The Functional Significance of Bacterial Predators

The word “predator” may conjure images of leopards killing and eating impala on the African savannah or of great white sharks attacking elephant seals off the coast of California. But microorganisms are also predators, including bacteria that kill and eat other bacteria.

such as Bdellovibrionales and Vampirovibrionales, increased in growth rate in response to added substrates at the base of the food chain, strong evidence of trophic control. This work provides quantitative measures of predator activity and suggests that predatory bacteria-along with protists, nematodes, and phages-are active and important in microbial food webs. KEYWORDS 18 O-H 2 O, Bdellovibrio, food webs, predator, qSIP, stable isotope probing, top-down control, trophic interactions B acteria that prey on other bacteria are too small to engulf their victims, yet they consume them no less ferociously. Members of the Bdellovibrionales attach to prey cells, penetrate the cell membrane, and then take up residence in the host cytoplasm, consuming cellular constituents while growing filaments and producing daughter cells that eventually lyse and kill the prey (1). Some bacterial predators have names that tell their mode of predation; Vampirovibrio (2, 3) and Vampirococcus (4) insert cytoskeletal protrusions, "fangs," which extract the cytoplasm from the attacked cell. Some members of the genus Cytophaga are "cell eaters" (5,6), and Lysobacter are "lysers of bacteria" (7). These and members of the Myxococcales are social organisms which hunt in packs (8,9). Many of these organisms can also subsist as saprotrophs and thus are facultative predators (10), in contrast to Vampirovibrio and Bdellovibrio, which are obligate predators (11). Most of what we know about the physiology, growth, and activity of predatory bacteria has been learned from laboratory studies because of the difficulty of measuring taxon-specific bacterial activity in situ.
Predators are thought to be functionally significant in microbial food webs, but quantitative estimates in situ have been very difficult to obtain. It is possible to use fluorescent markers and plate counts to estimate growth rates of predators in artificial media (12), but applying such approaches in the field is challenging. For example, it is known that phages prey upon cyanobacteria in rice paddy soils, but the rates of predation are unknown (13). Experimental manipulations of soil protozoa in mesocosm studies demonstrate the importance of these eukaryotic predators for nitrogen cycling (14) and for decomposition of plant litter (15), but the quantitative impacts on these ecosystem processes under field conditions are difficult to measure experimentally. Various environmental conditions also influence predator-prey interactions; changing moisture content alters soil connectivity, stabilizing or destabilizing predator-prey dynamics (16). It is important that predator activity and growth be measured under realistic and varied conditions.
Although protists (17), rotifers (18), nematodes (19), and phages (11,20,21) are thought to function as the dominant predators in microbiomes, predatory bacteria are common in both soil (8,22) and aquatic (23) systems. But beyond their common occurrence in these habitats, we know little of their activity in the wild, how rapidly they grow, their functional significance in food webs, and how they respond to enrichment at the base of the food web through substrate additions.
DNA sequencing and other 'omics techniques can provide detailed information on the composition and functional potential of the microbiome (24), but most measurements of in situ bacterial growth rates lack taxonomic resolution and are conducted at the scale of the entire microbial assemblage (25,26). Such aggregate measurements mask the contributions of genetically and functionally distinct populations. Even in macroscopic assemblages, taxa are known to vary in their influences on ecosystem processes (27). Techniques that combine isotopes and genetic sequencing hold promise for parsing the contributions of individual microbial taxa to interactions within microbial assemblages and to biogeochemical processes (28,29).
Here, we synthesized measurements using quantitative stable isotope probing (qSIP), a technique that quantifies the isotopic composition of DNA after exposure to an isotope tracer (30). qSIP with 13 C-labeled organic matter tracks the rate of labeled carbon assimilation into DNA, and qSIP using 18 O-water tracks the incorporation of oxygen from water into DNA. Recovery of the isotope tracer in taxon-specific DNA sequences reflects rates of growth and carbon assimilation of individual microbial taxa (28,31). The survey conducted here included qSIP measurements conducted in natural microbial assemblages from sites in North America, including 14 soils (1 arctic, 1 boreal, 11 temperate, and 1 tropical), along with one temperate stream ( Fig. 1,  Table 1. Inset shows a cluster of sites in Arizona (box scale is 1 by 1°).  Table 1). We evaluated this data set to compare rates of growth by predatory and nonpredatory bacteria, and their responses to substrate addition.

