Quantitative Analysis of Fitness Costs Associated with the Development of Resistance to the Bt Toxin Cry1Ac in Helicoverpa armigera
Guangchun Cao1☺, Hongqiang Feng2☺, Fang Guo1, Kongming Wu1*, Xianchun Li3, Gemei Liang1, Nicolas Desneux4
1. State Key Laboratory for Biology of Plant Diseases and Insect Pests, Institute of Plant Protection, Chinese Academy of Agricultural Sciences, Beijing 100193, China
2. Institute of Plant Protection, Henan Academy of Agricultural Sciences, Zhengzhou, 450002, China.
3. Department of Entomology, University of Arizona, 85721 Tucson, AZ, USA.
4. French National Institute for Agricultural Research (INRA), UMR1355-ISA, 06903 Sophia-Antipolis, France.
* Corresponding author. Fax: +86 1062815929. E-mail address:
☺ These authors contributed equally to this work.
Supplementary text
Methods
Screen of 11 H. armigera strains
Life talbes of the 11 strains on non-Bt diet
Bayesian statistics of survival rate
Analysis of the relationship between overall fitness cost and resistance level
Multiple regressions
Supplementary References
Supplementary information
Table S1. The resistance to Cry1Ac toxin and the intrinsic rate of population increase rm in eleven H. armigera strains.
Table S2. Statistical mean and 95% Credible Interval for proportional data of fitness components (data used to produce Fig. 3).
Table S3. Matrix of the posterior probability that {rc > rr}, where rc is the survival rate from neonates to the 6th instar larvae for strains listed in the first row, and rr is that for strains listed in the first column.
Table S4. Matrix of the posterior probability that {rc > rr}, where rc is the survival from 6th instar to pupa for strains listed in the first row, and rr is that for strains listed in the first column.
Table S5. Matrix of the posterior probability that {rc > rr}, where rc is the emergence rate for strains listed in the first row, and rr is that for strains listed in the first column.
Table S6. Matrix of the posterior probability that {rc > rr}, where rc is the copulation rate for strains listed in the first row, and rr is that for strains listed in the first column.
Table S7. Matrix of the posterior probability that {rc > rr}, where rc is the hatching rate for strains listed in the first row, and rr is that for strains listed in the first column.
Table S8. Matrix of the posterior probability that {rc > rr}, where rc is the proportion of female adults for strains listed in the first row, and rr is that for strains listed in the first column.
Table S9 Fitness components of XX, LF and XJ strain series.
Fig. S1. Pairplot (pairwise scatterplot with correlation coefficients) of all variables. The upper panel contains estimated pair-wise correlations, and the font size is proportional to the absolute value of the estimated correlation coefficient. The diagonal panel contains histograms and the lower panel scatterplots with a LOESS smoother added to aid visual interpretation. The thirteen influential variables were fitness cost in survival rate from the 1st to 6th instar larvae R16, from the 6th instar larvae to pupae CR6p, pupal weight CWp, emergence rate to healthy moths CRe, sex ratio of male to female CRsex, copulation rate CRc, fecundity CFec, hatching rate CRh, larval duration CDl, developmental duration of female pupa CDpf, developmental duration of male pupa CDpm, developmental duration of female adults CDaf and developmental duration of male adults Dam.
Fig. S2. Pairplot of all variables after combination of influential variables. The upper panel contains estimated pair-wise correlations, and the font size is proportional to the absolute value of the estimated correlation coefficient. The diagonal panel contains histograms and the lower panel scatterplots with a LOESS smoother added to aid visual interpretation. The ten influential variables were fitness cost in larval survival rate CRl, pupal weight CWp, emergence rate to healthy moths CRe, sex ratio of male to female CRsex, copulation rate CRc, fecundity CFec, hatching rate CRh, larval duration CDl, pupal duration CDp, and adult duration CDa.
Fig. S3. The scatter plots of fitted overall fitness cost against observed data with line of y = x as a reference.
Fig. S4. The scatter plots of fitted log resistance ratio with a multiple linear model against observed data with line of y = x as a reference.
