Integrating complementary methods to improve diet analysis in fishery‐targeted species

Abstract Developing efficient, reliable, cost‐effective ways to identify diet is required to understand trophic ecology in complex ecosystems and improve food web models. A combination of techniques, each varying in their ability to provide robust, spatially and temporally explicit information can be applied to clarify diet data for ecological research. This study applied an integrative analysis of a fishery‐targeted species group—Plectropomus spp. in the central Great Barrier Reef, Australia, by comparing three diet‐identification approaches. Visual stomach content analysis provided poor identification with ~14% of stomachs sampled resulting in identification to family or lower. A molecular approach was successful with prey from ~80% of stomachs identified to genus or species, often with several unique prey in a stomach. Stable isotope mixing models utilizing experimentally derived assimilation data, identified similar prey as the molecular technique but at broader temporal scales, particularly when prior diet information was incorporated. Overall, Caesionidae and Pomacentridae were the most abundant prey families (>50% prey contribution) for all Plectropomus spp., highlighting the importance of planktivorous prey. Less abundant prey categories differed among species/color phases indicating possible niche segregation. This study is one of the first to demonstrate the extent of taxonomic resolution provided by molecular techniques, and, like other studies, illustrates that temporal investigations of dietary patterns are more accessible in combination with stable isotopes. The consumption of mainly planktivorous prey within this species group has important implications within coral reef food webs and provides cautionary information regarding the effects that changing resources could have in reef ecosystems.


| INTRODUC TI ON
Prey acquisition is a fundamental biological process that drives development and behavior (e.g., growth, reproduction, foraging) of individuals, and contributes to population-level characteristics (e.g., migration, trophic position, habitat selection). Prey selection and availability can also have ongoing and multiplicative ecological effects within an ecosystem (e.g., trophic cascades; Estes et al., 2011) because consumers are often resource-limited or have overlapping dietary preferences (Ross, 1986;Sale, 1977). Empirical diet data help quantify the relative importance of prey items and characterize ecological interactions (e.g., resource partitioning, trophodynamics, competition) that occur within and among species (Connell, 1980;Schoener, 1974).
For fishes, there are several ways to identify or quantify diet.
There are also several considerations in selecting methods to characterize diet. These vary on a case-by-case basis and the goals of the research, but are constrained by the cost of approach, lethal vs nonlethal sampling, number of samples/individuals required, necessity of repeat sampling, and/or resolution provided by approach (e.g., temporal or identification resolution). One of the most direct methods is a visual examination of identifiable prey from stomach contents, and while this provides a snapshot of feeding (e.g., hours-days), digestion limits identification, stomachs are often empty, and large sample sizes and lethal sampling are generally required (St John, 1999;Vinson & Budy, 2011). However, advances in molecular approaches provide a potential alternative to visual stomach content analysis (Carreon-Martinez, Johnson, Ludsin, & Heath, 2011;Leray, Meyer, & Mills, 2015). The ability to sequence prey items from degraded stomach contents enhances diet data and has the capacity to reduce inefficiencies caused by unidentifiable samples. Nevertheless, this metabarcoding approach is still limited by the completeness of reference sequence databases and the choice of genetic markers (Devloo-Delva et al., 2018); consequently, prior validation is needed for newly studied species/systems. Another method to characterize diet is stable isotope analysis (e.g., δ 15 N and δ 13 C), a biogeochemical indicator of prey assimilation in the tissues of consumers (see Newsome, Clementz, & Koch, 2010 for review). Due to different metabolic processing within tissues, the timeline (or turnover) representing prey assimilation varies depending on the tissue sampled. For example,  found that 50% incorporation times (or 50% turnover) of δ 15 N in plasma, red blood cells (RBC), and muscle tissues of the predatory coral reef fish Plectropomus leopardus, were 66, 88, and 126 days, respectively. As δ 15 N and δ 13 C values change from prey to consumer by conserved amounts, the identity (e.g., species, family, habitat) and relative importance of different prey sources can be estimated (e.g., mixing models; Chiaradia, Forero, McInnes, & Ramírez, 2014). This approach requires methodical sampling of potential prey items, and standardization of assimilation parameters (e.g., diet-tissue discrimination factors) that may not exist for that species; thus, it often requires additional sampling/testing over other methods. Stable isotopes also reflect assimilation patterns of often confounding dietary sources over relatively long periods of time and therefore is a representation of broad-scale patterns (i.e., does not necessarily identify exact prey) over the temporal scale pertinent to the tissue sampled. Each method for analyzing diet includes limitations; a combination of approaches has the potential to provide greater resolution and clarity at multiple spatial and temporal scales.
The first objective of this study was to compare three dietary sampling approaches (i.e., visual, genetic, stable isotope analysis) to identify the advantages and weaknesses of each technique in isolation and combined. A congeneric group of coral trout (Plectropomus spp.), were selected because they are widespread mesopredators found throughout the Indo-Pacific with significant fishery value (Sadovy de Mitcheson et al., 2013).
Multiple past studies using visual stomach content analysis have shown that the diet of adult P. leopardus, the most abundant Plectropomus species in the Great Barrier Reef Marine Park (GBRMP) in Australia, consists of >25 prey families, but is mainly comprised of Clupeidae, Pomacentridae, and Labridae (Kingsford, 1992;St John, 1999). Dietary comparisons between sympatric Plectropomus are of interest because they can reflect competitive interactions or niche partitioning, which can help elucidate small-scale distributional patterns and capacity for hybridization (e.g., Harrison et al., 2017). However, dietary comparisons between sympatric Plectropomus are scarce; isotopic (δ 15 N and δ 13 C) niche differed between P. laevis and P. leopardus (Matley, Tobin, Simpfendorfer, Fisk, & Heupel, 2017), and P. maculatus and P. leopardus (Frisch, Ireland, & Baker, 2014) at reefs off Townsville and Northwest Island, respectively. However, isotopic niche between P. maculatus and P. leopardus was similar at Orpheus Island Reef . Examination of stomach content has yet to be completed for Plectropomus species in sympatry. Therefore, the second objective of this study was to identify and quantify the composition of prey consumed by Plectropomus spp. to explore niche segregation and further inform on prey consumption patterns of an iconic species group.

