Creating novel food webs on introduced Australian acacias: indirect effects of galling biological control agents


  • Ruan Veldtman,

    Corresponding author
    1. Centre for Invasion Biology, Department of Botany and Zoology, Stellenbosch University, Stellenbosch, South Africa
    2. Applied Biodiversity Research, South African National Biodiversity Institute, Claremont, Cape Town, South Africa
    3. Department of Conservation Ecology and Entomology, Stellenbosch University, Stellenbosch, South Africa
    Search for more papers by this author
  • Thomas F. Lado,

    1. Centre for Invasion Biology, Department of Botany and Zoology, Stellenbosch University, Stellenbosch, South Africa
    2. Applied Biodiversity Research, South African National Biodiversity Institute, Claremont, Cape Town, South Africa
    3. Department of Zoology and Entomology, University of Cape Town, Cape Town, South Africa
    Search for more papers by this author
  • Antoinette Botes,

    1. CapeNature, Scientific Services, Jonkershoek, South Africa
    Search for more papers by this author
  • Şerban Procheş,

    1. School of Environmental Sciences, University of KwaZulu-Natal, Durban, South Africa
    Search for more papers by this author
  • Alicia E. Timm,

    1. Department of Conservation Ecology and Entomology, Stellenbosch University, Stellenbosch, South Africa
    Search for more papers by this author
  • Henk Geertsema,

    1. Department of Conservation Ecology and Entomology, Stellenbosch University, Stellenbosch, South Africa
    Search for more papers by this author
  • Steven L. Chown

    1. Centre for Invasion Biology, Department of Botany and Zoology, Stellenbosch University, Stellenbosch, South Africa
    Search for more papers by this author

Ruan Veldtman, South African National Biodiversity Institute, Kirstenbosch Research Centre, Private Bag X7, Claremont 7735, South Africa. E-mail:


Aim  The use of host-specific biological control agents is widely considered an effective option for the management of invasive alien plant species. However, the formation of novel associations between released biological control agents and indigenous species poses risks. Here, we investigate whether native food webs associated with two galling biological control agents on Acacia longifolia and A. saligna are similar to those found in their introduced range.

Location  Gall inhabitants recorded from South Africa and Australia.

Methods  Non-targeted insects were collected from galls in introduced ranges for comparisons to that of the agents’ native ranges.

Results  We find that two host plant-specific galling biological control agents accumulate food web links with higher trophic levels in their introduced range that are similar in number, taxonomic/phylogenetic pattern and guild composition to those in their native range. Bray-Curtis percentage similarity between native (Australia) and novel (South Africa) food webs was 30–50% and 50–75% at the family and superfamily taxonomic level, respectively, and 45–50% if considering shared phylogenetic diversity.

Main conclusions Trichilogaster acaciaelongifoliae and Uromycladium tepperianum accumulated food webs in South Africa that are strikingly similar in complexity and structure to those that occur in their native ranges. This indicates that the structure of food webs in the introduced range could be predicted by studying food webs in the native range of a biological control agent, potentially paving the way for more effective risk assessment of weed biological control.


The release of biological control agents to suppress invasive alien organisms (especially plants) has been lauded as both successful and safe (Cruttwell McFadyen, 1998; Hoddle, 2004; Moran et al., 2005). Although cases exist that illustrate negative effects of agents on indigenous non-target host plants (Louda et al., 1997), it has been argued that these negative effects could have been predicted from adequate host-specificity trials (Pemberton, 2000). However, because studies on the long-term effects of agent release are the exception rather than the rule in classical biological control (Gillespie et al., 2006), the risk of potential long-term ecological impacts (especially the indirect effects on higher trophic levels, see Pearson & Callaway, 2005, 2008; Carvalheiro et al., 2008) is seldom determined, and consequences thereof remain largely unknown (Louda et al., 1997; Pearson & Callaway, 2003; Messing & Wright, 2006; Thomas & Reid, 2007). This is disquieting, given that continued monitoring is vital for identifying both direct and indirect non-target effects of biological control, and has been identified as a significant component of understanding the risks posed by agent release, a crucial part of any cost-benefit assessment process (Howarth, 1991; McGeoch & Wossler, 2000; Louda et al., 2003; Delfosse, 2005; Paynter et al., 2010).

