18S rRNA metabarcoding diet analysis of a predatory fish community across seasonal changes in prey availability

Abstract Predator–prey relationships are important ecological interactions, affecting biotic community composition and energy flow through a system, and are of interest to ecologists and managers. Morphological diet analysis has been the primary method used to quantify the diets of predators, but emerging molecular techniques using genetic data can provide more accurate estimates of relative diet composition. This study used sequences from the 18S V9 rRNA barcoding region to identify prey items in the gastrointestinal (GI) tracts of predatory fishes. Predator GI samples were taken from the Black River, Cheboygan Co., MI, USA (n = 367 samples, 12 predator species) during periods of high prey availability, including the larval stage of regionally threatened lake sturgeon (Acipenser fulvescens Rafinesque 1817) in late May/early June of 2015 and of relatively lower prey availability in early July of 2015. DNA was extracted and sequenced from 355 samples (96.7%), and prey DNA was identified in 286 of the 355 samples (80.6%). Prey were grouped into 33 ecologically significant taxonomic groups based on the lowest taxonomic level sequences that could be identified using sequences available on GenBank. Changes in the makeup of diet composition, dietary overlap, and predator preference were analyzed comparing the periods of high and low prey abundance. Some predator species exhibited significant seasonal changes in diet composition. Dietary overlap was slightly but significantly higher during the period of high prey abundance; however, there was little change in predator preference. This suggests that change in prey availability was the driving factor in changing predator diet composition and dietary overlap. This study demonstrates the utility of molecular diet analysis and how temporal variability in community composition adds complexity to predator–prey interactions.


| INTRODUC TI ON
Characterization of predator diets and food web interactions is important to the understanding of community functioning and management of freshwater systems (Thompson, Dunne, & Woodward, 2012;Vaughn, 2010).
Quantifying the dietary composition of predator fishes in the context of relative prey availability in the environment is necessary to investigate the ecological relationships between predators and their prey and to determine predator-prey preference. Traditionally, diet analyses have been conducted through morphological identification of prey collected from predator gastrointestinal contents, but identification of diet contents using these methods is often inaccurate (Buckland, Baker, Loneragan, & Sheaves, 2017;Schooley et al., 2008). More recent molecular methods have been applied to overcome some of these shortcomings (Berry et al., 2015;Carreon-Martinez, Johnson, Ludsin, & Heath, 2011;Sheppard & Hardwood, 2005). Application of a molecular approach to quantify diet compositions of multiple species in a predator community, combined with data on prey resource availability, can lead to greater understanding of possible competitive interactions between predators as well as how changes in prey abundance affect these interactions.
Predator preferences and dietary overlap among predators are driven in part by the prey relative abundance. High abundance of prey leads to higher encounter rates and in some cases can cause abundant prey taxa to be targeted by predators (i.e., positive frequency dependence; Ims, 1990;Murdoch, 1969). High abundance of prey can reduce interspecific competition, allowing predators to coexist despite high dietary overlap (Gray, Boltz, Kellogg, & Stauffer, 1997;Kelling, Isermann, Sloss, & Turnquist, 2016;Michaletz, 1997).
Seasonal fluctuations in prey abundance and species composition are common features of riverine communities (Brown & Armstrong, 1985;Gray et al., 1997;Smith & King, 2005). Synchronized emergence and dispersal of larval fishes and aquatic macroinvertebrates may be an adaptive strategy to swamp predators, leading to periods where foraging predators are saturated by prey (Frank & Leggett, 1983;Ims, 1990). The seasonal influx of prey, comprised of the early life stages of river spawning fishes and emergence of certain aquatic macroinvertebrate taxa, can alter trophic interactions between predators and prey, and change the diet composition and dietary overlap of the predatory fishes in rivers. Understanding how the variation in the prey community affects these relationships is important to conservation, as predator preference can indicate what members of the community act as important energetic links between trophic levels (Chesson, 1978;Ivlev, 1961), and estimates of diet similarity between two predator species may indicate the degree of interspecific resource competition (Schoener, 1970).
DNA-based molecular methods are useful tools for analyzing the diets of fishes in freshwater food webs with greater accuracy and resolution than traditional morphology-based methods (Carreon-Martinez & Heath, 2010;Pompanaon et al., 2012).
Metabarcoding is one molecular technique, utilizing conserved regions of DNA to amplify sequences in samples that are unique in different taxa (King, Read, Traughott, & Symondson, 2008). Molecular techniques have advantages over morphological analyses of diets that require visual identification of partially digested prey items. Molecular barcoding is capable of identifying prey items to a greater taxonomic resolution and for longer periods after consumption (Berry et al., 2015;Carreon-Martinez et al., 2011;Schooley et al., 2008;Sheppard & Hardwood, 2005). The greater diet breadth and taxonomic resolution that can be achieved through metabarcoding diet analysis can allow characterization of the degree of niche partitioning among species, revealing how predators can partition resources to reduce interspecific competition (Albaina, Aguirre, Abad, Santos, & Estonba, 2016;Katzinel et al., 2015;Leray, Meyer, & Mills, 2015). Dietary overlap estimated by molecular methods could also be significantly different than nonmolecular studies estimated depending on the prevalence of soft-bodied prey items in predator diets (Gebremedhin et al., 2016;Soininen et al., 2015), which are often difficult to detect in morphological diet studies due to rapid digestion times (Carreon-Martinez et al., 2011;Ley et al., 2014).
In temperate streams in northern Michigan, USA, where this study was conducted, the period of high prey dispersal in the drift (mid-May to early June) is predominated by the emergence of larval suckers (Family: Catostomidae) and larval lake sturgeon (Acipenser fulvescens Rafinesque 1817), a species of conservation concern (Auer & Baker, 2002;Smith & King, 2005). This period also coincides with the emergence of several aquatic insects, including families Heptageniidae, Isonychiidae, and Perlidae (Scribner, unpublished data). This study examined associations between abundance of prey in the drift and the diet composition of predators that prey upon larval lake sturgeon. The goals of this research were to (a) characterize the diets of predatory fish during and after the high prey biomass drift period using metabarcoding molecular diet analysis, (b) measure dietary overlap between predator species during and after the drift period, and (c) quantify predator diet preferences and changes in preference using metabarcoding diet data combined with composition estimates from stream surveys of the prey community.

