Experimental design and data collection
The data came from a highly replicated field trial (The Farm Scale Evaluations, FSE) designed to test whether the adoption of genetically modified, herbicide-tolerant (GMHT) crops would have led to significant changes in UK farmland biodiversity by comparison with the conventional crops and herbicide management then used23. The FSE sampled the biodiversity in and around 187 spring sown crop fields across the UK using a standard methodology (Supplementary Fig. S5). Post-hoc analyses demonstrated that the trial design had more than sufficient statistical power to test the null hypothesis of no change in biodiversity31 and indicated that the species nodes were fully sampled32,33. Full details of the experimental design and protocols for data collection can be found in Champion et al. 34 and Bohan et al. 35, which we briefly detail here.
The count data for the carabids, weed seedbank and seed rain, and gastropod molluscs came from 66 spring-sown beet, 55 spring maize and 66 spring oilseed rape fields. The fields were distributed across the UK (Supplementary Fig. S5) and each field was sampled for one cropping year23 between 2000 and 2004. The great majority of fields were below 15 Ha in size, and typically varied between 2 and 10 Ha depending upon the crop34,35. The trial used a half-field design, and each field was divided in two with one half sown with a conventional crop and the other a GMHT variety of the same crop. Data from both treatments are used for the analyses presented in this study, giving a total of 374 half-fields.
The pitfall trapping of soil-surface-active invertebrates employed the method described by Brooks et al. 36. Pitfall traps, spaced at 2, 8 and 32 m points along 4 transects running from the field-edge into the cropped area, were opened in the spring (April ⁄ May) and summer (June ⁄ July) and in late summer (August). The weed seedbank was measured by taking 8 soil samples in each field, at 2 and 32 m points on four transects, prior to the crops being sown and after harvest37. The soil samples in each half-field were bulked, placed in germination trays in a greenhouse and the seeds they contained allowed to germinate over the course of 18 weeks37,38. The viable weed seeds available to the carabids were measured as the seed rain onto the soil surface from the weed plants in the field. Eight seed rain traps, placed at 2 and 32 m along 4 of the transects, sampled the rain of weed seed throughout the growing season37. Gastropods were sampled using baited refuge traps at the same positions used for the pitfall trapping in late April and in early August for spring oilseed rape, and in May and August for maize and beet36. All carabids, molluscs and weed seeds sampled were identified as species and counted. For the carabids and molluscs, counts were then pooled, by summation, to give a year-total estimate for each species in each half-field, from which a relative abundance of each species was calculated by dividing the count of that species by the total count for the group of carabids and molluscs, respectively. Total, monocotyledon and dicotyledon weed species counts were pooled, by summation, to give an estimate of these weed seedbanks before (t0) and after (t1) the crop was sown and harvested, respectively.
Network construction
The species sample data were supplemented with carabid dietary information recovered from the literature. We assumed that where a carabid species, A, was noted to consume a resource species, B, in the literature and both these species were present in a half-field, then this trophic interaction was realised6,39,40,41. To standardise the (trophic interaction) sampling effort across all carabid species and reflect the generalist nature of carabid consumers it was assumed, following Honek et al. 42, that each carabid species would consume the same resources as other carabid species within the same genus39,41. A similar generalisation was made at the resource level: where a particular species of carabid was recorded to feed upon one species of gastropod or weed in the literature, we assumed that this carabid would also consume other resource species of the same genus43. This assumption of generalisation was done to reduce the numbers of artefactually isolated species within each network, and to avoid the bias towards more highly-studied species44,45.
