No guts, no glory: Gut content metabarcoding unveils the diet of a flower‐associated coastal sage scrub predator

Invertebrate generalist predators are ubiquitous and play a major role in food‐web dynamics. Molecular gut content analysis (MGCA) has become a popular means to assess prey ranges and specificity of cryptic arthropods in the absence of direct observation. While this approach has been widely used to study predation on economically important taxa (i.e., pests) in agroecosystems, it is less frequently used to study the broader trophic interactions involving generalist predators in natural communities such as the diverse and threatened coastal sage scrub communities of Southern California. Here, we employ DNA metabarcoding‐based MGCA and survey the taxonomically and ecologically diverse prey range of Phymata pacifica Evans, a generalist flower‐associated ambush bug (Hemiptera: Reduviidae). We detected predation on a wide array of taxa including beneficial pollinators, potential pests, and other predatory arthropods. The success of this study demonstrates the utility of MGCA in natural ecosystems and can serve as a model for future diet investigations into other cryptic and underrepresented communities.


INTRODUCTION
Predatory arthropods can have profound impacts on pollinator-plant communities as their presence may alter the behavior and abundance of other flower-visiting insects and indirectly affect plant fitness (Dukas 2005, Jones 2010, Wirsing et al. 2010, Huey and Nieh 2017. Generalist ambush predators can engage in a variety of trophic interactions, ranging from direct predation on both pests and beneficial organisms to complex trophic cascades through intraguild predation (Polis and McCormick 1987, Rosenheim et al. 1993, Finke and Denno 2004, Gagnon et al. 2011. A refined understanding of trophic interactions not only has implications for pest management, but also has implications for the conservation of biodiversity and endangered species (Polis and Holt 1992, Bampfylde and Lewis 2007, Hurd 2008, Gagnon et al. 2011, Chisholm et al. 2014. Fundamentally, it enables biologists to unravel the complex dynamics and functions of ecosystems (Agrawal 2000).
While much research has been devoted to natural enemies that specialize on pests (Sheehan 1986, Landis et al. 2000, Snyder and Ives 2003, Choate and Lundgren 2015, Morgan et al. 2017, trophic interactions involving generalist arthropod predators in natural systems remain vastly understudied. One such system is the coastal sage scrub (CSS) or soft chaparral of Southern California. With a great diversity of endemic flora and fauna, this habitat covers lower elevation portions of a complex Mediterranean-type scrub ecoregion that is part of one of Earth's biodiversity hotspots (Cowling et al. 1996, Myers et al. 2000. Once widespread, CSS communities have become increasingly fragmented and altered through anthropogenic disturbance and the introduction of non-native species over the past two centuries (Westman 1981, Minnich and Dezzani 1998, Lambrinos 2000. As a result, many endemics of CSS have become rare and some even endangered such as the California gnatcatcher, Polioptila californica californica Brewster (McCormack and Maley 2015), and the Quino checkerspot butterfly, Euphydryas editha editha (Boisduval; Parmesan et al. 2015).
