Methods
Study sites : The coastal regions included in the study were (a) Uppland north of Stockholm with the lowest salinity (≈5‰, northern region) and (b) Kalmar and Öland in southeastern Sweden with somewhat higher salinity (≈7‰, southern region) (Fig. 1, Supplementary Information Table 1). The numbers of coastal sites were 13 (Uppland) and 7 (Kalmar). Among these, two sites respectively had thick wrack beds, and the other sites were similar but without thick wrack beds and often with short-cut grass due to grazing. The thick wrack beds had a thickness of more than 20cm with a considerable extension (several 10s of meters). The non-wrack sites either lacked wrack almost completely (as in the Uppland region) or that wrack occurred in scattered patches and never so thick as to provide a suitable habitat for detritivores (as in the Kalmar region). The inland regions included 15 and 23 shoreline sites in Uppland and southern Halland (same latitude as Kalmar) respectively (Fig. 1, SI Table 1), as part of a broader study focusing on both insect and spider communities in wetlands.
Field sampling : Coastal wolf spider communities were sampled using 10 pitfall traps per site placed in the wrack (wrack sites) or in open ground (non-wrack sites) for three nights in early May 2022 (Kalmar) or early-mid June 2021 and late May 2022 (Uppland). Inland wolf spider communities were sampled during June 2020. These times were chosen because wolf spiders are then adults or subadults, which simplifies species identification. Spiders were placed in 70% ethanol and brought to the laboratory for identification. Spiders from the inland sites were identified by R. Vicente and those from coastal sites by M. Langbak and A. Hoffmann, with assistance from R. Vicente for complicated cases (mainly involving Pardosa agrestis/agricola/monticola ).
Spiders used for diet analyses were only collected from 13 sites, six from Uppland and seven from Kalmar, including all wrack sites. Spiders were individually collected by hand (30 per site), to reduce contamination risk, at two times (June and August 2019) and directly transferred to 95% ethanol. In the lab, samples were placed in a freezer (-20oC) until DNA extraction and further processing. Finally, prey densities were estimated by placing one SLAM (Sea Land Air Malaise) trap for two nights at the same time when collecting spiders for diet analyses. SLAM trap catches were placed in 70% ethanol, and brought to the laboratory for sorting to a family level.
Diet analyses : To metabarcode prey content of the hand-collected spiders, DNA was extracted from either a dissected abdomen (larger spiders) or the whole specimen (small spiders). To reduce the DNA yield of the focal spiders, we used a forward primer designed not to amplify wolf spider DNA (NoSpi2, Lafage et al. 2020) in combination with a general reverse primer (fwhR2n, Vamos, Elbrecht & Leese 2017) to amplify a section within the Folmer region of COI (Folmer et al.1994). Procedures for PCR amplification and library building follow Hambäck et al. (2021), and sequencing of the spider samples was performed in one batch on the Illumina MiSeq3 platform at SciLifeLab in Stockholm. To detect individual samples after sequencing, a dual tagging approach was used where the 5’-end of both primers included an 8 base-pair tag (Binladen et al. 2007). Illumina-adaptors bearing unique indices were then ligated to the phosphorylated amplicons without a PCR step to preclude tag jumping errors (Bohmann et al. 2021). Due to problems with low DNA content, we had to change strategy and add a second PCR step with a low cycle number (6). Because this additional step increases the risk of tag jumping errors, we built libraries separately for each site, using SMARTer ThruPLEX DNA-seq library preparation kit excluding fragmentation of DNA (Takara Bio), as tag jumps between spiders within site do not affect the results due to pooling at this level before analysis. In each library, we also included at least 25% empty combinations to estimate tag jumping errors (which was about 6%). After sequencing, we used ObiTools (Boyer et al.2016) within the Galaxy Platform (Jalili et al. 2020) to assemble paired end sequences of high quality (score > 40), trim primers, clean sequences using ‘obiclean’, and demultiplex resulting sequences to individual samples using ‘NGSFILTER’ after filtering for size. These procedures resulted in a data set of 367 spider individuals and about 384,600 prey sequences that were grouped based on 97% similarity and where representative sequences were taxonomically assigned using BoLD (Ratnasingham & Hebert 2007) before further analyses.
Statistical analyses : Spider communities were modelled as the abundance of each spider species per site in a multivariate analysis with region, inland/coast, wrack and the region-by-inland/coast interactions as independent variables using the command manyglm (package: mvabund, Wang et al. 2012) with a negative binomial error distribution. Prey communities were similarly modelled as the abundance of major groups in a multivariate analysis with manyglm between regions with wrack as an independent variable and a negative binomial error distribution, but these tests additionally included season (June and August) as independent variable. Finally, the proportional number of prey sequences (logit-transformed) of major groups were pooled for each species within site and season, and was modelled using adonis2 (package: vegan, Oksanen et al. 2019). To compare diet composition between spider species, we also compared gut contents while controlling for effects of region. To examine model assumptions, we used plot.manyglm and all tests showed no pattern in errors which confirm the model appropriateness. Significant relationships were further explored using anova with adjusted p-values, to identify which groups that explained the variation. In all these tests, prey communities and spider diets were included at the level of family or higher taxonomic unit and not at a species level.
To study prey diversity and diet consistency within and among species, we first calculated individual diets using the dynamic threshold model in Cirtwill and Hambäck (2021). We then compared species accumulation curves in spider guts using specaccum with spider individual as sampling unit (package: vegan, Oksanen et al. 2019), and then estimated diet consistency by calculating the Jaccard similarity index between diets of individual spiders’ prey species and prey families, first between pairs of all spider individuals and then between individual pairs of the same species. Diet similarity was compared between region, wrack, and their interaction, firstly, depending on if pairs included all spider individuals or where restricted to within species comparison, and secondly, depending on if diets were based on prey species or prey family. If the interaction terms did not contribute, models were re-fit without the interaction. We then tested for pairwise differences between region-wrack combinations using a Tukey’s HSD test applied to the analysis of variance of the above linear models, including the interaction term between region and wrack. All tests were performed using R 3.6.3 (R Core Team 2020).