Food web structure of three guilds of natural enemies: predators, parasitoids and pathogens of aphids


  • F. J. F. Van Veen,

    Corresponding author
    1. NERC Centre for Population Biology and
    2. Division of Biology, Imperial College London, Silwood Park Campus, Ascot, Berks, SL5 7PY, UK; and
    Search for more papers by this author
  • C. B. Müller,

    1. Division of Biology, Imperial College London, Silwood Park Campus, Ascot, Berks, SL5 7PY, UK; and
    Search for more papers by this author
    • Present address: Institute of Environmental Sciences, University of Zürich, Winterthurerstrasse 190, 8057 Zürich, Switzerland.

  • J. K. Pell,

    1. Plant and Invertebrate Ecology Division, Rothamsted Research, Harpenden, Herts, AL5 2JQ, UK
    Search for more papers by this author
  • H. C. J. Godfray

    1. NERC Centre for Population Biology and
    2. Division of Biology, Imperial College London, Silwood Park Campus, Ascot, Berks, SL5 7PY, UK; and
    Search for more papers by this author
    • Present address: Department of Zoology, University of Oxford, South Parks Road, Oxford OX1 3PS, UK.

*Correspondence author. E-mail:


  • 1Most communities of insect herbivores are unlikely to be structured by resource competition, but they may be structured by apparent competition mediated by shared natural enemies.
  • 2The potential of three guilds of natural enemies (parasitoids, fungal entomopathogens and predators) to influence aphid community structure through indirect interactions is assessed. Based on the biology, we predicted that the scope for apparent competition would be greatest for the predator and least for the parasitoid guilds.
  • 3Separate fully quantitative food webs were constructed for 3 years for the parasitoid guild, 2 years for the pathogen guild and for a single year for the predator guild. The webs were analysed using standard food web statistics designed for binary data, and using information-theory-based metrics that make use of the full quantitative data.
  • 4A total of 29 aphid, 24 parasitoid, five entomopathogenic fungi and 13 aphid specialist predator species were recorded in the study. Aphid density varied among years, and two species of aphid were particularly common in different years. Omitting these species, aphid diversity was similar among years.
  • 5The parasitoid web showed the lowest connectance while standard food web statistics suggested the pathogen and predator webs had similar levels of connectance. However, when a measure based on quantitative data was used the pathogen web was intermediate between the other two guilds.
  • 6There is evidence that a single aphid species had a particularly large effect on the structure of the pathogen food web.
  • 7The predator and pathogen webs were not compartmentalized, and the vast majority of parasitoids were connected in a single large compartment.
  • 8It was concluded that indirect effects are most likely to be mediated by predators, a prediction supported by the available experimental evidence.


Ecological communities may be haphazard collections of weakly interacting species or they may be structured by ecological processes such as competition, predation and parasitism. While resource competition has played a major role in explaining the structure of many ecological communities [e.g. plants (Tilman 1982); intertidal invertebrates (Menge 1976; Paine 1984)] it is unlikely to play a significant role in contemporarily structuring communities of herbivorous insects. The latter often consist of species that are restricted to feeding on nonoverlapping sets of plant species and therefore cannot interact through interference or resource competition. However, interactions may occur through shared natural enemies – predators, parasitoids or pathogens – and these indirect interactions may be important in influencing community structure. The simplest type of indirect interaction is apparent competition in which the presence of one species leads to higher densities of a common natural enemy, which consequently causes increased mortality and a lower population density of a second species. This idea, which has a long pedigree in ecology (Jeffries & Lawton 1984; Abrams 1987; Holt & Kotler 1987; Holt & Lawton 1994; Abrams & Matsuda 1996; van Veen, Morris & Godfray 2006b), was first formally explored by Holt (1977) and its principles were experimentally demonstrated by Bonsall & Hassell (1997, 1998). Just like interference and exploitation competition, apparent competition and related interactions can also structure ecological communities (Jeffries & Lawton 1984; Lawton 1986; Godfray 1994; Hawkins 1994).