Coastal sage scrub and surrounding communities support a particularly high number of unique arthropods. Approximately 500 native bee species have been documented here (Michener 1979), many of which provide important services for natural, urban, and agricultural landscapes (Kremen et al. 2002b, Hernandez et al. 2009). Despite recent concerns surrounding the general decline of pollinators (Committee on the Status of Pollinators in North America 2007, Potts et al. 2010), little is known about the current status of native pollinator populations in CSS. The fitness of many flowering plants relies largely on interactions with pollinating insects; mutually, many insects require pollen and/or nectar from flowers for sustenance and development (Tepedino 1979, Kearns andInouye 1997). Flowering plants which have evolved mutualistic relationships with specific pollinating insects are particularly vulnerable to environmental changes (Gilman et al. 2012), and a reduction in the services provided by a unique pollinator can negatively impact plant populations (Wilcock andNeiland 2002, Romero and. Predatory arthropods are also diverse and ubiquitous in CSS and influence ecosystem dynamics (Burger et al. 2001(Burger et al. , 2003. Predation on pollinators may have strong indirect effects given that pollinators provide essential services for natural communities and help maintain healthy ecosystems (Klein et al. 2007). The occurrence of predators on flowers may reduce pollinator visitation, causing some pollinators to spend less time at or avoid certain flowers, or potentially diminish the numbers of pollinators that share mutualisms with rare native plants (Elliott and Elliott 1991, Reader et al. 2006, Tan et al. 2013. California buckwheat, Eriogonum fasciculatum Bentham (Polygonaceae), is a dominant and widespread perennial of CSS that serves as an important resource for many arthropods including over 30 bee species (Kremen et al. 2002a, Montalvo andBeyers 2010). This plant is commonly, although patchily, frequented by Phymata pacifica Evans (Hemiptera: Reduviidae), an ambush bug native to CSS and a presumed generalist predator of other flower-associated arthropods. Like crab spiders (Thomisidae; Llandres and Rodr ıguez-Giron es 2011, Llandres et al. 2013, Huey andNieh 2017) and weaver ants (Gonz alvez et al. 2013), ambush bugs can alter the foraging behavior of other flower visitors. For example, pollinators will spend significantly less time foraging on flowers harboring these ambush predators than on vacant flowers Elliott 1991, 1994). Given their flower-dwelling niche, a diverse range of prey taxa are available to ambush bugs in CSS. Despite this, the composition of their diet remains unclear.
Over the past two decades, novel methods to delineate food-web linkages have been devised that eliminate the need for direct observation or the visual inspection of gut or fecal matter (King et al. 2008, Pompanon et al. 2012, Gonz alez-Chang et al. 2016, Birkhofer et al. 2017). Among the most commonly applied and successful approaches for gathering qualitative prey data is DNA-based MGCA; Seri c Jelaska et al. 2014, Rondoni et al. 2015, Roubinet et al. 2015, Schmidt et al. 2016, Curtsdotter et al. 2018, Eitzinger et al. 2018. This method provides a reliable means to examine the diets of small, cryptic arthropods that pre-orally digest their food such as spiders and true bugs (Heteroptera). To capture the wide taxonomic range of a generalist predator's diet, DNA metabarcoding can be used to accumulate large amounts of prey data (Ji et al. 2013, Brandon-Mong et al. 2015. In metabarcoding, large numbers of amplicon sequences are derived via high-throughput sequencing and compared to existing barcode databases for identification (Blaxter 2016). Of studies on terrestrial arthropods that have utilized MGCA, many have focused on one or several specific prey taxa (often pests) and relied on prey-specific primers to determine which natural enemies are consuming those taxa in agroecosystems (Harwood et al. 2007, Fournier et al. 2008, Juen et al. 2011, Szendrei et al. 2014. A disproportionately small number of studies have attempted to assess the diet range of generalist predators in natural communities ( Seri c Jelaska et al. 2014).
For this study, our attention focused on four main objectives: I. determine whether native bees constitute the main group of prey of ambush bugs in a CSS community; II. document the diet breadth of P. pacifica with respect to (1) the taxonomic diversity of prey (i.e., the number of different families, genera, and species consumed) and (2) the trophic category of prey (pollinators, herbivores, or entomophagous) found on the dominant host plant, E. fasciculatum, and search for any indication that certain arthropod groups also associated with California buckwheat are absent from their diet; III. estimate the detectability half-life of DNA recovered from the guts of P. pacifica; and IV. employ MGCA using DNA metabarcoding and test the effectiveness of a predator-specific blocking primer.

