Trophic links between functional groups of arable plants and beetles are stable at a national scale


Correspondence author: E-mail address:


1. There is an urgent need to accurately model how environmental change affects the wide-scale functioning of ecosystems, but advances are hindered by a lack of knowledge of how trophic levels are linked across space. It is unclear which theoretical approach to take to improve modelling of such interactions, but evidence is gathering that linking species responses to their functional traits can increase understanding of ecosystem dynamics. Currently, there are no quantitative studies testing how this approach might improve models of multiple, trophically interacting species, at wide spatial scales.

2. Arable weeds play a foundational role in linking food webs, providing resources for many taxa, including carabid beetles that feed on their seeds and weed-associated invertebrate prey. Here, we model associations between weeds and carabids across farmland in Great Britain (GB), to test the hypothesis that wide-scale trophic links between these groups are structured by their species functional traits.

3. A network of c. 250 arable fields, covering four crops and most lowland areas of GB, was sampled for weed, carabid and invertebrate taxa over 3 years. Data sets of these groups were closely matched in time and space, and each contained numerous species with a range of eco-physiological traits. The consistency of trophic linkages between multiple taxa sharing functional traits was tested within multivariate and log-linear models.

4. Robust links were established between the functional traits of taxa and their trophic interactions. Autumn-germinating, small-seeded weeds were associated with smaller, spring-breeding carabids, more specialised in seed feeding, whereas spring-germinating, large-seeded weeds were associated with a range of larger, autumn-breeding omnivorous carabids. These relationships were strong and dynamic, being independent of changes in invertebrate food resources and consistent across sample dates, crops and regions of GB.

5. We conclude that, in at least one system of interacting taxa, functional traits can be used to predict consistent, wide-scale trophic links. This conceptual approach is useful for assessing how perturbations affecting lower trophic levels are ramified throughout ecosystems and could be used to assess how environmental change affects a wider range of secondary consumers.


Global biodiversity declines are being driven by rapid, anthropogenically mediated change (Chapin et al. 2000). Although there has been a bias towards considering biodiversity trends alone, models describing effects on ecosystem function are now regarded as indispensible precursors to understanding and managing the consequences of such change (Purvis & Hector 2000; Kremen 2005). However, the development of such models presents considerable challenges (Lavorel et al. 1997). For example, modelling of ecosystem dynamics is critically impeded by a lack of knowledge of how trophic levels are linked in ecosystems (Tylianakis et al. 2008). A number of theoretical approaches to modelling such interactions have emerged, ranging from mechanistic modelling of pairwise species interactions (May 1974) and consumer resource models (Holt, Grover & Tilman 1994) through to complex assessments of food webs (McCann 2007). Opinion is, however, divided about which approach is best, or indeed whether any of them can contain the complexity of interactions in multi-species systems (Simberloff 2004). One way of improving this theoretical framework, that is, increasingly gaining favour, is the incorporation of taxa functional traits into models (De Bello et al. 2010). For example, carabids can have trait-mediated responses to landscape composition (Vandewalle et al. 2010), land disturbance and stress induced by habitat changes (Ribera et al. 2001), and flooding intensity (Lambeets et al. 2008). This approach provides a tractable route to modelling complex interactions because species traits are usually linked to the functions that they perform within ecosystems. Furthermore, these functions often vary in their importance, both for the viability of an ecosystem or the services that it delivers (Zavaleta & Hulvey 2004).

Lavorel & Garnier (2002) described the hypothesis that plant functional traits will determine the response of plant communities to the environment and, in turn, affect ecosystem processes. This concept can be extended to analyse the effect of plant traits on the responses of other functionally related taxa, predisposed to such links by their own traits. Evidence is growing that this hypothesis is a powerful tool (Kremen 2005), because when there is concordance between these ‘effect’ and ‘response’ traits modelling their relationship is a tractable way of understanding profound changes in ecosystem function (Larsen, Williams & Kremen 2005; Petchey & Gaston 2006; Schmitz 2008). This hypothesis has not, however, been extended to investigations of trophic linkages at large spatial scales. This is necessary to test the wider validity of this approach (Kremen et al. 2007), not least because quantifying geographical variation in ecological functioning is crucial to understanding how environmental change will alter biodiversity (Symstad et al. 2003).

