Insect colonies and life cycles

Laboratory colonies of C. cactorum were maintained to provide host larvae for testing parasitism by G. legneri and A. opuntiarum. Insect rearing was conducted under 25°C, 70% RH, and a photoperiod of 16:8 (L:D) at the Fundación para el Estudio de Especies Invasivas (FuEDEI), Hurlingham, Buenos Aires, Argentina. Cactoblastis cactorum larvae were originally collected in a large O. ficus-indica plantation located near Santiago del Estero city, Santiago del Estero province, Argentina (S 27° 62′ 24.2″, W 62° 34′ 39.1″). Larvae were transported to the laboratory to initiate a laboratory colony of the cactus moth. Females lay eggs as a linear structure called an ‘eggstick’ (Dodd, Reference Dodd1940) that hatch simultaneously; larvae gregariously chew a hole to enter the plant and tunnel through the cladode while feeding. Larvae complete their development by building galleries as they feed on soft interior tissues and discard their feaces outside the plant. The final (sixth) instars leave the cladode, spin a cocoon, and pupate inside the cocoon mostly within plant litter near the base of the host plant. The complete life cycle takes from 4 to 6 months, depending on the host plant species, giving rise to two or three generations per year (Zimmermann et al., Reference Zimmermann, Bloem and Klein2004; Varone et al., Reference Varone, Aguirre, Lobos, Ruiz Pérez, Hight, Palottini, Guala and Logarzo2019).

Females of A. opuntiarum came from wild larvae of C. cactorum collected in the same O. ficus-indica plantation during the last host larval generation before winter, previously determined to have high parasitism rates (Varone et al., Reference Varone, Aguirre, Lobos, Ruiz Pérez, Hight, Palottini, Guala and Logarzo2019). Larvae were lab-reared in groups of 50 until pupation when parasitised larvae were identifiable. Parasitised C. cactorum were individually stored in 20 ml plastic cups and checked daily until emergence of the adult parasitoids. Upon wasp emergence, females were confined with double the amount of males (4–8 females with 8–16 males) for 24 h to allow mating in a 3 L plastic canning jar with a modified 80-mesh lid. Added to the mating jar were a slice of O. ficus-indica (6 × 10 cm), a tablespoon of C. cactorum larval frass to stimulate female oviposition (Varone et al., Reference Varone, Mengoni Goñalons, Faltlhauser, Guala, Wolaver, Srivastava and Hight2020), and a damp strip of paper towel (1 × 3 cm) saturated with water and honey to provide moisture and food.

Individuals of G. legneri were obtained from the Instituto Nacional de Tecnologia Agropecuaria mass-rearing institute located in Alto Valle, Río Negro, Argentina. Mass-rearing of this parasitoid is carried out with fifth instar larvae of Galleria melonella (L.) (Lepidoptera: Pyralidae) as hosts because oviposition is higher in this species and produces a more efficient generational increase (Garrido et al., Reference Garrido, Cichón, Lago, Navarro, Herrera and Becerra2018b). Parasitoids were received at FuEDEI as pupae in 20 ml plastic cups. When female parasitoids emerged, they were immediately used in the experiments. Copulation was ensured because males of G. legneri emerge before females and enter the female cocoons for mating. After mating, the female seeks out a host larva and injects a poison through its stinger, causing paralysis that facilitates oviposition (Gordh et al., Reference Gordh, Woolley and Medved1983).

Study 1: Lethal effect of G. legneri on C. cactorum

Three experiments were conducted, with increasing experimental spatial scale, to assess the lethal effect of G. legneri on C. cactorum larvae (fig. 1a–c). The first experiment was performed to test the ability of G. legneri to attack (kill and/or parasitise) C. cactorum in the presence or absence of a small refuge for the host larvae, and with different larval instars exposed to the parasitoid. Although C. cactorum larvae have internal feeding habits in the cladodes, they occasionally exit the cladode, either to move to a new one or to pupate, exposing themselves to parasitoids. Cactoblastis cactorum larvae were confined in a 300 ml vented container with or without a piece of cactus and with a single mated G. legneri (fig. 1a) female. Treatments with cacti present in the container consisted of either three to five second instar larvae (L2) or three to five fourth instar larvae (L4) together with a 3 × 3 cm cube-shaped piece of O. ficus-indica that served as a refuge for the host larvae. Larvae were placed together with the cactus cubes 24 h before the release of the female parasitoid, allowing larvae to feed and penetrate the cactus. For the treatment without cactus cubes (no refuge), only L4 host larvae were placed in the 300 ml test arena because they could survive starvation during the parasitoid exposure time. Female parasitoids were confined with larvae for 24 h and removed. Between 10 and 12 replicates of each treatment were conducted. An additional set of five replicates with three to five L4 host larvae plus cactus cubes, without parasitoids, served as controls. Cactoblastis cactorum larvae were fed with fresh O. ficus-indica cladodes as needed until they either completed development and pupated, or died. Containers were checked three times a week to record parasitised, dead, or pupated larvae. Parasitised larvae were isolated in 20 ml plastic cups until the emergence of G. legneri adults. The number and sex ratio of adult wasps per container were recorded.