There is as yet only limited experimental evidence for apparent competition and related indirect effects among herbivorous insects (Berdegue et al. 1996; van Veen et al. 2006b). The three main categories of natural enemies that might mediate these interactions are predators, parasitoids and pathogens. Short-term apparent competition between herbivorous arthropods mediated by predators has been demonstrated in field experiments (Karban, Hougen-Eitzmann & English-Loeb 1994; Müller & Godfray 1997; Rott, Müller & Godfray 1998) and has been shown to prevent the establishment of some aphid species in a field even though they were common in the aerial plankton (Müller & Godfray 1999). There is also evidence for apparent competition mediated by parasitoids among leafhoppers (Settle & Wilson 1990) and among rainforest leafminers (Morris, Lewis & Godfray 2004). The advantages of providing alternative hosts for the natural enemies of crop and forestry pests that presupposes indirect effects between herbivores have also received much attention (e.g. Landis, Wratten & Gurr 2000; van Veen, Memmott & Godfray 2006a).

Food webs provide a valuable means of assessing the potential for indirect effects and for generating testable hypotheses (Cohen, Brian & Newman 1990). Traditional food webs consist of sets of binary links between ‘trophic species’ representing feeding interactions. These are particularly sensitive to sampling effort (Polis 1991; Bersier, Banaek-Richter & Cattin 2002). The use of trophic species also makes interpreting food web structure difficult (Sugihara, Bersier & Schoenly 1997; Solow & Beet 1998; Yodzis & Winemiller 1999), particularly when studying population-level processes where distinguishing among species is important. The alternative is to construct quantitative webs where the densities of all biological species and trophic links are given in the same units (Memmott & Godfray 1994; van Veen et al. 2006b).

Communities of hosts and parasitoids provide good systems for study using quantitative food webs, because the trophic links between hosts and parasitoids are relatively easy to establish and to quantify and a number of such parasitoid webs have now been described (Memmott, Godfray & Gauld 1994; Müller et al. 1999; Rott & Godfray 2000; Schönrogge & Crawley 2000; Henneman & Memmott 2001; Valladares, Salvo & Godfray 2001; Lewis et al. 2002). However, herbivorous insects are attacked not only by parasitoids but also by predators and pathogens and it is as yet unclear whether webs of the links between these guilds and their host/prey will show similar properties to those of the parasitoids.

The aim of this study was to construct sets of quantitative food webs for the three guilds of natural enemies attacking a community of aphids and to compare their properties, in particular with regard to the potential for apparent competition. The aphids (Aphidoidea) are a superfamily of phloem-sucking insects (Hemiptera: Sternorrhyncha). They are extensively attacked by predators, parasitoids and pathogens. We reasoned that natural enemies that required a prolonged intimate association with the host would evolve over time to be highly specialized with a narrower host range. The two groups of koinobiont parasitoids that attack aphids (Aphidiinae (Ichneumonoidea: Braconidae) and Aphelinidae (Chalcidoidea)) lay single eggs inside live aphids and the larva may suspend development until the aphid has become an adult (Powell 1982; Askew & Shaw 1986). During this period the parasitoid is susceptible to attack by the host immune system and to survive it has to tolerate or disable the host defences, representing an intimate relationship with the host. The most important aphid pathogens are fungi from the Zygomycetes, order Entomophthorales. The conidia (spores) germinate on the aphid cuticle and invade the haemolymph and internal organs. However, unlike the parasitoids, the fungus does not suspend development and hence the interaction may be less intimate (Pell et al. 2001). Finally, specialist aphidophagous predators such as ladybirds (Coccinellidae); hoverfly larvae (Syrphidae); predatory midge larvae (Cecidomyiidae); lacewing larvae (Chrysopidae) and species of Anthocoridae (Hemiptera) (Dixon 1985) exploit aphids as prey rather than as hosts for reproduction. Predators generally do not require being adapted to a close and prolonged interaction. We hypothesized that connectance would decrease from predator guilds through pathogen guilds to parasitoid guilds on the grounds that the more intimate the relationship between aphid and natural enemy, the greater the degree of specialization that would evolve. Hence, we expected predators to show the greatest potential to mediate apparent competition. As a corollary of this, we expected species diversity to increase from predator guilds through pathogen guilds to parasitoid guilds as niche space became more finely divided.