RESULTS AND DISCUSSION
Bacterial taxa identified as potentially predatory were detected at all sites and amounted to 7.4 6 6.0% of taxa detected at each site (median 6 standard deviation). We refer to these as "predatory bacteria" henceforth, acknowledging the limitations of that designation based on 16S rRNA sequence variation (see Materials and Methods). Most of the predatory bacteria detected were facultative, with 64.7% from the order Myxococcales, 16% from the class Cytophagia, and 9.2% from the order Streptomycetales; 8% were obligate predatory bacteria, with 7.0% from the order Bdellovibrionales and 1.0% from the order Vampirovibrionales.
Across all sites and experiments, predatory bacteria assimilated isotope tracer into their DNA at rates 23.1 6 7.0% higher than nonpredatory bacteria (meta-analysis, P = 0.002, n = 407; Fig. 2). Climate appeared to have little discernible influence on the differential isotope uptake between predatory and nonpredatory bacteria, with weak and nonsignificant relationships across sites for mean annual temperature (P = 0.336) and for precipitation (P = 0.738). Soil pH (P = 0.871) and soil water content (P = 0.165) also had no statistically discernible influence on the relative isotope assimilation between predators and nonpredators. Given the current design (15 sites), power may have been limited for detecting such environmental effects.
Predator identity significantly influenced isotope assimilation (P , 0.0001; Fig. 2); although both obligate and facultative predators assimilated the isotope tracers at rates higher than nonpredatory bacteria, the difference was larger for obligate (57.7 6 8.4%, P , 0.001) than facultative (17.6 6 7.1%, P = 0.019) predatory bacteria. Finer resolution revealed taxon-specific patterns, with especially high isotope uptake in the members of the obligate predator order Vampirovibrionales (2, 3) and in the genus Lysobacter, which is known to exhibit wolf-pack type predation (7)(8)(9). Isotope uptake was also higher in the Bdellovibrionales, Streptomycetaceae, and Cytophagia, whereas rates of isotope uptake for the Myxococcales, many of which are thought to function as saprotrophs (10), were similar to rates of nonpredators. The higher values Across the 15 sites, bacterial growth rates were log-normally distributed, with a median growth rate of 0.035 per day (d 21 ), and 95% confidence from 0.003 to 0.198 d 21 , a range consistent with past estimates (31). The difference in growth rates between predators and nonpredators was higher for obligate predators than for facultative predators (Fig. 3A). The pattern held for rates of C uptake from 13 C-labeled substrates; obligate predators had significantly higher C uptake than facultative predators and nonpredatory bacteria (Fig. 3B).
Adding a source of energy for heterotrophs, in the form of carbon substrates, disproportionately stimulated growth rates of obligate predatory bacteria, whereas responses were indistinguishable between facultative predatory and nonpredatory bacteria (Fig. 4). This indicates that higher productivity increases top-down (predatormediated) control in food webs, that added energy disproportionately flows to the predator trophic level, and that predators exhibit functional responses to shifts in prey resource availability. These findings are consistent with long-standing ecological theory that predicts the functional importance of predators increases with productivity (32)(33)(34), theory that also has support in macroscopic food webs (35,36), and are consistent with observations in polar ocean systems where boom-bust cycles suggest viral response to increased algal productivity (37). The similar response of obligate predators from phylogenetically distant clades (i.e., protebacteria Bdellovibrionales and cyanobacteria Vampirovibrionales) implies that the mode of feeding determines the response. As such, similar results may be expected for other obligate predatory clades such as the widely distributed marine clade OM27 (Deltaproteobacteria) and family Halobacteriovoraceae. Across all predator taxa, adding nitrogen and carbon together elicited a larger (P , 0.001) growth response (38.6 6 7.5%) than adding carbon alone (19.1 6 10.4%), indicating that carbon-nitrogen stoichiometry of resources affects energy transfer to predatory bacteria (38).
Our findings indicate that predatory bacteria are highly active in microbial food webs, synthesizing DNA with elements derived from added isotope tracers at rates Functional Significance of Bacterial Predators ® higher than nonpredatory bacteria, consistent with evidence from experimental microcosms (39). These results suggest that bacteria should be considered alongside eukaryotes and viruses as important predators in microbial food webs. Similarly, a recent metagenomic qSIP analysis using a 13 C-CO 2 tracer introduced via plant root exudates found that 13 C recovery in metagenomes associated with putative predator bacteria was comparable to the recovery in viruses and substantially higher than that in predatory eukaryotes (40). Slower growth might be expected if bacterial predators were inactive or dormant, as are many soil microorganisms (41). The results presented here indicate that bacterial predators grow, metabolize, and feed at higher rates than most bacteria in the soil food web and that predatory bacteria may exert top-down effects in microbial food chains. Though our analysis focused on predation, techniques that combine isotopes and gene sequencing can also quantify evidence of other ecological interactions in microbiomes and how they shape carbon flow and nutrient cycling in microbiomes. Multiple signatures of interactions among bacteria have now been identified (42)(43)(44), informing the use of qSIP, metagenomics, and traits to evaluate the functional significance of interactions in diverse microbiomes.
