Money spider dietary choice in pre‐ and post‐harvest cereal crops using metabarcoding

Money spiders (Linyphiidae) are an important component of conservation biological control in cereal crops, but they rely on alternative prey when pests are not abundant, such as between cropping cycles. To optimally benefit from these generalist predators, prey choice dynamics must first be understood. Money spiders and their locally available prey were collected from cereal crops 2 weeks pre‐ and post‐harvest. Spider gut DNA was amplified with two novel metabarcoding primer pairs designed for spider dietary analysis, and sequenced. The combined general and spider‐exclusion primers successfully identified prey from 15 families in the guts of the 46 linyphiid spiders screened, whilst avoiding amplification of Erigone spp. The primers show promise for application to the diets of other spider families such as Agelenidae and Pholcidae. Distinct invertebrate communities were identified pre‐ and post‐harvest, and changes in spider diet and, to a lesser extent, prey choice reflected this. Spiders were found to consume one another more than expected, indicating their propensity towards intraguild predation, but also consumed common pest families. Changes in spider prey choice may redress prey community changes to maintain a consistent dietary intake. Consistent provision of alternative prey via permanent refugia should be considered to sustain effective conservation biocontrol.


Introduction
Effective pest control requires an integrated, augmentative approach, aimed at maximising the effectiveness of natural enemies (Bale et al., 2008;Peterson et al., 2016). Spiders are polyphagous generalist predators naturally abundant at densities of 200-600 m −2 in UK crops (Nyffeler & Sunderland, 2003;Shayler, 2005). As a community, they employ a diversity of foraging techniques (Turnbull, 1973;Riechert & Lockley, 1984) which influence food webs via a range of hunting strategies including passive sit-and-wait predation from webs, and active hunting . Acknowledgement that spiders are an effective biocontrol agent has existed for causing major changes to invertebrate community composition and interactions through immigration/emigration and potential exposure to other predators Davey et al., 2013). The large degree of turnover in invertebrate communities following harvest profoundly affects the diet of generalist predators, with the changes in spatial co-occurrence of predator and prey fundamentally influencing predation events (Bell et al., 2010). Given that many spiders over-winter in the field, field margins, and hedgerows (Sunderland & Samu, 2000), the post-harvest provision of prey for these predators will influence their abundance and ability to suppress early pest populations in the subsequent crop cycle . To understand more precisely how harvest affects spider behavioural dynamics and how to optimise prey availability to support over-wintering and early-season spiders, the prey choice and dietary dynamics of spiders during this period must first be analysed.
Web-building spiders are effective models of prey choice, with webs providing a proxy for foraging investment (Welch et al., 2016). Spiders are known to forgo abundant prey in favour of less locally abundant taxa (Agustí et al., 2003;Welch et al., 2016). Studies of spider prey choice have mostly consisted of laboratory feeding trials (e.g. Mayntz et al., 2005;Rendon et al., 2019). Given that spiders are fluid feeders, morphological analysis of gut contents is impossible, thus field studies are restricted to direct observation and molecular methods (Symondson, 2002;Harwood et al., 2004;Bell et al., 2010;Pompanon et al., 2012;Chapman et al., 2013;Birkhofer et al., 2017). The parallel identification of many species by variations in short sections of variable genes, termed DNA metabarcoding (Pompanon et al., 2011;Taberlet et al., 2012), is one such molecular method for field-based analysis of spider diet (Piñol et al., 2014;Lafage et al., 2019). A target gene must be selected and a region identified which is variable between target species (the barcode) but flanked by two conserved regions, commonly in the cytochrome c oxidase subunit I (COI) gene for studies of animals due to its mutation rate and the extensive reference libraries available (Hebert et al., 2003;Deagle et al., 2014). These short sections of DNA, typically of 200-400 base pairs, can then be amplified using polymerase chain reactions (PCRs) with primers. Primers are short synthetic oligonucleotides which are complementary to the regions either side of the barcode that are conserved across the target taxa; these catalyse the amplification of the short sections of DNA (Folmer et al., 1994;Piñol et al., 2018). Using high-throughput sequencing (HTS), such PCR-amplified short sections of DNA can be identified in parallel from a single sample, such as the prey in a spider's gut. This ultimately provides an efficient and accurate method for analysing the prey range of predators (Pompanon et al., 2012;Piñol et al., 2018).
Two factors, however, must be overcome if HTS and DNA metabarcoding are to be used for dietary analysis of arthropod-consuming arthropods: (1) if the predator is an arthropod, then there is a high probability that existing general arthropod PCR primers will amplify the predator as well as the prey in its guts; (2) tissue from the spider predator will be undegraded and will hence swamp amplification of the prey (Vestheim & Jarman, 2008;Piñol et al., 2014). Applying universal primers and accepting a loss of data to amplification of the predator is feasible with sufficient read depth and allows analysis of interactions between closely-related species (Piñol et al., 2014). Blocking primers can be used alongside universal primers for the prevention of amplification of the predator whilst still amplifying the prey (Vestheim & Jarman, 2008;Deagle et al., 2009), but these can introduce strong taxonomic biases (Piñol et al., 2015(Piñol et al., , 2018. Primers can be designed to exclude amplification of the predator whilst still amplifying a broad range of prey species, and such primers have been designed for wolf spiders (Lycosidae; Lafage et al., 2019). However, these primers amplify money spider DNA and are thus not appropriate for analysing linyphiid gut contents. Taxonomically-similar predators and prey, such as intraguild spider-spider interactions, may also be undetected when using primers that exclude amplification of the predator DNA (Vestheim & Jarman, 2008;Piñol et al., 2014).
In this study, we aimed to analyse the diets and prey choices of linyphiid money spiders (Linyphiidae) in cereal crops preand post-harvest using DNA metabarcoding. To facilitate this, we developed novel PCR primers for the analysis of spider diet using high-throughput sequencing, with a specific focus on the diet of linyphiids. We hypothesised that linyphiid prey availability and diet would change following harvest, with diet largely reflecting prey assemblage turnover. However, we expected the prevalence of some prey species in the diet of the linyphiid predators to be disproportionate to their availability, reflecting prey choice.

