Gall wasp biocontrol of invasive longifolia: implications of strong bottom-up effects

MARIE V. H ENRIKSEN, DAVID G. CHAPPLE,STEVEN L. CHOWN, AND MELODIE A. MCGEOCH

School of Biological Sciences, Monash University, Clayton, 3800 Australia

Citation: Henriksen, M. V., D. G. Chapple, S. L. Chown, and M. A. McGeoch. 2017. Gall wasp biocontrol of invasive : implications of strong bottom-up effects. Ecosphere 8(12):e02043. 10.1002/ecs2.2043

Abstract. The population dynamics of herbivore biocontrol agents is central to the successful con- trol of invasive weeds. Although the importance of agent population dynamics is recognized, it is rarely considered in assessments of the biocontrol potential of herbivore agents. Herbivore insect population dynamics are influenced by a combination of top-down effects from natural enemies, bottom-up effects from resource availability (resource quality and quantity), and potential interactions between these effects. To better understand the tri-trophic interactions that are likely to determine biocontrol success in a host plant–gall wasp–parasitoid system, the relative importance of top-down and bottom-up effects for the survival of a herbivore biocontrol agent (Trichilogaster acaciaelongifoliae), on two Acacia host in their native range, was estimated using path analysis. On both host plants, there was a strong positive relation- ship between gall mass per chamber and gall wasp survival and a strong negative relationship between gall mass per chamber and gall parasitism, with parasitoids being less common in large than small galls. There was, however, no significant correlation between parasitism and gall wasp survival and, therefore, no evidence for top-down effects in this system. Strong bottom-up effects of host plant resources on both gall wasp survival and gall parasitism have implications for the spatio-temporal variability of biocontrol success. Such variation should be considered in pre-release assessments and post-release monitoring of gall wasps used as herbivore biocontrol agents.

Key words: biological control; herbivore performance; parasitoid wasps; plant–herbivore interactions.

Received 2 November 2017; accepted 3 November 2017. Corresponding Editor: Robert R. Parmenter. Copyright: © 2017 Henriksen et al. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. E-mail: [email protected]

INTRODUCTION (such as parasitoid wasps), and bottom-up effects from plant resource availability (resource Knowledge of insect herbivore population quality and quantity; Hunter and Price 1992). dynamics is central to the success of weed man- Host plant resources (e.g., plant tissue nutrient agement programs using herbivore biocontrol and water status) and intra-specific competition agents (Gassmann 1996, Sheppard 2003, Zalucki affect herbivore performance between and and van Klinken 2006). Although the general within plant individuals (Gassmann 1996, Fritz theory of insect herbivore ecology, especially et al. 2000, Briese 2004, Zalucki and van Klinken population dynamics, is well developed, spatio- 2006, Cornelissen et al. 2008). Introduced herbi- temporal variation in the strength of insect herbi- vore may also be exposed to top-down vore–plant interactions is rarely considered when effects because novel generalist natural enemies evaluating herbivore agent biocontrol potential adopt them as prey (Denoth and Myers 2005, (Gassmann 1996, Lakatos et al. 2017). Herbivore Veldtman et al. 2011). Studies have shown that insect demography is influenced by a combina- the relative importance of top-down and bottom- tion of top-down effects from natural enemies up effects for herbivore performance varies

❖ www.esajournals.org 1 December 2017 ❖ Volume 8(12) ❖ Article e02043 HENRIKSEN ET AL. substantially, driven by spatial and temporal parasitoids attempting to lay their eggs on devel- variation in abiotic and biotic conditions (Hunter oping wasp larvae inside the gall chambers (Price and Price 1992, Forkner and Hunter 2000, Hunter and Clancy 1986). Thus, large galls on high- 2001, Denno et al. 2002, Stiling and Moon 2005). quality plant tissue are less often parasitized than In some herbivores, top-down and bottom-up small galls (Price and Clancy 1986, Albarracin and effects may also interact and the influence of host Stiling 2006). The consequences of plant-mediated plant resources (such as plant quantity, quality, top-down effects on herbivore performance and morphology) on the relative strength of top- remain poorly understood, with only a few exam- down impacts from natural enemies of the herbi- ples from biological control of crop pests (Agrawal vores is well studied (Price et al. 1980, Forkner 2000, Cortesero et al. 2000, Hosseini et al. 2010). and Hunter 2000, Denno et al. 2002). In insect gall Bottom-up impacts of plant resources on para- makers, between- and within-plant variation in sitoids in the third trophic level may often have gall size positively influences gall maker perfor- been overlooked in weed biocontrol studies mance through increased resource quality and because herbivore agents are traditionally quantity, buffering against water loss and protec- assumed to be introduced to an enemy-free space tion against parasitoid attack (Price and Clancy (Lawton 1985), and thus free from the negative 1986, Price 1991, Sumerford et al. 2000, Ito and impact of a third trophic level. When generalist Hijii 2004, Albarracin and Stiling 2006, Laszl oand enemies are present (e.g., Veldtman et al. 2011, Tothm eresz 2013). Gall development (growth rate Lopez-N unez~ et al. 2017), however, bottom-up and final size) has been shown to be limited by impacts of host plant resources on the strength of the resource status of host plant tissue with galls top-down effects could be essential to herbivore on high-quality host plants, and host plant mod- agent survival and the eventual success of ules, growing faster and to a larger final size than biocontrol programs. One such example is the galls on low-quality host tissue in accordance with bud-galling wasp Trichilogaster acaciaelongifoliae the plant vigor hypothesis (Price and Clancy 1986, (Froggatt) (: ; Fig. 1A) Price 1991, Albarracin and Stiling 2006). The thick that has been used for biocontrol of the invasive layer of gall tissue is a physical barrier to Australian Acacia longifolia (Andrews) Willd. in

Fig. 1. (A) Reared gall wasp (Trichilogaster acaciaelongifoliae) female. (B) Parasitoid (large) and hyper-parasitoid (small) larvae in the gall chamber of a consumed female gall wasp larvae. On the right side, the radial line shows an oviposition trace of a parasitoid extending from the gall surface to the gall chamber.

❖ www.esajournals.org 2 December 2017 ❖ Volume 8(12) ❖ Article e02043 HENRIKSEN ET AL.

