Intraguild predation (IGP) – predation between generalist predators (IGPredator and IGPrey) that potentially compete for a shared prey resource – is a common interaction module in terrestrial food webs. Understanding temporal variation in webs with widespread IGP is relevant to testing food web theory. We investigated temporal constancy in the structure of such a system: the spider-focused food web of the forest floor. Multiplex PCR was used to detect prey DNA in 3,300 adult spiders collected from the floor of a deciduous forest during spring, summer, and fall over four years. Because only spiders were defined as consumers, the web was tripartite, with 11 consumer nodes (spider families) and 22 resource nodes: 11 non-spider arthropod taxa (order- or family-level) and the 11 spider families. Most (99%) spider-spider predation was on spider IGPrey, and ~90% of these interactions were restricted to spider families within the same broadly defined foraging mode (cursorial or web-spinning spiders). Bootstrapped-derived confidence intervals (BCI’s) for two indices of web structure, restricted connectance and interaction evenness, overlapped broadly across years and seasons. A third index, % IGPrey (% IGPrey among all prey of spiders), was similar across years (~50%) but varied seasonally, with a summer rate (65%) ~1.8x higher than spring and fall. This seasonal pattern was consistent across years. Our results suggest that extensive spider predation on spider IGPrey that exhibits consistent seasonal variation in frequency, and that occurs primarily within two broadly defined spider-spider interaction pathways, must be incorporated into models of the dynamics of forest-floor food webs. 

Study system

We collected spiders and potential non-spider prey from the oak-dominated (Quercus alba and Q. rubra) Swallow Cliff Woods (41° 40.519’ N, 87° 51.437’ W) within the 320-ha Swallow Cliff nature preserve in Palos Township, Illinois (USA). The preserve, which is within the Chicago metropolitan region, is managed by the Cook County Forest Preserve District. Forests in this region are actively managed for several invasive plants (23), and the forest floor at Swallow Cliffs contains a thick leaf-litter layer with an abundant and diverse arthropod community. 

Collecting spiders and non-spider prey

Our goal was to search the ground layer and low understory as thoroughly as possible, so that we would collect enough spiders from less-abundant families to yield the same number of spiders per family analyzed for prey DNA. We did not estimate spider densities. All collections were made between 1000 and 1600 hours. We collected from a different location each day. The size of the area searched each day was not measured and varied with the number of searchers. Collecting areas were widely distributed throughout Swallow Cliff Woods, but we did not subdivide the Woods into sampling regions. Most terrain was upland forest, but some collections were taken from a few scattered wet/marshy areas. The number of collecting days in each season was spring (31), summer (33), and fall (29) over the years 2009, 2010, 2011 and 2012; the number of days per year was 33, 12, 34 and 14, respectively.

On each collecting day, we used both litter sifting and simple searching to capture spiders from several microhabitats. For litter sifting, we placed litter collected by hand into a flat tray (58 cm x 17 cm x 15 cm) with a screen bottom. This tray was shaken over a second tray of the same size with a solid bottom, allowing arthropods to fall through the screen to be collected by hand or aspirator. Sifted litter was returned to its original location. Spiders were also collected by hand from the litter surface, open areas in the litter, logs, low vegetation up to ~1m, and tree trunks up to ~2m. Individual spiders were placed in separate labelled vials.

Of the spiders that were eventually analyzed for prey DNA (see below), 81% were captured from either leaf litter (70%) or adjacent bare ground/logs (11%). Thus, most spiders were collected from the litter layer broadly defined. The litter layer is a fairly distinct subsystem with respect to rates of migration of arthropod predators and prey (24). Nevertheless, we did not limit our definition of the “forest floor” to the litter layer because many spiders spin webs in vegetation close to the ground. Also, some cursorial species move back and forth between the ground and lower understory vegetation and tree trunks (for example, 84% of the Corinnidae, a guild of “foliage runners” (25), were collected from leaf litter). Therefore, we also analyzed spiders that had been collected from low vegetation (10%) and tree trunks (9%).

All specimens were placed on ice within one hour of capture. On the same day, spiders collected for detection of consumed prey using PCR were taken to the laboratory where they were weighed and stored at -20◦C in a 1.5-mL microcentrifuge tube containing 95% ethanol (EtOH). Spiders and non-spider prey (see below) intended for primer development or assay optimization (see below for details) were kept alive, weighed, placed individually into 60-mL glass vials, and provided with water ad libitum at room temperature. Spiders were identified to family and genus using identification guides (26-29). Voucher specimens (one adult male and female) for each genus (when available) were archived at The Field Museum (Chicago, Illinois).