However, surprisingly little is known about how introductions of classical biological control agents affect local food webs (Pearson & Callaway, 2005, 2008; Carvalheiro et al., 2008). A proper assessment of the costs and benefits of biological control release can only be made when the likelihood and significance of developing food web associations are understood. This could be made a priori by investigating the food web associations in the agent’s native range, as will be demonstrated here with two different types of galling biological control agent of two invasive Australian acacia.

At least 23 species of Australian acacias (taxa in the genus Acacia subgenus Phyllodinae; see Miller et al., 2011 for a review of different nomenclatures) are known to be invasive in different parts of the world (Richardson & Rejmánek, 2011). The widespread planting of many Australian acacias outside their natural range can be seen as a natural experiment which has huge potential for shedding new light on numerous facets of invasion ecology (Richardson et al., 2011). South Africa in particular provides an interesting study arena with at least 70 species introduced (Richardson et al., 2011) – ten of which are widespread invaders, and a further six have naturalized or are invasive at only a few sites (van Wilgen et al., 2011; Wilson et al., 2011) These species occur in nearly all regions of the country, except for the arid north and humid tropical eastern lowlands, and have a wide range of impacts in invaded ecosystems (Richardson & van Wilgen, 2004; Gaertner et al., 2009; Le Maitre et al., 2011). One facet of impact of these acacias that has not yet been well explored is that resulting from the introduction of biological control agents. South Africa being the only country that has released biological control agents on Australian acacias (Wilson et al., 2011), thus provides an opportunity to explore the effects of the introduction of these biological control agents. In general, biological control agents released on wattles in South Africa offers a diversity of case studies (e.g. long-history, multiple agents, varying degrees of control achieved, several different life-forms) to study long-term impacts of biological control agents and to improve risk assessment procedures.

Agents with sufficiently narrow host plant ranges, such as gall-inducing agents, are widely considered to be low risk, with negligible non-target host species impacts (Cruttwell McFadyen, 1998; Pemberton, 2000; van Klinken & Edwards, 2002; Messing & Wright, 2006). Here, we focus on two galling agents of Australian acacia which are well established in South Africa, were released more than 2 decades ago, are easy to collect, and represent a sizeable biomass in ecosystems where their hosts occur (McGeoch & Wossler, 2000; Impson et al., 2009; Seymour & Veldtman, 2010).

The flower-bud galling wasp, Trichilogaster acaciaelongifoliae Froggatt is a successful biological control agent of Acacia longifolia (Andr.) Willd. in South Africa, spread throughout the range of its host (Dennill, 1987; Hoffmann et al., 2002; Moran et al., 2005; Veldtman et al., 2010). Subsequently, four other galling agents on Australian acacias were released in South Africa (i.e. the related Trichilogaster signiventris on A. pycnanthaHoffmann et al., 2002; a rust fungus Uromycladium tepperianum (Sacc.) McAlp on A. salignaMorris, 1999; and midge Dasineura dielsi on A. cyclopsAdair & Neser, 2006), and Dasineura rubiformis on Acacia mearnsiiImpson et al., 2009). Although all species released were found to be host-specific to wattles in pre-release tests and have not been seen to cause direct effects on species other than wattles, the potential risk of forming non-target associations with higher trophic levels was not considered (Pearson & Callaway, 2003; Joy & Crespi, 2007). Galling insects may be particularly prone to forming such associations, as the galling habit is conducive to the formation of parasitoid and predator associations (Price et al., 1987), can be considered ecological engineering (Crawford et al., 2007), and provides a resource for gall inquilines (McGeoch & Chown, 1997; Bashford, 2002), some even being agricultural pests (Seymour & Veldtman, 2010).

A pioneering approach to assess the impacts of alien insects on native insect species in general has been to analyse food webs (Memmott et al., 1994; Schönrogge & Crawley, 2000; Morris et al., 2004). This approach has been applied successfully in evaluating the impacts of biological control agents on non-target species through quantitative post-release studies (Henneman & Memmott, 2001; Carvalheiro et al., 2008). A biological control agent can interact with its host community directly (natural enemies) or indirectly (inquilines or shared parasitoids) via the effects it has on its host plant (Fig. 1) (Pearson & Callaway, 2003; Messing & Wright, 2006). Here, we investigate whether a gall-forming wasp and a gall-forming rust, T. acaciaelongifoliae and U. tepperianum (both agents established for more than 2 decades), have non-target associations in South Africa mirroring those in their native food chains (Australia). This is the first time that any such a comparison of an agents’ non-target interactions (and not just parasitism; see Adair & Neser, 2006; Paynter et al., 2010) is made between its introduced and native range. In future, quantitative food web analyses of the species associated with prospective weed biological control agents in their native range could be used to a priori predict possible associations where they are planned to be introduced.

