds per plant or stem and number of seeds per fruit. If studies reported herbivore damage to seeds and fruits in the absence and the presence of carnivores, these data were also included in our meta-analysis, but the sign of the effect (− or +) was reversed (i.e. reduced herbivore damage to fruits and seeds in the presence of predators indicated positive effect of predators on plant fitness). Data on plant growth, plant biomass, plant nutrition, seed or fruit biomass, or bud and flower damages were not included in our meta-analysis because these measures do not assure that plant fitness is affected. Fourth, we only included studies on indirect effects of carnivores on plant fitness through trophic cascades via herbivores or pollinators; we disregarded studies in which predators acted as pollinators or nectar thieves and studies on parasitic castration in ant-myrmecophyte systems. Finally, to be included, studies had to report mean fitness estimates for control and experimental treatments, some measure of variation (SD, SE or CI) and sample sizes.
In addition to the estimates of the effects of predators on plant fitness components, we also recorded several study characteristics (sources of variation). These characteristics included predator hunting mode, predator habitat domain, predator taxa and mechanism of their effects (via pollinators or via herbivores). Carnivore effects on plant fitness were classified as mediated via herbivores if the study compared degrees of seed or fruit damage in the presence and the absence of herbivores and the authors of the original study explicitly interpreted the effect as herbivore mediated. Similarly, carnivore effects were classified as pollinator mediated if the study compared numbers of undamaged seeds or fruits in the presence and the absence of carnivores and the authors of the original study explicitly interpreted the effect as pollinator mediated. Five papers (see Appendix S1, Supporting Information) evaluated both effects via herbivory and effects via pollinators; these data were generally obtained from different flowers or inflorescences per plant or using different experimental sets. Carnivore taxa included were spiders, ants, bugs, wasps, beetles, dragonflies, birds and reptiles (lizards and snakes); we also included parasitoids (parasitic Hymenoptera) in our analysis. Hunting mode was classified as sit-and-wait (e.g. crab spiders, lizards), sit-and-pursue (e.g. lynx spiders) or active hunters (e.g. ants, jumping spiders, dragonflies, birds, wasps); Appendix S1 (Supporting Information) provides hunting modes for each carnivore species evaluated. Predator habitat domain was classified as narrow for those taxa that typically forage locally upon foliage or flowers (spiders, beetles, bugs) and broad for those taxa that forage in relatively large areas and are not restricted to one or few individual plants (e.g. birds, dragonflies, wasps, ants and reptiles). Parasitoids were excluded from the analyses of effects of hunting mode and habitat domain because they are very specialized in their foraging and not comparable in this respect to carnivores. We also recorded whether the plant species in question uses carnivores as indirect defences (see Heil 2008) and thus offers food rewards (floral and EFNs) and other carnivore attractors (fruit juice and presence of honeydew-producing insects), plant structures used for shelter (domatia) or foraging sites (glandular trichomes). We defined tropical studies as those undertaken between the tropics of Cancer and Capricorn. Additional sources of variation examined (plant habit, predator diversity, system [natural vs. managed], type of study [observational and experimental], length of the experimental studies and invasive vs native predators) are presented in the Appendix S4 (Supporting Information).
Data Extraction from Studies
Many studies reported more than one estimate of effects of predators on plant fitness, which may be nonindependent. Nonindependence of the data may lead to underestimation of the standard error of the mean effect and thus may inflate significance levels for statistic tests (see Gurevitch & Hedges 1999; Koricheva 2002). We minimized nonindependence of multiple estimates in our data base as following: (i) for time series measures, we only included data from the last estimate to decrease the likelihood of evaluating transient dynamics (cf Schmitz, Hambäck & Beckerman 2000); (ii) independent experiments reported in the same paper that were conducted in different seasons or years or in distinct geographical regions were treated as independent comparisons because predators under study migrate relatively short distances and have short life cycles, and thus, it is likely that their effects are local and seasonally restricted; (iii) When multiple response variables (e.g. fruits in different phenophases) were reported, we considered only those which are most representative for plant fitness (e.g. fruits in oldest phenophase). The above steps allowed us to reduce the number of estimates per study to 2·4 on average. To avoid violation of the assumption of independence of statistical tests owing to inclusion of multiple estimates per study, we used the ‘shifting units’ approach recommended by Cooper (1998). First, to estimate the overall effect of carnivores on plant fitness, we have conducted meta-analysis using one effect estimate per study calculated as a mean of effect size estimates reported in each study. Second, to analyse sources of variation in the magnitude of the effect, we have included multiple estimates per study. This allowed us to retain the maximum information from studies and hence to examine more sources of variation in the magnitude of the effects. The problem of nonindependence of multiple comparisons from the same study is much less severe for the analyses of sources of variation because in such analyses, multiple estimates of the effect from the single study are included in different subgroups to be compared, i.e. each study would contribute only one data point to each of the categories distinguished by the explanatory variable (Cooper 1998).