Figure 1. Experimental designs to assess the lethal effect of Goniozus legneri on Cactoblastis cactorum larvae with increasing experimental spatial scale (a to c), and the interaction between G. legneri and Apanteles opuntiarum (d).

A second experiment tested the mortality of C. cactorum with increasing numbers of G. legneri females and larger pieces of O. ficus-indica (fig. 1b). Because C. cactorum larvae reduced cactus mass by feeding inside the cladode, the piece of cactus was weighed before and after the trial to estimate weight loss due to feeding. Groups of 20 L4 C. cactorum larvae were placed in an 8 l plastic vented container with half a cladode of O. ficus-indica of similar size, and larvae were allowed to penetrate the plant for 24 h. After 24 h, 5, 10, 15, or 20 G. legneri mated females were added to the containers with a solution of honey and water as a food source. After the death of all the females, containers were checked three times a week to record parasitised, dead, and pupated C. cactorum larvae until all larval hosts completed development or died. Groups of 20 larvae feeding on similar sized cladodes, unexposed to parasitoids, were placed in 8 l containers and served as controls for larval survival. Ten replicates of each treatment (and control) were conducted.

The third experiment was conducted in cages to evaluate the lethal effect of G. legneri on C. cactorum when the larvae were feeding inside live prickly pear plants (fig. 1c). Two-year-old O. ficus-indica potted plants (with two cladodes each) were infested with 40 L4 larvae of C. cactorum. Prior to the release of the parasitoids, and so that they could penetrate into the plant, larvae were placed in an open 20 ml plastic cup, and attached to the plant in a manner that the larvae were in contact with the surface of a cladode. Potted plants were placed inside a 120 × 50 × 50 cm rectangular mesh cage. After larvae feed for 48 h, 40 G. legneri mated females were released in the centre of the cage. A solution of honey and water was also provided as food for the females. Five replicates were conducted and five plants with C. cactorum larvae, without parasitoids, were used as controls. Since larvae leave the plant to pupate, a piece of cloth was placed at the base of each plant, which served as a pupation substrate. After the parasitoids' death, cages were checked three times a week to record the number of pupating C. cactorum, meaning that those larvae survived exposure to the parasitoids. Parasitised or dead larvae remaining inside the plants were not recorded since this would have required the destruction of the entire living plant system. After all larvae pupated (or died), damage to the plants produced by larval feeding was estimated as the percentage of visible surface damage to plant tissue (categories of 0, 25, 50, and 100% surface damage). The damage estimate, along with the number of successfully pupated larvae, were compared between treatments with and without G. legneri to identify the impact of the parasitoid on the test host larvae.

Study 2: Interaction between G. legneri and A. opuntiarum

The mortality produced by G. legneri and A. opuntiarum, acting together or separately, was tested with increasing amounts of C. cactorum host larval densities. In addition, the influence of refugia on parasitoid interaction was included in the host density tests; with a piece of cactus that served as a refuge for the larvae or without a piece of cactus (no refugia) (fig. 1d). Figure 2 highlights the trophic relationships between species that were investigated in this study.

Figure 2. Venn diagram of trophic relationships in this study. The yellow circles are the species, and the blue lines indicate the trophic relationships between species. The light pink filled circle indicates the entire group of hosts/prey of Cactoblastis cactorum (Cc) exposed to natural enemies. The medium-light pink circle indicates the subgroup attacked only by Apanteles opuntiarum (Ao, shown as Cc -> Ao), while the medium-dark pink circle indicates the subgroup attacked only by Goniozus legneri (Gl, shown as Cc -> Gl). The dark pink area indicates the Cc larvae parasitised by Ao and attacked by Gl (Cc -> Ao -> Gl).

