Predator effects on a detritus-based food web are primarily mediated by non-trophic interactions



  1. Predator effects on ecosystems can extend far beyond their prey and are often not solely lethally transmitted. Change in prey traits in response to predation risk can have important repercussions on community assembly and key ecosystem processes (i.e. trait-mediated indirect effects). In addition, some predators themselves alter habitat structure or nutrient cycling through ecological engineering effects. Tracking these non-trophic pathways is thus an important, yet challenging task to gain a better grasp of the functional role of predators.
  2. Multiple lines of evidence suggest that, in detritus-based food webs, non-trophic interactions may prevail over purely trophic interactions in determining predator effects on plant litter decomposition. This hypothesis was tested in a headwater stream by modulating the density of a flatworm predator (Polycelis felina) in enclosures containing oak (Quercus robur) leaf litter exposed to natural colonization by small invertebrates and microbial decomposers. Causal path modelling was used to infer how predator effects propagated through the food web.
  3. Flatworms accelerated litter decomposition through positive effects on microbial decomposers. The biomass of prey and non-prey invertebrates was not negatively affected by flatworms, suggesting that net predator effect on litter decomposition was primarily determined by non-trophic interactions.
  4. Flatworms enhanced the deposition and retention of fine sediments on leaf surface, thereby improving leaf colonization by invertebrates – most of which having strong affinities with interstitial habitats. This predator-induced improvement of habitat availability was attributed to the sticky nature of the mucus that flatworms secrete in copious amount while foraging. Results of path analyses further indicated that this bottom-up ecological engineering effect was as powerful as the top-down effect on invertebrate prey.
  5. Our findings suggest that predators have the potential to affect substantially carbon flow and nutrient cycling in detritus-based ecosystems and that this impact cannot be fully appreciated without considering non-trophic effects.


Predators face disproportionate risk of extinction and their loss can affect deeply and durably the provision of ecosystem services (Dobson et al. 2006; Estes et al. 2011). Decades of research on predation have demonstrated that carnivorous animals play a critical role in the regulation of prey population dynamics with potential indirect effects cascading downwards along food chains (Terborgh & Estes 2010). However, predator effects on ecosystems are not solely lethally transmitted and are often explained by non-consumptive indirect effects (Werner & Peacor 2003; Schmitz, Krivan & Ovadia 2004; Preisser, Bolnick & Benard 2005). Those trait-mediated indirect interactions are determined by changes in prey traits (e.g. behaviour, morphology, physiology) in response to predation risk that subsequently affect the interaction between the prey and its resources, competitors and/or habitats. In addition, predators can also strengthen bottom-up controls through direct effects on elemental cycling (e.g. nutrient retention/excretion and carcass distribution; Ngai & Srivastava 2006; Hawlena et al. 2012) and habitat alterations (i.e. ecological engineering effects; Zhang, Richardson & Negishi 2004; Sanders & van Veen 2011; Statzner 2012). All these non-trophic interactions (i.e. non-feeding links sensu Kéfi et al. 2012) have the potential to dampen or reinforce trophic cascades, implying that both trophic and non-trophic pathways should be evaluated in concert to better predict net predator effects on ecosystems (Preisser, Bolnick & Benard 2005; Werner & Peacor 2006; Hawlena et al. 2012; Zhao et al. 2013).

Predators influence trophic dynamics and elemental cycling through effects on plant litter decomposition, a key ecosystem-level process in many ecosystems (Moore et al. 2004; Gessner et al. 2010). However, the underlying mechanisms of this top-down control are not fully understood – partly because empirical findings often considerably deviate from expectations based on classic trophic cascade concept. In fact, recent findings from terrestrial realm highlight that non-trophic interactions may prevail in the regulation of invertebrate spatial distributions and elemental cycling rates by predators (e.g. Hawlena et al. 2012; Zhao et al. 2013 but see Nichols 2013). This principle may be extended to detritus-based aquatic food webs that share common features with the terrestrial ‘brown’ world (Gessner et al. 2010). Forested headwater streams receive large amounts of terrestrial leaf litter used as food and habitats by aquatic microbial decomposers, invertebrate consumers and their predators (Wallace et al. 1997). In these ecosystems, predators have been found to reduce litter decomposition through trophic cascade with and without consumption of large detritivorous prey (e.g. Malmqvist 1993; Greig & McIntosh 2006; Woodward et al. 2008). However, these previous studies did not thoroughly determine the relative importance of trophic versus non-trophic predator effects and how they propagate down through microbial decomposers and microbivores.