| Study area and sample collection
Sampling was conducted in the Upper Black River (UBR; Cheboygan County, MI, USA), the largest tributary of Black Lake, a 4,100 ha inland lake in the northern lower peninsula of Michigan. Black Lake supports a population of ~1,200 adult lake sturgeon (Pledger, Baker, & Scribner, 2013), which spawn solely in the UBR. Larval lake sturgeon disperse from the UBR in late spring, often coinciding with the outmigration of larval white suckers [Catostomus commersonii (Lacepède, 1803)] and silver redhorse [Moxostoma anisurum (Rafinesque, 1820)], and the emergence of several species of aquatic insects (e.g., Families: Heptageniidae, Isonychiidae, Perlidae), leading to a high abundance and diversity of available prey for predatory fishes in the system. This high prey abundance contrasts with the comparatively lower abundance of available prey present in the drift by mid-summer in the UBR.
Sampling of drifting prey was conducted during 2015 at four sites downstream of lake sturgeon spawning sites. Two sites consisted of predominately habitats composed of gravel substrate (Figure 1; PD1 and PD3), and two sites further downstream were located in habitats composed predominately of sand substrate (Figure 1; PD4 and PD5). Sampling dates were divided into two periods. The first period, "drift," occurred when larval lake sturgeon and catostomids were observed in survey samples. "Drift" samples were collected for five days during the lake sturgeon and catostomids drift period in 2015 (24 May, 4-7 June). The second period, "postdrift," occurred when larval lake sturgeon and catostomids were no longer observed in the survey samples. The "postdrift" period began 2 days after no larval lake sturgeon or catostomids were observed in the drift surveys and included drift sampling on two nights (3 July, 5 July). The abundance of drifting larval lake sturgeon and co-distributed larval fish and macroinvertebrate prey taxa was quantified using D-frame drift nets (Auer & Baker, 2002). Beginning at 21:00, five D-frame drift nets with 1,600 µm mesh and detachable cod ends were set at one of the sampling sites each night. To estimate the proportion of the river sampled by the drift nets, total river discharge (m 3 /s) and the discharge entering nets were measured using a Marsh McBurney Flow-Mate 2000 (Hach Company, Loveland, CO, USA). Contents of the cod ends were collected hourly between 22:00 and 02:00.
Larval lake sturgeon were counted on site and returned to the river. 5% subsamples of the cod end contents were collected for each hour and preserved in 95% ethanol. Sucker larvae and invertebrates in the preserved samples were later counted and macroinvertebrate larvae were morphologically identified to the family level. Dry weight biomass estimates for individual fish and aquatic insect larvae were collected for most families observed during drift sampling (Table 1).
These estimates or the estimate from a closely related family were used to estimate total nightly catch biomass by multiplying the nightly catch counts by the individual dry weight biomass for each taxon and adjusting for subsample size ( Figure 2).
Electrofishing surveys were conducted the day following drift sampling to collect diet samples of predatory fishes (n = 367 samples from 12 predator species). A barge electrofishing unit including a three-person crew sampled a 0.5 km stream transect directly downstream of the site where drift sampling was conducted the previous night ( Figure 1; Transects A, B, C, and D). Electrofishing voltage and amperage were set to 400 V at 4 A, respectively. Two crew members carried anodes and collected fish, and the third crew member moved the barge upstream and stored captured fish in a live well.
Predator fish were sacrificed with an overdose of MS222 (0.4 mg/ ml). Total length and species of all fish captured during the survey were recorded. Sacrificed fish were placed in Whirl-Paks (Nasco, Fort Atkinson, WI, USA) and stored in a −20°C freezer within 2 hr.
Predators were dissected, the entire GI tracts were removed, and contents were carefully extracted to minimize the amount of predator tissue in the sample. Diet samples were preserved in 95% ethanol and stored at −20°C prior to DNA extraction.

