Fish as predators and prey: DNA‐based assessment of their role in food webs

Abstract Fish are both consumers and prey, and as such part of a dynamic trophic network. Measuring how they are trophically linked, both directly and indirectly, to other species is vital to comprehend the mechanisms driving alterations in fish communities in space and time. Moreover, this knowledge also helps to understand how fish communities respond to environmental change and delivers important information for implementing management of fish stocks. DNA‐based methods have significantly widened our ability to assess trophic interactions in both marine and freshwater systems and they possess a range of advantages over other approaches in diet analysis. In this review we provide an overview of different DNA‐based methods that have been used to assess trophic interactions of fish as consumers and prey. We consider the practicalities and limitations, and emphasize critical aspects when analysing molecular derived trophic data. We exemplify how molecular techniques have been employed to unravel food web interactions involving fish as consumers and prey. In addition to the exciting opportunities DNA‐based approaches offer, we identify current challenges and future prospects for assessing fish food webs where DNA‐based approaches will play an important role.


| INTRODUCTION
Fish are embedded in complex trophic networks, both as consumers of plants, invertebrates or other vertebrates and at the same time as prey and hosts for a wide range of species. Understanding these food web interactions is vital to comprehend the underpinning mechanisms driving alterations in fish communities in space and time (Baker et al., 2014). Compared to long-term changes in community composition, such as species replacements or varying abundances of taxa, trophic shifts can happen rapidly, typically before transitions in the species assemblages are recognizable (Tylianakis et al., 2008). This means that the direct and indirect interactions between species and the species' functional roles do change without major changes in terms of community assemblages or before these occur. Therefore, quantifying these trophic links is essential to understand the effect of environmental perturbations, such as climate warming, on fish populations (Baker et al., 2014). To predict their future changes is furthermore crucial for the sustainable management of fish stocks (Baker et al., 2014). The importance of trophic links for the sustainable management of populations of both fish and piscivores has long been acknowledged.
For example, large-scale examinations of the diet of commercially relevant fish such as cod and herring have been implemented in long-term monitoring programmes (e.g., Korneev et al., 2015) to provide dietary data for setting Barents Sea-wide catching rates. Regarding piscivorous birds and mammals, their long-standing perception as food competitors by humans (Duffy, 1995;Gosch et al., 2014) has led to an increase of shooting, culling, and repellent measures to reduce their populations in areas of conflict (Boudewijn and Dirksen, 1999;Bowen and Iverson, 2013). However, with many of them being endangered species, their management needs to be based on empirically generated dietary data to optimize and justify management interventions.
For example, Grandquist et al. (2018) found no evidence of salmonids in the diet of harbour seals (Phoca vitulina) using a combination of DNA-based and morphology-based gut content analysis, calling for a reassessment of seal culling programmes that were implemented to reduce phocid predation on salmonids.
Key to any food web-based approach is how trophic links are identified and quantified. The long-term dietary patterns of both fish and piscivores have been examined by stable isotope and fatty acid analyses (Nielsen et al., 2018). Although these approaches can depict the feeding history of individuals, they are usually quite limited in the taxonomic resolution of the diet. More detailed diet information of fish has traditionally been obtained by examining the gut content for morphologically identifiable prey remains such as bones, scales, otoliths, exoskeletons and other hard part prey remains (Nielsen et al., 2018). The same is true for dietary studies on piscivores where, depending on species, the gut content of killed specimens (Boström et al., 2012;Labansen et al., 2007), regurgitated pellets (Dias et al., 2012;Gonzalez-Solis et al., 1997;Thalinger et al., 2016) or faeces have been examined for identifiable hard parts (Jedrzejewska et al., 2001;Sinclair and Zeppelin, 2002). However, this type of diet analysis can be hampered by difficulties in identifying the prey remains. For example, small prey remains can be too digested to be identifiable while larger bits can be too distorted to achieve prey assignment on lower taxonomical levels as the food remains are too degraded for proper identification (Barrett et al., 2007;Bowen and Iverson, 2012). This is especially true for soft-tissued prey such as jellyfish, small prey items and food which is easily digested often remains undetected by visual means (Berry et al., 2015;Zingel et al., 2012), leaving us with incomplete diet information. Moreover, the level at which this prey can be identified largely depends on the skills of taxonomists. Consequently, the trophic data generated may vary considerably between studies due to the differences in the capability of the persons involved.