In addition to multiple effects from the same study, the explanatory variables considered in the meta-analysis may also be nonindependent from each other. Associations among these variables were tested using G-test (Zar 1996). In the case of significant associations, effects of each variable on the strength of the response variable were evaluated separately at each level of the other explanatory variable (see Koricheva 2002).
Data on means, standard deviations and sample sizes for experimental treatments (carnivores present) and controls (carnivores absent or at low density) were obtained from texts, tables or graphs. The graphs were enlarged and saved as JPEG figures and then imported to ImageJ 1.42q software (http://rsb.info.nih.gov/ij) for data extractions.
We first converted values of means, standard deviations and sample sizes of each study to effect sizes and their associated variances to place the data from primary studies on a common scale (Gurevitch & Hedges 2001). Here, we estimated the magnitude of carnivore effect on plant fitness by using log response ratio (ln R), ln R = ln(XE/XC), where XE and XC denote mean plant fitness in the presence (or higher density) and absence (or reduced density) of carnivores, respectively (Hedges, Gurevitch & Curtis 1999). Positive or negative effects indicate that carnivores increase or decrease plant fitness, respectively. The condition for the use of ln R is that the and >3 (Hedges, Gurevitch & Curtis 1999), where N, X and SD denote, respectively, sample size, mean and standard deviation from experimental (E) or control groups (C). Our data met this requirement as 90% of the values of and 86% of the values of were >3. We conducted analyses using ln R as a measure of effect size and then back transformed ln R to % difference between control and treatment [as (EXP ln R−1) × 100%] for the ease of interpretation.
To ensure that the results of our meta-analysis are not biased owing to the choice of ln R as a metric of the effect size, we have also conducted analyses using standardized difference between the means, Hedges’d, as metric of the effect size (Hedges & Olkin 1985). The results of the analysis were the same for both metrics, and therefore, we report only ln R and corresponding % fitness changes because of the ease of interpretation and because ln R has been used in previous meta-analyses on trophic cascades (Schmitz, Hambäck & Beckerman 2000; Shurin et al. 2002; Borer et al. 2005; Knight et al. 2006; Rosumek et al. 2009).
We used a mixed effects meta-analysis model, which assumes that there is a random variation among studies within each category of explanatory variable (e.g. within each type of hunting mode or type of habitat domain), but variation among categories is fixed (Gurevitch & Hedges 2001). Mixed models are often preferable to fixed effects models because their assumptions are more likely to be satisfied in ecological data (Gurevitch & Hedges 2001). Ninety-five per cent confidence intervals (CI) were estimated by bootstrapping based on 4999 iterations, and the effect sizes were considered significant when the CI did not overlap with the zero. P values for the between-group heterogeneity (Qb) tests were obtained by randomization tests based on 4999 iterations. The meta-analysis was carried out by using the MetaWin 2.0 statistical software (Rosenberg, Adams & Gurevitch 2000). Analyses of publication bias are presented in the Appendix S5 (Supporting Information).
