Three different micro-predators and three bacterial species were used in the study: a ciliate protist (Tetrahymena pyriformis), a BALO (B. bacteriovorus strain 109J), and a Klebsiella phage originally isolated from a wastewater treatment plant (WWTP) in Langenreichenbach, Germany. Two of the prey strains were isolated from the inflow of the same WWTP and identified as Klebsiella sp. and as Staphylococcus sp. by 16S rRNA gene sequencing. Pseudomonas putida KT2440 from our laboratory collection was also used as a prey. The micro-predators are characterized by different prey ranges: while the protist feeds on all three prey species, the BALO only preys upon the two Gram-negative bacteria (Klebsiella sp. and P. putida), and the phage infects only the Klebsiella strain (Figure 1).

Experiments were conducted in custom-built mMBRs (Supplementary Figure 1; Segev et al., 2013). We used a 0.2-μm membrane filter that enables the washout of the phages, whereas all other microorganisms remain in the system. Each reactor was filled with 100 ml of artificial wastewater (Patil et al., 2010). An inflow and outflow rate of 10 ml d⁻¹ was kept constant over the course of the experiment. Ambient air was introduced continuously into the system through a 0.2-μm filter and diffused into the medium, thus allowing a steady mixing (~0.5 l h⁻¹).

T. pyriformis was cultured axenically (i.e., without any bacterial prey) in proteose peptone yeast extract medium (PPY, 20 g l⁻¹ proteose peptone, 2.5 g l⁻¹ yeast extract, Smith and Doerder, 1992) prior to the experiments. For each experiment, cells from the stock culture were washed by centrifugation for 10 min at 1,000 × g and resuspension in artificial wastewater. Approximately 1 × 10³ cells ml⁻¹ were added to the reactor at the start of an experiment (minimum amount: 517 cells ml⁻¹ and maximum amount: 1,133 cells ml⁻¹).

Phages were obtained after the infection of the host Klebsiella strain and subsequent cultivation in LB broth over night at 37°C. On the day of the experiment, the culture was filtered through a 0.2-μm filter to remove all the remaining host bacteria. At least 1.7 × 10⁵ PFU ml⁻¹ (maximum amount: 2.51 × 10⁶ PFU ml⁻¹) were added to a reactor at the beginning of an experiment.

B. bacteriovorus strain 109J was maintained as previously described (Jurkevitch, 2005) and with Klebsiella as host. On the day of the experiment, BALOs were passed through a 0.45-μm filter to remove the remaining prey cells and subsequently washed with artificial wastewater. At least 3.1 × 10³ PFU ml⁻¹ (maximum amount: 1.25 × 10⁶ PFU ml⁻¹) were added to a reactor at the beginning of an experiment.

Single colonies of bacterial prey cultures were re-suspended in 5 ml of LB and incubated over night at 37°C with gentle shaking. Cells were washed in equivalent amount of artificial wastewater. At the start of each experiment, 1.5 ml of prey culture adjusted to an optical density (OD₆₀₀) of 1.5 was added to each reactor. In experiments including all three prey organisms, 0.5 ml of each prey species were added to maintain an equal total volume.

Different predator–prey diversity levels (Figure 2) were established in triplicates, and experiments were run for 72 h. To control bacterial growth without predation, single prey (C I) as well as multiple prey controls (C II) were used in duplicates. This constrain in control replication was mainly due to the limited amount of reactors that we were able to run at the same time but was considered appropriate as the control reactors showed almost no variation.

Samples were taken every 24 h for a total of 72 h by injecting a sterile needle into a rubber septum located at the side of each reactor. For counting bacterial colony forming units (CFU), 200 μl samples were serially diluted and 20 μl drops (four times for each dilution) were incubated on LB plates over night at 37°C. Klebsiella and Pseudomonas colonies were distinguished on chromogenic E. coli/Coliform agar (Fluka). Since Staphylococcus colonies cannot grow on this medium, cell numbers were obtained by subtracting the number of Klebsiella and Pseudomonas colonies from the total number of colonies observed on LB plates. In addition, Staphylococcus appeared as distinctive yellow colonies on the LB plates; counting these colonies served as a control for the subtraction method. All plates were incubated overnight at 37°C and examined for changes of the colony morphology.

Protists were quantified in a Sedgwick-Rafter chamber with 500 μl samples which were fixed with 5% Lugol solution. At least 2 × 20 squares of the chamber were counted.