The Farm Scale Evaluations (FSE) tested the effect of managing genetically modified crops on numerous groups of plants and invertebrates across most arable areas of Great Britain (GB). This study represents the largest field ecology experiment to date, providing paired data sets of arable plants and invertebrates, under contrasting crop management regimes, at an unprecedented spatial scale (Perry et al. 2003). These data therefore offer a unique opportunity to test different theoretical approaches to modelling links between trophic levels, thereby assessing the scope and universality of their application. In particular, the FSE generated data from conventionally managed half-field plots linking 236 species of arable weeds with 123 species of carabid beetles, covering 521 618 and 618 384 individuals of these groups, respectively. These plots accurately reflected the distribution and commercial production of four arable crops (Perry et al. 2003). These data provide an ideal model system for testing the coherence of a trait-based approach for investigating how trophic levels interact. For example, responses of weeds to agricultural perturbations are strongly linked to their functional traits (Storkey, Moss & Cussans 2010). Similarly, carabids have trait-mediated responses to successional stages of plant communities in undisturbed habitats (Gobbi et al. 2010) and have facultative responses to feeding on weeds, including a preference for granivory in some species (Slansky & Panizzi 1986). For simplicity, throughout this paper, we describe any non-crop plants occurring in cultivated fields as weeds, but it should be noted that although all these species are important components of agro-ecosystem biodiversity, only a few are likely to cause significant reductions to crop yields.

Here, we use the FSE data to test the hypothesis described by Lavorel & Garnier (2002) within the context of two interacting trophic levels. To do this, we test the ability of arable weeds to respond to agricultural perturbations according to their traits and, in turn, affect a higher trophic level via links between these traits and the traits of carabid beetles. Within this system, the weed traits act as both ‘response’ and ‘effect’ traits, as defined by the hypothesis, by ‘responding’ to changes in the environment and then ‘effecting’ carabids. Similarly, the carabid traits act as the ‘response’ traits of a functionally related group, in our case by mediating their trophic responses to changing weed assemblages. We do this by grouping the numerous weed and carabid species available by their functional traits to test the following predictions: (i) weed taxa grouped by their traits will have similar functional responses across crops and regions of GB; (ii) carabid traits can be used to link patterns of their community variation to such wide-scale weed responses; (iii) groups of carabids expected, a priori, to have little response to weeds, owing to carnivorous feeding traits, will not have such links and (iv) groups of omnivorous carabids will have robust functional links with weeds, regardless of fluctuations in their invertebrate prey.

Materials and methods

Experimental Design

Data were derived from the FSE project, which surveyed three spring crops between 2000 and 2002 and one winter crop between 2000 and 2003. Here, we provide a synoptic overview of the experiment; see Perry et al. (2003) and Firbank et al. (2003) for detailed descriptions of its design and methodology, respectively. The experiment used blocks of whole fields, with treatments allocated to half-field plots. Here, we use data from 249 conventionally managed half-field plots, which had an average size of 5·37 ha (SE = 0·193) and comprised of 65 beet, 58 maize, 67 spring oilseed rape and 59 winter oilseed rape plots. Half-field totals were used to model the wide-scale responses of taxa because variation was much higher between plots than within plots.

Weed Explanatory Variables

All weed species were counted in a 0·25 m × 0·50 m quadrat at points 2, 4, 8, 16 and 32 m from field edges, along 12 transects in each half-field, during surveys conducted in April–May, June–July and August for spring crops, and February–March, April–May and June–July for winter oilseed rape; see Firbank et al. 2003; for schematics of the sampling positions. Similarly, seed rain was sampled continuously from first anthesis, using eight mesh traps of 0·1 m diameter, sunk into the ground and divided between locations at 2 m and 32 m on four transects. Common UK weeds have been screened for a range of eco-physiological traits and seven emergent functional groups identified on the basis of life-history traits, seed mass and maximum height (Storkey 2006). These groups and weed diversity measures were used to form explanatory variables for carabid responses (Tables 1 and S1, Supporting information). For emerged weeds and seed rain, partial principal component analyses (pPCA) were also conducted to assess community variation, using the canoco V4.5 program (Ter Braak & Šmilauer 2002). Smaller-seeded, autumn-germinating species (Group W3 and most species with these traits in Groups W2/W4) were separated from larger-seeded, obligate spring-germinating species (Group W5) (Figs 1 and S1, Supporting information). The sample scores on the first two pPCA axes were used to form additional variables describing these distinct communities (Table S1, Supporting information).

