Hybridizing transgenic Bt cotton with non-Bt cotton counters resistance in pink bollworm

Significance

Crops genetically engineered to produce insecticidal proteins from the bacterium Bacillus thuringiensis (Bt) kill some major pests and reduce use of insecticide sprays. However, evolution of pest resistance to Bt proteins decreases these benefits. We report a strategy for combating resistance by crossing transgenic Bt plants with conventional non-Bt plants and then sowing the second-generation seeds. This strategy yields a random mixture within fields of three-quarters of plants that produce Bt protein and one-quarter that does not. An 11-y field study in China shows this strategy countered resistance to Bt cotton of pink bollworm, one of the world’s most devastating pests. This outcome illustrates that non-Bt plants in a seed mixture can boost survival of susceptible insects and delay resistance.

Abstract

Extensive cultivation of crops genetically engineered to produce insecticidal proteins from the bacterium Bacillus thuringiensis (Bt) has suppressed some major pests, reduced insecticide sprays, enhanced pest control by natural enemies, and increased grower profits. However, these benefits are being eroded by evolution of resistance in pests. We report a strategy for combating resistance by crossing transgenic Bt plants with conventional non-Bt plants and then crossing the resulting first-generation (F₁) hybrid progeny and sowing the second-generation (F₂) seeds. This strategy yields a random mixture within fields of three-quarters of plants that produce Bt toxin and one-quarter that does not. We hypothesized that the non-Bt plants in this mixture promote survival of susceptible insects, thereby delaying evolution of resistance. To test this hypothesis, we compared predictions from computer modeling with data monitoring pink bollworm (Pectinophora gossypiella) resistance to Bt toxin Cry1Ac produced by transgenic cotton in an 11-y study at 17 field sites in six provinces of China. The frequency of resistant individuals in the field increased before this strategy was widely deployed and then declined after its widespread adoption boosted the percentage of non-Bt cotton plants in the region. The correspondence between the predicted and observed outcomes implies that this strategy countered evolution of resistance. Despite the increased percentage of non-Bt cotton, suppression of pink bollworm was sustained. Unlike other resistance management tactics that require regulatory intervention, growers adopted this strategy voluntarily, apparently because of advantages that may include better performance as well as lower costs for seeds and insecticides.

Genetically engineered crops that produce insecticidal proteins from the bacterium Bacillus thuringiensis (Bt) have been planted globally on a cumulative total of over 732 million ha since 1996 (1). The benefits of these Bt crops include pest suppression, reduced insecticide use, enhanced biological control, and increased farmer profits (2–7). However, increasingly rapid evolution of resistance to Bt crops by pests has eroded these benefits (8–11). The main strategy for delaying pest resistance to Bt crops aims to increase the survival of susceptible insects with “refuges” of host plants that do not produce Bt toxins (9). Although refuges can delay insect adaptation to Bt crops (2, 3, 9, 12), the optimal spatial scale for planting refuges remains unresolved (13–15). Also, because refuges are often perceived to cause short-term economic sacrifices for growers, they are usually imposed by regulations.

Before 2010, regulations in the United States mandated refuges of non-Bt plants in blocks consisting of separate fields, rows, or strips (14). In 2010, the regulations were modified to include mixtures of Bt and non-Bt seeds generating a random array of Bt and non-Bt plants side by side within fields (14). By 2015, all farmers surveyed in the US Corn Belt planted such seed mixtures for some or all of their corn (16). Seed mixtures have several advantages relative to block refuges, particularly convenience for farmers and elimination of noncompliance with block refuge requirements (15, 17). However, seed mixtures increase opportunities for larval movement between Bt and non-Bt plants or plant tissues, which could accelerate resistance evolution by raising the dominance of resistance or by reducing the survival of susceptible insects (13–15, 17). Despite potentially profound implications of the seed mixture strategy for the sustainability of Bt crops, large-scale field tests of its efficacy have been lacking.

Here, we test a version of the seed mixture strategy by comparing predictions from computer modeling with field monitoring data for the pink bollworm (Pectinophora gossypiella) based on evaluation of resistance in 96 sets of insects collected during 2005–2015 from 17 sites in six provinces in the Yangtze River Valley of China (total n = 66,648 larvae tested in bioassays). The millions of small-scale farmers in this region first planted transgenic cotton producing Bt toxin Cry1Ac in 2000, with the pink bollworm as one of the primary targets (18). The larvae of this devastating worldwide pest feed primarily on cotton seeds within bolls, and susceptible larvae are killed by the Cry1Ac in Bt cotton (6, 19).