Finally, the possibility of automatizing morphological diet identification is limited by these conventional approaches, thus making diet analysis a time-consuming and cumbersome process that hampers the routine analysis of the large data sets necessary to assess the dynamics in real-world food webs and within routine monitoring programmes.
DNA-based diet analysis constitutes a solution to most of these problems: DNA of the food of fish as well as DNA of their parasites or the one of fish being consumed by piscivores can be detected and identified at high specificity and sensitivity in different samples, including gut content, faeces or regurgitates. This enables the investigation of feeding interactions across trophic levels even within complex food webs at unprecedented resolution. Noteworthy for this special issue, aquatic ecologists have been at the forefront of DNAbased diet analysis. Already in the 1990s the consumption of stone flounder Kareius bicoloratus by the sand shrimp Crangon affinis was examined by PCR-based gut content analysis of the decapod (Asahida et al., 1997), while the field of DNA-based diet analysis started to expand in the early 2000s (Agusti et al., 1999;Roslin and Majaneva, 2016;Zaidi et al., 1999).
Compared to nonmolecular approaches DNA-based assessment of trophic interactions includes a range of advantages: (a) higher specificity and sensitivity at which food DNA can be detected and identified; (b) the ability to standardize the methodology; (c) the ability to verify the food and parasites detected via DNA sequences; and (d) the possibility of employing high-throughput analyses. Due to the high sensitivity of the molecular methods, even a few molecules of food DNA can be detected , which allows a comprehensible assessment of what has been consumed compared to morphology-based identification (Deagle et al., 2009). Moreover, DNA typically facilitates identifying prey at much lower taxonomic levels than hard part prey analysis (Berry et al., 2015), again leading to a more complete food spectrum of the species of interest. The application of DNA-based methods for prey detection also allows for assessing trophic interactions in an unambiguous and verifiable way as the detected DNA is or can be sequenced and identified. Applying a well-characterized methodological framework to the analysis of such diet samples paves the way for standardized trophic data across samples, circumventing bias introduced by visual inspection of prey remains. Moreover, molecular prey identification also decreases the per-sample effort for dietary analysis, i.e., the time needed to find and identify prey remains . Finally, the use of highthroughput technology (e.g., Wallinger et al., 2017) for processing and analysing diet samples allows large sample numbers (i.e., several thousand samples, e.g., McInnes et al. 2019, Roubinet et al. 2018) to be examined cost-and time-effectively. This is a key feature to obtain reliable and quantifiable information on trophic networks including fish both as consumers/hosts and prey as well as to examine and predict the dynamics in these food webs across space and time.

| MOLECULAR ASSESSMENT OF TROPHIC INTERACTIONS: THE WORKFLOW
When investigating trophic interactions molecularly (Figure 1), one of the most important things to consider is the fact that the DNA of interest, i.e., the DNA of the consumed food, is digested and thus will be degraded over time. Consequently, the detection of multiple copy genes, i.e., genes that are present in hundreds to thousands of copies per cell such as mitochondrial, chloroplast or ribosomal DNA, promotes the probability of finding semi-digested food DNA due to the overall higher number of target molecules in the consumed tissue.
Furthermore, ongoing digestion leads to fragmented DNA strands such that the detection of long DNA fragments will become increasingly difficult the longer the prey is digested (Deagle et al., 2006). Thus, the use of shorter DNA fragments as targets is preferred to extend the time window post consumption during which a food item can be detected (Hoogendoorn and Heimpel, 2001) and increase the overall detection probability (Keskin, 2016). It has been therefore suggested that the use of DNA fragments longer than approx. 300-400 bp should be avoided in trophic studies to maximize the detection probability (King et al., 2008). For the application of metabarcoding, where millions of DNA sequences are read in parallel and then matched to reference databases, this limitation to short amplicons is additionally driven by methodological constraints as only a few sequencing platforms are capable of sequencing longer amplicons. However, if specific primers are used to detect certain prey taxa, the 300-400 bp should not be seen as an absolute limit, as the overall performance of the applied assay (i.e., its sensitivity, which is influenced by potential mispriming, primer dimer formation, etc.) will also strongly influence the ability to detect food. It is thus not uncommon that a well performing assay targeting fragments of 380-480 bp can outperform a less effective one targeting 100-200 bp fragments (Waldner et al., 2013). If the investigated organisms are parasites or symbionts, the focal DNA is not digested and thus the length of the analysed DNA fragment is anyway of reduced importance. Nevertheless, it is also advisable to target multiple copy genes to make sure that the symbionts and parasites are detected even if only a few cells are present (Ye et al., 2017). If early developmental stages are to be analysed, mitochondrial genes are preferred, as eggs are especially rich in mitochondria with up to 15,000 times more organelles than can be found in somatic cells (Tourmente et al., 1990).
