Bottom-up effects of soil quality on a coffee arthropod interaction web

. Nutrient availability and soil quality influence herbivores through changes in plant traits and can have cascading effects on herbivore interactions. In complex systems, with many positive and negative interactions, the consequences of these bottom-up effects are still not well established. We carried out a set of studies to determine the impact of soil quality (organic compost amendments) on a hemipteran herbivore ( Coccus viridis ), two ant mutualists, predators, pathogens, parasitoids of C. viridis , and an arboreal arthropod community on coffee ( Coffea arabica ). We also determined the impact of Azteca instabilis ants on the arthropod community with an exclusion experiment. In an observational study, the carbon to nitrogen ratio (C:N) of leaf tissue correlated negatively with C. viridis density, however caffeine content did not correlate with C. viridis . In a field experiment with coffee seedlings, both C. viridis and total arthropod abundance were greater on high-quality plants than on low-quality plants. Excluding A. instabilis resulted in higher C. viridis abundance and parasitism rate, and higher spider and total arthropod abundance. Although A. instabilis attendance of C. viridis only marginally differed across soil quality treatments, in a second experiment, Pheidole synanthropica ants recruited more workers per C. viridis individual on high-relative to low-quality plants. Soil quality treatments did not impact predator abundance or fungal pathogen prevalence. These results suggest soil quality impacts C. viridis herbivores, P. synanthropica ants, and total abundance of arthropods on coffee, but did not impact the third trophic level. Thus this study provides a complex case study of pathways in which bottom-up effects influence arthropod interaction webs.


INTRODUCTION
Ecologists have long debated how biotic and abiotic factors control populations and provide structure to ecological communities (Hairston et al. 1960, Hunter andPrice 1992). For many terrestrial systems, nutrient availability can have important consequences on communities through direct effects on the quality and quantity of primary producers (plants), and indirect effects on herbivores that consume plants (Awmack and Leather 2002). Theory also suggests that increased productivity can increase the length of food chains (Fretwell 1977, Oksanen et al. 1981, Abrams 1993. At the same time, the presence of some organisms at higher trophic levels, such as top-predators and mutualists of herbivores, can have strong community wide effects that cascade down to influence plant abundance and quality (Gratton and Denno 2003, Gruner 2004, Kay et al. 2004, Mü ller et al. 2005. It is now commonly acknowledged that both bottom-up and top-down forces work together to shape ecological communities, yet our understanding of how these forces interact with one another is still lacking because systems are often only studied from one point of view (Hunter and Price 1992).
Nutrient enrichment typically increases populations of herbivores that are limited by elements such as nitrogen (N) and other limiting nutrients (Mattson 1980). For instance, plant nutritional traits that are altered by fertilization or soil quality have strong effects on hemipterans because hemipterans feed on plant phloem, which is low in amino acid content (Ponder et al. 2000, Awmack and Leather 2002, Stadler et al. 2002, Morales and Beal 2006, Nowak and Komor 2010. Increases in herbivore density from enrichment can result in increases in predator density and pathogen prevalence (Forkner andHunter 2000, Raubenheimer andSimpson 2009). Nutrient enrichment also impacts other interactions like mutualisms between ants and plants, as well as between ants and honeydew-producing insects (Baylis andPierce 1991, Heil et al. 2002). Ant-hemipteran mutualisms, in particular, structure tropical arboreal communities (Floren et al. 2002, Davidson et al. 2003) and can be mediated by the indirect effects of nutrient availability on host-plant quality (Baylis andPierce 1991, Cushman 1991). Ant abundance increases on plants with high nutrient availability or with high plant quality due to increases in honeydew-producing herbivore abundance (Strauss 1987) or subsequent increases in the quality of honeydew rewards.
Despite this knowledge of nutrient influences on species interactions, many food webs are complicated and complexity may attenuate the effects of nutrient enrichment or productivity. For instance, fertilization can increase defenses in some plants, which can actually make plants more toxic to some herbivores, thereby preventing higher productivity at consumer levels (Dyer et al. 2004). Predators can also exhibit intra-guild predation that may attenuate fertilization effects at higher trophic levels (Letourneau et al. 2009). Additionally, defense mutualists of herbivores can guard herbivores from predators and thereby prevent the third trophic level from becoming more productive. Understanding how complex ecological communities respond to enrichment is therefore a goal of future studies.
In coffee plantations of southern Mexico, the complex interaction web of arthropods on coffee has been well described (Vandermeer et al. 2010). However, few have investigated the bottom-up effects of fertilization, soil quality, or plant quality on the interactions that structure the web. Here, we investigate the direct and indirect influence of soil quality on the interactions of an arthropod community on coffee (Coffea arabica L. [Rubiaceae]) (Fig. 1). First, in an observational study, we examined if coffee plant traits are correlated with the abundance of a honeydewproducing hemipteran (Coccus viridis Green [Hemiptera: Coccidae]) throughout the coffee agroecosystem. Second, in a seedling experiment, we determined whether improved soil quality would increase the quality of coffee host plants, abundance of C. viridis, and in turn the recruitment of Azteca instabilis F. Smith [Hymenoptera: Dolichoderinae] ants, abundance of predators, pathogens, and parasitoids of C. viridis, as well as the overall abundance of the arthropod community. In a second experiment, we tested if soil quality would mediate the recruitment of a second ant-mutualist (Pheidole synanthropica Longino [Hymenoptera: Myrmecinae]) to C. viridis. We hypothesized that increased soil quality would lead to increased plant quality and result in higher densities of C. viridis. We predicted that the presence of A. instabilis ants would also increase C. viridis density because previous studies have suggested a positive relationship between A. instabilis and C. viridis (Vandermeer and Perfecto 2006). We hypothesized that tending ants (A. instabilis and P. synanthropica) would be more abundant on higher-quality plants and have a higher per capita rate of attendance to C. viridis on high-quality plants relative to low-quality plants because of the increased quality of honeydew rewards. We hypothesized that predators and parasitoids would exhibit an interaction between soil quality and ant exclusion treatments, where they would be more abundant with increased soil quality, but only on ant exclusion plants. In the presence of ants, predators and parasitoids would be lower under high-quality treatments because of increased ant guard of C. viridis. We expected that pathogens of C. viridis would be more prevalent in the presence of ants and under increased soil quality, where C. viridis would be more abundant. Finally, overall abundance of other arthropods within the community should also exhibit an interaction between soil quality and ant treatment, where there are more arthropods on high-quality plants, but only in the absence of ants. High-quality plants should provide more resources to host more arthropods, however A. instabilis indiscriminately removes arthropods from foliage (Liere and Perfecto 2008) limiting the effect of soil quality on plants with A. instabilis.