Previous work in the Yangtze River Valley documented a small but statistically significant increase in resistance of the pink bollworm to Cry1Ac from 2005–2007 to 2008–2010 (18). The percentage of populations with one or more larvae surviving at the diagnostic concentration (9 μg of Cry1Ac per milliliter of diet) increased from 0% in 2005–2007 to 56% in 2008–2010 (total n = 51 populations; Fisher’s exact test: P < 0.0001; Fig. 1A and SI Appendix, Fig. S1 and Table S1). In addition, the median percentage survival at the diagnostic concentration increased from 0% in 2005–2007 to 1.6% in 2008–2010 (Mann–Whitney U test: U = 504, P < 0.001; Fig. 1B). The resistance ratio, which was calculated as the concentration of Cry1Ac killing 50% of larvae (LC₅₀) for a field population divided by the LC₅₀ for the susceptible laboratory strain QJ-S tested in the same year, rose from a mean of 2.5 in 2005–2007 to 4.1 in 2008–2010 (t test: t = 2.2, df = 49, P = 0.03; Fig. 1C and SI Appendix, Table S2).

Fig. 1.

Pink bollworm resistance to Bt toxin Cry1Ac in the Yangtze River Valley, 2005–2015. (A) Percentage of populations with one or more survivors at the diagnostic concentration of Cry1Ac (SI Appendix, Table S1). (B) Percentage of survivors in each population (SI Appendix, Table S1). (C) Resistance ratio, the LC₅₀ divided by the LC₅₀ for the susceptible QJ-S laboratory strain tested in the same year as an internal control (SI Appendix, Table S2). Values in B and C are means with SE. Data for 2005–2010 are from ref. 18.

Without major changes, rapid increases in resistance were anticipated after 2010 because refuges of non-Bt cotton varieties had decreased to only 6% of all cotton hectares planted (18) and evolution of resistance is an exponential process expected to accelerate after more than 1% of the population is resistant (20). However, the results reported here show that pink bollworm resistance to Cry1Ac did not increase from 2008–2010 to 2011–2015 (Fig. 1). The percentage of populations with one or more larvae surviving at the diagnostic concentration declined from 56% in 2008–2010 (n = 27 populations) to 0% in 2011–2015 (n = 45 populations) (Fisher’s exact test: P < 0.0001; Fig. 1A). Also, the median percentage survival at the diagnostic concentration declined from 1.6% in 2008–2010 to 0% in 2011–2015 (Mann–Whitney U test: U = 945, P < 0.0001; Fig. 1B). The mean resistance ratio, which is a less sensitive indicator than survival at the diagnostic concentration, decreased slightly but not significantly, from 4.1 in 2008–2010 to 3.7 in 2011–2015 (t test: t = 0.77, df = 70, P = 0.55; Fig. 1C). In control tests, survival at the diagnostic concentration was 0% for the susceptible laboratory strain QJ-S every year from 2005 to 2015 (n = 72 larvae per year, total of 792 larvae), indicating consistency of the bioassays over time.

To understand the unexpected decrease in survival at the diagnostic concentration of Cry1Ac, we investigated the possibility that the percentage of non-Bt cotton plants increased in 2010 and subsequent years because growers purchased and planted seed from second-generation (F₂) cotton hybrids. Because of heterosis, F₁ and F₂ cotton hybrids typically have a higher yield than their parent varieties (21, 22). Whereas production of F₁ hybrid seeds requires costly hand pollination, self-pollination of F₁ hybrids produces F₂ hybrid seeds (21, 22). Crossing Bt cotton with non-Bt cotton creates F₁ hybrids that make Bt toxin and are hemizygous for this trait because they have one copy of the Bt transgene and no corresponding allele (23, 24). Self-pollination by such F₁ hybrids creates F₂ hybrid seeds expected to consist of 25% Bt homozygotes and 50% hemizygotes that produce Bt toxin and 25% non-Bt homozygotes that do not (23).

