Exploiting pathogens and their impact on fitness costs to manage the evolution of resistance to Bacillus thuringiensis

Authors


Ben Raymond, Department of Zoology, South Parks Road, Oxford OX1 3PS, UK (fax +44 1865 271281; e-mail benjamin.raymond@zoo.ox.ac.uk).

Summary

  • 1Sustainable insect control requires effective management of the evolution of resistance to pesticides. Resistance management may ultimately depend on a range of diverse strategies. We explored how combining the use of a pathogen with an integrated pest management (IPM)-compatible pesticide, Bacillus thuringiensis (Bt), could affect the evolution of resistance.
  • 2We used fitness and laboratory selection experiments to explore whether the use of a nucleopolyhedrovirus could alter the rate of evolution of resistance to a Bt toxin. These results were incorporated into simulation modelling to investigate how pathogens could best be exploited in resistance management.
  • 3Simultaneous exposure to virus and Bt toxin (mixed sprays) reduced the fitness of Bt-resistant insects compared with treatments with toxin alone. Moreover, Bt resistance incurred additional fitness costs, in terms of egg fertility, in the presence of virus.
  • 4In a selection experiment with caged insects, spraying toxin-free refugia with virus slowed the evolution of resistance relative to unsprayed refugia, confirming the results of two fitness experiments that indicated that virus would increase the fitness costs of resistance. The impact of virus-mediated costs was explored further in simulation models.
  • 5Simulations showed that large virus-mediated fitness costs (a reduction in fecundity of 35%) with partially dominant inheritance could lead to effective resistance management with a Bt/virus rotation. Modest fitness costs could, however, markedly improve a rotation strategy with a pathogen that could replicate post-application.
  • 6Synthesis and applications. Mixed sprays of virus and resistance-prone pesticides such as Bt have the potential to slow the evolution of resistance within an integrated pest management context. However, the efficacy of mixed-spray strategies depends upon precise dosages and/or the presence of a spray-free refuge. In contrast, rotations of pesticides with pathogens that could replicate after application slow the evolution of resistance over a wider range of conditions and control insects cost-effectively. This efficacy is dependent on the co-occurrence of persistent pathogen and pesticide reducing the relative fitness of resistant individuals.

Introduction

Ecological research can contribute to the management of resistance to pesticides and bio-pesticides in several ways. For instance, pleiotropic fitness costs associated with resistance to pesticides can be exploited by rotating alternative products; the relative fitness of resistance genes can be altered by using mixed sprays; selection pressure can be alleviated by reducing the number of pesticide applications; the population structure of pests may be changed with the provision of selection-free refugia (Tabashnik 1989; Roush & Daly 1990; Curtis, Hill & Kasim 1993; Roush 1994).

Bacillus thuringiensis (Berliner) (Bt) is a spore-forming entomopathogenic bacterium that produces proteinaceous toxins with narrow host activity, usually against members within specific orders. It is widely sprayed as a microbial pesticide and the genes encoding crystal toxins have now been engineered into insect-resistant transgenic (or genetically modified; GM) crops, particularly cotton and maize (Shelton, Zhao & Roush 2002). Many insect species have evolved resistance to Bt strains or specific toxins under artificial selection, although only two species, one being the diamondback moth Plutella xylostella (Linnaeus), have evolved substantial resistance to Bt sprays outside the laboratory (Ferré & Van Rie 2002; Janmaat & Myers 2003). Resistance to a Bt-engineered crop has not yet evolved in the field (Bates et al. 2005; Tabashnik, Dennehy & Carrière 2005), although a number of laboratory-selected strains from several species can survive on transgenic plants (Tabashnik et al. 2003). Current resistance management in GM crops is based upon the high dose/refuge strategy (Gould 1998) and the use of stacked Bt varieties (e.g. Bollgard II cotton) that express two Bt toxins (Zhao et al. 2003). Environmentally dependent fitness costs of resistance genes, imposed, for example, during over-wintering, have contributed to preventing the increase in resistance in Pectinophora gossypiella on Bt cotton (Carrière & Tabashnik 2001; Carrière et al. 2001, 2004). An understanding of ecological factors that can alter the fitness costs of resistance may therefore help improve future resistance management in transgenic crops.

There is, however, little current research on strategies for managing the evolution of resistance to Bt sprays, despite their importance in forestry and horticulture. The low toxicity of Bt sprays makes them an important resource for growers using organic methods or integrated pest management (IPM) who need to preserve natural enemies (Entwistle et al. 1993). Additional entomopathogens (e.g. fungi and nucleopolyhedroviruses) may be valuable tools in Bt-resistance management because their use will not disrupt IPM. In addition, because these organisms impose stress upon host insects, we hypothesized that infection or exposure to a non-Bt pathogen would magnify the fitness costs associated with resistance to Bt.