| DNA extraction and sequencing
Diet samples were mixed by hand, and pieces of tissue were broken apart with forceps and sterile toothpicks and thoroughly vortexed to homogenize the samples and to ensure representative subsampling. About 50-100 mg of tissue from the GI tract diet samples was used in each DNA extraction and washed with sterile water to remove excess ethanol. This was usually the entire diet sample. DNA was extracted according to a modified version of the QIAamp Stool Mini Kit (QIAGEN, Hilden, Germany) protocol. Lysis in InhibitEx Buffer from the QIAmp Stool Mini Kit was extended to 30 min at 72°C. Samples were also further homogenized with a 10-min beadbeating step using 0.70 mm garnet beads (MOBIO, Carlsbad, CA, USA) after lysis buffer and proteinase K were added to the sample.
DNA was eluted, and DNA concentration was quantified using an ND-1000 nanodrop spectrophotometer (NanoDrop Technologies Inc., Wilmington, DE, USA). If the nanodrop spectrophotometer revealed a high concentration of contaminants (260/280 < 1.7) in the sample, a salt precipitation using cold 100% ethanol and 0.15 M sodium acetate was used to clean samples. All samples were diluted using sterile water to a standard concentration of 20 ng/µl of DNA.
An empty microcentrifuge tube was used as a negative control for each extraction, and three negative controls were randomly selected for sequencing.
The coding region for 18S V9 rRNA (~200 bp; Stoeck et al., 2010) was amplified with universal eukaryotic primers 1391F (5′-GTACACACCGCCCGTC-3′; Lane, 1991) and EukB (5′-TGATCCTTCTGCAGGTTCACCTAC-3′; Medlin, Elwood, Stickel, & Sogin, 1988). PCR amplification of the 18S V9 region was carried out in 50 µl reactions using 20 ng of template DNA, 0.5 µmol of each primer, 200 µmol of dNTPs, 5 U of Taq polymerase, and 1X Taq reaction buffer (Invitrogen, Carlsbad, CA, USA). Reactions were amplified starting with an initial 5 min incubation at 95°C, followed by 30 cycles of 94°C for 30 s, 57°C for 45 s, and 72°C for 60 s before a final elongation step of 72°C for 2 min. These primers were chosen for their relatively short target sequence (~200 bp), the large taxonomic breadth encompassed, and because preliminary screening indicated sequences from lake sturgeon, suckers, and all major invertebrate families identified in UBR drift survey samples for the target region were available on GenBank (NCBI observed OTUs (Supporting Information Figure S1). Twelve samples with insufficient sequence numbers were discarded from further analysis.
OTUs appearing in the 2,500 most common unique OTU sequences were identified to the lowest taxonomic classification identifiable from matches on GenBank (>95% sequence similarity with 100% sequence coverage), with family being the lowest taxonomic classification used if lower classifications met these criteria. Only bilaterian DNA sequences were considered as potential prey items, as microbial sequences were more likely parasites or incidentally ingested. Prey taxa were divided into ecologically significant units (ESUs) based on the lowest taxonomic level that could be confidently identified. The number of OTU sequences from the same ESU was summed together within each sample. Sequences matching the identity of the predator the GI tract sample was taken from were removed from that sample, but retained in samples from predators of different taxa. Rarefied GI tract diet samples with <20 sequences (1% of sequences) from likely