Table 1.   Functional groups of weed taxa included in multivariate analyses. The summed abundance across each group is shown for the listed species, together with their percentage coverage of the wider count of all taxa in the respective group. Within groups, taxa are listed by rank abundance. Their total counts fall within the ranges: A = 35 001–c.111 500; B = 3001–35 000; C = 1001–3000; D = 501–1000 and E ≤ 501, with letters indicating emerged weed abundance given first and seed rain second
Functional group/speciesAbundanceEcological traits
Total%GerminationSeed mass: mean (range) in mmFirst flowering
Group W1
Galium aparine (C/D), Fumaria officinalis (D/–), Anisantha sterilis (E/D), Avena fatua (E/D), Galeopsis tetrahit (E/–)493293·3Autumn7·26 (8·11–12·00)May–June
Group W2
Sonchus asper (B/A) (S. asp), Tripleurospermum inodorum (C/B) (T. ino), Papaver rhoeas (E/B), S. oleraceus (E/C), S. arvensis (E/D)101 43798·6Autumn0·34 (0·20–0·49)April–June
Alopecurus myosuroides (B/B)   1·99 
Group W3
Capsella bursa-pastoris (B/A) (C. b-p), Poa annua (A/A) (P. ann), Stellaria media (B/A) (S. med), Senecio vulgaris (C/B) (S. vul), Veronica persica (B/C) (V. per), Lamium purpureum (C/D) (L. pur)260 74299·7Autumn0·45 (0·10–0·96)March–April
Group W4
Matricaria discoidea (C/B) (M. dis), Viola arvensis (B/B) (V. arv), Solanum nigrum (C/E), Myosotis arvensis (D/C) (M. arv), Urtica urens (D/D) (U. ure), M. recutita (E/E), Spergula arvensis (E/E)64 23997·6Autumn0·38 (0·04–0·70)April–June
Geranium dissectum (C/B), Aethusa cynapium (C/C), G. molle (C/–)   1·65 (1·09–2·28) 
Group W5
Chenopodium album (B/B) (C. alb), Sisymbrium officinale (E/B) (S. off),70 91298·9Spring0·45 (0·29–0·60)May–July
Polygonum aviculare (B/B) (P. avi), Persicaria maculosa (C/B) (P. mac), P. lapathifolia (C/B) (P. lap), Sinapis arvensis (C/C) (S. arv), Atriplex patula (C/D) (A. pat), Fallopia convolvulus (C/C) (F. con)   2·54 (1·30–6·70) 
Group W6
Anagallis arvensis (C/E)154785·2Spring0·50June–July
Euphorbia helioscopia (E/E)   2·31 
Group W7
Cerastium fontanum (E/C), P. trivialis (C/–), Holcus lanatus (–/C), H. mollis (–/C), Trifolium repens (E/E), U. dioica (E/E), Taraxacum officinale agg. (E/E), Equisetum arvense (E/–), G. pusillum (E/E), Dactylis glomerata (E/–)17 80975·6Perennial0·44 (0·10–0·90)June–July
Elytrigia repens (C/E), Cirsium arvense (D/D), Ranunculus repens (D/–), Rumex crispus (–/D), C. vulgare (E/D), Lolium perenne (D/–), R. obtusifolius (E/E)   1·96 (1·30–2·90) 
Figure 1.

 Partial principal component analysis (pPCA) of emerged weeds, across all crops in the summer, showing species with the strongest fit to the first two axes, after correcting for year, geographical and crop effects. Figures in brackets indicate weed functional groups (see Table 1 for groups and species abbreviations). Seed rain species across all crops and emerged weeds in other data sets have similar ordination patterns (see Fig. S1, Supporting information).

Invertebrate Prey Explanatory Variables

All carabids can prey on a wide-range of invertebrates (Sunderland 2002). Therefore, to assess the independence of carabid responses to weeds, we also modelled associations with the majority of their prey, using wide-ranging FSE invertebrate data. Variables were constructed for Collembola, Parasitica, spiders, aphids, Auchenorrhyncha, Heteroptera, Staphylinidae, Diptera and lepidopteron larvae, sampled by combinations of pitfall trapping, Vortis suction sampling or direct counts. Variables for counts of earthworm casts in quadrats and gastropods in baited refuge traps, consisting of large, inverted plastic plant-pot saucers, were also included (see Supporting Information, Table S1 for full descriptions of prey variables and sampling methodology).

Carabid Response Variables

Carabid variables consisted of total counts from 12 pitfall traps, located on four of the transects used to sample weeds, at 2, 8 and 32 m from the crop edge. Traps were run for 2 week-periods, on three separate occasions, divided between May, July and August for spring crops and late September to October, late April to May and late June to early July for winter oilseed rape. Data across all crops were divided into sets covering one round of sampling in spring (using April–May traps) and one in summer (using June–July traps). Half-fields were included in analyses when at least 80% of traps were present and comprehensive data for explanatory variables were available, with totals adjusted to account for any missing traps. Variables describing the functional similarity of carabids were formed prior to the analysis by grouping species according to their traits, using relevant literature (Table 2). Taxa were classified as carnivores if only limited evidence of ingestion of plant material could be found, as omnivores with a mixed diet if feeding on seeds and invertebrates was recorded, and as omnivores with a preference for granivory, if a predilection for seeds over invertebrates was demonstrated (see, e.g., Tooley & Brust 2002; Honek et al. 2007). Carnivores were included to avoid any erroneous assumptions about functional responses of the whole carabid community. Total abundance of functional groups was modelled using all species. Very rare species, with >75% of their data comprised of zeros, were dropped from other analyses, leaving species accounting for at least 97% of the total abundance within each group (Table 2).