We selected Acal MNPV (NPV) as our supplementary pathogen because this baculovirus has a wide host range, has been studied intensely and has been registered as a bio-pesticide in the USA. Our study insect, the diamondback moth (DBM) P. xylostella, is an excellent model system for investigating the evolution of resistance to Bt because a considerable number of field-selected resistant DBM populations have been described (Ferré & Van Rie 2002; Sayyed et al. 2004). In addition, diverse mechanisms of resistance to Bt do not produce cross-resistance to NPV in DBM (Raymond, Sayyed & Wright 2006). Unlike Bt sprays, baculoviruses commonly produce secondary cycles of infection in pest species, persist longer in the field (Ali, Young & Yearian 1987; Dwyer & Elkinton 1993; Fuxa & Richter 1994; Moscardi 1999) and can reduce pest populations in the long term (Entwistle et al. 1983; Dwyer & Elkinton 1993; Pingel & Lewis 1997). A commercially available baculovirus (Gemstar) also exists for the Heliothine cotton pests. There is no known field resistance to baculovirus bio-pesticides (Abot et al. 1996) and we argue that baculoviruses are less resistance-prone than Bt products.

We conducted a fitness experiments to determine how application of NPV and the Bt toxin Cry1Ac, separately and in combination, would affect the relative fitness of Bt-resistant and -susceptible DBM. From this we hypothesized that use of NPV in a toxin-free refuge and combined sprays of toxin and virus had the potential to slow the evolution of resistance: these hypotheses were tested in a selection experiment with caged laboratory DBM populations and in further fitness experiments. Experimental data were used to inform and parameterize simulation models that explored the efficacy of diverse resistance management strategies and the consequences of these strategies for population control. Finally, we modelled the replication and persistence of a supplementary pathogen with diverse ecological properties and differing impacts on the costs of resistance.

Methods

fitness experiment

Insect strains

The two insect strains in this study were of independent genetic origin. The Cry1Ac-resistant Karak had been collected from the field in Malaysia in 2001; rearing and selection of this population and the laboratory susceptible strain Laboratory-UK have been described previously (Sayyed & Wright 2001; Sayyed et al. 2004). Resistance in the Karak strain is recessive at high doses and is based on reduced binding of toxin in the mid-gut (Sayyed et al. 2004). The Laboratory-UK strain was originally a gift of Professor Juan Ferré (University of Valencia). Prior to experiments, resistance levels to Cry1Ac were checked using leaf-dip bioassays (Sayyed et al. 2000).

Experimental protocol

Eggs from resistant, susceptible and heterozygote Laboratory-UK × Karak mass crosses were collected on flexible plastic film (Parafilm®, American Can Company, Chicago, IL) dipped in cabbage extract. We used a three-way factorial design with three insect strains (resistant Karak, susceptible Laboratory-UK and an F1 hybrid with pooled reciprocal crosses) crossed with a two-level virus treatment (presence or absence) and a two-level toxin treatment (presence or absence). Treatments were replicated five times, each replicate being a whole plant. Toxin and virus were applied using a handheld sprayer (30 mL of 0·2 µg mL−1 Cry1Ac and 6·0 × 106 Occlusion bodies (OB) mL−1 NPV, respectively, with 50 µl L−1 surfactant Triton X-100, Sigma, UK) to greenhouse-reared, 5-week-old Chinese cabbage plants Brassica pekinensis (Lour.) Rupr. var. ‘One Kilo S.B.’. Control plants were sprayed with water and surfactant only. After plants had dried they received 35 DBM neonates and were bagged individually. Purified toxin degrades relatively rapidly under standard experimental conditions (25 °C, 65% relative humidity, 16:8 h photophase). However, sprayed toxin is still potent after 3 days (B. Raymond, unpublished data) and toxin was re-applied after 2 days.

We recorded the numbers of adults emerging, and their sex, daily. The fecundity of 1–3 pairs of survivors from each individual plant was measured by confining adults in 50-mm Petri dishes with 20% honey solution and with Parafilm (as above) for oviposition. Parafilm was changed every 3 days until females died: the total number of eggs laid and the numbers remaining unhatched after 1 week were counted. Egg development time for the first batch of eggs in each pair was measured by inspecting dishes twice daily and recording egg laying and the first emergence of neonates.

selection experiment: the effect of npv on the evolution of resistance to a bt toxin in cages