We studied the aphid–natural enemy community in a damp field (Rush Meadow) at Silwood Park, Berkshire, UK. The study site was 18 000 m2 in size, trapezoidal in shape, and surrounded on three sides by damp woodland dominated by Alnus, Salix and Betula, and on the fourth side by a higher, dryer, heavily grazed field. The plant community in the field was typical for a poorly drained mesotrophic grassland and consisted of patches of MG9 Holcus lanatus–Deschampsia cespitosa grassland and MG10 Holcus lanatus–Juncus effuses rush-pasture, according to the British National Vegetation Classification (Rodwell 1998). A full list of plant species from which aphids were recorded is given in the Appendix.

From April to October all plants in the field were scanned for aphids. When noted on a particular plant species, the density of each aphid species on that plant species was estimated by counting twice a month the number of individuals per ‘plant unit’. The plant unit was chosen to be a relatively constant unit of vegetation, often the individual plant in the case of herbaceous forbs, a flowering stem in the case of grasses fed on by aphids that congregate on flower spikes, or a 30 cm terminal shoot in the case of shrubs where aphids feed on young growth. On every fortnightly sampling occasion a minimum of 300 units of every plant species were chosen at random, stratified over forty 20 × 20 m grid cells covering the whole site. At the same time as the aphids were surveyed, predator, parasitoid and pathogen densities were estimated by counting the number of predator individuals, parasitoid mummies (parasitoid pupae attached to the plant substrate) and pathogen-infected cadavers associated with each plant unit. Plant abundance was estimated by recording the number of plant units per m2 at 3-m intervals along 15 parallel transects. The transects were placed at 20-m intervals over the whole site and care was taken that they ran through each of the forty 20 × 20 m grids. Estimates of plant density per m2 allowed us to translate all the relative estimates of aphid and natural enemy densities to absolute figures with units per m2.

To assess the composition of the parasitoid and pathogen communities, a sample of parasitoid mummies and infected cadavers was taken to the laboratory on each sampling date for identification. For each plant species on which parasitoid mummies were observed we collected a maximum of 200 mummies every fortnight or fewer when they were rare. The target sample size for fungus-infected cadavers was 50. The parasitoid mummies were reared individually in gelatine capsules and the adult insects identified using the keys in Mackauer (1959), Starý (1966), Gärdenfors (1986) and Pungerl (1983). Some species boundaries were resolved using cellulose acetate allozyme electrophoresis (see Appendix in Müller et al. 1999). The fungus-infected cadavers were placed on wet filter paper and inverted over microscope slides in 9 cm diameter Petri dishes overnight at room temperature. The discharged conidia on the microscope slide were fixed and stained in cotton blue (diluted to 10% in lactophenol) and identified from their gross conidial morphology using the descriptions of Wilding & Brady (1984), Humber (1989) and Keller (2006). Although 29 species of Entomophthorales have been recorded from aphids worldwide (Keller 2006), many are extremely rare and we are confident that our identifications are accurate in the vast majority of cases. The aphids were identified using the keys of Heie (1986, 1992, 1994, 1995).

For the predators we concentrated on specialist predators of aphids, specifically Coccinellidae (adults and larvae), Syrphidae (larvae), Chrysopidae (larvae), Anthocoridae (adults and nymphs) and Cecidomyiidae (larvae). These were identified using the standard identification manuals for each group. The abundance of each species and developmental stage was estimated as counts per plant unit and later translated into density per m2, as described above.

Unlike the parasitoid and fungal pathogens where the aphid mummy/cadaver is a natural unit to compare across species, predators vary greatly in size and obtain different amounts of sustenance from aphids of different size. To correct for this, in calculating link density we assumed that the volume of food eaten is proportional to metabolic rate, and that this scales with the 3/4 power of body size, the most common though not the only reported relationship (reviewed in Glazier 2005). Thus in the predator food webs the aphids are scaled by their volume, and the predators by their ‘collective appetite’.

Another challenge with building a predator web is that the unit of measurement is of a predator individual on a host plant rather than the actual predation event itself. This is different from the data underlying the pathogen and parasitoid webs where the mummy/cadaver is the interaction incarnate. Difficulties arise when there are more than one prey species present on the same plant at the same time. Therefore, we constructed two alternative interpretations of the predator web that are likely to bracket what actually occurs in the field. In the first we assume that each predator feeds on the aphids on a particular host plant in proportion to their abundance. In the second we assume that the predator feeds only on the most abundant prey species (leading to a reduced number of trophic links). Below we shall call the two types of web proportional and specialized predator webs, respectively. We present data for predators in 1996, pathogens in 1997 and 1998, and parasitoids for 1996–98.