Primer development and testing
Existing PCR primers were tested and ultimately redesigned to match the target taxa of this study. Two novel primer pairs were used for amplification of DNA for the dietary analysis of spider gut contents to overcome the problems associated with the taxonomic proximity of spiders and their prey (particularly other spider species). Novel PCR primers were adapted for the exclusion of all spider DNA, with a focus on linyphiids (henceforth spider exclusion primers, titled TelperionF-LaurelinR), based upon a primer site slightly 3 ′ of the general animal barcoding primers LCO1490 (forward primer Folmer et al., 1994), and mICOIintR (Leray et al., 2013). A second primer pair was employed for broad amplification of both spiders and their prey (henceforth general primers, titled BerenF-LuthienR), based upon mICOIintF (Leray et al., 2013) and HCO2198 (Folmer et al., 1994). Both primer pairs were adapted via base changes designed with reference to mass alignments of invertebrate COI sequences and tested in silico and in vitro. The spider-exclusion primers were designed to overcome the loss of reads to predator DNA, whilst the general primers were designed to avoid the taxonomic biases associated with the exclusion primers.
Mass-alignments of COI sequences were batch-downloaded from GenBank (NCBI) and BOLD (Ratnasingham & Hebert, 2007) using PrimerMiner (Elbrecht & Leese, 2016) in R v.3.3.4 (R Core Team, 2020 to aid visual inspection of existing ACTTCWGGRTGWCCAAARAAYCA Adapted from HCO2198 (Folmer et al., 1994) 2173 R 23 Note: The designed primer pairs, TelperionF-LaurelinR and BerenF-LuthienR, with amplicon sizes of 302 and 314 bp, respectively. and novel primer sites. PrimerMiner clusters batch downloads into operational taxonomic units (OTUs) based on sequence similarity and visualises mass alignments of sequence data for primer design. By merging overrepresented and duplicate sequences through taxonomy-independent clustering, Primer-Miner accounts for within-species variation and cryptic species whilst ignoring rare haplotypes (Elbrecht & Leese, 2016). Sequences were downloaded for all terrestrial invertebrate orders available, and consensus sequences were created by clustering these into OTUs for each order. The COI sequences were trimmed to include only the Folmer region (Folmer et al., 1994) using Geneious R10 (Kearse et al., 2012) for subsequent use in PrimerMiner. Alignments of prey sequences created via Primer-Miner included cereal crop spiders, in order to find primer sites conserved between a wide range of potential prey, but different for spiders. Where these sites were 100-400 base pairs apart on sequences from one another or from existing primer sites, they were paired, and primers designed (Table 1, Figs. S2 and S3). Existing general invertebrate primer sites were compared against the PrimerMiner alignments to identify any potential improvements to the primers for the amplification of cereal spider prey. The coverage of primers (% amplified) was determined via PrimerMiner using the same mass alignments used for primer design. PrimerMiner uses a taxonomy-independent database and accounts for adjacent base mismatches and the position of each base in the primer. Primers were also analysed using the online ThermoFisher Scientific Multiple Primer Analyzer tool. After the primers were deemed successful in silico, they were tested in vitro.
The primer pairs were tested in vitro against a wide range of extracted invertebrate DNA including spiders, common spider prey, and additional invertebrates. For this, invertebrate samples included those collected from the field site at Burdons Farm, Wenvoe, South Wales, the study site used for subsequent ecological analysis. Invertebrates were found via manual searching, collected via aspirator, and placed in microcentrifuge tubes of 100% ethanol. These were identified at 20-50X magnification using a light stereomicroscope and taxonomic keys (Goulet & Huber, 1993;Roberts, 1993;Unwin, 2001;Ball, 2008;Barber, 2008;Duff, 2012;Dallimore & Shaw, 2013). Additional invertebrates and DNA were taken from existing archived collections within Cardiff University. Extraction of DNA used DNeasy Blood & Tissue Kits (QIAGEN Inc., Chatsworth, CA) following the manufacturer's protocol for animal tissue. For predatory invertebrates, DNA was extracted from the lower legs, excluding the femur, to avoid the inclusion of prey DNA in the gut diverticulae and leg coxae (Macıas-Hernández et al., 2018).
To verify successful extraction, the DNA and negative controls were amplified via PCR with the Qiagen PCR Multiplex Kit (Qiagen) with 95 ∘ C for 15 min to activate the HotStarTaq ® DNA polymerase, 35 cycles of 95 ∘ C for 30 s, 40 ∘ C for 90 s and 72 ∘ C for 90 s, respectively, followed by a final extension at 72 ∘ C for 10 min using universal invertebrate primers LCO1490 and HCO2198 (Folmer et al., 1994). PCR reactions comprised 25 μl reaction volumes containing 12.5 μl Qiagen PCR Multiplex kit, 0.2 μmol (0.5 μl of 10 μM stock) of each primer, 6.5 μl DNase-free, water and 5 μl template DNA. Amplification was confirmed by gel electrophoresis.
Primers were initially tested against a small selection of spider and non-spider DNA. Temperature-gradient PCRs were used to determine the optimal annealing temperatures for the primer pairs selected, with temperatures between 40 and 60 ∘ C considered and initial tests starting 5 ∘ C lower than the mean melting temperature of both primers. Inclusion of the Q reagent supplied with Multiplex Kits was trialled for each pair to ascertain whether this could improve performance, but was ultimately excluded in all cases. PCR conditions were: 95 ∘ C for 15 min to activate the HotStarTaq ® DNA polymerase, 35 cycles of 95 ∘ C, the annealing temperature and 72 ∘ C for 30, 90, and 90 s, respectively, and a final extension at 72 ∘ C for 10 min. PCR reactions comprised 25 μl reaction volumes containing 12.5 μl Qiagen PCR Multiplex kit, 0.2 μmol (0.5 μl of 10 μM stock) of each primer, 6.5 μl DNase-free, water and 5 μl template DNA. Successful amplification was confirmed by gel electrophoresis. Once optimised to amplify a range of non-spider species whilst amplifying few spiders, or amplify a broad range of all species included, primers were further tested on a broader range of DNA (Table S3).
The TelperionF-LaurelinR primer pair has well-conserved sites, facilitating broad coverage with few degenerate bases necessary. The terminal base at the 3 ′ end of Laurelin, being a thymine base, critically mismatches with the guanine base present for most spider taxa tested; this should theoretically prevent or at least severely reduce amplification of spiders with little cost to amplification breadth otherwise. The BerenF-LuthienR pair similarly makes use of conserved primer sites employed in other studies but adapted for universal amplification of the focal taxa of this study.

