The world and its shades of green: a meta-analysis on trophic cascades across temperature and precipitation gradients


  • G. Rodríguez-Castañeda

    Corresponding author
    1. Department of Ecology and Environmental Science, Umeå University, Umeå, Sweden
    • Ecology and Evolution Department, Stony Brook University, Stony Brook, NY, USA
    Search for more papers by this author

Correspondence: Genoveva Rodríguez-Castañeda, Ecology and Evolution Department, Stony Brook University, 650 Life Sciences Building, Stony Brook, NY 11794, USA.




To assess effects of current global temperature and precipitation gradients on the trophic function of plant–herbivore–predator interactions. Specifically, I study effects of climatic gradients on factors that control herbivore abundances: top-down, bottom-up trophic cascades and plant defences. I include predictions of climate change on shifts in trophic function, under the assumption that temperature and precipitation affect the physiology and performance of plants, herbivores and predators.




A search of the relevant experiments on trophic interactions was conducted using the Web of Science and Scielo databases. Strength of trophic interactions from each experiment was studied by the calculation of the log ratio effect size (Ln R) of the control and experimental means. Each study was georeferenced and mean annual temperature (MAT) and total annual precipitation (TAP) were determined for each study location. Effect size of trophic interaction studies across the world were correlated with these environmental variables.


In total, 387 effect sizes were extracted from the literature. With the exception of bottom-up trophic cascades, trophic interactions and factors controlling herbivore abundance exhibited significant linear or quadratic relationships with either temperature or precipitation gradients: plant growth, predation and the effect of plant defence on herbivores increased with temperature. In contrast, plant growth and herbivory increased with precipitation across ecosystems. Finally, top-down trophic cascades increased towards the extremes of MAT and TAP gradients.

Main conclusions

This study shows climatic gradients not only affect species geographic distributions and physiological tolerance but also the strength of their trophic functionality. This is especially true for the main biotic controls of herbivore populations (i.e. predation, top-down trophic cascades and plant defences). These results suggest future climate change will cause shifts in the strength of trophic interactions, resulting in increased or reduced population control of herbivores across global ecosystems.


Plant productivity, or the shades of green in the world, is limited by abiotic factors such as light, temperature and precipitation, and biotic factors such as herbivory and the biotic cycles that determine nutrient availability. Herbivores and predators are also limited by abiotic factors such as temperature and precipitation (Hodkinson, 2005). The close relationship between plant productivity and strength of trophic interactions was summarized by the Hairston, Smith and Slobodkin (HSS) hypothesis which proposed that herbivores control plant populations and predators in turn control herbivores; thus predation on herbivores increases plant biomass through the depletion of plant consumers (Hairston et al., 1960). However, it is not just predators that control herbivores; the apparent verdant tropical and temperate forests are actually primarily brown by weight (i.e. trees are primarily composed of woody lignin and cellulose) (Polis, 1999). Further, there is also population control of herbivores through plant defences that either limit resources to herbivores or force them to cope with a wide array of defences. Control of herbivores by predators is thought to increase towards the equator (Jeanne, 1979; Dyer & Coley, 2002); however, the traditional notion that control of herbivores by plant defences is stronger in the tropics (i.e. Coley & Aide, 1991) has been recently challenged by a thorough meta-analysis (Moles et al., 2011). However, this study found a latitudinal gradient in which the effect of alkaloids on herbivores was stronger at tropical latitudes.

Biogeographically, evidence suggests that trophic cascades are stronger in aquatic ecosystems (Strong, 1992; Polis, 1999; Halaj & Wise, 2001). In terrestrial ecosystems, Oksanen et al. (1981) argued that predators may be able to regulate herbivores in productive ecosystems but fail to do so in the most unproductive, arctic and alpine ecosystems (i.e. the ecosystem exploitation hypothesis EEH). However, empirical evidence has generated controversy on the EEH hypothesis: Borer et al. (2006) reviewed research on trophic cascades and found that herbivore biomass declined and plant biomass increased in the presence of predators, regardless of the productivity levels of the ecosystem. In contrast, a recent literature review on effects of intraguild predation on the ability of intermediate predators to control herbivores found this relationship was not dependent on latitude but was dependent on primary productivity, hence supporting EEH (Mooney et al., 2010).

