Introduction

A major argument for the conservation of intact ecosystems is that high levels of food web complexity maintain community persistence (Duffy 2009; Loreau and de Mazancourt 2013; Thompson and Starzomski 2007). But why should more complex ecosystems be more stable, with species persisting for long periods of time? Natural food webs are often characterised by many weak interactions among species, as opposed to fewer, stronger interactions in simpler food webs (Edwards et al. 2010; Hillebrand and Cardinale 2004). Theoretical analyses show that a relatively high abundance of weak interactions reduces the chances of extinctions caused by high levels of predation and competition (McCann et al. 1998; McCann 2000). Further evidence for the stabilising effect of more complex food webs comes from the other end of the spectrum—very simple assemblages. Maintaining predator–prey pairs under simplified laboratory conditions is often difficult as predators tend to over-exploit prey (Bonsall et al. 2002).

Homogeneous environments lacking structural heterogeneity and temporal or spatial refuges for prey ought to be intrinsically unstable, allowing predators to over-exploit their prey (Holyoak et al. 2005; Hutchinson 1961). Increasing the structural complexity of an ecosystem is associated with reductions in predator efficiency (Srivastava 2006), that can allow co-existence of predators and susceptible prey (Crowder and Cooper 1982). Similar to physical complexity, increasing the complexity of food webs by including ‘nonprey’ species, resistant to predation can reduce the foraging efficiency of predators when predators spend time encountering and handling nonprey (Kratina et al. 2007; Vos et al. 2001). Nonprey species may also form a cryptic background against which prey items are difficult to identify (Wootton 1992). Modifying the ability to detect prey can alter the handling time and attack rate of a predator’s functional response, reducing the per capita prey consumption rates (Kratina et al. 2007). Despite the demonstrated effects of nonprey species on predators’ functional responses, it is currently unknown whether and how these short-term reductions in consumption rates translate into long-term persistence in ways similar to changes in structural or spatial complexity.

To test the hypothesis that weakening trophic interactions by nonprey can increase persistence in a simple four-species food web, we manipulated complexity and recorded species’ long-term persistence. Using a system of protist consumers and a turbellarian predator, we used eight different combinations of predators, prey, and two nonprey species to assemble aquatic microcosms that varied in richness from one to four species. We specifically assess whether: (1) the presence of nonprey species reduces predation pressure on prey and consequently increases prey persistence time; and (2) multiple nonprey synergistically or additively alter prey persistence.

We here define a food web as persistent if all initially present species remain present for many generations (e.g. >100). If a species is lost from a closed food web then there is no possibility for it to return (Staddon et al. 2010). The functions within the food web exclusively performed by that species are therefore lost (Petchey et al. 2004) until another species evolves to fill the vacant niche. Should a species be driven to very low population densities, it may be able to continue to be ecologically relevant by performing an ecosystem function (Lyons and Schwartz 2001), and the species’ continued presence ensures the potential for the population to rise in the future. Hence, although we track population dynamics of individual species, we focus on the persistence of species as a measure of food web stability. We adopt this approach as a species’ continued persistence within a food web maintains the potential for its density to rise to an ecologically important level, regardless of its current density.

Materials and methods

To understand how food web stability is affected by the addition of extra species, we used experimental 200-mL microcosms varying in number of interacting species. We assembled food webs based on trophic and competitive relationships elucidated from previous experiments (Fenchel 1980; Kratina et al. 2009; Hammill et al. 2010). We cultured a predatory turbellarian flatworm (Stenostomum virginianum), its ciliate prey Paramecium aurelia and two nonprey species, the ciliate Euplotes patella (hereafter referred to as nonprey 1 or Euplotes) and the bdelloid rotifer Philodina roseola (hereafter nonprey 2 or Philodina). Previous short-term experiments have shown that neither of the two non-prey species are consumed by Stenostomum, likely due to their large body size (Kratina et al. 2007). The body width of Euplotes in the presence of Stenostomum predators is 93.7 ± 4.3 µm (mean ± SE, n = 3 cultures), and predation by gape-limited Stenostomum on individuals larger than 80 µm is negligible (Altwegg et al. 2006). The body width of Philodina ranges between 75 and 150 µm, making the majority of individuals too large to be consumed by Stenostomum. In contrast, the maximum body width of Paramecium is 42.25 ± 1.64 µm (mean ± SE, n = 4 cultures) making it highly susceptible to predation by Stenostomum. Previous studies have demonstrated that Stenostomum is a voracious predator of Paramecium, with a single predator able to consume 15 prey individuals in 4 h (Kratina et al. 2007). In addition, the presence of cues from Stenostomum causes many ciliates including Paramecium to induce morphological and behavioural defences (Hammill et al. 2009, 2010; Kusch and Kuhlmann 1994). We have also directly observed Stenostomum consuming Paramecium, in agreement with earlier observations that Stenostomum readily consumes ciliated protozoans (Archbold and Berger 1985).