| DNA sequence processing
Sequences were processed in mothur v 1.38 (Schloss et al., 2009). Similar paired-end reads (<2 bp difference) were merged to generate a list of unique sequences. Sequences were screened for quality by removing sequences that were longer than the target size after primer sequences were trimmed (>175 bp), unique sequences that appeared only once, sequences with homopolymer regions ≥8 bp, and chimera checking. Sequences were clustered into unique OTUs if there were ≤2 bp differences between sequences.
To standardize sequence sampling coverage between samples, all samples were rarefied to 1950 sequences. Rarefaction subsamples a consistent number of reads from each sample to standardize each sample to the same number of sequences while still accurately reflecting the relative abundance of each unique sequence present in the sample. Rarefaction curves for all samples were created to ensure the rarefaction did not artificially reduce the number of F I G U R E 1 Map of the study area highlighting the D-frame drift net survey sites of the prey community (black points; PD1, PD3, PD4, and PD5) and the 0.5 km predator electrofishing transects (bold gray lines; A, B, C, and D) in the upper Black River, Cheboygan County, MI. Transects A and B were characterized by gravel substrate and transects C and D were characterized by sand diet items were removed from the dataset. These samples were likely taken from fish with empty or near empty stomachs, so most reads came from their own tissue, ingested environmental DNA, or resident parasites in their GI tract. All remaining samples were standardized so sequences of each diet item were represented as the proportion of all prey sequences in a sample. TA B L E 1 Dry weight biomass (g) estimates for individual prey for each family represented in the D-frame drift net surveys and the estimated catch biomass of each prey family for each night. Some prey families were grouped together under the same ecologically significant unit (ESU), indicated in parentheses after the family name

| Examination of relationship between sequence count and biomass
To test for bias due to differences in copy number of prey rRNA or differential amplification of prey sequences by the 18S rRNA universal primers and to empirically demonstrate the relationship between relative sequence abundance and biomass for a number of prey taxa, a homogenate of invertebrate and fish tissue was created using 16 of the most abundant families collected during the drift survey. Preserved samples were removed from ethanol and air-dried for 24 hr. Three mixtures of roughly equal biomass from each of family were held at −80°C for 1 hr and mechanically homogenized with a mortar and pestle for 20 min (Table 2). When possible, insect heads and limbs were used to create the homogenate to avoid possible PCR inhibitors and contaminants in digestive tracts. Two 100 mg subsamples were taken from each mixture for DNA extraction using the same process as described previously. DNA sequencing of the drift homogenate was the same as described for the diet samples.
Sequences were processed in mothur v 1.38 with the same protocol described previously with the only difference being that sample were rarefied to 9,450 sequences (the number of sequences present in the sample with the fewest sequences) instead of 1,950.
Relative correction factors (RCFs) were calculated for each fam- (1) Mean percentages of biomass and sequences and the relative correction factor (RCF) for 15 families used to test differential amplification of the universal primers in homogenized mixed samples of tissue. RCFs >1 indicate a family was overrepresented by sequence abundance and RCFs <1 indicate a family was under-represented by sequence abundance