Study system
We conducted all research at Finca Irlanda (15811 0 N, 92820 0 W; 900 m asl; 4500 mm/yr rain; hereafter Irlanda), a shade coffee plantation in the Soconusco region of Chiapas, Mexico in February to July of 2008. There, C. viridis feeds on the phloem of coffee and forms mutualisms with ants, including the dominant arboreal ant, A. instabilis and the ground-nesting P. synanthropica (Vandermeer et al. 2010). Coccus viridis reaches its highest densities near to shade trees with A. instabilis nests (Vandermeer and Perfecto 2006). Many natural enemies attack C. viridis, including parasitoid wasps, adults and larvae of Azya orbigera Mulsant (Coleoptera: Coccinellidae), and entomopathogenic fungus (Lecanicillium lecanii [Zimmerman] Zare and Gams) (Vandermeer et al. 2010). There are also a number of rare or low-density coffee herbivores that include aphids (Toxoptera sp. [Hemiptera: Aphidae]), Lepidoptera, Orthoptera, and Coleoptera. Coffee also produces the xanthine alkaloid caffeine that has insecticidal properties in some herbivores, but weak effects on specialists (Nathanson 1984, Guerreiro Filho and Mazzafera 2000, Magalhães et al. 2008. Research suggests v www.esajournals.org caffeine is induced by C. viridis infestations and it may deter C. viridis in feeding assays (Lemes Fernandes et al. 2011).

Field survey
To investigate the hypothesis that plant quality can impact C. viridis distribution, we determined the relationship between plant nitrogen and caffeine and C. viridis in coffee plantations. We located 11 shade trees housing A. instabilis nests and surveyed nearby coffee plants for C. viridis (n ¼ 38 coffee plants). In each site, we counted C. viridis individuals (both healthy and infected with L. lecanii ) on three randomly selected branches on at least one plant with high densities and one plant with low density of C. viridis to capture variation in density. All coffee plants were within 5m of ant nests. To correlate caffeine content and C. viridis, we collected phloem exudates from 2-3 leaves from each plant using a method modified from King and Zeevart (1974) because C. viridis is a phloem feeder. Following 1 h of darkness, we cut the petiole of one newly expanding leaf and submerged the petiole tip in 5 ml of 20 mmol EDTA for 8 h and then collected the solution for analysis. We also collected newly expanding leaf samples for leaf level caffeine, N, and C analysis.

