Cascading effects of predator–detritivore interactions depend on environmental context in a Tibetan alpine meadow

Authors


Summary

  1. Studies of grazing food webs show that species traits can interact with environmental factors to determine the strength of trophic cascades, but analogous context dependencies in detrital food webs remain poorly understood.
  2. In predator–detritivore–plant interaction chains, predators are expected to indirectly suppress plant biomass by reducing the density of plant-facilitating detritivores. However, this outcome can be reversed where above-ground predators drive burrowing detritivores to lower soil levels, strengthening their plant-facilitating effects. Here, we show that these trait-mediated indirect interactions further depend on environmental context in a Tibetan alpine meadow.
  3. In our study system, undulating topography generates higher (dry soil) patches interspersed with lower (wet soil) patches. Because the ability of detritivores to form deep burrows is likely to be limited by oxygen availability in low patches (wet soil), we hypothesized that (i) burrowing detritivores would undergo a vertical habitat shift, allowing them to more effectively avoid predation, in high – but not low – patches, and (ii) this shift would transmit positive effects of predators to plants in high patches by improving conditions in the lower soil layer.
  4. We tested these hypotheses using complementary field and glasshouse experiments examining whether the cascading effects of above-ground predatory beetles (presence/absence) on the density and behaviour of tunnel-forming detritivorous beetles, soil properties, and plant growth varied with patch type (low/high).
  5. Results revealed that predatory beetles did not reduce the density of detritivores in either patch type but had context-dependent trait-mediated effects, increasing the tunnelling depth of detritivores, improving soil conditions and ultimately increasing plant biomass in the high but not low patches.
  6. This study adds to an emerging predictive framework linking predators to plants in detritus food webs, demonstrating that these indirect interactions depend not just on the relative habitat domains of predators and prey, but also on environmental conditions that can predictably constrain the behavioural response of detritivores to predation risk.

Introduction

Since the emergence of the green-world hypothesis (Hairston, Smith & Slobodkin 1960), understanding the functional role of predators has been a major goal in ecological research (e.g. Oksanen et al. 1981; Paine 2002; Schmitz 2008; Griffin et al. 2011). Predators are now well known to indirectly facilitate plant growth in various living plant-based ecosystems by regulating the strength or nature of herbivore–plant interactions (e.g. Schmitz, Hambäck & Beckerman 2000; Silliman & Bertness 2002). In contrast, predators in tri-level detritus food webs are assumed to negatively impact plant growth by suppressing the density and/or activities of detritivores that usually promote soil-nutrient properties for plants; dubbed the ‘brown-world’ role of predators (Moore et al. 2003; Allison 2006; Wu et al. 2011). Such ‘trophic cascades’ result from predator suppression of intermediate consumer density (‘density-mediated indirect interactions’; DMIIs; Hairston, Smith & Slobodkin 1960; Oksanen et al. 1981) and/or predator-induced changes in intermediate consumer traits (e.g. morphology and behaviour; ‘trait-mediated indirect interactions’; TMIIs; Schmitz, Krivan & Ovadia 2004; Griffin et al. 2011), where predators, intermediate consumers (herbivores or detritivores) and plants serve as initiators, transmitters and receivers of the indirect interactions, respectively (sensu Pearson 2010).

The nature and strength of trophic cascading effects have been demonstrated to be largely context-dependent (Hunter & Price 1992; Strong 1992; Spooner, Vaughn & Galbraith 2012). One important source of context-dependency is variability in the traits of species involved. Traits such as the hunting mode of predators, the habitat domains of predators and intermediate consumers, and the palatability of producers are all known to affect trophic cascades (see review by Berlow et al. 2004; Borer et al. 2005; Preisser, Orrock & Schmitz 2007; Schmitz 2008; Spooner, Vaughn & Galbraith 2012). Physical abiotic conditions (e.g. temperature and precipitation) may also strongly influence trophic cascades, by differentially affecting the growth rates of species across trophic levels (e.g. Bridgeland et al. 2010) and by modifying the strength of interactions between trophic levels. As examples of this latter point, experimental warming has been shown to strengthen cascades in aquatic systems by differentially increasing the metabolic demand of intermediate consumers and predators relative to producers (O'Connor 2009; Kratina et al. 2012), and in grasslands, warming was found to modify a TMII by triggering a shift in predator habitat domain (Barton & Schmitz 2009). Notably, however, our understanding of context dependency of trophic cascades is predicated almost exclusively on examples from grazing components of food webs (e.g. Schmitz 2008; Barton & Schmitz 2009). Despite the acknowledged importance of detrital components of food webs to ecosystem functioning (Moore et al. 2004; Hagen et al. 2012), little is known about whether and how species traits and abiotic factors affect trophic cascades in these food web components, particularly those cascades linking predators to plants via detritivores (but see Moore et al. 2003; Wardle et al. 2004; Zhao et al. 2013).