| Statistical analysis
Multivariate analyses were conducted in R Statistical Software v.
3.2.2 (R Core Team, 2015) using the vegan library (Oksanen et al., 2016). A principal coordinates analysis (PCoA) was conducted on the proportions of diet items in each diet sample using Bray-Curtis distances. Correlations between the original matrix of diet item proportions and the eigenvectors of the first two principal coordinates were calculated to analyze which prey items were explaining most of the variation in the diet. The first two principal coordinates were also plotted by predator species and by time period collected with 80% confidence intervals around each category.
To test the effects of predator species, sampling period, and substrate on the diet composition among members of the predator community, a PERMANOVA analysis was performed on the Bray-Curtis distance matrix of the diet proportions using the adonis function (Oksanen et al., 2016). Each PERMANOVA was run with 1,000 permutations. Predator species, sampling period, and substrate were all treated as fixed effects and all interactions among the fixed effects were analyzed. If the three-way interaction was not significant, the in-  (Schoener, 1970 Predator species  Chesson, 1983).
where m is the number of prey types in the environment and α i is the Manly's selection index for the ith prey type (Equation (4); Manly, 1991).
where r is the proportion of the ith prey taxa in the predator diet

| Relationship between biomass and sequence abundance
Sequences of larval fish and invertebrates were recovered from all subsamples of the homogenates collected during the drift surveys.
Samples rarified to 9,450 sequences contained 413 unique OTUs.

| Diet characterization
Predator diets contained between 1 and 10 diet items from different ESUs (mean = 3.8). The average proportion of reads from each ESU was calculated for each species of predator (Table 4). PCoA of the diet proportions revealed that diets segregated mainly by prevalence of a handful of prey items (Figure 4). High prevalence of mayflies versus fish and rotifers was the most important prey taxa contributing to the variation in diet composition across all species, explaining 21.1% of the variation (PC1; Figure 4). The second most important distinction in diet composition was between diets that contained more otomorph fishes (encompassing the cyprinids and catostomids observed in this study) compared to diets which contained more simuliid fly sequences (PC2; Figure 4).
PERMANOVA results indicated that there was no significant three-way interaction between substrate, sampling period, and predator species in influencing predator diets (pseudo-F = 1.04, p = 0.351). With the three-way interaction removed, there were significant interactions between the predator species and sampling period (pseudo-F = 1.46, p = 0.001; Figure 5a) and between substrate and sampling period (pseudo-F = 2.22, p = 0.013;

| Dietary overlap
Schoener's index of dietary overlap (α) was calculated for each pair of predator species during each time period (Table 7) Table 3 for three-letter predator fish codes.

| Prey availability and diet selectivity
The proportions of total drift biomass and the proportions of total reads in the predator GI tract samples for 15 prey ESUs were estimated for each night (Table 8). On average, prey biomass was higher during the drift period than during the postdrift period ( Figure 2; Table 1). Catostomid larvae and mayflies in the "Other Ephemeroptera" ESU (primarily the family Isonychiidae) were the most abundant prey by biomass during the drift period (mean nightly catch dry weight biomasses of 11.69 and 2.12 g respectively). During the postdrift period, Trichoptera and Plecoptera became the most abundant prey taxa (mean nightly catch dry weight biomasses of 1.09 and 1.66 g, respectively). Mean biomass of Trichoptera and Plecoptera did not change substantially from drift to postdrift periods; however, biomass of other prey taxa declined, largely due to emergence of several mayfly families throughout June.
Chesson's selectivity index value (ε) was calculated for each prey ESU by pooling reads by predator species for each day of sampling. PERMANOVA of the ε values indicate that there was only a significant interaction between sampling period and predator species affecting selectivity of prey items. PCA of the Chesson's ε distance matrix revealed diet preferences were largely consistent between periods, as there was no obvious difference in the distribution of predator references along the first PCA axis that explained most of the observed variation (PC1, Figure 6). There did seem to be a shift in some predator preferences toward baetid mayflies and away from other mayflies during the postdrift period (PC2, Figure 6).