The majority of food web statistics have been designed for qualitative, binary webs where species and links among them are scored as either present or absent. Quantitative webs contain richer information. We follow Bersier et al. (2002; see also Ulanowicz & Wolff 1991) and use a series of metrics based on Shannon (1948) information theory. To prevent repetition we introduce them when they are used in the Results but they are largely based on the statistic


where x is a vector of quantities such as species densities indexed by i = 1 ... n, and inline image is the sum of the quantities. When all species have the same density then h(x) reaches its maximum, which is equal to n. The further quantity h′(x) = h(x)/n, which takes a value between 0 and 1, is a scaled measure of evenness. We shall refer to these below as the diversity metrics h and h′.


food webs

Quantitative food webs were constructed to describe the interactions between aphids and their primary parasitoids in 3 years (Fig. 1), aphids and entomopathogenic fungi in 2 years (Fig. 2) and aphids and the guild of predators specialized on this food resource in 1 year (Fig. 3).

Figure 1.

The quantitative aphid–parasitoid food webs constructed in each of the 3 years of the study. The conventions used for drawing the webs are described in the first section of the Results. The species identities of the codes in the figure are given in Table 1.

Figure 2.

The quantitative aphid–fungal pathogen food webs constructed in 1997 and 1998. The conventions used for drawing the webs are described in the first section of the Results. The species identities of the codes in the figure are given in Table 1.

Figure 3.

The quantitative aphid–predator food web constructed in 1996. The conventions used for drawing the webs are described in the first section of the Results and are similar to the parasitoid and pathogen webs except (i) aphids and predators are scaled by biomass and appetite, respectively, and (ii) two versions of the predator web are drawn assuming narrow prey range on a plant (specialist web) and random predation (proportional web). In both cases see the Methods for more detail. The species identities of the codes in the figure are given in Table 1.

In drawing the webs we use the same conventions as in our earlier work (Müller et al. 1999). Hosts are arranged as a series of bars in a lower register with the width of each bar proportional to the aphid's cumulative abundance over the year. The total host density is given in the legend below, the units being cumulative aphids per m2 (aphids were sampled every 2 weeks and their densities added together to get a cumulative total). Natural enemies are arranged as a series of bars in an upper register with again the width of each bar being proportional to species’ cumulative abundance. The width of the natural enemy bars are magnified relative to those of the aphids by the factor given in the upper legend. The predator web has been scaled to account for differences in body size and the units of abundance are now related to biomass rather than number of individuals. Natural enemies and aphids are linked by triangular wedges, the relative widths of which at the natural enemy register represent the contribution of each aphid species to the diet or host range of the natural enemy. All species are numbered and their identities are provided in Table 1.