A wide range of dietary sample types can be employed in DNAbased trophic analysis (Figure 1). For example, samples can be invasively collected by extracting the digestive part of a consumer for analysis (Braley et al., 2010;Jo et al., 2014;Marshall et al., 2010;Siegenthaler et al., 2019;Tverin et al., 2019) or flushing it to obtain its content, which often works well for piscivores such as penguins (Alonso et al., 2014). Alternatively, noninvasive sampling approaches, such as the collection of faeces and regurgitates, have successfully been applied for dietary analysis, having the advantage of minimizing the impact on the species under investigation. Noninvasive samples should be collected as fresh as possible to avoid additional breakdown of food DNA by ongoing digestion through omnipresent DNases (Palumbi, 1996), exposure to UV-light (Deagle et al., 2005;Oehm et al., 2011) or microbial activity (Barnes et al., 2014). Hence faeces have been analysed to track consumption of fish (Granquist et al., 2018;Hong et al., 2019;Jarman et al., 2013;Sommer et al., 2019) and their diet (Corse et al., 2010;Guillerault et al., 2017). Additionally, pellets produced by piscivorous birds that regurgitate indigestible prey remains proofed as suitable source of trophic information Thalinger et al., 2018). Likewise, lost food items or dropped prey remains can provide valuable information on a diet. Especially at times when young are reared, prey remains are found close to nests (Bowser et al., 2013;Oehm et al., 2017).
Besides employing optimized detection assays (Figure 1), the probability to successfully identify trophic interactions via DNA can be enhanced by the application of high-quality DNA isolation protocols. Such high-end extraction protocols which are specifically tailored towards the extraction of DNA from dietary samples effectively remove potential PCR inhibitors such as phenolic compounds , humic acids or other complex substances known to be often present in diet samples, such as faeces (Abu Al- Soud and Rådström, 2000;Sidstedt et al., 2019). Likewise, the use of high-quality PCR chemistry and the addition of PCR enhancers for overcoming inhibition are essential to maximize food DNA detection success. Contrastingly, the often suggested dilution of DNA extracts to avoid PCR inhibition is not advised when dealing with degraded target DNA, being present only in traces (King et al., 2009). In such cases, a dilution of the sample could reduce the probability of detecting the food or parasite DNA. While general primers match short areas of DNA which are (often not perfectly) conserved for all targeted species, specific primers are purposely designed to fit only to DNA of a certain species or taxon. There are two approaches to how the DNA of the food and/or parasites can be detected and identified: (a) diagnostic PCR and (b) diet metabarcoding ( Figure 1). Cannibalistic interactions go undetected as the DNA of the consumer cannot be differentiated from that of conspecific prey with the technologies currently applied.

| Diagnostic PCR
Assays employing specific primers are known as diagnostic PCR. They provide the possibility of discovering directly whether DNA of this taxon is present depending at the presence/absence of a specific amplicon. The merit of this type of assay is that no further analysis is necessary for answering the question of whether a specific prey has been consumed or if a specific parasite uses a fish species as a host.
Apart from that, diagnostic PCR can be employed almost independently of the platform of choice (e.g., standard PCR, quantitative PCR, droplet digital PCR) and the potential use of fluorescently labelled probes provides the opportunity to increase assay specificity further.
Multiplex PCR assays, although challenging to be developed and optimized, even allow the detection of several different food or parasite taxa simultaneously in one reaction, making diagnostic PCR a very fast and cheap way to analyse large numbers of samples (Gariepy et al., 2008;Harper et al., 2005).
