Mutualism with aphids affects the trophic position, abundance of ants and herbivory along an elevational gradient

. Varied interspecific interactions have a tremendous impact on the population dynamics of related species. However, the context dependency of interspecific interactions and their ecological effects have not been well-characterized. To understand the context dependency of ant-aphid mutualism and the corresponding ecological effects, we explored the association of ant abundance, trophic position, and plant herbivory along an elevational gradient. We hypothesized that a higher abundance of ants would be associated with a lower trophic position and lower plant herbivory level along the gradient. We detected a large variation in ant abundance and herbivory along the elevational gradient. Consistent with our prediction, a higher ant abundance was associated with a lower trophic position, and herbivory decreased linearly with ant abundance. A tight positive relationship between the abundance of ants and aphids was observed. These findings suggest that the context dependency of ant-aphid mutualism can be an important factor affecting the pattern of ant abundance and plant herbivory across the elevational gradient. This study highlights the significance of mutualistic interactions in shaping the pattern of certain key ecological processes, such as herbivory, at a relatively larger spatial scale.


INTRODUCTION
Context dependency is a major theme in the study of interspecific interactions (Chamberlain et al. 2014). Varied interspecific interactions can have a tremendous impact on the population dynamics of related species (Maron et al. 2014). To date, the context dependency of interspecific interactions and their ecological effects have not been well-characterized, especially along environmental gradients (Maron et al. 2014). Exploring how interspecific interactions and their ecological effects vary along an environmental gradient is important for gaining insight into a community's response to a changing world (Chamberlain et al. 2014).
The ant-hemipteran interaction is a common mutualistic interaction in nature (Way 1963, Stadler andDixon 2005). In these interactions, ants protect hemipterans against predators; in return, hemipterans offer ants honeydew, which is rich in carbohydrates, as food (Way 1963, Stadler andDixon 2005). Honeydew is an important food source for ants. Recent studies have found that the experimental addition of carbohydrate foods can enhance the activity, aggressiveness, population size, and dominance of ants within a community (Grover et al. 2007, Heil 2008, Gibb and Cunningham 2009, Ness et al. 2009, Byk and Del-Claro 2011, Pringle et al. 2011, Wilder et al. 2011a. Ants are omnivorous insects in the field, and their trophic position can range from pure herbivores to pure predators (Blü thgen et al. 2003, Rico-Gray and. Because honeydew is a plant-based food source, the more dependent that ants are on honeydew, the lower their trophic position will be (Davidson et al. 2003). Recent studies indicate that the extraordinarily high abundance and dominance of ants in certain habitats are closely related to their lower trophic position: These ants depend more on plant-based food sources, such as extrafloral nectaries (EFNs) and honeydew (Davidson et al. 2003, Wilder et al. 2011b). These findings suggest that shifts in mutualistic interaction with hemipterans may have a considerable impact on the population dynamics of ants in nature (Davidson et al. 2003, Wilder et al. 2011b.
Ant-hemipteran interactions have a variety of ecological effects on the host plants as well as on other arthropods feeding on foliage (Styrsky andEubanks 2007, Zhang et al. 2012b). Ants can reduce plant herbivory by consuming or driving off other insects when tending hemipterans, but this anti-herbivory effect is context-dependent (Styrsky andEubanks 2007, Zhang et al. 2012b). Honeydew has a similar effectiveness when it comes attracting ants and reducing herbivory compared to EFNs (Fiala 1990, Chamberlain andHolland 2009). Ant identity and abundance are key factors determining the magnitude of the protective effect; more ants often means greater protection for plants (Moreira and Del-Claro 2005, Frederickson and Gordon 2009, Yamawo et al. 2012. Nevertheless, the protective effect of ants on plants can be saturated as ant abundance increases (Palmer and Brody 2013). If shifts in mutualistic interaction with hemipterans can lead to varied ant abundance, it is reasonable to predict that the herbivory level of the host plants may also be impacted. Trophic position can be used as an indicator for the dependence of ants on the exudates of hemipterans, with a lower trophic position indicating a tighter relationship (Davidson et al. 2003, Wilder et al. 2011b. Several recent studies have found that habitat type, ecological corridor, logging, and species invasion can lead to shifts in trophic position for ants (Gibb and Cunningham 2011, Wilder et al. 2011b, Resasco et al. 2012, Woodcock et al. 2013. However, the shifts in trophic position and corresponding ecological effects of ants along environmental gradients have not been studied previously.
How mutualistic interactions respond to the changing world is one of the questions focused on in ecology (Kiers et al. 2010). Despite the ubiquity of ant-hemipteran interactions in many communities, their macro-scale effects in varied environments have received little attention (Styrsky and Eubanks 2007). Only recently two experimental studies confirmed the negative effects of warming on the ecological effects of ant-aphid mutualism Ives 2014a, Marquis et al. 2014), and another two studies evaluated the effects of ants on plants in large scales (Hanna et al. 2015, Sendoya andOliveira 2015). The effects of ant-aphid interactions under environmental gradients are still far from clear. Elevational gradients provide ''natural experiments'' for exploring shifts in interspecific interactions in varied environments (Korner 2007, Sundqvist et al. 2013, Rasmann et al. 2014. Numerous biotic and abiotic factors vary with elevation, such as biodiversity, temperature, and precipitation; this can lead to varied biotic interactions (Sundqvist et al. 2013). Both ants and their food sources can change with elevation (Samson et al. 1997, Sanders 2002, Hodkinson 2005, Lach et al. 2010. Therefore, it is reasonable to predict that the effects of ants on host plants can also vary with elevation (Miller et al. 2009, Lach et al. 2010. Using aphid-mediated ant-plant interaction as the study system, we aimed to explore the association of ant trophic position, abundance, and plant herbivory along an elevational gradient at Donglingshan Mountain in Beijing, China. Based on our previous experimental work, the aphid-tending ants Lasius fuliginosus can effectively reduce the herbivory level of the host plant Quercus liaotungensis at the study site (Zhang et al. 2012a). We hypothesized that along the gradient: (1) Higher ant abundance should be associated with a lower trophic position but a higher abundance of aphids, and (2) higher ant abundance should be associated with lower levels of plant herbivory.