Table 1.  Identity of species in the food webs
CodeAphid name (host plants in brackets: for names see Appendix)CodeParasitoid nameCodePathogen name 
1Capitophorus carduinis (1) 1Aphidius matricariae 1Pandorra neoaphidis 
2Aphis fabae cirsiiacanthoides (1, 2) 2Aphidius urticae 2Entomophthora planchoniana
3Uroleucon cirsii (1, 2) 3Aphidius microlophii 3Zoophthora phalloides 
4Microlophium carnosum (3) 4Aphidius picipes 4Conidiobolus obscurus 
5Aphis urticata (3) 5Praon dorsale 5Neozygites microlophi 
8Brachycaudus cardui (4) 6Aphidius eadyi   
9Acyrthosiphon pisum (5, 6, 14) 7Aphidius erviCodePredator nameFamily
10Macrosiphum funestum (7) 8Aphidius rubi   
11Amphoraphora rubi (7) 9Aphelinus abdominalis 1Adelia bipunctataCoccinellidae
12Aphis ruborum (7)10Aphidius rhopalosiphi 2Anthocoris nemoralisAnthocoridae
13Aphis rumicis (8)11Aphelinus varipes 3Aphidoletes aphidimyzaCecidomyiidae
14Metopolophium albidum (9)12Ephedrus plagiator 6Chrysoperla carneaChrysopidae
15Sitobion spp. (avenae & fragariae) (10)13Praon volucre 7Coccinella septempunctataCoccinellidae
16Megoura viciae (6)14Trioxys acalephae 8Episyrphus balteatusSyrphidae
17Aphis spp. on Epilobium (11)15Lysiphlebus fabarum 9Epistrophe eligansSyrphidae
18Sitobion ptericolens (12)17Praon abjectum11Eupeodes lunigerSyrphidae
20Aphis jacobaeae (4)19Ephedrus lacertosus16Orius sp.Anthocoridae
21Aphis salicariae (13)20Aphidius funebris19Propylea quattuordecimpunctataCoccinellidae
23Aphis sarothamni (14)23Ephedrus cerasicola21Scaeva pyrastriSyrphidae
24Megourella purpurea (6)24Trioxys centaureae22Syrphus ribesiiSyrphidae
25Sitobion fragariae (on primary host) (7)25Ephedrus niger23Syrphus vitripennisSyrphidae
26Macrosiphum euphorbiae (15, 4, 16)26Trioxys pallidus   
27Metopolophium dirhodeum (9, 17)29Trioxys sp. 1   
28Diuraphis holci (18)30Aphidius salicis   
29Hydaphis foeniculi (16)     
30Hyalopteroides humilis (17)All Braconidae except   
36Aphis fabae (15, 19, 16)9 and 11, Aphelinidae   
39Myzus sp. (15, 20, 21, 22, 1)     
40Cavariella angelica (16)     

Visual comparison of the food webs in Fig. 1 shows some support for our hypotheses. The predator food web for 1996 is clearly much more connected than the parasitoid web of the same year. Below we first present statistics on the annual variation in host/prey abundance and then present the qualitative and quantitative food web statistics to compare the natural enemy guilds.

host/prey abundance

In the 3 years of the study the number of aphid species recorded in the site remained approximately equal: 23–25 of a total of 29 (Table 2). Aphid densities were approximately the same in the first 2 years but about seven times higher in the final year (Table 2). Part of the reason for the higher density in 1998 was the colonization of the field site by Diuraphis holci, which became the most abundant aphid (55% of all individuals). However, even omitting this species, abundances were about three times higher in year 3.

Table 2.  Numbers, density and diversity of aphids in the 3 years of the study
Number of aphid species 23 23  25
Total aphid density2853072014
Diversity h(aphid density)  3·6  8·6   4·9
Diversity h′(aphid density)  0·16  0·38   0·20
Diversity h1(aphid density)  9·8  8·6   7·5

The abundances of individual aphid species were highly correlated between the first 2 years (Pearson, ρ23 = 0·85) but even after omitting D. holci there was only a weak nonsignificant correlation between abundances in the final year and in either of the two earlier years (ρ26 = 0·13, ρ26 = 0·14, respectively). There were greater differences in evenness among years (Table 2) with the diversity metrics h and h′ showing relatively low diversities in years 1 and 3. We have already mentioned that in year 3 D. holci was particularly common, and in year 1 the thistle aphid Capitophoris carduinis was relatively common making up 70% of all aphid individuals recorded. If these two dominant species are omitted and the metric h recalculated (h1 in Table 2) then all 3 years show similar diversities.

comparison of guilds

Species numbers and predator–prey ratios

The results for this section are summarized in Table 3. We recorded 24 species of parasitoid in total of which between 12 and 19 were found in any particular year. Parasitoid densities varied, roughly in line with those of the aphids, and were particularly high in 1998. Only five species of entomopathogenic fungi were recorded, all but one present in both years. Pathogen densities were considerably higher in 1998 compared with 1997. We found 13 species of specialized aphid predator in the one year we studied this guild (Table 3). We have not attempted to calculate an overall predator density because of their great variation in mass. The marked unevenness in natural enemy densities meant that the diversity metrics h took values considerably lower than the species numbers. The parasitoid figure for 1998 is strongly affected by the presence of the common A. varipes (omitting this species increases the value of h to 7·2).