The ecological importance of non-trophic predator effects on detritus-based food webs is likely to be promoted: (i) by the compactness and porous structure of litter habitats and (ii) by the poor elemental quality of detrital resources. Trait-mediated indirect interactions potentially emerge because structurally complex habitats, such as leaf litter layer or packs, may affect negatively the foraging success of predators and offer opportunity for prey to develop antipredator behavioural responses (Kalinkat, Brose & Rall 2013). Such altered detritivore behaviour combined with the foraging activity of predators may then affect deeply microhabitat features (e.g. Zhang, Richardson & Negishi 2004; Zhao et al. 2013). This ecological engineering effect involves, for instance, biotic disturbance of coarse and fine particles and the formation of burrow holes and tracks on litter. Additionally, as refractory organic matter with high carbon-to-nutrient ratios (i.e. plant litter) is in excess in detritus-based food webs (Moore et al. 2004; Wardle et al. 2004), animal-originating nutrients and labile carbon (e.g. excretion products and prey carcasses) have the potential to stimulate substantially the growth and activity of microbial decomposers (Guenet et al. 2010; Schmitz, Hawlena & Trussell 2010). This effect may propagate upwards along trophic chains through consumption of microbial biomass by microbivores and through litter conditioning effects on larger detritivores (Moore et al. 2004; Gessner et al. 2010).

In this study, our core motivation was to gain a better mechanistic understanding of the trophic and non-trophic predator effects on aquatic detritus-based ecosystems. For this purpose, we assessed experimentally the influence of a free-living planarian flatworm predator on litter decomposition rate and litter-associated fauna and microbial decomposers in a forested headwater stream. The functional importance of flatworms has been suggested by their high abundance in leaf packs, where they use sticky mucus secretions to crawl and to trap small prey (e.g. Jennings 1957; Reynoldson & Young 1963; Armitage & Young 1990; Cash, McKee & Wrona 1993). In the experimental food web considered here, we a priori hypothesized a theoretic size-structured food chain composed of (ranked in decreasing order): flatworm as the top predator, macrofauna as main prey, microphagous meiofauna (i.e. 50–500 μm invertebrates; Giere 2009), microbial decomposers (fungi and bacteria) and finally leaf litter as the basal resource (Fig. 1). Leaf litter surface is a quite homogeneous landscape for minute organisms like meiofauna and bacteria, which thus may heavily rely on complex interstitial habitat formed by fine sediment entrapped in litter accumulation (Wood & Armitage 1997; Robertson & Milner 2001; Gaudes et al. 2009). The largest invertebrates (flatworm and their prey) may affect the amount of deposited sediments through bioturbation or bioconsolidation (Zhang, Richardson & Negishi 2004; Statzner 2012), thereby providing the basis for an ecological engineering effect, through which flatworm predators may drive trophic dynamics (dashed arrows in Fig. 1).

Figure 1.

A conceptual model of the aquatic food web associated with decaying leaf litter in experimental enclosures set in stream. Macrofauna considered in this study includes Hydrachnidia (water mites; M), Oligochaeta (O), non-predatory Chironomidae (C), biofilm-grazing Ephemeroptera and Coleoptera (E), leaf-shredding Plecoptera (P), predatory Diptera (D). Meiofauna includes Rotifera (R), Gastrotricha (G), Tardigrada (T), Nematoda (N) and Copepoda Harpacticoida (H). The scale bar placed next to invertebrates represents 1 mm. Arrows depict direct and indirect interactions through which the predatory planarian flatworm Polycelis felina was hypothesized to influence detritus-associated communities, their habitat (sediments) and the stock of the basal resource (leaf litter). Signs denote the direction of expected changes in biomass/amount (depletion −, accretion +, undefined ±). Solid arrows represent a cascading top-down effect of flatworms determined by both trophic (biomass changes) and non-trophic (e.g. behavioural changes) interactions. Dashed arrows represent ecological engineering effects mediated by sediment deposition on leaf surface. For instance, the largest animals (flatworm and macrofauna) were thought to influence sedimentation through bioturbation (−) or bioretention/consolidation (+) (see e.g. Zhang, Richardson & Negishi 2004; Statzner 2012).