Figure 1.

 Food web guilds interacting with two galling biological control agents. (a) Dense stand of Acacia longifolia; (b) Uromycladium tepperianum galls on Acacia saligna; (c) Trichilogaster acaciaelongifoliae galls on A. longifolia; (d) Pseudotorymus sp. – parasitoid on T. acaciaelongifoliae; (e) Eupelmus sp. – hyperparasitoid on Pseudotorymus sp.; (f) Cryptophlebia peltastica– inquiline feeding on galls of both biological control agents; (g) Rogadinae sp. – parasitoid on C. peltastica.


We determined non-target species associated with T. acaciaelongifoliae and U. tepperianum by collecting galls (host trees larger than 15 cm basal circumference and more than 5 years old) throughout the invaded range of target hosts for both species. For T. acaciaelongifoliae we sampled a total of 3270 galls on 218 A. longifolia trees (at least 100 m apart) at 19 distinct localities in four different biogeographical regions, as well as 16 trees from the introduced, but non-invasive, A. floribunda (four localities) (May 2005–April 2006). For each tree, 15 galls were randomly sampled over the entire canopy. The host plant of U. tepperianum, A. saligna, has a smaller distribution range and was sampled at seven representative localities, yielding a total of 232 galls collected from an average of five trees per locality (July–November 2003). Insect identification was verified by the South African National Collection of Insects and Iziko Museums of Cape Town, South Africa. Each insect species was assigned to a trophic level based on published species or family natural history (Scholtz & Holm 1985), or from field observations during this study (see Supporting information Table S1). Other invertebrates or associated bacteria and fungi were not recorded during sampling. Trophic levels used were parasitoids, hyperparasitoids, inquilines, herbivores, predators and mutualists.

To assess sampling efficacy, sample-based rarefaction curves for species using U. tepperianum and T. acaciaelongifoliae galls (or as a host species in the latter case) in South Africa (see Supporting information Fig. S1) were created for one and several biogeographic zones, respectively, using sample-based rarefaction curves, using EstimateS V5 (Colwell, 1997; Gotelli & Colwell, 2001). The zone classification for T. acaciaelongifoliae was based on previous work quantifying the abundance and distribution of the host plant species, A. longifolia (Veldtman et al., 2010). These curves are constructed based on Monte Carlo resampling. The datasets were resampled 1000 times with samples drawn at random without replacement during each resampling (Colwell, 1997). A curve is then generated by plotting the mean number of species represented by the different number of samples (Colwell, 1997).

The number and guild type of species associated with these two biological control agents in their native range (Australia) were extracted from the literature (see Supporting information Table S2). In both studies, Bashford (2002, 2004) sampled over the course of whole year with more than 10 localities and 1000 galls collected. It was thus assumed that his sampling was exhaustive. Another Australian study on the inquilines of U. tepperianum galls does exist (Hosking & Edwards, 2010) but did not consider parasitic wasps, rendering it less useful for the purposes of this study. This paper did however find most of the species of inquilines described by Bashford (2004) for Tasmania, also occurred in New South Wales plus several new species from related families (Hosking & Edwards, 2010).

The percentage similarity between insects associated with T. acaciaelongifoliae and U. tepperianum galls in South African with those associated with these galling species in their native range (Bashford, 2002, 2004) was determined using the Bray-Curtis Similarity index calculated in PRIMER v5.0 (Clarke & Warwick, 1994; Clarke & Gorley, 2001). Analysis included inquilines, parasitoids and hyperparasitoids, but not herbivores, predators and mutualists. The reason for omitting the second group of guilds was inconsistent data availability in both the native and introduced ranges (neither this study nor Bashford, 2002, 2004 recorded data for each of these groups). The first mentioned group of guilds are always found inside the gall for at least one life stage and are thus more reliably sampled during gall collections.