| D ISCUSS I ON
Metabarcoding of the 18S V9 region of rRNA combined with field surveys of the prey community allowed quantification of changes in predator diets as the availability and taxonomic composition of prey changed. The taxonomic makeup of diets as characterized by metabarcoding analyses were largely concordant with the current knowledgebase for the diets of the predator species sampled in this study.
Sequencing analysis revealed that many of the riverine fish predators analyzed in this study had diverse diets, with nine or more prey taxa identified at least to the class level contributing at least 1% of the prey sequences within the diet samples of each predator species.
This high diversity observed in this study can largely be attributed to the ability of metabarcoding to detect quickly digested soft-bodied prey that are often difficult to identify in morphological analyses of diet samples (Albaina et al., 2016;Alonso et al., 2014;Moran, Orth, Schitt, Hallerman, & Aguilar, 2016;Sakaguchi et al., 2017).
Combining the dietary composition data with information on the abundance of prey taxa provided insight into how changes in the prey community affected the dietary patterns of predators.
There was a significant shift in the diet composition of the predator community as a whole as the overall biomass of the prey community decreased and the relative abundances of prey taxa changed between drift and postdrift periods. Dietary overlap between predator species decreased as prey decreased in abundance, possibly due to niche partitioning to avoid intense competition for less abundant prey resources. Finally, while dietary composition changed, predator preferences remained stable. Predator preferences were not dependent on the availability of prey in the environment. High biomass of a few prey taxa preferred by many predators seemed to drive the relatively higher diet overlap during the drift period. As these preferred prey taxa declined in abundance and the prey community was not dominated by a few taxa, differences in predator preferences drove the reduction in diet overlap seen in the postdrift period. Including surveys of prey communities with diet analysis can greatly improve metabarcoding diets studies, both from a technical stance by enabling researchers to test for bias in their molecular assays, but also by providing a greater ecological context to interpret diet composition and dietary shifts.

| Diet characterization
While diets of riverine fish predators were more diverse based on metabarcoding analysis than morphological diet analysis of these species has previously recorded, the identities of the most prevalent prey items were largely consistent with previous dietary Diets of minnows and chubs (Family: Cyprinidae) were quite diverse, but largely characterized by small dipteran larvae, consistent with past studies (Johnson, 2015;Quist, Bower, & Hubert, 2006).
Central mudminnow [Umbra limi (Kirtland, 1841)] exhibited a diet similar to other analyses that focused primarily on midge larvae crustacean zooplankton, and mollusks (Chilton, Martin, & Gee, 1984;Martin-Bergmann & Gee, 1985). White sucker were shown to prey largely on ostracod crustaceans, a common prey item observed in other studies (Ahlgren, 1990), but the molecular diet analysis also revealed that white sucker in the UBR may also engage in piscivory or consume the eggs and larvae of spawning bass, perch, and darters, which has also been observed in other studies (Baldridge & Lodge, 2013 was not analyzed in this study, so findings do not take into account ontogenetic diet shifts that may occur in these species (Amundsen et al., 2003;Dauwalter & Fisher, 2008;Paterson, Drouillard, & Haffner, 2006) that could explain dietary breadth. The smaller size classes of these species may prey heavily on aquatic macroinvertebrates while larger fish account for most of the piscivory seen in the data from this study (Amundsen et al., 2003;Dauwalter & Fisher, 2008;Paterson et al., 2006). As a result, the diets of some size classes may overlap more strongly with other species.
Additionally, cannibalism and consumption of other Perciformes are a component in the diets of rock bass and smallmouth bass (Clady, 1974;Frey, Bozek, Edwards, & Newman, 2003) (Dauwalter & Fisher, 2008;Paterson et al., 2006). Adult crayfish were present in the UBR during the entire sampling period, but most predator fish sampled for this study would have been too small to prey on adult crayfish. Juvenile crayfish were a significant part of the drift in mid-late June (data not shown), but numbers had dramatically decreased by July when sampling during the postdrift period occurred.  (Cardona, 2001;Jacobs, Madenjian, Bunnell, & Holuszko, 2010;Raborn, Miranda, & Driscoll, 2004). Niche theory actually predicts that the high abundance of prey reduces interspecific competition pressure, allowing predators to utilize the same resources (Pianka, 1974(Pianka, , 1976Schoener, 1974).
No predator species exhibited a high degree of dietary overlap during the postdrift period, possibly indicating niche partitioning among the predator species when prey became relatively scarce (Gray et al., 1997;Raborn et al., 2004).
Niche partitioning as prey becomes scarce would explain the patterns seen in the predators that exhibit the greatest shifts in diet between the drift and postdrift periods. Blackside darter, logperch, and rock bass had significantly altered diets between the two periods (Figures 5b-d) Predator preferences for certain prey items were not significantly different between the drift and postdrift periods ( Table 9).
Most of the variation in predator preferences was associated with preferences for small-bodied fly larvae (Families: Chironomidae, Simuliidae) or larger macroinvertebrates (Orders: Coleoptera, Plecoptera, Trichoptera; PC1, Figure 6). Biomass for these groups remained stable or slightly declined in the postdrift period. The largest change in predator preferences came from reduction in preferences for nonbaetid Ephemeropterans (PC2, Figure 6), which faced a much steeper decline in biomass, to preference for baetid mayflies, which contributed more biomass to the drift during the postdrift period (