Field collection and identification
Linyphiids were visually located on transects through two adjacent spring barley fields at Burdons Farm, Wenvoe in South Wales (51 ∘ 26 ′ 24.8 ′′ N, 3 ∘ 16 ′ 17.9 ′′ W), and collected from occupied webs and the ground, in August and September 2017. Each transect comprised 4 m 2 searching areas at least 10 m apart and all observed linyphiids were collected. Spiders were taken from 20 locations along the aforementioned transects, 10 pre-harvest, and 10 post-harvest. Spiders were collected 2 weeks prior (7-13th August) to harvest (∼20th August) of the crop and 2 weeks after harvest (4-8th September) in crop stubble and placed in 100% ethanol using an aspirator. Ground-active linyphiid spiders were collected when webs were not abundant. Spiders were taken to Cardiff University, transferred to fresh 100% ethanol, adults identified to species-level and juveniles to genus, and stored at −20 ∘ C until subsequent DNA extraction.
Invertebrate prey communities were collected using a converted McCulloch GBV 325 G-vac leaf blower suction sampler for 1 min over 4 m 2 areas near to those from which spiders for DNA analysis were collected. Samples were taken in transects, with 10 samples each pre-and post-harvest (20 total), split evenly between two adjacent fields. Invertebrate prey community samples were taken approximately 10 m apart, in sites near to those from which spiders were collected, with different sites used pre-and post-harvest. Invertebrates were killed with ethyl acetate and stored in 70% ethanol at −20 ∘ C, as these samples were for measurement of the invertebrate community and not for molecular analysis. All invertebrates were identified to family level under an Olympus SZX7 stereomicroscope using morphological keys, except for springtails of Sminthuroidea (Sminthuridae and Bourletiellidae, which were often indistinguishable following vacuum sampling and preservation due to the fine features necessary to distinguish them) which were left at super-family and mites (many of which were immature or in poor condition), which were identified to order level.