Table 2.   Functional groups of carabid taxa included in multivariate analyses: (C1) small, spring-breeding carnivores; (C2) autumn-breeding carnivores; omnivores with a mixed (M) diet of seeds and invertebrate prey, divided between; (C3) smaller, spring breeders and (C4) larger, autumn breeders; omnivores with preferences for granivory (G), divided between; (C5) smaller, spring breeders and (C6) larger, autumn breeders. The summed abundance across each group is shown for the listed species, together with their percentage coverage of the wider count of all taxa in the respective group. Within groups, taxa are listed by rank abundance, where total counts fall within the ranges: A = 30 001–c. 347 000; B = 10 001–30 000; C = 1001–10 000; D = 501–1000 and E ≤ 501
Functional group/speciesAbundanceEcological traits
Total%Feeding guildSize: mean (range) in mmBreeding period
Group C1
 Bembidion lampros (B), B. tetracolum (B), Anchomenus dorsalis (C), Notiophilus biguttatus (C), B. quadrimaculatum (C), B. obtusum (C), Loricera pilicornis (C), B. guttula (E), B. aeneum (E), Pterostichus strenuus (E), Demetrias atricapillus (E), B. lunulatum (E), Clivina fossor (E), B. properans (E), Stomis pumicatus (E), Asaphidion flavipes (E), P. nigrita (E), N. quadripunctatus (E)49 20399·1Carnivore5·29 (3·00–10·50)Spring
Group C2
 Nebria brevicollis (B), Trechus quadristriatus (C), N. salina (C), Leistus spinibarbis (E), Synuchus vivalis (E), Carabus violaceus (E), Patrobus atrorufus (E), Abax parallelepipedus (E)30 39699·4Carnivore12·29 (3·85–25·00)Autumn
Group C3
 Poecilus cupreus (B) (P. cup), Agonum muelleri (D) (A. meu)13 18399·9Omnivore-M10·00 (8·00–12·00)Spring
Group C4
 Pterostichus melanarius (A) (P. mel), P. madidus (A), P. niger (B) (P. nig), Calathus fuscipes (C) (C. fus), C. melanocephalus (C) (C. mel), C. rotundicollis (E) (C. rot)491 72799·9Omnivore-M13·04 (7·25–18·50)Autumn
Group C5
 Amara similata (C) (A. sim), A. ovata (D) (A. ova), A. aenea (E) (A. aen), A. plebeja (E) (A. ple), A. familiaris (E) (A. fam)532397·4Omnivore-G7·69 (6·40–8·75)Spring
Group C6
 Harpalus rufipes (B) (H. ruf), Curtonotus aulicus (E) (C. aul), Ophonus spp. (E) (Oph)28 57999·9Omnivore-G11·83 (9·50–13·50)Autumn
 H. affinis (C) (H. aff)10·50Spring/Autumn

Links Between Functional Groups

Carabids responded strongly to geographical gradients and different crops in the FSE (Brooks et al. 2008). To robustly model the wide-scale consistency of carabid trophic responses, we therefore included a baseline component in all models to describe significant spatial, crop and survey year effects. To assess the unified responses of species within functional groups and thereby provide evidence that links are consistent across agglomerations of taxa, symmetric matrices of the distances between half-fields were formed for each of the groups listed in Tables 1 and 2, using counts of their constituent species. Additionally, matrices were calculated to account for geographical distances between half-fields, refined by truncation to their furthest inter-node distance in a Delaunay triangulation network (Borcard & Legendre 2002). Presence–absence matrices were also used to correct for crop and sampling year, together with matrices based on total weed, seed and invertebrate counts. Multiple regressions were performed on these distance matrices according to the methods described by Legendre, Lapointe & Casgrain (1994). This method uses an extension of the Mantel test to compute partial correlation coefficients between multiple matrices and test their significance by repeatedly permuting the rows and columns in the dependent matrix. Separate models were developed for each carabid group, by fitting significant year, crop and geographical matrices. A stepwise procedure was then used to select matrices describing all weed and invertebrate groups with significant additive effects. Matrices were fitted by forward selection and dropped by backwards elimination, according to a predefined Bonferroni corrected P-value of 0·05, using a permutation test of their multiple coefficient of determination. Regressions were developed using matrices calculated with four commonly used distance coefficients (Legendre & Gallagher 2001) and the most parsimonious method used to parameterise final models.