Field sampling and specimen vouchering
We collected Phymata pacifica specimens from two field sites along Lytle Creek in San Bernardino National Forest over three visits during late June and early July of 2016. Each of the field sites was visited at least once in the morning (08:00-12:00 hours) and again at least once in the afternoon (12:00-16:00 hours). We exclusively sampled from Eriogonum fasciculatum. Anticipating taxonomic gaps among the barcode sequences available online for local fauna, we collected other buckwheat-associated arthropods for sequencing by means of beating, sweeping, and aerial netting on and around blooming flowers. Upon capture, P. pacifica specimens were immediately placed into separate vials containing 95% ethanol and cooled with dry ice. In the laboratory, all P. pacifica specimens were stored in a À80°C freezer to retard DNA degradation until they could be dissected. All dissected ambush bugs were given unique specimens identifier numbers and databased using the Plant Bug Planetary Biodiversity Inventory instance of the Arthropod Easy Capture Specimen Database (http://www.research.amnh.org/pbi/locality/index. php) and deposited in the Entomology Research Museum at the University of California, Riverside (UCR). Specimen information can be accessed through the Heteroptera Species Pages (http://research.amnh.org/pbi/heteropteraspecies page). Non-reduviid buckwheat-associated arthropods were also mounted and identified to the lowest taxonomic level possible using reference specimens from UCR's Entomology Research Museum, online searches (BugGuide.net), taxonomic keys (Goulet and Huber 1993, Triplehorn and Johnson 2005, Lawrence et al. 2010, and advice from specialists of various groups. These specimens were also deposited in the Entomology Research Museum at UCR.

Gut and DNA extraction
Sterilized forceps were used to remove midand hindguts from 225 specimens, and each was placed into individual crosslinked 1.5-mL Eppendorf tubes and homogenized using sterile pestles. DNA extraction was conducted using a QIAGEN DNeasy Blood and Tissue Kit. DNA was also extracted from the legs of 60 non-reduviid buckwheat-associated arthropods to construct a de novo COI reference library (hereafter denoted as our "local database" or "LocalDB") for taxa for which COI barcoding sequences were unavailable from BOLD or GenBank (as of January 2019).

Primer design, PCR, and sequencing
Major challenges include finding a set of universal primers that can accommodate the entire, often unknown, prey range, and limiting the amplification of predator DNA so that signal from the prey is not overwhelmed. To address the first issue, we used a universal primer pair that amplifies a 313bp sequence of the COI barcoding region: mlCOIintF (Leray et al. 2013) and HCO-2198(Folmer et al. 1994. When used in tandem, these primers can amplify a wide range of metazoan taxa (Leray et al. 2013). To overcome the second challenge, predator-specific blocking primers were developed and added to the PCR cocktail to limit non-target DNA amplification. These oligonucleotides contain a C3 spacer at their 3 0 end that inhibits polymerization during the elongation phase of PCR (Vestheim and Jarman 2008). The program PrimerMiner (Elbrecht et al. 2017) facilitated selection of these oligonucleotides over other possible primer pairs and design of a blocking primer for this study. In the process of developing a blocking primer, we downloaded all available Lepidoptera, Hymenoptera, and Diptera (known prey groups of P. pacifica) COI barcode sequences from NCBI and BOLD (available as of June 2016) and compared these sequences with that of P. pacifica to find a region suitable for a blocking primer. We designed a blocking primer (mlCOIintF-BLK-Phymata: 5 0 -TCCACCACTATC AAGAAATCTTGC/3SpC3/-3 0 ) that contains a C3 spacer at its 3 0 end to inhibit elongation of P. pacifica DNA. This oligonucleotide competes for binding sites with the mlCOIintF primer in its 5 0 region and spans into a Phymata-specific region along its 3 0 end. Since test PCR trials and Sanger sequencing demonstrated that the P. pacifica-specific blocking primer does limit the amplification of non-target host DNA (Appendix S1: Fig. S1), this oligonucleotide was used in conjunction with fusion primers that contain Illumina adaptor sequences at their 5 0 ends and the universal primers mentioned above at their 3 0 ends (see Appendix S1: Table S1 for primer information) during the initial round of PCR for metabarcoding.