Phage and BALO numbers were quantified in plaque assays. Five hundred micro liter samples were filtered through a 0.2-μm filter (phages) and a 0.45-μm filter (BALOs), respectively. Phage plaque forming units (PFU) were obtained using the double agar technique with LB plates (1.5% agar) and LB soft agar (0.7% agar; Kropinski et al., 2009). Triplicates of multiple dilutions were plated for proper counting. Phage samples were diluted with saline (0.9%) and plates were incubated at 37°C overnight. BALO plaque assays were performed as described elsewhere (Jurkevitch, 2005). Plates were incubated at 30°C until plaque formation was observed.

To investigate fitness costs of acquired resistances, we recorded and compared the growth kinetics of prey bacteria for the ancestral bacterial strains (t = 0 h) and the evolved bacteria (t = 72 h). Single colonies of prey bacteria were incubated in 5 ml LB overnight at 37°C with gentle shaking to provide a starter culture. Growth measurements were then performed by preparing cultures with an initial OD₆₀₀ of 0.05 and then measuring OD every 30 min for 24 h. To connect the growth data with the colony doubling time, multiple dilutions of the respective prey culture were incubated on LB plates for 24 h and colonies were counted to calculate the prey CFU per ml of the culture used in the OD measurement.

We re-isolated Klebsiella cells, phages, and BALO cells at the end of the experiments from each reactor. Resistance of potentially evolved Klebsiella cells to ancestral phages or BALOs and of ancestral Klebsiella cells to potentially evolved phages or BALOs was tested in cross-infectivity tests (adapted from Friman and Buckling, 2014). A pure phage or BALO filtrate was obtained by filtering the sample through 0.2 or 0.45 μm filter, respectively. The phage filtrate was streaked out in lines on LB agar plates to dry. Ten micro liter of the respective bacterial overnight culture was streaked across the phage lines, cultivated at 37°C overnight, and subsequently checked for bacterial growth.

The BALO filtrates were tested against the respective overnight prey cultures in a plaque assay as described earlier. Assays with evolved Klebsiella strains and ancestral phages or BALOs were performed with at least 10 cultures from single Klebsiella colonies to calculate the percentage of resistant strains.

All statistical analyses were performed within the R software environment (R. Core Team, 2016) using the package “vegan” (Oksanen et al., 2007) or the included “stats” package. Possible factors accounting for the changes in the prey removal rates were analyzed with ANOVA as the variances in prey removal between groups of data were similar and normally distributed. The prey removal rate per hour was calculated as the average log slope between all sampling points. Using the aov() function, we tested the correlation between the degree of predator specialization (“generalist,” “semi-specialist,” “specialist,” “generalist + semi-specialist,” “generalist + specialist,” “semi-specialist + specialist,” or “generalist + semi-specialist + specialist”) and the prey removal rate per hour. The same function was used to correlate the number of additional predators (“none,” “one,” or “two”) and the predator removal rate due to normal distribution of the data and variance differences that still allow a robust analysis.

The differences between the BALO and phage growth rates with one additional predator were tested with the two-sample t-test as well as the differences in prey removal of all three bacteria individually in the presence or absence of a generalist predator, since the data were normally distributed and showed little variance differences. Significant differences in the generation times of the respective prey bacteria compared to the ancestral Klebsiella generation time were tested with the one sample t-test due to the normal distribution and low variance differences of the data.

All prey from the reactors containing the generalist protist predator showed a reduction in cell numbers to a more or less similar extent and independent of the presence of additional predators (Figures 2c,f,i,j,l). Both specialized predators reduced only the concentration of Klebsiella cells when all three prey species were present. Klebsiella went extinct in these assemblages after 48 and 72 h, respectively (Figures 2g,h,k), even though the BALO has the potential to grow on Pseudomonas (Figure 1). Reactors containing one or two specialists as well as the generalist predator showed reductions in the cell numbers of all three prey species. However, no extinction of any prey species was observed (Figures 2i,j,l), suggesting that the presence of a generalist predator allows for the coexistence of all prey bacteria and their predators. Contrary to reactors containing all prey strains (Figures 2f–h), the presence of only one predator and one prey (Figures 2c–e) did not lead to a similarly strong reduction of Klebsiella, indicating the appearance of resistant prey strains.