Despite these advantages, so far only relatively few studies have applied diagnostic PCR to investigate fish as predator or prey (Brandl et al., 2015;Casper et al., 2007b;Keskin, 2016;Lamb et al., 2017). This might be due to the drawbacks of diagnostic PCR: apart from requirements regarding sufficient experience and skills for assay development, the detection is restricted to those taxa that are looked for and only a limited number of food taxa can be investigated in parallel, even when stepwise multiplex PCR approaches are applied .

| Diet metabarcoding
If general food-or parasite-specific barcoding primers are employed in PCR, the amplicons need to be further analysed to gain information on the consumed taxa. Usually this is done by reading the respective DNA fragment and comparing it with known DNA sequences in a reference database (DNA barcoding). Until recently, this was mainly achieved through cloning of PCR products and Sanger sequencing of individual clones (Braley et al., 2010;Hardy et al., 2010;Leray et al., 2013b). Since the costs for the parallel sequencing of millions of individual DNA molecules via high-throughput sequencing (HTS) approaches dropped significantly, targeted amplicon sequencing (i.e., diet metabarcoding) has replaced the aforementioned cloning approach. Nowadays, most trophic studies including fish as consumers or prey/host are conducted based on the metabarcoding approach (e.g., Bessey et al., 2019;McInnes et al., 2017b;Riccioni et al., 2018;Schwarz et al., 2018;Waraniak et al., 2019). Even the combination of several general barcoding primer pairs is possible, to cover a very wide food spectrum of generalist consumers (Deagle et al., 2009). The application of general barcoding primers has the advantage that no or very little a priori knowledge on potential food is needed to elucidate the diet of a species and also unexpected food sources might be detected (Leray et al. 2013). However, this comes at the cost that the primers often also amplify the consumer DNA. If consumer DNA is overrepresented compared to prey DNA in a sample, this can lead to a dramatic drop in the number of informative diet DNA sequences obtained by metabarcoding (Krehenwinkel et al., 2017). Hence, the application of blocking primers, which can prevent or reduce the amplification of a certain taxon (e.g., the consumer), is particularly useful for dietary studies (Su et al., 2018;Vestheim and Jarman, 2008). A metabarcoding study on the gut content of stickleback (Jakubaviciute et al., 2017;Leray et al., 2013a) demonstrated that blocking primers should be preferentially used over restriction digestion for predator DNA removal as they recover greater prey diversity. However, potential co-blocking of some food taxa or shifts in read numbers across taxa can be a problem (Piñol et al., 2015). If consumer and prey are closely related, as it is the case in piscivorous fish, their DNA sequences might be similar in the target region, increasing the risk for unwanted effects.

| Comparison of diagnostic PCR and metabarcoding
There are advantage and drawbacks in both the two basic principles for molecular diet analysis and parasite detection, and the decision of which method is better suited strongly depends on the research question asked (Table 1). Provided the specific primers employed in diagnostic PCR are well designed, the results of these assays are usually very robust and reproducible (Rennstam Rubbmark et al. 2019). This is because the ability to detect a certain prey/parasite taxon is barely influenced by the presence or absence of other DNA types such as the consumer, other prey or microbes (Sint et al., 2012). Only close to the detection limit, which can be below 20 molecules per PCR , can variability in prey/parasite DNA detection success be observed (Sint et al., 2012). This effect depends mainly on whether the primers managed to attach to the few template molecules during the first few PCR cycles (Ruano et al., 1991).
General barcoding primers, such as those used for metabarcoding, however, will likely have an imperfect match to some of the taxa targeted. While still being able to bind to and amplify these DNA strands, the efficiency can be strongly reduced, dependent on the number and location of the mismatches at the priming sites, leading to biased amplification of the different species being present in a biological sample (Piñol et al., 2015). Even if a perfect match is realized for all potential targets, the base composition or number of template molecules of an individual target can influence the amplification simply by stochastic mechanisms. This means that abundant types of DNA are more likely to be amplified than rare ones, or certain DNA sequence variants are preferentially amplified and thus sequenced over others (Kelly et al., 2019;Ruano et al., 1991). This stochasticity plays an important role in the limited repeatability of metabarcoding results (Rennstam Rubbmark et al., 2019) and shows that metabarcoding approaches, while highly suited to describe the general prey spectrum, are often not the best choice to detect rare targets. Moreover, closely related species often differ only by a few bases in the target genome region, which can make it challenging to distinguish the actual presences of these two species from a PCR or sequencing error. Furthermore, the ability to identify the diet sequences derived by metabarcoding depends on the sequence reference databases used.