South Africa (Impson et al. 2011) and Portugal (Sumerford et al. 2000), and more likely to have (Marchante et al. 2017) and has been proposed high fecundity (Ito and Hijii 2004). Furthermore, for biocontrol of A. longifolia in other parts of the we predicted an interaction between top-down world including New Zealand (Hill 2005) and and bottom-up effects with gall mass per chamber areas of Australia where A. longifolia has spread being negatively related to the presence of para- beyond its native range (Adair 2008). A rich sitoids in the galls (Price and Clancy 1986, Albar- novel enemy assemblage has been shown to be racin and Stiling 2006). related to the gall wasp in the introduced range in South Africa (Veldtman et al. 2011), poten- METHODS tially exposing the biocontrol agent to strong top-down effects. The biology and ecology of Trichilogaster By examining the relationship between gall acaciaelongifoliae properties, parasitism, and gall wasp survival Trichilogaster acaciaelongifoliae lays its eggs in across multiple sites in the native range, spatial flower buds and vegetative buds of its host variationinthesystemcanbeusedtobetterunder- plants and a gall forms, inhibiting flower and stand the mechanisms that may be responsible for phyllode development (Noble 1940, Dennill et al. the success of biocontrol agents in their introduced 1993). When most successful, it limits up to 100% range. Here, we, therefore, estimated the relative of seed production, inhibits vegetative growth, importance of top-down and bottom-up effects and occasionally causes tree mortality (Dennill within trees for the survival of this herbivore gall 1985, Hoffmann et al. 2002). Knowledge of the wasp biocontrol agent (T. acaciaelongifoliae)ontwo tri-trophic Acacia–gall wasp–parasitoid interac- Acacia () host plants (A. longifolia subsp. tion in its native range is, however, limited longifolia and Acacia floribunda (Vent.) Willd.) across (Noble 1940, Bashford 2004), and most empirical the edge of its native range. We also evaluated studies have been conducted in South Africa for the possibility of a direct effect of gall mass per biocontrol purposes (e.g., Dennill 1985, Manongi chamber on gall parasitism. We did this to better and Hoffmann 1995, McGeoch and Wossler 2000, understand the tri-trophic interactions that are Veldtman et al. 2011). Both top-down and likely to influence the biocontrol success of this bottom-up effects apparently influence the popu- gall wasp beyond its native range in Australia lation dynamics of this gall wasp (e.g., Dennill (Adair 2008) and elsewhere in the world (Hill et al. 1993, Manongi and Hoffmann 1995, Bash- 2005, Impson et al. 2011, Marchante et al. 2017). ford 2004). Yet no previous study has quantified Based on patterns of top-down and bottom-up the relative influences of top-down and bottom- effects observed in other gall makers, we pre- up effects on its performance. dicted that female gall wasp survival at the gall Females of the gall wasp are parthenogenetic level is positively related to gall mass per chamber and can reproduce without males (Noble 1940), (used as a measure of resource availability in and male-to-female sex ratios vary considerably galls; Abrahamson and Weis 1997, Stone et al. from about 0.01:1 to 0.94:1 (Noble 1940). Females 2002, Price 2003), negatively related to the pres- are therefore critical to gall wasp reproduction and ence of parasitoids (Stiling and Moon 2005, Albar- population viability and were the focal individuals racin and Stiling 2006) and negatively density of this study. The female larva of the gall wasp dependent due to intra-specific competition induces gall formation (Dorchin et al. 2009), and between larvae within the galls (i.e., the individ- the gall tissue grows into a thick sphere surround- ual likelihood of survival decreases with the num- ing the female larva (Fig. 1B; Noble 1940). Each ber of individuals per gall; Fritz et al. 2000). Gall female gall chamber is visible externally as a lobe mass per chamber is an informative proxy for on the gall surface, and female chambers can, resource availability in individual galls in the therefore, be counted prior to wasp rearing (Noble form of plant tissue quality and quantity, with 1940). Male larvae develop in chambers in the multiple demonstrated benefits including individ- periphery of the gall tissue (Noble 1940) with no uals in large compared to small galls being less externally visible indication of the number of often parasitized (Price and Clancy 1986, Albar- males in the gall (Noble 1940, Bashford 2004). racin and Stiling 2006), less likely to desiccate Male chambers can, therefore, only be observed by

❖ www.esajournals.org 3 December 2017 ❖ Volume 8(12) ❖ Article e02043 HENRIKSEN ET AL. dissecting the gall, killing any developing larvae the edge of the native range of the Acacia host or pupae. Given that our principle aim was to plants (Walsh and Entwisle 1996). The study measure gall wasp survival to adulthood, that is, extent includes three Australian bioregions, the to record the number of adults emerged, galls Gippsland Plain (Melbourne and Mornington couldnotbedissectedimmediatelytoquantify Peninsula), the Highlands—Southern Fall (Mel- male abundances. By the time all individuals have bourne), and the Greater Grampians (Halls Gap) emerged from the galls, the small peripheral male that vary in climate, geology, and native vegeta- chambers were no longer visible when galls were tion (IBRA7; Australian Government Depart- dissected. Female chambers are, however, sur- ment of the Environment and Energy 2016). We rounded by substantially larger amounts of both sampled across this large spatial extent with the storage and vascular tissue than male chambers purpose of capturing natural variation in the sys- (1.5 times more storage tissue and 3.5 times more tem to examine the relationships between gall vascular tissue; Dorchin et al. 2009) and it was, size and gall wasp survival. therefore, assumed that males would have a For biocontrol purposes, the primary interest negligible effect on within-gall competition for of the study was the A. longifolia–gall wasp inter- resources compared to females. Thus, male sur- action but given much overlap in the distribution vival was not estimated and gall wasp survival of A. longifolia and A. floribunda (in the Mel- was quantified as female survival to adulthood. bourne region) trees of both host plants were Herein, the terms “chamber” and “gall wasp” are chosen for collection and are dealt with sepa- used to refer to female chambers and wasps, rately in the analysis. Galls were collected from respectively, unless specifically mentioned as male. 17 sites (circular and 1 km in diameter) across The natural enemies of the T. acaciaelongifoliae the three regions (Melbourne: n = 15, Halls Gap: gall wasp include a rich assemblage of parasitoid n = 1 and Mornington Peninsula: n = 1). In Halls species (Hymenoptera). In Australia, the most Gap and Mornington Peninsula, only A. longifolia abundant parasitoids found in association with was present while both A. longifolia and A. flori- the gall wasp are members of the Eurytomidae bunda co-occurred across sites in the Melbourne and Torymidae families (Noble 1940, Bashford region. Acacia floribunda generally occurred in 2004). Parasitoid-induced gall wasp mortality much higher densities across sites (Appendix S1: can be up to 61% (Bashford 2004), but varies sub- Table S1) and more A. floribunda trees were, stantially, even between neighboring trees (Noble therefore, sampled overall (n = 44 for A. flori- 1940). The reason for such variation, and the bunda and n = 27 for A. longifolia). Up to six trees impact of high rates of parasitism on gall wasp were sampled within each site except when less population dynamics, is unknown. The para- than six galled trees were present (Appendix S1: sitoids use their ovipositors to penetrate the gall Tables S2 and S3) and these trees were chosen to tissue and lay their eggs in the gall chambers be as far apart as possible within the extent of the (Fig. 1B) and male chambers that lie in the site. The total number of trees (galled and periphery of the gall, close to the surface, are ungalled) per site was counted. For each sampled more often parasitized than female chambers tree, the circumference at the base was measured (Manongi and Hoffmann 1995). The parasitoid and the number of galls per branch was estimated larvae that hatch will consume the gall wasp as a mean of ten randomly chosen branches (with larva at some point during larval development. a “branch” defined as the terminal 30-cm segment Herein, the term “parasitism” refers to para- of a shoot following Dennill 1985). The sampled sitoid-induced mortality of gall wasp larvae. trees were located in urban environments, that is, along road sides and in urban reserves and parks Study area where galls occur in high densities, which enables Galls were collected from two Acacia host spe- a relatively high sample size per tree. cies, A. floribunda and A. longifolia, across an area of 9370 km2 between the Mornington Peninsula Gall collection and rearing (38°20043″ S, 145°00025″ E), Melbourne (37°50045″ Galls were collected for rearing and dissection S, 145°04025″ E), and Halls Gap (37°08012″ S, over a three week period in November 2014 142°31009″ E), Victoria, Australia, which spans (Table 1) after gall wasp larvae had pupated and

❖ www.esajournals.org 4 December 2017 ❖ Volume 8(12) ❖ Article e02043 HENRIKSEN ET AL.