This experiment tested two hypotheses: (i) that the ability of a toxin-free refuge to slow the evolution of resistance would be improved by applying virus to a toxin-free refuge; and (ii) that simultaneous exposure to virus and toxin would slow the evolution of resistance relative to toxin alone. Discrete generations were propagated in culture cages (1 × 0·75 × 0·55 m). We applied five treatments replicated three times: toxin with unsprayed refuge; toxin with virus-sprayed refuge; toxin without refuge; toxin + virus mixture without refuge; and an unsprayed control. Virus was sprayed at a lower dose (1·5 × 106 OB mL−1) than in the first fitness experiment; toxin doses were increased gradually in order to avoid driving populations extinct. These doses were: generation 1 (G1) 0·2 µg mL−1 ; G2 0·8 µg mL−1 ; G3 and G4 3·2 µg mL−1; G5 and G6 10 µg mL−1 0. The initial dose was the same as applied in the fitness experiment. Each cage population began with 26 pupae, half Laboratory-UK susceptible insects and half Laboratory-UK × Karak resistant F1 insects. Each generation, adults laid eggs on three to four plants in each cage for 24 h in the presence of a 20% honey solution. Plants were sprayed with toxin twice, 4 and 6 days after oviposition; virus was applied with the first spray only. In refuge treatments, a single plant was bagged after oviposition and removed from the cage while toxin was applied. At the end of each generation, pupae were recovered over 4–5 days. In treatments without refugia, all survivors were collected to propagate the next generation; control cages were subsampled to constrain numbers of survivors to below 100. Pupae from cages with refugia were subsampled so that an equal number of pupae from refuge and toxin plants was used to initiate the next generation (typically 20–25 from each part of the cage). The overall number of pupae used to begin each generation varied from 40 to 100. We bio-assayed the pooled population for susceptibility to Cry1Ac at the start of the experiment (G0). Cages were bio-assayed individually every two generations by removing larvae before spraying and applying standard leaf-dip protocols (Sayyed et al. 2000). Bioassays used four to five doses with controls where possible and 25–35 third instar larvae per dose; occasionally a shortage of larvae constrained us to three doses.

virus-mediated fitness costs of resistance

An additional experiment measured the virus-mediated fitness costs using populations with similar genetic background. The selected populations (SEL) were derived from the survivors of the toxin-only treatment at the end of the selection experiment, and the unselected populations (UNSEL) were derived from the three control cages. We measured the difference in fitness (net replacement rate R0) and development time between the three SEL and UNSEL populations in the presence and absence of virus (1·5 × 106 OB mL−1) using 12 individually bagged plants per population.

data analysis

All data analysis and modelling was carried out in R (http://www.r-project.org, accessed May 2005). In the fitness experiments, mean values for each individual plant were used in analyses to avoid pseudo-replication. Fitness components and net replacement rate, R0 (Birch 1948), were analysed with three-way anovas, with insect population, presence/absence of toxins and presence/absence of virus as factors. Proportional data were analysed with binomial errors; F-tests were applied to correct for over-dispersion. Bioassay data from each generation of the selection experiment were analysed with mixed-model ancovas. Cage was treated as a random effect, log-transformed bioassay dose and treatment as fixed effects. Doses were nested within cages and arc-sine transformed mortality was used as the response variable. Models were initially fitted with restricted maximum likelihood estimation (REML), although comparison of mixed-effect models after simplification was carried out using models fitted with maximum likelihood (Crawley 2002). The results of model simplification agreed qualitatively with individual t-tests on the slope and intercept of mortality plotted against dose. These t-tests have been presented when informative. Final model fitting and model assumptions were checked with standard graphical analyses in all cases.

simulation modelling

The experimental results suggested a number of ways in which virus, or a similar pathogen, could be exploited in Bt-resistance management. First, could virus-mediated fitness costs be exploited by applying virus to a toxin-free refuge in a GM crop (model I; Table 1)? Secondly, a range of different strategies were explored in an IPM scenario with microbial Bt sprays, a density-dependent spray threshold and a fixed level of natural enemy attack (model II; Table 1). The importance of virus-mediated resistance costs, as quantified above, were explored in a standard rotation strategy (applying Bt and virus sprays in alternate generations). Mixed-spray strategies were also investigated; refugia size was varied in these models because it is vital for their efficacy (Roush & Daly 1990). We also modelled a third scenario, the ‘rotation + epizootic’ strategy, in which the pathogen was allowed limited persistence post-spraying. In this scenario virus was rotated alternately with Bt sprays. When persistent virus and Bt were present simultaneously, virus reduced the relative fitness of Bt-resistant insects. The model recorded pest population size and the frequency of resistance after 200 generations, the number of generations for resistance frequency to exceed 0·5 (G0·5) and the number of Bt sprays applied every 200 generations. Model parameters were varied as described in Table 1, and most models were run with two realistic initial Bt-resistance gene frequencies (0·01 and 0·001) (Gould et al. 1997; Andow et al. 2000; Tabashnik et al. 2000).