The experimental communities were sustained on a basal resource of bacteria, microflagellates and other small protozoa (1–5 µm length). Paramecium (prey) and Euplotes (nonprey 1) both consume particles between 0.2 and 1.5 µm in diameter (Fenchel 1980). However, within this size range, interspecific differences in preferred food size exist, with Paramecium able to more efficiently consume particles <1.2 µm in diameter, but Euplotes better able to consume particles at the higher end of the distribution spectrum (Fenchel 1980). Conversely, Philodina (nonprey 2) prefer food items at the upper end of the size spectrum preferred by Euplotes (and larger still), and with their complex feeding apparatus are able to efficiently harvest food from the ecosystem (Ricci 1984). The prey and both nonprey species therefore likely compete with each other over the basal resource (Lawler and Morin 1993; Worsfold et al. 2009). Predatory Stenostomum possess muscular mouthparts, geared towards the capture of relatively large, fast, motile prey (Nuttycombe and Waters 1935), as opposed to the filtering apparatus associated with capture of bacteria. Previous studies have shown that, when presented with a wide variety of prey choices, Stenostomum guts contained a high percentage of animal and protist matter, with very little bacteria or algae (Nandini et al. 2011). Although we have observed Stenostomum surviving on the basal resource we used in our experiment, the morphological descriptions of Stenostomum mouthparts show that the basal resource is at the low end of the diet breadth of Stenostomum, and likely not their preferred food when Paramecium are available. The consumptive and competitive interactions present in the experimental food web are depicted in Fig. 1.

Fig. 1
figure 1

Consumptive and potential competitive interactions among predators (Stenostomum), prey (Paramecium) and two nonprey species (Euplotes and Philodina) making up the experimental food web. Consumptive links and lack thereof are confirmed from a previous short-term experiment (Kratina et al. 2007), competitive links are suggested by (Fenchel 1980) and the results of this study. Arrow widths are indicative of interaction strength

Full size image

To assess how over-exploitation of prey is altered by different levels of nonprey, we established the following treatments: (1) prey (Paramecium) alone (i.e., control to ensure that prey persist over the experimental duration); (2) prey and nonprey 1 (Euplotes); (3) prey and nonprey 2 (Philodina); (4) prey and both nonprey; (5) prey and predators; (6) prey, predators and nonprey 1; (7) prey, predators and nonprey 2; and (8) all four species present. We incubated five replicates of each treatment for 78 days (approximately 130 prey generations). The experiment was performed in media consisting of 0.4 g/L crushed protozoa pellets (no. 13-2360; Carolina Biological Supply, Burlington, NC, USA) dissolved in Purelife™ mineral water (Nestle, Canada). Prior to species inoculations, media was filtered through standard coffee filters (Thrifty Foods, Canada) and sterilised by autoclaving (Hammill et al. 2009).

In each of the 200-mL microcosms, 100 individuals of the appropriate prey and/or nonprey species were introduced after the cultures had been inoculated with 0.5 mL of mixed bacterial culture and left to stand for 24 h. Within 2 h of adding prey and nonprey, 20 Stenostomum predators were introduced into the predator treatments. We sampled 10 mL (5 %) from each vessel every 3 days and replaced with 10 mL of fresh media. During the course of the experiment, the same amount of abiotic nutrients were added to each of the microcosms. Replacing 5 % every 3 days made some nutrients present for the duration of the experiment. However, different combinations of prey, non-prey and predators presumably reduced the resources to different levels. We recorded the abundance of each consumer species using a Leica MZ8 dissecting microscope, and then calculated the total abundance of each consumer species in the microcosm from the abundance within the sample. Microcosms were swirled prior to sampling to ensure homogenous mixtures. At the end of the experiment, we searched all media remaining in the microcosms using the dissecting microscope and recorded the abundance of each consumer species. During this final sorting, we never “re-discovered” species in replicates in which they had earlier dropped below detection limits.