Materials and methods

Study Site

Our study site was in an oligotrophic headwater stream (‘Le Bernazobre’: lat. 43°28′53′′N, long. 02°13′06′′E, elevation 400 m a.s.l.) situated in the ‘Montagne Noire’, a midland region in south-western France covered by broadleaf deciduous forest and coniferous plantations. Stream width (2·7 m), water depth (0·2 m), water temperature (8·1–8·5 °C), pH (7·8–7·9) and conductivity (83–95 μS cm−1) were fairly constant throughout the study period from 29 March to 14 May 2012. This stream showed a pluvial flow regime, with winter–spring maximum discharges followed by a summer–autumn low-flow period. It received substantial inputs of plant litter (550 gDM m−2 in 2011; unpublished data) from riparian forest primarily composed of beech (Fagus silvatica L.), oaks (Quercus spp.), alder (Alnus glutinosa Gaertn.) and poplars (Populus spp).

Polycelis felina (Dalyell) is the most abundant predatory flatworm species in headwater streams in the ‘Montagne Noire’ (11 headwater streams surveyed). P. felina occurs in most benthic habitats with average density by stream ranging from 5 to 76 individuals per square metre and with leaf packs hosting up to 20 individuals per gram of leaf litter dry mass.

Experimental Design

We manipulated the density of P. felina (0, 3, 9 ind.) in small (10 × 10 × 2 cm) semi-rigid litter bags set in the study stream. These enclosures were constructed using 500-μm-square nylon mesh netting, which efficiently prevented flatworms from escaping. Bags were filled with 2·5 g of air-dried oak leaves (Quercus robur L.) collected at abscission in autumn 2011. Due to its slow decomposition and high toughness, oak leaves form structurally stable microhabitats for benthic organisms. Fifty-six litter bags secured on iron sticks and wedged onto the stream bottom were incubated in the stream for 22 days to ensure suitable litter colonization by small invertebrates and micro-organisms (Gaudes et al. 2009). Then, a total of 192 adult P. felina individuals (length ~ 8 mm, dry mass ~ 1·2 mg) were collected in a nearby stream and were introduced in litter bags with no delay. There were 16 pre-colonized litter bags by treatment consisting of zero, three or nine P. felina individuals enclosed. Eight litter bags were not assigned to any of the three treatments and were returned to the laboratory to control response variables at the beginning of the experiment (day 0: i.e. 20 April 2012). Eight randomly picked bags by treatment were sampled after 14 and 24 days following the addition of predators (i.e. the 4 and 14 May 2012). Samples were stored individually in plastic zip-lock bags, were transported to the laboratory in a cool box and were processed within two hours following harvesting.

Sample Processing and Response Variables

Oak leaves contained in litter bags were gently rinsed with dechlorinated water over a 50-μm-mesh sieve to collect fine sediments and invertebrates. Two sets of five leaf discs (Ø 13 mm) were cut out from randomly picked leaf fragments, avoiding the central vein. One set was preserved in a 4% formaldehyde solution for later bacterial biomass determination, and the second set was kept frozen until fungal biomass analyses (Appendix S1, Supporting information). Ten additional discs were also cut out from leaves and were kept fresh to assess fungal assemblages from the spores they released (Appendix S1, Supporting information). The remaining litter was oven-dried at 60 °C for 72 h and weighed to the nearest 0·1 mg. Detrital dry mass was converted into carbon (C) mass after the C content of each sample was estimated from an aliquot of finely grounded litter using a C/N analyser (NA 2100 Protein, Carlo Erba Instruments, Wigan, UK). Remaining leaf C mass was calculated as the difference of detrital C mass to decomposer C biomass (fungi and bacteria; see Appendix S1, Supporting information) and was expressed as the ratio of final-to-initial leaf C (Gessner & Chauvet 1994).

Polycelis felina and the largest invertebrates were hand-picked on the 50-μm-mesh sieve and were preserved in 70% ethanol until identification and biomass determination. Smaller invertebrates and fine inorganic sediments were separated using a density-gradient centrifugation technique (Appendix S2, Supporting information). The inorganic pellet made of fine (50–500 μm) sediment particles was rinsed, oven-dried at 60 °C for 72 h and weighed to the nearest 0·1 mg. This variable was used as a proxy to measure the amount of drifting particles trapped onto leaf surface biofilm. The organic supernatant containing small macro- and meiofauna was preserved in a 4% formaldehyde solution until identification and biomass determination (procedures further detailed in Appendix S2, Supporting information).