Table 3.  Numbers, density and diversity of natural enemies in the years they were studied, and enemy:aphid ratios
Number of species1219174 513
Density of natural enemy13 461384*
Diversity h(enemy density) 2·4 6·1 3·42·4 2·3 4·2
Diversity h′(enemy density) 0·23 0·32 0·200·60 0·46 0·32
Enemy:aphid ratio 0·52 0·83 0·680·17 0·20 0·57
h(enemy):h(aphid) density ratio 0·65 0·70 0·690·27 0·47 0·56

The different numbers of species in each guild of natural enemies are reflected in changes in the natural enemy to aphid ratio (the equivalent of the standard predator–prey ratio) though the differences are less marked for the ratio of the diversity statistics (Table 3). To calculate the quantitative predator–prey ratio we measured predator density not in terms of individuals but of biomass (see Methods). Doing this increases the quantitative measure of aphid diversity from 3·6 to 7·5 because larger aphids tended to be rarer than smaller aphids.

connectance and compartmentalization

The number of links in the seven food webs varies from 20 in the 1996 parasitoid web to between 61 and 83 in the predator web (depending on assumptions about predator specialization) (Table 4). There was considerable variation between years for the guilds we studied in more than one season. There were fewer trophic links in 1996, the year when the least number of parasitoid species was recorded. More surprising is the over threefold increase in number of links involving fungal pathogens between 1997 and 1998. As noted above, pathogens were much more common in the second year and through a purely sampling effect it was possible to detect many more relatively rare associations. Also in 1998 the nettle aphid Microlophium carnosum was the most important host of all species of pathogens. In 1997 M. carnosum was rare, which might account for the paucity that year of pathogenic fungi.

Table 4.  Numbers and diversity of aphid–enemy trophic links, and measures of host/prey range, connectance and compartmentalization
Number of links20394215488361
Diversity h(links) 2·38 8·16 3·76 2·94 4·4730·8727·97
Diversity h′(links) 0·12 0·21 0·09 0·20 0·09 0·37 0·46
Host/prey range 1·67 2·05 2·47 3·75 9·60 6·38 4·69
h(host/prey range) 1·24 1·48 1·38 1·41 1·98 3·89 3·67
Connectance 0·07 0·09 0·10 0·16 0·38 0·28 0·20
h(Connectance) 0·009 0·019 0·009 0·032 0·036 0·103 0·094
Number of compartments 3 1 5 1 1 1 1
h1(compartments) 1·89 1 2·35 1 1 1 1
h2(compartments) 2·17 1 1·99 1 1 1 1

Variation in link densities means that the diversity metric, h, is always considerably lower than the number of trophic links. We checked to see if any particularly common association had a major effect on the statistic and found that only one link made a major difference, that between D. holci and A. varipes in the 1998 parasitoid web – omitting this increased link diversity to 10·25 (h′ = 0·24). In the pathogen webs the h statistic increased between years by a factor of 1·5, much less than the threefold increase in raw link numbers because h is far less affected by the discovery of a few instances of rare associations. Compared with both the parasitoid and pathogen webs, the quantitative measure of the number of links involving predators is high, and (see h′ in Table 4) nearer the qualitative measure. There are thus more predator links, and they tend to be more even. Finally note that there is relatively little difference in the quantitative measure of link abundance in the ‘proportional’ and ‘specialized’ predator webs. This is reassuring as it suggests our precise assumptions about predator behaviour will not affect the conclusions, at least those derived from the quantitative statistics.

From the food web we can calculate the average host or prey range of the different natural enemies. For the parasitoids it varies from 1·5 to 2·5 between years while for the pathogens it is 3·7 in the first year and nearly 10 in the second. For the predator web it is 4·7 if we assume specialized feeding and 6·4 for proportional feeding. Calculating the average quantitative measure of host range (mean h(links per host)) gives a simpler picture: parasitoids and pathogens have relatively narrow host ranges with equivalent host ranges between 1 and 2 while the figure for predators is c. 3·8 and only slightly affected by our assumption about feeding specialization.

Connectance is defined as the proportion of all possible links that is realized. Because we only recorded host–natural enemy associations we define connectance here as the number of links divided by the product of host and natural enemy numbers (Müller et al. 1999; Lewis et al. 2002). If every natural enemy fed on every host or prey then this measure equals 1. Adapting Bersier et al. (2002) we define the quantitative (weighted) equivalent to be h(links) divided by aphid times natural enemy species. This measure ranges from 0 to 1 and reaches the maximum when every natural enemy attacks every host or prey with equal intensity.