Table 1.  Survival of larvae in simulation models. The parameter c determines the impact of virus on insect fitness and can reflect changes in virus dose or pathogenicity
I : GM crop
GenotypeGM cropRefuge–high-virus doseRefuge–low-virus doseRefuge–unsprayed
SS, RS0·010·040·080·25
RR0·10·040·080·25
II: microbial spray with density-dependent application
GenotypeBt sprayVirusBt and virus mixtureNatural enemies
SS, RS0·020·25–0·46c0·020·25
RR0·160·25–0·46c0·16–0·31c0·25

Resistance was determined by a single, completely recessive, diallelic locus (R for resistant, S for susceptible). Fitness costs occurred only in the presence of virus. Selection was imposed by the differential survival of larvae and differential fecundity of adults. We used a stochastic population genetics model; gene frequencies followed a Poisson distribution; expected gene frequencies were determined from panmictic random mating. There was no migration in and out of the modelled population and generations were non-overlapping. Fecundity and the effects of Bt and virus on fitness were based on the above experiments. We used data from Bt crops (Tabashnik et al. 2003) or commonly used baculoviruses (Hunter-Fujita et al. 1998) to determine survival parameters (Table 1). Twenty simulations of 200 generations were run with each parameter set. The initial pest population size was 10 000 adults; we imposed a carrying capacity of 200 000 adults. The density-dependent spray threshold was arbitrarily set at 350 000 larvae, we did not specify the number of crop plants. The sensitivity of the rotation + epizootic model to spray thresholds was explored in additional simulations (see Appendix 1 in the supplementary material).

Fecundity was determined by parental genotype × environment interactions (Bt resistance and virus presence/absence). The fitness consequences of exposure to virus were additive; Bt had no effect on fecundity. Relative fecundity can be expressed by adapting conventional population genetic equations (Lenormand & Raymond 1998):

FSS  = 1 − cv(x)
FRS = 1 − cv(x) − hccrv(x)
FRR = 1 − cv(x) − crv(x)

where c is the reduction in fecundity caused by exposure to virus. The function v(x) is 1 if virus is present and 0 if no virus is present. The coefficient hc is the dominance level of the fitness cost cr. When relative fecundity was equal to 1·0, females laid 120 eggs.

The parameter c (the reduction in fecundity on exposure to virus) was used as a measure of virus dose or pathogenicity. Thus survival parameter and fitness costs of exposure to virus were determined as linear functions of c (Table 1). The persistence of virus on crop plants was also expressed in relation to c:

c = cmax − td

where cmax is the value of c (or dose) at the time of application, t is the number of generations since virus was applied and d is the decline in c per generation. Virus always declined post-spraying (d was positive) and did not persist for more than three larval generations.

Experimental results

fitness experiment

Prior to the fitness experiment the Cry1Ac LC50 (lethal concentration killing 50% of larvae) of the Karak resistant population was 61·1 µg mL−1 [95% confidence limit (CL) 30·6–122 µg mL−1]. The LC50 of the Laboratory-UK strain was 0·02 µg mL−1 (95% CL 0·004–0·026 µg mL−1). Resistance and its fitness costs were recessive as the F1 hybrids and the susceptible strain could be pooled into a single factor without significant loss of variation in R0 (F4,45 = 0·93, P > 0·45) and development time (F4,47 = 1·42, P = 0·24). Toxin-resistant insects had higher fitness (R0) when exposed to toxin alone (Fig. 1). Application of virus changed the direction of the interaction between resistance and toxin, i.e. resistant insects were fittest in the presence of toxin alone and susceptible insects were fittest when both toxin and virus were present (Fig. 1). This was supported by a significant three-way interaction between phenotypic resistance, virus and toxin (F1,49 = 4·28, P = 0·044). There were also significant interactions between virus and resistance (F1,49 = 4·4, P = 0·041) and virus and toxin (F1,49 = 6·94, P = 0·011) on R0. Full anova tables for analysis of fitness parameters are presented in Appendix 1 (see the supplementary material).

Figure 1.

Variation in fitness (R0) and two fitness components of Cry1Ac-resistant DBM (Karak or ‘res’), a laboratory susceptible strain (Laboratory-UK or ‘sus’) and their F1 hybrids (‘het’) exposed to NPV and the Bt toxin Cry1Ac. Data are means ± SE.