Spatial and temporal heterogeneity in abiotic factors, such as climate, can often alter the deterministic outcome of interactions between plants, herbivores and predators (Andrewartha & Birch, 1954; Menge & Sutherland, 1976; Hunter & Price, 1992). The varying effect of climate on the strength of interactions makes it difficult to determine if weak bottom-up or top-down control of trophic interactions is due to a biotic factor or the conditions at the particular location where the interaction is studied (Ovadia & Schmitz, 2004). In addition, when responses of individual species are studied, factors such as plant phenology and metabolic rates are known to change across the species geographic range (Fox & Morrow, 1981; Hodkinson, 2005).

The distribution of herbivore abundance and their ability to outbreak is not necessarily a function of bottom-up and top-down forces but is confounded by the climatic conditions determining variations in plant quality, plant defences and predation at a landscape scale (reviewed by Gripenberg & Roslin, 2007). Despite of all these efforts to understand variation in trophic cascade strength, how global climatic gradients will affect trophic interactions remains elusive.

Global warming is perhaps the most dramatic anthropogenic disturbance to natural ecosystems (Thomas et al., 2004). Effects of climate change have already been documented as shifts in species geographical distribution (e.g. Parmesan & Yohe, 2003; Chen et al., 2009). However, in order to understand the magnitude of the effect and possible biological feedbacks is important to understand how changes in temperature and precipitation will affect the biology of organisms. Predictions of how trophic interactions will change are not straight forward since high thermal tolerance is similar for ectotherms across latitudinal gradients (Addo-Bediako et al., 2000). Thus, unprecedented temperatures in rain forests and the narrow temperature tolerance ranges of tropical organisms increases the risk of biotic attrition in the lowland rain forests (Deutsch et al., 2008; Colwell et al., 2008). In the face of climate change, increasing our understanding of the effects of global climate gradients on trophic interactions is imperative.

Landscape and global level studies of plant–herbivore–predator interactions are rare because of the difficulties in conducting species exclusions simultaneously at different sites. In this case a meta-analysis where location and strength of trophic interactions are recorded and analysed may enable testing for the climatic effects on strength of interactions across broad spatial scales (Cooper, 1998). Here I review publications on trophic cascades published in the past 50 years and georeference each study's location in order to address how global temperature and precipitation gradients affect trophic interactions between plants, herbivores and predators. Further, I ask if the same climatic factors facilitating primary productivity are correlated with stronger trophic interactions.

Materials and Methods

In order to test how trophic interactions affect herbivore distribution change across precipitation and temperature gradients, I reviewed studies that measured effects of resources on plants, herbivory, predation, top-down and bottom-up trophic cascades. Further, to study how climate affects main controllers of herbivore abundance, I included studies on the effects of plant defences on herbivores (Fig. 1; for details on measured variables see Tables 1 & 2).

Figure 1.

Trophic interactions measured in the study: (a) plant response to resources, (b) herbivory, (c) predation, (d) top-down trophic cascades, (e) bottom-up cascades, and (f) plant defences on herbivores. Since most of the effect sizes represent trophic interactions among insects and plants, plants are represented by a legume, plant defences are represented by a skull, herbivores are represented by caterpillars and natural enemies are represented by ants. Direct effects are indicated by a solid line between two trophic levels, and indirect effects (cascades) by a dashed line. A negative effect of one trophic level on another is drawn with a solid dot, and a positive effect is drawn with an arrowhead.

Table 1. Trophic cascade interactions with the predicting and response variables included in the analysis
Interaction measuredPredicting variablesResponse variables included
1. Resources on plant biomassFertilizer, richer soils, sunlight, waterBiomass and growth
2. Plant resources on herbivore biomassFertilizer, richer soils, sunlight, waterBiomass, abundance, growth, developmental time
3. Plant chemical defences on herbivore biomassPlant chemical and mechanical defencesBiomass, survival, developmental time, deterrence/acceptance
4. Herbivores on plant biomassHerbivore exclusions, high versus low herbivore loads, top predator additionsBiomass, growth, damage, quality, reproduction, survival
5. Predators on herbivore biomassPredator exclusions, predator additionsBiomass, abundance, damage, survival
6. Top-down trophic cascadesPredator removal, predator additionsBiomass, damage, quality, survival
Table 2. Number of publications included for each analysis
Interactions measuredPapers included in the meta-analysis (n)
Strength of trophic interactions in tropical versus temperate zones  
a. Resources on plant biomass3020
b. Herbivores on plant biomass5833
c. Predators on herbivore biomass3938
d. Top-down trophic cascades2524
e. Resources to plants and its effects on herbivores2016
f. Plant chemical defences on herbivore biomass4338
Strength of trophic interactions in lowland (L < 1000 m a.s.l.) versus montane (M > 1000 m a.s.l.) elevations nested within tropical or temperate zonesTemperateTropical
a. Resources on plant biomass285166
b. Herbivores on plant biomass4216258
c. Predators on herbivore biomass318299
d. Top-down trophic cascades204223
e. Resources to plants and its effects on herbivores155142
f. Plant chemical defences on herbivore biomass384344

