The structure of an aphid–parasitoid community


H. C. J. Godfray, Department of Biology and NERC Centre for Population Biology, Imperial College at Silwood Park, Ascot, Berkshire SL5 7PY, UK. Tel. & Fax 01344 294215 & 01344 294339. E-mail:


1. A quantitative parasitoid web was constructed describing the trophic relationships between the community of aphids, parasitoids and secondary parasitoids in an abandoned field in southern England. Root aphids were omitted and secondary parasitoids were associated with aphids rather than primary parasitoids. All aphids, parasitoids and trophic links were expressed in the same units (m−2). Over a 2-year period, separate webs were constructed for every month that aphids and parasitoids were active in the field.

2. Twenty-six species of plants were attacked by 25 species of aphid which were parasitized by 18 species of primary parasitoids. The primary parasitoids were attacked by 28 species of secondary parasitoids, of which 18 directly attacked the still living aphid (hyperparasitoids) and 10 attacked the aphid after mummification (mummy parasitoids). The aphids were attended by three species of ants.

3. Eleven webs were constructed for the months May to September 1994 and May to October 1995. Aphids and primary parasitoids were most abundant and diverse in June, and secondary parasitoids one month later.

4. The ratios of the number of aphid species to the number of species of primary parasitoid and hyperparasitoid were relatively constant across webs, as was the ratio of the number of links involving hyperparasitoids and mummy parasitoids. The ratio of the total number of links to the total number of species increased with web size.

5. The relative abundance of the species in the different webs was well described by a Dirichlet distribution with a common parameter. This implies a gamma distribution of aphid abundances in the field with few common and many rare species.

6. Predator (i.e. parasitoid) overlap graphs were constructed to assess the potential for indirect interactions between aphids. Mummy parasitoids were the most important group linking different aphids. A quantitative overlap diagram was developed to illustrate the potential strengths of indirect linkages. Common aphid species shared few strong, indirect links via primary parasitoids or hyperparasitoids, but could be strongly linked by mummy parasitoids. Symmetrical links were uncommon, and rare species were potentially strongly influenced by the presence of common aphids with which they shared parasitoids.

7. Ant-attended aphids were attacked by fewer species of primary parasitoids and hyperparasitoids than those species not attended by ants. Species of mummy parasitoids attack, on average, approximately twice the number of host species than species of either primary parasitoids or hyperparasitoids.


Recent reviews of foodweb theory and experimentation (Cohen, Briand & Newman 1990; Pimm, Lawton & Cohen 1991; Cohen et al. 1993; Polis & Winemiller 1996) have stressed the importance of: (i) collecting data specifically to build foodwebs (as opposed to using data collected for other purposes); (ii) using a common methodology that can be applied to different communities; and (iii) incorporating quantitative information. Because host–parasitoid interactions are relatively easy to establish and quantify compared to other trophic interactions, we believe that a comparison of different quantitative host–parasitoid foodwebs may be very valuable in testing general ideas in foodweb theory. Of course, the advantages gained from concentrating on host–parasitoid interactions come at the cost of ignoring other species found in the same habitat. To emphasize that only part of a larger community is being studied the webs constructed are referred to as quantitative parasitoid webs (Memmott & Godfray 1992, Memmott et al. (1994).

In this study a quantitative foodweb is constructed and analysed which describes the interactions within a community of aphids, their primary parasitoids and secondary parasitoids, in an abandoned field in the south of England. The data were collected in order to answer a set of questions about host–parasitoid community structure. First, changes in the structure of the foodweb are studied to see whether there are patterns in the community that remain constant over time. Quantitative foodwebs were constructed for each month that aphids were present over a 2-year period, giving a time series of 11 separate webs. The study investigates whether community statistics such as the ratio of the number of species and links, or the number of species at different trophic levels, remain constant across the different webs.

Second, we investigate the dynamic complexity of the community and the potential for indirect dynamic interactions between species via their natural enemies. The vast majority of experimental and theoretical studies of host–parasitoid population dynamics consider pairs or, occasionally, very small communities of hosts and parasitoids. The quantitative webs here were used to assess the degree to which the community is divided into subunits that are isolated or weakly connected with the rest of the foodweb and whose dynamics may be partly or wholly independent. Predator overlap graphs (Cohen 1978; Sugihara 1984) are used as a means to assess the potential for apparent competition (indirect interactions via natural enemies; Holt 1977; Holt & Lawton 1994) and to compare its importance at different trophic levels. An extension of predator overlap graphs has been developed to make use of the quantitative information available for the study community.

Parasitoid community ecology has been dominated for the last 10 years by a single question regarding the statistical correlates of the number of parasitoid species attacking different host species (Hawkins 1994). The reason for this is the ready availability in the literature of parasitoid lists for many species of hosts. This study investigates whether, in the present community, the number of parasitoid species attacking different species of aphid is influenced by host abundance, phenological range and the presence of attending ants. We also examined whether parasitoid host range is influenced by phenological range and by mode of attack (described below).

Aphids are attacked by hymenopterous parasitoids belonging to the Aphidiinae (Ichneumonoidea, Braconidae) and Aphelinidae (Chalcidoidea) (Starý 1970; Powell 1982). Both groups of primary parasitoids deposit their eggs internally in the aphid, the larva normally killing the later instar or adult, which it attaches to the plant surface to form a mummy inside which pupation occurs. The primary parasitoids are parasitized by two groups of secondary parasitoids. Members of the first group attack the aphid before mummification and lay their eggs inside the body of the primary parasitoid larvae. These parasitoids, largely Alloxystinae (Cynipoidea, Charipidae), delay their development until the primary parasitoid has caused mummification. A further group of secondary parasitoids attack the mummy, irrespective of whether it contains primary or secondary parasitoids. They paralyse the mummy inhabitants and their larvae develop immediately. Several unrelated genera (Asaphes, Coruna and Pachyneuron in the study site) of Pteromalidae (Chalcidoidea), and Dendrocerus (Ceraphronoidea, Megaspilidae) contain species with this life history. Secondary parasitoids in the genus Syrphophagus (Chalcidoidea, Encyrtidae) have a life history that is not perfectly understood but they appear to attack both living and mummified aphids. They are not common in the study community and they are included here provisionally in the second group of parasitoids. In the rest of the paper, the term hyperparasitoid will be reserved for species of secondary parasitoid that attack the living aphid and show delayed development (koinobiont life history), and the term mummy parasitoid will be used for species that attack the mummified aphid and develop without delay (idiobiont life history). Apart from parasitoids, the aphids in the study site were attacked by a number of predators and pathogens, some of which also prey on or infect mummies (Dixon 1985). Unfortunately, the work involved made it unfeasible to collect simultaneously quantitative information on shared predation and disease and this study is restricted to the parasitoid component of the aphid natural enemy complex.


Foodweb construction

We studied the aphid–parasitoid community in a damp field (Rush Meadow) at Silwood Park, Berkshire, England in 1994 and 1995. 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 dryer, heavily grazed field.

Parasitoid webs were constructed for every month in which adult aphids were present in the field site. Quantitative data were collected on the density of aphids and aphid mummies per ‘unit’ (see below) of their host plant, and this was converted to absolute density by surveying host plant density. The composition of the parasitoid community was obtained by rearing aphid mummies. We ignored one class of aphids, those feeding on plant roots. It was not possible practically to sample this guild of species, and they are rather infrequently attacked by a specialist group of parasitoids (Starý 1970).

Aphids and aphid mummies were surveyed twice a month and the data were combined to obtain monthly averages. On each sampling occasion all potential host plant species in the site were scanned for aphid colonies (possibly transient, winged aphids that had not produced offspring were ignored). Those plant species that were attacked by aphids were then sampled quantitatively. In each square of a 20 × 20 m grid covering the site, random ‘plant units’ of each species were selected and the numbers of aphids and aphid mummies counted nondestructively. The number and nature of the plant unit used varied amongst species but was typically the whole plant, or in the case of grasses a ramet. Separate counts were taken of nymphal, alate and apterous aphids for each species present on a plant. When counting mummies, care was taken to check for individuals that had mummified away from the host plant.

The densities of plants in the site were surveyed in August/September. Parallel transects, 20 m apart, were run through the site and the number of ‘plant units’ of each species (the same units used in the aphid survey) were recorded in every third m2-quadrat along the transect. Aphid and parasitized aphid densities were made absolute by multiplying the numbers per plant unit by the estimated density of plant units per m2 in the site.