Trophic cascades in detritus food webs appear to be highly context-dependent (Moore et al. 2004), and recent work suggests that species traits may play an important role in explaining and ultimately predicting this context dependency. Specifically, in Tibetan alpine meadows, we initially found evidence in support of a ‘brown world’ predator role: above-ground predatory beetles indirectly decreased plant growth by suppressing the density of dung-dwelling coprophagous beetles and reducing nutrient availability (Wu et al. 2011). However, in a subsequent study (Zhao et al. 2013), we found that above-ground predatory beetles actually promote plant growth via their effects on earthworms (see also Nichols 2013). Predatory beetles did not suppress earthworm density, but instead induced their movement to lower soil layers where their positive per capita effects on plants were stronger. Accordingly, the nature of above-ground predator – detritivore interactions and resultant cascades appears to be affected by the traits of the detritivore species (i.e. habitat domains of above-ground vs. below-ground detritivores). Consumptive effects and negative DMIIs are thus expected to dominate for above-ground species, and nonconsumptive effects and potentially positive TMIIs expected to dominate for below-ground detritivores in the Tibetan alpine meadows.

Cascading effects of predators in detritus food webs may also depend on soil condition, which has the potential to modify interactions in the predator-detritivore-plant pathway. For example, the ability of burrowing detritivores to undergo a vertical habitat shift and seek refuge in lower soil layers could be constrained by physical soil conditions (e.g. highly compacted or wet and anoxic soil), in turn both leaving them more vulnerable to predation and preventing the emergence of a positive TMII. Furthermore, detritivore-mediated changes in soil conditions may not necessarily translate into enhanced plant growth; for example, increased soil nutrients will not benefit plants if soil nutrient use is indirectly limited by other factors, such as soil oxygen (Gerard, Sexton & Shaw 1982; Chapman et al. 2012). Soil conditions can vary widely within and among ecosystems, resulting from factors such as climate, vegetation and small-scale variation in topography and drainage (Lavelle & Spain 2001). Elucidating how cascades in detritus food webs vary with soil conditions is, therefore, an important step towards a general predictive framework.

The eastern part of the Tibetan plateau is characterized by extensive alpine meadows. These meadows have undulating topography, creating widespread small-scale variation in soil conditions, with high (dry soil) patches interspersed within low (wet soil) patches. Plant growth in the meadows is often nutrient limited, as demonstrated for the high patches in our previous studies (Wu et al. 2011; Zhao et al. 2013); it can also be limited by low soil oxygen, particularly within the low patches (Cai & Jin 1963). Thus, deep burrowing of soil macrofauna that enhance soil nutrient status and oxygenation may facilitate plant growth in both high and low patches within meadows (Cai & Jin 1963; Zhao et al. 2013).

We focus here on interactions between the predatory beetle species Philonthus purpuripennis (i.e. Philonthus rubripennis in Wu et al. 2011) and the tunnelling (i.e. burrowing) coprophagous beetle Onthophagus yubarinus. Philonthus has been previously found to drive a negative DMII via an above-ground dung-dwelling coprophagous beetle (Aphodius erraticus), but the cascading interactions between Philonthus and Onthophagus have not been previously studied. Based on the habitat domains of Philonthus and Onthophagus and developing theory (Zhao et al. 2013), we hypothesized that Onthophagus would undergo a vertical habitat shift to deeper soil to avoid Philonthus, in turn improving soil conditions and increasing plant growth (a positive TMII). Previous work suggests soil organisms are less able to burrow in wet soil due to oxygen limitation (Collis-George 1959; Sparks & Strayer 1998), we therefore further hypothesized that the positive TMII would be context-dependent – that is, the behavioural response of the tunnelling beetles to predation risk would be constrained in the low patches and a positive TMII would be more evident within high patches. Moreover, in the low patches, the lessoned ability of Onthophagus to seek a refuge from predation could leave it more vulnerable to attack from Philonthus, potentially reducing Onthophagus density and further weakening its positive effect on plants. To test these hypotheses, we factorially manipulated predators (presence/absence) in the field under two different environmental contexts (high, dry soil patches and low, wet soil patches) in an alpine meadow. After a growing season of the field experiment, we quantified detritivore density, dung mass loss, above-ground plant biomass, and soil physical and chemical properties in response to these treatments. Furthermore, we also performed two complementary glasshouse experiments that allowed examination of detritivore behaviour in response to the same factorial treatments.

Materials and methods

Study Background

This study was conducted in Hongyuan Alpine Meadow Research Station (32°48′N, 102°33′E), Sichuan province in the eastern Qinghai-Tibetan Plateau. Further details of the site and plant community can be found in previous studies (Liu et al. 2011; Wu et al. 2011).