Literature search

I used Thompson Reuters search engine for electronic publications, Web of Science, and searched for papers published between 1960 and 2010. To increase the number of tropical studies, I searched through all issues of Journal of Tropical Ecology and Biotropica and collected papers on trophic interactions. Further, I used the Brazilian search engine Scielo, which contains publications of local research in Portuguese and Spanish. The search was performed using the key words used singularly or in a combination: ‘water’, ‘nitrogen’, ‘shade’, ‘herb*’, ‘pred*’, ‘defens*’, ‘chem*’, ‘bottom-up’ and ‘top-down’.

Subsequently, I searched reference lists of papers obtained for additional studies that measured the effects listed above. Articles on tropical ecology published in regional German, French and Asian journals were excluded due to the lack of access. I initially identified c. 38,000 publications.

Then, I excluded trophic interaction studies conducted on aquatic ecosystems such as streams, lakes and oceans, but included studies on the shores of these environments whenever they involved non-aquatic plants. Since the hypotheses to be tested concerned natural environmental conditions, experimental studies conducted in greenhouse and laboratory conditions and most agricultural and forestry studies were excluded. Studies on below ground trophic interactions were also excluded. In a review of the remaining 700 studies, I excluded 62% of them because they dealt with introduced species, were inconsistent in reporting the results, or did not report sample size and a measurement of variance. Therefore, this study was based on 249 publications from which I extracted 387 effect sizes on trophic interactions (Appendix S1 in Supporting Information).

To maximize generality and to have a wide sample of trophic interactions across the globe different manipulations of resources, herbivores, predators and plant defences were included. Details on how experimental manipulations and response variables were measured are shown in Table 1.

Georeferencing studies

Locations from each published study were recorded and georeferenced with the highest precision possible. When the geographical coordinates of the study location were not published, I used biogeomancer ( in order to locate the study area and obtain latitude and longitude in decimal degrees. Once all the studies were georeferenced, I used the ASTER digital elevation model (DEM) to estimate the elevation. I extracted individual values of mean annual temperature (MAT), and average total annual precipitation across 7 years for total annual precipitation (TAP) for each study by using bioclimatic data at a scale of 2.5 arcmin. Bioclimatic data was downloaded from Worldclim ( All mapping and data extraction was conducted using ArcGIS version 9.3 (ESRI, Redlands, California, USA).

Effect size calculation

From each paper I obtained the mean, sample size and standard deviation of the treatment and the control. Because most studies presented means and standard errors in graphs, I measured figures to the nearest millimetre and determined the mean and the standard error, and finally the standard deviation. Since the meta-analysis includes mixed response variables such as biomass, survival, herbivory and growth (Table 1). I tested for systematic differences in the meta-analysis of using the different response variables singularly at first; I grouped the response variables once I checked that they reacted similarly to the environmental gradients.

In order to include as much information as possible on the differences between the experiments, when available I added information that could also affect the strength of trophic interactions: (1) duration of study manipulation, (2) type of resource manipulated in the study (nutrients, water, light), (3) plant growth form (tree, shrub, grass, liana, herb), (4) whether the response was measured for a single species or the entire guild of herbivores/predators, and (5) whether the herbivores/predators were vertebrates or arthropods. Most of the studies did not include the magnitude of the manipulation, hence it was not possible to standardise and analyse effects magnitude of resource addition and herbivore/predator exclusions had on the correlation between strength of trophic interactions and the environmental variables.