Exposure to virus infection reduced fecundity (Fig. 2; F1,43 = 19·22, P < 0·001) and affected each population differently (Fig. 1; interaction, F2,43 = 3·68, P = 0·033). F1 crosses exposed to virus had a mean fecundity intermediate between the resistant and susceptible populations (Fig. 1). The resistant Karak population had reduced egg fertility on exposure to virus relative to the other populations (Fig. 1; resistance × virus interaction, F1,46 = 5·28, P = 0·026). Virus and/or toxin also had significant consequences for additional performance parameters (development time, sex ratio and egg development time) (see Appendix 1 in the supplementary material). Mean sex ratios for each treatment were incorporated in the calculation of R0.

Figure 2.

The change in susceptibility to the Bt toxin Cry1Ac relative to controls in the selection experiment. Each treatment was replicated with three caged populations of DBM that were bioassayed independently. To enable comparison between generations, we calculated the standard errors of mortality at a standard dose (the LC50 of controls in each generation). This was carried out by shifting the x-axis such that the REML-fitted line in the mixed-model ancova intersected the y-axis at the same dose within each generation.

selection experiment: the effect of npv on the evolution of resistance to a bt toxin in cages

The initial selection experiment led us to hypothesize that the co-application of virus and toxin or the use of virus in a toxin-free refuge could slow the evolution of resistance. These hypotheses were tested in a selection experiment with caged insect populations. After two generations of selection, bioassay mortality did not vary significantly among treatments (bioassay dose × treatment interaction, d.f. = 4,10, likelihood ratio = 3·41, P = 0·49; treatment main effect, d.f. = 4,6, likelihood ratio = 6·55, P = 0·016; Fig. 2) (see Appendix 1 in the supplementary material). After four generations of selection most treatments did not differ from controls, although post-hoc pairwise contrasts indicated lower mortality in the virus + toxin treatment (difference in intercept =−0·14, SE = 0·06, t =−2·23, P = 0·03; Fig. 3); the treatment main effects were near significance (d.f. = 4,6, likelihood ratio = 9·11, P = 0·058).

Figure 3.

The effect of NPV on the two fitness components and the overall fitness (R0) of two DBM populations, resistant (SEL) and susceptible (UNSEL) to Cry1Ac. These populations were derived from the survivors of the selection experiment and were initiated with populations with similar genetic background.

After six generations of selection there were differences in the susceptibility to toxin between several treatments and the controls (Fig. 2; treatment × dose interaction, d.f. = 4,10, likelihood ratio = 10·44, P = 0·034; treatment main effect, d.f. = 4,6, likelihood ratio = 6·89, P = 0·14). However, insects from the virus refuge treatment had similar mortality to the controls (difference in intercept =−0·012, SE = 0·044, t =−0·11, P = 0·91; difference in slope =−0·012, SE = 0·044, t =−0·28, P = 0·78). Model simplification confirmed that treatment levels could be reduced to two groups (control and virus refuge treatments being one group; all other treatments the second group) without significant loss of deviance (d.f. = 6,8, likelihood ratio = 0·65, P = 0·95). Thus refugia in which insects were exposed to virus controlled resistance more effectively than refugia with randomly culled populations. However, results for sprays of toxin alone were indistinguishable from mixed sprays of virus and toxin.

virus-mediated fitness costs of resistance

Using the SEL and UNSEL populations, derived from cultures with similar genetic backgrounds, we found that although virus decreased the fitness (R0) of resistant larvae this was not a significant trend (population × virus interaction, F1,50 = 1·4, P = 0·24; Fig. 3). Similarly, there was no interaction between population and virus on the fecundity of survivors (F1,48 = 1·52, P = 0·46). In general, fecundity was reduced and data more variable in the second fitness experiment (Figs 1 and 3). However, the SEL population had reduced egg fertility in the presence of virus (Fig. 3; population × virus interaction, F1,46 = 4·52, P = 0·039; virus, F1,47 = 6·3, P = 0·016; population, F1,48 = 7·15, P = 0·010).

Simulation results

model 1: virus use and gm crops

Virus-mediated fitness costs in GM refugia had little effect on the evolution of resistance when recessive. The time for resistance to evolve (G0·5) was 112 generations with an unsprayed refuge, 100 generations with a high-dose virus spray (c = 0·5, cr = 0·35) and 105 generations with a low-dose (c = 0·25, cr = 0·17). Only partially dominant fitness costs could substantially slow the evolution of resistance, although the magnitude of these costs could be small (hc = 0·03 at the high dose; hc = 0·3 at a low dose; Fig. 4).

Figure 4.