The meadow has been mainly grazed by yaks (Bos grunniens) for decades (Xiang et al. 2009). Possibly because of high density of yak dung (up to 5900 pats per hectare; Wu & Sun 2010), the decomposer community responsible for dung removal is diverse and complex, as in other areas (e.g. Mohr 1943). Coprophagous beetles are the most effective detritivores, while predacious beetles are the dominant predators in this decomposer community (Wu & Sun 2010). The alpine frog (Rana kukunoris) often feeds on the beetles in low-lying grassland but typically not in the study meadows. According to feeding behaviours, coprophagous beetles are usually classified into the following four species groups (Doube 1990): tunnellers (paracoprids), species that dig channels under dung pats and transport dung particles into the tunnels for feeding and breeding; rollers (telecoprids), species that roll dung balls away at various distances from the dung pats; dwellers (endocoprids), species that feed and breed within the dung pat; and kleptoparasites (kleptocoprids), species that use dung stores already buried by tunnellers or rollers. In our study area, the two most abundant beetle groups are dwellers including A. erraticus, Aphodius rectus, Aphodius rusicola, Aphodius edgardi, Aphodius frater, and tunnellers including O. yubarinus, Onthophagus tabidus, Germarostes sp. Predaceous beetles include P. purpuripennis, Platydracus sp. and Sphaerites sp.

In this study, we focus on the indirect effects of predaceous beetles on dung removal rate, soil properties and associated plant growth rate. We established a three-level interaction chain of predator–detritivore–plant, where the intermediate detritivore was the most abundant tunneller beetle, O. yubarinus, and the predaceous beetle was one of the dominant predator species in the meadow, P. purpuripennis (Wu et al. 2011). The tunneller species O. yubarinus (15·1–19·4 mm in adult body length) is active from June to October in dung pats and account for more than 10% of the total number of coprophagous beetles. P. purpuripennis is a large (14·6–15·8 mm in body length) and strong predaceous species, making up about 15% of the carnivorous beetles in abundance. We (X. Wu and S. Sun) observed in the field the predator directly attacking the tunneller adults, splitting the body into two parts at the neck and subsequently eating the abdomen part, with only the hard shell and wings remaining. An independent survey, in which 30 artificial and fresh dung pats (1- to 2-day old, 17 and 5 cm in diameter and height, respectively, about 1000 g in fresh weight) were examined, showed that there were averagely three and eight adult individuals (per pat) of P. purpuripennis and O. yubarinus, respectively.

The Primary Experiment

Experimental design

In a fenced plot of 1 ha, we conducted an orthogonal two factor-two level experiment in which we crossed predator presence (present vs. absent) with patch elevation (high vs. low), yielding a total of four treatments: (i) high elevation, predator absent (H, −P), (ii) high elevation, predator present (H, +P), (iii) low elevation, predator absent (L, −P), and (iv) low elevation, predator present (L, +P). Each treatment consisted of five replicates, each replicate having one dung pat and eight tunneller beetles, with or without three predator individuals (see Fig. S1 in Supporting information).

In addition, we established a control that did not include either tunnelling beetles or predatory beetles and was otherwise treated exactly the same as the experimental plots. This allowed us to establish the effects of tunnelling beetles on plant growth in the absence of predators and examine whether this varied across low vs. high patches. Control replicates were spatially interspersed with the other treatments and replicates located on both low- (= 5) and high- (= 5) elevation patches.

The high- (dry soil) and low- (wet soil) elevation patches were located on the crest and trough, respectively, of natural undulations in the topography of the meadows and had a difference in elevation of c. 2 m. Both types of patches were dominated by graminoids including sedge (Kobresia setchwanensis, Blysmus sinocompressus, and Kobresia pygmaea) and grass species (Elymus dahuricus, Festuca rubra, Deschampsia caespitosa and Agrostis matsumurae), and forbs (Saussurea nigrescens, Geranium wilfordii, Lancea tibetica, Gueldenstaedtia delavayi and Potentilla anserina), but forbs were more dominant (by c. 20% in cover) in high patches and graminoids more dominant (by c. 20% in cover) in low patches. Each replicate was located within a single patch; any two patches were separated by at least 2 m. Soil water content (0–20 cm of soil layer) measured using TDR (Time Domain Reflectometry, Midwest-G, JZZ1-TDR-3, Peking, China) in a typical low and high patch at the beginning of the experiment was 53% and 35% (volume), respectively. Furthermore, measurements over a 10-day period during the experiment showed that soil moisture at 20-cm soil depth was consistently higher in the low treatment (54%) than in the high treatment (36%; see Fig. S2 for details).