Element flux through the microbiome is central to its functioning, and results from macroecology show how ecological interactions-competition (45), mutualism (46), and predation (47, 48)-strongly influence those fluxes. The evidence presented here synthesizing isotope-enabled microbiome analysis couples predator identity and activity in situ and demonstrates that predatory bacteria are highly active in environmental microbiomes, more active than the average bacterial member. Patterns observed across the sites surveyed indicate that top-down trophic interactions are an active force that may structure the composition of element flow in microbiomes and clearly suggests the functional significance of predatory bacteria in microbial food webs.

MATERIALS AND METHODS
Atom fraction excess (AFE) values for 18 O and 13 C were extracted from qSIP measurements. AFE values were used to estimate bacterial growth rates based on 18 O assimilation from 18 O-labeled water, and the 13 C assimilation rate from 13 C-labeled organic substrates, using methods described in references 30, 49, and 50. All qSIP measurements involved parallel incubations with samples receiving either isotopically labeled (e.g., 97 atom % 18 O-H 2 O, 99 atom % 13 C-glucose) or unlabeled substrates (e.g., water with natural abundance 18 O or glucose with natural abundance 13 C). Incubations lasted for 7.1 6 1.8 days (average 6 SD). After each incubation, DNA was extracted and subjected to density separation via isopycnic centrifugation. Density fractions were collected, the 16S rRNA gene was sequenced, and the total  abundance of 16S rRNA gene copies in each fraction was quantified using qPCR. Quantitative stable isotope probing calculations were then applied to estimate the atom fraction excess 18 O or 13 C of each sequenced taxon (30,31).
Across the 15 sites, multiple qSIP measurements were conducted, including experiments within each site. Across all sites and experimental treatments, there were a total of 82 qSIP data sets, and each data set contained estimates of 18 O or 13 C AFE for hundreds of bacterial taxa from a particular site and under a given experimental treatment. The identities of bacterial taxa were used to assign taxa to bacterial groups known to be capable of predation or to nonpredatory taxa. Predators were assigned based on belonging to one of six bacterial taxonomic groups known to exhibit predatory behavior: Bdellovibrionales, Cytophagales, Lysobacter, Myxococcales, Streptomycetales, and Vampirovibrionales. We recognize that assuming these taxa are unambiguously predatory based on their taxonomic assignment is uncertain. In particular, the facultative groups are known to vary in substrate utilization; the designation of "facultative" acknowledges the range of feeding behaviors exhibited by large groups, such as the Cytophagales (51), Streptomycetales (52), and Myxococcales (51). Not all taxa in these groups have been documented to be predatory; we use such broad groups because finer divisions are not available for the trophic behaviors of these organisms. Also, our approach relies on taxonomic assignments based on 16S rRNA gene sequences, which can be unreliable for delineating species or strain (53). In 98% of cases, we were able to assign taxa to possible predator groups based on name occurrences in class, order, or family, the higher levels of taxonomic resolution where 16S rRNA gene assignments have been found to be more robust (54).
Growth rates were estimated using 18 O qSIP after accounting for potential differences in the sources of 18 O among organisms functioning at different trophic levels. qSIP-derived estimates of growth rate using 18 O-H 2 O begin with the observation that some of the oxygen in DNA is derived from the oxygen in water, so the assimilation of 18 O from water into DNA reflects its rate of replication, a proxy for cellular growth (55). Ribose sugars, nitrogenous bases, and phosphate (56) all acquire oxygen from water (55). Therefore, the DNA of predators will likely contain oxygen both from water in their growth environment as well as from cellular constituents of prey; these two potential sources of 18 O in predator DNA may or may not be additive.