Soil quality-Azteca instabilis experiment
To investigate hypotheses related to how soil quality impacts the arthropod interaction web, we conducted a factorial designed field experiment with coffee seedlings crossing soil quality with ant exclusion treatments (3 3 2). We obtained 8-month old coffee (C. arabica var. Catuai and Catimor) seedlings (mean 16 leaves) from the Irlanda plant nursery (Feb 2009) and repotted seedlings in pots (11.5 cm diam. by 15 cm height) with one of three treatments: (1) low quality that contained a 3:1 mixture of sand and soil, (2) intermediate quality that contained a 1:2:1 mixture of sand, soil, and organic compost, and (3) high quality that contained a 1:3 mixture of soil and organic compost. We obtained soil from the study site, sand from a nearby riverbank, and organic compost (worm vermiculture of coffee parchment, chicken manure, and calcium carbonate) from the Irlanda composting facility. Weekly, the high-(100 ml) and intermediate-treatments (50 ml) received compost 'tea' (stewed compost). Watering was unnecessary because rainfall saturated potted soil almost daily. We confirmed the quality of soil treatments by running soil analyses (N ¼ 3) at Universidad Autonoma de Chiapas in Huehuetan, Mexico (Appendix A: Table A1). After two months, we infested seedlings with C. viridis by placing them in contact with other heavily infested seedlings. Adults of C. viridis are sessile, but nymphs (crawlers) are mobile and readily colonize new plant tissues (Fredrick 1943). Prior to the experiment, we thinned C. viridis to 30 individuals per plant with cotton swabs. We placed six randomly selected seedlings, two of each soil quality treatment, in random order 1 m from the tree base of 19 selected independent A. instabilis nests on Inga micheliana trees. Then we randomly excluded ants from half of plants (one per soil treatment) by painting Tanglefoot (Grand Rapids, MI) around the stem 3 cm above soil. To control for any effect of Tanglefoot, we painted a half circle of it on control plant stems. We reapplied Tanglefoot and removed encroaching vegetation as necessary. Each week, for five weeks after set-up, we counted the number of C. viridis (excluding crawlers and individuals infected with L. lecanii ), Toxoptera sp., A. instabilis workers, A. orbigera, L. lecanii infected C. viridis (only last 3 wk), parasitized C. viridis (only last 3 wk), and all orders of arthropods on plants; except Diptera because of the difficulty of observing them. All measurements were taken at sites between 6:30 AM and 1:30 PM under similar weather conditions.
To determine the effect of treatments on plant traits we examined plant growth rates by dividing the number of leaves at the end of experiment by the number at the onset of treatments. At the completion of the experiment, we also collected phloem exudates from 2-3 coffee leaves per plant and collected old leaves for caffeine and nutritional analysis as described below.

Soil quality-Pheidole synanthropica experiment
To further investigate the hypothesis that soil quality can indirectly alter the per capita attendance of ants to C. viridis on coffee, we conducted a second field experiment. Upon completion of the A. instabilis field experiment described above, we moved all plants to the outdoor plant nursery v www.esajournals.org and investigated the effect of soil quality on recruitment of P. synanthropica ants to C. viridis. We removed Tanglefoot from all plants, divided them into two areas (blocks) of the nursery (separated by 30 m), and evenly interspersed them in rows (0.25 m apart). Independent colonies of P. synanthropica often occur at small distances (3-4 m; K. Ennis, personal observation), so we assumed the two blocks were foraged by independent P. synanthropica colonies. On two dates (a week apart), we counted P. synanthropica workers on each plant and calculated means across the two sample dates.

Plant trait analyses
To aid our understanding of how soil quality alters plant quality, we measured several plant nutritional and chemical traits. To prepare samples for both leaf caffeine and nitrogen analysis, we dried tissues at 40-458C for 48 h and ground them in liquid nitrogen with a mortar and pestle. We analyzed caffeine using a maximized ultrasonic assisted extraction method at the University of Toledo. Briefly, we combined 0.1 g of prepped leaf sample with 20 ml of methanol and extracted caffeine by ultra-sonication for 4 h at 658C and passed extracts through 13 mm (0.45 lm pores) filter paper. We diluted aliquots with methanol within the linear working range of the instrument. We spiked samples with an internal standard ( 13 C 3 -caffeine) to control for instrumental error. We analyzed samples using liquid chromatography tandem mass spectrometry (LC-ESI-MS/MS, Model Varian 1200L, Agilent, Santa Clara, CA). For the field survey, we compared new leaves; however for the soil quality experiment we used old leaves because there were too few new leaves. We ran each leaf sample in triplicate to control for variability within plant. For phloem caffeine, we filtered un-diluted liquid exudates and analyzed samples with LC-ESI-MS/MS. To determine leaf nitrogen and carbon, we analyzed samples with C-H-N combustion analysis (Perkin Elmer model 2400) at the Application Technology Research Unit at the United States Department of Agriculture's research facility in Toledo, OH.