To generate amplicons during the first round of PCR, we used a touchdown protocol with the following conditions: initial denaturation for 5 min at 95°C, followed by denaturation for 30 s at 95°C, and then annealing starting at 62°C for 30 s and decreasing by 1°C over 16 subsequent cycles until reaching a minimum annealing temperature of 46°C, with intervening extension phases run for 60 s at 68°C. Once an annealing temperature of 46°C was reached, we then continued with a 95°C-48°C-68°C regime for 24 cycles and ended with a final 7 min of extension phase at 68°.
The resulting products were run on a 2% agarose gel and then cleaned using solid-phase reversible immobilization with carboxylated Sera-Mag SpeedBeads (GE Healthcare UK Limited, Little Chalfont, Buckinghamshire, UK) in NaCl-and PEG-containing buffer (Rohland and Reich 2012) and indexed with dual index primers from NEB-Next Multiplex Oligo kits (New England Biolabs, Ipswich, Massachusetts, USA). Normalization was carried out with Charm Biotech Just-a-Plate 96well clean-up kits. A PureLink PCR Purification Kit was used to concentrate the final library after pooling and remove DNA fragments of less than 300 base pair in length. To confirm fragment size, the pooled samples were analyzed on a Bioanalyzer. The library was then sequenced on a single run of Illumina MiSeq v3 2 9 300 bp at the UCR Institute for Integrative Genome Biology.

Bioinformatics and prey identification
MiSeq reads were demultiplexed on UCR's Linux Cluster at the High-Performance Computing Center. Adaptor primers, barcodes, and low-quality ends were cut from reads using Trimmomatic v0.36 (Bolger et al. 2014). Pairedend output reads were then filtered, trimmed, dereplicated, and merged using the DADA2 package v1.6.0 (Callahan et al. 2016) in RStudio. Briefly, the DADA2 pipeline is designed to filter/ denoise amplicon data and infer sequence variants by modeling and correcting errors present after Illumina sequencing. Following merging, chimeras were removed using the removeBimeraDenovo() function. We included a negative (blank) sample in our sequencing run and, after processing with DADA2, recovered no sequence variants from it. Because of this and the fact that we recovered many rare and unique amplicon sequence variants, we opted not to set a minimum abundance threshold.
All resulting output amplicon sequence variants were queried against the Barcode of Life Data System (BOLD) COI database (Ratnasingham and Hebert 2007). Sequences for which no close matches were found on BOLD (<95% identity) were then searched against both NCBI GenBank and the local database of buckwheat-associated arthropods with BLAST. To assign identifications to sequence variants, we used identity thresholds. Only sequences sharing 100% identity with BOLD/NCBI/LocalDB matches were classified as identified to species. Matches below this threshold were only identified to genus, family, or order level depending on confidence values estimated using a taxonomic classifier (see below). Generally, NCBI sequences that matched our queried sequences in combination with the smallest E values, greatest nucleotide percent identity, and longest query cover were used to make taxonomic designations.
To measure the confidence of these identifications, we then used the insect (informatic sequence classification trees) R package (Wilkinson et al. 2018), a tool designed to assign rankbased taxonomic identifications to amplicon sequence variants generated by DADA2. For classifying the sequence variants of this study, we used the trained COI classifier (i.e., classification tree) specific to mlCOIintF/jgHCO2198 (Leray et al. 2013) barcoding amplicons (classifier.rds v5 20181124) provided through the insect package. The classify function was set to a threshold value of 0.6 as many sequences (including those of P. pacifica) returned uninformative taxon identifications when run with the default threshold parameter of 0.9. In addition to outputting a taxon name and rank, the classify function also reports an Akaike weight value (i.e., confidence score ranging from 0 to 1) for each of the final taxon assignments and this was used to judge the accuracy of initial BOLD/NCBI/LocalDBbased identifications.