Table 1. Counts of trees, galls, and reared individuals were preserved in 96% ethanol and stored for from two host plants, Acacia floribunda and Acacia subsequent identification. Trichilogaster acaciae- longifolia. longifoliae gall wasps were identified using Prinsloo and Neser (2007). Parasitoids were iden- Gall rearing A. floribunda A. longifolia tified to family, or lower taxonomic level when Trees sampled 44 27 possible, using relevant identification keys Galls reared 1416 679 (Boucek 1988, CSIRO Division of Entomology Gall wasps reared 1991, Lawrence and Slipinski 2013) and grouped Females 422 (0.65) 250 (0.67) Males 72 (0.11) 42 (0.11) into morphospecies. A range of inquilines (e.g., Male-to-female ratio 0.17 0.17 Hymenoptera, Lepidoptera, and Coleoptera indi- Parasitoids reared viduals feeding on the gall tissue; Noble 1940, Torymidae Boucek 1988, CSIRO Division of Entomology Megastigmus sp. 90 (0.14) 50 (0.13) 1991) also emerged from the galls. Whereas para- < Monodontomerinae sp. 1 ( 0.01) 0 (0.00) sitoids depend on reaching the gall wasp larvae Eurytomidae in the gall chambers to complete their develop- Eurytoma sp. 1 33 (0.05) 16 (0.04) Eurytoma sp. 2 27 (0.04) 14 (0.04) ment to adulthood, inquilines complete their Pteromalidae development in the tissue of the gall. Inquilines Coelocybinae sp. 6 (0.01) 3 (0.01) were, therefore, not expected to be restricted by Galls parasitized (%) 11.0 11.7 gall size in the same way as parasitoids. Since the Total individuals reared 651 375 focus of this study was to quantify the influence Note: For each group of individuals reared (gall wasps of gall size on gall wasp survival, inquilines and parasitoid morphospecies), their proportion out of all were, therefore, not considered further. individuals reared per host plant (i.e., relative abundance) is given in brackets. To determine if parasitoids preferentially attacked male chambers in the galls (as has been while gall tissue was still fresh (i.e., not yet shriv- observed in the introduced range; Manongi and eled). Galls were, as far as possible, collected Hoffmann 1995), the male-to-female sex ratio of evenly across trees of each host plant within a emerged individuals was compared to the ratio site except when individual trees had very low of male to female gall chambers in a separate set gall abundances (see Appendix S1: Tables S2 and of dissected galls. If parasitoids attack male and S3 for exact number of galls sampled per tree). A female chambers with the same frequency, the total of 30.9 16.9 (mean SD) galls per tree relative number of male compared to female gall were collected for rearing. This did not include wasps should be similar between reared (indi- galls from which individuals had already viduals surviving to adulthood) and dissected emerged (as determined by emergence holes on (observed number of chambers) galls. Thus, the gall surface). Galls collected for rearing were while collecting galls for rearing, 24.6 10.5 weighed and, to quantify the number of female (mean SD) galls per tree were collected for chambers, the number of lobes on each gall was dissection, from a subset of the sampled trees counted. With an increasing number of lobes per (n = 11 for A. floribunda and n = 12 for A. longifolia; gall, it becomes more difficult to distinguish each see Appendix S1: Tables S2 and S3). These galls lobe from the others. To minimize error in the were dissected in the laboratory, and the number estimated number of gall chambers, we therefore of female and male chambers was counted. avoided using galls with four or more lobes (ex- cluding <5% of the galls sampled for rearing). Data analysis Galls were then stored individually in resealable For each reared gall, gall wasp survival to adult- plastic bags, and gall wasps and parasitoids were hood was quantified as the proportion of cham- reared from the galls at room temperature bers from which a gall wasp successfully emerged, (~20°C). Emergence of individuals was checked that is, the number of emerged adult females from 25 November, at least once a week for the divided by the number of female chambers in each first three months of rearing and then once a gall (estimated as the number of gall lobes). Gall month for the next four months until emergence mass per female chamber (i.e., the tissue available ended (early July 2015). Emerged individuals to each female gall wasp in the gall) was

❖ www.esajournals.org 5 December 2017 ❖ Volume 8(12) ❖ Article e02043 HENRIKSEN ET AL. quantified as the total mass of each gall divided the overall fit of the hypothesized causal network by the estimated number of female chambers. Gall was tested. For each response variable in the causal wasp density was included as a categorical predic- network (i.e., each variable with at least one ingo- tor defined as the number of chambers in a gall ing arrow), a separate component model was con- (one, two, or three chambers). Parasitism was structed. Each response variable was then tested as scored as a binary variable, that is, either present a function of all its hypothesized direct causes (i.e., or absent in each gall. A binary score was chosen variables with arrows connecting directly to, and over a density measure because the number of pointing toward, the response variable). The com- parasitoids in a gall is likely to be highly related to ponent models of the piecewise SEM were ana- gall wasp sex ratio (Manongi and Hoffmann lyzed as mixed models including individual trees 1995). Parasitoid density is, therefore, more likely as a random effect. A mixed model approach was to overestimate the strength of the top-down effect chosen to correct for spatial aggregation of the data in the system. Parasitoid presence as a measure of (Zuur et al. 2009) originating from the sampling of parasitism is, therefore, also only an estimate of multiple galls within each tree and because we the likelihood that female gall wasps are exposed were testing the within-tree effect of gall size on to top-down effects. If effects of parasitism on sur- gall wasp survival. The component models (Fig. 2) vival are absent, it may be due to either low levels are as follows: (1) a gall wasp survival model (gen- of parasitism overall or low levels of parasitism on eralized linear model [GLMM] with binomial dis- female gall wasps specifically. tribution and logit link function) testing gall wasp The hypothesized direct and indirect effects on survival as a function of gall mass per chamber, within-tree variation in gall wasp survival in the gall wasp density, and parasitism; (2) a parasitism reared galls were structured as a causal network model (GLMM with binomial distribution and (Fig. 2; for detailed description see Appendix S2) logit link function) testing parasitism as a function and tested as a piecewise structural equa- of gall mass per chamber; and (3) a gall mass tion model (piecewise SEM) for each Acacia host model (linear mixed model [LMM]) testing gall (Lefcheck and Duffy 2015). Unlike a traditional mass per chamber as a function of gall wasp den- path analysis, a piecewise SEM enables the inclu- sity. The effect of gall wasp density on gall mass sion of random effects and variables with non- per chamber was included as a path in the causal normal error distributions (Shipley 2009). network because there was a non-linear association The causal network was divided into component between gall mass per chamber and number of models, which were analyzed separately, before chambers in a gall (see Results and Fig. 3). Pairwise comparisons between all three levels of gall wasp density (i.e., one, two, or three chambers per gall) i in the gall mass models were performed with Parasitism – Tukey’s post hoc test. Finally, to test for the impor- Gall wasp tance of the direct link between gall resources and density gall parasitism in the tri-trophic interactions of the Gall wasp i ii – survival gall wasp, we also constructed and tested a com- – petingcausalnetwork(AppendixS2:Fig.S2).This Gall mass causal structure did not include the direct link per chamber + iii between gall mass per chamber and gall parasitism i – (see Appendix S2 for further details). The overall fit of a piecewise SEM is tested with Shipley’s test of d-separation (Shipley 2009). In the d-separation test, a significance for each Fig. 2. Hypothesized causal network of bottom-up independence claim in the causal network is and top-down effects on gall wasp herbivore survival found. Any two variables in the causal network to adulthood. The causal network can be divided into that do not share an arrow are included as an three component models, (i) predictors of gall wasp independence claim; that is, based on the knowl- survival, (ii) predictor of gall parasitism and (iii) pre- edge used to construct the network, the two dictor of gall mass per chamber. variables are considered to be independent

❖ www.esajournals.org 6 December 2017 ❖ Volume 8(12) ❖ Article e02043 HENRIKSEN ET AL.