Percentage similarity was determined at the species, genus, family, superfamily and order level, and number of occupant species (thus using number of morphospecies only) leading to a sample size reduction with an increase in classification category. Further descriptive statistics using PRIMER were not possible as only one native and one introduced species assemblage were used in the analysis. Identical samples will have 100% similarity, while for one shared species out of a possible 20, it is about 5%.

For the Bray-Curtis Similarity comparison of native and invasive food chain interactions associated with T. acaciaelongifoliae and U. tepperianum, occurrence data were used, and when available abundance data as well. This allowed the results when using abundance data (standardized) versus when using only occurrence data to be compared. Abundance data were available in both ranges for all trophic levels associated with T. acaciaelongifoliae, but only for inquilines associated with U. tepperianum galls.

The percentage similarity of associated insect fauna of the native wasp and rust galls, and those associated with both in their introduced range were also each assessed. This was performed to determine the degree to which native range information on food web links improved the prediction of the number in the introduced range, compared to the number and composition predicted from studying food webs of other introduced galling agents.

Given the subjective nature of taxon delimitation and rank allocation, we substantiated the taxonomic rank analyses with one examining shared phylogenetic branch length. A phylogenetic tree (Supporting information Appendix S1) was assembled based on information from Grimaldi & Engel (2005) and Hunt et al. (2007), with branch length values approximated to the closest 10 Myr. Other studies (e.g. Nylin & Wahlberg, 2008) suggest that these values represent slight underestimates, but the underestimation is likely to be fairly consistent throughout and should not affect our comparisons. Given the lack of fully resolved phylogenetic trees for insects, we limited our tree to family level, with the additional inclusion of subfamilies in one family where these are likely to represent ancient lineages (Horak & Komai, 2006). We are confident that all lineages older than 50 Myr are incorporated in our tree, and branch lengths within families and subfamilies – if values were available – would only increase the phylogenetic diversity value for any of our four assemblages by less than 10% (cf. Procheşet al., 2006 for plants).


Few of the species accumulation curves approximated asymptotes (Supporting information Fig. S1). This indicates that more samples are needed to collect the remaining species occurring in those areas and that the food webs are thus likely to be even more complex.

Novel and native food webs associated with T. acaciaelongifoliae and U. tepperianum revealed qualitative similarities (Fig. 2). For both galling agents, 33% and 44% (wasp and rust respectively) of the insect species were from the same families, with greater overlap observed at the superfamily and order levels, as expected (Fig. 2a–d, e–h & i–l). Bray-Curtis percentage similarity between novel and native assemblages associated with these galling agents revealed similar patterns when using occurrence data (Fig. 3). At the species and genus level, percentage similarity was very low but increased rapidly at the family and superfamily level, reaching a maximum at the order level. This was also reflected in percentage shared phylogenetic branch length, with the same lineages often colonizing the structures associated with the control agents in their native and introduced range (see Fig. 4 and Supporting information Table S3). Species within different trophic levels sampled in the novel range associated with these organisms were similar in number and taxonomic composition to those found in their native range (percentages given in Fig. 3, Table S4 in Supporting information). In addition, the optimal trade-off between minimum taxonomic predictor level and maximum percentage similarity was at the superfamily level. In contrast, when comparing the associations of the two agents in their native range, there was a high percentage of similarity at the species level (Fig. 3).

Figure 2.

 Food webs associated with Trichilogaster acaciaelongifoliae and Uromycladium tepperianum (left- and right-hand columns respectively) in their native Australian range (a, b, e, f, i, and j) and introduced range in South Africa (c, d, g, h, k and l), shown at decreasing taxonomic resolution: family (a–d), superfamily (e–h) and order (i–l), respectively. Colours represent different orders, different shades representing specific families (a–d), and superfamilies (e–h) (see Supporting information Tables S1 & S2 for family and super family names); and orders (i–l) (orange = Hymenoptera; blue = Lepidoptera; green = Coleoptera; brown = Hemiptera). ‘?’ indicates host of parasitoid is unknown.

Figure 3.