The results of the multiple matrix comparisons were corroborated by modelling the links between weeds, invertebrates and carabids using total abundances of their groups. This was done with log-linear regressions using a generalised linear model (GLM) assuming a Poisson distribution for each response and specifying the logarithmic link function to relate the mean response to the linear predictor (McCullagh & Nelder 1989). Multiple regression models were developed for each carabid group by stepwise selection of year, crop and geographical variables to form a baseline component, and additive effects of weed and invertebrate variables were tested after further refinement by stepwise selection; see Table S2 (Supporting information) for a full description of this methodology.

Links Between Taxa

To test the validity of models connecting functional groups, patterns in the trophic relationships of individual taxa were investigated, without prior expectations about how species traits and responses might be linked. Links between carabid communities and multiple weed and invertebrate explanatory variables (Table S1, Supporting information) were modelled with partial redundancy analyses (pRDA) (Ter Braak 1995). Final models were simplified by forward selection of the explanatory variables and assessed by Monte Carlo permutation tests of canonical axes and retained variables. First, responses of carnivores and omnivores were modelled to investigate how the combined effects of weeds and invertebrates structured the whole carabid community. Second, models were reduced to omnivorous carabids and weed variables. To further investigate trends in pRDAs and their consistency across species, logistic regression modelling was used to test links between taxa traits and their responses to the trophic resource gradients represented by lower canonical axes. This was done by regressing the proportional abundances between individual carabid species and counts of all carabids within samples (half-fields) against correlations of those samples with axes in the pRDA, with an additional interaction term between categorical descriptions of the functional groups of these species and the axes fitted to test whether carabid traits explained significant contrasts in their responses to trophic gradients. Additionally, univariate responses of each carabid species were investigated by the methodology used to model total counts of functional groups. To assess whether any trophic links were because of similarities in the phenology of taxa, splined abundance data of weeds and carabids were plotted against time across the season.


Responses of the Whole Carabid Community

When seeds were unavailable in the early part of the season, all carabids were predominantly associated with invertebrate prey in matrix and log-linear regression analyses (Tables 3 and S2a, Supporting information). Comparisons of distance matrices, derived from multiple taxa, demonstrated the importance of collembolan prey for all carabids at this time (Table 3). Isotomidae and, to a lesser extent, Entomobryidae were particularly important, accounting for over two-thirds of species responses to collembolan families in the spring (Table S3, Supporting information). Forward selection of variables in a pRDA confirmed the importance of invertebrate prey, especially Collembola, for the whole carabid community at this time (Fig. S2, Supporting information).

Table 3.   Multiple matrix regression models comparing half-field differences of: species within carabid functional groups (dependent matrix); crops (CR) and geographical locations (GEO) of half-fields (independent baseline model matrices shown in italics), and species within emerged weed or seed rain functional groups, emerged weed and seed rain total counts, families of Collembola or counts of other invertebrate groups (independent, additional model matrices, shown in bold), where COL = Collembola, WOR = earthworms, SPI = spiders, TEW = total emerged weeds and weed seed rain functional groups, shown by W followed by respective group number. Independent parameters are shown with their standardised partial regression coefficients; number and letter suffixes, for baseline and additional terms, respectively, indicate their associated probabilities (P), tested by a permutation test (999 randomisations of the rows and columns of dependent matrices), under the null hypothesis of no concordance between matrices. Weed group terms entering final models were derived from seed rain matrices: those describing emerged weed counts of the same group were also significant when substituted in the models, shown by the respective suffix underlined
Carabid functional group/data setDistance metricFitted terms in final modelMatrix-based permutation test
Baseline model termsAdditional model termsFit of final model
t-value (d.f.)Pt-value (d.f.)PR2P
Carabid Group C1
 All crops/springBray-Curtis0·206•CR(1) + 0·223•GEO(2) + 0·096COL(a)(1) 34·659 (225)0·002(a) 16·041 (225)0·0180·1110·002
(2) 37·116 (225)0·002
 All crops/summerBray-Curtis0·151•CR(1) + 0·079•GEO(2) + 0·073COL(a)(1) 24·142 (224)0·002(a) 11·652 (224)0·0280·0370·002
(2) 12·640 (224)0·020
Carabid Group C2
 All crops/springBray-Curtis0·040•CR(1) + 0·115COL(a)(1) 6·320 (225)0·002(a) 18·398 (225)0·0100·0160·008
 All crops/summerEuclidean0·120•CR(1) + 0·127•GEO(2) + 0·058WOR(a)(1) 19·304 (224)0·002(a) 9·358 (224)0·0120·0360·002
(2) 20·486 (224)0·004
Carabid Group C4
 All crops/springBray-Curtis0·053•CR(1) + 0·106•GEO(2) + 0·122SPI(a) + 0·148COL(b)(1) 8·617 (225)0·002(a) 19·741 (225)0·0040·0630·002
(2) 17·279 (225)0·008(b) 23·788 (225)0·006
 All crops/summerBray-Curtis0·104•CR(1) + 0·301•GEO(2) + 0·106TEW(a) + 0·077W5(b)(1) 17·524 (224)0·002(a) 17·765 (224)0·0060·1340·002
(2) 50·941 (224)0·002(b) 12·967 (224)0·006
Carabid Group C5
 All crops/springBray-Curtis0·210•CR(1) + 0·137COL(a)(1) 34·529 (225)0·002(a) 22·760 (225)0·0020·0650·002
 All crops/summerEuclidean0·122•CR(1) + 0·263W3(a)(1) 20·042 (224)0·002(a) 43·354 (224)0·0020·0900·002
(a) 28·640 (224)0·002
Carabid Group C6
 All crops/springBray-Curtis0·143•GEO(1) + 0·113COL(a)(1) 22·918 (225)0·002(a) 18·157 (225)0·0100·0380·002
 All crops/summerBray-Curtis0·043•CR(1) + 0·226•GEO(2) + 0·136W5(a)(1) 6·977 (224)0·002(a) 22·193 (224)0·0020·0740·002
(2) 37·087 (224)0·002(a) 18·559 (224)0·004