Traditionally, aquatic invertebrates passing through 500-μm meshes are classified as meiofauna (Giere 2009). However, due to strong body size disparity among invertebrates found in our leaf bags, we took a more pragmatic classification, considering two groups which were based on mean taxa body mass: macrofauna (5–50 μgC ind−1) and meiofauna (0·02–0·2 μgC ind−1) (Appendix S3, Table S4, Supporting information). Macrofauna, both hand-picked and centrifuge-sorted, comprised taxa usually classified as macro-invertebrates and dominated by insect larvae (Appendix S2, Table S2, Supporting information). Most individuals probably entered enclosures as eggs or as tiny larval instars. Unlike macrofauna, meiofauna was only made up by non-insect taxa, likely too small to be preferentially preyed by P. felina as supported by the extremely high predator-to-prey body size ratio (>3000). In addition to discriminate between prey and non-prey items for P. felina, the distinction between macrofauna and meiofauna was justified by fundamental differences expected in response/effect traits (see Appendix S3, Supporting information).

Species level identification was performed in two key communities, nematodes (meiofauna) and aquatic hyphomycetes (fungi), to infer predator indirect effects on community assembly. Nematodes were the dominant group in meiofauna (48·6% of meiofaunal biomass; Table S2, Supporting information), whereas aquatic hyphomycetes are recognized to be the main fungal decomposers in streams (Barlöcher 1992).

Statistical Analyses

Linear model was used to assess the main and interactive effects of P. felina density and experiment duration on oak leaf decomposition, community biomass (fungi, bacteria, meiofauna, macrofauna) and sediment deposition on leaf surface. P. felina density was introduced as a covariate in the model; the assumption of linear relationship with the response variables was met after the square-root transformation of P. felina density. In addition, response variables were log-transformed to achieve homoscedastic and normally distributed residuals.

Predator-induced changes in nematode and aquatic hyphomycete communities were assessed using non-metric multidimensional scaling (nMDS) ordinations. In addition, permutational multivariate analysis of variance (permanova) was performed to test for main and interactive effects of P. felina density (covariate, square-root transformed) and time. Then, a distance-based test for homogeneity of multivariate dispersion (PERMDISP2) was used to check whether statistically significant predator effects were not due to a deviation from the condition of homogeneous dispersions within density levels (Anderson 2006). P-values were determined based on 999 permutations. Lastly, similarity percentages analysis (SIMPER) was conducted on each community to identify the species that contributed the most to the separation between the high predator density treatment and the treatment without predator. Community analyses were all based on the Bray–Curtis distance calculated from untransformed nematode species densities or fungal species’ spore production in each litter bag.

Partial least squares path modelling, a robust form of structural equation modelling (Esposito-Vinzi, Tinchera & Amato 2010), was used to assess hypothesized direct and indirect predator effects on community biomass and composition, the basal litter resource and fine sediment deposition on leaves (Fig. 1; Appendix S3, Supporting information). Experimental data were fitted to cause-and-effect models to potentially account for patterns of covariance in biomass (or inorganic sediment mass) of system components. To gain better mechanistic insight into predator effects on invertebrate communities while limiting model complexity, macrofauna and meiofauna were specified as latent variables to take within-community functional differences into account. The latent variable for macrofauna was estimated based on the biomass of six groups which were defined using their main feeding modes and food: biofilm-grazing Ephemeroptera and Coleoptera; non-predatory (mostly biofilm-feeding) Chironomidae; deposit-feeding Oligochaeta; leaf-shredding Plecoptera; predatory Diptera; predatory/parasites Hydrachnidia (Appendix S3, Table S4, Supporting information). Such trophic groups cannot be readily defined for meiofauna, especially given the coarse taxonomic resolution and limited knowledge of the trophic ecology of some groups (Schmid-Araya et al. 2002). Consequently, the latent variable for meiofauna was estimated based on the biomass of the five broad taxonomic groups encountered (Nematoda, Rotifera, Tardigrada, Copepoda Harpacticoida and Gastrotricha). Paths were set to represent both trophic (direct feeding link) and non-trophic (e.g. ecological engineering effect) pathways (Fig. 1). The modelling procedure, the rationale behind the construction of causal path diagrams, and the results and their interpretations are fully detailed in Appendix S3 (Supporting information). Briefly, we first fitted our a priori hypothetical model combining top–down interactions and ecological engineering effect (Fig. 1). As model outcomes did not match our qualitative predictions and failed to explain the response of certain biological communities to predator, we constructed an alternative model that also included a trait-mediated indirect interaction (suggested by correlation analyses) and bottom-up regulation (Appendix S3, Supporting information). Variables entered into path models were centred by date to remove temporal variations from the data set.