The final data base consisted of 163 estimates of effect sizes from 67 papers or book chapters published in 1967–2009 (Appendices S1 and S2, Supporting Information). Most of the studies examined effects of invertebrate carnivores (95% of all comparisons), particularly ants and spiders (79%), and 54% of the studied plants were woody (shrubs, trees, vines). Seventy-seven per cent of the 163 estimates referred to plants that had some type of attractors for predators (e.g. EFNs, domatia, honeydew-producing herbivores and glandular trichomes). Most of the estimates were from studies conducted in natural field conditions (84%), and the rest of studies were conducted in agricultural fields, greenhouses and common gardens (Appendix S1, Supporting Information).
The overall effect of carnivores on plant fitness was positive and significantly different from 0 (analysis based on single mean estimate per study: ln R = 0·27, 95% CI = 0·14–0·44, n = 67; analysis based on multiple estimates per study: ln R = 0·28, 95% CI = 0·19–0·37, n = 163); this corresponds to 31–32% increase in plant fitness in the presence of carnivores. However, total heterogeneity among studies (Qt) was very high (analysis based on single mean estimate per study: Qt = 159·01, d.f. = 66, P < 0·0001; analyses based on multiple estimates per study: Qt = 443, d.f. = 162, P < 0·0001), indicating that effects varied significantly among studies, and thus, analysis of explanatory variables is warranted. The magnitude of the effect did not depend on the fitness measure used (no. of flowers, fruits and seeds) (Qb = 3·4, d.f. = 3, P = 0·683). Therefore, in the rest of the analyses, we did not differentiate between different fitness measures. As expected, predators had opposite effects on plant fitness depending on whether they affected plant mutualists (pollinators) or antagonists (herbivores). When predators interfered in plant–pollinator mutualism, they decreased plant fitness by 17% (ln R = −0·19, 95% CI = −0·29 to −0·09, n = 36). In contrast, when predator effects on plant fitness were mediated by their negative effects on herbivores, plant fitness was increased by 51% (ln R = 0·41, 95% CI = 0·38 to 0·60, n = 126). For studies in which both effects via pollinators and herbivores were evaluated in the same system (Appendix S1, Supporting Information: Louda 1982, Norment 1988, Altshuler 1999; Romero & Vasconcellos-Neto 2004, Romero, Souza & Vasconcellos-Neto 2008), the effect of carnivores on plant fitness via pollinators was nonsignificant (ln R = 0·01, 95% CI = −0·07 to 0·08, n = 10), but positive and significant via herbivores (ln R = 0·40, 95% CI = 0·14–0·65, n = 7; Qb = 14·2, d.f. = 1, P = 0·002). The overall net effect of carnivores via pollinators and herbivores in these studies was positive and significant (ln R = 0·10, 95% CI = 0·0005–0·24, n = 17).
Predator hunting mode influenced the strength of the trophic cascade (Qb = 18·8, d.f. = 2, P = 0·041). The sit-and-wait and sit-and-pursue predators had no significant effects on plant fitness, whereas the active hunters increased plant fitness by 36% (Fig. 1). However, the predator hunting mode was significantly associated with the presence of plant attractors; sit-and-pursue predators occurred only on plants with some type of attractors, whereas sit-and-wait predators and active hunters occurred on both plants with and without attractors. When we restricted the analysis to plants that had no attractors for predators, the effect of sit-and-wait and active hunters became marginally significant and negative (Fig. 1), and the difference between the two types of predators was nonsignificant (Qb = 0·17, d.f. = 1, P = 0·706).