The structure of the parasitoid community was assessed by rearing mummies collected in the quantitative sampling programme, supplemented by additional nonquantitative collections made in a stratified manner throughout the site. Mummified aphids were placed in gelatine capsules until the parasitoids emerged.

Aphids were identified using Heie (1980, 1982, 1986, 1992, 1994, 1995), Stroyan (1984) and Taylor (1984). Despite their economic importance, the parasitoids and hyperparasitoids of aphids present a number of taxonomic problems and our treatment of these is discussed in the Appendix.

Statistical methods

There are relatively few standard techniques available for analysing data such as those from the present study and in this section the methods used are described.

Ratio constancy

An important and controversial issue in foodweb theory is whether certain ratios, for example between the number of species in different trophic webs, or the number of species and the number of trophic connections (links), take constant values across different webs (for references see below). However, clear statistical criteria for constancy are absent. Wilson (1996) has argued that previous demonstrations of constant ratios of predators and prey cannot be distinguished from a simple null model (see also Goldwasser & Roughgarden 1997). To test for constancy, a statistical method is used that is equivalent to Wilson's randomization test of his null model. Specifically, it asks whether the ratio of, for example, aphids and primary parasitoid species is more regular than would be expected if the number of species of each was a Poisson-distributed variable conditional on fixed species totals. If the number of aphid species is defined as Y and the number of parasitoid species as X, then under the Poisson assumption with (X + Y) fixed, the distribution of X/(X + Y) is binomial and we thus ask whether the distribution X/(X + Y) across the 11 foodwebs has significantly less than binomial variance. The most straightforward estimate of the parameter of the binomial distribution, p, is ΣXi/Σ(Xi + Yi) where Xi and Yi are the number of parasitoids and hosts in the ith parasitoid web. However, communities with no aphids would not be recognized, leading to a slight bias in the estimation of p. This can be corrected by solving for p in the implicit formula:

image(eqn 1)

where Ni = Xi + Yi. Under the assumption of a binomial distribution, the log-likelihood deviance statistic

image(eqn 2 )

is asymptotically χ2 distributed with the degrees of freedom one less than the number of communities. Constancy, in approximately Wilson's (1996) sense, is a significantly small deviance statistic. We also ask whether the deviance is reduced significantly by making p a function of Ni using generalized linear modelling techniques (McCullagh & Nelder 1989): in other words whether there is scale dependence in the ratio of Xi and Yi. In the Discussion we return to two statistical points: whether the criterion of sub-binomial variance is an overly stringent definition of constancy, and the problem of the temporal nonindependence of the foodwebs analysed here.

Dirichlet distribution

Because quantitative parasitoid webs have been constructed, it is possible to compare the relative abundance and diversity of species across foodwebs. Statistically, this is more difficult than simply comparing species numbers because the null expectation of species distributions is harder to define. Consider the densities of the si species of aphids sampled in month i and let pi = {pi1, pi2 . . . pis} be the vector representing the proportions of each of the si aphids in the community. A reasonable null hypothesis is to assume that the aphid densities are described by si independent gamma variates with common shape parameter, κi, in which case (Engen 1978; Fang, Kotz & Ng 1990) the probability distribution function f (pi) is the Dirichlet distribution:

image(eqn 3 f)

which has the two parameters si (which is known) and κi (which can be estimated from the data). When the Dirichlet distribution applies, the marginal distribution of the species proportions, pij, is given by the Beta distribution with parameters [κi, κi(si − 1)].

The Dirichlet distribution has been used most frequently in the statistical study of compositions, in particular the mineral make-up of geological samples (Aitchison 1986), and the division of household budgets (Goodhardt, Ehrenberg & Chatfield 1984). Compositions present statistical difficulties because, by definition, the sum of components (expressed as proportions) must equal one. The advantage of the Dirichlet distribution is that it provides a relatively simple, analytically tractable description of a composition that can be derived from simple statistical assumptions (Goodhardt et al. 1984). The disadvantage is that it assumes that the elements of the composition are statistically independent (Aitchison 1986). Unfortunately, techniques that allow the covariance structure of a composition to be estimated (Aitchison 1986) are difficult to apply in situations such as that presented here where particular components often take zero values.

We asked whether fitting separate Dirichlet distributions to the aphid proportions in each of the monthly foodwebs provided a significantly better fit to the data than fitting a single distribution with a common κ. The densities of primary parasitoids, hyperparasitoids and mummy parasitoids were analysed in the same way. A good fit for a common distribution can be interpreted as evidence for constancy in community structure. We also asked whether a common distribution could be fitted to all the data from the four categories of species. The parameter κ was estimated using maximum likelihood and the significance of the simplified models were tested using likelihood ratio statistics, −2(Ln[L1] − Ln[L2]), where Ln[L1] and Ln[L2] are the maximized values of the log-likelihood functions under the simple and more complicated models, respectively. The statistic is asymptotically χ2 distributed with the degrees of freedom equal to the difference in the number of fitted parameters. Communities with single species in any category contribute neither to the likelihood function nor degrees of freedom. The standard error for κ was obtained as usual from the second derivative of the log-likelihood function.

Evenness statistic

A number of statistics have been suggested to describe the evenness of the numbers of individuals of different species in a community (e.g. Pielou 1969). One due to Alatalo (1981) based on Hill (1973) was used,

image(eqn 4)

where pk are species proportions. This statistic is particularly robust to the presence of outliers.

Variance of link numbers in predator-overlap graphs

A predator-overlap graph (Cohen 1978; Sugihara 1984) is a collection of points (vertices) representing prey joined by lines (links) if they share a common predator (or, here, parasitoid). If the numbers of species (n) and links (e) are known, then the expected number of links per prey is simply 2e/n. The variance of the number of links per prey can be calculated from the data and compared with the null hypothesis of a random distribution of links among prey pairs. The distribution of links under the null hypothesis is equivalent to picking (n − 1) links without replacement from a pool of (


) links, e of which represent successes, and is thus described by a hypergeometric distribution. The variance can be shown to be:

image(eqn 5 )

The statistical distribution of the variance was obtained by random simulation and the predator-overlap graphs were tested for greater or lesser than expected variance.

Quantitative parasitoid-overlap diagrams

Traditional predator-overlap graphs include no quantitative information about the strength and polarity of the links between two resource species, and we know of no quantitative equivalents in the literature. A new type of graph has therefore been developed to explore these data, where the links represent the potential importance of one species as a source of natural enemies attacking the other. Specifically, we ask what fraction of the parasitoids attacking species i are likely to have developed on species j. If this fraction is high, then the indirect effect of host j on i is possibly great. Note that the links between pairs of species will now often be asymmetric (i.e. the graph edges are directed) and in the diagrams this is represented by making the links of variable thickness, with the width of the link connecting i to j at the species i end representing the potential impact of species j on i, and the width at the j end representing the reciprocal indirect effect.

Technically, the magnitude of the effects of species j on i is defined as:

image(eqn 6)

where αik is the strength of the link between host i and parasitoid k in the quantitative parasitoid web. The quantity dij summarizes interactions between two hosts via all possible shared parasitoids and hence the outer summation in eqn 6 is taken over all parasitoids. The first quantity within the square brackets is the fraction of parasitoids of host i that belong to species k and the second quantity is the fraction of parasitoids of species k that develop on host species j. If species i and j share no parasitoids then dij = 0. If species i is attacked by one or several species of parasitoids, and if members of these parasitoid species develop very commonly on host j (which might be a much more abundant species in the environment), then dij→1. Clearly, alternative definitions for dij can be envisaged, but we believe eqn 6 provides a good description of potential indirect effects. Note, dii is defined by eqn 6 and represents the fraction of parasitoids attacking a host that are likely to have developed on aphids of the same species, and Σjdij = 1, which just states that all the parasitoids attacking a host must have developed on some host within the community (ignoring parasitoid immigration).


The results are presented in four sections. The first is purely descriptive and presents the data on the species present and the quantitative parasitoid webs. The following sections describe the analysis of the web to answer the three sets of questions listed in the introduction.