 Percentage similarity between insects associated with biological control agents in South Africa compared to their native associations in Australia. Similarity between these two native and two invasive webs respectively is also shown. The wasp denotes T. acaciaelongifoliae, and the rust U. tepperianum. Similarity is based on occupancy of introduced and native insect assemblage’s (using Bray-Curtis Similarity by Primer V5) and shared phylogenetic diversity (PD) (see Supporting information Table S3 for branch lengths used in calculations).

Figure 4.

 Phylogenetic tree including the insect species associated with the biological control agents in their native range and introduced range. Thick lines denote the wasp and thin the rust; solid and dashed lines indicate native and introduced range respectively (see Supporting information Appendix S1 for tree construction; tree was drawn using an informal supertree approach, see Bininda-Emonds, 2004).

In both cases where abundance and occurrence data were available for the native versus introduced food chain comparison, occurrence data resulted in slightly higher percentage similarity (2–5%) for all taxonomic groupings except for the order and guild level, where percentage similarity was higher (approximately 20%) (Supporting information Table S4).


The similarity observed between native and novel food webs at the family and superfamily level (plus similar patterns quantified using data on shared phylogenetic diversity) indicates promise in using the number and type of food web links of prospective insect weed biological control agents in their native range, to predict the establishment of feeding links in a novel environment. An exact forecast of the species or even genera that will become associated with biological control agents is obviously not required because it is unlikely that there will be species similarities between two such disparate ranges. However, if predictive capability exists at the family level, the type of association and potential other food web interactions (and thus potential effects) could be predicted.

The results for phylogenetic branch length values are most similar to the ones for genus level (Fig. 3; Table S3 in Supporting information). If this is indicative of an equivalence in the predictive power of phylogenetic measures and that of genera (as suggested by Procheşet al., 2009), this would mean that the simple incorporation of relationships at family level and higher could replace the effort of identifying all taxa to genus level. With novel methods (e.g. DNA barcoding –Valentini et al., 2009), the availability and accuracy of insect phylogenies are likely to improve, thus providing an additional tool for analyses involving assemblage similarity, some of which, as suggested here, could have broad applicability.

The number of species and type of food chain links formed with biological control agents in this study (Table 1; Fig. 2) can be expected to increase with more sampling, given the species accumulation curves generated here (Supporting information Fig. S1) and the characteristic spatial variability in novel food webs (Mitchell et al., 2006). Nonetheless, this will be less of a concern when making comparisons at a coarse scale such as national distributions as done here.

Table 1.   Number of new versus native galling-agent interactions.
Food chain link typeNumber of species links in invaded range*Number of species links in native range†
  1. *Total documented number of species links found in South Africa; species sampled in this study appears in brackets if different (see Supporting information Table S1; Mangoni & Hoffmann 1995; McGeoch & Wossler 2000).

  2. †See Supporting information Table S2; Noble 1940; Bashford 2002, 2004).

  3. ‡Invertebrates feeding externally on gall; na, not sampled.

  4. §Although the paper in question lists this number of host species, each strain or genotype of U. tepperianum is known to be host specific to one or a few species (Hosking & Edwards, 2010) and the difference between the native range and South Africa is likely due to differences in number of strains present.

Trichilogaster acaciaelongifoliae (released 1982)
 Host plants4 (2)3
 Parasitoids6 (2)9
 Hyper parasitoids3 (2)3
Uromycladium tepperianum (released 1987)
 Host plants1118§
 Hyper parasitoids12

That the percentage similarity increased between native and novel food webs associated with a biological control agent as taxonomic resolution declined is expected. This does not, however, invalidate the similarity approach. Even if the new links formed in the introduced range can only be predicted with confidence at the order level, this still highlights potentially unwanted or problematic relationships with parasitoids or inquilines (Hymenoptera and Diptera or Lepidoptera and Coleoptera respectively) before release.

A recent study in South Africa found that the important agricultural pest false codling moth (Thaumatotibia leucotreta) uses the galls of U. tepperianum on A. saligna as a larval food resource in agricultural ecosystems (Seymour & Veldtman, 2010). With current biological control agent risk assessment procedures, it is impossible to consider such undesirable effects. Therefore, even by using a simple food chain approach (compared with more complicated food webs approaches - Schönrogge & Crawley, 2000; Morris et al., 2004; Tylianakis et al., 2007), we have demonstrated the potential for accumulating similar multi-trophic food chain links with introduced biological control agents, as well as marked similarities between these links and those observed in the native range.