To compare the magnitude and direction of the effects for predators with broad and narrow habitat domains and among different carnivore taxa, we removed studies on plants with attractors from the analysis because carnivore taxon, domain and presence of attractors were not independent from each other (i.e. vertebrate carnivores and many predators with broad habitat domain usually occurred on plants without any attractors; Gdomain vs. attractor = 7·08, d.f. = 1, P = 0·008). In addition, habitat domain was not independent of hunting mode (G = 5·32, d.f. = 1, P = 0·021). Therefore, we compared effects of broad vs. narrow habitat domain predators within each type of hunting mode category. Within the sit-and-wait category, predators with broad habitat domain decreased plant fitness by 52%, whereas narrow-domain predators had no significant effect (Fig. 2; Qb = 14·5, d.f. = 1, P = 0·013). Within the active hunting mode, broad-domain predators had significantly negative effect on plant fitness, whereas narrow-domain predators had significantly positive effect (Fig. 2; Qb = 21·5, d.f. = 1, P = 0·002). The broad-domain carnivores included reptiles, birds, dragonflies, wasps and ants, and all of them had significant negative effects on plant fitness (Fig. 3; wasps: ln R = −0·53). Narrow-domain carnivores included spiders, bugs, parasitoids and beetles. While bugs, beetles and parasitoids had positive effects on plant fitness, the effects of spiders did not differ from 0 (Fig. 3). Overall, vertebrates had a strong negative effect decreasing plant fitness by 40%, whereas invertebrates (including parasitoids) in the absence of attractors increased plant fitness by 33% (Fig. 3, Qb = 32·3, d.f. = 1, P < 0·001). Also, parasitoids had stronger and more positive effects on plant fitness than predators (Qb = 44·9, d.f. = 1, P < 0·001): while parasitoids increased plant fitness by 99% (Fig. 3), other carnivores on average reduced plant fitness by 18%.
Presence and type of plant attractors for carnivores significantly influenced the magnitude and direction of the effects of predators on plant fitness (Qb = 64·3, d.f. = 5, P < 0·001). Predators decreased by 19% fitness of plants that lacked any attractors (ln R = −0·21, 95% CI = −0·36 to −0·05, n = 24), but increased by 42% fitness of plants that had attractors (ln R = 0·35, 95% CI = 0·25–0·46, n = 125). When studies on ants (the most commonly studied predators) were excluded from the analysis, the effects of carnivores on the fitness of plants that have attractors were still positive (ln R = 0·15, 95% CI = 0·05–0·26, n = 22). When the effect of carnivores via pollinators was evaluated separately, carnivores had no significant effect on fitness of plants that had attractors (ln R = 0·02, 95% CI = −0·08 to 0·03, n = 21), whereas in the absence of attractors, plant fitness was reduced by 32% (ln R = −0·39, 95% CI = −0·54 to −0·24, n = 15). As predator taxon was not independent of the presence and type of plant attractor, we analysed the influence of plant attractors separately for spiders and ants because these arthropods were the most representative in our meta-analysis. When analysis was restricted to studies on spiders, plant fitness was increased by 13% in the presence of attractors (ln R = 0·13, 95% CI = 0·03–0·23, n = 18), but the effect did not differ from 0 in the absence of attractors (ln R = −0·13, 95% CI = −0·32 to 0·006, n = 4). Spiders that live on plants with EFNs improved plant fitness by 23%, while spiders that live on plants with glandular trichomes and on plants with no attractors had no significant effect on plant fitness (Fig. 4). However, when the effects via pollinators were removed from the analysis, effects of spiders on the fitness of glandular plants were higher than those on plants with EFNs, improving plant fitness by 54% (ln R = 0·43, 95% CI = 0·16–0·68, n = 5, Qb = 0·004, d.f. = 1, P = 0·97). For studies on ants, fitness of plants without attractors was negatively affected by the predator (ln R = −0·13, 95% CI = −0·33 to −0·01, n = 4), whereas fitness of plants with attractors was increased (ln R = 0·40, 95% CI = 0·29–0·54, n = 103; Qb = 10·3, d.f. = 1, P = 0·048). Note, however, that the number of studies without attractors was very small, and results of this comparison should be interpreted with caution. There was no statistical difference among different types of plant attractors (Qb = 8·04, d.f. = 2, P = 0·177) (Fig. 4). Predatory bugs inhabiting leaf domatia also improved plant fitness by 23% (ln R = 0·21, 95% CI = 0·15–0·25, n = 4).
The strength of negative effects of predators on plant fitness via pollination did not differ between the tropics and the temperate regions (Qb = 0·11, d.f. = 1, P = 0·82). Similarly, there was no significant difference in predator effects on plant fitness via herbivory reduction between the tropics and the temperate zone (Qb = 0·16, d.f. = 1, P = 0·79) (Fig. 5).