Dietary analysis
Erigone atra and E. dentipalpis (Erigoninae), and Tenuiphantes tenuis (Blackwall, 1852; Linyphiinae) were the focus of our study, although a few juveniles were included from other genera due to the difficulties associated with morphological identification of linyphiid juveniles; these misidentifications were confirmed in the subsequent metabarcoding. In total, 66 spiders were screened (Table S2), unevenly split across the 20 corresponding prey sampling sites. Spiders were washed in and transferred to fresh 100% ethanol to reduce external contaminants prior to identification using Roberts (1993) morphological key. Abdomens were removed from spiders and washed again in fresh 100% ethanol. Only abdomens were used for molecular analysis of their gut contents given their higher concentration of prey DNA than that of the cephalothorax (Krehenwinkel et al., 2016;Macıas-Hernández et al., 2018). To ascertain optimal extraction technique, samples were split into two groups. From one group, DNA was extracted from the abdomens via Qiagen TissueLyser II as per the DNeasy Blood & Tissue Kit (Qiagen, Hilden, Germany) manufacturer's protocol and abdomens kept in the lysis buffer during incubation. From the other group, DNA was extracted by splitting the abdomen with a sterile micropestle, swilling it in the lysis buffer, and then removing the bulk tissue. Neither method ultimately afforded a significantly greater proportion of prey DNA reads post-amplification (Fig. S1), so were combined for analysis. Post-lysis, all extractions followed the DNeasy Blood & Tissue Kit (Qiagen) manufacturer's protocol but with an extended lysis time of 12 h (recommended: 1-3 h) to account for the complex and branched gut system in spider abdomens (Krehenwinkel et al., 2016). Per 12 spiders, each DNA extraction session included at least one negative control consisting of an empty tube treated identically to the samples.
Primers were labelled with unique 10 bp molecular identifier tags (MID-tags) and samples had a unique pairing of forward and reverse tags for identification of each sample post-sequencing. PCR reactions of 25 μl reaction volumes contained 12.5 μl Qiagen PCR Multiplex kit, 0.2 μmol (2.5 μl of 2 μM) of each primer, 2.5 μl DNase-free, water and 5 μl template DNA. Reactions were carried out in the same Veriti Thermal Cycler (ThermoFisher Scientific, Waltham, USA), with annealing temperatures optimised via temperature gradient PCRs in the same machine. PCRs comprised 15 min at 95 ∘ C, followed by 35 cycles of: 95 ∘ C for 30 s, the primer-specific annealing temperature for 90 s, and 72 ∘ C for 90 s; followed by a final extension at 72 ∘ C for 10 min. The new primers, designated BerenF-LuthienR (universal) and TelperionF-LaurelinR (spider-excluding), used annealing temperatures of 52 and 42 ∘ C, respectively.
Within each PCR 96-well plate, 12 negative (extraction and PCR), and two positive controls were included following Taberlet et al. (2018). Negative PCR controls consisted of DNase-free water. Positive controls comprised known-concentration mixtures of the invertebrate DNA used for primer testing, detailed above, quantified using Qubit dsDNA High-sensitivity Assay Kits (ThermoFisher Scientific) to ascertain any effects of primer bias. All concentrations were standardised at 0.1 ng μl −1 by diluting the DNA in DNase-free water. Five mixtures of different species richness and proportions were prepared (Table S1). A negative control was present for each MID-tag to identify any contamination of primers. Each plate was pooled according to concentrations determined by Qiaxcel Advanced System (Qiagen). Each pool was cleaned via SPRIselect beads (Beckman Coulter, Brea, USA), with a left-side size selection using a 1:1 ratio (retaining ∼300-1000 bp fragments). The concentration of the pooled DNA was determined via Qubit dsDNA High-sensitivity Assay Kits, quality-checked via TapeStation 2200 (Agilent, Santa Clara, USA), and all pools sharing the same primer pair were pooled again into a 'super pool', thus forming one pool per primer pair. Library preparation for Illumina sequencing was carried out on these cleaned 'super pools' via NEXTflex Rapid DNA-Seq Kit (Bioo Scientific, Austin, USA) and samples were sequenced on an Illumina MiSeq via a Nano chip with 2 × 250 bp paired-end reads (expected capacity ≤1 000 000 reads).