3 If marginal R2 is low compared to conditional R2, a and most of the variance explained is due to the n = 373 random effect, there is large between-tree varia- tion in the magnitude of the tested response. b b n = 100 All statistical tests were performed in R 3.2.2 2 n = 206 A (R Core Team 2015). Linear mixed models were n = 859 analyzed in the nlme package (version 3.1-121; B B Pinheiro et al. 2015), and GLMMs were analyzed n = 433 n = 124 with the lme4 package (version 1.1-9; Bates et al. 1 2015). The overall piecewise SEM fit(d-separa- tion test), marginal R2, and conditional R2 were Gall mass per chamber (mg) calculated in the piecewiseSEM package (version 1.0.0; Lefcheck 2015) using P-values based on ’ 0 Satterthwaite s approximations for the d-separa- 1 2 3 tion test. Tukey’s test was performed in the mult- Number of chambers per gall comp package (Hothorn et al. 2008) while vif was calculated using the vif.mer function (avail- Fig. 3. Average gall mass per chamber (mg) with able online at https://github.com/aufrank/R-hac standard errors of reared galls with one, two and three ks/blob/master/mer-utils.R). chambers. Average mass is shown for both Acacia flori- bunda (white) and Acacia longifolia (gray) and the num- RESULTS ber of galls (n) is given for each category. Significant differences in gall mass per chamber across galls with Reared and dissected galls varying number of chambers were tested within each Between late November 2014 and early July fi host plant. Different letters indicate signi cant differ- 2015, 651 and 375 individuals emerged from fl ences between groups (uppercase letters for A. ori- reared Acacia floribunda and Acacia longifolia galls, bunda and lowercase letters for A. longifolia)as respectively (Table 1). Most gall wasp individuals ’ determined with Tukey s post hoc tests. (female and male) emerged within the first four weeks of rearing from both A. floribunda and (Shipley 2009). A composite probability of all A. longifolia galls (Fig. 4). The reared male-to- independence claims is then calculated with a female sex ratio was 0.17 on both host species. Fisher’s C test statistics using a v2 distribution. A The male-to-female chamber ratio (mean SD) resulting P-value lower than the chosen signifi- of dissected galls was much higher, at 0.75 0.16 cance level (here a = 0.05) leads to a rejection of for A. floribunda trees and 0.69 0.20 for A. longi- the proposed causal network. If the d-separation folia trees. Of the emerged parasitoids, 96% test fails to reject the causal network, the belonged to the families Torymidae (two mor- observed data are consistent with the hypothe- phospecies; Table 1) and Eurytomidae (two mor- sized causalities. This does not exclude the possi- phospecies; Table 1) with the remainder being bility of other causal structures having created members of the Pteromalidae family (one mor- the observed pattern. phospecies; Table 1). Parasitoids made up 24% Multicollinearity between variables was tested and 22% of all individuals reared from A. flori- and excluded for the gall wasp survival model bunda and A. longifolia,respectively(relative (variance inflation factors [vif] ≤1.4). Coefficient abundances of reared individuals from each mor- estimates (B) were considered significant at P- phospecies are given in Table 1). Parasitoids values <0.05. Variance explained by the models is emerged throughout the rearing period (up to reported as marginal and conditional R2 as rec- 33 weeks) from galls of both host plants (Fig. 4). ommended by Nakagawa and Schielzeth (2013). Marginal R2 considers the variance explained by Piecewise SEM the fixed effects only while conditional R2 For both host plants, gall wasp survival describes the overall variance explained by the increased significantly with gall mass per chamber model (including both fixed and random effects). but was unaffected by gall wasp density and the

❖ www.esajournals.org 7 December 2017 ❖ Volume 8(12) ❖ Article e02043 HENRIKSEN ET AL.

0.8 P < 0.001 for A. floribunda; z = 0.681, P < 0.001 ’ 0.7 for A. longifolia; Tukey s test for one- vs. three- chambered galls: z = 0.536, P < 0.001 for 0.6 A. floribunda; z = 0.679, P < 0.001 for A. longifo- 0.5 Gall wasp females lia; Fig. 3). Gall wasp males The conditional R2 was much higher than the 0.4 Parasitoids marginal R2 for both host plants (Table 2iii) indi- 0.3 cating that there is large between-tree variation in the size of a one-, two-, or three-chambered gall, 0.2

Proportion of individuals respectively (i.e., the variance explained was due 0.1 to the random rather than the fixed effects). In the 2 0.0 other models, conditional R was not much higher 2 0.8 than the marginal R (Table 2i and ii), indicating that most of the variance was explained by the 0.7 fixed effects (i.e., gall mass per chamber, gall wasp 0.6 density, and parasitism) and that the examined relationships were consistent across the sampled 0.5 trees. For example, for a given gall mass per 0.4 chamber the likelihood of survival was similar

0.3 across individual trees within each host plant spe- cies. When comparing across hosts plants, the log 0.2 odds for survival in A. floribunda galls was more Proportion of individuals 0.1 than one and a half times as high as the log odds for survival in A. longifolia galls (B = 1.32 vs. 0.0 = 02468 10 12 14 16 18 20 22 24 26 28 30 32 34 B 0.84, respectively; Fig. 5). Thus, even though Weeks of rearing the average gall mass per chamber was higher on A. longifolia than on A. floribunda (Fig. 3), survival Fig. 4. Cumulative proportion of all individuals per mass was higher in A. floribunda galls (Fig. 6). reared (three types: adult gall wasp females, adult gall When testing the independence claims of the wasp males and parasitoids) from galls of each host hypothesized causal network (variables not con- plant, that is, Acacia floribunda (top) and Acacia longifo- nected by arrows), Shipley’s d-separation test lia (bottom), over 33 weeks of rearing (starting late failed to reject the hypothetical causal network for- November 2014). Points along the lines indicate weeks mulated in Fig. 2 (A. floribunda: C = 2.76, df = 2, when galls were checked for emerged individuals. P = 0.252; A. longifolia: C = 2.66, df = 2, P = 0.265; Fig. 5). The proposed causalities therefore provide presence of parasitoids (Table 2i, Fig. 5; in this sec- a valid interpretation of the relative influences of tion, “gall wasp” and “chamber” refer to female top-down and bottom-up effects on the survival of gall wasp and female chamber unless specifically the gall wasp on both host plants. The competing mentioned as male). In the parasitism model, par- causal network, excluding the direct impact of gall asitoid presence decreased significantly with mass per chamber on gall parasitism, was rejected increasing gall mass per chamber in both host as a valid explanation of the observed tri-trophic plants (Table 2ii, Fig. 5). There was a significant interactions (see Appendix S2). difference in gall mass per chamber at the different levels of gall wasp density in the gall mass model DISCUSSION (Table 2iii, Fig. 5). For both host plants, two- and three-chambered galls did not differ in mass per Here, we show that there was a strong positive chamber (Tukey’stest:z = 0.078, P = 0.541 for relationship between gall mass per chamber A. floribunda; z = 0.002, P = 1.000 for A. longifo- (a proxy for resources provided by the host plant) lia), but they both had a significantly smaller mass and the survival of a biocontrol agent to adult- per chamber than one-chambered galls (Tukey’s hood on two host plant species in its native range, test for one- vs. two-chambered galls: z = 0.458, while gall mass per chamber was negatively

❖ www.esajournals.org 8 December 2017 ❖ Volume 8(12) ❖ Article e02043 HENRIKSEN ET AL.

Table 2. Direct and indirect top-down and bottom-up effects on larval survival for each host plant species (Acacia floribunda and Acacia longifolia).