Our analyses show that the overall effect of carnivores on plant fitness depends on the relative effects of carnivores on plant mutualists (pollinators) and antagonists (herbivores). In systems where carnivores disrupted plant–pollinator interactions, plant fitness was reduced by 17% in the presence of carnivores (cf Knight et al. 2006). In contrast, when carnivores affected plants through their effects on herbivores (the classical trophic cascade), their effect on plant fitness was positive and stronger than the effect via pollination (51% increase) (cf Schmitz, Hambäck & Beckerman 2000). Our meta-analysis has shown for the first time that the overall net effect of carnivores on plant fitness is positive (32% increase) and is similar in magnitude or even stronger than previously reported carnivore effects on plant biomass and herbivory rate (Schmitz, Hambäck & Beckerman 2000; Shurin et al. 2002; Borer et al. 2005). A similar result (positive effect of carnivores) emerged when we restricted analysis only to studies in which carnivore effects on plant fitness via pollinators and herbivores were studied in the same system. Therefore, our findings add to a growing consensus that trophic cascades are present in terrestrial ecosystems (Schmitz, Hambäck & Beckerman 2000; Halaj & Wise 2001; Shurin et al. 2002; Borer et al. 2005; Shurin, Gruner & Hillebrand 2006), although they are weaker than in many aquatic ecosystems (Shurin et al. 2002; Borer et al. 2005; Shurin, Gruner & Hillebrand 2006). Moreover, the results from the present study also support our prediction that the strength and direction of terrestrial trophic cascades are strongly influenced by relative effects of carnivores on pollinators vs. herbivores, predator hunting mode, carnivore habitat domain and taxonomy, and presence and type of plant attractors.
The net positive effect of carnivores on plant fitness suggests that carnivore effects on herbivores were stronger than on pollinators. This could be because of several factors. First, most herbivores are more sessile and remain longer on plants than pollinators and thus are more vulnerable to carnivores (e.g. Romero, Souza & Vasconcellos-Neto 2008), while direct lethal effects of carnivores on pollinators may be less common (Suttle 2003; Morse 2007; Gonçalves-Souza et al. 2008; Ings & Chittka 2009). Second, many pollinators have well-developed sensory systems that allow more detailed assessment of the foraging areas when compared to herbivores (Gonçalves-Souza et al. 2008; Ings & Chittka 2009; but see Sendoya, Freitas & Oliveira 2009). Third, plants may be buffered against loss of pollination by attracting different types of pollinators, some of which are inaccessible to carnivores (i.e. larger sizes) (Dukas & Morse 2005; Morse 2007). Fourth, predators (e.g. ants) may increase relocation frequency of pollinators and thus increase the rate of flower visitation (Altshuler 1999) and plant fitness. Fifth, stronger effects of carnivores on herbivores than on pollinators may be an artefact. Only 4 of 17 studies that examined effects of carnivores on plant fitness via pollinators controlled for herbivore presence or reported that herbivores were absent. If herbivores that feed on reproductive tissues were present in the majority of studies that evaluated carnivore–pollinator interactions, their effects may be underestimated. Positive effects of herbivore removal might have to some extent offset negative effects of predation on pollinators. We therefore suggest that future studies aiming at elucidation of the mechanisms behind carnivore effects on plant fitness control for herbivore presence on the plants.
The strength of carnivore effects on plant fitness also depended on carnivore hunting mode. We expected that cues from stationary predators (sit-and-wait) could be more indicative of imminent predation risk and thus trigger stronger negative effects on floral visitors than cues from actively hunting predators (Preisser, Orrock & Schmitz 2007; Schmitz 2007). In contrast, we found that the effects of sit-and-wait and sit-and-pursue predators on plant fitness were not significant, whereas active hunters increased plant fitness when plant attractors were present. Although sit-and-wait predators like crab spiders have been shown to affect pollination and seed set of individual flowers within the plant (e.g. Gonçalves-Souza et al. 2008; Brechbühl, Kropf & Bacher 2010), their effects on the whole-plant fitness appear to be weaker. Plants have various chemical and mechanical strategies to restrict foraging by active hunters (e.g. ants) to foliage rather than flowers (Willmer et al. 2009), thus reducing negative effects on pollinators. Therefore, our findings do not support our prediction that stationary predators are more prone to disrupt plant–pollinator mutualism, but instead suggest that active hunters have more positive effects on plant fitness.