To distinguish between these two sources, we compared 18 O versus 13 C enrichment in predatory taxa-since many of our SIP studies included treatments with both labeled water and labeled organic C substrates (Table 1). It is standard in food web studies using isotope tracers to treat the 13 C isotope composition of predator taxa as a conservative indicator of the 13 C composition of their prey (57). The qSIP data sets we evaluated included a subset of dual-isotope measurements, where both 18 O and 13 C were determined in parallel experiments with 18 O-labeled H 2 O and 13 C-labeled carbon substrates. These measurements occurred in separate incubations, with identical conditions and resource availability but with different isotope labels applied; in one case, 18 O water was added with a natural abundance carbon substrate, and in the other, the carbon substrate was 13 C-labeled, while the added water was at natural abundance 18 O. With these parallel measurements, we were able to estimate both the 13 C and 18 O for multiple taxa.
Across 5 sites and 12 experiments, there were 2,197 simultaneous measurements of 13 C and 18 O, including 2,060 cases of nonpredatory taxa and 137 cases of predatory taxa. We evaluated the relationships between 18 O and 13 C for both predator and nonpredator taxa, reasoning that the two sources of 18 O to predators (compared to one source for nonpredators) would result in predator DNA that was relatively higher in 18 O compared to 13 C, to the extent that these sources were additive. As expected, for a given value of 13 C, predator taxa had higher values of 18 O than nonpredator taxa (Fig. S1). We used the difference in the relationships (model II linear regressions) between 18 O versus 13 C for predators and prey (Fig. S1) to predict what the 18 O composition of predator taxa would have been based on growth on 18 O-labeled H 2 O alone. This approach resulted in the following correction, which was applied to all predator taxa in the data set: where 18 (31,55). An additional consideration is that oxygen concentration can affect 18 O assimilation from labeled water (58). Although oxygen concentrations were not measured in the incubations, for the mixed conifer, ponderosa, pinyon-juniper, and grassland sites included here, median final CO 2 concentrations were 0.31% (0.81%, 95th percentile) (59), which translates to a small change in atmospheric O 2 and suggests that oxygen depletion during the incubations was unlikely to have reached levels shown to affect 18 O assimilation from labeled water (58).
Experiments with 18 O were conducted by adding 97 atom % 18 O-H 2 O to the experimental system and incubating for several days. Because background levels of unlabeled water were present, the 18 O composition of water in each incubation was determined as a function of the amount of 97 atom % 18 O water added and the amount of background water. The relative growth rate for each taxon was estimated according to equation 7 from reference 31, using the AFE 18 O m of individual bacterial taxa, the AFE 18 O of water during the incubation, and the duration of the incubation in days.
We compared AFE, growth rates, and carbon assimilation rates of predatory and nonpredatory bacteria using meta-analysis (metafor package in R [60]), using the log ratios of predator:nonpredator as the Functional Significance of Bacterial Predators ® metric of difference between trophic strategies. This analysis was tested across all sites, treatments, measurement conditions, and tracers. Because some sites included experiments with both 18 O and 13 C tracers, isotope treatment was nested within sites to preserve independence. For all analyses, site was included as a random effect, because sites included multiple effect sizes which were not independent from each other. Computing multiple estimates with the same control group induces dependency on sampling errors, requiring the use of a variance-covariance matrix in the analysis (61). We computed the covariance in log response ratios as where SD C is the standard deviation of the control group, C is the mean, and N C is the sample size. We tested for the effect of predator identity on AFE, growth rate, and carbon assimilation rate. Predator identity was evaluated by taxonomic assignment and functional group-obligate predators (Bdellovibrionales and Vampirovibrionales) and facultative predators (Cytophagales, Lysobacter, Myxococcales, and Streptomycetales). The effect of predator identity was nested within experiment, because multiple predator groups occurred in the same data set, so their assimilation rates were not independent of each other.
We used a similar meta-analysis model to evaluate the influence of added carbon substrates on the relationship between growth rates of predatory and nonpredatory bacterial taxa. Twenty-four of the compiled qSIP data sets included experimental substrate additions, in which 18 O-H 2 O qSIP was conducted in soils amended with various carbon substrates compared to a control. Substrates included glucose (6 experiments), oxalic acid (2), ground plant litter (6), a mixture of glucose and ammonium (4), and a mixture of sugars, organic acids, and amino acids simulating root exudates (6). Across all substrate addition experiments and predator taxonomic groups, there were 113 log ratios comparing predator and nonpredator growth rates with substrates added, and 187 log ratios comparing predator and nonpredator growth rates without substrates added. (The compiled data set also included experimental manipulations of temperature and of leaf litter species, but the sample sizes were too small to evaluate these as potential drivers.) We evaluated the effect of substrate addition on the growth rates of predators using models with both predator identity and substrate as moderators.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only. FIG S1, PDF file, 0.3 MB.