Bioinformatic analysis
The Illumina run generated 405 270 and 482 249 reads using BerenF-LuthienR (universal) and TelperionF-LaurelinR (spider-excluding), respectively. All reads were quality-checked and trimmed in Trimmomatic v0.38 (Bolger et al., 2014) with a minimum quality score and sliding window of 20 and 4 bp, respectively, and a minimum length of 135 bp. The read pairs were aligned via FLASH v1.2.11 (Magoč & Salzberg, 2011) and demultiplexed via Mothur v1.39.5 (Schloss et al., 2009), removing the MID and primer sequences. Replicates were removed, and denoising and clustering to zero-radius OTUs (ZOTUs; clustered without % identity to avoid multiple species represented within a single OTU) completed via Unoise3 in Usearch11 (Edgar, 2010). The resultant sequences were assigned a taxonomic identity from GenBank via BLASTn v2.7.1. (Camacho et al., 2009) using a 97% identity threshold (Alberdi et al., 2017). The BLAST output was analysed in MEGAN v6.15.2 (Huson et al., 2016). Where the top BLAST hit, determined by lowest e-value, was resolved at a higher taxonomic level than species-level, the results were checked by blasting the sequence manually in GenBank and comparing the results; where possibly erroneous entries were preventing species-level assignment (e.g. poorly-resolved identifications on GenBank), finer resolution was considered. Where ZOTUs were assigned the same taxon, these were aggregated. Given the prevalence of family-level assignments (e.g. Chloropidae), the data were eventually converted to family-level, but were retained at their respective output assignments for clean-up.
To clean data prior to statistical analysis, all read counts less than the maximum read count present in blanks (negative controls and unused MID-tag combinations) for its respective ZOTU were removed. Instances of non-positive control taxa present in positive controls were calculated as a percentage of the maximum read count for that taxon. The greatest of these percentages was used to guide a universal percentage of the maximum read within each taxon to be removed. This accounts for tag-jumping and "bleeding" of over-represented taxa into other samples during sequencing. For BerenF-LuthienR, 0.54% was optimal, whilst there were no obvious instances for TelperionF-LaurelinR, so the conservative 0.54% was also applied for that library.
Simultaneously, known lab contaminants (e.g. German cockroach Blatella germanica (L.) and various species for which molecular analysis was recently undertaken that could be differentiated from the target taxa in this study, such as tropical species) were identified and the percentage of these occurrences of the total read count for their respective samples was calculated. The highest of these percentages was used as a guide for the universal percentage of each total sample read count to be removed. This accounts for environmental and lab contamination, and artefacts and errors of the sequencing process, which for BerenF-LuthienR and TelperionF-LaurelinR were 0.43% and 0.45%, respectively. The data from the two libraries were then aggregated together, first removing non-target taxa (e.g. fungi) and instances in which predator DNA was amplified (i.e. ZOTUs matching the individual spider's morphological identity). All taxa were converted to family-level to standardise the taxonomic level since many ZOTUs could not be resolved further; this also increases evenness for subsequent analyses. Whilst all conspecific reads were removed to account for predator amplification, interspecific linyphiid interactions were still retained, thus any counts of linyphiids in the diet exclusively represent the consumption of other species. Finally, read counts were converted to presence-absence data.

Statistical analysis
All analyses were conducted in R v.3.3.4 (R Core Team, 2020). Invertebrate communities were compared between pre-and post-harvest using multivariate generalised linear models (MGLMs) via 'manyglm' in the 'mvabund' package (Wang et al., 2012) with a binomial error family and Monte Carlo resampling. Spider diets were compared between pre-and post-harvest, and spider life stage (juvenile, sub-adult, and adult) and sex, using MGLMs that included all two-way interactions between these variables. Whilst the samples selected for sequencing primarily comprised Tenuiphantes tenuis and Erigone spp. with the intention of comparing these taxa, samples sizes were not sufficient for taxonomic comparisons due to sample drop-out in the sequencing process and the misidentification of some juvenile linyphiids (pre-harvest: Tenuiphantes = 19, Erigone = 5, other = 3; post-harvest: Tenuiphantes = 16, Erigone = 1, other = 2). Models were simplified using 'step' from the base R 'stats' package to determine an optimal model based on the lowest AIC value by removing variables. Dietary differences were also visualised by non-metric multidimensional scaling via metaMDS in the 'vegan' package (Oksanen et al., 2016) with Jaccard distance in two dimensions (stress = 0.082). Spider plots were created with nMDS results via 'ordispider' and plotted through 'ggplot2' (Wickham, 2016) to illustrate the distinction in communities between categories. Spider prey choice was analysed using network-based null models in the 'econullnetr' package (Vaughan et al., 2018) with the 'generate_null_net' command and visually represented with the 'plot_preferences' command. This detects whether prey taxa are consumed more or less frequently than expected based on their relative abundance in the community.