The quantitative parasitoid web

In the 2-year study 25 species of aphid were recorded breeding in the study field site. Their names and foodplants are given in Table 1. The aphids were attacked by 18 species of primary parasitoids and 28 species of secondary parasitoid, of which 18 were hyperparasitoids and 10 were mummy parasitoids (using the terms as described in the aphid parasitoid biology section above). In constructing the parasitoid webs, two pairs of aphid species were joined to make the composite taxa ‘Sitobion spp.’ (S. avenae and S. fragariae) and ‘Aphis (Epilobium)’ (A. epilobii and A. grossulariae). Members of each pair were found on the same foodplant and in neither case was it possible to identify unambiguously mummified aphids. Table 2 gives the matrix of the associations between aphid and parasitoid that were recorded, and also the numerical codes used to identify species in the parasitoid webs. As is discussed further below, we can only associate secondary parasitoids with the genus of primary parasitoid that they attack, and the associations that we have established are listed in Table 3. Three species of ants were found attending aphids in the site. The ant species were, with, in parentheses, the codes of the aphids (from Table 1) with which they were observed: Lasius niger (1, 2, 12, 13, 20, 21, 22); Myrmica ruginosus (1, 5, 8, 13, 20) and Formica fusca (1, 20).

Table 1. 
The host plants of the aphids occurring in the study site
CodeAphidHost plants
1Capitophorus carduinisCirsium palustre
2Aphis fabaeCirsium arvenseCirsium palustre
3Uroleucon cirsiiCirsium arvenseCirsium palustre
4Microlophium carnosumUrtica dioica
5Aphis urticataUrtica dioica
6Cavariella pastinacaeAngelica sylvestris
7Myzus persicaeSenecio jacobaeaRanunculus repens
8Brachycaudus sp.ASenecio jacobaea
9Acyrthosiphon pisumLotus uliginosusCytisus scoparius
10Macrosiphum funestumRubus sp.
11Amphorophora rubiRubus sp.
12Aphis ruborumRubus sp.
13Aphis rumicisRumex obtusifoliusRumex crispus
14Metopolophium albidumArrhenatherum elatius
15Sitobion fragariae & avenaeRubus sp. (S. fragariae only)Dactylis glomerataHolcus lanatusHolcus mollisPoa trivialisPoa pratensisAnthoxanthum odoratumAgrostis stoloniferaDeschampsia cespitosa
16Megoura viciaeLathyrus pratensis
17Aphis epilobii & grossulariaeEpilobium ciliatumEpilobium montanum
18Sitobion ptericolensPteridium aquilinum
19Ovatus glechomaeGlechoma hederacea
20Aphis jacobaeaeSenecio jacobaea
21Aphis salicariaeChamaenerium angustifolium
22Brachycaudus sp.BSenecio jacobaea
23Aphis sarothamniCytisus scoparius
Table 2.  Host associations recorded during the study (with codes for hosts and parasitoids in the parasitoid webs). Hosts are listsed across the top; primary parasitoids are listed first, followed by secondary parasitoids. Among the secondary parasitiods, the codes for mummy parasitoids (as opposed to hyperparasitoids) are shown in bold. Thumbnail image of
Table 3.  The host associations (at generic level) of the secondary parasitoids in the study site. Hyperparasitoids are given first, followed by mummy parasitoids
CodeSecondary parasitoidGenus of host
14Alloxysta brachypteraAphidius
26Alloxysta circumscriptaPraon
 8Alloxysta cursor
 5Alloxysta f1
22Alloxysta f3Binodoxys, Praon
20Alloxysta fulvicepsAphidius, Ephedrus, Praon
17Alloxysta fuscicornisAphidius
 7Alloxysta halterata
16Alloxysta m2Aphidius
 6Alloxysta macrophadna
15Alloxysta pleuralisBinodoxys
12Alloxysta r2Aphidius
27Alloxysta r3Aphelinus, Aphidius
28Alloxysta ramuliferaAphelinus
 1Alloxysta v2
 9Alloxysta victrix
13Phaenoglyphis villosaAphidius, Ephedrus, Aphelinus
23Phaenoglyphis xanthochroaAphidius, Praon
18Asaphes suspensusAphelinus, Aphidius, Ephedrus, Praon
 2Asaphes vulgaris
10Coruna clavataAphelinus, Aphidius, Binodoxys, Ephedrus, Praon
 3Dendrocerus aphidum
11Dendrocerus carpenteriAphidius, Ephedrus, Lysiphlebus, Praon
 4Dendrocerus dubiosus
19Dendrocerus laevisAphidius
24Pachyneurone aphidisAphidius, Lysiphlebus
25Syrphophagus aphidivorusAphidius, Lysiphlebus, Praon
21Syrphophagus mamitusAphelinus, Aphidius, Praon

Aphid colonies were first found at the beginning of May in both years, and continued until September (1994) or October (1995). Eleven monthly parasitoid webs were thus constructed which are shown in Figs 1 and 2 for 1994 and 1995, respectively. To see how this information has been displayed, consider the May 1994 web. In that month, six species of aphid were recorded in the study site (species 1, 2, 3, 4, 5 and 13) and their relative abundances are indicated by the widths of the six numbered bars in the centre of the diagram. Because aphid densities vary greatly from month to month, webs are not all plotted on the same scale. The bottom line of text states that the absolute density of all aphids in May 1994 was 27 m−2. The bottom series of bars represents primary parasitoids and the upper series of bars secondary parasitoids. Secondary parasitoids were associated with aphids and not with primary parasitoids because it was not possible in all cases to link unambiguously secondary parasitoids with primary parasitoids without making a number of assumptions about parasitoid biology (see Discussion). In May 1994 only two of the six aphid species were parasitized, species 1 which was attacked by a single primary species (1) and species 4 which was attacked by three (2, 3 and 4). There was no overlap in host range of the four parasitoids. The relative widths of the four parasitoid bars represent relative abundance within that trophic level. The line of text below the primary parasitoid bars gives scale information: the primary bars are drawn at 470 times the scale of the aphids and hence the total primary parasitoid density is 27/470 = 0·057 m−2. Early in the year parasitoids are very scarce. Secondary parasitoids are only associated with aphids from which primary parasitoids were recorded (very occasionally only secondary parasitoids were reared from aphid species of which very few mummies were found). In May 1994, four species (1, 2, 3, and 4) were reared from aphid 1 and seven species (2, 4, 6, 7, 9, 10 and 11) from aphid 4. Secondary species 2 and 4 were reared from both hosts; the relative basal widths of the ‘wedges’ linking the parasitoid to its two hosts represent the relative numbers developing on each species. As before, the relative widths of the secondary parasitoid bars represent relative abundances, and the scale statement at the top of the web allows conversion to absolute densities. Secondary parasitoid bars are coloured black to represent mummy parasitoids and grey to represent hyperparasitoids. In drawing the webs, the aphids were displayed in a fixed order and the parasitoids arranged by eye to minimize trophic link overlap. All webs were produced using low-level graphics routines in the computer package Mathematica.

Figure 1.

Figure 1.

Parasitoid webs for 1994. Relative aphid abundances are shown in the centre with primary parasitoids below and secondary parasitoid above (hyperparasitoids in grey and mummy parasitoids in black). The numbers are the species codes from Table 2. Species densities are shown to scale within each month for aphids and the two categories of parasitoids. Densities in different months (see Fig. 3) and of different species categories have different scales. A fuller description of the interpretation of the web diagrams is given in the text

Figure 1.

Figure 1.

Parasitoid webs for 1994. Relative aphid abundances are shown in the centre with primary parasitoids below and secondary parasitoid above (hyperparasitoids in grey and mummy parasitoids in black). The numbers are the species codes from Table 2. Species densities are shown to scale within each month for aphids and the two categories of parasitoids. Densities in different months (see Fig. 3) and of different species categories have different scales. A fuller description of the interpretation of the web diagrams is given in the text

Figure 2.

Figure 2.

Parasitoid webs from 1995.

Figure 2.

Figure 2.

Parasitoid webs from 1995.

An overview of the seasonal changes in the aphid community is provided by Fig. 3. In both years total aphid densities peaked in June and then declined over the late summer. Primary parasitoid densities also peaked in June, while secondary parasitoids were most common in July. In 1995, primary parasitoids were again relatively abundant in September, an effect that was almost entirely due to a single host association (Fig. 2e). Both aphid and parasitoid species diversity peaked in June and July, with the highest density of trophic links occurring in July. The monthly webs for the two years can be combined to give a summary quantitative web (Fig. 4). The webs were combined by simply adding together the estimated density of aphids and parasitoids in each month.

Figure 3.