For all three prey species, independently we found a highly significant correlation between the prey removal rate and the degree of predator specialization [ANOVA: Klebsiella F_{(6, 14)} = 216.86, P < 0.001; Pseudomonas F_{(6, 14)} = 28.82, P < 0.001; Staphylococcus F_{(6, 14)} = 46.569, P < 0.001]. However, the relationship between specialization and prey removal rate differed for the three prey bacteria. Klebsiella removal by specialist and semi-specialist predators occurred at a higher rate than in the presence of a generalist (P < 0.001, two-sample t-test), whereas Pseudomonas and Staphylococcus removal rates were higher in the presence of the generalist than in the absence of a generalist (P < 0.001, two-sample t-test, Supplementary Figure 2).

Presence of additional predators had no effect on the growth of protists [Figure 3a, ANOVA: F_{(2, 9)} = 0.081, P = 0.923]. Phage proliferation was also not significantly influenced by additional predators [Figure 3c, ANOVA: F_{(2, 9)} = 1.037, P = 0.393]. Note, however, that our measurements of phage growth rate are likely to underestimate the true rate since phage particles are small enough to be washed out of the reactor.

We found a negative relationship between the number of predators and BALO growth [Figure 3b, ANOVA, F_{(2, 9)} = 6.597, P = 0.0172]. In addition, the range of growth rates of BALOs and phages in the presence of one additional predator was rather broad and depended on the type of predator present in the system (Figures 3b,c). Both BALO and phage growth rates were significantly lowered by the presence of the protist (Figure 3d, P = 0.044, two-sample t-test, and Figure 3e, P = 0.013, two-sample t-test, respectively).

As the prey dynamics of some reactors indicated the presence of Klebsiella strains resistant to predation (Figures 2d,e), we performed cross-infectivity tests to identify if bacteria developed resistance to the two specialized predators and if the predators subsequently adapted; the latter is implied by the loss of the ability to infect ancestral prey strains (Friman and Buckling, 2013).

Only the phage and Klebsiella from the reactors containing no additional predator or prey species showed evidence of potential short-term coevolutionary dynamics, i.e., bacteria became resistant against sympatric phages, whereas the phages showed reduced infectivity against ancestral Klebsiella (Figure 4a). We found resistant Klebsiella strains in reactors that contained protists, phages, and all three bacteria (Figure 4c), and in reactors containing all predators and prey species (Figure 4e). In contrast, we did not obtain any resistant Klebsiella strain in reactors containing either phages and all three prey bacteria (Figure 4b) or phages, BALOs, and all three prey bacteria (Figure 4d). Furthermore, none of the Klebsiella strains developed resistance against the ancestral BALO (Figures 4f–h). However, infectivity of BALOs from the reactors that only contained BALOs and Klebsiella was reduced (Figure 4f).

Unfortunately, we were not able to evaluate the resistance of ancestral Klebsiella to BALOs that were in the reactor with phages, as the presence of large numbers of phage plaques prevented the clear identification of BALO plaques.

Inability to infect ancestral prey by evolved predators was only observed when no additional predators and prey species were present (Figures 4a,f). The dynamics in these reactors indicate coevolution of the phage in response to resistant Klebsiella (Figures 2e, 4a) and potential coevolution of BALOs as response to a plastic resistance phenotype of Klebsiella (Figures 2d, 4f).

We compared the generation time of evolved and ancestral Klebsiella in order to identify the costs of acquired resistances in the fitness of the prey bacteria (Figure 5). In general, bacteria did not show reduced fitness in terms of a prolonged generation time. In contrast, we found a significant reduction in the generation time of Klebsiella strains from reactors that contained (i) additional prey species (P = 0.021, one sample t-test), (ii) the protist and Klebsiella (P = 0.004, one sample t-test), (iii) the phage and Klebsiella (P = 0.003, one sample t-test), and (iv) the protist, BALO, and all prey species (P = 0.033, one sample t-test).