Supplementary text
Methods
Screen of 11 H. armigera strains
The 96S was a strain susceptible to Bt originally collected from cotton fields in 1996 at Xinxiang (Henan province, China) and has been reared under laboratory conditions on an artificial diet with no contact to any Bt toxinS1. BtR was a Bt-resistant strain selected from the 96S strain for 105 generations using Cry1Ac-contaminated artificial dietsS2 (Cry1Ac toxin was provided by the Biotechnology Group, Chinese Academy of Agricultural Sciences). In this strain, the dose of Cry1Ac toxin imposed on each generation of H. armigera larvae increased gradually from 0.008 to 161 mg/L basing on the criteria that 70% of the neonates could successfully survive to adult stages and reproduce next genrationS3.
LF was a series of Bt resistant strains which originated from a H. armigera population that was collected in 1998 at the Langfang experimental station (Hebei province, China). The field-collected population had been reared on a sequence of Bt-contaminated artificial diet containing up to 3.0 mg Cry1Ac / L for 38 generations under laboratory conditionsS4. From the 39th generation onward, it was subjected to selection through rearing on artificial diets containing Cry1Ac toxin at a dose of 5 mg Cry1Ac / LS3. When 80% pupation success was observed in 3 successive generations, the population was designated as the resistant strain LF5 (reached at 45th generation). At the 46th generation, half of the LF5 larvae were reared continuously on the same dose of Cry1Ac-contaminated diet (i.e., 5 mg Cry1Ac / L) and the others was subjected to a higher Cry1Ac dose of 10 mg / L, to select a more resistant strain LF10 (reached at 56th generation). Similarly, the LF20, LF30 and LF60 strains were established using Cry1Ac-contaminated diets at concentrations of 20, 30 and 60 mg Cry1Ac/ L from the 57th, 60th and 77th generations, respectively (Fig. 1).
XJ was a series of strains originating from a H. armigera population collected in a cotton field at Xiajin (Shandong province, China) during the Cry1Ac resistance monitoring program carried out in 2004S5. The XJF strain was originated from the non-Bt fed progeny (the control treatment in resistance gene frequency monitoring) of the family that showed the highest resistance to Cry1Ac in our previous studyS5 and has been reared for 50 generations on non-Bt artificial diet in the lab since 2004. Using the same method as for LF series, the resistant strains XJ1, XJ5, and XJ10 were selected from the XJF strain using Cry1Ac-contaminated diet at concentrations of 1, 5, and 10 mg / L from the 6th, 17th, and 20th generations, respectively (Fig. 1).
During selection of resistance, the neonates were reared on the above described dose of toxin artificial diet for 7 days and then well-developed larvae were transferred to non-Bt artificial diet until moth emergence. The last generation of each strains were used in the present study (Fig. 1).
Life talbes of the 11 strains on non-Bt diet
Life tables were established for each H. armigera strains with rearing neonates on non-Bt diet. Two hundred forty H. armigera neonates of each strain were reared on a non-Bt artificial diet as described by Liang et al.S2. Each neonate was placed individually with 1.25±0.25g of artificial diet in plastic wells (depth: 1.5cm, vol. 3ml) using 24-well plates (they were covered with a plastic lid). On the 7th day, we counted and recorded larvae that had survived until the 3rd instar. The survivors were placed individually into clear glass tubes with non-Bt artificial diet until they pupated. The developmental duration and the survival from the 3rd to 6th instar and from 6th instar to pupa were recorded. Within 24-48h of pupation, pupae were weighed and sexed. The emergence of adults was recorded and 15 adult pairs from each strain were monitored daily for survival and oviposition until the death of female adults. The eggs were collected on the gauze that was covered in the plastic cup during mating, and the hatching rate of eggs was calculated as the number of newly hatched larvae divided by the number of eggs.
Bayesian statistics of survival rate
Basing Bayesian methodS6, the expectation of survival rate E[θ|y] = (a + y)/(a + b + n), where y is the number of survivors, n is the total number of tested individuals, a and b are the parameters of prior distribution of beta(a = 1, b = 1). The 95% credible interval of survival rate was obtained from the 0.025 and 0.975 quantiles of posterior distribution beta(a + y, b + n - y), that was calculated with R function qbeta(c(0.025, 0.975), a + y, b + n - y). While comparing two survival rates, a sequence of 10000000 Monte Carlo samples from the beta(a + y, b + n - y) distribution was produced with R function rbeta(a + y, b + n - y) for each survival rates, i.e., theta1 and theta2. The probalility of survival rate theta1 > survival rate theta2 was calculated with R function mean(theta1 > theta2).