The abundance and diversity of aphids and parasitoids in different months. Aphids and primary parasitoids were most abundant in June 1994, and secondary parasitoids in July 1994. The abundances in other months are shown as proportions of their maximum densities (i.e. densities are to scale within but not between species categories). Under ‘number of species’ the number of (i) aphids, (ii) primary parasitoids and (iii) secondary parasitoids found in each month are shown, and under ‘number of links’ the number of associations involving (i) primary parasitoids and (ii) secondary parasitoids are listed.

Figure 4.

A summary diagram (see text for method of summation) describing the complete parasitoid web sampled over the two years.

Patterns in the webs

The aim of this section is to look for invariant patterns present in the 11 parasitoid webs in Figs 1 and 2. Such patterns may reflect processes that structure the web and distinguish it from a random assemblage of species and trophic links. First, patterns are sought in the ratios of the number of species in the four categories of insects (that is hosts and primary, hyper- and mummy parasitoids), and in the ratios of the number of species in these different categories and the number of different types of trophic links. The motivation here is the often reported observation that equivalent ratios in traditional foodwebs are constant, or have a simple scale relationship with the size of the web (e.g. Cohen 1978; Briand & Cohen 1984; Cohen & Briand 1984; Sugihara, Schoenly & Trombla 1989; Cohen 1990; Pimm et al. 1991; Schoenly, Beaver & Heumier 1991; Martinez 1992, 1993, 1994; Goldwasser & Roughgarden 1993, 1997; Hall & Raffaelli 1993). These analyses just use presence/absence data and in the second part of this section we look for equivalent invariant patterns in the quantitative community data, using a statistical distribution to describe the species compositions. A complicating factor in analysing time series of foodwebs is the possible presence of temporal correlation, a problem that will be examined further in the Discussion section.

Species and link ratios

Six independent ratios were studied: (i) the numbers of aphid and primary parasitoid species; (ii) the numbers of aphid and hyperparasitoid species; (iii) the numbers of aphid and mummy parasitoid species; (iv) the numbers of links between aphids and primary parasitoids and between aphids and secondary parasitoids; (v) the numbers of links involving hyperparasitoids and mummy parasitoids; and (vi) the total number of species and the total number of links.

The estimated values of the six ratios (as binomial parameters) are given in Table 4 with the associated deviance statistic. For the first two comparisons, those involving the number of aphid species and the number of primary and hyperparasitoid species, the ratios are significantly more constant than expected under the null hypothesis, as is the ratio of links between hyperparasitoids and mummy parasitoids. The other comparisons do not meet the constancy criterion. The ratio of species to links is almost exactly 1 : 1 but is by far the most variable statistic calculated with a deviance close to the average expected under the null hypothesis. In only one case was there a major reduction in deviance when p was made a function of the total number of species or links in the web: the ratio of total links to total species increases with web size. The change in deviance (


) was 11·41 which is significant at P < 0·001. The residual deviance is very small (1·31) strongly suggesting less than binomial error variance, although the goodness-of-fit statistic is not asymptotically well defined in logistic regression with a continuous variate (McCullagh & Nelder 1989). The scale dependence of the ratio of species to links is shown in Fig. 5, where linkage density (number of links divided by number of species) is plotted against the number of species in the web.

Table 4.  Comparison of community statistics. The four columns are the parameter from the binomial distribution [X/(X + Y)]; the log-likelihood deviance statistic; the probability that the deviance statistic is significantly less than the binomial expectation; and the degrees of freedom (communities where X or Y were less than four were omitted)
Comparison (X and Y)Binomial parameterDeviance statisticProbabilityDegrees of freedom
(i) Primary parasitoid and aphid species0·450·680·00157
(ii) Hyperparasitoid and aphid species0·412·150·0248
(iii) Mummy parasitoid and aphid species0·362·610·245
(iv) Primary and secondary parasitoid links0·362·990·117
(v) Hyperparasitoid and mummy parasitoid links0·421·730·0277
(vi) Total links and total species0·4912·300·868
Figure 5.

Linkage density (the number of links divided by the number of species) in the 11 monthly foodwebs plotted against the number of species in the foodweb.

Quantitative patterns

Are there patterns in the quantitative data in the 11 webs? As described in the statistical methods sections, the Dirichlet distribution was used to summarize the species composition in the different webs. Examination of the residuals and the marginal distributions of species proportions (Fig. 6) suggests that the Dirichlet distribution provides a good description of the data.

Figure 6.

The fit of the Dirichlet distribution to the species proportion data. For each month and each species category (aphid, primary parasitoid, hyperparasitoid and mummy parasitoid) we tally the number of species whose proportional representation lies in the range 0–0·05, 0·05–0·1 and so on in categories of width 0·05. The histogram shows their frequency distribution across all month and categories. The distribution of proportional representations when the Dirichlet distribution (with parameter κ) applies is beta-distributed with parameters κ and s, where s is the number of species in the community. We plot three beta distributions with the common κ estimated from the data and with (a) s = 5, (b) s = 10 and (c) s = 15, to span the communities sampled in this study (the data were combined as there were too few communities of any particular size to analyse individually). The Dirichlet distribution is obtained by repeated sampling from a gamma distribution with shape parameter κ. The inset shows the probability density function [p(x)] of a gamma distribution with shape parameter 0·442, the estimated common κ (and scale parameter—which cannot be estimated from the data—arbitrarily set at one).

The fitted parameters for the four categories of insects are shown in Table 5. The parameters are in the range 0·3 < κ < 0·6, which implies that the underlying distribution of aphid abundance is sharply and monotonically decreasing from zero (see inset in Fig. 6). In no case was a significantly improved fit obtained by fitting separate parameters to the monthly foodweb data. Hence, while the densities of species vary markedly over the season, their relative proportions remain constant (or in terms of the underlying assumption of a gamma distribution of abundance, the shape parameter remains constant while the scale parameter varies). Fitting separate distributions to the four species categories fails to improve significantly the fit of the model, although the significance value (P = 0·10) suggests some heterogeneity. The latter is caused by a tendency for there to be a greater number of rare aphids, and a relatively more equitable distribution of mummy parasitoids.

Table 5.  The results of fitting the Dirichlet distribution to data for the proportion of species in different communities. The first and second columns are the value and standard error of the Dirichlet parameter (κ) fitted to data from each category of species and (final row) all data combined. The last three columns assess the significance of fitting models with separate parameters. In the case of the first four rows, the comparison is between a common parameter for all communities and separate parameters for each month. In the case of the last row, the comparison is between a common parameter for all species categories and separate parameters for aphids and each of the three types of parasitoids. The third column gives the χ2 statistic for the reduction in deviance resulting from fitting the more complicated model and the fourth column the reduction in the degrees of freedom (communities with only one member do not affect the fit). The last column gives the approximate significance of the improvement in fit provided by the fuller model
DataκS.E. (κ)χ2d.f.Probability
Primary parasitoids0·4710·0643·81100·96
Mummy parasitoids0·5970·10811·91100·22
All species0·4420·0316·2230·10

Because the value of the Dirichlet distribution in modelling biological data has received comparatively little study, a different approach was also used to confirm the similarity of species compositions amongst the four categories of species. The Alatalo–Hill evenness statistic (see statistical methods) was calculated for each type of insect in the 11 webs (except when only one species was present) and the results were compared using the Kruskal–Wallis nonparametric anova. The differences in the evenness statistic among the four types of species were not significant (P = 0·66).

Web complexity and indirect interactions

The aim of this section is to explore how different parts of the community are connected through shared parasitism by the three categories of parasitoids. The extent of shared parasitism will indicate the potential dynamic complexity of the system, and the degree to which aphid species on different host plants may interact indirectly via shared natural enemies. The chief tool used were predator-overlap graphs (or in this context parasitoid-overlap graphs), in which vertices representing different aphid species were linked by lines if they were both attacked by a species of the same category of parasitoid (Cohen 1978; Sugihara 1984). Both traditional predator-overlap graphs, which utilize only qualitative data, and a new type of quantitative parasitoid-overlap diagram, which allows assessment of the potential strength of indirect effects, were used.

Parasitoid-overlap graphs

The three traditional parasitoid-overlap graphs for the different parasitoid types are shown in Fig. 7 together with the graph for all parasitoids. Three species of aphid share no primary parasitoids with other aphid species. Of these, two species (codes 1 and 19) are attacked by monophagous parasitoids, while a third species (21) was rarely parasitized and all the mummies that were collected produced secondary parasitoids (no vertex is drawn for this species). Five species of aphid were not attacked by hyperparasitoids and from three species no mummy parasitoids were recorded. However, all aphids from which parasitoids were reared were connected by shared parasitism of some kind with at least one other aphid.