With increasing weed seed availability in the summer, carabids diversified their trophic responses and distinct preferences emerged between functional groups. As expected, carnivorous carabids maintained their strong associations with invertebrate prey (Tables 3 and S2a, Supporting information). Abundant, mixed-feeding omnivores (Group C4) retained some associations with invertebrate prey, but weed resources became important for many of these species (Tables 3 and S3, Supporting information). Other omnivores shifted their preferences even more strongly from invertebrates to weeds in the summer (Tables 3, S2a and S3, Supporting information). Multivariate models highlighted these contrasting responses, confirming that invertebrates and weeds were important for structuring the whole community at this time, with Collembola and earthworms particularly important in the ordination of carnivorous species, and weed resources mediating the responses of most omnivores (Fig. S2, Supporting information).

Responses of Omnivorous Carabids

Omnivorous species accounted for around 87% of total carabid abundance (Table 2). Weed seeds were the most valuable food for these carabids during the summer (Tables 3 and S2a, Supporting information), but these resources were not utilised in a uniform way. Preferences of carabid taxa differed markedly according to both their functional grouping and that of the weed community. Autumn-breeding, omnivorous carabids (Group C4) were correlated with total weeds across all crops (Table S2a, Supporting information). However, when the consistency of responses between these taxa was emphasised by matrix regressions and ordinations, there was a preference for spring-germinating weeds in Group W5 (Table 3 and Figs 2a and 3a). Responses of more specialised seed-feeding omnivores became strongly divergent, with contrasting but even stronger links established between weeds in different functional groups. Larger, mainly autumn-breeding carabids (Group C6) were strongly correlated with larger-seeded, spring-germinating weeds in Group W5 (Table S2a, Supporting information). Matrix regression, pRDA ordination and log-linear analyses demonstrated the consistency of such links across species (Tables 3 and S3, Supporting information and Figs 2 and 3). Of the autumn-breeding, omnivorous carabids, Calathus and Harpalus spp., were noticeable for their consistent and significant relationships with Group W5 weeds (Table S3, Supporting information and Figs 2a and 3a).

Figure 2.

 Omnivorous carabid responses to weeds in spring crops in the summer: (a) partial redundancy analysis (pRDA) of carabid responses; figures in brackets indicate carabid functional groups (see Table 2 for species abbreviations); underlined genera are projected passively onto the diagram; significant explanatory weed variables are shown by bold vectors; WAX1 = PCA axis representing small-seeded weed taxa in Groups W2, W3 and W4 (conditional eigenvalue = 0·012, F(10,154) = 2·68, P = 0·022), and WAX2 = PCA axis for Group W5 weeds (marginal eigenvalue = 0·021, F(9,155) = 4·81, P = 0·004); other explanatory variables are shown by grey line vectors and are projected passively onto the diagram; WE and WR = logit-transformed ratios of emerged weed and seed rain counts, respectively, within the functional group indicated by the proceeding number; RPANN = Poa annua seed rain; envelopes contain carabid and weed variables with significant univariate relationships (see Table S3, Supporting information); and (b) logistic regression model, with fitted binomial error structure, where dependent variables are the proportion of counts of each carabid taxa to total carabid counts at sites, categorised by their functional groups; C4 (upward triangles, trend = solid grey line); C5 (circles, trend = solid black line) and C6 (crosses, trend = broken black line) and independent variables are half-field scores on the second axis of the pRDA; the interaction between a factor for carabid groups and the second pRDA axis is significant in an accumulated analysis of variance (deviance ratio (2,489) = 18·29, P ≤ 0·001); with the parameter for Group C5 being significantly different from Groups C4 and C6 (coefficient = −0·086, t-value(2,489) = −2·81, P = 0·005), a dispersion parameter of 4·23 from the residual deviance was fitted.