Study site
The study area, the Beijing Forest Ecosystem Research Station of the Chinese Academy of Sciences, is located at 40800 0 to 40803 0 N and 115826 0 to 115830 0 E. The area has a typically warm, temperate, continental monsoon climate with an average annual precipitation of 500-650 mm and a mean annual temperature of 5-108C. The altitude of most of the area is over 1000 m above sea level and peaks at 2303 m. The zonal vegetation of the Donglingshan Mountain region is a highly heterogeneous, warm temperate deciduous broadleaf forest, and it mainly includes oaks (Q. liaotungensis), mixed tree species (e.g., Tilia spp., Ulmus spp., Acer spp., Juglans mandshurica, and Fraxinus rhynchophylla, among others), birches (Betula spp.), and poplars (Populus davidiana). Some conifers and shrubs (e.g., Prunus spp. and Vitex negundo var. hetertophylla, among others) are also present.

Experimental design
At first, 10 western slopes dominated by Q. liaotungensis were located in the study area. Those 10 slopes form a single elevation gradient that ranges from 1020 to 1770 m and that completely overlaps the distribution range of Q. liaotungensis in the study area (Qi et al. 2009). Then, a transect was set up from the base to the top of every slope in the study area. The width of each transect was 10 m, with the length ranging from 80 to 180 m. Along the gradient, the mean annual temperature decreased about 4.78C, but the mean annual precipitation increased about 84 mm (Qi et al. 2009).
Each transect was divided into 10 3 10 m plots, resulting in a total of 119 plots. Plant herbivory and ant abundance were surveyed in each plot. Based on previous observations, L. fuliginosus and Formica sinensis were the dominant ant species in the study area. L. fuliginosus is a typical honeydew-feeding ant and often nests in the tree holes in the basal part of Q. liaotungensis (Hopkins and Thacker 1999). Workers of L. fuliginosus forage on both the ground and the canopy of Q. liaotungensis. Aphids included Lachnus tropicalis and Tuberculatus sp. in the canopy and Stomaphis japonica on the trunk of Q. liaotungensis (Zhang et al. 2012a). Aphids were the key food source attracting L. fuliginosus in the canopy of Q. liaotungensis at the study site (Zhang et al. 2012a).