Figure 7.

Parasitoid-overlap graphs. The vertices represent the aphids from which at least one species of parasitoid were reared. Vertices are joined by edges when the aphids share at least one species of parasitoid. The four graphs show linkages by shared primary parasitoids, hyperparasitoids, mummy parasitoids, and by all parasitoids together.

The extent of shared parasitism differed among the three types of parasitoid. Parasitoids were reared from 15 species of aphid and hence there is a maximum of 105 pairs of aphid (i.e. the binomial coefficient (2Ümh-920Ý15) that may be linked by parasitoids. The number of linked pairs observed was 34 (32%), 26 (25%) and 57 (54%) for primary, hyper- and mummy parasitoids, respectively. Thus the mummy parasitoids, despite having the fewest species, were the most important group in connecting the web. Considering all groups together, 64 aphid pairs (61%) share parasitoids. If the eight other aphid species, from which no parasitoids were reared, are also included, then 25% of the possible species pairs are linked by parasitoids.

The mean numbers of links per aphid for the different types of parasitoids are shown in Table 6, with the observed variance and the variance calculated from eqn 5. In all cases the variance is considerably higher than expected under the null hypothesis (randomization test: P < 0·001). The rejection of the null hypothesis might be interpreted as indicating ‘community structure’, although, as discussed below, the presence of rare species of aphid attacked by few species of parasitoid is sufficient to lead to rejection.

Table 6.  Distribution of the links in the predator-overlap graph. The null model variance is based on the assumption that links are randomly distributed amongst species (see text)
 Total linksMean links per aphidObserved varianceNull model variance
Primary parasitoids344·5311·722·68
Mummy parasitoids577·6116·773·04
All parasitoids648·5312·252·92

Quantitative parasitoid-overlap diagrams

In the statistical methods section we describe the construction of quantitative parasitoid-overlap diagrams which illustrate the potential importance of one host as a source of parasitoids attacking another. Figure 8 shows these diagrams for the three categories of parasitoid. Different aphid species are represented as before by numbered vertices except aphids are now depicted by circles of different diameters where the area of the circles indicates the relative abundance of the aphid species. Pairs of aphids are linked if the two species share a common parasitoid. The links are now of variable thickness, with the width of the link at species i being proportional to dij (the fraction of parasitoids of species i whose parents are expected to have developed on species j). The magnitude of the dii values, the self loops, are indicated by the extent to which the vertices are shaded black. If most individuals of the parasitoid species attacking an aphid develop on that aphid species (dii→1), then the vertex is shaded largely black; if most individuals of those parasitoid species develop on other aphid hosts (dii→0), then the vertex is largely white.

Figure 8.

Quantitative parasitoid-overlap diagrams. As in Fig. 7, the vertices represent aphids and they are linked if the two species share a parasitoid. The size of each vertex is proportional to aphid abundance. As is more fully explained in the text, the extent to which the vertices are coloured black, and the strength and symmetry of the links, measure the importance of each aphid species as a source of parasitoids attacking other species of aphids and itself. The three diagrams represent primary parasitoids (top left), hyperparasitoids (top right) and mummy parasitoids (bottom).

To what degree do aphid species act as sources of their own parasitoids (dii large) and hence are likely to show dynamic independence? Figure 8a shows that most primary parasitoids attacking common species of aphids would have developed on the same species of host. Although there are some links between common aphids, they tend to be relatively weak. By far the strongest indirect links are from common to rare aphid species (and so, as a corollary, there are very few, strong symmetric links). The quantitative overlap diagrams thus suggest that the aphid–primary parasitoid dynamics of the relatively common species are not strongly linked, but that the common species are likely to influence strongly the dynamics of the rarer species in the community.

In comparison to Fig. 8a, the strengths of the self loop terms for the common aphid species in the hyperparasitoid diagram (Fig. 8b) are weaker, although the same general pattern emerges: common aphids are largely attacked by wasps which would have developed on the same species of host, strong symmetric links are rare, and the parasitoid complex attacking less common aphid species tends to be largely composed of wasps which would have developed on other hosts. The picture from the mummy parasitoid diagram (Fig. 8c) is very different. As the parasitoid overlap graph showed, the density of linkages is much higher. In comparison with the other parasitoid types, the communities attacking the common aphid species tend not to be dominated by wasps which would have developed on that host. Aphid species 1 (Capitophorus carduinis) is the only species with dii > 0·5. While there is still a marked tendency for the strongest indirect effects to be from common to less common aphids, even the more abundant aphids (with the exception of C. carduinis) are linked to other species by potentially strong links.

Parasitoid species per host and parasitoid host range

In this section we look for statistical correlates of the probability that an aphid species avoids parasitism altogether, and try to explain the number of parasitoid species reared from those aphid species that are attacked. We also analyse the number of host species attacked by different species and categories of parasitoids. The analyses described here are thus in the tradition of Hawkins's (1994) investigations of parasitoid communities, although using data from a single community rather than a range of literature sources.

Numbers of parasitoid species per host species

Of the 23 aphid species found in this study, no parasitoids at all were reared from eight species. The total numbers of primary and secondary parasitoid species found on the remaining 15 aphid taxa varied from 1 to 24 (Table 2). We sought to explain this variation in terms of a series of explanatory factors: (i) the logarithm of total aphid density (the sum of the individual monthly totals); (ii) the temporal range of the aphid (in how many months it was recorded); (iii) aphid size (mean size of wingless adults in Heie 1980, 1982, 1986, 1992, 1994, 1995); and (iv) the presence of attending ants. For secondary parasitoids, a further variable was included, (v) the logarithm of total primary parasitoid density.

Although the mean log abundance of aphid species from which parasitoids were reared (2·17; 0·56 S.E.) was greater than that of unparasitized aphids (1·09; 0·89 SE), the difference was not significant (t-test, P = 0·30). The temporal range of parasitized aphid species (4·73; 0·56 SE) was also greater than unparasitized aphids (3·25, 0·75 SE), although again the difference was not significant (Mann–Whitney test, P = 0·12; a nonparametric test was used because of violations of the assumption of normality). The average size of the two categories of aphid species, their probability of ant attendance, were both very similar. We have thus had little success in understanding why some aphid species escaped parasitism.

We analysed separately the number of primary parasitoid, hyperparasitoid, and mummy parasitoid species recorded from the 15 species of aphid from which at least some parasitoids were reared. Preliminary analysis of the explanatory variables revealed significant correlations between aphid density and temporal range, and aphid and primary parasitoid density. Ant-attended aphid species were significantly smaller than those not ant-attended (P < 0·005). The data were analysed using stepwise analysis of deviance using generalized linear modelling techniques (implemented in the statistical package GLIM; specifically a Poisson error distribution was assumed but correction was made for overdispersion by allowing the variance to be proportional rather than equal to the mean; McCullagh & Nelder 1989). The significance values for the different factors refer to the change in deviance (asymptotically χ2 distributed) after dropping the factor from the minimally sufficient model.

The number of primary parasitoid species was best described by a model containing the single explanatory variable ant attendance (


 ≈ 6·40, P < 0·025). Fewer species of primary parasitoid were reared from ant-attended species. The number of hyperparasitoid species were best described by a model including ant attendance (


 ≈ 10·01, P < 0·005) and temporal range (


 ≈ 10·88, P < 0·001), with more species being found on aphids that were not ant attended and had a wide temporal range. Finally, none of the explanatory variables that were explored had a significant effect on determining the numbers of mummy parasitoid species found on different aphid taxa.

Parasitoid host range

The mean number of aphid species attacked by different primary parasitoid species was 2·1, by hyperparasitoids 2·3, and by mummy parasitoids 4·9. The differences were highly significant (


 ≈ 16·0, P < 0·001) as a result of the larger number of aphid species attacked by mummy parasitoids (the difference between primary parasitoids and hyperparasitoids was not significant). We explored the influence of (i) parasitoid density and (ii) the temporal range of the parasitoid (the number of months in which it was recorded) on the number of aphid species attacked by members of the three classes of parasitoid. Stepwise analysis of deviance in GLIM was again used as described in the analysis of the number of parasitoid species per aphid. The host range of primary parasitoids was significantly influenced by parasitoid density (


 ≈ 4·55, P < 0·05) and temporal range (


 ≈ 6·24, P < 0·025), species with a wider temporal range attacking more hosts and, surprisingly, rarer species having a broader host range. Neither of the two explanatory variables were significantly correlated with hyperparasitoid host range. Finally, parasitoid density and temporal range together had a highly significant effect on the number of aphid species attacked by different species of mummy parasitoids (


 ≈ 10·41, P < 0·01) but because the two variables were highly correlated (ρ = 0·87) it was not possible to decide their separate contributions.