To compare differences in persistence times among individual treatments, we used parametric survival analyses, a branch of generalised linear models. Analysing time-to-event data (such as species extinctions) is difficult, as the variance tends to increase with the mean, confounding the constant variance assumption of linear model analyses. In addition to the increasing variance, time-to-event data may also contain censored data if an event, such as extinction, does not occur over the experimental duration. Parametric survival analyses were carried out by calculating a survival object using the function surv(), and analyzing differences using population stochastic modeling [function psm()] in the R programming language (R Development Core Team, Vienna). These functions were specifically designed for use with time-dependent and potentially censored data (Crawley 2007; Harrell 2001). Each of our survival analysis models incorporated a time-specific hazard function and lognormal error distribution as this generated the best fit to the data. Because certain treatments were used in two analyses, we applied a Bonferroni correction to avoid inflating the chance of finding spurious significant results.

In order to understand the competitive interactions among prey, predators, and both nonprey in treatments where a species of interest persisted to the end of the experiment, we analysed its densities between 20 days and the termination of the experiment. By 20 days, the dynamics of populations tended to have plateaued, suggesting they reached a carrying capacity (Fig. 2). Densities were analysed using mixed effects ANOVA with the presence of other species and date being included as fixed factors, and “microcosm” included as a random factor to account for multiple samples being taken from the same microcosm (Pinheiro and Bates 2000).

Fig. 2
figure 2

Population dynamics of predator, prey and two nonprey species in assemblages with different composition, n = 5 for each different treatment. a Prey (Paramecium) alone, b prey and predators (Stenostomum), c prey and nonprey 1 (Euplotes), d prey, nonprey 1 and predators, e prey and nonprey 2 (Philodina), f prey, nonprey 2 and predators, g prey and nonprey 1 + 2, h prey, nonprey 2 and predators in microcosms containing all species, i nonprey 1 in microcosms containing all species, illustrated separately from other species to improve clarity. For all species in all treatments, central thick lines represent means of five replicates within a treatment, thinner lines represent standard errors. The end of the line denotes the last time a species was detected in all replicates of the treatment

Full size image

Results

The prey survived to the end of the experiment in all treatments without predators, but rapidly dropped below detection limits when housed only with predators (Figs. 2, 3). Prey persistence times with predators significantly increased in the presence of nonprey 1 (Figs. 2d, 3a; P < 0.001, z = 18.44). We observed no increase in prey persistence when housed with predators and nonprey 2 compared to the predator-only treatment (Figs. 2f, 3a; z = 0.833, P = 0.41). The presence of both nonprey species increased prey persistence in the presence of predators to a greater extent than would be predicted from the effects of each nonprey in isolation (Figs. 2h, 3a; interaction term between nonprey 1 and nonprey 2, z = 2.53, P = 0.02).

Fig. 3
figure 3

Number of days above detection limit and modelled survival probabilities of extinction prone species: a prey (Paramecium) persistence time (days above detection limit), b nonprey 1 (Euplotes) persistence time (days above detection limit). Labels under bars denote the other species present in the experimental treatment, each treatment contained five replicates. Data are means ± standard errors, dotted line denotes the end of the experiment; there are no error bars in last columns as species survived till the end of the experiment in all replicates. Persistence times of predators and nonprey 2 are not shown as both species survived to the end of the experiment in all treatments in which they were present

Full size image

Nonprey 1 populations in the presence of predators and prey were significantly lower than when housed with prey alone (Table 1; Fig. 2c, d), although populations of nonprey 1 remained stable in the presence of predators after prey had been driven to extinction. When housed with prey and nonprey 2, nonprey 1 dropped below the detection limit in all replicates by day 9 (Figs. 2g, 3b; z = 13.27, P < 0.001, parametric survival analysis), suggesting they were unable to survive in the combined presence of prey and nonprey 2. In contrast, when all four species were present, nonprey 1 populations dropped below the detection limit in only 2 out of 5 replicates on days 48 and 75 (Figs. 2i, 3b; z = 31.77, P < 0.001), although nonprey 1 densities were significantly lower than when housed with either prey alone or predators and prey (Table 1; Figs. 2c, d).