Marginal Conditional Host plant Component model Predictor variable n df B (SE) P R2 R2

A. floribunda (i) Call wasp survival 1416 Gall mass 1.32 (0.10) <0.001 0.26 0.34 Gall wasp density (2 chambers) 1 0.28 (0.14) 0.054 Gall wasp density (3 chambers) 1 0.34 (0.20) 0.082 Parasitism 1 0.41 (0.24) 0.090 (ii) Parasitism 1416 0.12 0.14 Gall mass 1 0.77 (0.14) 0.001 (iii) Gall mass 1416 0.07 0.33 Gall wasp density (2 chambers) 1 0.45 (0.04) <0.001 Gall wasp density (3 chambers) 1 0.53 (0.07) <0.001 A. longifolla (i) Call wasp survival 679 0.30 0.42 Gall mass 1 0.84 (0.08) <0.001 Gall wasp density (2 chambers) 1 0.25 (0.21) 0.226 Gall wasp density (3 chambers) 1 0.07 (0.24) 0.771 Parasitism 1 0.16 (0.35) 0.639 (ii) Parasitism 679 0.25 0.37 Gall mass 1 0.74 (0.14) <0.001 (iii) Gall mass 679 0.05 0.25 Gall wasp density (2 chambers) 1 0.68 (0.12) <0.001 Gall wasp density (3 chambers) 1 0.68 (0.16) <0.001 Notes: Component models of the path analysis were tested as mixed models (linear mixed models and generalized linear mixed models) to account the sampling of multiple galls within trees. Coefficient estimates (B SE) with P-values above 0.05 are considered significant (shown in bold). The categorical variable gall wasp density was tested with one-chambered galls as the reference level.

A. floribunda A. longifolia

Parasitism Parasitism Gall wasp Gall wasp density density Gall wasp Gall wasp

survival survival

–0.77 –0.74

Gall mass Gall mass per chamber per chamber +1.32 +0.84 –0.45 2ch –0.68 2ch

–0.53 3ch –0.68 3ch

Fig. 5. Observed patterns of direct and indirect effects on gall wasp survival when tested as a piecewise structural equation model for each of the two host plants (Acacia floribunda and Acacia longifolia). Dashed arrows indicate non-significant partial regression estimates and full arrows indicate significant partial regression estimates. Unstandardized coefficient estimates are given for each significant partial regression and arrow thickness corresponds to the relative size of the coefficient estimates. Gall mass is defined as gall mass per female chamber. Gall wasp density is tested as a categorical variable (one, two, or three chambers) with one-chambered galls as the reference level; that is, results are shown for both two-chambered (2ch) and three-chambered galls (3ch) tested against one-chambered galls.

❖ www.esajournals.org 9 December 2017 ❖ Volume 8(12) ❖ Article e02043 HENRIKSEN ET AL.

1.00 bottom-up effects emphasizes the need for post- release management and monitoring of herbivore agent populations when host plant resource qual- ity is poor or highly variable. 0.75 Top-down and bottom-up impacts on gall wasp survival The population dynamics of gall makers are 0.50 often strongly related to bottom-up effects from host plant resources due to the influence that resources have on gall maker survival, fecundity, 0.25 and oviposition preferences (e.g., Fritz et al. 2000, Stone et al. 2002, Price 2003). Here, we found a Probability of female gall wasp survival strong relationship between within-tree variation in gall mass per chamber and gall wasp survival. This 0.00 0 246 8 bottom-up effect was dominant on both host plants, Gall mass per chamber (mg) but the likelihood of survival with increasing gall mass per chamber was higher in A. floribunda galls, Fig. 6. Probability of female gall wasp survival suggesting that A. floribunda may provide higher (95% CI) to adulthood with increasing gall mass per resource quality per unit mass of gall tissue. chamber in Acacia floribunda (black) and Acacia longifolia Despite a strong effect of gall mass per chamber (gray) galls. The increase in survival with each addi- on survival, there was no evidence that gall wasp tional gram of gall mass is higher in A. floribunda galls survival was density dependent within galls. No compared to A. longifolia galls. For each host plant, the previous study has examined density-dependent probability of survival is predicted from the gall wasp survival in A. floribunda. At the scale of individual survival component model in the piecewise SEM cover- gall, gall wasp density had no effect on pupal ing the observed range of gall sizes (i.e., up to 8.59 mg). mass in A. longifolia (Dennill et al. 1993). Thus, using a different performance measure (i.e., pupal related to gall parasitism (here and below, “gall mass), the results of Dennill et al. (1993) support wasp” and “chamber” refer to female gall wasp our finding of absence of intra-specificcompeti- and female chamber unless specifically mentioned tion within galls. The gall wasp has, however, as male). Bottom-up effects were not only directly been observed to switch to sub-optimal hosts (i.e., related to the survival of the potential biocontrol to the closely related Acacia melanoxylon R. Br. and agent, but also related to the strength of the top- lophantha (Willd.) I. C. Nielsen; down effect of parasitoids in the system. The Fabaceae) at sites with particularly high gall wasp importance of bottom-up and top-down effects, densities in South Africa (Dennill et al. 1993). and their interactions, for herbivore population Though potentially a rare event, host switching dynamics are well established (Hunter and Price could indicate that intra-specific competition can 1992, Forkner and Hunter 2000, Hunter 2001, occur at the tree level in this species (Dennill et al. Denno et al. 2002, Stiling and Moon 2005), but 1993). Herbivore performance can vary across rarely accounted for in assessments of herbivore spatial scales (McGeoch and Price 2005) and life agent biocontrol potential (Gassmann 1996, Hos- stages (Ohgushi 1995). Thus, negative density seini et al. 2010). When bottom-up effects are dependence could be an important contributor to strong, between- and within-tree variation in host gall wasp population dynamics without being plant resources are likely to be a strong predictor evident at the scale and with the performance of gall wasp survival and establishment in a target measure used here (i.e., larval survival to adult- range. Furthermore, if gall wasps are exposed to hood at the scale of individual galls). novel enemies when introduced, low resource No evidence was found for top-down impact availability could further suppress gall wasp pop- of parasitism on female gall wasp survival on ulations by increasing parasitism in small galls. A either host plant, despite the high abundance of strong dependence of biocontrol success on parasitoids among the emerged individuals (24%