Connectance is low (≤ 0·1) in the parasitoid web, variable in the pathogen web (0·16 and 0·38) while in the predator web it varies between 0·20 and 0·28. However, adopting the quantitative measure of connectance, the parasitoid web is clearly the least connected, the pathogen web intermediate and the predator web the most connected (Table 4). The two different assumptions we made about predator feeding had only a minor effect on connectance (0·10 vs. 0·09). The only particular link having a significant strong influence was the D. holci–A. varipes association in the 1998 parasitoid web; without it, quantitative connectance is 0·024.

Compartmentalization is the degree to which a food web is divided into nonconnected subwebs. Both the pathogen and predator webs are fully connected, as is the 1997 parasitoid web. In 1996 one host–parasitoid association (the most common) and in 1998 four pairs of associations (including the two most common) were not linked to the rest of web giving two and five compartments, respectively.

We can measure how the disparity in compartment size affects a quantitative measure of compartmentalization by calculating either h(species per compartment) or h(number of associations per compartment); in Table 4 these are labelled h1(compartments) and h2(compartments), respectively. Using the second quantitative measure the 1996 web is now more divided than the 1998 web.


We predicted that the predator guild would show the highest degree of linkage and connectedness, parasitoids the least, and with fungal pathogens intermediate but nearer parasitoids than predators. The food web data confirm our initial hypothesis. The predator web contains more links than the parasitoid or pathogen webs, the average number of aphid species attacked per predator species tends to be greater than the host range of parasitoids and pathogens, and predator food web connectance tends to be higher than for the other guilds. The exception to these trends is the pathogen web for 1998 when a large number of rare associations increases both average host range and connectance. However, if we adopt the quantitative measures of these two statistics, which de-emphasizes rare and so presumably less important interactions, the data are in very good agreement with the hypothesis. Quantitative link diversity is approximately the same for the parasitoid and pathogen guilds, and nearly double for predators. Quantitative connectance is highest for predators and lowest for parasitoids, and intermediate for pathogens. In the year that both the predator and the parasitoid web were sampled the quantitative connectance differed by a factor of 10.

The main motivation behind this study was to compare the potential for apparent competition mediated by the three guilds of aphid natural enemies. The majority of aphids present in our community cannot compete directly because they feed on different host plants (375 of the 406 possible pair-wise aphid competition coefficients are zero for this reason). If our community is structured by biotic interactions then they must be mediated by processes other than resource and interference competition. A likely route is that aphid species on different host plants interact through shared natural enemies. The analysis of the webs clearly shows that predators have the greatest potential for mediating apparent competition.

While we present parasitoid data for 3 years, a weakness of the study is that we studied the predator and pathogen guilds for only 1 and 2 years, respectively. However, over the 3 years the number of aphid species remained relatively constant despite marked differences in overall aphid density. The major difference between the 3 years was that in the first and last years two different species of aphid were particularly abundant. If these species are omitted then aphid diversity (measured by the h statistic) as well as species number are similar across years. Although this does not avoid the problem of studying different guilds in different years it is reassuring that interannual differences are not more pronounced. Moreover, the magnitude of the differences observed among the webs of the different guilds, as opposed to differences among years within guilds, strongly indicate that it is unlikely that the observed effects were down to chance.

We conclude that indirect interactions such as apparent competition which might structure aphid communities are most likely to be mediated by predators. This is supported by a series of experiments we have carried out with aphids from this community in which we have compared the dynamics of aphids feeding on one species of host plants in the presence or absence of a second species of aphid feeding on a different plant species (Müller & Godfray 1997; Rott et al. 1998; and unpublished). We also know that some species that are absent from our study community despite the presence of their host plant are able to persist if protected from aphid–specialist predators (Müller & Godfray 1999).

We have so far failed to demonstrate parasitoid-mediated (unpublished results) or pathogen-mediated (Pope et al. 2002) apparent competition. However, the pathogen food web described here suggests that one species, the nettle aphid Microlophium carnosum, may be particularly important as a source of fungal pathogens attacking other aphid species. In 1998 when nettle aphids were abundant they were responsible for a large proportion of the total fungal cadavers recorded. In Fig. 4 we display the 1998 fungal food web in a different way, as a quantitative overlap graph (Müller et al. 1999). The figure shows the importance of M. carnosum (species ‘4’ in the diagram): it is more strongly connected to other aphids through shared fungal pathogens than any other species, and it acts as a source of pathogens attacking these other species. We predict that under meteorological conditions favouring fungal pathogens the presence of nettle aphids will negatively influence the population dynamics of other species feeding on nearby host plants.