| Additional considerations
Analyses in this study were conducted assuming that the number of sequence reads in a diet sample was proportional to prey OTU biomass. Evidence from other studies suggest that the number of TA B L E 9 Results of PERMANOVA analysis testing effects of predator species (n = 13), substrate (sand or gravel), and time period (during or after drift) on the predator preferences for prey observed in the drift sequencing reads is generally a good approximation of the relative biomass of organisms in a sample (Clarke, Beard, Swadling, & Deagle, 2017;Elbrecht & Leese, 2015;Evans et al., 2016;Hänfling et al., 2016), including with the same set of universal primers used in this study (Albaina et al., 2016). However, the relationship between biomass and number of sequence reads can be highly variable among taxa due to amplification bias of the primers (Albaina et al., 2016;Elbrecht & Leese, 2015). Amplification biases are heavily dependent on the primers and prey taxa in a study, and this study showed that the sequencing procedure consistently overrepresented some taxa (e.g., Family: Cambaridae) and under-represented others (e.g., Family: Perlidae) relative to biomass. Additionally, taxonomic resolution of the prey items could be further improved through the use of different sets of barcoding primers targeting different regions (Albaina et al., 2016;Hänfling et al., 2016).
Although metabarcoding can detect a wide array of prey items, there are some drawbacks. For example, incidental consumption of environmental DNA in the water by predators and secondary predation (detection of prey of prey) can be mistaken as predation on some prey taxa (King et al., 2008;Pompanaon et al., 2012). It can be difficult to determine whether fish are actually targeting some prey items (e.g., rotifers) or whether the sequences from those taxa are showing up in fish diets because aquatic insects or other prey items were consuming certain prey taxa. Furthermore, because metabarcoding relies on unique sequences to detect prey items, prey with the same DNA sequence as the predator cannot be distinguished from predator sequences (King et al., 2008). Therefore, incidence of cannibalism or predation on related species was not represented in predator diets, which could particularly affect estimates of rock bass and smallmouth bass diets in this study. Bass were observed to prey upon darters, and likely prey on other centrarchids (Dauwalter & Fisher, 2008), but all of those fishes have indistinguishable sequences at the 18S V9 region used in this study. Using a primer targeting a more variable sequence in fishes (e.g., cytochrome oxidase I) would improve the taxonomic resolution and make a more comprehensive analysis of piscivory in these predators possible (Trebitz et al., 2015). Likewise, cannibalism has been shown to be an important component of burbot diets (Jacobs et al., 2010), but could not be detected using metabarcoding techniques employed.
The relative abundance of the prey community estimated from the drift survey could have been biased and may not have represented the true availability of prey in the UBR. D-frame drift nets were deployed to maximize the catch of larval lake sturgeon (Auer & Baker, 2002;Smith & King, 2005), so drift surveys might have overestimated the abundance of taxa with similar benthic drifting behaviors.
Prey that drifted near the surface (e.g., catostomid larvae ;Corbett & Powles, 1986) are likely under-represented in the prey community relative abundance data. Prey taxa that do not drift (e.g., Unionidae) were too small to be sampled by the 1,600 µm mesh of the D-frame drift nets (e.g., Rotifera) or could escape from the drift nets (e.g., Perciformes) were not represented in the prey community relative abundance estimates. Only 15 of the 33 ESUs identified in this study were represented in the drift survey. As a result, the selectivity values based on the prey community composition should only be interpreted in the context of the 15 ESUs identified in the drift surveys.