The meta-analysis

Effect size was calculated by using log ratio analysis (Ln R) whenever more than 70% of the database satisfied the sample size/variance requirement; that is, the smaller of the values of either math formula or math formula when greater than three, where math formula, n, and SD denote the mean, sample size and standard deviation, respectively, of experimental group e and control group c (Hedges et al., 1999). In this study, 85% of the values of math formula and 75% of the values of math formula were within the accepted sample size and variance requirement to calculate effect size as the logged difference between control and treatment (Hedges et al., 1999). Publication bias was studied by inspecting funnel plots. I then calculated the effect size and bias controlled confidence intervals by resampling 999 times. Calculations were conducted with the MetaWin software (Rosenberg et al., 1999).

Studies across MAT and TAP gradients

In order to study how the strength of trophic interactions varied across gradients of climatic variables driving primary productivity, I used two approaches: (1) regression of Ln R of each trophic interaction and variance against MAT and TAP; and (2) categorizing MAT and TAP by dividing the data into intervals with even sample size within the environmental variable categories and calculating effect size and bias corrected confidence intervals for each category. Divisions and sample size for each category are included in Appendix S2, Table S2. Then I tested for differences between categories with a random effect model meta-analysis. Finally, I tested whether the effect size differences between categories were linear or quadratic with MAT and TAP. In order to test if other factors could explain or interfered with the relationship between strength of tropic interactions and MAT and TAP, respectively, I conducted a series of analyses of covariance (ancova) with the log ratios extracted from each study as a response variable and geographic location (tropical vs. temperate), categorical MAT and TAP as main factors. Duration of the manipulation, plant growth form, type of resource added, community versus single species and vertebrates versus arthropods were covariates. Statistical analyses were conducted with SAS 9.2.


The oldest experiment was the study of plant response to the manipulation of resources like water, light and nutrients representing 15% of the 249 studies. The most prevalent experiment studied herbivore effects on plants (26%), followed by the study of biotic interactions that control herbivore populations such as plant defences on herbivores (19%) and effects of predators on herbivores (19%). Trophic cascades were the least and more recently studied trophic interactions, top-down (12%) and bottom-up (9%). In total 46% of the predation experiments included in this meta-analysis studied ant predation. Experimental studies of top-down trophic cascades were relatively recent (1980 to 2010) resulting in a smaller sample size (n = 49; Table 1), but sample size between studies conducted in tropical (n = 24) and temperate latitudes (n = 25) was similar. Bottom-up trophic cascades are also conformed of new recent studies (1985–2007); resulting in the smallest sample size (n = 36; temperate latitudes n = 20; tropical latitudes n = 16).

Biases in the literature data

The key words and search methods used biased studies towards a greater inclusion of arthropods (e.g. 82% in the herbivore studies included in this study). Likely the search excluded large vertebrate herbivores since ‘grazing’ was not included as a key word in the search; thus the database mainly includes vertebrate herbivores with small body mass, such as rabbits and rodents. Arthropods also dominated among predation studies (c. 83% of the studies). Moreover, ants represented 70% of the predation studies conducted at tropical latitudes whereas for the temperate zone ants represented only 24%. Studies on parasitoids in natural ecosystems represented only 10% of the predation studies with only one tropical study, thus this meta-analysis does not represent how parasitism rates change with MAT and TAP. Ant bias in the data was also observed in tropical top-down cascades, where 88% of the studies included manipulated ant abundance.

