Arthropod predator nutrient content changes with crop sowing period with implications for biocontrol

Methods Fieldwork and identification Arthropods were collected from Cockle Park (Morpeth, UK; 55°13'00.8" N, 1°41'28.7" W) in July 2024. Transects were established in a field containing adjacent plots of spring-sown and winter-sown wheat. Six pairs of belt transects were established 20 m apart within each crop with 4 m between adjacent transects in the spring and winter sown crops. One pitfall trap for each pair of transects (i.e., three per sowing period) was filled with soapy water and deployed for 72 h. Pitfall trapping was conducted once nine days prior to individual arthropod collection (to represent past prey availability) and once in parallel with individual arthropod collection (to represent present prey availability). Invertebrates within these traps were filtered in-field using an AeropressⓇ and stored in 80 % ethanol for subsequent morphological identification. Within each transect, hand-searching was conducted for approximately 10 minutes and individual arthropods collected using a pooter. Figure 1: Experimental design for invertebrate sampling. Paired sampling locations were established 4 m apart in directly adjacent plots of winter- and spring-sown wheat. Created in BioRender. Cuff, J. (2025) https://BioRender.com/6rd9udt Following collection, the arthropods were frozen onsite at -20 °C to humanely kill them and preserve their macronutrient contents for subsequent analysis. The samples were transported to the Molecular Diagnostics Facility at Newcastle University for morphological identification and nutritional analysis. All invertebrates were identified using a stereomicroscope and morphological keys (Barber, 2008; Dallimore & Shaw, 2013; Luff, 2007; Roberts, 2016; Tilling, 2014). Most invertebrates were identified to at least family level or finer resolution (e.g., genus and species), except where access to appropriate identification resources or damage to specimens limited identification. Individuals from each taxa were then enumerated to produce species by site abundance data. Macronutrient determination Macronutrient analysis followed the Macronutrient Extraction and Determination from Invertebrates (MEDI) method (Cuff et al., 2021; Cuff and Wilder, 2021) which streamlines sulfo-phospho-vanillin, anthrone and Lowry colorimetric assays for lipid, carbohydrate and protein estimation, respectively. To extract macronutrients from whole arthropods, individually collected samples were placed in 2 mL tubes of a 96-well rack and heated until dry (~ 1 h) in an oven at 60 °C. To each sample, 500 μl of 1:12 chloroform:methanol solution was added and left at room temperature for 24 h. After 24 h, 400 μl of the chloroform:methanol solution was removed and retained separately for lipid analysis. A further 500 μl of chloroform:methanol solution was added to the sample and incubated at room temperature for 24 h to remove any residual lipids. All chloroform:methanol solution was removed by pipetting, and any residue evaporated in an oven at 60 °C for ~ 15 min. To each dried sample, 500 μl 0.1 M NaOH was added, after which the samples were incubated at 80 °C for 30 min and at room temperature overnight (~ 16 h). The samples were briefly centrifuged (~ 2 min, 2,000 x g) to avoid taking solid material through to the subsequent assays. From each sample, 400 μl of the supernatant was transferred into a separate tube for protein and carbohydrate determination. All assays were carried out in 384-well plate format. For each assay, a stock standard dilution series of a known concentration was made with lard oil, glucose and bovine serum albumin (BSA) for lipid, carbohydrate and protein quantification, respectively. Stock solutions of 2 mg ml-1 glucose and BSA were made using distilled water and subsequently diluted in 0.1 M NaOH, whereas 2 mg ml-1 lard oil was made and diluted with 1:12 chloroform:methanol. Dilution series comprised 0-1 mg ml-1 in nine increments (0, 12.5, 62.5, 125, 250, 375, 500, 750, 1000 μg ml-1). A quarter of each 384-well assay plate (96 wells) was reserved for repeats of these standards. Lipid quantification The sulfo-phospho-vanillin method was used to determine lipid content. The vanillin reagent was prepared by mixing 0.108 mg of vanillin with 18 μl of hot water and 72 μl 85 % phosphoric acid for every 100 μl reagent. From each sample, three repeats of 20 μl were added to a 384-well plate alongside standards and heated to 100 °C for 5 min to evaporate the chloroform:methanol carrier. To each well, 10 μl of concentrated sulfuric acid was added, mixed