| Visual stomach content identification
Stomachs were removed upon collection and frozen (−20°C).
Stomachs were thawed, dissected, and prey items classified based on the digestion level (1-4 = low-high digestion: 1-little or no digestion except superficially, for example, skin and fins; 2-moderate digestion with head and tail mostly digested and possibility of parts broken off and oval fleshy remains; 3-major digestion with small fleshy remains and abundance of broken parts; 4-complete digestion with very small fragments of prey remaining or empty stomach and clean lining). Prey (digestion level 1 and 2) were weighed (0.001 g) and identified to the lowest taxonomical level possible using Randall, Allen, and Steene (1997) and Froese and Pauly (2016).
An additional 81 stomachs were collected for visual stomach con-

| Stable isotope analysis
Stable isotope sampling procedures and quantification followed Matley et al. (2017). Briefly, three tissues (plasma, red blood cells, and muscle) were collected from Plectropomus individuals and frozen (−20°C) until processing. Muscle tissue (no skin) was sampled from the dorsal musculature using sterile forceps and scalpel, and blood components were sampled from the 2nd or 3rd gill arch with a sterile needle/syringe. Frozen samples were freeze-dried for 48 hr and ground into a powder, then samples were lipid-extracted using

| Data analysis
Unless indicated otherwise, samples from each species were pooled between reefs and dates due to the limited number of individuals sampled. Previous research indicated different color phases of P. laevis (bluespot and footballer) have different feeding ecology (Matley et al., 2017); therefore, color phases were analyzed separately. Prey items were grouped by family when visually identified due to low numbers. The family Labridae was subdivided into Scarinae and "all others" because of the different feeding modes exhibited (e.g., To investigate whether DNA-identified stomach contents included a sufficient number of samples to formally analyze, the cumulative number of new prey families within each consecutive stomach sampled (randomly ordered) was plotted for each species using the specaccum function within the "vegan" package (Oksanen et al., 2016) in the R environment (R Development Core Team 2014).
Samples were considered adequate to characterize the diet if curves approached an asymptote (Ferry & Cailliet, 1996).
Comparison of DNA-identified stomach contents among species and color phases was facilitated by nonmetric multidimensional scaling (nMDS) based on the presence/absence of prey families using the Bray-Curtis dissimilarity index within the "vegan" package (Oksanen et al., 2016). An analysis of similarity (ANOSIM) tested for significant differences among species and color phases (reefs and sampling periods pooled separately for TSV and OI reefs); a global R-statistic value between −1 and +1 was produced with an associated significance level (α = 0.05). More positive R-statistic values indicate betweengroup differences, whereas values close to zero indicate random grouping (i.e., within-and between-group dissimilarities are indistinguishable). The degree of DNA-based dietary overlap between species was tested using the simplified Morisita index and Plectropomus species combinations with values above 0.60 were considered to have significantly overlapping diets (Langton, 1982). Differences in DNA stomach contents between TSV reefs (all species and sampling periods combined) and sampling periods (all species and TSV reefs combined) were also tested by ANOSIM as described above.
In addition, prey family composition was plotted after Plectropomus

| RE SULTS