Following identification, prey taxa were assigned to one of five general trophic categories (or a combination of) based on their biology and affiliation with California buckwheat: pollinators, herbivores, parasitoids, predators, or other (e.g., scavengers or fungivores). Pollinators were categorized by taxa that typically only visit buckwheat to acquire nectar or pollen. Herbivores were classified as phytophagous arthropods that feed primarily on plant material other than nectar or pollen (but may also feed on nectar or pollen as adults). Any taxa that exhibit entomophagy during some stage of their life cycle were subsequently categorized as either predators or parasitoids. See Table 1 for a breakdown of trophic assignment per prey taxon.

DNA detectability half-life feeding trials
Adult P. pacifica were collected alive from our CSS field site along the North Fork of Lytle Creek in late June of 2017. Predators were housed in petri dishes with a photoperiod of 15:9 L:D and held at a constant temperature of 27°C and starved for seven days prior to beginning the feeding trial. Each P. pacifica was fed a single house fly (Musca domestica Linnaeus) and allowed to feed for one hour. Ambush bugs that failed to feed were dropped from the experiment. At t = 0 h, five unfed individuals were placed immediately into a À80°C freezer to serve as a negative control. Due to substantial mortality during the starvation period, only four P. pacifica were available for each time interval post-feeding: 0, 6, 12, 24, 48, 72, and 96 h. Only three fed P. pacifica were available for the remaining 120-hr time point. After death by freezing, specimens were placed into 100% EtOH and stored at À80°C until their mid-and hindguts could be extracted (as described previously). Phymata pacifica-specific primers were used to confirm the success of DNA extractions, and M. domesticaspecific primers (MuscaF1: 5 0 -TGAATTAGGA CACCCTGGTGCTCTA-3 0 and MuscaR1_268: 5 0 -AG TTCAACCTGTTCCAGCTCCCTT-3 0 ) were designed by comparing Musca COI sequences downloaded from GenBank to those of ambush bugs to test for the presence of prey DNA. The presence or absence of~268-bp bands was verified with electrophoresis on a 1% agarose gel. The DNA detectability half-life and predicted 95% confidence intervals were determined using a linear regression model in RStudio (function lm()).

Predation on flower visitors
Unlike many predatory arthropods that are limited by their size when hunting, ambush bugs can take prey of an extensive size range. This is reflected in our prey data and can be attributed to the fact that Phymata employ fast-acting paralytic venom while holding their prey in place with powerful raptorial forelegs (Walker et al. 2016). Identified prey taxa range in length from roughly 2 mm (e.g., Orius tricolor Fabricius (Anthocoridae)) to over 10 mm (e.g., Apis mellifera Linnaeus (Apidae)), roughly twice the length of P. pacifica. Like other ambush bugs, it is evident that P. pacifica is an opportunistic generalist predator and consumes a wide range of prey, as those analyzed fed on members of at roughly 46 families of arthropods spanning 10 orders (Fig. 1, Table 1).