We studied how generalist and specialist micro-predators influence the dynamics of a bacterial community consisting of three different species. Due to the predators' partially distinct prey range and a potential trophic complementarity, we expected the greatest prey removal in the presence of all three predators (Poisot et al., 2013). However, we did not find the number of different predators to be a trigger for increased prey removal, but rather the prey range of a predator. The generalist predator decreased all three prey species, whereas a specialist predator removed only one prey species, i.e., its preferred prey in case of the BALO and the only susceptible prey in case of the phage (Figure 2). Presence of the protist prevented the extinction of the most preferred prey species (Figures 2d,e,g). In particular, Klebsiella, which can be preyed upon by all predators, coexisted with the other prey bacteria in the presence of the generalist and specialist predators. In contrast, we found that the occurrence of predators of greater specialization and the concurrent absence of a generalist predator led to the extinction of the shared prey Klebsiella. This is contrary to the expected resource complementarity (Loreau et al., 2001; Hooper et al., 2002) that would posit conservation of the diversity. However, it seems that the BALO preferred Klebsiella over Pseudomonas despite its experimentally confirmed ability to infect Pseudomonas (Supplementary Table 1). Preferential predation by B. bacteriovorus 109J was also described for a mixture of two potential prey species resulting in significant shifts of each prey's abundance (Rogosky et al., 2006).

As expected, only the protist grazed upon all prey species and decreased their cell concentrations to a similar extent. This effect appears to be strong enough to dampen the effect of Klebsiella predation by the BALO and the phage and might, therefore, be the reason for the observed coexistence of all predator and prey species. Our results indicate that loss of a generalist predator can have a large impact on the diversity of a prey community and might lead to the extinction of prey species that are shared by more specialized predators. The strong stabilizing effects of generalist predators on predator–prey cycles were already described in a theoretical study using a reaction-diffusion-advection modeling framework (Vitense et al., 2016). In addition, a meta-analysis of 60 predator-exclusion experiments with different specialized aphid predators showed similar results (Diehl et al., 2013). Assemblages that contained either only specialists or specialists and generalists had the greatest effect on the reduction of the aphid population compared to an assemblage of only generalists.

Overall, the results indicate that predator–prey dynamics depend on the resource specialization of the predators. Predator–prey dynamics can either result in the extinction of a prey species in case of the specialist predator and Klebsiella or enable coexistence of all prey and predator species in the presence of a generalist predator.

We compared the growth rates of the different predators alone and in the presence of one or two more predators (Figure 3). As expected from its broad prey range, the protist growth rate was not affected by the presence of additional predators (Figure 3a). BALO and phage growth, however, significantly decreased in the presence of the protist (Figures 3d,e) and also in the case of BALO when both protist and phage were present (Figure 3b). This indicates that the protist either ingested free BALOs (Johnke et al., 2017), phages (Deng et al., 2014), or BALO- or phage-infected prey, or was the strongest resource competitor. Another possibility is an indirect effect of the overall prey density reduction by protist predation, which might have substantially lowered the resource availability for the BALO and phage and consequently reduced the growth of the latter. However, the shared resource bacterium Klebsiella was able to survive in reactors containing all three predators and we, therefore, can assume a continuous resource availability for the two specialists. Even though these results may indicate direct negative interactions between the different predators, such as, intraguild predation (Holt and Polis, 1997), further experiments are required to draw definitive conclusions. Independent of the ultimate cause, the presence of a protist had a negative impact on the growth of the more specialized predators possibly resulting in decreased predation pressure on Klebsiella and coexistence of all species.

We observed Klebsiella strains resistant to ancestral phage predation after 72 h in multiple reactors. Bacterial resistance to phage predation can be acquired by multiple mechanisms that either prevent phage attachment (e.g., blockage of phage receptors, production of extracellular matrix, or production of competitive inhibitors) or phage DNA entry (e.g., the Sis system); degrade phage genetic material after entry (e.g., restriction-modification systems); or induce death of phage-infected cells in order to protect the remaining population (e.g., abortive infection systems; Labrie et al., 2010). The mechanism responsible for the resistance of Klebsiella in our experiments and the mechanisms leading to the resistance of bacteria against BALOs are currently unknown. Resistant Klebsiella strains appeared only in reactors that contained only the phage and Klebsiella (Figure 4a), the protist and phage together with all prey bacteria (Figure 4c), or all predators and all prey bacteria (Figure 4e). One explanation for the occurrence of resistant Klebsiella strains in only those treatments might be the availability of resources that allow a bacterium to develop resistance, a theory adapted from Thingstad et al. (2014). This resource availability is reduced in reactors that include all prey bacteria due to resource competition between them. However, the presence of a generalist predator might counterbalance this effect by reducing the amount of competing bacteria. Contrary to another study that investigated the effect of protist and phage predation on resistance evolution, we did not observe a negative association between the appearance of resistant prey bacteria and the number of predators (Friman and Buckling, 2013). However, this study was performed with a single prey bacterium and this may explain the differences in our results. The authors argued that the reduced appearance of resistant prey in experimental set-ups that contained both predators vs. set-ups that only contained the phage was potentially governed by reduced prey densities due to protist predation. As an outcome, prey encounter rates with predators were lower and might have resulted in decreased mutation rates (de Visser et al., 1999). Less encounters between the phage and its prey decrease the selection strength for prey resistance and phage infectivity (Hochberg and van Baalen, 1998; Brockhurst et al., 2003; Lopez-Pascua and Buckling, 2008). Since the protist was able to graze on all three prey bacteria in our set-up, this density-mediated effect was most likely much weaker and might have had no influence on the resistance evolution of Klebsiella.