Thirty days before the beginning of the experiment, half of 30 circular soil cores (30 cm in diameter and 20 cm in depth) including plants were carefully removed from the low patches, and another half from the high patches randomly. Each circular core was dug out as a whole after creating a circular trench around the soil core to minimize the impact during the digging process. Plant roots are usually less than 20 cm in depth and are very dense in the meadow, such that the whole system was almost intact. Subsequently, the circular core was tightly covered by steel screen (0·3 mm thick), with mesh size of 0·5 × 0·5 mm, which was small enough to prevent escape of the beetles and entry of other macro- or meso-faunas, while still allowing water to drain. Yet, the cylinder net was sufficiently high (30 cm) for plant growth (see Fig. S3 for the structure). Then, the 30 caged chambers were placed into 30 holes (approximately the same diameter of the core, 20 cm in depth), which were randomly dug in two types of distinct vegetation patches when the soil cores were removed. Half of the holes (that were removed from high patches) were randomly assigned to the high-elevation patches and the other half to low-elevation patches. Each chamber was fumigated to remove soil animals using 40% chlorpyrifos [O, O-diethyl O-(3, 5, 6-trichloro-2-pyridyl) phosphorothioate], which has been shown to effectively kill soil fauna while having no apparent adverse effects on plant growth (Wu et al. 2011).

Dung was collected in the early morning (before 07:00 a.m.) on 5 June 2012 from fresh droppings by yaks in a stall of a Tibetan family, such that the dung was free of beetles and other macro-decomposers. The dung was thoroughly mixed in a big bucket and then was divided into individual pats using a circular metallic mould. The pats were 17 cm in diameter and 1000 g (± 20 SE) in fresh weight (c. 5 cm thick). Beetles were collected by hand until enough individuals were captured to conduct the experiment. We selected only the medium-sized beetles of each species for the experiment. In addition, eight individuals of O. yubarinus and three individuals of P. purpuripennis were placed in the centre of each cage in predator-present treatments, and eight O. yubarinus were added to the predator-absent treatments. No beetles were added to controls.

Data collection

The experiment began on 5 June and ended on 5 September 2012. At the end of the experiment, the remaining dung was collected and the above-ground plant parts occupying a circular belt of 10 cm (from the edge of dung pats to the edge of cages; except those that senesced under the original dung pats) were harvested from each chamber. During the harvest, all plants were assigned to two species groups: graminoids (including sedge and grass species) and forbs. We separated the soil in the chamber into two halves: the upper layer (0–10 cm depth) and the lower layer (10–20 cm depth). Three soil cores were sampled for each chamber to measure soil bulk density, soil organic matter and nutrients. For each layer, the soil was manually smashed and sieved, and plant root and beetles (including the living adults and larvae, if any) were picked out and counted. Subsequently, the soil was mixed and sampled (more than 500 g), and transported to the laboratory along with the biological materials. In addition, all the soil cores that were sampled for bulk density measurements were also separated into two layers, and hereafter, the beetles (including the living adults and larvae, if any) were sorted out and counted, and the soil placed in containers and taken to the laboratory. All the replicates were harvested and sampled within 24 h for each treatment. For the controls, the above-ground plant parts (also assigned to graminoids and forbs species groups), but not soils, were sampled due to logistical constraints. Because this was upon termination of the experiment, it did not confound comparisons of plant biomass across control and experimental plots.

The residual dung, above-ground parts of plants, as well as plant roots, were dried at 75 °C for 48 h and weighed. Dung mass loss was calculated by the initial dung dry weight minus the residual dry weight. The soil samples were weighed for each soil layer immediately after being taken to laboratory, and then, they were dried at 105 °C for 48 h and weighed for the calculation of soil water content. Soil bulk density was calculated as the dry mass of the soil core. Moreover, for each soil sample, total N and P were determined by the Kjeldahl method and spectrophotometric colorimetry (Unicam-200; Unicam, Cambridge, UK), respectively; soluble N and P concentrations were determined using the alkaline KMnO4 method and 0·5 M NaHCO3 (pH 8·5) solutions, respectively.

Complementary Experiments

The field experiment lasted for about 3 months, and any destructive sampling would negatively influence plant growth within the chambers. This makes it impossible to accurately estimate the predator effect on the behaviour of the prey species during the experiment, although this is critical to interpreting how the predators potentially affect plant growth. We therefore conducted two complementary glasshouse experiments, with the aim of clarifying the behavioural response of the detritivores to predators.

The first glasshouse experiment included the same four experimental treatments as the field experiment. Each treatment was replicated (= 5) within individual experimental mesocosms (total = 20). Each mesocosm consisted of two cylinders: a small one (20 cm in diameter and 40 cm in height), which was an aluminium bucket serving to contain experimental materials (i.e. soil core, dung pat and beetles); and a large one (40 cm in diameter and 50 cm in height), which was made of steel sheet (mesh size was 0·5 × 0·5 cm, small enough to prevent beetle escape) and placed outside the small one (Fig. S4), resulting in a space between the large and small cylinders. The experimental beetles might emigrate from the small container falling into the space, where they could be counted. The mesocosm was similar to the set-up used by Sowig (1995).