The parasitoid web described in this paper contains 71 species of aphid and parasitoid, and 100 species exactly if the food plants of the aphids and attending ants are also included. Compared with the majority of foodwebs included in Cohen's (1989) database (ECOWeB) of over 200 foodwebs, and many of the webs collected more recently, it is unusual in its size, in being fully quantitative, in virtually all taxa being distinguished at the species level, in being collected at a single site, and in being resolved into a temporal series of individual webs. On the negative side, it includes aphids and their parasitoids but not other predators, pathogens and competitors of aphids, nor predators or pathogens of the parasitoids; and the extent to which trophic links between secondary and primary parasitoids have been resolved is poor.

Pimm et al. (1991), Cohen et al. (1990, 1993) and several of the authors in Polis & Winemiller (1996) have all argued that for foodweb studies to progress, new ‘purpose-built’ webs are required, collected using a common set of techniques. Their pleas have already been partially answered by the series of new high quality webs published in the last 10 years (e.g. Warren 1989; Winemiller 1990; Hall & Raffaelli 1991; Martinez 1991; Polis 1991; Havens 1992; Goldwasser & Roughgarden 1993, 1997; Deb 1995; Tavares-Cromar & Williams 1996). For all but the simplest community there is a trade-off between the inclusiveness of the foodweb, and the level of quantification and species resolution that can be achieved. Most of the new webs are at the inclusiveness end of this spectrum. We believe that the construction of parasitoid webs is a means of obtaining highly resolved and quantitative foodwebs, albeit of only a component of a much larger community, using a methodology that can be applied uniformly to different natural systems.

This is the second quantitative parasitoid web to be published (see Memmott, Godfray & Gauld 1994) and we are currently writing up two others, and are aware of at least one more nearing publication. It appears premature at this stage to try to find common patterns in the two published webs, or to look for similarities between the structure of our quantitative parasitoid webs and the few published qualitative or semiquantitative parasitoid webs (Askew 1961; Askew & Shaw 1974; Rejmánek & Starý 1979; Hawkins & Goeden 1984; Dawah, Hawkins & Claridge 1995), or indeed between our webs and traditional foodwebs [very few of which include any data on parasites, for an exception see Huxham, Raffaelli & Pike (1995) and Huxham, Beaney & Raffaelli (1996)]. Instead, in the rest of the Discussion we will first critically review our data and list some of its shortcomings, and then discuss the three specific sets of questions about parasitoid community structure that we have asked of our data.

Criticism of the parasitoid webs

There are a number of possible sources of error in the construction of the foodwebs described here.

Sampling bias

The accuracy of the webs relies on being able to sample, without bias, the 71 species of aphids and parasitoids in the community. Were some species to be harder to locate or apt to be overlooked, they would be under-represented in the foodwebs. We believe the risk of under-sampling aphid species to be relatively small. Aphids typically feed externally on the plant and are easy to spot. We knew in advance from the literature the species that might occur in the study site and hence were able to check for the presence of species with more cryptic biology. By excluding root aphids from the study, the difficulties of sampling subterranean herbivores were avoided.

Assessing the risk of bias in the parasitoid community is harder. It is well known that some species of parasitoid cause their hosts to wander after parasitism and mummify away from the aphid colony or aphid food plant (Brodeur & McNeil 1989, 1992; Müller, Völkl & Godfray 1997). This might cause underestimation of parasitism, and lead to biased estimates of community structure when one parasitoid species causes its host to move further than another. When assessing parasitism rates and species composition, we endeavoured to search not only the whole host plant but also the surrounding vegetation for mummified aphids, but we cannot exclude some bias in community composition caused by biased sampling.

Parasitism was assessed by collecting mummified aphids. The category of secondary parasitoids we call hyperparasitoids attack the primary parasitoids prior to mummification but the mummy parasitoids in this study attack the mummies directly. There is thus a possibility that we have underestimated mummy secondary parasitism by collecting mummies prior to their being attacked. Mummies containing secondary parasitoids are present in the field, and hence liable to be collected, for a longer period than primary parasitoids. This may lead to an overestimation of secondary parasitoid densities. Currently we are attempting to collect the quantitative data necessary to assess these potential sources of bias.

Sampling error

The foodwebs do not include an explicit estimate of the sampling errors in the density estimates, nor an estimation of the likelihood that there are further species missed by the sampling programme (and what their densities might be). However, there is an implicit estimate. It is highly unlikely that common species are missed through sampling error (although conceivably overlooked for the reasons discussed above) and their estimated densities will be relatively accurate. The densities of the rarer species will be less accurately estimated, and some may be overlooked. Compared with a binary connectance foodweb, a quantitative foodweb thus provides some information about the accuracy of numbers of species and links. Nevertheless, it is impossible to sample 71 species simultaneously with the precision of a single or a few species study and the possibility of sampling errors influencing foodweb statistics must always be borne in mind.

Identification of trophic links

It is relatively straightforward to associate a primary or secondary parasitoid with an aphid host. For only two groups of hosts was it necessary to work with ‘combined’ host taxa because mummies could not be identified in the field (see Results). Most morphological features of the living aphid are preserved by mummification and, in combination with information from the host plant, we believe few errors were made in host aphid identification. Where one aphid is attacked by several primary parasitoids, linking secondary parasitoids with specific primary parasitoids is much more difficult. Typically, mummy structure provides information about the genus of the primary parasitoid, but not the species. Two options were available to deal with this uncertainty. The first was to assume no within-genus secondary parasitoid specialization and to use the relative proportions of potential hosts in estimating the strength of different primary/secondary parasitoid links. The second option was to work chiefly with the strengths of the secondary parasitoid/aphid links. The second option was chosen because it involved no secondary interpretation or averaging on our part, although we acknowledge that it results in the trophic links in the parasitoid web not being fully resolved.

There are some more subtle questions of trophic links in this community. Most of the mummy parasitoids are idiobionts (Askew & Shaw 1986), the female killing or permanently paralysing the inhabitant of the aphid mummy at oviposition. Although the details will vary from species to species, the majority of mummy parasitoids are likely to be able to develop on whatever parasitoid is in the mummy, including hyperparasitoids and other mummy parasitoids of their own or another species. There are thus almost certainly trophic loops within the guild of mummy parasitoids, including self loops. Polis & Strong (1996) have argued against the simple use of the concept of trophic levels in foodweb studies. In our web, aphids, primary parasitoids and hyperparasitoids form a straightforward series of trophic levels, but mummy parasitoids do not feed at a single level as they are able to develop on primary parasitoids, hyperparasitoids and other mummy parasitoids.

Other natural enemies

For reasons discussed in the Introduction, this study is restricted to aphids and their parasitoids, while other natural enemies—predators and pathogens—are not considered. Clearly the dynamics of the aphid–parasitoid community cannot be considered in isolation from the other natural enemies (see, for example Müller & Godfray 1997) and we are currently extending the foodweb to include these other components. Other natural enemies will influence aphid–parasitoid community structure in several ways. First, predators will directly consume parasitoids, either before or after mummification. In some circumstances mummified aphids may be preferentially consumed, leading to an underestimation of parasitism. It is possible that predators may selectively attack certain parasitoid mummies and hence distort community composition, although we know of no definite evidence for this. We do know that some aphid parasitoids spend less time ovipositing in aphid colonies where predators are present (Taylor, Müller & Godfray 1998). However, the major impact of other natural enemies on the parasitoid community is likely to be in determining the numbers of aphids that are available for parasitoid attack.


Despite their economic importance, many of the parasitoids of aphids are poorly known taxonomically. In particular, the braconid subfamily Aphidiinae, and the charipid subfamily Alloxystinae, respectively, major primary parasitoids and hyperparasitoids of aphids, present a series of unsolved problems. The reason for these difficulties is that both groups consist of exceptionally small insects, derived from larger relatives. The evolution of small size in parasitic Hymenoptera is frequently accompanied by structural simplification and a loss of useful morphological character states. Our treatment of the taxonomic problems we encountered is described in the Appendix. Further research should provide names for some of the species simply given codes in Table 2, although this will not influence community structure. We have checked the species boundaries of the more common primary parasitoids using molecular techniques (see Appendix), but it is possible that the species concepts of one or two of the rarer species of parasitoids, particularly the Alloxystinae, may change with further study.