Statistical analysis
To determine if plant traits influenced C. viridis within the field survey, we performed general linear mixed models (GLMM). Azteca instabilis nests were treated as a random factor within model and the carbon to nitrogen ratio (C:N), and phloem and leaf caffeine concentration were treated as fixed covariates in model. We performed type III F-tests of significance for main effects with restricted maximum likelihood (REML) to estimate the fixed effect parameters and the variance of random effects (West et al. 2007). Non-significant factors were removed from the model and we compared all possible models with Akaike Information Criteria (AIC). We centered the dependent variable (C. viridis abundance) around the mean to improve the fit of GLMM.
To examine treatment effects within the field experiment, we performed GLMM as described above. Specifically, models considered soil quality, exclusion, and sampling date as fixed factors, and interactions between factors. We included nest as a random effect. We corrected for multiple comparisons by performing Bonferroni corrections. To do so, we grouped the arthropod community into four assemblies: mutualists (C. viridis, Toxoptera sp., A. instabilis, and A. instabilis per C. viridis), C. viridis enemies (A. orbigera, parasitized scales (per C. viridis), and L. lecanii infected scales (per non-infected C. viridis), other arthropods (spiders, hemipterans [without C. viridis and Toxoptera sp.], beetles, other ants), and total arthropods (minus ants and hemipteran mutualists). Then we corrected P-values by dividing them by the number of dependent variables compared within each group. To meet assumptions of normality, we transformed all variables with the natural log (variable þ 0.001); except we square root transformed total arthropods. We also compared plant growth rate, leaf C:N, and caffeine (phloem and leaf ) using GLMM as above. To meet the assumptions of normality, we natural log transformed (value þ 0.001) growth rate, and phloem caffeine. We used regression analysis to predict which plant traits were important to C. viridis abundance. Small sample size and lack of adequate plant material resulted in many missing cases and prevented multiple regression analysis from being conducted. Instead, we performed individual regressions between independent variables C:N, leaf caffeine, and phloem caffeine against C. viridis abundance (natural log transformed [þ0.001] for all analyv www.esajournals.org ses).
For the soil quality-Pheidole synanthropica experiment, we compared mean P. synanthropica per plant and P. synanthropica per C. viridis across soil quality treatment with GLMM. Because we were using plants from the previous experiment, we ensured that the previous exclusion treatment did not impact results by including this factor in the model. Area (block) was included as a random effect. We natural log transformed variables (value þ 0.001) to meet assumptions of normality. We analyzed all statistics with SPSS (16.0).

Soil quality-A. instabilis experiment
Hemipteran mutualists and A. instabilis ants.-Coccus viridis was influenced by soil quality, anttreatment, and time ( Fig. 2A, Table 1). Pair-wise comparisons revealed high-quality plants had 64% more C. viridis than low-quality plants (P , 0.001), and 19% more C. viridis than intermediate-quality plants (P ¼ 0.001). There was 61% more C. viridis on ant-excluded plants relative to controls. There was no clear pattern of C. viridis abundance over time, yet pair-wise comparisons revealed lower C. viridis abundance in the second (P , 0.001) and the last time points (P ¼ 0.019) compared with the first. Unlike C. viridis, Toxoptera sp. aphids were not strongly influenced by soil quality, exclusion treatment, or time effects (Table 1).
Azteca instabilis abundance per plant was significantly influenced by exclusion and marginally influenced by soil quality (Fig. 2B, Table  1). Control treatment plants had 26 times more A. instabilis than exclusion treatment plants. Azteca instabilis abundance tended to be higher on high-quality plants relative to low-quality plants. Only the obvious effect of exclusion treatment influenced the number of A. instabilis per C. viridis individual ( Table 1).
Enemies of C. viridis.-Ant exclusion and time impacted enemies of C. viridis, however there was no effect of soil quality treatment (Fig. 3, Table 2). Azya orbigera larvae abundance increased over time on ant-excluded plants, but did not change with time on control plants (Fig.  3A). The proportion of L. lecanii infected C. viridis to un-infected C. viridis was not influenced by soil quality or exclusion treatments, however it increased significantly over the three weeks sampled (Fig. 3B, Table 2). The proportion of parasitized C. viridis to un-parasitized C. viridis varied by exclusion treatment and time, but at no time were more than 3% of C. viridis parasitized, suggesting parasitism was rare (Fig. 3C, Table 2).