❖ www.esajournals.org 10 December 2017 ❖ Volume 8(12) ❖ Article e02043 HENRIKSEN ET AL. and 22% for A. floribunda and A. longifolia, (Hymenoptera: Tenthredinidae) was due to the respectively). Each emerged parasitoid has killed combined effects of gall size and toughness in a gall wasp to complete its development (Noble excluding parasitoid species with both short and 1940). Thus, the absence of a top-down effect is long ovipositors. The dependence of gall mor- unexpected, but may be due to high levels of par- phology on the resource quality of host plant tis- asitism of male gall wasp larvae. The focus on sue does, however, suggest that there is potential females in our study may therefore obscure the for modified top-down impact across environ- true strength of the top-down impact from para- mental gradients that affect host plant quality sitoids. The large discrepancy in male-to-female such as soil nitrogen and water status (Albarracin sex ratio of reared (0.17 for both host plants) and Stiling 2006, Price and Hunter 2015). compared to dissected (0.75 and 0.69 for A. flori- The path analysis provided the first evidence bunda and A. longifolia, respectively) galls sug- for the importance of gall mass per chamber, par- gests that survival of males was low compared to asitism, and the number of chambers per gall for females in the reared galls. Previous findings of a determining gall wasp survival in the system. relatively higher exposure of male than female The tested relationships are, however, likely to chambers to parasitism due to their peripheral represent a subset of a more complex network of placement in the gall tissue (Manongi and Hoff- causal relationships. Future work may focus on mann 1995) provide further support for the idea. linking plant nutrient quality to both between- Thus, many of the parasitoids reared in the pre- and within-tree variation in gall wasp preference sent study may have emerged from male, rather and performance to further explore the determi- than female, chambers. Nonetheless, female gall nants of biocontrol success in the gall wasp. wasps are principally responsible for gall forma- Although the direct link between gall mass per tion (Noble 1940, Dorchin et al. 2009) and their chamber and the nutrient status of host plant exposure to parasitism is, therefore, critical for tissue still needs to be demonstrated for this determining the gall wasp biocontrol potential. system, the dominant within-tree effect of gall Gall mass per chamber was negatively related mass per chamber on gall wasp survival that we to gall parasitism (Table 2ii, Fig. 5), suggesting found suggests that significant variation exists in that bottom-up effects from gall size can influence gall resource availability in this system. the strength of top-down effects on gall wasp sur- vival at the gall level. An effect of gall morphol- Implications for biological control ogy on parasitoid access to gall chambers has Indicators of resource availability in this gall been observed in a range of different gall makers wasp–host plant interaction, such as gall size, are (Weis et al. 1985, Price and Clancy 1986, Albar- likely to be important predictors for its biocontrol racin and Stiling 2006). Gall size is also one of the potential in an introduced range. Our results sug- gall morphological traits that is known to be gest that bottom-up effects of gall size will be related to the bottom-up effects of host plant qual- strongly related to gall wasp survival and through ity between and within individual plants (Price this, the ability of a gall wasp population to estab- and Clancy 1986, Albarracin and Stiling 2006). lish successfully on invasive host species. Gall The rate of development of galls determines the makers are often used in biocontrol due to of their temporal window of opportunity for parasitoids ability to suppress plant reproduction, especially with short ovipositors to reach gall chambers among species that affect flower development while the final size of the fully developed gall can and seed set (Harris and Shorthouse 1996). limit the access of later attacking parasitoids with Trichilogaster acaciaelongifoliae is generally consid- longer ovipositors (Price and Clancy 1986, Albar- ered to be a successful biocontrol agent in South racin and Stiling 2006). The weak top-down Africa (Impson et al. 2011), where it has become effects from parasitoids that are often found in widespread across the invasive range of A. longi- gall makers (Stone et al. 2002, Price 2003) may be folia (Veldtman et al. 2010). Impacts on host popu- due to the effectiveness of gall tissue in restricting lations can, however, be inconsistent, with seed parasitoid access. Price (2003) suggested that the production being unaffected in some localities weak influence of parasitism on the population (Dennill and Gordon 1990, Dennill et al. 1999). dynamics of the sawfly Euura lasiolepis Smith This low effectiveness has been ascribed to

❖ www.esajournals.org 11 December 2017 ❖ Volume 8(12) ❖ Article e02043 HENRIKSEN ET AL. regional variability in aridity and temperature, A strong impact of bottom-up effects on suggesting that bottom-up effects do indeed influ- gall wasp survival suggests that post-release man- ence biocontrol potential in the introduced range. agement may need to be implemented in regions In areas where gall mass per chamber is low or where plant condition may result in low gall mass highly variable, the composition of native para- per chamber, or where it is highly variable. This sitoid assemblages may also be critical to biocon- could involve fertilization of low-quality hosts, or trol success. Even though parasitism was not continued addition of the biocontrol agent over significantly related to female survival in the time. Biocontrol agent establishment under low- native range studied here, plant stress or novel resource conditions has been shown to benefit enemy assemblages could result in significant top- from weed fertilization due to increased herbivore down effects in the introduced ranges of potential survival and fecundity (Van Hezewijk et al. 2008, biocontrol agents (Veldtman et al. 2011). The high Hovick and Carson 2015). However, as Hovick exposure of male chambers to parasitism is and Carson (2015) showed, if resource addition is anotherpotentiallyimportantaspectoftop-down too high, the increased resistance of the weed to impacts on gall wasp populations. In the short herbivore damage could reduce the impact of the term, parthenogenetic females sustain population biocontrol agent while increasing weed abun- growth and dispersal to new host trees. Long-term dances (Hovick and Carson 2015). Fertilization of exposure of males to high levels of parasitism weeds is therefore potentially risky and requires could, however, make the gall wasp more vulnera- careful management. An alternative strategy may ble to environmental variability if asexual repro- be ongoing targeted release of additional individ- duction affects the adaptive potential of the uals of the biocontrol agent (i.e., agent augmenta- population (West et al. 1999). This could be espe- tion). This is likely to be necessary in the early cially critical if gall wasps experience a genetic bot- stages of biocontrol establishment, when agent tleneck when introduced to a target range, which populations are small and vulnerable to stochastic is not uncommon for introduced biocontrol agents changes in the environment (Grevstad 1996). (Debach and Rosen 1991). In the introduced range The need for post-release management of bio- in South Africa, high levels of parasitism (47.6%) control agent populations can only be assessed havebeenfoundinassociationwiththeintro- through continued monitoring of released agents. duced gall wasp (McGeoch and Wossler 2000) and Long-term post-release monitoring of released a food web of similar structure and complexity to agents is often neglected in biocontrol programs theoneinthenativerangenowexists(Veldtman despite its importance for the evaluation of bio- et al. 2011). Veldtman et al. (2011) also found 33% control success (Morin et al. 2009). The strong of the acquired natural enemies of the gall wasp to correlation between gall mass per chamber and belong to the same families as its native enemies. gall wasp survival found here provides a poten- High similarity between native and novel food tial cost-effective tool for such post-release moni- webs suggests that the gall wasps could be experi- toring of gall wasp agent populations. Gall mass encing equivalent top-down impacts in native and per chamber can be readily measured across mul- introduced ranges. However, the strength of the tiple sites and used as an indicator of relative gall top-down effect will depend on the ability of novel wasp survival. This measure could be used to enemies to access gall chambers. If novel enemies predict biocontrol agent population growth or to in the introduced range on average have longer identify localities and years in which additional ovipositors, the impact of top-down effects could management of agent populations is needed. be elevated, especially when gall mass per cham- ber is low. Thus, the choice of optimal release areas CONCLUSIONS should include evaluations of local host plant resource quality and quantity. Biocontrol is more Here,wehaveshownthatthesurvivalofthe likely to succeed if both herbivore agent popula- gall wasp Trichilogaster acaciaelongifoliae on two tion dynamics and the suitability of the release host plant species is strongly affected by within- area are assessed prior to agent release (Sheppard tree variation in gall resource availability and that, 2003, Zalucki and van Klinken 2006, Lopez-N unez~ by comparison, parasitism plays a minor role. The et al. 2017). effect of parasitism on this gall wasp is, however,