The interaction study consisted of four treatments: host larvae exposed to only one species of the two parasitoids, host larvae exposed to one parasitoid species for 24 h and then the sequential exposure to the other parasitoid species (G. legneri followed by A. opuntiarum, and vice versa). In all four treatments, five larval densities (one repetition each) of 20, 25, 30, 35, and 40 L3 C. cactorum were confined with a single female parasitoid for 24 h in a 500 ml vented arena. In the treatments with sequential exposure of the two species of parasitoids, the female was also with the larvae for 24 h. After the first 24 h with one parasitoid species, the parasitoid was removed and the other species was immediately introduced for another 24 h. A piece of cactus was added in the refugia-present design when host larvae were placed in the arena, allowing them to enter and use it as food and refuge (fig. 1d). The same design was repeated (four treatments of parasitoids exposed to five larval densities) but only a slim piece of cactus was available as a food source, insufficient in thickness to be used as a refuge. At the end of all parasitoid 24 h exposures, host larvae were transferred to 3 l vented containers with a more significant piece of O. ficus-indica cladode for feeding and development. Larvae were checked thrice weekly to record parasitised, dead, or pupated larvae.

Data analysis

The lethality of G. legneri on C. cactorum was analyzed in the first and third experiments of Study 1 which compared larval mortality among treatments. For the first experiment, the number of G. legneri adults and the sex ratio of the F1 that emerged from the parasitised C. cactorum larvae were compared. All comparisons were made using a Kruskal–Wallis test because of the non-normal distribution of the data, and groups were identified post hoc using Duncan's test. For the second experiment of Study 1, a linear regression was conducted between larval survival, or relative loss of cladode weight, and the increasing number of female G. legneri exposed. All the analyses were conducted in Infostat version 2020e (Di Rienzo et al., Reference Di Rienzo, Casanoves, Balzarini, Gonzalez, Tablada and Robledo2008).

The interaction between G. legneri and A. opuntiarum of Study 2 was evaluated by measuring the effectiveness in the order of the release of the species causing host mortality through the development of a series of multiplicative models of cumulative mortality coupled with functional responses. The best-explaining model in terms of more explanatory power and less complexity was selected by means of the Schwartz criterion, or Bayesian Information Index, called BIC hereon (Schwarz, Reference Schwarz1978).

Model formulation

The starting point of the model development was a null model in which the host mortality is constant:

(1)$$N_d\sim B( {N_o, \;p_d} ) $$

where N_o is the number of offered hosts and the number of dead hosts (N_d) is a random variable which follows a binomial distribution (B), with a probability of death equal to p_d. To add the effect of parasitoids, we modified the parameter p_d. The mortality caused by a given parasitoid i is:

(2)$$p_r( q ) = ( {q_i > 0} ) p_{di}$$

where p_r is the resulting mortality caused by parasitoid i, q_i is the number of parasitoids of species i in the experimental arena (zero or one), and p_di is the mortality caused by parasitoid i. If parasitoid i is absent, q_i is zero, and then, the resulting mortality is zero, otherwise, p_r is p_di.

As presented here, equation 2 assumes that the parasitoids follow a type-I functional response, with a constant attack rate resulting in a constant proportion of hosts attacked regardless of the number of hosts offered (Holling, Reference Holling1959). More realistically, if the proportion of attacked hosts decreases with the number of hosts (N) because the handling time of the host is required, it reaches a saturation curve. Replacing equation 2 with a type-2 functional response model (Holling, Reference Holling1959):

(3)$$p_r( {q_i, \;N} ) = q_ip_p/ ( {1 + p_phNq_i} ) $$

where q_i is the number of parasitoids in the experiment, p_p is the attack rate, and h is the handling time. Because at very low N, p_r is almost the same as p_r from equation 2, in the following equations all the p_r(q_i) or p_r(q_i, N) will be called p_r for simplicity.

If the ‘natural’ mortality of equation 1 is combined with mortality caused by parasitoids from equations 2 and 3, the following model is:

(4)$$\eqalign{& p_{di}( {q_i} ) = 1 - ( {1-p_o} ) ( {1- p_r} ) \cr & {\rm with} \cr & N_d\sim B( {N_o, \;p_{di}( {q_i} ) } ) } $$

where p _o is the natural mortality of the hosts. Consequently, the proportion of the dead hosts is given by those that remain after removing hosts that survived the other causes of mortality and the action of parasitoids.