Of the 226 stomachs visually examined 100 (44%) contained prey, 31 (23-P. leopardus; 5-P. laevis; 3-P. maculatus) had identifiable prey items (39 different items), which were identified to family or lower (11 of these identified to species). Caesionidae, Labridae, and Pomacentridae were the main prey families and comprised ~80% of identified prey (Table 2; Supporting information Figure S1) and at least one of these families was found in ~71% of individual stomachs with identifiable prey.
Of the stomachs (n = 101) sampled for genetic metabarcoding of prey (Table 1), 187 prey items (digestion level 1: n = 41, 2: n = 33, 3: n = 68, 4: n = 45) from 81 individuals (40-P. leopardus; 32-P. laevis; 9-P. maculatus) were identified which included 50 species from 20 families (Supporting information Table S1; Supporting information Figure S2). Cumulative prey curves for P. leopardus and P. laevis approached asymptotes at ~20-25 samples, suggesting sufficient samples to characterize diet (Supporting information Figure S3). The footballer phase of P. laevis had <20 samples but was treated separately from the bluespot phase due to previous investigations indicating distinct feeding ecology. Likewise, sample sizes for P. maculatus and P. leopardus at OI Reef were not adequate, Acanthuridae (Bierwagen et al.-in press). Prey selection patterns, as determined by Jacobs' Electivity Index showed selection for Labridae (not including Scarinae) for all Plectropomus at TSV reefs ( Figure 4). Also, no strong selection or avoidance patterns were readily apparent for Pomacentridae despite its high abundance. Otherwise, the bluespot P. laevis selected for Siganidae, Serranidae, and Lutjanidae, whereas P. leopardus demonstrated an affinity to Lethrinidae ( Figure 4); however, these families contributed only a small portion within the diet of Plectropomus (Figure 2).
Caesionidae and a few other families found in the stomachs of Plectropomus were not included in these abundance surveys and were not included in this analysis.
Prey contribution based on stable isotope mixing models using DNA-identified stomach content as prior information mainly consisted of Caesionidae and Pomacentridae for all species and tissues ( Figures 5 and 6). At TSV reefs, due to similar isotopic values between these two prey families it was difficult to distinguish their contribution values. Nevertheless, both comprised >60% of diet in P. leopardus and P. laevis for all tissues sampled ( Figure 5). Prey composition overlap between mixing models with and without prior information was significant (>0.60 index) for all tissues of P. leopardus (at TSV reefs and OI Reef), P. maculatus, and P. laevis (footballer); however, prey composition differed for P. laevis (bluespot) (all tissues). The main difference between mixing model outputs at TSV reefs was that Pomacentridae and Caesionidae contributed less to the diet when prior information was not included in mixing models, particularly for bluespots which generally showed a greater input of benthic consumers such as Scarinae, Acanthuridae, and Siganidae (Supporting information Figure S7). At OI Reef, Serranidae contributed a larger portion of the diet when prior information was not considered (Supporting information Figure S8).
Prey composition estimated from spatially and temporally equivalent DNA-identified stomach contents and stable isotopes was similar ( Figure 7). DNA-identified stomach contents from August and  Note. Stomachs of 226 individuals (-171-P. leopardus, -43-P. laevis, -12-P. maculatus) were examined but only 31 had stomach contents identifiable to family. Prey weight was not adjusted for partial digestion, and consequently, %W and %IRI are likely underestimated for some families.