Over the four years, ~14,000 spiders (juveniles and adults) from 20 families were collected. Presence of prey DNA was tested for adult spiders from 11 abundant families (those with at least 300 adults) that live primarily on the forest floor. Spiders from six of these families (Corinnidae, Gnaphosidae, Lycosidae, Pisauridae, Salticidae, and Thomisidae) do not spin webs to capture prey (“cursorial” spiders). The other five families (Agelenidae, Dictynidae, Hahniidae, Linyphiidae, and Theridiidae) are “web spinners.” This dichotomy reflects basic differences in foraging behavior (16, 17), but the distinction is not absolute. The web spinners in our food web include genera of spiders that also forage for prey off their web (18).

Non-spider arthropod prey were also collected for primer development. They were not sampled quantitatively, but were simply selected due to their apparent abundance in leaf litter and/or activity just above the litter layer, and their likely occurrence in the diets of at least one spider family (15-17, 30). Non-spider nodes of the food web were broadly defined taxonomically (at the Order level except for Gryllidae):  flies (Diptera), moths/butterflies (Lepidoptera), springtails (Collembola), ants/bees/wasps (Hymenoptera), jumping bristletails (Archaeognatha), crickets (Gryllidae), pseudoscorpions (Pseudoscorpiones), harvestmen (Opiliones), beetles (Coleoptera), earwigs (Dermaptera), and pillbugs (Isopoda).

Molecular techniques

Primer development and optimization

We utilized multiplex PCR to sequence DNA from at least ten spiders from each family and at least ten specimens from each non-spider prey taxon. Each spider was first starved for at least ten days to eliminate any gut-content DNA that may have been present. Specimens were then homogenized in 180 μL of phosphate-buffered saline (PBS) (Hoefer, San Francisco, CA). DNA was then extracted with a Qiagen DNEasy Tissue Kit (Valencia, CA) using the manufacturer’s protocol. Upon completion of DNA extraction, the 200μL of eluate was well-mixed, separated into 20μL aliquots, and stored at -20°C until analysis.

The general arthropod primers LCO-1490 and HCO-2198 (31) were used to amplify DNA from the mitochondrial genome’s cytochrome oxidase I (COI) region. Eluate from DNA extractions was amplified and sequenced by The Field Museum (Chicago, IL) or Research Resources Center (RRC) at the University of Illinois, Chicago. Sequences were used to conduct BLASTN searches following the protocol developed by (32) using the databases GenBank and BOLD (the Barcode of Life Database). Following (33), database sequences were used only if they showed ≥97% match to submitted sequences. Sequences were aligned using the CLUSTALW or AMPLICON programs. Primers were designed with the assistance of the IDT (Integrated DNA Technologies, Coralville, IA) program PrimerQuest and tested for melting temperature and CG content using Sci-Tools OligoAnalyzer (IDT).

Spider gut-content testing

After a PCR assay was developed and optimized for a particular prey taxon (spider family or non-spider arthropod), frozen field-caught adult spiders were tested for the presence of the target-prey DNA. Spiders were thawed to room temperature and underwent DNA extraction and PCR amplification as described above. The entire spider was homogenized, except for the largest individuals, for which legs were removed to increase the prey/predator DNA ratio; coxae were left attached to the body when possible because spider guts often extend into the coxae (17). The homogenate was then mixed and 4uL were added to a well (on a 96-well plate) that contained 21 uL of Master Mix. Every run also included positive, negative, and blank controls to ensure that target DNA was amplified and that no contamination existed on the run. Positive controls consisted of DNA specific to the target taxon in question, negative controls contained the PCR Master Mix without DNA template, and blank controls were created from MBG water. A sample was considered positive for target-prey DNA within the spider’s gut if the Ct value of the amplification curve was above the background threshold, if the shape of the curve was sigmoidal, and if the positive and negative controls were acceptable. Samples that did not show amplification were re-analyzed using arthropod-general primers (31) before identifying them as negative results; questionable samples (low amplification or a non-sigmoidal shape) were re-tested.

For constructing the food web, adult spiders (as close as possible to 1:1 :: male:female)  were selected at random from genera within the most abundant families to yield 25, 50, 75, 100, or 150 per genus (roughly in proportion to relative abundances) for a total of 300 / family. Each spider was analyzed for DNA from the entire range of potential prey, i.e., the 11 non-spider arthropod taxa and the other 10 spider families. Only 17 out of 3300 spiders tested positive for two different prey items; these were treated as two separate interactions (as if there were two spiders). Only one spider tested positive for more than two prey: a thomisid tested positive for 7 different prey taxa, which was considered an outlier and was removed from the analysis (refer to S1 File for a discussion of possible causes of the low number of spiders testing positive for more than one prey item).  Because of these minor adjustments, the number of spiders tested for prey in our statistical analysis ranged from 299-304 per family.