Habitat domain of the predator also significantly affected the strength of trophic cascades on plant fitness. Predators with broad habitat domain (e.g. wasps, dragonflies, ants, birds and reptiles) had negative effects on plant fitness, whereas the majority of carnivores with narrow habitat domain had positive effects (Fig. 3). Broad habitat domain predators typically forage by flying (except ants and reptiles) and are likely to affect the abundance and behaviour of pollinators, which also reach flowers by flying. In contrast, most of the narrow-domain carnivores (e.g. bugs, lady birds and parasitoids) are likely to have stronger effects on herbivores than on pollinators, and hence, their positive effects outweigh the negative effects on plant fitness.
Parasitoids increased plant fitness by 99%, whereas predator effects were considerably weaker: 30% increase in plant fitness in the presence of attractors and 18% reduction in fitness in the absence of plant attractors. Among different predator taxa, only beetles tended to have stronger effects on plant fitness than parasitoids in the absence of attractors (Fig. 3). Stronger and more positive effects of parasitoids on plant fitness may be because of their higher feeding specialization and lack of negative effects on pollinators. Therefore, parasitoids might be the only guild of carnivores that have no negative effect on plant fitness.
Some plant species have evolved mutualistic relationships with carnivores and actively recruit them as indirect defences against herbivores by providing various rewards in the form of food (EFN, fruit juices), ‘accommodation’ (domatia) and foraging sites (glandular trichomes). Although honeydew-producing insects can be costly to the plants, the presence of such insects may also attract carnivores to the plants and ants are known to be very protective of the plants bearing colonies of honeydew-producing insects (Styrsky & Eubanks 2007). Use of carnivores as indirect plant defences can be effective in improving plant reproduction (reviews in Heil 2008; Romero & Benson 2005; Rico-Gray & Oliveira 2007) and is likely to be the crucial determinant of the direction of the trophic cascades on plant fitness in terrestrial ecosystems. However, none of the previous meta-analytical studies on trophic cascades (Schmitz, Hambäck & Beckerman 2000; Halaj & Wise 2001; Dyer & Coley 2002; Shurin et al. 2002; Bell, Neill & Schluter 2003; Borer et al. 2005; Knight et al. 2006) has evaluated the role of indirect plant defences comprehensively. It has been proposed that plant chemical defences may buffer trophic cascades in terrestrial ecosystems by dampening lower interactions (herbivores producers) (see Shurin et al. 2002; Shurin, Gruner & Hillebrand 2006 and references therein). Here, we showed that indirect defences may have opposite effect, i.e. they triggered positive trophic cascades on plant reproduction, whereas the overall magnitude of top-down effects was negative in the absence of them.
While spiders improved plant fitness in the presence of EFNs, they had no significant effect on fitness of plants with glandular trichomes. This might be because all the studies on glandular trichomes evaluated both the costs and benefits of spiders on plant fitness, while some studies on plants bearing EFNs evaluated only the positive effects of spiders on plant fitness via suppression of herbivores and not their potential negative effects via pollinators. When we excluded effects via pollination from the analysis, the magnitude of the effects of spiders on fitness of plants with glandular trichomes (ln R = 0·43) was higher than that of spiders on plants bearing EFNs (ln R = 0·20). Thus, the actual overall effect of EFNs as mediator of spider–plant mutualisms remains to be investigated. As expected, in the presence of plant attractors, ants improved plant fitness. The previous meta-analysis by Rosumek et al. (2009) evaluated only the effects of food rewards (EFNs, food bodies) on plant reproduction, whereas we have also evaluated the role of other plant attractors, such as reproductive structures (e.g. flower, fruit) and presence of ant-tended insects on plant fitness. While ant effects on plant fitness were stronger in the presence of ant-tended insects and EFNs, the presence of reproductive structures had no significant effect. It is possible that by foraging on reproductive structures (flowers), ants may remove herbivores, but also chase away pollinators, thus reducing positive effects on plant fitness. In contrast, although both ant-tended insects and EFNs may be costly for plants (reviewed in Rutter & Rausher 2004; Styrsky & Eubanks 2007), their costs may be offset by the benefits the ants provide by removing herbivores. Costs and benefits of plant associations with ants are still poorly explored and deserve further attention.