The interaction frequency for each realised link between consumer and resource was calculated as the multiplicative product of the consumer and resource relative abundances, under the assumption that species abundance is a predictor of the strength of interaction between species46. Applying a frequency weighting to the links in this manner integrates a quantitative estimate of the interaction strength between species in each network, enriching the simple binary network structure built from presence/absence data. Future consideration of carabid interaction strength prey may include traits, such as body size and mandible type47. Following construction of the carabid-weed seed (herbivore) and carabid-gastropod (carnivore) layers, each carabid species was assigned to an empirical trophic group based upon their role in each replicate multilayer network. Carabid nodes linked only to gastropods were assigned to the ‘carnivore’ grouping, while those consuming only weeds were ‘herbivores’, and ‘omnivores’ were generalist species linked to both gastropods and weeds. Thus, a particular carabid species might be a carnivore, an herbivore or an omnivore in different replicate multilayer networks. Some species may therefore appear in different classifications across the collection of networks; a particular carabid species may be a weed consuming herbivore in some networks, a gastropod carnivore in others and an omnivore in yet other networks.
Regulation of the weeds in the seedbank
Regulation was calculated from the change in total, monocotyledon and dicotyledon seedbank counts between t0 and t1, as:
$${rm{regulation}} = {mathrm{ln}}left( {frac{{t_1 + 0.5}}{{t_0 + 0.5}}} right)$$
(1)
so that for each half-field three metrics of regulation for the total, dicotyledon and monocotyledon seedbanks were calculated. Negative values of this metric indicate a decline in the size of the weed seedbank from t0 to t1. Following Bohan et al. 14, negative relationships between the metric of regulation and carabid counts are indicative of seedbank regulation by these beetles.
Statistics and reproducibility
All analysis was done in R (R Core Team 2019) using the cheddar48, bipartite49 and vegan50 packages. Network plots were created with the HiveR package51. Interaction frequencies were calculated separately for herbivores, carnivores and omnivores in each network.
Species and link turnover across the collection of multilayer networks were measured using Bray-Curtis dissimilarity in the vegan package50. Each network layer is a realisation of the interactions drawn from the composite network52, with the interactions only being contingent on local species composition and abundances. To assess how species and link turnover changes across the herbivore/carnivore gradient, we used the number of herbivores and carnivores within each network as factor levels with which to categorise the networks (i.e. networks with 1 herbivore, 2 herbivores, 3 herbivores, etc.). We ensured that no single network appeared in more than one group by randomly assigning networks to either their herbivore or carnivore grouping, and then calculated the Bray-Curtis dissimilarity between the herbivore and carnivore groups.
The regression modelling between species and link variables was done using Generalised Linear Mixed-effects Models (GLMM) with Gaussian errors. Each field was treated as a replicate as there was no repeat sampling from any one site23. All analyses were appropriately subjected to tests of normality and behaviour of the model residuals53.
Species richness
GLMMs were used to assess the effect that increasing numbers of gastropod species has on the number of weed species, and the number of herbivorous carabids implicated in each multilayer network. Field identity was nested within crop type (spring-sown beet, spring maize or spring oilseed rape), which was nested within management type (conventional or GMHT) and included as a random effect. The species richness of all groups was log(x+0.5) transformed to conform to normality. The relationship between the number of carabid species acting as herbivores and carnivores in each network was assessed using the same model structure.
Link richness
The relationship between the number of links to plant resources and links to gastropod resources for the omnivorous carabid nodes was assessed using GLMMs, with field identity nested within crop type (spring-sown beet, spring maize or spring oilseed rape), which was nested within management type (conventional or GMHT) and included as a random effect.
Link frequency
Due to the extreme distribution pattern of the carnivore and herbivore predation interaction frequencies, the correlation between these two variables was examined on the log(x + 0.5) scale using LOESS smoothing53.
Weed seed regulation
The relationship between weed seed regulation and the count of herbivores or herbivory interaction frequency was modelled with GLMMs. The count of herbivores and herbivore interaction frequency were log(x+0.5) transformed to attain normality. Field identity was nested within crop type (spring-sown beet, spring maize or spring oilseed rape), which was nested within management type (conventional or GMHT) and included as a random effect.
Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article.
Source: Ecology - nature.com