Human interference

Constructing the foodwebs involved the collection of mummified aphids for rearing, as well as a certain amount of trampling and disturbance, although this was kept to a minimum. For most species, the fraction of parasitoids removed from the community is only a small proportion of the numbers estimated to be present in the site, but the possibility cannot be discounted that the sampling may have had some influence on community structure.


For logistic reasons it was only possible to sample one community—the sampling and rearing required at least two people full time during the field season. Clearly this affects the inferences that can be drawn from these data on general issues of aphid–parasitoid community structure.

Patterns in the webs

We compared the properties of the 11 separate monthly webs looking for constancy and patterns in the ratios of the different groups of species and trophic links, and in the relative proportion of different species. The greatest problem with the approach adopted by this study is that the 11 webs were treated as though they were statistically independent. In the past, the possibility of nonindependence of webs has largely been ignored but Bengsson 1994) has recently highlighted the problems that can arise. Faced with nonindependence, one can either: (i) try to model the statistical autocorrelations within the data; (ii) select for analysis a subset of data including only independent webs; or (iii) ignore the nonindependence at the analysis stage but interpret the data carefully. With a limited number of webs, the first two options are not practical and we have taken the third approach.

The ratios of the number of species of primary parasitoid, hyperparasitoid and aphid were remarkably constant across the 11 webs, with significantly sub-binomial variance. Part of this low variance may be explained by nonindependence of the webs: species present in one month are likely to be present in the next. However, the constancy is also likely to reflect structural properties of the community, with different hosts species tending to be attacked by similar numbers of the two more host-specific classes of parasitoid. The ratios of the numbers of links involving hyperparasitoids and mummy parasitoids also had subbinomial variance, an observation for which we do not have an explanation. Finally, the ratio of total species to total links was very variable, with linkage density increasing with web size. Linkage density has been observed to increase with web size in a number of traditional foodwebs (Cohen et al. 1990), although the linkage density observed here (≈ 1) is about half that observed in the ECOWeB data. However, as discussed above, the number of trophic links involving secondary parasitoids are underestimated in the web and hence comparisons with other systems need to be made cautiously.

Wilson (1996) has criticized previous claims of constant predator–prey ratios on the grounds that they lack a rigorous null hypothesis. His null model approach is similar (possibly identical) to the approach adopted here, although the test statistics employed are different. Wilson believes that constancy requires less than binomial variance, something we found in several comparisons but which he failed to discover in the communities he examined. However, a strong argument could be made that natural communities are likely to be assembled in a way that, in the absence of biological interactions, would give rise to greater than binomial variance: overdispersion is a very prevalent feature of logistic biological analyses. There might thus be a danger in treating communities with binomial species ratios as lacking biologically interesting assembly rules, although obviously it is difficult to choose a specific level of overdispersion to act as a new null hypothesis. A better approach is probably to model statistically the variance in species numbers and to look for patterns across different communities.

Using the quantitative data, it was found that the relative proportions of species in the different communities were described by a Dirichlet distribution, and that the explanatory power of the model was not increased by fitting individual parameters to the monthly webs. The Dirichlet distribution can be obtained from the expected composition of independent variables sampled from a gamma distribution. The results obtained by the present study imply that aphid population densities are heavily skewed, with many rare species and relatively few common species, and that the shape of this distribution, although not its mean, was constant over the course of the sampling. The same distribution described the relative abundance of the parasitoids as well as the aphids, but this is perhaps less surprising as the density of aphids will be a strong determinant of the density of their parasitoids. The estimated parameter of the Dirichlet distribution was 0·44, which was intermediate between that expected from a logarithmic (parameter → 0) or broken stick (parameter = 1) model (Engen 1978). Naeem & Hawkins (1994) have compared the distribution of parasitoid abundances recorded from one host species against a range of niche partitioning models, an approach that is difficult to apply to data in the present study because of the abundance of secondary parasitoids.

Web complexity and indirect interactions

We recorded 23 taxa of aphid (i.e. 21 species and two species pairs, see Results) from our study site over the two years of sampling. There are thus 253 possible interactions, direct or indirect, between pairs of aphid species. Direct interactions via exploitation competition require that the aphids share at least one host plant and from Table 1 it can be seen that this occurs for only 19 pairs of species. Of the 23 aphid species, 15 are attacked by parasitoids and hence there is a maximum of 105 pairs of indirect interactions via shared primary or secondary parasitoids. Evidence of shared parasitism was found in 64 of these 105 pairs, although primary parasitoids and hyperparasitoids linked only 34 and 26 species, respectively, with mummy parasitoids being the most important, linking 57 pairs. Parasitoids are only one component of the aphid natural enemy complex and further indirect interactions between different species will occur through shared pathogens and, perhaps most importantly of all, shared predators.

The distribution of indirect interactions is described by the parasitoid overlap graph (Fig. 7). For all three types of parasitoids, the distribution of links was found to be more variable than expected from a simple statistical null model. One explanation for this pattern is that the community is divided into strongly connected species groups or compartments that are only weakly connected with each other. However, the quantitative data suggest that the variance is largely a result of differences in the abundance of species, with commoner aphids supporting more parasitoids, particularly mummy parasitoids, and hence being more likely to share parasitoids with other common aphids.

Using the quantitative data we can go beyond merely stating that two aphids share natural enemies to assessing the strength and directionality of the indirect interaction. To do this a statistic dij was calculated, which measures the role of species j as a source of parasitoids that attack species i. If nearly all of the parasitoids of species i are host specific, then dii→1 and all dij→0. If host species i is attacked by a parasitoid species, the vast majority of the population of which develops on host species j, then dij→1. The accuracy of the statistic in assessing the scope for indirect interactions rests on the assumption that indistinguishable host races of parasitoids do not occur in the field, and that the parasitoid populations are well mixed at the level of the field. The different dijs can be used to construct a new type of predator overlap graph with weighted and directed links between species pairs, as well as providing an assessment of the importance of a species as a source of its own parasitoids. In drawing the diagrams, information was also included about the relative abundance of different aphid species (Fig. 8).

Examination of the primary parasitoid and hyperparasitoid diagrams shows that the common aphids are largely attacked by specific parasitoids in these categories and that there are few strong interactions between the more abundant aphids. However, the common aphids can act as a source of the parasitoid species that constitute the majority of individuals that attack some of the rarer species of aphids. Thus the most important indirect interaction for these groups of parasitoids is likely to be the asymmetric effect of common on rare species. For mummy parasitoids, the picture is different. Not only does this category of parasitoid link more aphid pairs, but the strength of the indirect interactions are stronger and fewer species act as the major source of their own parasitoids (i.e. fewer high diis).

What do these results suggest about the dynamic interactions among aphid parasitoids? Again, it must be stressed that static foodwebs can only suggest the role of different dynamic processes and suggest experiments; they cannot verify their presence (Paine 1988, 1992; Menge 1995). We hypothesize that the community consists of a series of relatively tightly linked interactions between aphids, primary parasitoids and hyperparasitoids centred on different host plant species (or closely related groups of plant species), which we shall call an aphid–parasitoid–hyperparasitoid (APH) group. Sometimes, more than one aphid, primary parasitoid or hyperparasitoid may be present, but links with other APH groups are weak. The primary parasitoids and hyperparasitoids in the APH groups are attacked by mummy parasitoids, which are far more catholic in their host preferences and provide a dynamic linkage between different APH groups. Considering the low parasitism rates attained by most primary parasitoids, and the high rate of hyperparasitism, we believe it likely that secondary parasitoids regulate primary parasitoids but that primary parasitoids do not regulate aphids (see also Mackauer & Völkl 1993). We have argued that different APH groups are linked together at the top of the trophic chain by mummy parasitoids; we also believe that shared predators link the APH groups at the aphid level (Godfray & Müller 1999; Müller & Godfray 1997). An important question that can be answered by experimentation is whether mummy parasitoids respond to an increase in the abundance of primary parasitoids in an APH group by density-dependent attack (switching, Murdoch 1969). Such a response could be an important factor stabilizing the host–parasitoid dynamics within the individual APH groups.