From 2010 to 2015, we determined the percentage of cotton seeds containing Cry1Ac using immunoassays with 100 seeds tested individually per assay in 140 assays of 16 varieties and 68 hybrids recommended for planting in the Yangtze River Valley (total of 14,000 seeds tested; Fig. 2 and SI Appendix, Fig. S2 and Tables S3–S6). We detected Cry1Ac in 99% (SE = 0.6%) of seeds from six Bt cotton varieties, which is close to the expected 100%. In 10 non-Bt cotton varieties, Cry1Ac occurred in 23% (SE = 2) of seeds, which indicates contamination of the non-Bt cotton seed supply. For the 68 hybrids, the percentage of seeds containing Cry1Ac clusters into two groups: 23 hybrids with a range of 89–100% (mean = 96%, SE = 0.5%) and 45 hybrids with a range of 39–84% (mean = 67%, SE = 1%; SI Appendix, Fig. S2). We refer hereafter to these groups as F₁ and F₂ hybrids, respectively, because they are likely to consist predominantly of these hybrids. However, the observed percentage of seeds containing Cry1Ac is lower than the expected 100% for F₁ hybrids and 75% for F₂ hybrids, which implies that some of these F₁ and F₂ hybrids contain seeds from non-Bt cotton varieties, other hybrids, or both. Based on our classification of the hybrids, the price was 35% higher for seeds from F₁ than F₂ hybrids (SI Appendix, Table S7), which corresponds with the higher production cost of the F₁ hybrids. Although our analysis focuses primarily on 2010–2015, we also surveyed 12 major seed companies to estimate the percentage of F₂ hybrids planted from 2004 to 2009 (SI Appendix, Table S8).

Fig. 2.

Cotton planted in the Yangtze River Valley of China from 2004 to 2015. (A) Percentage of cotton hectares planted to F₁ hybrid cotton, F₂ hybrid cotton, non-Bt cotton varieties, and Bt cotton varieties (SI Appendix, Tables S8–S11). (B) Percentage of cotton hectares consisting of plants not producing Cry1Ac: total and contributions by F₂ hybrids and non-Bt cotton varieties. The contribution of F₁ hybrids plus Bt cotton to the percentage of cotton hectares consisting of plants not producing Cry1Ac was ≤2.5% each year (mean = 1.4%).

To calculate the total percentage of non-Bt cotton plants for each year, we multiplied the proportion of non-Bt cotton seeds for each type of cotton (SI Appendix, Tables S3–S6 and S8) by the percentage of cotton hectares planted with the corresponding type of cotton during that year (Fig. 2A and SI Appendix, Tables S8–S11), and then summed the resulting percentages. The overall percentage of non-Bt cotton plants declined from a maximum of 37% in 2004 to 12% in 2008 and 2009, then nearly doubled to 23% in 2010, and was 25–27% from 2011 to 2015 (Fig. 2B). The initial decrease in non-Bt cotton percentage reflects reduced planting of non-Bt cotton varieties, whereas replacement of Bt cotton varieties and F₁ hybrids with F₂ hybrids caused the subsequent increase (Fig. 2). In each year from 2010 to 2015, F₂ hybrids accounted for 78–84% of the total non-Bt cotton plants (mean = 81%; Fig. 2B), which provided an exceptional opportunity to test hypotheses about how this seed mixture affects the evolution of resistance.

We used computer modeling to evaluate two hypotheses: (a) all non-Bt cotton plants act as refuges, including the non-Bt cotton plants in fields of F₂ hybrid cotton, or (b) only non-Bt cotton plants in fields of non-Bt cotton varieties act as refuges. In both scenarios, we simulated the period from the end of the season in 2008 to 2015 using a previously described one-locus, two-allele population genetic model, with fitness values for each genotype on Bt and non-Bt cotton plants based on empirical data (25, 26) (SI Appendix, Table S12). Under hypothesis (a), which yielded an annual mean refuge of 23% non-Bt cotton plants, resistance evolution was delayed, which corresponds with the observed outcome in the field (Fig. 3). In striking contrast, hypothesis (b) yielded an annual mean refuge of only 4.1%, and the predicted percentage of resistant individuals surged to 100% by 2013 (Fig. 3). These comparisons between the observed and predicted outcomes support hypothesis (a) but not hypothesis (b). Sensitivity analyses varying assumptions about refuges and insect fitness show that the predicted delay in resistance evolution associated with the increased percentage of non-Bt cotton plants provided by F₂ hybrids is robust (SI Appendix, Figs. S3 and S4).

Fig. 3.

Predicted versus observed evolution of pink bollworm resistance to Bt toxin Cry1Ac in the Yangtze River Valley of China from 2008 to 2015. The predictions are from computer simulations of hypotheses proposing that (a) all non-Bt cotton plants act as refuges, including the non-Bt cotton plants in fields of F₂ hybrid cotton or (b) only non-Bt cotton plants in fields of non-Bt cotton varieties act as refuges. The observed values are bioassay data for survival at the diagnostic concentration of Cry1Ac (Fig. 1B and SI Appendix, Table S1).