Figure 3.

 Omnivorous carabid responses to weeds in winter oilseed rape in the spring and early summer: (a) partial redundancy analysis (pRDA) of carabid responses: see Fig. 2 for variable descriptions and diagram formatting, additionally; ECBP and RCBP = emerged and seed rain counts of Capsella bursa-pastoris, respectively, and EURE = emerged counts of Urtica urens; significant explanatory variables under independent tests; WAX1 (eigenvalue = 0·042, F(6,52) = 4·05, P = 0·002) and WAX2 (eigenvalue = 0·041, F(6,52) = 4·02, P = 0·002); and (b) logistic regression model, see Fig. 2 for descriptions of the symbols, trend lines and parameterisation of the model, additionally; Group C3 carabids are shown by downward triangles and broken grey trend line; the interaction between a factor for carabid groups and the first pRDA axis is significant in an accumulated analysis of variance (deviance ratio (3,228) = 10·71, P ≤ 0·001); with the parameter for Group C5 being significantly different from Groups C3, C4 and C6 (coefficient = −0·538, t-value(228) = −4·08, P ≤ 0·001), a dispersion parameter of 7·12 from the residual deviance was fitted.

In contrast to other omnivorous species, smaller, spring-breeding carabids, with a preference for granivory in Group C5, had exclusive and noticeably uniform responses to small-seeded, autumn-germinating weeds in Groups W2, W3 and W4 (Tables 3, S2a and S3, Supporting information and Figs 2a and 3a). Comparisons between weed matrices and log-linear analyses demonstrated that short, early flowering weeds in Group W3 were most important for Group C5 carabids, as responses were at their strongest and most consistent across taxa in this functional group (Tables 3 and S2). Group W3 weeds were, however, associated with other small-seeded and autumn-germinating, but later flowering weeds in Groups W2 and W4 (Figs 1 and S1, Supporting information). Ordinations revealed close links between Group C5 carabids and community descriptors of weeds sharing these traits (Figs 2a and 3a), and log-linear analyses demonstrated correlations between a number of weed species with these traits in Groups W2/W4 and these beetles (Tables S2a and S3). This suggests that periodicity of emergence and seed size, rather than flowering time, are the main weed traits driving the responses of Group C5 carabids. Like other generalist omnivores, Group C3 carabids had preferences for Group W5 weeds, but the strongest response of the dominant species in the group, Poecilus cupreus (Linnaeus), was towards diverse plant communities (Table S3).

The trait-mediated links between trophic levels observed here, as well as being applicable to multiple taxa, are also consistent at wide spatial scales, being robust to ecosystem perturbations caused by changing crops, regions and years (Table S2b). Phenological synchronicity between different weed and carabid functional groups could not account for the quantitative links established between them. Although there were seasonal coincidences in the peaks of trophically associated functional groups, they were not mirrored by concomitant dissimilarities in the actual availability of different seed resources. For example, the abundance of seed rain from Group W3 weeds was always much higher than that of Group W5 in winter oilseed rape, regardless of the phenology of the taxa (Fig. S3, Supporting information).


Our results validate the hypothesis postulated by Lavorel & Garnier (2002) that traits determining the response of one group of organisms to the environment can, in turn, mediate effects on other functionally related taxa. Elucidating the form of this functional relationship is regarded as crucial to understanding and managing the impacts of anthropogenic change on ecosystem processes (Loreau et al. 2001; Kremen 2005). In a recent review of studies taking a functional trait approach, however, only 2% were applicable to trophic interactions (De Bello et al. 2010). Nonetheless, finding the mechanisms underlying trophic links is critical for modelling ecosystem dynamics because of cascading and amplification of effects between trophic levels (Thébault & Loreau 2003), which can affect community stability (Ives, Cardinale & Snyder 2005). Our results increase evidence that Lavorel & Garnier's (2002) hypothesis can be extended to trophic systems (Schmitz 2008), particularly as our study additionally shows that it can contain the inherent complexity of multiple, interacting species in diverse plant and insect communities.