Figure 4.

Quantitative overlap diagram for the 1998 fungal pathogen web. Aphid species are represented as nodes arrayed in a circle; the nodes are linked if they share a common natural enemy. A statistic (dij) is calculated that describes the probability that a natural enemy attacking one species of aphid (i) developed on a second species (j). It is assumed that the natural enemy population is well mixed and does not consist of host races. The width of the line connecting species i and j at i is proportional to dij. The diameter of the node represents aphid abundance, and the degree to which it is coloured black di. The species identities of the codes in the figure are given in Table 1. The diagram shows that species 4 (nettle aphid Microlophium carnosum) may act as a major source of fungal spores attacking other aphid species.

The density of pathogens between the 2 years was dramatically different with much higher densities in 1998, both in absolute terms and relative to the aphids. Although outbreaks (epizootics) of entomopathogenic fungi are often associated with wet conditions we cannot attribute the low densities in 1997 to unsuitable meteorological conditions. Weather data for the study site show that temperature, total rainfall and number of rain days are very similar in the summers of 1997 and 1998. It is possible that the high pathogen densities in 1998 were due to the higher aphid densities, especially of M. carnosum, which would have allowed for more efficient transmission of the pathogens.

Food webs contain information only about the trophic interactions in a community while populations may interact by other means, which can affect community structure. For example, plants responding to herbivore attack by releasing volatile chemicals that attract parasitoids (Turlings, Tumlinson & Lewis 1990; Guerrieri et al. 1999). Interactions among the guilds of natural enemies may also play a role. It is known that predators may enhance transmission of fungal pathogens, but also interfere with pathogen and parasitoid efficacy by consuming infected and parasitized hosts (Pell et al. 1997; Roy et al. 1998; Roy, Pell & Alderson 2001; Baverstock, Alderson & Pell 2005). How such trait-mediated indirect effects affect the populations within a complex community is still unclear but in simple experimental insect communities, both trophic and trait-mediated interactions determine species persistence and community structure (van Veen, van Holland & Godfray 2005).

Phytophagous insect assemblages are some of the most diverse ecological communities on the planet. While much progress has been made in recent decades in understanding the dynamics of communities structured chiefly by resource competition (Tilman 1994; Menge 1995; Hubbell 2001) progress has been slower in the case of plant-feeding insects. One problem has been the difficulty in describing community structure, and in identifying the patterns that require an ecological explanation (if of course they exist). We hope that the construction of quantitative food webs and related approaches to determine systematically the structure of insect communities (Novotny et al. 2002, 2006) may help address this and stimulate the experimental work that will be required to test hypotheses about the nature of the processes concerned.


Financial support for this project was provided by the UK Natural Environment Research Council. A team of field assistants lead by Irma Adriaanse helped with data collection. The predator data were collected by Richard Cooke. Robert Belshaw assisted with parasitoid identification.


Plant species from which aphids and natural enemies were sampled. Codes correspond to codes listed after the aphids in Table 1

CodePlant name
1Cirsium palustre
2Cirsium arvense
3Urtica dioica
4Senecio jacobea
5Lotus uliginosus
6Lathyrus pratensis
7Rubus fruticosus
8Rumex obtusifolius, R. crispus + hybrids
9Arrhenatherum elatius (vegetative stage)
10Flowering grasses: Agrostis spp., Anthoxanthum odouratum,Arrhenatherum elatius, Dactylus glomerata, Deschampsiacespitosa, Holcus lanatus, Holcus mollis, Poa spp.
11Epilobium spp.
12Pteridium aquilinum
13Chamerion angustifolium
14Cytisus scoparius
15Galium aparine
16Angelica sylvestris
17Dactylus glomerata (vegetative stage)
18Holcus mollis (vegetative stage)
19Lysimachia vulgaris
20Glechoma hederacea
21Ranunculus repens
22Veronica chamaedrys