All statistical analyses were performed with R (R Development Core Team 2012). The ‘vegan’ (functions: ‘monoMDS’, ‘adonis’, ‘betadisper’ and ‘simper’; Oksanen 2009) and ‘plspm’ (Sanchez & Trinchera 2013) packages were used for multivariate community analyses and partial least squares path modelling, respectively.


Net Predator Impact

Oak leaves lost on average 21% of their initial C mass during the 22-day period of pre-colonization (day 0) and the percentage of leaf C remaining continued to decline after predator addition to enclosures (Fig. 2a). At the end of the experiment, leaf C remaining ranged from 51 to 67% of initial leaf C across treatments (Fig. 2a). Biomass of leaf-associated biota and amount of sediments tended to increase with incubation time (Fig. 2), although fungal biomass did not increase as much as the biomass of other assemblages (Fig. 2b). In fact, fungal biomass remained fairly constant between 14 and 24 days after predator addition (F1,44 = 0·4; P = 0·53), whereas all other biota showed significant biomass increase through time (F1,44 > 8·4, P < 0·01).

Figure 2.

Effects of the presence and density of the predatory flatworm Polycelis felina on oak leaf mass (a), on the biomass of fungi (b), bacteria (c), meiofauna (d) and macrofauna (e), and on fine sediments deposited on leaves (f). Values are means by sampling date (N = 8, ±1 SE). Open, shaded and solid bars show leaf pack enclosures without flatworm, with three flatworms and with nine flatworms, respectively. Abbreviations: carbon (C), dry mass (DM) and not determined (nd).

Polycelis felina density accelerated significantly oak litter decomposition (F1,44 = 4·7, P = 0·036; Fig. 2a). It also enhanced bacterial biomass (F1,44 = 32·1, P < 0·001; Fig. 2c), meiofaunal biomass (F1,44 = 5·0, P = 0·03; Fig. 2d) and the amount of fine sediment deposited on leaves (F1,44 = 4·3, P = 0·043; Fig. 2f). Bacteria and meiofauna showed the greatest response to P. felina with a twofold increase, at the end of the experiment, between controls without predators and the high predator density treatment (Fig. 2c,d). In contrast, predator effect was detected neither on fungal biomass (F1,44 = 0·1, P = 0·75; Fig. 2b) nor on macrofaunal biomass (F1,44 = 1·6, P = 0·21; Fig. 2e). All these results were not dependent on the time as the predator-by-time interaction was not significant for any tested response variable (F1,44 < 3·1, P > 0·085).

Community Structure

Multivariate analyses based on the Bray–Curtis distance indicate that P. felina affected the community structure of aquatic hyphomycetes (permanova: F1,44 = 5·0, P = 0·002; Fig. 3a; Table 1) and nematodes (permanova: F1,44 = 3·6, P = 0·015; Fig. 3b; Table 1). The response of nematode community was observed only at high predator density (Fig. 3b; Table 1). Both hyphomycete and nematode communities also changed through time (permanova F1,44 > 4·0, P < 0·01), but assembly trajectories were not affected by the predator (predator-by-time interaction: F1,44 < 1·2, P > 0·29). The permanova assumption of homogeneous multivariate dispersions was met, as the spread of points in ordination plots did not differ significantly across predator density levels for any of the two communities (Fig. 3a,b; PERMDISP2: P > 0·78).

Table 1. Mean (N = 16) density of nematode species and spore production of aquatic hyphomycetes in treatments with either zero, three or nine flatworms. Summary of similarity percentages analysis (SIMPER) is given between the treatment without predator and that with nine predators. Only species contributing >1% to the separation between these treatments were shown. The main diet of nematode species was extrapolated from their feeding types following Traunspurger (1997)
 Mean density (ind gLeaf C−1) or Spore production (μgC gLeafC−1 day−1)SIMPER (%)
0 flatworm3 flatworms9 flatworms0 versus 9 flatworms
  1. B, bacterial feeders; O, omnivores.