Survey of ant abundance
Pitfall traps were used to quantify the abundance of ants in the community in mid-July of 2010 and 2011. July is the middle of the growing season and a highly active period for ants. Pitfall traps are a sampling technique extensively used to sample such surface-foraging invertebrates as ants, beetles, and spiders (Cardoso et al. 2007, Gonzalez-Megias et al. 2008, Wardle et al. 2011). In each 10 3 10 m plot, three traps were set in a triangular shape (with 5 m between vertices). If there were ant nests in a plot, the traps were set at least 2 m away from the nest, to avoid the impacts of nest on the sampling. For each trap, a cup (diameter ¼ 7.9 cm, depth ¼ 9.7 cm) containing 50 ml of an alcoholic solution (5% in concentration) was buried even with the soil surface. Two days after the traps were set, we retrieved the cups, and the samples were taken back to the lab in order to identify the ants to species and the abundance of each ant species could be recorded.

Analysis of ant trophic position
Stable isotope analysis was used to estimate the trophic position of ants. The nitrogen from the tissue of consumers is enriched in dN 15 relative to that of prey, which means that dN 15 is positively correlated with a consumer's trophic position (Post 2002). Ants in the middle portion of each transect, within an area of approximately 10 3 50 m, were collected by hand. At each transect, at least 30 individual ants were collected as a sample, and three samples were collected for each ant species within a transect. For rarer species, we collected as many ants as we could to create a sample. Herbivores (mainly caterpillars) were collected from the canopy by hand and by beating branches (Campos et al. 2006). Both ants and herbivores were immediately taken back to the lab and dried at 608C for at least 48 hours until the weight of the sample stabilized. Thereafter, we first removed the abdomens of each ant to prevent recent stomach contents from influencing dN 15 values (Tillberg et al. 2006, Wilder et al. 2011b; this approach provides more relevant information about the v www.esajournals.org long-term assimilation and incorporation of nutrients from foods into tissues (Tillberg et al. 2006, Wilder et al. 2011b. Each sample was ground with a mortar and pestle, and approximately 1 mg of each sample was packed into a tin capsule for isotopic analysis. Samples were analyzed for dN 15 using a continuous-flow isotope-ratio mass spectrometer (Finnigan Delta V Advantage, Finnigan MAT, San Jose, California, USA). The d-value is expressed as d(%) ¼ (R sample À R standard )(R standard ) À1 3 1000%, where R sample and R standard refer to the molar ratios of the heavier isotope to the lighter isotope ( 15 N/ 14 N) of the sample and the standard molar ratio (which corresponds to atmospheric air). The isotope ratio of herbivores was used as a baseline (trophic position ¼ 2) to predict the trophic position of ants (Wilder et al. 2011b). The trophic position of ants was calculated as follows: trophic position ¼ (d 15a N À d 15h N)/3.4 þ 2, where d 15a N is the value for ants, and d 15h N is the value for herbivores. According to a previous review, the mean enrichment in d 15 N for one consumer trophic-level transfer was set as 3.4 (Post 2002).