The consequences of increasing dominance of fitness costs imposed by exposure to virus on the evolution of resistance to Bt toxins of a simulated pest population on a GM crop. Initial frequency of resistance R = 0·01; proportional size of virus-sprayed refuge = 0·25. Each data point is the mean frequency of R after 200 generations in 20 model runs.

model 2: combined use of bt and virus in ipm

Without fitness costs, rotational strategies increased the time taken for resistance to evolve by no more than 40 generations relative to Bt-only strategies (Table 3). Mixed sprays with equal fitness of resistant and susceptible insects (c = 0·45) prevented the evolution of Bt resistance. Similar efficacy in resistance management and pest control was provided by the rotation + epizootic model, with fewer applications of Bt and virus (Table 3). At an initial resistance frequency of 0·001, there was considerable stochastic variation in the evolution of resistance under rotational strategies (Table 3). Results were more consistent for the mixed spray and the rotation + epizootic strategies despite the absence of fitness costs.

Table 3.  The evaluation of different resistant management strategies for mitigating the evolution of resistance to Bt microbial sprays (model II). Values are the means of 20 model runs of 200 generations each. G0·5 is the number of generations taken for resistance gene frequency to exceed 0·5. c = 0·5, except in mixed sprays; cr = 0
Simulation typeFrequency of resistance after 200 generations, mean (range)G0·5 (SE)Mean adult population sizeMean no. Bt sprays (max. 200)
  1. NA, resistance gene frequency never exceeded 0·5.

Initial frequency of R = 0·01
Bt only1·0 (1·0, 1·0) 19·8 (0·3)1·8 × 106200
Rotation: Bt/NPV1·0 (1·0, 1·0) 38·8 (0·8)4·9 × 105100
Mixed Bt/NPV0·010 (0·001, 0·02)NA4·1 × 104174
c = 0·45
Rotation + epizootic0·012 (0·002, 0·03)NA3·9 × 104 81
d = 0·05
Initial frequency of R = 0·001
Bt only0·71 (0, 1·0)153 (11)6·2 × 105200
Rotation: Bt/NPV0·35 (0, 1·0)189 (5)7·1 × 104100
Mixed Bt/NPV0·0014 (0, 0·007)NA4·2 × 104174
c = 0·45
Rotation + epizootic0·0005 (0, 0·008)NA4·0 × 104 82
d = 0·05

the mixed-spray strategy: variation in virus dose and refuge size

The effectiveness of mixed sprays was highly sensitive to the impact of virus on insect fitness (c) in the absence of a refuge. Simulations in which c was below 0·4, which implied unequal fitness of RR and SS individuals in the presence of Bt and virus, led to resistance management failure (Fig. 5). Increasing refuge sizes decreased the sensitivity of this strategy to c (Fig. 5). However, when the refuge size exceeded 8% there were marginal benefits of using a mixed spray in addition to a refuge (Fig. 5).

Figure 5.

Combining an unsprayed refuge with a mixed-spray strategy for managing the evolution of resistance to a Bt microbial spray. The impact of refuge size and dose of an additional pathogen (virus) on the evolution of resistance. Virus dose is expressed in terms of its impact on host fitness; initial frequency of resistance gene R = 0·001.

rotation and rotation + epizootic strategies: the effect of fitness costs and virus persistence

Fitness costs led to small improvements in the efficacy of the rotation strategy unless these costs were large and partially dominant (Table 4). In contrast, small, recessive fitness could extend the efficacy of the rotation + epizootic strategy, even at a high initial gene frequency (Table 4). At high initial gene frequencies of 0·01, the epizootic strategy was very sensitive to d (the rate of decline of the virus) (see Appendix 1 in the supplementary material), although a high dose virus (high c) with good persistence could still curtail the evolution of resistance (Table 3). When the initial frequency was lower (0·001), there were very small increases in resistance over a wider range of values of c but the evolution of resistance was still sensitive to d, virus decline rate (Fig. 6).

Table 4.  The impact of virus-mediated fitness costs on the evolution of resistance to Bt under two resistant management strategies in model II (microbial sprays). Fecundity reduction on exposure to virus (c) = 0·5. cr, fitness cost; hc, dominance of fitness cost; d, virus decay rate post-spraying. Initial gene frequency of R = 0·01
Management strategydResistance frequency after 200 generations, mean (range)Mean adult population sizeMean no. Bt sprays (max. 200)
cr = 0·1, hc= 0 (recessive)
Rotation1 (1, 1)4·1 × 105100
Rotation + epizootic0·050·008 (0, 0·02)3·9 × 104 83
Rotation + epizootic0·180·022 (0, 0·17)3·1 × 104 86
cr = 0·1, hc = 0·25 (partially dominant)
Rotation1 (1, 1)3·9 × 105100
Rotation + epizootic0·050·0003 (0, 0·002)3·9 × 104 83
Rotation + epizootic0·180·006 (0, 0·026)3·2 × 104 86
cr = 0·35, hc= 0 (recessive)
Rotation1 (1, 1)2·2 × 105100
Rotation + epizootic0·050·006 (0, 0·14)3·8 × 104 84·5
Rotation + epizootic0·180·011 (0, 0·04)3·3 × 104 86
cr = 0·35, hc = 0·25 (partially dominant)
Rotation0·05 (0, 1)4·7 × 104100
Rotation + epizootic0·050·0 (0)3·8 × 104 84·4
Rotation + epizootic0·180·0 (0)3·3 × 104 86
Figure 6.