Contrary to our expectations, of the resulting 280 total prey amplicon sequence variants obtained from the 225 gut samples sequenced, only a small proportion (41/280:~15%) were identified as native bees. Regardless of this relatively low number, the ambush bugs examined fed on a broad diversity of Hymenoptera. Of the eight genera of Apoidea consumed, Lasioglossum Curtis (Halictidae; 31/280:~11%) and non-native A. mellifera (14/280:~5%) were recovered most frequently. Among the~80 unique buckwheatassociated arthropod morphospecies collected from CSS, we obtained nine genera of native apoids (Appendix S1: Table S2). Of these, four were also sequenced from P. pacifica gut contents. Additionally, several amplicon sequence variants were identified to apoid genera not collected ❖ www.esajournals.org  Perhaps most surprising is the great diversity of entomophagous arthropods found as prey. Approximately 35% (99/280) of detected prey were classified as either parasitoids or predators. Multiple genera of tachinid flies and braconid wasps, several of which are common lepidopteran parasitoids such as Agathis Latreille and Cotesia Cameron (Whitfield 1995, Sharkey et al. 2006, fell victim to P. pacifica. Among entomophagous prey identified to at least genus, Chetogena parvipalpis (Wulp), a tachinid fly known to parasitize Hesperiidae, Pyralidae, and Gryllacrididae (Arnaud 1978), was recovered most often (9/280:~3%). Several taxa of predatory Notes: Phymata pacifica specimen identification numbers given. If multiple taxa were detected from a single ambush bug gut, the specimen number is listed with number of detected prey taxa in parentheses. Percent identity for matches found using searches against BOLD, GenBank, or our local buckwheat-associated arthropod barcoding dataset is listed, and the database used for taxonomic identification is given in the Det. by column (denoted as BOLD, NCBI, or LocalDB). Names in bold represent identifications supported by both database searches and the insect classifier. Prey unidentified at a particular taxonomic level or that were not assigned by the "Insect" classifier are denoted with an en dash.
heteropterans, including groups that are used in some systems as biological control agents such as Nabis Latreille (Cabello et al. 2009) and Orius Distant (Van De Veire and Degheele 1992), were also consumed by P. pacifica. Other examples of intraguild predation involved predation on Mecaphesa Simon, a genus of crab spiders that share a niche and are potentially direct competitors with Phymata; and predation on entomophagous Coleoptera such as Cleridae and Melyridae that visit blooming E. fasciculatum (Arnett et al. 2002). To our knowledge, intraguild predation between Fig. 1. Diversity of prey taxa identified from the guts of Phymata pacifica. Line thickness corresponds to the number of instances that a given taxon was detected from the 225 gut samples. Color shading represents the general trophic categories recognized in this study. Only taxa which matched with 90% or greater identity to the recovered amplicon sequence variants are displayed. Amplicon sequence variants which could not be identified below order level are not included.
Phymata and Thomisidae has never been formally documented until now.
Phytophagous insects also comprised a great proportion of the prey identified (91/280:~33%). The diversity of lepidopteran prey at the generic level is unrivaled by other arthropod groups identified from gut contents. Sequence amplicon variants were matched to 10 families and~22 genera of moths and butterflies. This diversity of lepidopteran prey is not surprising since E. fasciculatum serves as an important host resource for both immature and adult Gelechiidae (Chionodes H€ ubner), Lasiocampidae (Gloveria Packard), Geometridae (Glaucina Hulst, Nemoria H€ ubner, Synchlora Guen ee), and Saturniidae (Hemileuca Walker) (Powell and Opler 2009). Gelechiid moths were one of the most frequently detected types of prey (26/280:~9%). Other common nonlepidopteran herbivorous prey included potentially pestiferous chrysomelid beetles such as Zabrotes Horn (Meik and Dobie 1986) and plant bugs (Miridae; Culliney 2014). Balduf (1941Balduf ( , 1942Balduf ( , 1943 conducted an observational study in a tallgrass prairie community in Illinois on the diet of a related ambush bug, Phymata americana Melin, and reported many of the same families of prey detected here. All six prey insect orders identified by Balduf were also recovered in this study on P. pacifica (Coleoptera, Diptera, Hemiptera, Hymenoptera, Lepidoptera, and Neuroptera).

Prey detectability half-life
Musca domestica DNA fragments of~268 bp were successfully amplified from more than half of the fed P. pacifica for each post-feeding interval up until the 72 h mark (Appendix S1: Fig. S2). Based on the regression analysis, the DNA detectability half-life for prey in the guts of P. pacifica was estimated to be 90.6 h (intercept: 0.9781; slope: À0.005276). This detectability window is substantially longer than most insect predators for which half-lives have been estimated . Given their sitand-wait predation strategy, ambush bugs, like many spiders, may have a lower metabolic rate and digest food slower than active-foraging predators (Anderson 1970, 1996, New 1975, Greenstone and Bennett 1980, 2014, Kobayashi et al. 2011, Virant-Doberlet et al. 2011. A long DNA detectability half-life could help explain why numerous ambush bugs from the MGCA survey simultaneously yielded DNA from two or more prey taxa (Table 1; 82/  225:~36%).