Table 1 Results of mixed effects ANOVAs of species densities after day 20
Full size table

Nonprey 2 populations were not affected by the presence of nonprey 1, but were significantly reduced in the presence of predators (Table 1; Fig. 2e–h). Nonprey 2 populations remained stable while co-existing with predators isolation for ~70 days following extinction of prey (Fig. 2g).

Predator populations grew to carrying capacity in all replicates where they were present. Nonprey 1 had no effect on the carrying capacity of predators (Table 1; Fig. 2d); however, nonprey 2 reduced predator carrying capacity (Table 1; Fig. 2e, f). Predators could persist in the absence of prey (Fig. 2b), suggesting an intraguild predation system (see also Kratina et al. 2010). An intraguild system would also explain why predators were able to persist after they had driven prey to extinction or low levels.

Discussion

Our results show that predation-susceptible prey may persist for longer in more complex food webs. This effect was likely driven by the weakening of trophic interactions in more complex assemblages (Vos et al. 2001; Kratina et al. 2007). Specifically, we show that: (1) the presence of an inedible nonprey species increases the persistence time of predation-susceptible prey, although the effect of single nonprey was species specific; and (2) multiple nonprey act synergistically to increase prey persistence to a greater extent than would be predicted from additive nonprey effects. These findings suggest that non-trophic interaction modification is an important driver of long-term persistence in multi-trophic systems. We also found that the stabilising effects of species in combination cannot be accurately predicted from the effects of those species in simpler assemblages, suggesting that any species within a community can potentially be important for maintaining community dynamics. The loss of any species may therefore have unpredictable secondary effects leading to subsequent loss of other species.

In the absence of nonprey, predatory Stenostomum rapidly over-exploited populations of their prey (Paramecium), driving them below detection limits in all replicates within 9 days (~18 prey generations). This rapid reduction of prey agrees with short-term experiments showing Stenostomum are voracious predators of Paramecium (Kratina et al. 2007, 2009). The predation rates observed in these short-term experiments would be sufficient for Stenostomum to drive Paramecium (prey) to extinction within the time frame we observed here (for quantitative comparison, see Supplementary materials). This leads us to conclude that predation, rather than competition, was the primary cause of prey extinction.

In the presence of predators, nonprey 1 substantially increased prey persistence times in all replicates, with prey still being present after 48 days in one replicate. This increase in prey persistence in the presence of nonprey 1 provides further evidence that predation, not competition, is the mechanism causing prey to go extinct when housed with predators. If competition were driving prey extinctions, we would expect the addition of nonprey 1 to increase the speed of prey extinction, due to the combined competitive pressure of nonprey 1 and predators. Conversely, the reverse is true; nonprey 1 increased the ability of prey to co-exist with predators. The presence of nonprey 2 (Philodina), however, did not alter prey persistence, and no prey were detected in any replicate after day 9. This lack of an effect due to nonprey 2 may be because nonprey 2 were easily distinguishable from prey in isolation, and, in simpler food webs, predators were easily able to avoid unpalatable species and focus on prey (Ihalainen et al. 2012). As the presence of nonprey 2 did not affect the persistence time of prey when incubated with predators, we would expect that adding nonprey 2 to the three species food web (predator, prey, non-prey 1) would have no effect on prey persistence time. However, in the four-species community, prey persisted approximately 33 % longer than predicted from simple additive effects of the two nonprey. This suggests that nonprey 2 enhanced the effect of nonprey 1 on the strength of trophic interactions, with consequences for prey population dynamics. This synergistic effect of multiple nonprey may be because, in more complex food webs, predators have more difficulty avoiding multiple different inedible species that have different movement patterns and occupy different microhabitats (bottom and water column). Although not all nonprey species may be able to modulate the persistence of predator–prey interactions, our findings highlight the importance of understanding both, trophic and non-trophic interactions in the context of the wider overall food web.