The evolution of resistance to Bt in a simulated insect population controlled with the rotation + epizootic strategy and a second pathogen of varying persistence and pathogenicity. Virus dose is expressed in terms of its impact on host fitness. Initial frequency of resistance gene R = 0·001.

Discussion

The presence of NPV can increase the fitness costs of resistance to the Bt toxin Cry1Ac. In the initial fitness experiment, laboratory susceptible insects had higher performance than the resistant population when a high dose of virus was present, irrespective of the presence of toxin (Fig. 1). Karak resistant insects have 80–90% of the fitness of susceptibles in controls but only 25% that of susceptibles in the presence of virus (Fig. 1); resistant insects also had reduced egg fertility in the presence of virus. Exposure to virus led to heterozygote F1 crosses having fecundity intermediate between the resistant and susceptible insects (Fig. 2), a pattern consistent with a partially dominant fitness cost, although we cannot exclude the role of differences in genetic background. In the repeated fitness experiment, when we controlled for genetic background, resistant insects had significantly lower egg fertility in the presence of virus and a similar trend for reduced R0 (Fig. 3). The analysis of overall fitness (R0) is, however, much less powerful than those of single life-history parameters, as variation in R0 is compounded by variation in sex ratio, survival, fecundity and egg fertility. Other things being equal, we would expect differences in egg fertility to translate to overall reduced fitness. The reduced magnitude of fitness differences in the second experiment may be partly attributable to the lower NPV dose used, the presence of resistance genes in the unselected population or the homogenization in genetic background. The unselected population had an initial resistance gene frequency of 0·25 and therefore up to 38% of the individuals in the unselected population may have been resistant heterozygotes.

The selection experiment showed that the use of virus could slow the evolution of resistance as refugia were more effective at preventing the evolution of resistance if sprayed with NPV (Fig. 3). Relative population sizes on the sprayed and unsprayed plants were controlled and so differences between these treatments should be the result of stronger selection pressure against resistant alleles on exposure to NPV. This experiment therefore represents an independent, genetically controlled test of the hypothesis that NPV could increase the fitness costs of resistance to Cry1Ac, and was designed to reflect the efficacy of a virus-sprayed refuge relative to a refuge sprayed with a conventional pesticide. Simulations of the impacts of fitness costs on the efficacy of refuge strategies (this study; Carrière & Tabashnik 2001) indicate that increasing the dominance of fitness costs markedly increases their impact on the evolution of resistance. The frequency of resistance genes was probably too high initially for the unsprayed refuge to slow the evolution of resistance (Roush 1994). Fitness costs of resistance are also vital for rotational resistance management strategies (Tabashnik 1989; Curtis, Hill & Kasim 1993). Simulations in this study showed that recessive fitness costs did not markedly improve resistance management with a simple rotation strategy. However, the rotation + epizootic strategy was sensitive to small fitness costs (e.g. 10%). In the absence of fitness costs, rotations, at best, doubled the time taken for resistance to evolve, as might be expected with half the spraying frequency (Tabashnik 1989).

Unfortunately for resistance management, diverse Bt-resistance genes may have diverse, environmentally dependent fitness costs (Carrière et al. 2004; Raymond, Sayyed & Wright 2005). Rotational management strategies that depend upon these costs may ultimately select for resistance alleles with lower costs (Clarke & McKenzie 1987; Lenormand et al. 1998). Mixed sprays do not depend upon fitness costs for their efficacy and Bt and NPV fulfil many of the required assumptions of this strategy, i.e. independent action, no cross-resistance and recessive resistance (Tabashnik 1989; Curtis, Hill & Kasim 1993; Hoy 1998). However, mixed-spray strategies are very vulnerable to small variations in dosage in the absence of an unselected refuge population. Subsequent experiments showed that the virus dose used in the selection experiment reduced the fitness advantage of resistant insects but did not eliminate it (B. Raymond, unpublished data). Modelling a mixed-spray strategy with varying doses of a supplementary pathogen showed that when the survival of susceptible insects dropped below half that of resistant insects (i.e. when c < 0·4; Fig. 5), resistance evolved very rapidly. Maximizing the kill rate of both products in mixed sprays increases their efficacy dramatically (Curtis, Hill & Kasim 1993; Roush 1993). However, this ‘redundant killing’ approach may be unworkable for expensive bio-pesticides that cannot be applied systemically.