We did not find any BALO-resistant Klebsiella strains. However, BALO predation can result in a more resistant, but plastic prey phenotype (Shemesh and Jurkevitch, 2004), that can be rapidly lost when the bacterium is re-grown in the absence of BALOs. Unfortunately, our experimental setup required a BALO-free cultivation step for Klebsiella in order to obtain sufficient material for the sensitivity assay, and this step may have resulted in the loss of any resistance phenotype. This hypothesis is supported by the fact that we found BALOs that have lost the ability to infect ancestral Klebsiella strains in reactors that only contained BALO and Klebsiella. Plastic phenotypic resistance may, thus, explain the observation that Klebsiella cell numbers remained stable in the reactors containing BALO as the sole predator (Figure 2d). From the three replicate reactors containing BALOs and Klebsiella, BALO extinction during the last day of the experiment was observed in only one. This also suggests the development of resistance mechanisms by Klebsiella as there was no concurrent decrease in Klebsiella cell numbers. Similarly, we observed phages that lost infectivity of ancestral Klebsiella in reactors that only contained Klebsiella. However, the concentration of phages did not decrease even though a proportion of phages was most likely washed out of the reactors due to their small size. Selection, therefore, was not only by the appearance of resistant prey strains but also by the continuous loss of phage particles due to the washout.

The evolution of resistance in prey is often accompanied by trade-offs. A common trade-off for the development of bacterial defense mechanisms is a reduction in the competitive ability of bacteria, e.g., a decline in growth rate (Bohannan and Lenski, 1999; Bohannan et al., 2002; Meyer and Kassen, 2007). Surprisingly, we observed the exact opposite, namely a significant decrease in the generation time of some of the bacteria (Figure 5). All experiments were conducted in a resource-rich environment that may lead to selection for increased growth rates in small populations (Hairston et al., 1970). Exploitation of new resources had been already theoretically concluded to be a necessity for the evolution of bacterial strains under phage predation (Thingstad et al., 2015). It would be, therefore, interesting to see if the evolved Klebsiella strains would show reduced growth in less complex media. In one case, we saw both a significant reduction in Klebsiella generation time and resistance evolution of the Klebsiella in contrast to the usually observed trade-off. This unexpected result questions the benefit of resistance evolution for a fast growing strain. However, we found this to be the case only in Klebsiella from reactors that also contained phages. Here, predation pressure might have been particularly strong, since no other organisms were present. We, therefore, expect that phage–host encounters almost always led to an infection and the faster growth rate might have not induced a strong enough effect in order to prevent extinction. Another explanation might be related to what was observed by Hewlett (2015): bacterial resistance to a phage with an additional increase in growth rate was induced by a mutation within a protein that led to both resistance and an increased rate of sugar uptake (Hewlett, 2015). Interestingly, resistance without costs has already been shown in a number of studies on plants and in Daphnia (Bergelson and Purrington, 1996; Labbé et al., 2010).

Altogether, our results led to a theory of coexistence in a micro-predator system as described in the Supplementary Figure 3. The final cause that determines if a prey species indeed faces extinction seems to depend on the availability of enough resources. An environment containing multiple resource competitors can be made accessible by a generalist predator that reduces the overall density of competing bacteria. Our study indicates that coevolution of a specialist predator might be affected by the presence of additional competing predators in resource-rich environments. That is, the presence of specialist predators increases the predation pressure on shared prey, but only in the absence of a generalist predator. As a consequence, coevolution in multi-predator communities including a generalist might be less common or at least require more time due to competition effects between predators. Even though this hypothesis needs further testing, it could have a great impact on the design of augmentation trials used to combat pathogens in the context of agriculture or wastewater treatment. Further experiments over longer time scales and in more realistic environments should, therefore, be conducted.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.