Ecological plasticity governs ecosystem services in multilayer networks

The study addresses how ecosystem services (ES) in agro-ecosystems are governed by the structure and dynamics of multilayer ecological networks. Specifically, it examines whether regulation of the weed seedbank by carabid beetles is determined by their herbivory interactions and how this service co-varies with predation on gastropod molluscs, a potential alternative prey. The authors hypothesize that: (1) weed seed regulation is driven by the structure (species richness and interaction frequencies) of the carabid–weed seed herbivore layer; (2) weed seed regulation co-varies inversely with gastropod predation via changes in the carnivore layer; and (3) variation in service delivery is governed more by link turnover (rewiring) between herbivore and carnivore layers due to prey switching than by species turnover. Motivated by limited empirical demonstrations of network structural governance of ES, the study leverages a large, replicated field dataset to reveal how ecological plasticity and interlayer dynamics shape ES delivery and its management implications.

Prior work establishes that ecosystem functions arise from network structures and that managing species and their interactions can maintain functioning. Multilayer networks conceptually cluster interactions by function and predict that species switching among layers can affect service stability and balance. Classical studies link carabid abundance to weed seedbank changes, while molecular gut analyses show gastropods as key prey and that carabid predation alters slug distributions. Most ES studies are autecological, overlooking interdependencies among services. Approaches such as tripartite or coupled networks exist, but multilayer frameworks enable analysis of separate functional layers and shared species. The literature also indicates generalist predator diet adaptation in multi-prey contexts and that interaction rewiring is common, potentially enhancing network robustness. However, empirical datasets quantifying ES within network contexts remain scarce, motivating this study.

Design and data: The study uses data from the Farm Scale Evaluations (FSE) of genetically modified herbicide-tolerant (GMHT) crops across 187 UK fields in a half-field split design, yielding 374 half-fields in three spring-sown crops (beet, maize, oilseed rape) sampled once between 2000–2004. Biodiversity sampling employed standardized protocols with adequate statistical power demonstrated in previous analyses. Sampling: - Carabid beetles and gastropod molluscs were sampled via pitfall and baited refuge traps along transects at 2, 8, and 32 m from field edges during spring, summer, and late summer. Counts were summed to annual totals per species per half-field, then converted to relative abundances (species count divided by total group count).

• Weed seedbank was quantified pre-sowing (t0) and post-harvest (t1) by germinating bulked soil cores (8 samples per field) in greenhouse assays over ~18 weeks. Seed rain (viable seeds available at soil surface) was measured with traps along transects through the growing season. Network construction: For each half-field, two bipartite layers were constructed: herbivore layer (carabids–weed seeds) and carnivore layer (carabids–gastropods). Trophic links were compiled from literature. A link was realized when a documented consumer–resource pair co-occurred in the half-field. Generalization rules standardized trophic sampling: (i) within-genus generalization for carabids (a carabid species consumes resources known for congeners); (ii) within-genus generalization for resources (a carabid consuming one species of a resource genus is assumed to consume other congeneric resource species). These rules reduced isolated nodes and bias toward well-studied species. Interaction frequency: For each realized consumer–resource link, quantitative interaction frequency was estimated as the product of consumer and resource relative abundances, assuming abundance predicts interaction strength. Frequencies were computed separately for herbivore, carnivore, and omnivore roles in each network. Carabid trophic roles: In each half-field, carabid species were empirically assigned as herbivore (links only to weeds), carnivore (links only to gastropods), or omnivore (links to both), allowing species to occupy different roles across networks. Weed seedbank regulation metric: For total, monocot, and dicot seedbanks, regulation was calculated as ln((t1 + 0.5)/(t0 + 0.5)). Negative values indicate declines (i.e., regulation) from t0 to t1. Following prior work, negative relationships between regulation metric and carabid variables indicate regulation by carabids. Statistics: All analyses were in R using cheddar, bipartite, vegan, and HiveR. Generalized Linear Mixed-effects Models (GLMMs, Gaussian errors) tested relationships between: (i) weed seed regulation and carabid herbivore species richness or herbivory interaction frequency; (ii) numbers of carabid herbivores vs gastropod richness; (iii) numbers of links to weeds vs links to gastropods for omnivorous carabids; (iv) herbivory vs carnivory interaction frequencies (using LOESS on log(x+0.5) due to distribution). Random effects: field identity nested within crop type nested within management (conventional vs GMHT). Species and link turnover were quantified with Bray–Curtis dissimilarities across networks categorized along a herbivore/carnivore gradient, ensuring no network was double-counted by random assignment to either group. Model assumptions and residuals were checked for normality and behavior.