It could be argued that this study only shows that if a gall maker is attacked mostly by hymenopteran parasitoids in its native range, then this is also likely to be the case in its introduced range. Although this result is expected and does not reveal anything about the effect on the native food web and potential negative indirect effects on non-target species, the native insect-gall-inducing fauna South Africa are not satisfactorily described (Scholtz & Holm 1985; Veldtman & McGeoch, 2003), let alone their food web links. It is thus impossible to predict the subset of those native parasitoids that interact with the introduced species or give the opportunity to make predictions on negative non-target effects, e.g. through apparent competition with native galling insect species. Current assessments are susceptible to missing possible non-target species associations, with potential economic and ecological effects (e.g. Seymour & Veldtman, 2010).

Adair & Neser (2006) proposed monitoring the accumulation of native parasitoid species on biological control agents in their introduced range relative to their native range, to optimize agent survival and efficacy. The release of other gall-inducing flies from Australia in South Africa is considered acceptable, because a few parasitoid species accumulated after five years on Dasineura dielsi (on Acacia cyclops) compared with eleven species in the native range (Adair & Neser, 2006). Although most parasitoid species will accumulate on biological control agents within the first 3 years of release (Cornell & Hawkins, 1993; Hill & Hulley, 1995), a strategy whereby non-target associations with introduced agents are quantified post-release is ecologically risky (Pearson & Callaway, 2003; Delfosse, 2005; Thomas & Reid, 2007). The findings we present here potentially markedly alter how agents could be selected for biological control, by including assessments of upper trophic level non-target risks prior to agent release (see also Paynter et al., 2010), thereby improving the risk assessment process and limiting later concerns about unexpected non-target impacts.

A recent review of the effectiveness of biological control agents in controlling their invasive host plants identified agents that become superabundant on their host as a recipe for non-target associations (Thomas & Reid, 2007) as well as increasing the potential for host shifts (Dennill et al., 1993). Furthermore, another recent study (Carvalheiro et al., 2008) showed that the hosts of native parasitoids (seed feeders) may become extinct locally when such a superabundant biological control agent acts as a shared host. If in fact secondary associations with biological control agents (such as parasitism) present a threat to native biodiversity, in addition to potentially affecting agent effectiveness (Impson et al., 2009), then ways of predicting secondary associations (prior to release) need to be improved. Our approach thus provides framework for improving predictions of future associations developing with biological control agents after release and thereby improve the risk assessment process (at least for galling agents with the current data available).

We do not suggest that our results prove a link between associations and impacts, or that if there are impacts that these will always be negative; rather, they raise sufficient concern to merit application of thorough risk assessment. The aim should be to predict the impact of native species on agent effectiveness and also secondary impacts of agents on native species. Once similarity in species association is assessed, several further steps will be required to quantify similarity in expected impacts. This study has provided one means of so doing which we hope will stimulate future research in classical biological control and invasive plant management and hereby advance the current understanding of the formation of novel associations between native and introduced species.


We thank farmers, landowners, residents and individuals from the Western Cape, Eastern Cape, Mpumalanga and Gauteng for allowing or assisting in surveys; Cape Nature and Eastern Cape Department of Economic Affairs, Environment and Tourism for supplying collection permits; M.A. McGeoch for advice on research planning; S. van Noort for wasp identification; and M.A. McGeoch, C.L. Seymour, J.S. Donaldson, M. McLeish, A.G. Rebelo and L.G. Carvalheiro for comments on previous drafts. This research was funded by the Department of Science and Technology and the National Research Foundation (South Africa) which allocated funds to Centre for Invasion Biology. The Claude Leon Foundation is thanked for funding the postdoctoral fellowship of R.V. Funding from Stellenbosch University and the Oppenheimer Memorial Trust enabled A.B., T.F.L. and R.V. to attend the October 2010 Acacia workshop in Stellenbosch.


Ruan Veldtman works in the field of applied biodiversity research. His particular interests include ecological entomology, plant–insect interactions, pollination ecosystem services and biological control of plant invasions.

Author contributions: R.V. conceived the idea; R.V., T.F.L., A.E.T. and H.G. collected the data; R.V., A.B., S.P. and S.L.C. analysed various sections of the data, and R.V. led the writing with all authors editing the manuscript.

Editor : John Wilson