The correspondence between the outcome in the field and modeling results under hypothesis (a) demonstrates that the increase in the overall percentage of non-Bt cotton plants, including F₂ hybrids, is sufficient to explain the observed lack of increase in resistance from 2011 to 2015 (Fig. 3). Nevertheless, this result does not exclude the alternative hypothesis that an increase in the relative abundance of non-Bt host plants other than cotton contributed to this delay in resistance. To address this alternative hypothesis, we tested the potential of six other common crops in the Yangtze River Valley to serve as pink bollworm host plants. The results of laboratory experiments confirmed previous reports that okra (Abelmoschus esculentus) and tomato (Lycopersicon esculentum) support pink bollworm larval development (19, 27), but no survival occurred on eggplant (Solanum melongena), cucumber (Cucumis sativus), apple (Malus domestica), or kidney bean (Phaseolus vulgaris) (SI Appendix, Table S13). In field experiments, however, pink bollworm larvae and adults were common in cotton, yet none were found in okra or tomato (SI Appendix, Tables S14 and S15). Thus, it is unlikely that non-Bt host plants other than cotton contributed substantially to the observed delay in evolution of pink bollworm resistance to Bt cotton.

A potential drawback of the increased percentage of non-Bt cotton plants from 2011 to 2015 is higher population density of pink bollworm. However, the mean number of pink bollworm eggs per 100 cotton plants was not higher in 2011–2015 (5.4) than in 2006–2010 (11) (Fig. 4A and SI Appendix, Table S16). Furthermore, no significant association occurred between pink bollworm eggs per cotton plant and year from 2011 to 2015 (linear regression: r² = 0.05, df = 3, P = 0.72; Fig. 4A), indicating no steady rise in population density during this period. Relative to the 5 y before Bt cotton was planted (1995–1999), when the mean number of pink bollworm eggs per 100 cotton plants was 94 (SE = 9.4), the abundance in 2011–2015 declined by 96%.

Fig. 4.

Pink bollworm population density and insecticide sprays in the Yangtze River Valley, 1995–2015. (A) Number of F₁ pink bollworm eggs per 100 cotton plants was determined annually for 150 plants per county in each of 12–24 counties (mean = 20 counties, total sample = 63,300 plants; SI Appendix, Table S16). (B) Sprays targeting pink bollworm annually per cotton field based on ∼10 fields per county for 14–16 counties per year (SI Appendix, Table S17). Values in both panels are means with SE.

In parallel, the mean number of insecticide sprays applied yearly against pink bollworm per cotton field decreased from 2.0 in 1995–1999 to 0.64 in 2011–2015, a 69% reduction (Fig. 4B and SI Appendix, Table S17). Moreover, sprays were 27% fewer in 2011–2015 relative to 2006–2010 (mean = 0.88) (Mann–Whitney U test: U = 24, P = 0.02).

Whereas previous work has emphasized problems caused by interbreeding between transgenic and nontransgenic plants (13–15), we discovered that increased planting of F₂ hybrids from crosses between Bt and non-Bt cotton was associated with delayed evolution of pink bollworm resistance to Bt cotton and continued suppression of this voracious pest. Despite extensive global monitoring of resistance to Bt crops during the past two decades and some examples of fluctuations in resistance allele frequency (2, 9, 28), previous reports have not documented any cases where resistance was markedly delayed or reversed after the threshold of 1% resistant individuals was exceeded in a region for several years, as reported here.

The seed mixture generated with F₂ hybrids analyzed here may be especially effective against pink bollworm because of the substantial fitness cost associated with its resistance to Cry1Ac and its recessive inheritance of resistance (25), even when feeding on bolls from F₁ hybrids, Cry1Ac-producing bolls from F₂ hybrids (SI Appendix, Table S18), or a mixture of Bt and non-Bt seeds (24). Selfing of F₂ plants in subsequent generations is expected to maintain roughly 25% Bt homozygotes, 50% hemizygotes, and 25% non-Bt homozygotes. The F₂ hybrid seed mixture planted in China differs from the standard seed mixtures used now for managing resistance to Bt corn in the United States, which do not entail hybridization and have either 90% or 95% homozygous Bt seeds combined in seed bags with 10% or 5% homozygous non-Bt seeds, respectively (15).