Depending on the organisms of interest and the target genes, several public databases [e.g., GenBank (www.ncbi.nlm.nih.gov/genbank/), BOLD (www.boldsystems.org), SILVA (www.arb-silva.de) or MitoFish (http://mitofish.aori.u-tokyo.ac.jp)] can be used to identify the obtained sequences. However, taxonomic coverage in such databases and consequently the resolution of dietary composition can vary greatly across genes, taxa and the geographic region of interest (Weigand et al., 2019). Additionally, databases potentially contain misidentified DNA sequences, or sequences identified via "reversed taxonomy" (i.e., via DNA barcoding and not morphology). Still, the majority of taxonomic identifications seem to be correct (Leray et al., 2019), such that adverse effects for dietary studies may be unlikely. This calls for a thorough preassessment to evaluate which taxonomic resolution can be obtained by using public reference databases. Therefore, it is not surprising that many publications rely on custom curated databases containing only verified sequences of plausible species in the investigated geographic range to improve the fit and taxonomic accuracy of the results (e.g., Hanfling et al., 2016).
There are many different tools and strategies for processing raw metabarcoding reads until they are matched against a reference database. Some applications are specific to certain processing steps (e.g., "Cutadapt") and many researchers develop their own pipelines using various tools. Other applications, such as "obitools", "QIIME2", "SLIM" and "JAMP", provide a framework for all these individual steps to facilitate sequence processing for users. Addressing the many aspects necessary for successful diet metabarcoding and subsequent sequence analysis comprehensively goes beyond the scope of this article, but has been dealt with elsewhere (e.g., Alberdi et al., 2018;Deagle et al., 2019;Pompanon et al., 2012).

| The importance of primers
The design and selection of PCR primers needs considerable care and the requirements for such primers differ between the two basic approaches of molecular diet analysis ( Figure 2). For diagnostic PCR, a species or taxon/group specific primer pair is required, meaning that the binding sites for the primers need to be conserved within the targeted taxon. At the same time, it must be different to all nontarget taxa to prevent the amplification of their DNA, which would otherwise result in false-positive detections. For metabarcoding approaches on the other hand, not only the primer binding sites need to be considered, but also the resulting amplicon length needs to be suitable for metabarcoding. Depending on the HTS platform used, this can vary between less than 100 and several hundred base pairs (Ergüner et al., 2015). Finally, the selected DNA region needs to provide stretches that are conserved among taxa (to place the primers a suitable distance apart), but the sequence information between the primer binding sites needs to be variable enough to allow taxon identification based on these sequences (Clarke et al., 2014).

| QUANTIFICATION OF FOOD DNA
Due to the co-occurrence of consumer DNA, measuring total DNA concentration in a dietary sample is not informative regarding the amount of food or parasite DNA. Even so, several methods exist to quantify this specific target DNA ( Table 2). As the concentration of food DNA is typically not high in dietary samples, standard PCRs are usually terminated before reaching product saturation (Ruano et al., 1991). Thus, the amount of amplicon being generated correlates well with the amount of template DNA (Thalinger et al., 2019). The application of prey-specific diagnostic CE-PCR (i.e., capillary electrophoresis PCR, which stands for "diagnostic PCR" combined with frag-    (Bowles et al., 2011;Deagle and Tollit, 2007). Contrastingly, most of the recent fish dietary studies rely on metabarcoding approaches and RRA. Considerable effort has been put into identifying and accounting for amplification, sequencing, and bioinformatic biases which metabarcoding samples are exposed to (Lamb et al., 2017;Zinger et al., 2019).
Although there exist several options for quantifying food DNA, translating the amount of diet/parasite DNA to the amount of prey consumed is not trivial and has to be evaluated separately. In this con-  introduce other biases. Which type of quantitative measure is best to be used will depend on the predator-prey system studied and the characteristics of the molecular methodology employed. needs to be noted that the estimation of the defecation time was challenging (sampling impossible during the night). Two studies (Bowles et al., 2011;Deagle et al., 2005) found a good correlation between the amount of fish consumed and the number of prey DNA molecules amplified, with small meals being reliably detected. In a study on cormorants, earliest prey detection in faeces was reported 1.5-2.5 h post feeding and the maximum prey detection time in faeces was 75 h post feeding . In this case, however, small meals were not reliably detected and showed significantly lower detection probabilities across time than large meals.