For predators with an asymptotic functional response, rates of prey consumption are described by two parameters, attack rate and handling time (Holling 1959). Attack rates dictate the rate at which predators encounter and successfully capture prey, while handling time reflects the amount of time required to process a single prey before recommencing search. Handling time may include time required to physically overcome a prey and ingest and digest it before a predator is able to deal with the next prey item (Jeschke et al. 2002). Nonprey may mask or disrupt cues that predators use to detect prey, and therefore reduce the predator’s attack rate on focal prey. Short-term experiments investigating predator functional responses in the presence of nonprey demonstrate how nonprey reduce per capita predation rates (Kratina et al. 2007). Using the same species as our study, the previous experiment of Kratina et al. (2007) showed that multiple nonprey reduced predation rates to a greater extent than would be predicted from the results of single nonprey trials. This synergistic effect of multiple nonprey indicates how increased species number can reduce the strength of trophic interactions. The results of the present study show that this weakening of trophic interactions translates into long-term persistence of predator and prey in more complex assemblages.

Nonprey 1 persisted in all microcosms where they were housed with only one other species (prey alone, or predators alone following extinction of prey), although their populations were lower in the presence of predators, suggesting a dietary overlap. As population sizes of nonprey 1 were similar in the presence and absence of predators, our results agree with previous work showing nonprey 1 are not consumed by predators (Kratina et al. 2007). When nonprey 1 were incubated with both prey and nonprey 2 but in the absence of predators, their populations dropped below detection limits in all replicates by day 9, suggesting that may have been excluded by the combined competitive pressure of prey and nonprey 2. Although we cannot be certain competitive exclusion was the mechanism leading to the extirpation of nonprey 1, competitive exclusion is the simplest explanation given the diet breadths of prey (Paramecium), nonprey 1 (Euplotes), and nonprey 2 (Philodina) overlap considerably (Fenchel 1980; Ricci 1984).

When prey, predators and both nonprey were combined, nonprey 1 dropped below the detection limit in only 2 out of 5 microcosms, on days 48 and 75 (~90 and ~125 prey generations). This increase in nonprey 1 persistence may be mediated by predators reducing densities of prey. Although some prey individuals remained to compete with nonprey 1, prey populations were substantially reduced by predation. While we cannot categorically conclude that the predator-driven reduction in prey made resources available for nonprey 1 (preventing extinction), this explanation would appear the most parsimonious given the results of the current and previous experiments (Fenchel 1980). As prey are able to out-compete nonprey 1 at the lower end of nonprey 1’s diet breadth (Fenchel 1980), a reduction in prey by predators may increase nonprey 1’s access to small food items and increase their persistence. Therefore, while the addition of predators may increase the number of species competing with nonprey 1, the indirect effect of predators (through consuming prey) may increase nonprey 1’s access to a portion of the basal resource. Our results therefore agree with earlier research demonstrating that the presence of predators can alter the strength of interactions between competitors (Morin 1986; van der Stap et al. 2008), and improve community persistence (Borrvall and Ebenman 2006).

The persistence times of both prey and nonprey 1 in the four species system could not be predicted from the results of any of the three species treatments (prey + single nonprey + predator, or prey + both nonprey treatments). In all three species treatments, one species on average went extinct. In the four species food web, we may therefore predict nonprey 1 to decline as was observed when they were incubated only with prey and nonprey 2, potentially due to high levels of competition. This loss of nonprey 1 could then lead to exploitation of prey by predators, as we observed when prey were incubated with only nonprey 2 and predators. However, we detected significantly fewer extinctions in the four species system, suggesting that the increased level of complexity enhanced species persistence. The two nonprey species appeared to synergistically reduce predation pressure on prey, preventing their overexploitation. This weakening of multiple interspecific links is potentially the mechanism leading to all four species persisting for longer in the more complex food webs. These findings highlight the need to consider non-trophic interactions in predicting food web dynamics (Loreau and de Mazancourt 2013). However, this effect of nonprey interaction modifications on species persistence may prevail predominantly in ecosystems that are controlled by top–down forces.

Conclusions

We showed that, by weakening predation and potentially competition, additional consumers can increase persistence times of species susceptible to extinction. Species are currently being lost from ecosystems at rates far greater than historically observed (Cardinale et al. 2012). We found that the loss of species that initially appear irrelevant may have far-reaching consequences for food web persistence due to increasing the strength of trophic interactions among other species. These changes to interaction strengths could result in secondary extinctions and reorganisation of entire ecosystems (Dunne and Williams 2009; Golubski and Abrams 2011).

Author contribution statement

EH, PK, MV and BRA originally formulated the idea, EH and OLP carried out the laboratory work and along with BRA analysed the data. All authors contributed to the writing and editing of the MS. The experiment complies with all national laws.