• Composite multilayer network: 41 carabid, 96 weed, and 9 gastropod species with 811 hypothesized pairwise links. Carabid roles across the composite: 17 obligate herbivores (only weeds), 6 obligate carnivores (only gastropods), 18 omnivores capable of switching.
• Weed seed regulation and herbivore layer support the first hypothesis:
  • Seedbank regulation increased with carabid species richness in the herbivore layer for total weeds (F1,328 = 4.9, p = 0.0275), monocots (F1,328 = 6.4, p = 0.0117), and dicots (F1,328 = 5.1, p = 0.0246).
  • Regulation increased with summed herbivory interaction frequency for total (F1,328 = 5.2, p = 0.023), monocots (F1,328 = 3.9, p = 0.049), and dicots (F1,328 = 5.7, p = 0.017), confirming the herbivore layer’s driving role.
• Inverse co-variation with carnivore layer supports the second hypothesis:
  • Carabid herbivore species richness decreased as gastropod species richness increased (F1,367 = 182.1, p ≤ 0.001).
  • Numbers of links to weeds vs gastropods co-varied inversely across networks (F1,367 = 212.8, p ≤ 0.001).
  • Herbivory interaction frequency declined sharply as carnivory interaction frequency increased (LOESS on log(x+0.5)).
  • Among omnivorous carabids, there was a strong trade-off: no omnivore was highly linked to both weeds and gastropods within a network (t = -7.08, p ≤ 0.001).
• Link turnover exceeds species turnover, supporting the third hypothesis:
  • Carabid species turnover (mean Bray–Curtis dissimilarity): 0.44 ± 0.17.
  • Link turnover between carabids and resources: 0.65 ± 0.25, indicating network structural changes are driven more by rewiring than species replacement.
• Ecological plasticity and key species: Common carabids (e.g., Pterostichus melanarius, P. madidus, P. niger, Harpalus rufipes, H. affinis) exhibited trophic role plasticity, acting as herbivores, carnivores, or omnivores depending on community context, marking them as key ES-governing species.
• Management implication: Weed seed regulation co-varies inversely with gastropod predation due to prey switching; focusing solely on weed seed consumers is unlikely to stabilize service delivery.

Findings demonstrate that weed seedbank regulation is governed by the structure and quantitative interaction frequencies of the herbivore layer, but this service co-varies inversely with gastropod predation in the carnivore layer due to prey switching by behaviorally plastic carabids. The network-wide inverse trade-offs in links and interaction frequencies, together with stronger link turnover than species turnover, confirm that ecological plasticity and link rewiring across multilayer networks drive ES outcomes. This addresses the research question by showing that interlayer dynamics and flexible trophic roles, rather than static species lists, determine service delivery. The results emphasize the need to manage both species and their interactions to sustain ES. For agriculture, integrated management that enhances specialist carabid–weed seed interactions while addressing gastropod availability (e.g., targeted mollusc control) may stabilize weed regulation. More broadly, this supports moving beyond autecological approaches to consider multilayer network structure when designing sustainable management for multiple, potentially coupled, services.

This study provides empirical evidence that ecological plasticity and link rewiring in multilayer networks govern ecosystem service delivery in agro-ecosystems. Weed seed regulation increases with herbivory interaction frequency and species richness, but declines as carnivory interactions increase due to prey switching, revealing an interlayer trade-off. Link turnover surpasses species turnover in explaining variation, highlighting the primacy of interaction structure over composition. Management should target both nodes and links to promote stable ES, integrating weed and pest control strategies. Future research should: (i) experimentally manipulate herbicide and molluscicide regimes to test causal effects on network structure and ES; (ii) employ molecular gut-content analyses to refine trophic links and quantify diet plasticity; (iii) build higher-resolution, empirically validated networks to assess robustness and optimize management interventions.

• Trophic interaction data were compiled from literature and generalized at genus level for both consumers and resources, potentially introducing under- or over-estimation of specific interactions and reducing taxonomic resolution.
• Many prey species lacked documented trophic links and were excluded (notably 109 weed species and 3 gastropod species), which may bias network structure and interaction frequency estimates.
• Interaction frequencies were approximated as products of relative abundances, assuming abundance-driven encounter rates; trait-mediated interaction strengths (e.g., body size, mandible morphology) were not incorporated.
• Despite strong replication, the observational design limits causal inference; co-variation may be influenced by unmeasured environmental or management factors beyond those modeled as random effects.
• Carabid trophic roles were assigned per network based on realized links; detection limits and literature gaps may misclassify some roles.