Analysis of food web structure

Interaction matrix

PCR results were translated into an 11 x 22 interaction matrix in which each row (i) represented a consumer, i.e. one of the 11 focal spider families; and the ith cell in each column (j) contained the number of the ith consumer that tested positive for DNA for one of the other spider families or one of the 11 non-spider taxa. Intra-family predation, which includes cannibalism within species, no doubt occurred in our study system, but could not be detected by PCR. Thus, matrix cells (i,i) by necessity = 0, and the food web that was analyzed contained not 11 x 22, but  11 x (11+10) = 231 potentially detectable predator-prey interactions.

The entire matrix, with years and seasons pooled, constituted the summary web. To examine temporal variability, we created interaction matrices for each year (2009, 2010, 2011, and 2012; seasons pooled) and each season (spring, summer, and fall; years pooled). We did not analyze matrices for each year x season combination because small sample sizes for many combinations would have made such analyses to be of questionable value. 

Food web structure:  Graphical depictions

Interactions in our tripartite food web can most easily be depicted as an interaction network using a graphical technique that places the consumer categories (the 11 spider families) in the middle row, with triangles connecting each spider family with the non-spider prey depicted in the lower row, and spider prey in the upper row (34). These diagrams reflect the direction and magnitude of trophic interactions based solely upon spiders that tested positive for prey DNA. The width of each triangle reflects the relative proportion of that prey in the consumer’s diet. Proportions of tested spiders that had no detectable prey DNA are reported separately (S1 Table).

Food web structure:  Indices

Numerous indices have been proposed to measure food web structure. The goal often has been to determine whether the observed index indicates a degree of organization and distribution of links between nodes that differ from a null hypothesis of a random arrangement. We addressed a different question:  do measures of food web structure vary temporally?  Three indices were calculated that are appropriate to evaluating changes in a spider-focused, hybrid source-sink web. The first two, restricted connectance and interaction evenness, are abstract measures of overall network structure. The third measures the frequency of predator-predator interactions within the context of widespread IGP.

Restricted connectance -- Measures of connectance compare the number of realized links to the number of possible links in a food web (1, 35). The number of possible links varies according to assumptions about the nature and direction of predator-prey interactions in the web, yielding a range of specific terms for connectance (36). We used a definition, more restrictive than most, that (i) matches the hybrid sink-source structure of our spider-focused web, and (ii) reflects the limitations of our PCR analyses, i.e. we could not test for intra-family predation (analogous to cannibalism in a food web in which species are nodes). To make this distinction clear, we use the term restricted connectance, defined as: C_(R )=  L/(J(J-1)+JI)

where L is the number of realized links, J is the number of consumers (spider families), and I is the number of non-spider prey. Restricted connectance, which can vary from 0 to 1, is calculated from a binary (i.e. presence/absence ) interaction matrix. We re-sampled, with replacement, from subsets of the full data set to estimate margins of error (95% BCI) for yearly and seasonal estimates of restricted connectance. From 5,000 simulated values we extracted the median, and the 2.5 and 97.5 percentiles for the bootstrapped confidence interval (95% BCI). Analyses were performed in R version 3.2.1. For the yearly webs, three families (Corinnidae, Pisauridae and Gnaphosidae) were removed from the full data set due to low numbers collected in 2009; 25 spiders were then sampled, with replacement, from each family for each year. Three families (Corinnidae, Pisauridae and Agelenidae) were removed before bootstrapping seasonal data due to low numbers collected in one season; 30 spiders were then sampled, with replacement, from each remaining family for each season.

Interaction evenness: This metric reflects the relative intensity of interactions across a quantitative food web by incorporating the actual values in each cell of the interaction matrix. The metric was calculated using the “network-level’ function in the ‘bipartite’ R package (37). The metric scales between 0 and 1, with 1 = maximum evenness. We employed the “sum” version of the index, which calculates the index only for realized interactions.  Margins of error (95% BCI) for yearly and seasonal values of interaction evenness were estimated using the procedure employed for restricted connectance. 

% IGPrey: A spider preyed upon by another other spider was an IGPrey if both spiders shared at least one non-spider prey, i.e., if both spiders, along with the non-spider prey, formed an IGP module. We defined % IGPrey as the percentage of all detected prey that were classified as IGPrey. Margins of error (95% CI) for values of % IGPrey by year (seasons pooled) and by season (years pooled) were based upon the normal distribution because sample sizes and success/failure rates satisfied assumptions. Whether there was an interaction between year and season was evaluated by selecting the best model for a 4 x 3 x 2 contingency table using a log-linear model with a log link function and a Poisson response (38, 39). The analysis was carried out using SAS version 9.4.