Survey of plant herbivory, aphid abundance, and the occupation of food sources by ants
To investigate plant herbivory, we randomly selected three trees of Q. liaotungensis in each plot. On each tree, a twig was cut off from a south-facing branch 4-5 m in height. To prevent leaf-selection bias, the first to sixth leaves were collected from the tip of each twig. To examine aphid abundance, we cut off another twig 4-5 m in height on each tree and recorded the number of aphids found from the tip of the twig to the 10th leaf. For the herbivory analysis, all of the leaves were scanned using an EPSON Perfection 4870 Photo flatbed scanner (EPSON America, Inc., USA) and then used to determine the degree of herbivory. For each leaf, the gaps in the scanned leaf image (herbivore damage) were filled in Adobe Photoshop CS2 (Adobe Systems Inc., USA), according to the expected shape. The area of the original (a) and repaired (b) leaves was calculated using WinFOLIA Basic 2004a (REGENT Instruments Inc., Australia). The percentage of leaf-area loss was calculated as follows: L (%) ¼ (b À a)/b 3 100.
Q. liaotungensis trees are the main host plants for ants in the community; they provide ants with honeydew and other food sources (Zhang et al. 2012a). In the middle portion (approximately 10 3 50 m) of each transect, we surveyed 30 randomly selected trees and recorded whether or not there were ants active on the trees. Observations were conducted between 9:00 a.m. and 4:00 p.m., which is a highly active period for ants, on a sunny day. If ants were active on a tree, we assumed that the tree was occupied by ants. The proportion of trees occupied by ants (PTOA) and the species of ants were recorded. All of the above samples were collected in mid-July of 2011.

Data analysis
The data for herbivory and the abundance of ants and aphids were log-transformed to conform to the assumptions of parametric analysis. Ordinary regression was used to fit the variation in different variables along the elevational gradient. The averaged value of each plot was used to fit the relationship between the averaged herbivory and ant abundance by a linear mixed effect model. To account for possible heterogeneity among transects, different transects were allowed to have varied variances in the model. A Kenward-Roger approximation was used to adjust the degrees of freedom. Model diagnosis was also conducted by fitting residuals with predicted values and testing the normality of the residuals. The relationship between aphid and ant abundances was also analyzed using this model.
Only the ant abundance middle portion of each transect (within an area of approximately 10 3 50 m) was used to fit the relationship between ant abundance and ant trophic position, as the ants used in isotope analysis were collected in this area. Considering the small sample sizes for trophic position and PTOA (with each transect having only one data point), simple linear regression was used to evaluate the relationship between the two variables as well as their relationship with ant abundance. In addition to the linear regression, a randomization test was used to test the relationship between trophic position and PTOA, with a re-sampling of 1000 times. The Kruskal-Wallis test was employed to explore the difference in ant trophic position between transects with only one species and v www.esajournals.org transects in which more than one species had been found. To fully analyze the relationship among ant trophic position, the proportion of trees occupied by ants, ant/aphid abundance, and plant herbivory, a partial correlation among these variables was conducted, with altitude as the partial parameter. All analyses were conducted using SAS 9.2.

Variation in different variables along the elevational gradient
Along the elevational gradient, ant abundance for the two years had a significant positive relationship (R 2 ¼ 85.9%, P , 0.0001); a further analysis found that year had no effect on ant abundance, and neither did the interaction between year and elevation (both P . 0.4). The data for the two years were thus pooled for the analysis. In total, 34 549 worker ants were collected. There was a large variation in ant abundance in different transects: The total ant abundance varied approximately 87 times, and the dominant ant species L. fuliginosus varied approximately 250 times in abundance.
L. fuliginosus contributed 87.18% of the total ant abundance and was found in nine of the 10 transects. The red wood ants, F. sinensis, contributed to 8.25% of the total ant abundance. Several other species were found in very low abundance, including Camponotus japonicus Mayr, Formica fusca Linnaeus, and Tetramorium caespitum. Only in one transect (T3) were no workers of L. fuliginosus collected. In this transect, the dominant species was the highly aggressive red wood ant F. sinensis. This ant species monopolized the transect with a high abundance (mean ¼ 54.24 workers per trap, se ¼ 11.05, n ¼ 49), and no other ant species was collected via pitfall trap sampling.
Ant abundance peaked at middle elevations, and more than half of the variation in ant abundance was explained by elevation (Fig.  1A). For L. fuliginosus, elevation explained 48.22% of the variation in its abundance (Fig.  1B). The pattern of aphid abundance was similar to that of ants, with more than 60% of the variation explained by elevation (Fig. 1C).
Plant herbivory tended to decrease with elevation at first, and then it increased with elevation, with 20.39% of the variation explained by elevation (Fig. 1D). Among the 10 transects, the mean level of plant herbivory varied from 3.89% to 14.99%. The pattern of the proportion of trees occupied by ants was similar to the pattern of ant abundance as well as that of aphid abundance, with a peak at middle elevations (Fig. 1E).
In light of the widespread distribution (in nine of the 10 transects) and dominance of L. fuliginosus along the gradient, only the data for this species were used for analyses related to trophic position. The trophic position of L. fuliginosus ranged from 2 to 3 among different transects, indicating that this species is omnivorous and thus feeds on both plant-and animalbased food sources. The trophic position of L. fuliginosus showed no clear trends with elevation ( Fig. 1F).