Soil quality-Pheidole synanthropica experiment
Pheidole synanthropica recruitment per plant was three times greater on high-quality relative to low-quality plants (P , 0.001) (Fig. 5A, F 2, 106 ¼ 9.8, P , 0.001). Neither high-(P ¼ 0.059) nor lowquality plants (P ¼ 0.098) differed from intermediate-quality plants in terms of P. synanthropica per plant. There was no difference in P. synanthropica recruitment on plants previously assigned to exclusion and control treatments in the A. instabilis experiment (F 2, 106 ¼ 0.25, P ¼ 0.622). This suggests there was no effect of previous experimental legacies (ant pheromones or Tanglefoot residues). There was no correlation between the number of C. viridis per plant and v www.esajournals.org the number of P. synanthropica suggesting that although we did not control for the number of C. viridis per plant this was not an important factor. On high-quality plants there were 3 times as many P. synanthropica workers per C. viridis relative to low-quality plants (P ¼ 0.04) (Fig. 5B, F 2, 104 ¼ 3.9, P ¼ 0.024). The number of workers per C. viridis on intermediate-quality plants did not differ from high-(P ¼ 0.056) or low-quality (P ¼ 0.987) plants.

DISCUSSION
Our hypothesis that C. viridis would increase with soil quality was supported, which suggests changes in soil quality indirectly influenced C. viridis growth. This effect was likely channeled through changes in plant quality or productivity. In the field experiment, high-quality plants had greater growth rates and lower C:N than lowquality plants. Growth of phloem-feeding insects is often limited by amino acid content in the phloem (Ponder et al. 2000, Awmack and Leather 2002, Nowak and Komor 2010, which is correlated with leaf N in some studies (Nowak and Komor 2010). The field survey data also supported this hypothesis because C:N was correlated with C. viridis density across adult coffee plants (Table 1), although there was only a marginal correlation between C:N and C. viridis in the experimental study. Other greenhouse experiments support this finding (Lemes Fernandes 2007). However, unlike in Lemes Fernandes et al. (2011), caffeine concentration was not correlated with C. viridis abundance in our study. Caffeine may not have impacted C. viridis in our study  v www.esajournals.org because variation between treatments (0.4 to 10 mg g À1 leaf caffeine) was low compared to the range (4.2 to 28 mg g À1 ) found in Lemes Fernandes (2007). It is also possible that the presence of predators, pathogens, parasitoids, and ant-mutualists attenuated caffeine effects.
Surprisingly, C. viridis was more abundant on seedlings without ants than on plants with ants, rejecting our hypothesis that C. viridis density would increase in the presence of A. instabilis. Other studies have supported (Philpott et al. 2008) and contested (Reimer et al. 1993; Uno, unpublished data) this finding. The response of C. viridis to exclusion may depend on the given ant mutualist and predators within the community. Lower density of C. viridis on plants with A. instabilis may be due to predation by A. instabilis ants, as has been observed in other ant-hemipteran mutualisms (Offenberg 2001). It also could be that exclusion treatments lowered the abundance of non-flying predators, leading to increased C. viridis densities (Mueller et al. 1988, Piñol et al. 2009). Other combined factors, such as parasitism and infection, may have also contributed to differences observed between exclusion and control treatments.
Our results support our hypothesis that soil quality influences ant-hemipteran mutualisms because Azteca instabilis marginally responded to soil quality (at the per plant level) and P. synanthropica responded strongly to treatments  v www.esajournals.org (at the per plant and per C. viridis levels). For P. synanthropica and C. viridis, this is evidence that soil quality caused an interaction modification, which has important ecological and evolutionary implications ( Wootton 1994, Mooney andAgrawal 2008). Several other studies have demonstrated increased ant recruitment to honeydew-producing insects on fertilized host plants (Baylis and Pierce 1991, Billick et al. 2005, Morales and Beal 2006 or on plants varying by genotype or hybridization (Wimp andWhitham 2001, Mooney andAgrawal 2008). Here, the likely mechanism behind the increased recruitment per C. viridis on high-relative to lowquality plants is the increased N in honeydew that likely limits ants like P. synanthropica (Baylis andPierce 1991, Davidson et al. 2003). There are  v www.esajournals.org several possibilities for why we observed differences between the recruitment in the two ant species. First, the A. instabilis study had much greater variation between site locations and a smaller number of replications were used relative to the P. synanthropica experiment, therefore the former experiment might have had more environmental variation and reduced statistical power to capture treatment effects. Coccus viridis densities on our seedlings may not have been high enough to attract A. instabilis away from other honeydew-producing insects on adjacent plants (Livingston et al. 2008). Differences in ant response to soil quality could also be a result of differences in species dominance, discovery ability, or foraging strategy between the two ant species.