(B) Eumonhystera pseudobulbosa503428121014·96
(B) Eumonhystera vulgaris476548110814·42
(B) Eumonhystera dispar4454437179·19
(B) Eumonhystera barbata1021383174·51
(B) Eumonhystera longicaudatula73351652·70
(O) Fictor fictor3953981·69
(O) Tobrilus sp23731891·67
(B) Eumonhystera simplex17351011·29
(O) Dorylaimoides sp.519591·12
(O) Tobrilus sp13119181·00
Aquatic hyphomycetes
Clavariopsis aquatica 33637345810·47
Tricladium chaetocladium 7611270128210·01
Anguillospora filiformis 4525527407·80
Alatospora acuminata 1018203·30
Tetracladium marchalianum 1572121842·82
Flagellospora curvula 810222·75
Tetrachaetum elegans 63124981·64
Lunulospora curvula 121581·12
Figure 3.

Structure of aquatic hyphomycete (a) and nematode (b) assemblages assessed through two dimensional nMDS ordination based on Bray–Curtis distance calculated from untransformed nematode species densities or fungal species’ spore production in each litter bag. ‘Spider webs’ link each sample to the centroid of the predator treatment it belongs.

The dominant nematode species were bacterial feeders, whose global contribution to nematode assemblages averaged 88% (Table S3, Supporting information). Outcomes of SIMPER analyses indicated that the separation between control enclosures without predator and the high predator density treatment was due to a positive effect of flatworms on the dominant bacterial-feeding nematode species (mostly Eumonhystera pseudobulbosa, E. vulgaris and E. dispar) and on the dominant aquatic hyphomycete species (mostly Clavariopsis aquatica, Tricladium chaetocladium and Anguillospora filiformis) (see Table 1).

Path Modelling

Our a priori hypothetical causal model, once fitted to experimental data, provided only weak evidence of a top-down cascading regulation of consumer biomass by flatworm predators (Fig. 4a). Flatworms did not deplete macrofaunal biomass (path coefficient = +0·08), and the negative effect of macrofauna on bacterial biomass (−0·27) was not significant (95% CL:−0·73 to +0·11). Fungal biomass was positively affected by macrofauna (+0·38), and non-significant positive path coefficients were found between macrofauna and meiofauna and between meiofauna and bacteria (Fig. 4a). Macrofauna, fungi and bacteria had negative effect on leaf litter resource; however, this was only significant for bacteria. Flatworms and macrofauna increased significantly the amount of fine sediments deposited on leaf surface (Fig. 4a). Sedimentation had a positive and significant effect on meiofauna (+0·47), and a positive but not significant effect on bacteria (95% CL: −0·27 to +0·90).

Figure 4.

Path diagrams used to assess direct and indirect effect of flatworm on a detritus-based food web. The a priori hypothetical model (a) presented on Fig. 1 was fitted and a second model (b), which was developed to test alternative hypotheses (e.g. bottom-up regulation), improves the match between theory and observations (see Appendix S3, Supporting information for further details). Solid arrows indicate trophic interactions, whereas dashed arrows indicate non-trophic interactions. The path linking flatworm to bacteria in the second model (b) represents a possible trait-mediated indirect interaction (see Results). Macrofauna and meiofauna were specified each as a latent variable defined as a linear combination of trophic groups or taxa. Loadings are not shown on the figure since all taxa or trophic groups were positively correlated with the latent variable they represent (see Appendix S3, Supporting information). Values are path coefficients estimated by partial least squares path modelling analysis. Asterisks indicate values significantly different from zero determined based on 95% percentile confidence intervals calculated on 200 bootstrap samples. Abbreviation: goodness-of-fit (GOF).

Of the six invertebrate groups initially considered to represent macrofauna in the path model, the dominant taxa (non-predatory Chironomidae: >80% macrofaunal biomass) and two other minor trophic groups (biofilm-grazing Ephemeroptera and Coleoptera and predatory/parasites Hydrachnidia) showed very weak association with other variables and thus were removed from the final model to improve the goodness-of-fit (Appendix S3, Supporting information). If flatworms did not deplete the biomass of macrofauna, they might modify their feeding efficiency through trait-mediated interactions. To test this hypothesis, we assessed the correlation between the biomass of non-predatory Chironomidae and bacteria (their expected main food source; Appendix S3, Supporting information) for each level of predator density. Pearson's correlation was negative in the absence of predator (N = 16, r = −0·52, P = 0·04) and it became non-significant with three (N = 16, r = −0·05, P = 0·86) and nine predators (N = 16, r = −0·07, P = 0·78).