Twenty soil cores matching the size of the inner container were collected from the study meadow. Half (= 10) were collected from low-elevation sites and half (= 10) from high-elevation sites; collection sites were interspersed with the experimental patches to ensure the soil conditions represented those of respective patch types in the field. Cores were added to small containers immediately after they are moved from the field to the glasshouse. After removing the above-ground plants and the surface litter, we maintained the soil moisture differences evident in the field by watering mesocosms differentially. At the end of the experiment, the measured soil water content was 49% (± 1·6 SD) (volume) and 33% (± 1·6 SD) in the low- and high-elevation treatments, respectively; these levels were comparable with those in the field experiment. After the equipment and soil cores were installed in the glasshouse, 1000 g fresh dung and eight tunneller beetles were placed on the central part of all the 20 soil cores, and three predator individuals were additionally added to each chamber of the predator-present treatments.

We inspected the steel cage twice a day to remove and to count the trapped beetles that had emigrated from the pat. When more than half of beetle individuals had emigrated from a chamber, we terminated the observation for that chamber (the experimental duration was about 15 days on average). We took out the soil core and then recorded the number of tunnels and measured the depth of each tunnel; we also collected the dung from the tunnels, which was then dried and weighed as ‘tunnel dung weight’ (i.e. mass of the dung used for the beetle brooding in the tunnels, see Sowig 1996). In addition, the living beetles were also collected and counted to assess possible consumptive effects.

Because the soil cores used in the first glasshouse were removed from different types of habitats (low vs. high patches), plant species composition and hence the associated root density could be different, which would confound the effect of soil moisture on detritivore behaviours. To remove this potentially confounding factor, we conducted a second glasshouse experiment from 14 to 27 June 2013, experimentally creating low and high soil moisture conditions (through differential watering) while using the cores from only a single type of habitat: high-elevation patches. The experiment equipment and sampling methodology were identical to the first glasshouse experiment. Soil water content at the end of the experiment was 49% (± 1·1 SD; volume) and 36% (± 1·0 SD) in the low- and high-elevation treatments, respectively.

Data Analyses

We determined the effects of patch elevation (low/high), tunnelling beetles (presence/absence), and predators (presence/absence) on above-ground plant biomass in the primary experiment using a general linear model (GLM), in which the biomass was set as dependent variable and the independent terms included all the three factors and the interactions between patch elevation and tunnelling beetles and between patch elevation and predators. As the remaining response variables were not assessed in the control, we used two-way analysis of variance (anova) to determine the effects of patch elevation, predators and their interactions on dung mass loss, and soil properties (upper layer and lower layer, respectively). Similarly, for the two glasshouse experiments, two-way anova was used to examine the effects of patch elevation and predators on mean tunnel depth and tunnel dung weight. Dung mass loss, soil organic matter content and soil N and P concentrations in the field experiment and tunnel dung weight in the glasshouse experiment were log-transformed to achieve homogeneity of variance. The effects on number of living tunnellers in the field experiment and on tunnel number in the glasshouse experiment were determined using analogous two-way models with Poisson errors; that is, generalized linear models (GLM). In all the experiments, once a significant effect was detected, post hoc Tukey's tests were used to determine the significance level of the difference between the treatments. In addition, linear regression analyses were conducted to determine the relationships among soil bulk density, water content, soil soluble nitrogen concentrations and plant biomass accumulations in the field experiment. All the data analyses were conducted by r 2.14.1 (R Development Core Team 2011).

Results

The Primary Experiment

The presence of tunnelling beetles enhanced plant biomass (Controls vs. – P treatments in Fig. 1; = 0·001; tunnelling beetle effect in Table S1). This effect did not depend on patch elevation (tunnelling beetle × patch elevation = 0·109; Table S1). In support of our main hypothesis, the effect of predators on plant biomass depended on patch elevation (predator × patch elevation = 0·010; Table S1). Specifically, the effect of predators was greater within the high-elevation patches; indeed, predators significantly increased plant biomass within the high (Tukey's test, = 0·001) – but not the low (Tukey's test, = 0·337) – patches (Fig. 1). There was no evidence that this effect was driven by a reduction in tunnelling beetle density within low patches. In fact, neither predators, patch elevation nor their interaction affected the number of living adult tunneller beetles recovered at the end of the experiment (Table S2; Fig. 4b). This suggests that in both low and high patches tunnellers were able to effectively avoid being attacked and that observed effects of predators on plant biomass were not density mediated.

Figure 1.

The difference in above-ground plant biomass among the four treatments [(i) high elevation, predator absent (H, −P), (ii) high elevation, predator present (H, +P), (iii) low elevation, predator absent (L, −P), and (iv) low elevation, predator present (L, +P)], as well the control (i.e. no detritivores or predators) at high elevation (H, Control) and at low elevation (L, Control), at the end of the field experiment. The different letters above the error bars denote that the difference was statistically significant at the level of = 0·05, as revealed by GLM followed by Tukey's test for multiple comparisons. Sample size was five for all the treatments. Note that the difference was marginally significant (= 0·06) between treatment ‘L, −P’ and its control (L, Control).