Previous meta-analytical studies have shown that the effects of herbivores on plant biomass, effects of predators on prey, effects of predators on plant biomass and herbivory (Dyer & Coley 2002) and biotic defences provided by ants against herbivory (Rosumek et al. 2009) are all stronger in the tropics than in temperate regions. We also predicted that the strength of trophic cascade via both herbivores and pollinators on plant fitness would be stronger in the tropics. However, our analysis showed that the strength of cascading effects of predators on plant fitness is similar among the tropics and temperate regions. Although tropical plants receive more herbivory than temperate ones (review in Schemske et al. 2009), they may be better able to tolerate herbivory or compensate for it by producing more flowers and seeds after herbivore attack. In addition, only those plants that are more pollen limited may have been selected in temperate regions to test for the trophic cascade of carnivores via pollination. Thus, other factors not measured here may have influenced plant reproduction in addition to presence of predators.
Conclusions and Suggestions for Future Studies
Our meta-analysis evaluated the relative importance of factors that can affect the dynamics of the interactions between carnivores and plant fitness. We found that cascading effects of carnivores via herbivores were stronger than via pollinators, which is likely to be because of higher vulnerability of herbivores to predators or lower sensitivity to predation risk than pollinators. Details on natural history of pollinators and herbivores, especially on their avoidance behaviour against predation risk, are scarce in the literature and could help understanding the indirect cascade effects of predators on plant fitness. Comparative studies of predation risk sensitivity of herbivores and pollinators could be valuable. Predator hunting mode (stationary vs. active hunters), habitat domain and plant indirect defences emerged as important predictors of predator effects in our analysis. Future empirical studies could investigate simultaneously the indirect effects of stationary vs. active hunters on the pollinator behaviours and plant fitness, as well as the contrasting impacts of predators exhibiting broad vs. narrow habitat domain on herbivory and pollination. As expected, plant indirect defences caused strong variation in the strength and direction of the trophic cascades.
Although we have examined as many studies as possible to evaluate the contrasting trophic cascades on plant fitness through pollination and herbivory in terrestrial ecosystems, the majority of studies were restricted to trophic cascades via herbivory (75%). Several studies and variables have been included since a previous meta-analysis on trophic cascade via pollinators (Knight et al. 2006), but the predator–pollinator–flowering plant interactions are still poorly known. Moreover, ideally carnivore effects via both pathways (pollinators and herbivores) should be evaluated in the same system. To date, only 5% of studies evaluated both of these contrasting pathways. Furthermore, for most of the trophic cascades evaluated here, it is unclear whether the indirect interactions of carnivores with producers are via consumptive or nonconsumptive effects on prey. In addition, most of the studies have focused on invertebrate predators and herbivores or pollinators. Although many plants are pollinated by vertebrates (e.g. hummingbirds), which are vulnerable to several vertebrate predators (Lima 1991), very few studies to date evaluated the outcomes of vertebrate–vertebrate interactions on plant reproduction. Finally, although indirect plant defence has been considered as an important factor in determining plant fitness (e.g. Heil 2008), to date it has not been evaluated comprehensively in meta-analyses of terrestrial trophic cascades. Because our study provides evidence that biotic defences strongly determine the direction of trophic cascades in terrestrial ecosystems, we suggest inclusion of sources of variation concerning plant defences (chemical, mechanic and biotic) in future meta-analyses on trophic cascades. Our general understanding of terrestrial systems would be much improved by including above-mentioned sources of variation.
We thank I. Leal, T. Turlings, G. Bernardello, K. Del-Claro and R. Snell for providing pdfs of their papers and L. Lach and three anonymous reviewers for the constructive comments on the manuscript. G.Q.R. was supported by research grants from FAPESP (04/13658-5) and CNPq (309815/2009-6).