❖ www.esajournals.org 12 December 2017 ❖ Volume 8(12) ❖ Article e02043 HENRIKSEN ET AL. mediated by gall mass per chamber and may Briese, D. T. 2004. Weed biological control: applying change with variation in host plant quality and science to solve seemingly intractable problems. the composition of local parasitoid assemblages. Australian Journal of Entomology 43:304–317. Strong bottom-up effects of factors that determine Cornelissen, T., G. W. Fernandes, and J. Vasconcellos- the survival of herbivorous biocontrol agents have Neto. 2008. Size does matter: variation in herbivory between and within plants and the plant vigor implications for spatio-temporal variability in the hypothesis. Oikos 117:1121–1130. success of biocontrol programs and should be con- Cortesero, A. M., J. O. Stapel, and W. J. Lewis. 2000. sidered in pre-release assessments and post-release Understanding and manipulating plant attributes monitoring of herbivore insect biocontrol agents. to enhance biological control. Biological Control 17:35–49. ACKNOWLEDGMENTS CSIRO Division of Entomology. 1991. The insects of Aus- tralia: a textbook for students and research workers. This work was supported by Parks Victoria Research Cornell University Press, Ithaca, New York, USA. Partners Panel (grant number RPP1314P12) and an Aus- Debach, P., and D. Rosen. 1991. Biological control by tralian Government Research Training Program (RTP) natural enemies. Second edition. Cambridge Scholarship. We wish to thank A. Guilluame and S. University Press, Cambridge, UK. Fialkowski for help with field and laboratory work, and Dennill, G. B. 1985. The effects of the gall wasp Trichilo- Z. Squires, G. Latombe, C. Dickson, G. Horner, and K. gaster-acaciaelongofoliae (Hymenoptera, Pteromali- Monro for discussions and comments on the manuscript. dae) on reproductive potential and vegetative growth of the weed Acacia-longifolia. Agriculture LITERATURE CITED Ecosystems and Environment 14:53–61. Dennill, G. B., D. Donnelly, and S. L. Chown. 1993. Abrahamson, W. G., and A. E. Weis. 1997. Evolution- Expansion of host-plant range of a biocontrol agent ary ecology across three trophic levels: goldenrods, Trichilogaster-acaciaelongifolia (Pteromalidae) released gallmakers and natural enemies. Princeton Univer- against the weed Acacia longifolia in South Africa. sity Press, Princeton, New Jersey, USA. Agricultural Ecosystems and Environment 43:1–10. Adair, R. J. 2008. Biological control of Australian Dennill, G. B., D. Donnelly, K. Stewart, and F. A. C. native plants, in Australia, with an emphasis on Impson. 1999. Insect agents used for the biological . Muelleria 26:67–78. control of Australian Acacia species and Paraseri- Agrawal, A. A. 2000. Mechanisms, ecological conse- anthes lophantha (Willd.) Nielsen (Fabaceae) in quences and agricultural implications of tri-trophic South Africa. African Entomology Memoir 1:45–54. interactions. Current Opinion in Plant Biology Dennill, G. B., and A. J. Gordon. 1990. Climate-related 3:329–335. differences in the efficacy of the Australian gall Albarracin, M. T., and P. Stiling. 2006. Bottom-up and wasp (Hymenoptera: Pteromalidae) released for top-down effects on insect herbivores do not vary the control of Acacia longifolia in South Africa. Envi- among sites of different salinity. Ecology 87:2673– ronmental Entomology 19:130–136. 2679. Denno, R. F., C. Gratton, M. A. Peterson, G. A. Australian Government Department of the Environment Langellotto, D. L. Finke, and A. F. Huberty. 2002. and Energy. 2016. Interim biogeographic regionali- Bottom-up forces mediate natural-enemy impact in sation for Australia (IBRA) version 7. http://www. a phytophagous insect community. Ecology 83: environment.gov.au/land/nrs/science/ibra 1443–1458. Bashford, R. 2004. The insects associated with galls Denoth, M., and J. H. Myers. 2005. Variable success of formed by Trichilogaster acaciaelongifoliae (Frog- biological control of Lythrum salicaria in British gatt) (Hymenoptera: Pteromalidae) on Acacia Columbia. Biological Control 32:269–279. species in . Australian Entomologist 31: Dorchin, N., J. H. Hoffman, W. A. Stirk, O. Novak, 5–12. M. Strnad, and J. van Staden. 2009. Sexually dimor- Bates, D., M. Maechler, B. Bolker, and S. Walker. 2015. phic gall structures correspond to differential lme4: linear mixed-effects models using eigen and phytohormone contents in male and female wasp S4_. R package version 1.1-9. https://CRAN.R-pro larvae. Physiological Entomology 34:359–369. ject.org/package=lme4 Forkner, R. E., and M. D. Hunter. 2000. What goes up Boucek, Z. 1988. Australian Chalcidoidea (Hymenop- must come down? Nutrient addition and predation tera) – A biosystematics revision of genera of four- pressure on oak herbivores. Ecology 81:1588–1600. teen families, with a reclassification of species. Fritz, R. S., B. A. Crabb, and C. G. Hochwender. 2000. C.A.B. International, Wallingford, UK. Preference and performance of a gall-inducing

❖ www.esajournals.org 13 December 2017 ❖ Volume 8(12) ❖ Article e02043 HENRIKSEN ET AL.

sawfly: a test of the plant vigor hypothesis. Oikos Ito, M., and N. Hijii. 2004. Roles of gall morphology in 89:555–563. determining potential fecundity and avoidance of Gassmann, A. 1996. Classical biological control of parasitoid attack in Aphelonyx glanduliferae. Journal weeds with insects: a case for emphasizing agent of Forest Research 9:93–100. demography. Pages 171–175 in V. C. Moran and Lakatos, K. T., Z. Laszl o, and B. Tothm eresz. 2017. J. H. Hoffmann, editors. Proceedings of the IX Disturbance induced dynamics of a tritrophic novel International Symposium on Biological Control of ecosystem. Bulletin of Entomological Research. Weeds. University of Cape Town, Stellenbosch, https://doi.org/10.1017/S0007485317000621 South Africa. Laszl o, Z., and B. Tothm eresz. 2013. The enemy hypoth- Grevstad, F. S. 1996. Establishment of weed control esis: correlates of gall morphology with parasitoid agents under the influences of demographic attack rates in two closely related rose cynipid galls. stochasticity, environmental variability, and Allee Bulletin of Entomological Research 103:326–335. effects. Pages 261–267 in V. C. Moran and J. H. Hoff- Lawrence, J. F., and A. Slipinski. 2013. Australian bee- mann, editors. Proceedings of the IX International tles: morphology, classification and keys. CSIRO Symposium on Biological Control of Weeds. Univer- Publishing, Collingwood, Victoria, Australia. sity of Cape Town, Stellenbosch, South Africa. Lawton, J. H. 1985. Ecological theory and choice of bio- Harris, P., and J. D. Shorthouse. 1996. Effectiveness of logical control agents. Pages 13–26 in E. S. Delfosse, gall inducers in weed biological control. Canadian editor. Proceedings of the VI International Sympo- Entomologist 128:1021–1055. sium on Biological Control of Weeds. Agriculture Hill, R. 2005. Prospects for the biological control of Syd- Canada, Ottawa, Ontario, Canada. ney golden Wattle, Acacia longifolia,usingTrichilogaster Lefcheck, J. S. 2015. piecewiseSEM: piecewise struc- acaciaelongifoliae and Melanterius ventralis. Landcare tural equation modeling in R for ecology, evolu- research contract report LCO506/009. Landcare tion, and systematics. Methods in Ecology and Research New Zealand, Lincoln, New Zealand. Evolution 7:573–579. Hoffmann, J. H., F. A. C. Impson, V. C. Moran, and Lefcheck, J. S., and J. E. Duffy. 2015. Multitrophic func- D. Donnelly. 2002. Biological control of invasive tional diversity predicts ecosystem functioning in golden wattle trees () by a gall experimental assemblages of estuarine consumers. wasp, Trichilogaster sp. (Hymenoptera: Pteromali- Ecology 96:2973–2983. dae), in South Africa. Biological Control 25:64–73. Lopez-N unez,F.A.,R.H.Heleno,S.Ribeiro,~ Hosseini, M., A. Ashouri, A. Enkegaard, W. W. H. Marchante, and E. Marchante. 2017. Four-trophic Weisser, S. H. Goldansaz, M. N. Mahalati, and level food webs reveal the cascading impacts of an H. R. S. Moayeri. 2010. Plant quality effects on invasive plant targeted for biocontrol. Ecology intraguild predation between Orius laevigatus and 98:782–793. Aphidoletes aphidimyza. Entomologia Experimen- Manongi, F. S., and J. H. Hoffmann. 1995. The inci- talis et Applicata 135:208–216. dence of parasitism in Trichilogaster acaciaelongifoliae Hothorn, T., F. Bretz, and P. Westfall. 2008. Simultane- (Froggatt) (Hymenoptera: Pteromalidae), a gall- ous inference in general parametric models. Bio- forming biological control agent of Acacia longifolia metrical Journal 50:346–363. (Andr.) Willd. (Fabaceae) in South Africa. African Hovick, S. M., and W. P. Carson. 2015. Tailoring Entomology 3:147–151. biocontrol to maximize top-down effects: on the Marchante, H., F. A. Lopez-N unez,~ H. Freitas, J. H. importance of underlying site fertility. Ecological Hoffmann, F. Impson, and E. Marchante. 2017. First Applications 25:125–139. report of the establishment of the biocontrol agent Hunter, M. D. 2001. Multiple approaches to estimating Trichilogaster acaciaelongifoliae for control of invasive the relative importance of top-down and bottom- Acacia longifolia in Portugal. EPPO Bulletin 47:274–278. up forces on insect populations: experiments, life McGeoch, M. A., and P. W. Price. 2005. Scale-dependent tables, and time-series analysis. Basic and Applied mechanisms in the population dynamics of an insect Ecology 2:295–309. herbivore. Oecologia 144:278–288. Hunter, M. D., and P. W. Price. 1992. Playing chutes McGeoch, M. A., and T. C. Wossler. 2000. Range and ladders: heterogeneity and the relative roles of expansion and success of the weed biocontrol bottom-up and top-down. Ecology 73:724–732. agent Trichilogaster acaciaelongifoliae (Froggatt) Impson, F. A. C., C. A. Kleinjan, J. H. Hoffman, J. A. (Hymenoptera: Pteromalidae) in South Africa. Post, and A. R. Wood. 2011. Biological control of African Entomology 8:273–280. Australian Acacia species and Paraserianthes lophan- Morin,L.,A.M.Reid,N.M.Sims-Chilton,Y.M.Buckley, tha (Willd.) Nielsen (Mimosaceae) in South Africa. K. Dhileepan, G. T. Hastwell,T.L.Nordblom,and African Entomology 19:186–207. S. Raghu. 2009. Review of approaches to evaluate

