Numerous studies have found that predators can suppress prey densities and thereby impact important ecosystem processes such as plant productivity and decomposition. However, prey suppression by spiders can be highly variable. Unlike predators that feed on prey within a single energy channel, spiders often consume prey from asynchronous energy channels, such as grazing (live plant) and epigeic (soil surface) channels. Spiders undergo few life cycle changes and thus appear to be ideally suited to link energy channels, but ontogenetic diet shifts in spiders have received little attention. For example, spider use of different food channels may be highly specialized in different life stages and thus a species may be a multichannel omnivore only when we consider all life stages. Using stable isotopes, we investigated whether wolf spider (Pardosa littoralis, henceforth Pardosa) prey consumption is driven by changes in spider size. Small spiders obtained > 80% of their prey from the epigeic channel, whereas larger spiders used grazing and epigeic prey almost equally. Changes in prey consumption were not driven by changes in prey density, but by changes in prey use by different spider size classes. Thus, because the population size structure of Pardosa changes dramatically over the growing season, changes in spider size may have important implications for the strength of trophic cascades. Our research demonstrates that life history can be an important component of predator diet, which may in turn affect community- and ecosystem-level processes.

Study site and organisms

We conducted this study at a salt marsh near Tuckerton, New Jersey, USA (39° 30.8′ N, 74° 19.0′ W) that is dominated by the cordgrass Spartina alterniflora (Denno et al. 2002, now reclassified as Sporobolus alterniflorus). We focused on the dominant species in the live plant and epigeic (algal and detrital) food webs; relationships among these dominant arthropods have been well characterized (Finke and Denno 2002, 2003, 2004, 2005; Wimp et al. 2013). Pardosa hatch and grow during the summer, overwinter as late-instar juveniles, mature to adults in the spring and begin reproduction in early summer (Dobel et al. 1990). The grazing herbivore assemblage in S. alterniflora is dominated by two congeneric planthopper species, Prokelisia dolus and P. marginata, which constitute upwards of 80% of herbivore biomass (Denno et al. 2000). The planthopper Delphacodes penedetecta and the mirid bug Trigonotylus uhleri rank as the next most abundant herbivores. The mirid bug Tytthus vagus consumes planthopper eggs exclusively, and frequently falls prey to Pardosa.

We focused on three species to represent the marsh epigeic food web because of their abundance at the study marsh, similar to the high relative abundances of the herbivore species listed above: Orchestia grillus, Venezillo parvus, and Saldula interstitialis. The amphipod O. grillus is abundant and feeds on surface epiphyton as well as Spartina detritus (Agnew et al. 2003; Lopez et al. 1977); although amphipods may not be part of a typical terrestrial detrital food web, they are commonly found in systems with a terrestrial–aquatic interface, both freshwater and marine. The isopod V. parvus is less abundant than O. grillus at this marsh, but ingests substantial amounts of Spartina litter (Zimmer et al. 2004). Finally, the saldid bug S. interstitialis is saprophagous in early instars, while older nymphs and adults are active predators feeding on surface and subsurface arthropods, including amphipods (Griesinger and Bauer 1990).

Arthropod samples

Four times during the summer of 2009 we sampled marsh arthropods from six blocks, each composed of three 10-m² sample plots. Within each block, the three plots were located in different habitats. One plot was located near the upland border of S. alterniflora with S. patens (this border occurs at mean high water level), a second plot was located in the center of the S. alterniflorameadow, and the third in intermediate form S. alterniflora near a tidal creek. We sampled the entire arthropod community on 16 Jun 2009, 20 Jul 2009, 18 Aug 2009, and 15 Sept 2009 using a D-vac suction sampler that we placed in ten different locations within each plot for 3-s periods following the methods of Wimp et al. (2013) and Murphy et al. (2012). We immediately placed the collected arthropods into closed containers with ethyl acetate, then transported them to the laboratory in coolers with dry ice and stored in a − 20 °C freezer until they were processed.

We sorted arthropod samples by species, counted the abundance of each species, cleaned them of foreign debris, and dried them at 60 °C for 3 days. Pardosa were then weighed and assigned to one of six size classes by dry weight (Appendix A). Size classes were chosen to divide the collected Pardosa into roughly equally represented groups.

Stable isotope analysis

We used stable isotopes of carbon to assess the level of multichannel omnivory by Pardosa over the course of the summer, by measuring δ¹³C values of Pardosa and their grazing and epigeic prey. While δ¹⁵N values are useful for determining the trophic level of an organism and we measured them in this study for this purpose, δ¹³C values reflect changes in food web use. Carbon isotopes have proven useful in determining the relative importance of dual food webs in the diets of generalist predators (Newsome et al. 2007; Wimp et al. 2013; Wise et al. 2006), because carbon isotopic signatures, δ¹³C, remain relatively unchanged from food source to consumer (Fry 2006). Therefore, if the δ¹³C values of the bases of two food webs are sufficiently distinct, a predator’s δ¹³C value reflects its relative consumption from the two webs even when intermediate feeding links are unclear (DeNiro and Epstein 1981; Post 2002). In previous work, we found that δ¹³C values in salt marsh grazing and epigeic webs differed substantially for at least part of the summer (Wimp et al. 2013). In the present study, we examined both δ¹³C to examine whether Pardosa diet changes across different size classes, and δ¹⁵N to confirm trophic position. We had to combine individuals of the smaller size classes to obtain sufficient tissue for stable isotope analysis, but only individuals collected in the same plot on the same day were combined. We then ground samples and obtained their isotopic values as described in Wimp et al. (2013).

Statistical analyses

We performed an ANOVA on Pardosa δ¹⁵N values across the different collection months to confirm predator status and see how this status might change during the season or depending on the S. alterniflora habitat type (center, intermediate, edge; Appendix B). In all ANOVA and regressions involving Pardosa isotopes, we treated all Pardosa recovered from a plot as repeated measures from that plot. We also examined how Pardosa δ¹⁵N changed with spider size using a correlation analysis (Appendix B).

To examine how multichannel omnivory was related to the size of Pardosa, we used carbon stable isotopes. Initially, we collected Pardosa from different habitats, but we wanted to determine whether carbon isotopes differed according to habitat or could be combined. Thus, we performed a repeated measures ANOVA by SAS proc mixed (SAS 2002), which showed that habitat within block had no effect on the δ¹³C of grazing prey (F_2,17.1 = 0.03, P = 0.97), epigeic prey (F_2,14.3 = 2.45, P = 0.12), or Pardosa (F_2,12 = 1.32, P = 0.30), and that habitat had no effect on Pardosa size (F_2,15.5 = 1.24, P = 0.32). Thus, the different habitats in each of our transects did not affect δ¹³C isotopic ratios and were combined for further analysis.

We were able to collect a large range of Pardosa sizes only in June and July (Appendix A), so we examined the relationship between δ¹³C and size only during those months. Because we had to combine individuals of the smaller size classes (1–3) to obtain sufficient tissue for stable isotope analysis (explained above), we were not able to collect enough individuals of the smaller size classes to obtain even a single sample for the months of August and September, and we only had a single sample in size class 4 (Appendix A, Figs. A1, A2). We performed separate ANOVA analyses in June and July with size class as a discrete explanatory variable and δ¹³C as the response. We included only size classes for which we had at least five measurements in a given month. Variances were not equal across size classes, so the ANOVA was performed using separate variance estimates for each size class with SAS proc mixed.