Despite above-ground plant biomass effects, treatment effects were non-significant on root biomass (Table S2; Fig. S5). The ratio of graminoid biomass/forb biomass was significantly higher in low patches than in high patches reflecting known compositional differences; importantly, however, the graminoid/forb ratio was not affected by the presence of predators (Table S2; Fig. S6), suggesting that both plant groups did not differentially respond to predator presence. Corresponding to the plant biomass pattern, soils in the low patches were characterized by higher water content but lower soluble N and P (in both upper and lower layers) relative to the high patches (Table S2; Fig. 2). In the high-elevation patches (but not the low patches), predators significantly decreased bulk density and increased water content and soil soluble N (but not P) at the lower soil layer but not at the upper layer (Fig. 2). Such predator effects are consistent with our hypothesis that the predator-induced shift of detritivores to lower soil layers would occur most strongly in the high patches (see Complementary experiments below). However, predators had no significant effects on soil total N and P contents and soil organic matter content (Fig. S7).

Figure 2.

The difference in soil properties including (a) soil bulk density, (b) water content, (c) soluble nitrogen and (d) phosphorus concentrations for both upper and lower soil layers among the four treatments [(i) high elevation, predator absent (H, −P), (ii) high elevation, predator present (H, +P), (iii) low elevation, predator absent (L, −P), and (iv) low elevation, predator present (L, +P)] at the end of the field experiment. The different letters above the error bars denote the difference among treatments (but not between layers) was statistically significant at the level of = 0·05, as revealed by two-way anova followed by Tukey's test for multiple comparisons, respectively. Sample size was five for all the treatments.

Further analyses elucidated the links between predator-induced changes in soil properties in the lower soil layer of the high-elevation patches and ultimately plant biomass within these patches. At the lower soil layer of the high-elevation patches, bulk density was significantly and negatively correlated with soil water content, which was significantly and positively correlated with soil soluble nitrogen concentration that strongly affected plant biomass (Fig. 3). None of these relationships were evident within the low elevation patches.

Figure 3.

The regression relationships showing the effects of (a) soil bulk density on water content, (b) of water content on N availability, and (c) of N availability on above-ground plant biomass in lower soil layer among the four treatments [(i) high elevation, predator absent (H, −P), (ii) high elevation, predator present (H, +P), (iii) low elevation, predator absent (L, −P), and (iv) low elevation, predator present (L, +P)] at the end of the field experiment. The decision coefficient (R2) and statistical test (F and P values), as well as sample size (N), are provided for the relationships that are statistically significant.

Nevertheless, dung mass loss was significantly higher within low compared with high patches and predators slightly decreased the dung loss (Table S2; Fig. 4a), but there was no interaction between these factors (Table S2). This suggests that the improved soil conditions and associated increased plant biomass in the predator present and high-elevation treatment cannot be attributed to increased dung accumulation but potentially to the altered habitat domain of the tunnelling detritivores (see the Complementary experiments).

Figure 4.

The difference in (a) dung mass loss and (b) number of living tunnellers among the four treatments [(i) high elevation, predator absent (H, −P), (ii) high elevation, predator present (H, +P), (iii) low elevation, predator absent (L, −P) and (iv) low elevation, predator present (L, +P)] at the end of the field experiment. The different letters above the error bars denote that the difference was statistically significant at the level of = 0·05, as revealed by two-way anovas (a) and GLM (b) followed by Tukey's test for multiple comparisons. Sample size was five for all the treatments.

Complementary Experiments

Predator presence and patch elevation, as well as their interactions, significantly affected mean tunnel depth but not tunnel number in both of the glasshouse experiments (Table S2; Figs 5 and S8). Specifically, tunnel depth was lower in low patches compared with high patches; moreover, predators significantly increased the tunnel depth in high-elevation patches, while tunnel depth was not affected by predator presence in the low-elevation patches (Fig. 5a,c). This result indicates that the occurrence of predator-induced vertical habitat shifts is dependent on patch elevation.

Figure 5.

The difference in behavioural traits of tunneller beetles including mean tunnel depth and tunnel dung weight among the four treatments [(i) high elevation, predator absent (H, −P), (ii) high elevation, predator present (H, +P), (iii) low elevation, predator absent (L, −P), and (iv) low elevation, predator present (L, +P)] at the end of the first (a,b) and second (c,d) glasshouse experiment. The different letters above the error bars denote the difference among treatments was statistically significant at the level of = 0·05, as revealed by two-way anova followed by Tukey's test for multiple comparisons. Sample size was five and six for all the treatments in the first and second glasshouse experiment, respectively.