Global distribution of terrestrial trophic interaction studies

Biogeographically, studies on trophic interactions covered most of the globe, but the majority of experiments were concentrated in Europe and the Americas, with few in Russia, India, the Middle East and Africa (Fig. 2). Notably, there was no strong bias towards temperate region experiments in trophic interactions (Table 2). Nevertheless, the majority of experiments were conducted in lowland ecosystems (i.e. 81% of all interactions were from elevations < 1000 m a.s.l.; Table 2). I found a strong quadratic correlation between MAT and the latitude at which the studies were conducted (r2 = 0.79; P < 0.0001; Fig. 2a; range −11 to 28 °C). However, studies with low MAT values (i.e. MAT < 0 °C) were conducted primarily at northern latitudes Fig. 2a. Even though the gradient is not as steep, I also found a strong quadratic correlation between TAP and latitude (r2 = 0.35; P < 0.0001; Fig. 2b; range: 100–4500 mm); studies conducted near the equator had greater variation in TAP than the studies conducted at temperate latitudes Fig. 2b. There was a bias in study design correlated to tropical versus temperate latitudes: the duration of manipulations and plant growth. ancova results showed a significant interaction between adding resources and duration, with temperate latitudes having experiments of greater duration (model F = 5.16; d.f. = 36; P < 0.001, interaction F = 3.42; d.f. = 3; P < 0.03; Appendix S2, Table S3). Differences observed as a response of plant growth to resources between tropical and temperate studies were governed by studies of short term manipulations (0–23 months), with no significant latitudinal difference in plant growth when resources were added for more than 24 months (Appendix S2, Fig. S1). Even though several ancova models were significant, there was no other instance in which the covariates measured interacted with geographical location (Appendix S2, Table S3).

Figure 2.

Geographic distribution of studies of trophic interactions across the world. Black dots depict approximate spatial locations of the experiments included in this study. (a) Correlation between latitudinal degrees N and mean annual temperature (MAT). (b) Correlation between latitudinal degrees N and total annual precipitation TAP. Maps show distribution of the studies and the temperature and precipitation gradients across the globe.

Strength of Tropic Interactions and Environmental Gradients

Regressions using continuous data

Global gradient in mean annual temperature (MAT)

When all studies were included, higher mean annual temperatures were associated with a positive response of plants to resource addition (f (x) = 0.68 + 0.02(x); n = 63; Fig. 3a). The negative effect of herbivory on plants biomass, growth and reproductive output (f (x) = −0.65 − 0.01(x); n = 103; Fig. 3b) was not influenced by the global gradient in MAT. The negative effect of predation on herbivores did increase gradually with MAT (f(x) = −0.06 − 0.05(x); n = 74; Fig. 3c). Top-down trophic cascades had a quadratic relationship with MAT, with stronger indirect interactions in sites with the coldest and warmest annual temperatures (f(x) = −1.43 − 0.16(x) + 6 × 10−3(x)2; n = 49; Fig. 3d). Bottom-up trophic cascades were not significantly affected by global MAT gradient (f(x) = −0.94 − 0.01(x); n = 56; Fig. 3e). Plant defences had a stronger negative effect on herbivores in warm temperatures (f(x) = −0.27 − 0.04(x); n = 60; Fig. 3f).

Figure 3.

Regressions between the global gradient of mean annual temperature (MAT) and the Ln R ± vLn R (log ratio and the confidence intervals calculated from the standard error and sample size of the control and experimental means): (a) plant response to resources, (b) herbivory, (c) predation, (d) bottom-up trophic cascades, (e) top-down trophic cascades, and (f) plant defences on herbivores. R-squared and significance of regression in the top left corner of each graph.

Global gradient in total annual precipitation (TAP)

Response of individual plant species to resource addition increased toward places with higher total annual precipitation (f(x) = 0.48 + 2 × 10−4(x); n = 63; Fig. 4a). Negative effects of herbivory on plant biomass, growth and reproductive output were also stronger in wetter ecosystems (f(x) = −0.59 + 2 × 10−4(x); n = 103; Fig. 4b). Predation had a tendency to increase towards high annual precipitation, but it was not significant (f(x) = −1.07 − 3 × 10−4(x); n = 74; Fig. 4c). Most studies of top-down cascades were concentrated at a TAP range of 0–1000 mm; despite this there was a significant quadratic relationship between top-down trophic cascades and TAP (f(x) = 0.4 − 3 × 10−4(x); n = 46; Fig. 4d). Bottom-up cascades did not have a significant relation with TAP, even though a quadratic function was suggested (n = 56; Fig. 4e). Effect of plant defences on herbivores had no significant linear or quadratic relationship with TAP (n = 49; Fig. 4f).

Figure 4.

Regressions between the global gradient of total annual precipitation (TAP) and the Ln R ± vLn R (log ratio and the confidence intervals calculated from the standard error and sample size of the control and experimental means): (a) plant response to resources, (b) herbivory, (c) predation, (d) bottom-up trophic cascades, (e) top-down trophic cascades, and (f) plant defences on herbivores. R-squared and significance of regression in the top left corner of each graph.