We developed an alternative causal model (Fig. 4b): (i) to incorporate this possible trait-mediated indirect interaction of flatworms on bacteria (we added a direct path from flatworms to bacteria), (ii) to test for the occurrence of bottom-up rather than top-down regulation (we reversed the direction of the paths between macrofauna and meiofauna and between macrofauna and fungi) and (iii) to assess the possible control of sedimentation on macrofauna (we reversed corresponding path's direction). The alternative model had slightly better fit (goodness-of-fit: GOF = 0·43) than the a priori hypothetical model (GOF = 0·38). This was mostly due to an increase in explained variations for macrofauna (R2 changed from <0·01 to 0·54) and bacteria (R2 changed from 0·03 to 0·27). The alternative model hinted at the prevalence of non-trophic interactions (dashed arrows) for the propagation of predator effects across the detritus-based food web. Path coefficients revealed a strong significant trait-mediated indirect interaction between flatworms and bacteria (+0·54). Flatworm predators had also significant positive indirect effects on macrofauna and meiofauna through increased sedimentation on leaf surface (Fig. 4b). Bottom-up regulation of macrofauna by meiofauna was evidenced by a significant positive path coefficient (+0·35). The sum of indirect flatworm effects on macrofauna (sum of the products of path coefficients along all indirect pathways = +0·19) was found to balance out predator’ direct negative impact on prey biomass (−0·17). The balance of top-down and bottom-up forces provides a reasonable explanation for the non-significant net predator effect on macrofauna shown by univariate analysis (Fig. 2e). The alternative model confirmed that bacterial biomass was affected neither by sedimentation (+0·08) nor by direct predation (meiofauna: +0·03; macrofauna: −0·11). Lastly, macrofauna was found to respond positively to fungal colonization of leaf litter (+0·28).


Polycelis felina flatworms had noteworthy effects on litter decomposition and community assembly processes in the detritus-based food web studied here. Trophic interactions, in the narrow sense, were unlikely to be the chief driver of observed predator control, which was largely determined by non-trophic pathways. Specifically, we found evidence in support of trait-mediated indirect interaction between flatworms and bacteria possibly mediated by an antipredator response of non-predatory Chironomidae larvae (Stief & Hölker 2006). This dominant macrofaunal group relied on leaf biofilm as main food source and was likely the main prey of P. felina (Armitage & Young 1990). Additionally, flatworm affected indirectly macro- and meiofauna biomass through an ecological engineering effect on sedimentation. The importance of non-trophic predator effects have been highlighted in recent studies focusing on soil food webs (e.g. Hawlena et al. 2012; Zhao et al. 2013), hinting at similarities in the way that predators control the structure and the dynamics of detritus-based food webs in terrestrial and aquatic realms.

Results from path modelling suggest that predator-induced acceleration of litter decomposition was mediated by bacteria, which was surprising considering their small contribution (<0·1%) to the total biomass of microbial decomposers. Aquatic fungi are generally assumed to outperform bacteria in determining leaf mass loss and stimulating the feeding activity of leaf-shredding invertebrates (Gessner et al. 2010). It is thus questionable whether the changes in the structure of aquatic hyphomycete community in response to flatworm density were also involved in the acceleration of litter decomposition. The rationale behind this hypothesis is that fungal species differ in their efficiency to process leaf litter and that greater sporulation activity by functionally dominant species should lead to faster rates of litter decomposition (Gessner & Chauvet 1994; Dang, Chauvet & Gessner 2005; Dang et al. 2009). The genuine importance of fungi in mediating predator effects on decomposition may also have been underappreciated in our study partly because of the imprecise estimation of living mycelia biomass provided by ergosterol measurement (Mille-Lindblom, von Wachenfeldt & Tranvik 2004).

Quick prey population turnover in streams may also partly explain the apparent lack of top-down regulation of macrofauna by flatworms in enclosures (Englund & Olsson 1996). As predator density was not exaggerated in our study, predation rate was likely to be much smaller compared with the rate of prey immigration into enclosures through drift dispersal. Chironomidae larvae are known to show high passive drift rate; for instance, Palmer (1992) reports on average 61 drifting chironomids by cubic metre in a forested stream. Since thin-bodied chironomids can easily pass through 500-μm meshes, it is plausible that such continuous supply of colonizers compensated prey depletion by flatworm predators.