Heteropterans examined to date exhibit a wide spread of DNA detectability half-lives that range from less than a day to more than three days (Simmons et al. 2015), placing P. pacifica on the greater end of this spectrum. Zelus renardii (Kolenati), the only other assassin bug for which a detectability half-life has been estimated, also exhibits a rather long PCR prey detectability half-life time window of 51 h (Fournier et al. 2008). Our findings, as well as those from prior studies, are in line with the general notion that true bugs have relatively long detectability halflives compared to other predatory insects such as beetles (Agust ı et al. 2003, Anthocoris;Greenstone et al. 2007, Podisus;Hosseini et al. 2008, Nabis).

Efficacy of methods
In total, 280 different prey items were detected (Table 1). Of the 225 total P. pacifica gut samples sequenced, 203 remained after DADA2 filtering and denoising. Prey amplicon sequence variants were recovered from more than half of these (151 P. pacifica specimens). We also detected multiple prey items simultaneously from the guts of 82 different ambush bugs. Using a 95% identity threshold for sequences, 195 (~69%) of the 280 total prey items were identified to the generic or species level. Although the arthropod communities of CSS in Southern California have been relatively well sampled (Buffington and Redak 1998, Burger et al. 2003, Hung et al. 2015, it is clear that many taxa are yet to be COI-barcoded. The local database we compiled also facilitated identification; matches for 27 prey items (~10% of the 280 total) were obtained through BLAST searches against our COI database.
Approximately 80 morphospecies of buckwheat-associated arthropods were collected from CSS communities at our two field sites and were identified to the lowest taxonomic rank possible (Appendix S1: Table S2). We generated COI barcoding sequences for 51 of these specimens that did not have sequences available online (see Appendix S1: Table S2 for GenBank accession numbers). Among these, nine matched with 97% or greater identity to amplicon sequence variants recovered from the guts analyzed. Despite our efforts, sampling of non-Phymata arthropods was not comprehensive as we obtained prey sequences for numerous taxa that were not observed or collected in the field. This is likely a result of sampling time bias and/or ineffective collection methods (beating and sweeping vegetation and aerial netting).
The P. pacifica-specific blocking primer developed for this study appeared to greatly limit the amplification of host DNA as we witnessed a strong negative correlation between blocking primer concentration and the resulting visual signal from host DNA (Appendix S1: Fig. S1). This approach of coupling host-specific blocking primers with gut metabarcoding shows promise for use with predatory arthropods. Thus far, only a few molecular gut content analyses have been conducted on Reduviidae, the largest clade of non-holometabolous predators (~6800 spp.; Weirauch et al. 2014). While these studies have investigated the vertebrate host association of blood feeding kissing bugs (Reduviidae: Triatominae; Georgieva et al. 2017) and narrow diet range of termite assassin bugs (Reduviidae: Salyavatinae; Gordon and Weirauch 2016), this is the first study to evaluate the diet of a generalist assassin bug from a natural community using molecular gut analysis.

Limitations and solutions
Molecular gut content analysis can be limited or derailed by a host of issues. When conducting analyses that rely on universal primers, primer bias and taxonomic range are major concerns as they may fail to amplify certain taxonomic groups (Deagle et al. 2014, Sharma and Kobayashi 2014, Piñol et al. 2015. The universal primer set used here was highly effective and amplified DNA from 58 (~93.5%) of the 62 buckwheat-associated taxa for which PCR was attempted (Appendix S1: Table S2). We failed to amplify COI from two different hymenopterans, one heteropteran, and one coleopteran.