Dietary analysis
The general primers (BerenF-LuthienR) recovered an average of 4204 reads (8.39% reads were prey) and 1.93 prey taxa per spider, and the spider-exclusion primers (TelperionF-LaurelinR) recovered 3530 reads and 0.72 prey taxa per spider from those spiders from which prey were recovered by at least one of the primer pairs. The spider-exclusion primers successfully avoided amplifying any Erigone DNA but did amplify the DNA of the other predators (mostly Linyphiinae; in those cases, 2.99% reads were prey).

Prey choice analysis
Spiders exhibited prey choice (prey consumed by predators at a higher or lower relative frequency than expected based on availability) in both pre-and post-harvest periods (Figs 3  and S8). Pre-harvest spiders consumed Cicadellidae (leaf hoppers, Hemiptera), Linyphiidae, and Phoridae (humpbacked flies, Diptera) significantly more than expected, and Chloropidae (frit flies, Diptera), Ephydridae (shore flies, Diptera), and Isotomidae (Entomobryomorpha) significantly less than expected. Post-harvest spiders consumed Entomobryidae (Entomobryomorpha), Ephydridae (shore flies, Diptera), and Linyphiidae significantly more than expected, and Thripidae (Thysanoptera) and Delphacidae (plant hoppers, Hemiptera) significantly less than expected. All other taxa were consistent with the null models and were consumed at the rates expected from their density.