Association among ant trophic position, abundance, and plant herbivory
Among the different transects, a higher abundance of L. fuliginosus was associated with a lower trophic position and higher abundance of aphids. More than 20% of the variation in ant abundance at the middle portion of each transect can be explained by shifts in trophic position (Fig. 2). A clear bias in the residuals of the mixedeffect model was detected when the relationship between the abundances of ants and aphids was evaluated. This model was instead tested by linear regression. Aphid abundance explained nearly half of the variation in the abundance of L. fuliginosus (Fig. 2). The trophic position of L. fuliginosus was significantly lower in transects in which only this ant species was found than in other transects in which several species coexisted (Chi-square ¼ 5.4, df ¼ 1, P ¼ 0.02; Fig. 1F). Furthermore, compared to transects in which several ant species coexisted, ant abundance was significantly higher in transects in which only L. fuliginosus was found (F 1,43 ¼ 556.58, P , 0.0001).
The trophic position decreased with the proportion of trees occupied by ants. The proportion of trees occupied by ants explained 49.93% (P ¼ 0.03) of the variation in trophic position (P ¼ 0.032 for the randomization test; Fig. 3). Ant abundance and the proportion of trees occupied by ants also had a tight positive relationship (R 2 ¼ 78.25%, P , 0.0001).
v www.esajournals.org Plant herbivory decreased linearly with the total abundance of ants (F 1,53.2 ¼ 31.34, P , 0.0001), and nearly 20% of the variation in herbivory can be explained by ant abundance (Fig. 4A). The abundance of L. fuliginosus itself could explain 14% of the variance for plant herbivory (F 1,47.6 ¼ 17.01, P , 0.0001, Fig. 4B). The negative relationship between herbivory and ant abundance was more clear at the transect scale: The total ant abundance could explain about 58% of the variation in herbivory among different transects (P ¼ 0.0106 ), while the abundance of L. fuliginosus itself could explain 47% of the variation (P ¼ 0.041; Appendix: Fig.  A1). When the effect of altitude was controlled, the relationship between different variables was v www.esajournals.org still significant except for that between trophic position and herbivory (Table 1).