Categorical meta-analysis using random effect model

Global gradient in mean annual temperature (MAT)

The effect size of resource addition on plants when studies added resources to plants for a period of 0–24 months, increased linearly with MAT (Qe = 13.15; Qm = 31.75; Qt = 44.9; slope p (rand) = 0.05; Appendix S2, Fig. S2a). The herbivory effect size was significantly affected by MAT (Qe = 21.77; Qm = 154.7; Qt = 176.46; slope p (rand) = 0.02; Appendix S2, Fig. S2b). Predation effect decreased with MAT (Qe = 36.1; Qm = 191.88; Qt = 227.97; slope p (rand) = 0.02; Appendix S2, Fig. S2c) and top-down trophic cascades showed a linear increase with MAT (Qe = 396.42; Qm = 1391.01; Qt = 1787.44; slope p (rand) = 0.01; Appendix S2, Fig. S2d). Bottom-up trophic cascades were not significantly affected by MAT (Appendix S2, Fig. S2e). The strength of plant defences increased in strength of controlling herbivore biomass with MAT (Qe = 16.21; Qm = 104.18; Qt = 120.39; slope p (rand) = 0.01; Appendix S2, Fig. S2f).

Global gradient in total annual precipitation (TAP)

The effect size of resources on plants increased with TAP (Qe = 25.93; Qm = 62.13; Qt = 91.06; slope p (rand) = 0.01; Appendix S2, Fig. S3a). Herbivory on plants decreased linearly with TAP (Qe = 9.45; Qm = 87.52; Qt = 96.98; slope p (rand) = 0.04; Appendix S2, Fig. S3b). However, the strength of herbivory was also dependent on whether single species or whole herbivore communities were excluded (ANCOVA F = 2.68; d.f. = 94; P = 0.01; effect of single species vs. herbivore community exclusions F = 7.39; P = 0.01; Appendix S2, Table S3). Predation effect sizes increased linearly with TAP (Qe = 33.87; Qm = 250.65; Qt = 284.52; slope p (rand) = 0.05; Appendix S2, Fig. S3c). Top-down trophic cascades effect sizes had a quadratic relationship with TAP (Qe = 596.97; Qm = 812.63; Qt = 1409.61; slope p (rand) = 0.001; Appendix S2, Fig. S3d). However, duration of the manipulation was a significant covariate between TAP and top-down trophic cascades (F = 7.45; d.f. = 42; P < 0.0001; Appendix S2, Table S3), with longer manipulations of top-down control (> 12 months) occurring at 0–2000 mm and the pattern is a linear decrease as TAP increases. Shorter term top-down manipulations (< 12 months) were conducted throughout the TAP gradient with stronger effects of top-down cascades in sites with high TAP (i.e. TAP > 2000; Appendix S2, Fig. S4). Bottom-up trophic cascades had stronger effects at the driest ecosystems (Qe = 66.31; Qm = 179.04; Qt = 245.34; slope p (rand) = 0.01; Appendix S2, Fig. S3e). The strength of plant defences had a quadratic relationship with TAP; but it was not significant (Qe = 2.91; Qm = 88.26; Qt = 91.17; slope p (rand) = 0.65; Appendix S2, Fig. S3f). All relevant regressions and covariates are summarized in Table 3.

Table 3. Summary of the significant regressions between mean annual temperature (MAT) and total annual precipitation (TAP) and a. resources on plant biomass, b. herbivory, c. predation, d. top-down trophic cascades, e. bottom-up trophic cascades, and f. plant defences on herbivore biomass. Placement of the arrows under the climatic extremes (hot/cold, wet/dry) represents direction towards which the interaction increases in strength. Arrow heads represent a positive species interaction increases in strength; bullet heads represent a negative species interaction increases in strength. Black arrows represent a significant correlation between log ratios and the strength of the interaction (Log ratio reg. models); grey arrows represent significant differences between effect sizes (E+) in the random effect meta-analysis (M.p(rand) models) and a correlation coefficients (R-squared) between the categorical effect sizes and the climatic variables (Reg.E+ models). Asterisk signifies there was a category that was different (P > 0.05).