briefly on a plate shaker and incubated at room temperature for 15 min. Following this, 75 μl of vanillin reagent was added to each well, mixed briefly on a plate shaker and incubated at room temperature for 10 min. Absorbance at 490 nm for each was determined in a spectrophotometer. Carbohydrate quantification The anthrone method was used to determine carbohydrate content. The anthrone reagent was prepared by mixing 100 μg of anthrone with 100 μl of concentrated sulfuric acid for every 100 μl of reagent. From each sample, three repeats of 20 μl were added to a 384-well plate alongside standards, mixed with 80 μl of the anthrone reagent and incubated at 92 °C for 10 min. The plate was incubated at room temperature for 5 min and absorbance at 620 nm for each well was determined in a spectrophotometer. Protein quantification The Lowry method was used to determine protein content. From each sample, three repeats of 20 μl were added to a 384-well plate alongside standards, and 60 μl of modified Lowry reagent added to each well and mixed briefly on a plate shaker. The plate was incubated at room temperature for 10 min, and 6 μl of 1X Folin-Ciocalteu reagent added and mixed briefly on a plate shaker. The plate was incubated at room temperature for 30 min and absorbance at 750 nm for each well was determined in a spectrophotometer. Statistical analysis All analysis was conducted in R version 4.5.2 (R Core Team, 2025) and all data processing used the ‘tidyverse’ package for reproducibility (Wickham et al., 2019). Data visualisations were generated using ggplot2 (Wickham, 2016). The pitfall trapped invertebrate communities (herein considered representative of prey available to the hand-collected arthropod predators) were compared between winter and spring sown crops, pitfall trapping rounds and an interaction between the two using multivariate generalized linear models (MGLMs) with a Poisson error family and Monte Carlo resampling in the ‘mvabund’ package via the ‘manyglm’ function (Wang et al., 2012). These differences were visualised using non-metric multidimensional scaling with a Bray-Curtis dissimilarity matrix in the ’vegan’ package (Oksanen et al., 2022). Macronutrient contents of arthropods were compared between taxa, winter- and spring-sown crops and the interaction between the two using multivariate linear models in the ‘mvabund’ package via the ‘manylm’ function (Wang et al., 2012). Null network models were run to infer density-dependent interactions for each of the individually-collected arthropod predators based on prey availability and determine if prey availability explains differences in predator nutrient content. Using pitfall trapping prey availability data, separately for the past and present prey availability based on the two rounds of pitfall trapping, null networks were generated for each of the individually-collected predators included in the nutritional analyses above. These null network models assume that the trophic interactions of the predators were density-dependent (i.e., they interacted with the prey most available to them), providing a semi-realistic estimate of how prey availability would translate to trophic interactions. To construct null networks, null diets were simulated using the ‘generate_null_net' function in the ‘econullnetr’ package (Vaughan et al., 2018) using the prey availability data from the pitfall traps with random placeholder interaction data. To generate null networks, we required dummy interaction data. We simulated these dummy interactions for each individual arthropod predator by randomly allocating interactions with three of the available prey taxa (approximately reflecting the interaction richness of arable arthropod predators from previous work; Cuff et al., 2022). These dummy data were only required to run null network simulations and did not otherwise feature in any downstream analyses, with the exception of the fact that the node degree is invariable (i.e., a mean of three interactions per predator across all simulations). Using the ‘generate_null_net_indiv’ function (Cuff et al., 2023), just the simulation data were extracted irrespective of the identities of the random interaction data input. Null networks were visually constructed using ‘igraph’ (Csardi & Nepusz, 2006) and plotted using ‘ggnetwork’ (Briatte, 2021). Null network-inferred predator-prey interactions based on both past and present prey availability were compared against predator nutrient content using redundancy analysis (RDA) and model outputs analysed using analysis of variance (ANOVA) of the overall model, terms within the model and model axes.