The next model considers that both parasitoid species cause different mortality levels, thus the second term from equation 2 is incorporated into a new equation:

(5)$$\eqalign{& p_{d12}( {q_1, \;q_2} ) = 1 - ( {1-p_o} ) ( {1-p_{r1}( {1-p_{r2}} ) } ) \cr & N_d\sim B( {N_o, \;p_{d12}( {q_1, \;q_2} ) } ) } $$

where p_r1 and p_r2 are the proportion of hosts attacked by parasitoids 1 and 2, respectively. Given there are two parasitoids q₁ and q₂ acting sequentially the expected proportion of dead hosts is given by the remaining hosts that survived the natural cause of mortality and both parasitoids.

The following model considered that the mortality might also be affected by the availability of refuges (R) (pieces of cacti available in the experimental arena) and that the presence of refugees decreases the hosts' finding by the parasitoid at a certain ratio r. According to the function r(R) = r (R = 0), if there are no refuges R is zero, otherwise, R is one. In the case of R = 0, the mortality increases by r. Equation 4, is then modified as follows:

(6)$$p_{d12}( {q_1, \;q_2} ) = ( {1-r ( R ) } ) ( 1 - ( {1-p_o} ) ( {1-p_r( {q_1} ) ( {1-p_r( {q_2} ) } ) } ) $$

Finally, it is considered that the mortality might increase more than proportionally by the effect of the second parasitoid in the sequence:

(7)$$p_{d12}( {q_1, \;q_2} ) = ( {1-r} ) ( {1 - ( {1-p_o} ) ( {1-p_r( {q_1} ) ( {1-p_r( {q_2} ) } ) } ) p_{e12}} ) $$

where p_e12, is the extra mortality caused by the parasitoid 1, given that the host was first exposed to the parasitoid 2. Alternatively, the mortality might increase less than proportionally after the second parasitoid is added, for example, if there is a preference to attack hosts previously attacked as a form of interference competition. Thus the equation modifies as follows:

(8)$$p_{d12}( {q_1, \;q_2} ) = ( {1-r} ) ( {1 - ( {1-p_o} ) ( {1-p_r( {q_1} ) ( {1-p_r( {q_2} ) p_{e12}} ) } ) } ) $$

where p_e12 is the mortality caused by the second parasitoid, given that the attacked hosts were previously attacked by the first parasitoid.

Numerical methods (model selection and parameter calculation)

The data analysis was performed in two steps; first, a model selection procedure was performed to select the combination that produced the best explanatory power in terms of the log-likelihood of the model, and less complexity, in terms of the number of parameters. Second, the parameters of the selected model and their statistical distribution were calculated using Monte Carlo methods. A list of candidate models was created using different combinations of the equations described above (table 1). A Markov Chain Monte Carlo procedure using the Metropolis-Hastings algorithm was performed (MCMC) for all the candidate models, according to Gelman et al. (Reference Gelman, Carlin, Stern, Rubin, Chatfield, Tanner and Zidek2003). A total of 20,000 iterations were discarded, as a burn-in for optimisation of the parameters, and 50,000 iterations for the parameters and information indexes calculation, named traces.

Table 1. List of models proposed to evaluate the interaction between the parasitoids Goniozus legneri and Apanteles opuntiarum on the mortality of Cactoblastis cactorum (see model selection procedure)

Pr is the mortality caused by a parasitoid, BIC is the Schwarz criterion, and gcd is the generalized coefficient of determination. In bold is the selected model after the procedure.

The BIC information criterion was calculated using these traces that contained the values of the parameters and the log-likelihood function at each iteration (Schwarz, Reference Schwarz1978). The model with the lowest BIC value was chosen, and the trace of the iterations was used to calculate the parameters. Additionally, to describe how well the model described the data, we used a generalised r² calculated according to Cox and Snell (Reference Cox and Snell2018), because the variable was binomially distributed and therefore the classical r² was not applicable.

The model was written in the Python programming language version 3.10 (Van Rossum and Drake, Reference Van Rossum and Drake2009), with the libraries NumPy version 1.21.5 (Harris et al., Reference Harris, Millman and van der Walt2020), and scipy version 1.7.3 (Virtanen et al., Reference Virtanen, Gommers and Oliphant2020).