Discussion
Through the application of novel PCR primers, linyphiid diet was shown to differ following harvest, primarily due to changes in prey availability (cf. changing prey choice).
The primers presented in this study are the first designed specifically for gut content analysis of linyphiid spiders and provide an effective means of dietary analysis for a range of species. Both novel primer pairs performed well both in silico and in vitro, ultimately successfully detecting prey in the guts of field-captured linyphiid spiders with complementary coverage and overlap of detected prey taxa. The broad range of species amplified by the primers suggests applicability to other species and study systems. In the case of TelperionF-LaurelinR, this is particularly true for dietary analysis of spiders in the families Agelenidae, Pholcidae, and Clubionidae at least, as well as some success with ants and other predatory invertebrates given their lack of amplification. Whilst TelperionF-LaurelinR did ultimately amplify the DNA of Tenuiphantes, these primers facilitated effective gut content analysis for Erigone spiders, likely further Erigoninae species and possibly the other taxa listed above. Whilst no significant difference was found between the two extraction methods trialled, the mean percentage of prey reads per sample was slightly greater when the abdomen was left in the lysis buffer for the longer incubation period, aligning with the findings by Krehenwinkel et al. (2016).
This study provides an example of the use of general and predator-exclusion PCR primers together for the same dietary samples. That both primers exhibited taxonomic biases is unsurprising given this widespread phenomenon in metabarcoding (Piñol et al., 2018), but their complementary biases, determined via mock communities, is promising for their combined use, ultimately providing a more complete snapshot of diet. Given the increasing understanding of the effects of primer bias in metabarcoding (Piñol et al., 2018;Braukmann et al., 2019;Elbrecht et al., 2019), further studies should consider employing this approach to begin mitigating these effects without relying on the heavier biases associated with other methodologies, such as the use of blocking probes (Piñol et al., 2015). The low annealing temperature used for Telperion-Laurelin (42 ∘ C) was selected due to the exclusion of several prey species at higher temperatures, but could present issues such as dimerisation or non-specific binding, the latter of which can create differently-sized amplicons in some cases, including in another application of these primers (pers. obs. Jordan Cuff). Given the melting temperatures of around 54 ∘ C for these primers, other users could consider higher annealing temperatures when testing against their target organisms to avoid any associated issues, but, as highlighted by the data of this study, the primers will successfully amplify HTS-appropriate DNA from the guts of spiders even at these lower temperatures. Care must be taken, however, to ascertain whether the focal spider species is amplified prior to analysis. Modifications to the primer sequences may increase their specificity for the study of other spider species diets. These spider-exclusion primers, alongside NoSpi2 (Lafage et al., 2019), provide a complementary suite of primers for the metabarcoding of the diets of many spider taxa.
That distinct invertebrate communities were identified preand post-harvest indicates a substantial ecological impact of harvest on community composition and dynamics, and thus the provision of alternative prey for generalist predators. The significant decrease in isotomid abundance following harvest is noteworthy given that linyphiids regularly predate springtails such as Isotoma anglicana (Schäffer, 1896;Agustí et al., 2003;Harwood et al., 2003;Piñol et al., 2014). The loss of a major prey species could be detrimental to linyphiids immediately prior to winter, especially for those that overwinter as adults.
The relative avoidance of isotomids pre-harvest suggests that this may not be the case though, and that reduced abundance of isotomids may still be sufficient for linyphiid populations. Continuous provision of alternative prey not only reduces intraguild predation (Athey et al., 2016), but likely supports over-winter and early-season predation of pests, which is critical in curbing pest populations with short generation times such as aphids Korenko et al., 2010;Pekár et al., 2015). The similarity in linyphiid abundance pre-and post-harvest contrasts findings in a Mediterranean system which evidence declines in agrobiont linyphiid species ; climatic differences may account for this, but the importance of immigrant spiders in rapidly recolonising fields at the start of each cropping cycling and the substantial pest suppression that they provide cannot be overlooked .
The difference in diet between pre-and post-harvest spiders reflects the community turnover following harvest, with many crop-dependent species such as aphids disappearing from the diet post-harvest. This finding is consistent with Bell et al. (2010) in that community turnover, and thus changes in co-occurrences, facilitate dietary changes. Whilst only an overall significant difference was found, not specific taxonomic associations, some taxa were only found in diets in one of the two periods. Regardless of dietary differences, spiders differentially selected from the prey taxa available to them both pre-and post-harvest, inferring differences in prey choice between the two periods. A larger number of significant deviations from expected trophic interaction frequencies were found pre-harvest than post-harvest (7 vs. 4), suggesting that spiders can be more selective regarding their dietary intake, possibly due to greater prey abundance in this period.
Isotomids were predated less frequently than expected pre-harvest when their abundance was greatest, but not post-harvest, following the significant reduction in their abundance. Pre-harvest, spiders may choose alternative prey over isotomids to diversify their dietary intake despite the ease of consuming isotomids, possibly redressing nutritional deficiencies (including accumulation of prey toxins) that would result from consuming only one species (Mayntz et al., 2005). The greater-than-expected frequency of post-harvest consumption of ephydrids and entomobryids suggests that greater provision of these and similar taxa could benefit the health of linyphiid populations post-harvest. Such taxa could be supported by increased habitat complexity (Michalko et al., 2017) and continuous refuges following harvest, particularly grass margins and in-field 'beetle banks ' (MacLeod et al., 2004;Mansion-Vaquie et al., 2017). The relative avoidance of thrips post-harvest is likely a consequence of their dominance of the post-harvest invertebrate community, and the need to diversify dietary intake by the spiders. It should be noted that the primers used in this study may exhibit some bias against thrips in PCRs (Fig. S4), thus this result may alternatively be due to disproportional representation of thrips; greater sequencing depth or alternative PCR primers could mitigate this in future studies. Such biases are also observed for most metabarcoding primers, including poor amplification of thrips by the commonly-used primers ZBJ-ArtF1c and ZBJ-ArtR2c (Zeale et al., 2011;Gomez-Polo et al., 2016;Piñol et al., 2018;Silva et al., 2019).
Across both periods, linyphiids were prominent in the diets of the linyphiids screened. Since sequences conspecific with the predator were removed and contamination by tag-jumping accounted for, these were legitimate detections of intra-guild predation between linyphiid species. The inability to detect cannibalism via metabarcoding could mean that these instances of intra-guild predation are in fact an underestimate. That linyphiids are so prominent in their own diet, particularly post-harvest, and were preferentially consumed across both periods, could indicate increased competition for prey resources. Linyphiids will usurp one another for optimal web sites based on prey abundance, sometimes consuming one another in the process (Harwood & Obrycki, 2005), which is likely to increase during periods of prey scarcity. If prey choice is indeed influenced by nutritional requirement, linyphiids may also predate one another as perfect vessels for all of their nutritional needs. This intraguild predation is well-documented in linyphiids (Harwood & Obrycki, 2005;Davey et al., 2013) and is a concern regarding their effectiveness as biological control agents. Intraguild predation may, however, sustain linyphiid populations when suitable prey are absent. Increased habitat complexity could thus reduce intraguild predation through improved prey provisioning (Michalko et al., 2017) and improve the biocontrol capacity of linyphiids. Given that this harvest interface period lies beyond typical crop spraying times, the importance of natural enemies for biocontrol is arguably even greater. Linyphiids are consuming pests in this period (e.g. aphids, as shown in this study), so reductions in intraguild predation by altered management regimes could further enhance this primary control of pests as they begin over wintering, maximally impacting the early-season return of pests.