DISCUSSION
Exploring how biotic interactions and their ecological effects change in varied environments is one of the key focuses of ecological studies (Sundqvist et al. 2013). With this observational study, we show that spatial variation in ant trophic position, abundance, and plant herbivory are closely related. These results suggest that shifts in the mutualistic interaction between ants and aphids can account for the variation in ant abundance and its ecological effect across a continuous environmental gradient.
Our results suggest that shifts in the strength of ant-aphid interactions may be a key factor in  v www.esajournals.org the spatial population dynamics of ants across different environments. In lowland tropical forests, higher ant abundance is associated with a lower trophic position (Davidson et al. 2003).
Our study suggests that for ants, the negative relationship between abundance and trophic position may be a common phenomenon in different terrestrial ecosystems rather than being restricted to tropical areas. In our study, the plant-based foods used were mainly honeydew secreted by aphids (Zhang et al. 2012a). Several previous studies found that feeding on more plant-based foods, such as honeydew, can effectively enhance the growth rate (Wilder et al. 2011a), competitive ability (Heil 2008, Wilder et al. 2011b) and even the immunity of the ant colony (Kay et al. 2014). We found that at the community level, food source occupation and interspecific competition can be closely related to the trophic position and abundance of ants along the elevational gradient. At the three transects monopolized by L. fuliginosus, their trophic position was significantly lower, but their abundance was higher than it was in the other six transects in which several ant species coexisted. These results suggest that a more intimate interaction with aphids can not only enlarge the population size but also can enhance the ants' competitive ability in varied environments.
We suggest that the effect of biotic defense should not be neglected when accounting for the variation in herbivory across larger spatial scales (Moles et al. 2011, Andrew et al. 2012. A previous study found that the leaf traits of Q. liaotungensis varied with elevation in our study area, but none of the patterns were similar to the elevational patterns of herbivory (Qi et al. 2009). Therefore, it is unlikely that the pattern of herbivory is largely determined by leaf traits rather than ants along the elevational gradient. According to the resource availability hypothesis, herbivory should decrease with elevation because in nutrient poor environments (such as higher elevations), plants should invest more in defense against herbivores (Coley et al. 1985), as several recent studies found (Miller et al. 2009, Garibaldi et al. 2011, Metcalfe et al. 2014, Rasmann et al. 2014. However, our results show that plants at middle elevations have the lowest levels of herbivory, and we suggest that the variation in ant abundance may be an important determining factor in this pattern. More ants can provide plants with stronger protective effects (Oliveira et al. 1999, Heil et al. 2001, Yamawo et al. 2014). In our study, at higher elevations, where ant abundance was lower, the mean value of herbivory reached nearly 15%; however, at middle elevations, where ant abundance was higher, the value fell below 6%. For oaks, an herbivory level of more than 8% can significantly decrease the production of fruit (Crawley 1985). We suggest that the weaker protective effect of ants may be an important factor limiting the fitness of Q. liaotungensis at higher elevations. At lower elevations (from 1000 to 1400 m), opposite patterns between ant abundance and herbivory were also detected, suggesting the anti-herbivory effect of ants is also important in this area. We noticed that in general, compared to plants at lower elevations (,1400 m), plants at higher elevations (.1400 m) suffered higher levels of herbivory. The explicit mechanism underlying this pattern is unclear, but it is possible that at higher elevations, plants are more subject to herbivory. This is because in these stressful habitats, plants are limited in their ability to compensate for herbivory in accordance with the prediction of the compensatory continuum model (Appel and Cocroft 2014). Compared to higher elevations, plants at lower elevations may have higher levels of constitutive defense (Rasmann et al. 2014). This can partially explain why both Table 1. The partial correlation between ant/aphid abundance, herbivory the trophic position of ants and the proportion of trees occupied by ants (PTOA); * represents significant correlation (P , 0.05). herbivory and ant abundance were relatively lower in areas with elevation ,1200 m. Recent studies suggest that both warming and reduced precipitation have negative effects on ant-aphid mutualism and the corresponding effects on plants ( Ives 2014a, b, Marquis et al. 2014). In this study, we found there was a closer relationship between ants and aphids at sites at middle elevations, where the temperature and precipitation were also at an intermediate level (Qi et al. 2009). Thus, the suitable environments for ant-aphid mutualism may shift upward in the context of global warming.
In conclusion, through an observational study along an elevational gradient, we found that antaphid mutualism may be an important factor influencing the spatial variation in ant abundance and plant herbivory at a relatively larger spatial scale. The study suggests that the importance of biotic interactions in mediating certain key ecological processes and patterns in varied environments should not be neglected. It should be addressed that the observational nature of this study limits the confirmation of a mechanistic link between different patterns we found. Experimental studies are still needed to pinpoint the specific mechanisms causing the variance of mutualistic interactions in varied environments.