Our hypotheses that soil quality would impact organisms at the third trophic level (C. viridis enemies) and that there would be an interaction between plant quality and ant exclusion treatments were rejected. Nonetheless the predator A. orbigera increased in abundance after the third week on ant-excluded plants. Many of the A. orbigera individuals observed on the ant-excluded plants were early instar larvae, suggesting that they were eggs that hatched after adult oviposition. The higher abundance of A. orbigera on ant-excluded plants is particularly surprising because A. orbigera larvae have waxy extensions that deter ants from removing them, and parasitism of A. orbigera larvae is lower in the presence of A. instabilis relative to the absence of this ant (Liere and Perfecto 2008). The number of C. viridis exhibiting symptoms of infection from L. lecanii relative to the number of healthy C. viridis also increased with time, however there was no differences across exclusion or soil quality treatment. Lecanicillium lecanii was the most important top-down regulating agent of C. viridis in this study. By the end of the final week, there were three times more L. lecanii-infected C. viridis individuals than healthy C. viridis individuals. Although parasitism rates of C. viridis varied by time and exclusion treatment, rates were so low that these results may not be biologically relevant.
Soil quality, exclusion, and time influenced total arthropod abundance, however we did not observe a hypothesized interaction between soil quality and ant exclusion. Arthropod abundance likely increased on high-quality plants because these plants had greater leaf area than lowquality plants, which likely led to more arthropods per plant due to more plant resources per plant. The presence of A. instabilis also led to fewer total arthropods on coffee. Similarly, the number of C. viridis, Toxoptera sp., A. orbigera larvae, spiders, and parasitized scales was also reduced in the presence of A. instabilis. Other studies have demonstrated that A. instabilis reduces the abundance of specific arthropod groups on coffee (Philpott et al. 2008) or on shade trees (Philpott et al. 2004, Gonthier et al. 2010, however this study is the first to suggest that A. instabilis reduces overall arthropod abundance. This finding corroborates the work of others that show ants have community wide impacts on arboreal arthropod communities (Wimp and Whitham 2001, Floren et al. 2002, Kaplan and Eubanks 2005, Styrsky and Eubanks 2007, Rosumek et al. 2009).
Although our soil quality treatments impacted plant quality traits and the arthropod community, increased fertilization many not have been solely responsible. Increased N in the plant tissues of high-soil-quality plants might suggest that fertilization was important, but other soil characteristics, like organic matter, water retention, pH, and the microbial community may have also influenced plant-animal interactions. Dissecting the components of soil quality and evaluating their impact on plant-animal interactions remains an important research avenue with relevance to both ecologists and agricultural producers.
Bottom-up effects of soil quality, nutrient fertilization, and plant quality traits can have far reaching effects on higher trophic levels within ecological communities. Improved plant quality can increase abundance at the second trophic level and in some instances the third trophic level (Forkner andHunter 2000, Gruner 2004). Although we found strong impacts of soil quality on herbivores and some impacts on antmutualists, we did not find any effects on predators, parasitoids, and pathogens. Also, we did not observe interactions between soil quality and ant exclusion despite both treatments having important effects on some components of the arthropod community. Perhaps it is not surprising that our results are complicated given the v www.esajournals.org complexity of the coffee arthropod interaction web studied. Interactions between herbivores and mutualists may have attenuated the transfer of productivity from herbivores to predators or perhaps our study was too short to observe these effects. We conclude that bottom-up effects are important factors to consider within the coffee agroecosystem, but further research is needed in these systems and other systems where there is substantial variation in soil quality. Few other studies have investigated the bottom-up effects of soil quality on arthropod communities in human-dominated ecosystems, such as coffee plantations. Our results will lend important insights into the consequences of changes in soil quality given land-use change, agricultural intensification, nitrogen deposition, and forms of natural variation in soil quality.    v www.esajournals.org