The strong positive association we found between fine sediments and invertebrates revealed that habitat availability and complexity was probably the foremost constraint on leaf litter colonization. This makes sense, since most invertebrates encountered in leaf packs had clear affinity with interstitial habitats (Robertson & Milner 2001; Boulton et al. 2002). The possible consolidation effect of flatworm predators on sediments contrasted with the result of a previous study showing that actively foraging stonefly predators removed fine sediments in benthic habitats through their bioturbation activity (Zanetell & Peckarsky 1996). Omnivorous and detritivorous invertebrates are also known for their ability to resuspend fine particles settled on submerged leaves (Sanpera-Calbet, Chauvet & Richardson 2012). A reduction in macrofaunal activity in response to predation risk (e.g. Stief & Hölker 2006) may partly explain why fine sediments increased with P. felina density. Additionally, we suspect the coalescing mucus strands that P. felina secreted copiously while foraging, to act as sediment traps (Jennings 1957). In a recent review, Statzner (2012) indicates that animal secretions can substantially alter sediment deposition and transport in running waters. Although such effects are difficult to quantify in natura, observations of P. felina movement in laboratory suggest that large leaf areas were potentially covered by mucus secretions: At 10 °C, we determined that the mean crawling speed of P. felina was 1·6 mm s−1 (one hundred 5 s runs recorded). Assuming 1-mm-wide mucus tracks, a single individual of P. felina actively searching for its prey should be able to spread its mucus on a surface of 57 cm² of leaves in one hour, which was equivalent to about one-tenth of the total surface of oak leaves enclosed in our litter bags.

Planarian flatworms can release >50% of their assimilated carbon as mucus (Teal 1957) composed of high-quality organic compounds such as carbohydrates and proteins (McGee et al. 1998; Shagin et al. 2002). As microbial decomposers are strongly limited by the recalcitrance of leaf litter carbon (Moore et al. 2004; Romaní et al. 2006), the exploitation of the labile compounds secreted by P. felina by bacteria may be a plausible mechanism that have also sustained high bacterial biomass in enclosures with predators. This hypothesis is consistent with multiple lines of evidence, indicating that mucus trails secreted by gastropods (Herndl & Peduzzi 1989) and nematodes (Riemann & Helmke 2002; Moens et al. 2005) are hotspots of bacterial activity. If mucus tracks were really important in mediating flatworm effect on bacteria, then predator-induced acceleration of oak litter decomposition reported in this study would therefore elicit a ‘priming effect’ (Guenet et al. 2010; Danger et al. 2013), whereby inputs of labile organic matter (i.e. animal mucus) increased the mineralization rate of refractory organic matter (i.e. oak leaf litter).

The stability of predator–prey interaction is maintained by mechanisms controlled by life-history, behavioural and demographic traits of the predator and its prey (Murdoch 1994). Our quantification of direct and indirect predator effects on prey by the mean of causal path models hints at an interesting mechanism strengthening predator–prey coexistence. Specifically, P. felina maintained the local availability of its prey (macrofauna) by improving the availability of interstitial habitat (fine sediments) and its associated food (meiofauna and bacterial biofilm). In turn, this ecological engineering effect may have increased flatworm predator fitness owing to reduced prey searching time and intraspecific competition. Predator-induced prey aggregation may favour group foraging, a strategy used by free-living aquatic flatworms to maximize the success of prey capture within nested mucus webs (Medved & Legner 1974; Cash, McKee & Wrona 1993). The rationale is that the benefit outweighs the cost of group foraging when there is enough prey to share among group members (Packer & Ruttan 1988).


In our in situ enclosure experiment, evidence emerged that non-trophic mechanisms override lethally transmitted top-down predator impacts on ecosystem. The ecological relevance of non-trophic interactions is increasingly recognized in ecological literature (e.g. Werner & Peacor 2003; Schmitz, Krivan & Ovadia 2004; Schmitz, Hawlena & Trussell 2010; Kéfi et al. 2012), yet compelling empirical demonstration is still scarce especially for detritus-based food webs (Gessner et al. 2010). This can be explained by the difficulty to disentangle among the multiple components of predator effects occurring at different levels of biological organization. As shown in this study, such a problem can partly be overcome by combining experimental manipulation of predator density and data analysis based on causal path modelling, and by integrating ecologically relevant biotic and abiotic components into conceptual models.


We are grateful to Stefanie Gehner, Didier Lambrigot, Sylvain Lamothe and Frédéric Julien for technical assistance and to Olivier Dangles and two anonymous reviewers for comments on an earlier draft. This research was supported by CNRS Grant PICS05947 to AL and by ATER Assistant Professor Fellowship (University Paul Sabatier) to NM. Fieldwork and laboratory analyses were conducted within the framework of the SYLECOL project funded by the French Ministry of Ecology, Sustainable Development, and Energy.