| CON CLUS IONS
The 18S V9 rRNA metabarcoding approach implemented in this study shows promise as a powerful tool to investigate the diets of freshwater predatory fishes, especially if combined with other primers targeting more specific groups of taxa. Diet items could be identified to similar taxonomic levels as morphological diet analyses, with the potential for metabarcoding to have even higher taxonomic resolution as more sequences and longer reads become available.
Metabarcoding also revealed that predator diets were more diverse than previously thought, detecting predation on taxa such as larval fishes and rotifers that are unlikely to be accounted for using morphological diet analysis (Carreon-Martinez et al., 2011;Hunter, Taylor, Fox, Maillard, & Taylor, 2012;Ley et al., 2014).
This study also demonstrated how fluctuating seasonal abundance of drifting aquatic insects and larval fishes can impact predatory fish diets (Correa & Winemiller, 2014;Michaletz, 1997;Raborn et al., 2004;Sánchez-Hernández et al., 2017). High resource abundance could lead to a release from competitive pressure and reduce the niche partitioning expected under interspecific competition (Pianka, 1974). Seasonal drift serves as an important influx of energy and nutrients into riverine systems and as a competitive release for certain species, allowing them to utilize preferred prey resources without having intense resource competition from other predator species. The combination of more representative diet analysis using metabarcoding and the sampling of diets at very different periods of prey availability allow for a more complete understanding of the trophic links within complex riverine ecosystems.
The analyses conducted in this study suggest that seasonal changes in prey abundance and composition are mirrored in the diet compositions of predators and cause changes interactions between predator species. Altered flow regimes and climate change often lead to homogenization of stream habitats and lowered temporal variability, which disrupt natural macroinvertebrate and fish communities (Bunn & Arthington, 2002;MacNaughton et al., 2017;Mustonen et al., 2018) and spawning activity of fish that contribute large amounts of biomass and nutrients to river systems (Auer, 1996;Grabowski & Isely, 2007). The loss of seasonal variation in environmental conditions would likely lead to a reduction or elimination of the seasonal variation in prey community structure and abundance as seen in this study. Seasonal variation in food web structure appears to have an effect on the overall stability of riverine communities by reducing predator reliance on the presence of a certain prey resource (Saavedra et al., 2016. This study suggests seasonal variation may also be important by temporarily reducing competition between predators for preferred prey resources. Maintaining natural temporal variation and diversity of prey communities could be an important component in conservation of riverine ecosystems.
More research on the prevalence and effects of seasonal food web structural variation on the resilience of riverine communities is warranted. Jeanette Kanefsky provided laboratory and sequence processing support. James Bence, Brian Maurer, Gary Mittelbach, and Ed Baker, provided critical reviews of earlier drafts.

CO N FLI C T O F I NTE R E S T
None declared.

AUTH O R CO NTR I B UTI O N S
JMW and KTS designed the study and collected samples. JMW conducted laboratory work, performed statistical analyses, and wrote the paper with input from the other authors. TLM contributed new reagents and processed DNA sequence data.

DATA ACCE SS I B I LIT Y
FASTA sequences, datasets, and R code used to conduct the analyses in this manuscript are available on Dryad, https://doi.org/10.5061/ dryad.0jm1dt2