In this meta-analysis, five out of the six trophic interactions studied exhibited significant relationships across MAT, TAP or both climatic gradients (Table 3). Latitude explained 79% of the variation in MAT, (Fig. 2a); hence correlations between strength of trophic interactions and MAT were not independent from latitude. Thus, relationships between MAT and the strength of trophic interactions may be caused by factors covarying with latitude. For instance, tropical interactions may have had longer historical and coevolutionary time than those trophic interactions occurring in temperate ecosystems (Schemske et al., 2009). In contrast, only 35% of the variation in TAP was explained by latitude, with the greatest variation in TAP found at tropical latitudes (Fig. 2b); hence correlations between TAP and strength of trophic interactions were less dependent on latitude than the correlations between MAT and latitude.

Responses of plant growth to resources were stronger in warm and wet ecosystems (Figs 3a & 4a); empirical studies also demonstrate that adding resources to plants consistently resulted in enhanced growth or increased biomass (e.g. Gruner et al., 2008). My finding that plants at a global scale accumulate more biomass in environments with high MAT and TAP is in concurrence with multiple global studies in which primary productivity was found to be a function of both temperature and precipitation (Woodward et al., 1995; Knapp & Smith, 2001). However I also show a plant's responses to added resources depend on the latitudinal location of the studies (tropical/temperate) and the duration of the manipulations (Appendix S2, Fig. S1). This may mean that tropical plants are generally more limited by nutrient and light availability than plants in temperate soils (Chadwick et al., 1999; Reich & Oleksyn, 2004). Warm temperatures were also associated with stronger control of herbivores as suggested by (1) stronger negative effects of predators (Fig. 3c) and (2) stronger negative effects of plant defences (Fig. 3f) at biomes with high temperatures. Stronger herbivore control at higher temperatures, typical of tropical latitudes, provides some insight into the greater dietary specialization of herbivores observed at tropical latitudes (Dyer et al., 2007), since the tri-trophic niche concept states that herbivores, by specializing, would not only cope better with stronger plant defences (Cornell & Hawkins, 2003) but would also gain enemy free space (Singer & Stireman, 2005). Damage by herbivores was highest in wet ecosystems (Fig. 3b and Appendix S2 & S3b) but this effect depended on whether entire herbivore communities or single species were studied (Appendix S2, Table S3). This interaction may be due to a bias in the number of studies conducted with specialist versus generalist herbivores. Single species studies generally concern effects of specialists whereas entire communities of herbivores concern the global damage of herbivores on plants.

Top-down trophic cascades and predation were strongest in ecosystems with MAT > 25 °C and TAP > 2000 mm year–1 (Figs 3 & 4c,d); suggesting that climatic conditions of rain forests, favoured ant activity so that rain forests had the strongest suppression of herbivore damage, and a positive effect on plants. The effectiveness of ants in controlling herbivore damage in these ecosystems may explain why there is a higher proportion of plants investing in nutrient rewards such as food bodies and nectar in rain forests (Rosumek et al., 2009). Interestingly, top-down trophic cascades were also strong in cold and dry ecosystems (Table 3); strong top-down effects may thus not only be a result of high MAT and TAP but also extreme environments harbouring specialist interactions whose exclusion may have stronger top-down effects (Rodríguez-Castañeda, 2009). The interaction between duration of the top-predator exclusions and strength of top-down trophic interactions and TAP (Appendix S2, Fig. S4) call for long term exclusions of top-down predators in places with high annual precipitation so that we can determine whether the strong top-down control observed at TAP > 2000 mm is due to the climatic conditions or the strong immediate response of plants to top predator exclusions.

No climatic or latitudinal trend was found in bottom-up trophic cascades (Figs 3 & 4f), which may be due to the high number of secondary factors that may play a role in this interaction and the low sample size. In fact, bottom-up trophic cascades were stronger in dry ecosystems in which perhaps lower predation, competition and plant defences due to the harsh climatic conditions allow herbivores to take advantage of the effect that adding resources has on plants.

Plant defences increased with MAT (Fig. 3f and Appendix S2, S2f). Even though warmer temperatures enabling plants to invest more in defences to have a stronger negative effect on herbivores cannot be ruled out, it is more likely that stronger effects of defences on herbivore biomass is a result of warm ecosystems containing plants with stronger toxins, such as alkaloids, that negatively affect herbivore biomass (Moles et al., 2011).