| Food chain errors due to secondary predation
The food DNA detected in a dietary sample can either come directly from the food eaten by the consumer, the so-called primary prey, or stem from secondary prey, i.e., the food DNA which is contained in the digestive system of the primary prey . Secondary predation (also called hyperpredation) has received far too little attention in dietary studies of fish, albeit the topic is highly relevant and often addressed in review articles assessing the potential of (molecular) methods to study predator-prey relations and associated pitfalls (Deagle et al., 2019;Sousa et al., 2016). In practice, there are different ways of accounting for secondary predation during the interpretation of dietary field data. Often, taxa which are known to not be preyed upon are removed from the dataset prior to analysis (e.g., fungi, macroalgae and Chromista are not fed upon by sticklebacks) and taxa which are suspected to be derived from secondary predation (e.g., copepods in the diet of Adélie penguins) are addressed in the discussion (Jakubaviciute et al., 2017;Jarman et al., 2013). Additionally, the detection of diet taxa suspected to be secondarily predated is checked for consistent co-

| Scavenging of food
Another aspect involving the possibility for confusion in both molecular and nonmolecular diet analysis is scavenging. Per definition, scavenging is the consumption of already dead food, contrary to predation, which includes its killing. However, the DNA analysis of food present in the gut of the consumer does not allow differentiation between active predation and scavenging (Foltan et al. 2005). Similar to live food, the DNA of both carrion and decayed plant food can be detected at extended times post feeding (Juen and Traugott, 2005;Wallinger et al., 2013). As a result, molecular diet analyses may lead to an overestimation of the impact of predators on prey populations if primarily carrion is scavenged. Whether the differentiation of these two feeding strategies is important for a study depends on the questions asked. For the energy flow in a food web and the supplement of consumers it may be a minor factor whether the consumed prey had been dead or not. Wilson & Wolkovich (2011) recently highlighted that scavenging links have been underestimated in food webs ranging from marine to terrestrial ecosystems, and that substantially more energy is transferred per scavenging link than per predation link.
When it comes to the investigation of regulatory management, it is highly relevant whether a predator was actively hunting and killing its prey or taking advantage of an already dead animal. In terrestrial ecosystems, especially when dealing with agricultural systems, pest con- suggesting that this deep sea lobster has a broad diet spectrum, with a high reliance on scavenging a diverse range of pelagic and benthic species from the seafloor. Although scavenging plays a similar functional role in terrestrial and aquatic food webs, Beasley et al. (2012) suggested that several fundamental differences do exist. In particular, the movement of carcasses in marine ecosystems (e.g., wave action, upwelling and sinking) diffuses biological activity associated with scavenging and decomposition across large, three-dimensional spatial scales, creating a unique spatial disconnect between the processes of production, scavenging and decomposition, which in contrast are tightly linked in terrestrial ecosystems. The authors show that jellyfish predation may be more common than previously acknowledged, with jellyfish DNA detected in 27.6% and 11.6% of herring and whiting stomachs, respectively, representing the two most abundant fish species with the highest detection frequencies of jellyfish prey in this study. This work refutes the notion that jellyfish predation is rare and demonstrates the value of such comparably simple molecular detection systems to improve the understanding the trophic role jellyfish play in ecosystems and in predicting jellyfish blooms.
Diagnostic PCR assays can also be used to assess the consumption of multiple food types within one assay when primers for multiple prey targets are multiplexed (Harper et al., 2005;Sint et al., 2012).
Employing seven species-specific primer pairs in three multiplex PCR assays, Bade et al. (2014) tested the digestive tract samples of the Cownose ray (Rhinoptera bonasu) for DNA of commercially important bivalve species. The authors found that the rays ate stout tagelus and soft-shell clams but there was no evidence of consumption of commercially important oysters, hard clams and bay scallops. Another example is the extensive multiplex PCR system, allowing for the stepwise testing of diet samples for DNA of 31 Central European freshwater fish species, which was presented by Thalinger et al. (2016). This system comprises six PCR assays that were used to detect preyed fish in European otter spraints and kingfisher droppings as well as faeces and pellets of cormorants Thalinger et al., 2018).