Analysis of the relationship between overall fitness cost and resistance level
When analyzed for each series, the overall fitness cost, C, showed a positive linear trend with log10 transformed resistance ratio Log10Rr for the LF and XJ series; however, neither one was significant (C = -15.42 + 15.62 Log10Rr [p = 0.153] for the LF series; C = -51.02 + 41.38 Log10Rr [p = 0.164] for the XJ series). Because there were only two data points for XX series, no regression can be performed for this series. When the data from the three series were pooled together, the fitted linear model C = 0.09 + 8.71 Log10Rr was not significantly different from the linear model for the LF (F = 1.60, p = 0.375) or for the XJ (F = 1.04, p = 0.572) series, but it was statistically significant (p = 0.023) (Fig. 2). This indicated that there was no significant difference in linear tendency between series and it is better to analyze pooled data than to analyze data for each series separately.
When visually interpreted, the overall fitness cost of H. armigera logistically increased with the resistance level to Cry1Ac for the pooled data of the three series (pairwise scatter plot with a LOESS [local regression] smoother, Fig. S1ON). Therefore, we first fitted a four-parameter logistic model C = A + (B – A)/(1 + exp((xmid - Log10Rr)/scal)) to the pooled data. In the four-parameter logistic model, parameter A was not significantly different from zero (A = - 0.26, SE = 8.67, t = -0.03, p = 0.977); thus, again, a three-parameter logistic model C = Asym/(1 + exp((xmid - Log10Rr)/scal)) was fitted to the pooled data. In the three-parameter logistic model, Asym = 24.47 (SE = 5.22, t = 4.69, p = 0.00157), xmid = 1.57 (SE = 0.18, t = 8.64, p = 2.5e-05), and scal = 0.20 (SE = 0.16, t = 1.30, p = 0.231). Because the parameter scal was not significant, a two-parameter logistic model C = Asym/(1 + exp(xmid - Log10Rr)) was fitted to the pooled data again, but the AIC (Akaike’s Information Criterion) value rose.
Among the linear and nonlinear (logistic) models, the three-parameter logistic model had the lowest AIC value and was chosen to describe the relationship between overall fitness cost and resistance level.
Multiple regressions
Because there were more influential variables (i.e., 13 fitness component costs) than the number of data points i.e., 11 overall fitness costs or resistance levels, and there was no mathematic algorithm to estimate the regression coefficient parameters under such situation. Therefore, we need to exclude influential variables. The proportion of females (sex ratio) was neither significantly different among each strain (Fig. 3) nor different from 50% (with sex ratio of 1:1) for all strains; thus, this factor was confidently excluded from the multiple regression analyses. The developmental duration of female and male adults was not significantly different among strains and thus datasets were pooled. When datasets for both sexes were pooled to produce one single variable, CDa, the cost in developmental duration of adults, the correlation between overall fitness cost and development duration increased from 0.5 for each sex separately (Fig. S1LO & KO) to 0.6 for both sexes (Fig. S2IL). Thus, the two variables were combined into one variable CDa rather than being discarded. So, three approaches were carried out for multiple regressions.
Approach 1. The thirteen influential variables measured were divided into 6 groups based on developmental stages (i.e. stage of larva, pupa, adult, larva to pupa, pupa to adult, and adult to larva) to explain the overall fitness cost or resistance ratio, respectively. Stepwise regression was performed to choose the best model for each group. The most significant variables from each group were then chosen to explain the overall fitness cost or resistance ratio. Stepwise regression was performed again to choose the best model.
Approach 2. The fitness components with highly correlated fitness cost (Pearson correlation coefficient ³ 0.6) were combined into one biological parameter (e.g., the survival rate for the 1st-6th instars, and the 6th instar to pupae were combined to produce one fitness component: larval survival rate). We obtained ten influential variables, representing fitness cost for ten fitness components, to explain overall fitness cost, C, and resistance ratio, Rr (log10 transformed). The ten influential variables were fitness cost in larval survival rate, CRl, pupal weight, CWp, emergence rate to healthy moths, CRe, sex ratio of male to female, CRsex, copulation rate, CRc, fecundity, CFec, hatching rate, CRh, larval duration, CDl, pupal duration, CDp, and adult duration, CDa.