| Methodological implications
Visual stomach content analysis is typically an affordable approach to identify prey but relatively labor-intensive and limited by biases associated with digestion rates, regurgitation of prey, and empty stomachs (Arrington, Winemiller, Loftus, & Akin, 2002;Vinson & Budy, 2011).
These biases can be problematic when interpreting diet for large piscivores because a wide variety of prey is often consumed heterogeneously in space and time (Armstrong & Schindler, 2011 (Kingsford, 1992;St John, 1999). Unless sampling can be conducted on many individuals (e.g., >20-25 individuals with identifiable prey per sampling category), visual stomach content analysis alone may be impractical for fishes with conservation concerns such as Plectropomus.
The use of molecular approaches, especially next-generation sequencing (NGS) barcoding, to identify prey of fishes is relatively new.
However, these methods are increasingly utilized to identify prey and explore ecological implications of diet (e.g., Leray et al., 2013Leray et al., , 2015. Here the molecular approach identified prey in ~80% of individuals, including stomachs that were qualified as empty by visual analysis. Likewise, Barnett, Redd, Frusher, Stevens, and Semmens (2010)  Stable isotopes are now readily used as an alternative or supplement to stomach content analysis. The specific advantage is that broad-scale feeding patterns reflecting habitat and prey sources can be inferred at multiple temporal scales (Newsome et al., 2010). In addition, lethal sampling is not necessary and there is no bias associated with digestion rates or empty stomachs (Colborne, Clapp, Longstaffe, & Neff, 2015). A major limitation with isotope analysis is to obtain a comprehensive view of diet, isotopically distinct prey species typically need to be sampled, which can be difficult and inflate costs (~15-30AUD per sample). Here mixing models were difficult to statistically compare with stomach content results because of the greater temporal scale associated with tissue-specific isotopic assimilation. Thus, this study is unable to specifically compare prey composition among the different approaches because diet manipulation and standardization were not completed. In addition, not all prey types detected in the stomachs were sampled for stable isotope values. Nevertheless, based on the corresponding temporal proxies of diet between February 2014 muscle tissue and August 2013 DNA-identified stomach contents, and between February 2014 blood components and November 2013 DNA-identified stomach contents, the dietary output from mixing models was typically within estimated margins for DNA-identified stomach contents for P. leopardus at Helix Reef. Admittedly, stable isotope mixing models incorporated stomach content data, but conservative margins of error (i.e., 20% confidence interval) were used to not guide the models too strongly and mixing models without prior information still identified the main prey groups.
Results of this analysis highlight the value of using multiple complementary approaches. The visual analysis provided a baseline F I G U R E 4 Summary of resource selection as indicated by Jacobs' Electivity Index (D) for Plectropomus at TSV reefs. The proportion of prey consumed was calculated using DNA stomach contents and the proportion of prey available was estimated from standardized abundance surveys at four TSV reefs (Helix Reef, Rib Reef, Chicken Reef, and Knife Reef) during 2014 (Bierwagen et al.-in press).
The value of D varies from 1 (maximum avoidance) to +1 (maximum preference). Index values of 0 indicate that prey species are consumed in proportion to their abundance. Confidence intervals that fell between −0.25 and +0.25 were deemed to be consumed in proportion to its abundance (i.e., neutral selection) understanding of diet, but lacked the detail and resolution provided by the molecular approach. These results, in combination with longterm information supplied by isotope analysis, provide a more comprehensive understanding of feeding patterns. The inclusion of prior knowledge (i.e., stomach content data) into Bayesian mixing models has the ability to improve precision in estimating diet composition at monthly temporal scales (Chiaradia et al., 2014;Franco-Trecu et al., 2013). It was particularly useful identifying prey composition in P. laevis (bluespot), which appeared to underestimate the contri-