❖ www.esajournals.org 14 December 2017 ❖ Volume 8(12) ❖ Article e02043 HENRIKSEN ET AL.

the effectiveness of weed biological control agents. Improving the selection, testing and evaluation of Biological Control 51:1–15. weed biological control agents. CRC for Australian Nakagawa, S., and H. Schielzeth. 2013. A general and Weed Management, Glen Osmond, South Aus- simple method for obtaining R2 from generalized tralia, Australia. linear mixed-effects models. Methods in Ecology Shipley, B. 2009. Confirmatory path analysis in a gen- and Evolution 4:133–142. eralized multilevel context. Ecology 90:363–368. Noble, N. S. 1940. Trichilogaster acaciae-longifoliae (Frog- Stiling, P., and D. C. Moon. 2005. Quality or quantity: the gatt) (Hymenopt. Chalcidoidea), a wasp causing directandindirecteffectsofhostplantsonherbivores galling of the flower-buds of Acacia longifolia and their natural enemies. Oecologia 142:413–420. Willd., A. floribunda Sieber and A. sophorae R. BR. Stone, G. N., K. Schonrogge,€ R. J. Atkinson, D. Bellido, Transactions of the Royal Entomological Society of and J. Pujade-Villar. 2002. The population biology London 90:13–38. of oak gall wasps (Hymeoptera: Cynipidae). Ohgushi, T. 1995. Adaptive behaviour produces Annual Review of Entomology 47:633–668. stability in herbivore ladybeetle populations. Pages Sumerford,D.V.,W.G.Abrahamson,andA.E.Weis. 303–319 in N. Cappuccino and P. W. Price, editors. 2000. The effects of drought on the Solidago altissima- Population dynamics. Academic Press, San Diego, Eurosta solidaginis-natural enemy complex: population California, USA. dynamics,localextirpations,andmeasuresofselec- Pinheiro, J., D. Bates, S. DebRoy, D. Sarkar, and R Core tion intensity on gall size. Oecologia 122:240–248. Team. 2015. _nlme: linear and nonlinear mixed Van Hezewijk, B. H., R. A. De Clerck-Floate, and J. R. effects models_. R package version 3.1-121. http:// Moyer. 2008. Effect of nitrogen on the preference CRAN.R-project.org/package=nlme and performance of a biological control agent for Price, P. W. 1991. The plant vigor hypothesis and herbi- an invasive plant. Biological Control 46:332–340. vore attack. Oikos 62:244–251. Veldtman, R., S. L. Chown, and M. A. McGeoch. 2010. Price, P. W. 2003. Macroevolutionary theory on Using scale–area curves to quantify the distribution, macroecological patterns. Cambridge University abundance and range expansion potential of an inva- Press, Cambridge, UK. sive species. Diversity and Distributions 16:159–169. Price, P. W., C. E. Bouton, P. Gross, B. A. McPheron, Veldtman, R., T. F. Lado, A. Botes, S. Proches, A. E. J. N. Thompson, and A. E. Weis. 1980. Interactions Timm, H. Geertsema, and S. L. Chown. 2011. among three trophic levels: influence of plants on Creating novel food webs on introduced Australian interactions between insect. Annual Review of acacias: indirect effects of galling biological control Ecology and Systematics 11:41–65. agents. Diversity and Distributions 17:958–967. Price, P. W., and K. M. Clancy. 1986. Interactions Walsh, N. G., and T. J. Entwisle. 1996. Flora of Victoria. among three trophic levels – gall size and para- Volume 3: Dicotyledons Winteraceae to Myrtaceae. sitoid attack. Ecology 67:1593–1600. Inkata, Melbourne, Victoria, Australia. Price, P. W., and M. D. Hunter. 2015. Population Weis, A. E., W. G. Abrahamson, and K. D. McCrea. dynamics of an insect herbivore over 32 years are 1985. Host gall size and oviposition success by the driven by precipitation and host-plant effects: test- parasitoid Eurytoma gigantea. Ecological Entomol- ing model predictions. Environmental Entomology ogy 10:341–348. 44:463–473. West, S. A., C. M. Lively, and A. F. Read. 1999. A plu- Prinsloo, G. L., and O. C. Neser. 2007. Revision of the ralist approach to sex and recombination. Journal pteromalid wasp Trichilogaster Mayr (Hyme- of Evolutionary Biology 12:1003–1012. noptera: Chalcidoidea): gall-inducers on Australian Zalucki, M. P., and R. D. van Klinken. 2006. Predicting acacias. African Entomology 15:161–184. population dynamics of weed biological control R Core Team. 2015. R: a language and environment for agents: Science or gazing into crystal balls? Aus- statistical computing. R Foundation for Statistical tralian Journal of Entomology 45:331–344. Computing, Vienna, Austria. Zuur, A. F., E. N. Ieno, N. J. Walker, A. A. Saveliev, Sheppard, A. W. 2003. Prioritising agents based on and G. M. Smith. 2009. Mixed effects models and predicted efficacy: beyond the lottery approach. extensions in ecology with R. Springer, New York, Pages 11–21 in H. S. Jacob and D. T. Briese, editors. New York, USA.

SUPPORTING INFORMATION

Additional Supporting Information may be found online at: http://onlinelibrary.wiley.com/doi/10.1002/ecs2. 2043/full

❖ www.esajournals.org 15 December 2017 ❖ Volume 8(12) ❖ Article e02043