Parasitoid species per host and parasitoid host range

As discussed in the Introduction, there have been several studies that have looked for correlates of the number of parasitoid species attacking a particular species of host, and the size of different parasitoids species’ host ranges (reviewed by Godfray 1994; Hawkins 1994). Typically, these studies have been based on literature surveys and include a wide diversity of species, although sampled in many different ways. The work reported here contrasts with these in that it considers only a limited number of species, but sampled in a uniform manner. One problem that all of these studies share is that they treat species as statistically independent data points, ignoring their phylogenetic relationships (Harvey & Pagel 1991). There is an urgent need to extend the comparative method to deal with interacting sets of species, and for better phylogenies of insect parasitoids and their hosts. For the moment, the results of statistical analyses that do not take phylogenies into account must be treated with caution.

We were unable to find a reason why eight of the aphid taxa were not attacked by any species of parasitoid. A possible clue is given by Cavariella pastinacae, which was common in one year but absent in the other. We suspect that such species, which have highly variable population dynamics, may be able to escape parasitoid attack. We are continuing collecting data from the study site and hope to be able to test this hypothesis with a longer time series for analysis.

Aphids that are attended by ants are attacked by fewer species of primary parasitoids and hyperparasitoids, although mummy parasitoid numbers are not affected. Primary parasitoids and hyperparasitoids attack the living aphid and hence have to avoid any attending ants, while mummy parasitoids attack the mummified aphids which are not attended, although they may still lie within the aphid colony. Specialized parasitoids of ant-attended aphids have evolved a variety of adaptations to avoid ant attack and normally seem not to be recognized by the ants (Völkl 1992, 1994; Liepert & Dettner 1993, 1996; Völkl & Mackauer 1993; Völkl, Hübner & Dettner 1994; Hübner & Völkl 1996). However, ant attendance may restrict attack by less specialized species. One caveat to this result is that ant attendance is particularly common among aphids in the genera Aphis and Brachycaudus and hence the possibility of other phylogenetic correlates of parasitoid number must be considered.

The average host range of species of mummy parasitoids (4·9) is over twice that of primary parasitoids (2·1) or hyperparasitoids (2·3). Mummy parasitoids are idiobionts whose larvae develop on dead or permanently paralysed host tissue, while primary parasitoids and hyperparasitoids are koinobionts, which attack living hosts and have to resist the host immune response. As has been widely recognized (Askew & Shaw 1986; Pschorn-Walcher & Altenhofer 1989; Sato 1990; Sheehan & Hawkins 1991; Godfray 1994), koinobionts tend to have narrower host ranges than idiobionts, probably because their lifestyle requires greater adaptation to the host.

We found that host range in primary parasitoids was positively correlated with temporal range and negatively correlated with parasitoid density. The first relationship is expected: the longer time a parasitoid is active, the more potential hosts it will encounter. The second relationship is much less expected and seems to result from members of the genus Praon being both uncommon and relatively polyphagous (an example of an uncontrolled phylogenetic effect). None of the explanatory variables explored correlated with hyperparasitoid host range. Finally, there was a significant relationship between mummy parasitoid host range and a composite variable combining parasitoid density and temporal range, the two being too highly correlated for their individual effects to be distinguished. Again, common species active for a greater part of the year are likely to encounter more potential hosts; an effect that will be magnified if mummy parasitoids are relatively nonspecific.


We thank Richard Cooke, Peter Mayhew, Nathalie Tristem, Nigel Varndell and Frank van Veen for help in the field; Cliff Brookes, Henk Evenhuis, Nigel Fergusson, Ulf Gärdenfors, Hugh Loxdale, John Noyes, Wilf Powell, Frank van Veen, and Wolfgang Völkl for taxonomic and molecular help and advice; and Joel Cohen, Mike Hassell, Brad Hawkins, Dave Hodgeson, Lex Kraaijeveld, John Lawton, Stuart Pimm, Wilf Powell, Wolfgang Völkl and an anonymous statistical referee for discussions on parasitoid community and food web ecology. This work was funded by the Natural Environment Research Council (the taxonomy component as part of the NERC Initiative in Taxonomy Research and Training) and by the Swiss National Science Foundation (823A-040145).

Received 3 June 1997; revision received 3 July 1998


Parasitoid taxonomy

Aphidiinae (by RB)

Identification was based on Mackauer (1959), Starý (1966), Gärdenfors (1986) and Pungerl (1986) and proved straightforward for all genera except the notoriously difficult Aphidius. For this genus, species concepts were checked using cellulose acetate allozyme electrophoresis (following protocols of Richardson, Baverstock & Adams 1986). The three species rubi, urticae and eadyi were reared from different hosts but could not be distinguished morphologically. Current practice is to synonymise rubi with urticae but to treat eadyi as distinct (Starý 1973; Starý, González & Hall 1980). Examination of 50 individuals from the study site of the three species showed fixed allozyme differences in all three species at the phosphoglucomutase (PGM) and malic enzyme (ME) loci, and a fixed difference between rubi and the other two species at the isocitrate dehydrogenase (IDH) locus, leading us to treat them as separate species. Aphidius microlophii used to be regarded as a host race of ervi attacking Microlophium carnosum. Cross-mating, host-transfer, pheromone response and morphological studies all indicate a degree of biological isolation (Pennacchio & Tremblay 1986; Powell & Wright 1988). Atanassova et al. (1998) found a fixed mobility difference in the enzyme PGM among a large number of specimens collected at Rothamsted Experimental Station, Harpenden, UK. This difference was also found among a smaller sample from our field site, except for a single microlophii female, which had both alleles (sample composed of 13 ervi and 70 microlophii). It therefore appears that the gene flow between microlophii and ervi is low and that they should be treated as separate ecological entities. Phosphoglucomutase also clearly distinguished the morphologically similar picipes. A number of species attacking grass aphids have been described which are very similar to the common rhopalosiphi (frumentarius, Latteur & Rassel 1979; uzbekistanicus, Pennacchio & Höller 1990) and a species complex may be involved. The specimens we examined electrophoretically were homogeneous and similar to specimens from a rhopalosiphi culture maintained at Rothamsted. Electrophoresis showed that the common parasitoid of Capitophorus carduinis, called here Aphidius B, was composed of a single species. It may be the same as matricariae, which is recorded chiefly from aphids on Asteraceae including C. carduinis (Starý 1973), but many specimens lacked the three-segmented maxillary palp regarded as characteristic by Pungerl (1986), and hence for the moment we refer to it by code. Finally, we reared a very few specimens of a species we refer to as Aphidius E. This belongs to a group of morphologically indistinguishable species with 17-segmented antennae (Pungerl 1986). Further material is required to clarify the identity of this taxon.

Alloxystinae (by HCJG)cm

There are currently two very different views of species boundaries within the genera of aphid hyperparasitoids Alloxysta & Phaenoglyphis. Fergusson (1986) reviewed the British fauna and recognized 16 species, many of which were morphologically very variable and with wide host ranges. A much narrower view of species boundaries has been taken by continental authors led by Evenhuis (1985 and included references to a long series of papers). Part of the difference is because Fergusson chiefly worked with swept material and Evenhuis with reared material. Unfortunately there is no general treatment of Alloxystinae by the continental school. Our rearings of Alloxystinae are probably the largest from a single locality, which gives us an unparalleled opportunity to assess variation seasonally and across hosts. We strongly agree with the view of Evenhuis of species boundaries and record more species from Rush Meadow (18) than Fergusson lists for Great Britain. Species were separated using morphological characters, including a number used by Evenhuis but not by Fergusson. We have used Latin names for species that have been checked by Evenhuis, who has examined most of the extant European types. Although some of the names are the same as those in Fergusson, in only three cases do we believe our species concepts to be the same (A. pleuralis, P. villosa and P. xanthochroa). A number of taxa cannot yet be assigned to named species, and at least a couple are likely to be undescribed. The definitions of one or two of the rarer species, for which we have very few specimens, may require some revision. We have begun some molecular studies to test the species concepts used here. DNA sequencing of the nuclear ITS2 region in specimens of A. victrix from different hosts and of Aphidius sp. v2 from Capitophorus carduinis (two species that would key to victrix in Fergusson 1986) strongly supports their separate status (F. van Veen & HCJG, unpublished results).

Other groups (by HCJG)cm

The taxonomy of the other parasitoids reared was less problematic. Aphelinus were identified using Graham (1976), Pteromalidae using de Graham 1969), Dendrocerus using Fergusson (1980) and Syrphophagus using notes kindly provided by J.M. Noyes (Natural History Museum).