Our power to draw conclusions regarding trophic interactions ultimately hinges on the taxonomic breadth and reliability of databased sequences. Additional sampling of buckwheatassociated arthropods enabled us to identify some taxa for which limited sequence data are publicly available. However, even with additional sampling of flower-visiting taxa from CSS, we sequenced many amplicon variants that could not be classified below genus. It is clear that available COI sequence databases for CSS arthropods lack completeness, which is not surprising given the great biotic diversity associated with this community.
While na€ ıvely reporting secondary predation (i.e., committing a false-positive error) is a potential problem when conducting MGCAs on predators that engage in intraguild predation (Sheppard et al. 2005, Hagler 2016), many of the prey items were identified from guts which bore DNA from only a single prey taxon (70/280: 25%) or multiple taxa of which none are considered to be entomophagous (24/280:~8.6%). Ambush bugs, like all Hemiptera, possess piercing-sucking mouthparts and must extra-orally digest their food before siphoning it through a food canal formed by their maxillary stylets. Whether or not DNA from a previous meal in the prey's alimentary tract spills into the body cavity and is secondarily acquired by the true bug predator ultimately hinges on the time allowed for digestion and/or the ability of the stylet bundle to lacerate the gut (Cohen 1995).

Future directions
This study aimed to categorize ambush bug diet for only a short period in early summer and does not address the CSS community from a phenological prospective. The short timeframe allowed us to pool P. pacifica samples into one dataset and maximize sample size for the MGCA survey but inhibited us from comparing trophic interactions across an entire season. Future studies could potentially investigate temporal diet changes in generalist predators as different dominant plants come into bloom (e.g., California sage (Artemisia californica Lessing), chamise (Adenostoma fasciculatum Hook. & Arn.), or broomsage (Lepidopartum squamatum Gray)), as the temporal diversity of pollinators may vary (Hung et al. 2017).
In seeking to better understand natural systems, studies such as this provide useful data that can facilitate improved modeling of trophic networks. This study analyzes the diet of a single generalist from a community which supports a plethora of predatory arthropods, offering a unique perspective into trophic interactions in a diverse ecosystem. A diversity of predators were found hunting on buckwheat in relatively high abundance, including many crab, jumping, and lynx spiders (Thomisidae, Salticidae, and Oxyopidae, respectively) as well as other reduviids (Apiomerus californicus Berniker and Szerlip, Zelus renardii Kolenati, and Zelus tetracanthus St al).
Phymata pacifica engage in an array of trophic interactions with pollinators, herbivores, and other entomophagous arthropods found in CSS communities in Southern California. While a wide diversity of hymenopteran pollinators were preyed upon, we detected DNA more frequently from non-pollinating taxa. We advocate that more studies make use of gut content metabarcoding to categorize trophic interactions between generalist predators and their prey in other complex and understudied natural systems. Since predation by generalists can have cascading effects across multiple trophic levels, it behooves us to discover and characterize their feeding habits.

ACKNOWLEDGMENTS
This study was funded by the Shipley-Skinner Reserve-Riverside County Endowment. The authors wish to thank the following people: Paul Rugman-Jones and Erin Rankin for advice on molecular protocols; Chrissy Dodge for aid with specimen collection; Erin Rankin and Jong Soon Lee for allowing us to run several test samples on a MiSeq lane; Alexander Knyshov and Eric Gordon for their kind assistance with script writing and execution of post-sequencing analyses; Austin Baker, Jacob Cecala, Alexander Knyshov, Adrian Mayor, Kyle Whorrall, and Doug Yanega for assistance with identification of non-reduviid buckwheat-associated arthropods; and finally Stephanie Castillo, Chrissy Dodge, John Heraty, Rochelle Hoey-Chamberlain, Alexander Knyshov, Carlos Rosas-Sanchez, and Samantha Smith for draft editing and their contribution of helpful comments.