Classical DNA barcoding, i.e., the amplification and sequencing of DNA fragments to identify species (Hebert et al., 2003), can also be used to identify prey remains which are retrieved from diet samples.  (Guillerault et al., 2017), with the added advantage that this examination was noninvasive.
Apart from the identification of prey in fish, diet metabarcoding has also been widely used to examine the diet of piscivorous birds (e.   Thomsen and Willerslev, 2015). In comparison to eDNA, aDNA and fDNA, where usually only few samples are analysed, trophic studies often entail the assessment of much larger sample numbers ideally ranging between hundreds and thousands of samples.
This means that besides the provision of a high-quality diagnostic laboratory, high-throughput technology and laboratory routines need to be implemented for processing large sample numbers in a cost-and time-effective manner. Molecular diagnostic work also demands standardized sample processing to avoid introducing variability due to differently treated samples, for example when generating DNA extracts or conducting PCRs. Indeed, these are two of the major structural hurdles the field of DNA-based analysis of trophic interactions has to cope with, a topic rarely acknowledged in the literature at all (but see Kitson et al. 2019. In fact, the power of trophic data which is based on the assessment of dietary samples, using both molecular and nonmolecular approaches, is strongly correlated to the number of samples analysed, i.e., sufficiently large sample numbers are needed to statistically assess the effect of environmental changes on trophic links and/or to compare consumers with regard to their trophic ecology. Regarding fish as predator and prey, so far only a small number of DNA-based studies have built their analyses on sufficiently large data sets, which hampers their actual impact for obtaining a comprehensive understanding of fish-based food webs. The rapidly evolving field of molecular trophic analysis has opened up a wide field of varieties but also led to increasing methodological complexity (e.g., Alberdi et al., 2018). For example, the establishment of new diagnostic PCR assays from scratch entails the often cumbersome development of appropriate primers meeting all requirements regarding specificity and sensitivity. Diet-related metabarcoding work, however, needs to deal with the selection of suitable primers, but also comprehensive and sometimes sophisticated bioinformatic analyses for getting meaningful results. This situation puts high demands on the research teams involved and the decision of the method of choice for a specific research question alone can be challenging. One potential option to optimize both time and costs in projects where trophic interactions are to be analysed using DNAbased approaches is to (partly) outsource this work to the increasing number of specialist commercial providers, as is nowadays routine for Sanger sequencing.
Apart from technical issues, the lack of detailed information on the methodology provided in molecular dietary studies is another challenge for both wet and dry lab procedures. Unfortunately, such reporting standards which are comparable to the minimum information required for qPCR assays in medical applications (i.e., MIQE guidelines; Bustin et al. 2009). For example, the sensitivity of assays for detecting food DNA is rarely reported and standardized across targets when multiplexing is employed or different diagnostic PCR assays are compared within a study (but see Sint et al. 2012). Likewise, assay specificity is often poorly evaluated and documented, for example by sequencing obtained amplicons in diagnostic approaches or detailed reporting of target and nontarget sequence reads in metabarcoding studies. The lack of knowledge regarding such general criteria hampers the comparability of data across studies drastically, providing a significant drawback to utilizing the rapidly growing trophic information generated by DNA techniques.
When it comes to the interpretation of the molecularly generated trophic data several important issues need to be considered. Cannibalistic interactions cannot be depicted as the current approaches allow prey DNA to be differentiated between species but not within species, albeit the first successful efforts distinguishing between fish haplotypes in environmental samples were recently published (Tsuji et al. 2020). Generally, dietary studies still rely on morphological identification and counts of prey remains in parallel to the molecular trophic analyses (e.g., Eigaard et al., 2014) to assess intraspecific feeding interactions. Furthermore, the molecular approach cannot differentiate the life/developmental stage of the prey and again a combination of molecular and hard part prey remain analysis is advisable in this context, as exemplified by Thalinger et al. (2018).
Besides the undisputable advantages and exciting opportunities DNA-based approaches offer for studying food webs around fish, it should not be forgotten that the combination with other methodologies, such as identification of prey hard parts, stable isotope or fatty acid analysis, can deliver even stronger data sets and ecological insights. analysis (Horswill et al., 2018). The combined approach offered a less invasive sampling methodology and provided more in-depth information regarding prey species diversity, making it a promising approach for generating less-invasive dietary long-term data sets.