Tunnel dung weight was significantly higher in low compared with high elevation patches (Table S2; Fig. 5b,d) and predators significantly decreased the dung mass in tunnels in low elevation patches (Table S2; Fig. 5b,d) but not high-elevation patches. This is consistent with the trend of dung mass loss in the primary experiment.

In addition, the number of the living tunnellers remained at eight during the experimental period within all mesocosms in both of the two glasshouse experiments, which indicated that the predators produced a predation-risk but not a consumptive effect on the prey. This is consistent with the result obtained in the primary experiment.

Discussion

Our experiments revealed that predatory beetles can markedly enhance plant biomass (by 28%) in a Tibetan alpine meadow. Because predatory beetles did not cause any density changes in their detritivore prey, we are able to attribute the observed effects to a trait-mediated indirect interaction (TMII). More specifically, our results indicate that the predatory beetles altered the activity domain of tunneller beetles, in turn improving soil properties within the lower soil layer, consistent with trait-based predictions. Importantly, our results also reveal that trait-based predictions in these below-ground interaction chains may depend on environmental context: the positive TMII disappeared within low-elevation patches, where detritivores are apparently unable to respond to predator presence with an increase in burrow depth. Environmental context therefore determined whether the detritivore expressed a trait-shift (increased burrowing depth) that transmitted a positive indirect interaction between the initiator (predatory beetle) and receiver (plants). To the best of our knowledge, this is one of the few studies explicitly demonstrating the role of environmental context in the transmission of TMIIs.

Despite our prediction that tunneller beetles might be more vulnerable to predation in low patches where they had limited recourse to escape predation risk, we found no evidence of this effect in either the field or glasshouse experiments. Our observations suggest this can be explained by the ability of adult tunnellers to behaviourally avoid attack by leaning their heads against the soil within tunnels. However, as tunnelling beetles retreat to lower soil layers in the presence of predators when conditions allow (i.e. in high patches), it is logical to presume they incur a cost if they remain in the upper soil layer. This cost may be in the form of reduced activity or physiological stress (Hawlena & Schmitz 2010) that may reduce reproductive output and/or long-term survival of adults, and moreover, their offspring might suffer a higher mortality rate in shallow tunnels (Barkhouse & Ridsdill-Smith 1986; Sowig 1996).

Results from the greenhouse experiments provide support for our mechanistic interpretation of the field experiment, that is, within the high patches in the field experiment, above-ground predatory beetles induced an increase in the tunnelling depth of the detritivorous beetles. This behavioural response in the prey species is presumably to reduce direct contact with the predacious beetles and associated costs following perception of predator vibrations and odours (Young 1998). In addition, it is worthwhile to note that tunnel depth was negatively associated with dung mass removal in both glasshouse experiments (Table S3). This might be due to the trade-off between tunnelling time and dung-hauling time in the beetles, as suggested by the positive linear relationship between immigration time from the inner cages and tunnel depth and by the negative linear relationship between immigration time and tunnel dung weight (Table S3).

The fact that predators did not significantly change the depth of tunnellers living in the low-elevation patches can tentatively be attributed to higher water content at the lower soil layer, which may have been lethal for egg hatching and larval development (Doane 1967). Moreover, higher water content at the lower layer may induce low-oxygen concentrations affecting the metabolism of tunneller adults and larvae (Collis-George 1959; Sparks & Strayer 1998) and low-temperature limiting the normal physiological metabolic process rates for the tunneller adults that often have high thermal requirements under high moisture conditions (Lobo, Lumaret & Jay-Robert 2002). This lack of a behavioural response did not incur any detectable cost in terms of mortality for the tunneller adults, but substantially reduced the security for resources for next generation as indicated by the lower dung mass in tunnels (Fig. 5b,d). Importantly, the second glasshouse experiment provides evidence that differences in plant species composition across patch types, and potential differences in below-ground root density, was not the major cause of context-dependent habitat-shifting behaviour of detritivore beetles, supporting our interpretation that this was driven by mechanisms related to soil moisture.

Therefore, detritivores showed different behavioural responses to predators in high- vs. low-elevation patches in the mesocosms and also most likely in the field. In turn, this patch-dependent behavioural response to predators probably explains why predator presence improved soil properties and ultimately increased plant biomass in high – but not low – elevation patches. Specifically, the predator-induced enhanced tunnel depth of coprophagous beetles that was restricted to the high-elevation patches (observed in the glasshouse) resulted in reduced bulk density within the lower soil layer of such patches. Note that the mechanistic link between detritivore behaviour (tunnel depth; observed in glasshouse) and bulk density (observed in field) is supported by the tight relationship between the mean values of these variables across treatments (linear regression, = 4, R2 = 0·980, = 150·207, = 0·006). This detritivore-mediated reduction in bulk density in turn appears to have facilitated water drainage and nutrient leaching from the upper soil layer (see also Domínguez, Bohlen & Parmelee 2004). Moreover, the digestion activities of the detritivores and their larvae, together with improved soil aeration, might have improved microbial activity, contributing to increased levels of soluble nutrients in the lower soil layer (Jobbágy & Jackson 2001). These predator-induced improvements in lower soil conditions, which only occurred within the high-elevation patches, probably facilitated nutrient uptake by roots, ultimately increasing above-ground plant biomass. The mechanistic links between tunnel-modified conditions in the deep soil layer and plant biomass within the high-elevation patches were further evident in the continuous linear relationships among soil bulk density, water content, N availability and above-ground biomass in this soil zone (Fig. 3). Conversely, within the low-elevation patches, where the detritivores did not undergo a vertical habitat shift in response to predators, soil conditions at the lower soil layer were not improved by predator presence. This further suggests the importance of predator-induced habitat shifting to the observed pattern of soil properties and plant growth.