Conflict of interest
There are no conflicts of interest to declare and no disputes regarding ownership of the data presented in the manuscript.

Data availability
The data that support the findings of this study are openly available on Dryad at https://doi.org/10.5061/dryad.8cz8w9gng.

Author Contributions
JPC led the study, field work, and preparation of the manuscript. JPC, WOCS, PO-tW, JRB, IPV, and CTM designed and managed the study. JPC, LED, and JES contributed to the design and testing of the primers. JPC and LED processed and analysed the high-throughput sequencing data. JPC and MPTGT identified the invertebrates and processed these data. All authors contributed to the writing and editing of the manuscript.

Supporting Information
Additional supporting information may be found online in the Supporting Information section at the end of the article. Table S1: Mock community DNA mixtures. The volume of each 0.1 ng μl −1 sample added to each mock community mixture. Mock communities include some uniformly distributed volumes and others weighted in favour of specific taxa (most often spiders to reflect the abundance of spider DNA in spider diet extracts).  In vitro results for novel primer pairs. A) Arachnids, B) other invertebrates, C) vertebrates and marine. Both primer pairs show a broad amplification of many groups. Beren-Luthien proves to be very general, whilst Telperion-Laurelin avoids amplification of some spiders, but also a few other invertebrate species tested. Table S4: Invertebrate taxa identified from vacuum samples preand post-harvest. Vacuum samples were collected for 30 s over 4 m 2 . Each period represents the total from 10 samples evenly split between two adjacent fields. Table S5: Percentage of spiders in each category (pre-or post-harvest, sex, subfamily, and age class) that had consumed each of the families detected in gut content analysis. Table S6: Each spider from which a dietary sample was taken is given alongside the raw dietary data and the other information associated with that sample.

Fig. S2
PrimerMiner mass alignments of spiders and common cereal crop species for Beren and Luthien. Both primers utilise degenerate bases to increase the amplification of a broad taxonomic range. The conserved primer sites enable broad amplification with few degenerate bases.

Fig. S3
PrimerMiner mass alignments of spiders and common cereal crop species for Telperion and Laurelin. Both primers utilise degenerate bases to increase the amplification of a broad taxonomic range. The terminal base at the 3 ′ end of Laurelin is a critical mismatch for almost all spiders, making the primer pair efficacious for the amplification of all but spiders (G, rather than A).

Fig. S4
In silico analysis of primer bias for novel primer pairs compared against standard animal barcoding primers LCO1490 and HCO2198 (Folmer et al., 1994), and other primers used for linyphiid dietary analysis ZBJ-ArtF1c and ZBJ-ArtR2c (Zeale et al., 2011). The further from the centre that each coloured polygon extends at each anchor point reflects the relative performance of that primer for the respective taxon. Distances are relative, not absolute. Beren-Luthien show almost universally strong amplification potential for most taxa, except for thrips (Thysanoptera). Telperion-Laurelin show the expected bias against spiders (Araneae), but also weaker amplification of mesostigmatic mites and thrips; despite this, the amplification potential for most other groups is stronger than that of ZBJ, which has been used in other studies of linyphiid diet. The reduced amplification potential for thrips by both primer pairs illustrates a potential under-representation of them when paired for dietary analyses.

Fig. S5
Primer bias ascertained using mock community positive control and high-throughput sequencing. The distance of each point radially from the centre is the proportional success in amplifying the mock community component taxa. The "Expected" amplification is based on the proportion of each taxon's DNA in the mock community. Beren-Luthien shows some bias towards Lepidoptera (Euproctis similis) and some Diptera (Melieria crassipennis), but reduced amplification of linyphiids, springtails, wasps, and aphids. Telperion-Laurelin expectedly avoids amplification of spiders in most cases, although they are still amplified to some extent. The two primer pairs show moderately complementary biases, but both show some bias against linyphiids, and some springtails, flies, and wasps.

Fig. S6
Spider plot of invertebrate communities pre-and post-harvest showing distinction between communities (smaller nodes) and centroids of communities (larger nodes, mean coordinates for each category) within each category. Species present in the communities are plotted according to their alignment with sample dissimilarities.

Fig. S7
Spider plot of spider diets pre-and post-harvest showing some distinction with a degree of overlap between the prey families in the diets (smaller nodes) and centroids of diets (larger nodes) within both categories. Species present in the diets are plotted according to their alignment with sample dissimilarities.

Fig. S8
Full prey choice plot. Blue = lower consumption than expected (avoidance), white = as expected (in proportion to relative abundance), red = higher than expected (consumed more frequently than predicted from relative abundance). Horizontal lines denote 95% confidence limits of the observed frequency of predation.