| Ecological implications
This study demonstrated that planktivorous Pomacentridae and Caesionidae are important components of Plectropomus diet at short-and long-term temporal scales. This has been previously demonstrated for P. leopardus based on stomach content results (Kingsford, 1992;St John, 1999  ) using priors consisting of DNA stomach content analysis for each species F I G U R E 5 Prey contribution estimates (75% credibility intervals) for Plectropomus spp. at TSV reefs (Helix, Yankee, and Coil Reefs combined) based on Bayesian stable isotope mixing models (adjusted for plasma, RBC, and muscle discrimination factors, respectively ) using priors consisting of DNA stomach content analysis for each species. stomach issue will likely be amplified under predicted climate change scenarios if metabolic demands of large predators are not met due to prey availability (Johansen et al., 2015;Pörtner & Peck, 2010) and habitat degradation alters community composition (Jones, McCormick, Srinivasan, & Eagle, 2004;Wen, Bonin, Harrison, Williamson, & Jones, 2016). Although Plectropomus are likely capable of adapting to changing resource pools (Graham et al., 2007;Johansen et al., 2015), planktonic food sources are important. Indeed, the four main prey species (P. digramma, N. azysron, A. polyacanthus, and P. trichrourus) are predominantly planktivorous (Froese & Pauly, 2016). Changes in primary production and plankton-based trophodynamics (e.g., Doney, Fabry, Feely, & Kleypas, 2009) will likely have a strong effect on how mesopredators such as Plectropomus, select and partition prey (Audzijonyte, Kuparinen, Gorton, & Fulton, 2013;Hempson et al., 2017).
Despite major prey items being similar among species and color phases, their contribution, and that of lesser prey differed. For example, more planktonic prey (Clupeidae, Caesionidae) were detected in P. leopardus compared to P. maculatus, which consumed more benthic/midwater consumers (Gobiidae, Lethrinidae), at OI Reef. This difference may be a result of vertical segregation , but a larger sample size is needed to confirm these results. Bluespot P. laevis appeared to select predatory consumers such as Serranidae, Lutjanidae; however, low abundances of these families in surveys may have overinflated the few found in stomachs. Benthic herbivores (Acanthuridae, Blenniidae, Siganidae) were ~15%-20% more abundant in bluespot DNA stomach contents compared to other species and color phases. Differences in benthic carbon sources between P. leopardus and P. laevis (bluespot) were also found based on isotopic niche breadth, which showed limited overlap (0%-21%) for plasma, RBC, and muscle tissue (Matley et al., 2017 (Matley et al., 2017).
Consumers typically select prey that optimize energetic gains such as larger prey (offset by foraging costs; Pyke et al. 1977).
However, the size of consumed prey is often limited by consumer size due to limitations such as gape size (Mittelbach and Persson 1998

| CON CLUS ION
This study showed DNA stomach analysis was a more comprehensive tool to characterize diet compared to visual stomach analysis by increasing resolution of prey identification and detecting greater taxonomical diversity. The use of stable isotopes provided dietary estimates over longer periods of time and quantification of prey via Bayesian mixing models matched well with temporally congruent samples after incorporating stomach content information.
Plectropomus δ 15 N values in plasma and RBCs reflected the TL of prey; however, muscle δ 15 N values did not, highlighting the limitations of stomach contents to characterize diet over longer periods.
Thus, interpretation of muscle-derived mixing models should be treated cautiously because greater uncertainty in prey items exists. Lédée, J. Smart, M. Green, S. Bierwagen, and A. Schlaff.

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

JK Matley and AJ Tobin collected most samples; JK Matley and AT
Fisk processed and analyzed isotope samples; JK Matley, GE Maes, F Devloo-Delva, R Huerlimann, and G Chua processed and analyzed genetic samples; and all authors were involved in the concept and design of the project, data analysis, and contributed to writing/ revisions.