The fact that predatory beetles did not induce a detectable difference in root biomass between low and high patches can perhaps be attributed to the particular traits of the plant species. The alpine herbs are often characterized by a large root biomass allocation due to low-nutrient availability (Fan et al. 2008), such that their root growth might be insensitive to improved soil conditions (by the activities of tunneller beetles), at least in the short term.

In addition, we note there was a difference in plant composition (graminoid: forb ratio) between low and high patches, which could have potentially contributed to the context-dependent effect of predator beetles because graminoids tend to be more sensitive to nutrient availability than forb species (Henry, Freedman & Svoboda 1986). However, this can be largely ruled out in our experiment because graminoids were relatively less abundant in the high elevation patches (see 'Materials and methods') where the positive effect of predatory beetles occurred, and moreover, the biomass ratio of graminoid: forbs remained unchanged in both types of habitats during the experiment (Fig. S6). It is also worthwhile to note that plants could positively respond to activities of burrowing beetles in both types of elevation patches, as revealed by the field experiment, where plant biomass was greater in the treatment with the detritivores only than in the control (see Table S1; Fig. 1). This suggests that the non-significant effect of predators (on plant biomass) within low patches cannot simply be attributed to plant species traits.

Our study, together with many previous studies involving tri-trophic predator-detritivore-plant systems (e.g. Wise et al. 1999; Moore et al. 2003), indicate that cascading effects in brown food webs are widespread and that the mechanism underlying the effect include both DMIIs and TMIIs (Wu et al. 2011; Zhao et al. 2013), as in living-plant-based food webs. Our recent work shows that the sign of predator-detritivore-plant cascades can switch within the same system depending on the detritivore (transmitter) species considered. Importantly, however, these effects of species identity are not idiosyncratic but can be explained and predicted based on functional traits of predators and detritivores. A similar framework has been developed for predator–grazer–plant interactions in temperate grasslands (Schmitz 2008).

Strikingly, within high-elevation patches at the same study site and using the same predatory beetle species, we previously reported negative cascading effects from predator to plants (Wu et al. 2011) – results that directly contrast with those presented here (for the high elevation patches). The primary difference between Wu et al. (2011) and the current experiment lies in the traits of the prey species. Wu et al. (2011) used an above-ground dung-dwelling species, which is active only within dung pats, and therefore has an activity domain that is highly overlapping with the above-ground predator; it therefore had no recourse to escape and its density was reduced by the predators. In our study, however, the tunneller species not only directly consumes dung particulates, but also transports the particulates into soil tunnels as food for themselves and offspring. This particular trait of digging tunnels may produce additional change in soil properties, such as bulk density and nutrient status, as noted earlier. That is to say, the different species traits in the decomposer prey might have resulted in the contrasting ecological consequences of the same predators on plants in this detrital system.

While functional traits appear to be valuable in predicting cascading effects of predators in alpine meadows via the detritus food web, this present study shows that the manifestation of cascading effects further depends on an additional factor: patch elevation. Future experiments and models in this and other systems must therefore begin to include the spatially variable cascading effects of predators. By identifying key functional traits as well as the critical aspects of environmental variation that determine how functional traits influence species interactions (see also Barton & Schmitz 2009), we may develop a framework that ultimately allows prediction of cascading effects across naturally heterogeneous ecosystems.

Acknowledgements

We are grateful to Ales Smetana and Ming Bai for species identification and two anonymous reviewers for insightful comments on an early version of this manuscript. We thank Xinqiang Xi, Junpeng Mu, Kechang Niu, Chuan Zhao, Jiyan Zhao, Yangheshan Yang and Xiaohu Ase for field assistances. This study was supported by 973 Program (2013CB956300), Main Direction Program of Knowledge Innovation of Chinese Academy of Sciences (KSCX2-EW-J-22), National Special Transgenic Project (2011ZX08012-005) National Science Foundation of China (31100387 and 31325004) to Sun S. and Wu X., and the Climate Change Consortium of Wales (C3W) to J. Griffin.

Ancillary