Results from this study also provide partial support for the ecosystem exploitation hypothesis, which predicts predation to be highest in sites with climatic conditions favouring high primary productivity (Oksanen et al., 1981) and is consistent with experimental findings of predation being stronger under high primary productivity conditions (Forkner & Hunter, 2000; Gruner et al., 2008). However top-down trophic cascades in this study were also strong the lower end of the primary productivity gradient. Some of the strongest top-down control studies at low temperatures were studies conducted with plant–ant interactions (i.e. Warrington & Whittaker, 1985; Kelly, 1986). Even though, more studies would need to confirm this theory; ant–plant interactions may become more important at extreme temperature/precipitation regimes.

These results have relevancy when thinking about ecosystem changes under future climate regimes. For instance, control of herbivore abundance by predation and plant defences is predicted to increase in warmer conditions (Table 3c,f); this effect will be detrimental to herbivore taxa whose diversity peaks in tropical montane ecosystems such as geometrids (Brehm et al., 2005), where species thrive because of lower predation (Rodríguez-Castañeda, 2009). Experimental studies on the direct effect of temperature and precipitation on animal physiology and its effects on strength of trophic interactions are needed in order to distinguish effects of climatic variation from the effects trophic interactions on plants. For instance, from 2000 to 2009, global plant primary productivity was reduced as a response to increased drought in the southern hemisphere (Zhao & Running, 2010). These results make specific predictions on herbivore and predator performance under dry conditions (Fig. 4b,c); however, the question of what is the relative contribution of NPP versus the physiological effects of drought on herbivore and predator populations remains open.

According to Hillyer & Silman (2010), low seed predation allowed Amazonian trees to migrate upslope. It would be interesting to study the effects of having Amazonian plants with increased plant defences and higher predation rates on the herbivore communities currently living in the upslope Andean forests. In temperate zones, increased trophic interactions as a result of warming may also affect the geographic distribution of species. For instance, the composite Arnica montana is currently restricted in its altitudinal distribution by slug herbivory (Bruelheide & Scheidel, 1999); so that stronger control of herbivory would allow the species to expand its altitudinal distribution. Moreover, changes in strength of trophic interactions will likely affect the distribution of the holly leafminer Phytomiza ilicis since current geographical variation in host plant quality and parasitism rates rank as the main determinants of its geographic distribution (Brewer & Gaston, 2003; Gaston et al., 2004). These are some the studied examples that show the importance of understanding how climate impacts trophic interactions and the geographic distribution of plant species. Further, results from this study show that globally, trophic interactions are strongest in tropical ecosystems where species have high niche conservatism, small geographic ranges (Crisp et al., 2009), and are physiologically sensitive to changes in temperature regimes (Addo-Bediako et al., 2000; Deutsch et al. 2008).

Literature synthesis is the most feasible tool to understand how global climatic gradients affect strength of interactions. However, while literature syntheses can detect hidden correlations, this approach cannot estimate the magnitude of the effects of climatic gradients on the strength of trophic interactions since they are bound to be subject to other driving factors (Cooper, 1998). Therefore these results should be used to guide macroecological experiments that add knowledge to some of the patterns found in this study. A novel aspect of this study was georeferencing the studies from which measurements were taken because trophic interactions observed in individual experiments could then be placed in a spatially explicit framework.

Ideally the effect size of strength of trophic interactions observed under specific environmental conditions could be used to explore how trophic interactions affect the geographic distribution of species and the plasticity of functional traits in plants, herbivores and predators across their geographical distributions. Further, expected climate change and the strong effects of MAT and TAP on trophic interactions make it imperative to incorporate effects on ecological interactions to understand the impact of global warming not only on future biodiversity but also on crop production.


This study would not have been possible without the help of L. Santos-Oliveira, H. Connahs and M. Tobler who have always supported my ideas and research. I would also like to thank M. Blum, L. Dyer, R., L.E. Harding, Jansson, A. R. Hof, D. Jørgensen, M. Jonsson, M. Singer and anonymous referees for their valuable comments at various stages of this manuscript.

Genoveva Rodríguez-Castañeda is a postdoctoral researcher at the Ecology and Evolution department, Stony Brook University, NY. Her main research interests involve determinants of community structure, the macro-ecology and macro-evolution of terrestrial trophic interactions and the biogeographic patterns of the distribution of species.