More particularly, however, we show that the functional relationships described by Lavorel & Garnier (2002) are applicable not only to trophic interactions but to their robust modelling over wide spatial scales, something that was previously unknown (De Bello et al. 2010). This is vital for the wider application of Lavorel & Garnier's (2002) hypothesis because spatial scaling can fundamentally alter the dynamics of trophic links. For example, effects of landscape variation can be compounded through trophic levels, causing higher levels to respond at disproportionately wider spatial scales (Tscharntke & Brandl 2004; Purtauf, Dauber & Wolters 2005). Differences between the responses of carabid trophic groups have also been reported at this spatial scale, with granivores benefiting more from landscape diversity than carnivores (Vanbergen et al. 2010). Additionally, how trophically interacting species are linked spatially can determine their wide-scale responses to climate change (Schweiger et al. 2008). For our study system, however, we found that within-field variation in weed communities was sufficiently strong and consistent over large spatial scales to drive the wide-scale population dynamics of carabids. These patterns in weed communities will be a product of the timing of previous cultivations, largely determined by the proportion of spring cropping, and weed control in the current crop (Legere & Samson 1999). This enables a direct link to be made from crop management to carabid communities via plant traits, with autumn-germinating, small-seeded weeds associated with small, spring-breeding carabids, and spring-germinating, mainly large-seeded weeds, associated with larger, autumn-breeding carabids. Our results therefore highlight the utility of functional traits for modelling how the effects of perturbations cascade through trophic levels. More crucially, however, they also show that such functional relationships can be stable across wide gradients of disturbance and spatial variation, in our case caused by contrasting management and cropping regimes, and large-scale, regional differences in biogeography.

The trophic links found may extend the value of carabids as environmental indicators by allowing better predictions of how changes in weed communities influence wider biodiversity. For example, Amara spp. may indicate changes beneficial to insects more generally, as the small-seeded weeds that they are associated with provide resources for many phytophagous taxa, whereas Harpalus spp. may be good indicators of the availability of larger-seeded weeds, which are important resources for farmland birds (Marshall et al. 2003). Corroborating our results in winter cereals would further increase this indicator potential of carabids because these crops account for much wider proportions of the UK landscape than the break crops investigated here. Understanding the mechanisms by which carabids select seeds may also further our understanding of their trophic relationships. Our results are congruent with allometric relationships between the size of carabids and the seeds on which they feed (Honek et al. 2007). This alone could explain the preferences of small, spring-breeding carabids which probably have morphological constraints to eating larger seeds. Bigger, autumn-breeding carabids, however, select larger seeds even when they are much less abundant than smaller alternatives, tentatively supporting the theory that unfavoured food for carabids contains toxins limiting consumption rates or that essential nutrients are only supplied by preferred foods (Hendry et al. 1994; Toft 1996). Alternatively, selection may be influenced by the timing of cultivations of the crops. For example, many autumn-breeding carabids overwinter as larvae in the soil, and they may have evolved preferences for seeds that germinate better when there is less disturbance at this time.

The viability of many ecosystem processes is rapidly undermined when the most functionally important species decline within a community (Zavaleta & Hulvey 2004; Larsen, Williams & Kremen 2005). Therefore, links between species traits, such those highlighted here, may be useful for predicting and managing these losses in ecosystem function. For example, autumn-breeding, seed-feeding carabids are nearly 100 times more numerous than spring-breeding, seed-feeding carabids, yet their preferred seeds are in much lower supply in some crops than those of the later group. Loss of seeds with traits suitable for autumn-breeding carabids may, therefore, have disproportionately negative effects on the functioning of food webs. Indeed, such trait-mediated responses in plants are often pivotal in how ecosystems function (Tilman et al. 1997). Therefore, the approach developed here may have wider potential for modelling the responses of other taxa which rely on weed resources, such as pollinators and phytophagous insects. For example, pollinators can be related to flower traits (Potts et al. 2003) and leaf-chewers to leaf traits (Perez-Harguindeguy et al. 2003). These trait linkages may provide better quantification of the trade-offs and synergies between resources provided by weeds to different taxa, permitting more comprehensive and integrated modelling of arable plant–invertebrate interactions.


We thank everybody within the scientific consortium responsible for the FSE and the participating farmers, and Aline Boursault for her helpful and constructive comments on drafts of this paper. The FSE were funded by Defra and the Scottish Executive. Rothamsted Research is an Institute of the Biotechnology and Biological Sciences Research Council of the UK.