In contrast, the rotation + epizootic strategy, which rotated applications of Bt and virus and allowed mixed exposure to virus and Bt as the virus persisted, was effective over a wide range of conditions. The inclusion of fitness costs improved this strategy (Table 4) but, at an initial resistance frequency of 0·001, it effectively slowed resistance in the absence of fitness costs. Persistence of the supplementary pathogen reduced the frequency of Bt sprays by approximately 20% relative to standard rotation models (Table 3), a potentially substantial saving in cost of applied bio-pesticides. The most important variable for this strategy was the rate of decline of the virus in the environment. Persistent virus moderated the fitness advantage of resistant alleles in the presence of Bt. For example, a simple mixed spray approach using a pathogen with a moderate impact on pest fitness (c = 0·3) would lead to a rapid evolution of resistance. Use of the same pathogen in a rotation + epizootic strategy would not lead to the evolution of resistance provided the pathogen persisted well over three generations.

Are the assumptions of our models realistic? The qualitative effects of the rotation + epizootic strategy were relatively robust with respect to spray thresholds (see Appendix 1 in the supplementary material). Including immigration or varying population structure was beyond the scope of this study, although these factors can be vital processes in the evolution of resistance in several species (Roush & Daly 1990). Metapopulation structure can also influence the dynamics of recessive resistance mechanisms by providing local sources of homozygous resistant pests (Caprio & Hoy 1994). This is most likely to be important when the evolution of resistance is subject to strong stochastic variation and causes large differences in gene frequencies between neighbouring populations. We also assumed that resistance is less likely to evolve to our model virus than to Bt. Several insects (notably Heliothis spp.) have failed to evolve resistance to viruses under laboratory selection (Raheja & Brooks 1971; Ignoffo & Allen 1972; Whitlock 1977; Kaomini & Roush 1988; Fuxa 1993). In addition, relatively low levels of resistance to NPV (less than 10-fold) can lead to very high fitness costs (Boots & Begon 1993; Fuxa & Richter 1989). However, natural between-population variation in susceptibility to baculoviruses can respond to artificial selection (Fuxa 1993; Abot et al. 1996). Resistance to NPV pesticides is unknown probably because these products are applied to a relatively small proportion of their target populations (Abot et al. 1996). Natural transmission, for example secondary cycling, typically leaves a fraction of the population unexposed to virus (Dwyer, Elkinton & Buonaccorsi 1997; Hails et al. 2002) and natural disease outbreaks appear to exert a mild selection pressure on the evolution of resistance (Boots & Begon 1993; Fuxa 1993). Application strategies that depend on secondary cycling will therefore be less likely to lead to resistance than an inundative mixed spray approach with high doses.

Pathogens that cycle naturally in pest populations are plausible complementary bio-pesticides. Some combinations, however, may not be suitable. A fungal pathogen was shown to be capable of accelerating the evolution of resistance when applied onto a low Bt-toxin expression crop, because susceptible insects had increased restlessness and acquired more infections in the presence of toxin (Johnson, Gould & Kennedy 1997a,b). However, this type of application is not necessary for the current high-expression Bt crops. Damage-tolerant crops with higher pest populations are also likely to be more appropriate for the density-dependent transmission of pathogens. NPV and their relatives, the granulosis viruses, could potentially combine very well with Bt sprays. Baculovirus and fungi of several Lepidopteran species can help limit pest populations via natural outbreaks or secondary cycling after application (Entwistle et al. 1983; Ali, Young & Yearian 1987; Dwyer & Elkinton 1993; Pingel & Lewis 1997; Hajek et al. 2004). NPV can cycle very efficiently in greenhouse environments that have reduced UV and enhanced viral survival (Huber 1998). Importantly, replicating pathogens can be combined to manage the evolution of resistance to products other than Bt. When pathogens are combined with products with low toxicity or narrow specificity there will be additional benefits in terms of preserving natural enemy populations. In general, provided a pathogen can cycle effectively in a pest population, additional conditions for the efficacy of a epizootic + rotation strategy are relatively few: (i) co-application must reduce the relative fitness of resistant individuals; (ii) pesticide resistance should be recessive; and (iii) resistance gene frequencies must not exceed 0·001.

Acknowledgements

BBSRC grant D15960 funded this study. R. S. Hails was supported by the Department of Food and Rural affairs (HH 310TX foundation). Thanks to Dzolkifli Omar for the collection of the Karak strain in Malaysia and to the anonymous referees for helping to improve this manuscript.

Ancillary