The major advantage of a seed mixture generated with F₂ hybrids is a built-in refuge of roughly one-quarter non-Bt plants, which could be especially important in developing countries where planting of separate refuges is low or nil. Whereas compliance with refuge requirements is considered a key factor that delayed evolution of pink bollworm resistance to Bt cotton in the United States, the scarcity of non-Bt cotton refuges probably hastened this pest’s resistance to Bt cotton in India (8, 9). In the Yangtze River Valley, millions of farmers have planted F₂ hybrid cotton not as a resistance management strategy, but apparently for immediate benefits that may include higher yield relative to parental varieties, reduced expenses for insecticides relative to non-Bt cotton, and reduced cost of seeds relative to F₁ hybrids. The efficacy of this strategy for managing resistance in other regions and against other pests remains to be determined.

Materials and Methods

Monitoring Pink Bollworm Resistance to Cry1Ac.

We conducted field sampling and diet bioassays to monitor pink bollworm resistance to Bt toxin Cry1Ac as described previously (18). Briefly, for 2011–2015, we sampled 60–700 (mean = 329) pink bollworm larvae per site from seven to 10 sites per year in the Yangtze River Valley (SI Appendix, Fig. S1; totals: 45 sets of larvae and 14,799 larvae sampled). Larvae from each site were mass-reared on an artificial diet in the laboratory, and their F₁ progeny were tested on a diet containing concentrations of Cry1Ac ranging from 0 to 9 μg of Cry1Ac per milliliter of diet, which is a diagnostic concentration that kills all or nearly all susceptible larvae but few or no resistant larvae. Neonates were tested individually, with 24 neonates tested per concentration per replicate and three replicates for each concentration. After 21 d in darkness at 29 ± 1 °C, live fourth instars and pupae were scored as survivors. Each year, as an internal control, we used the same methods to test the susceptible strain QJ-S, which was started with insects collected from Qianjiang, Hubei province, in 2004 and reared in the laboratory without exposure to toxin. Results from 2005 to 2010 were reported previously (18) and are included here for comparison with the new data for 2011–2015.

Percentage of Cotton Seeds Containing Cry1Ac.

We used immunoassays to determine the percentage of cotton seeds producing Cry1Ac for cotton hybrids, Bt cotton varieties, and non-Bt cotton varieties (SI Appendix, Tables S3–S6). In each year from 2010 to 2015, we purchased 1,200–1,375 g of seeds of each of 20–30 cotton hybrids and varieties that were recommended that year by the local government. Each year, we randomly selected 100 seeds from each hybrid or variety and divided them into four groups with 25 seeds per replicate. We tested each seed individually for Cry1Ac in a total of 140 assays (total n = 14,000 seeds tested individually) of six varieties of Bt cotton, 10 varieties of non-Bt cotton, and 68 hybrids.

We used Cry1Ab/Cry1Ac immunoassay test strips (Aochuangjinbiao Biotech) according to the manufacturer’s instructions. Each replicate of 25 seeds had a separate positive and negative control. Each seed was stored in a 2-mL CryoPure tube (Sarstedt) with a 6-mm-diameter oil-free steel ball and 400 μL of leaching solution (Aochuangjinbiao Biotech). After centrifugation at 5,000 rpm for 18 seconds in a bead mill homogenizer (Precellys 24; Bertin Technologies), we added 600 more μL of leaching solution before analyzing each sample with a test strip. After 5 min, the appearance of a quality control line indicated the strip worked properly and the appearance of a second line indicated detection of Cry1Ac.

For each of the 68 hybrids, if the mean percentage of seeds producing Cry1Ac was >87.5% (mean of the expected 100% for F₁ seeds and 75% for F₂ seeds), the hybrid was classified as F₁. If the mean percentage of seeds producing Cry1Ac was <87.5%, the hybrid was classified as F₂. Based on this criterion, which aligns with a gap in the observed distribution (SI Appendix, Fig. S2), we classified 23 hybrids as F₁ (n = 38 assays, mean = 96.3%, SE = 0.5%, range = 89–100%; SI Appendix, Table S5) and 45 hybrids as F₂ (n = 82 assays, mean = 67.1%, SE = 1.4%, range = 39–84%; SI Appendix, Table S6).

Percentage of Cotton Planted to F₁ and F₂ Hybrid Cotton, Bt Cotton, and Non-Bt Cotton.

For each year from 2004 to 2015, we calculated the percentage of cotton hectares planted to hybrid cotton, Bt cotton, and non-Bt cotton using data from China’s Ministry of Agriculture for the six provinces in the Yangtze River Valley we studied (Anhui, Hubei, Hunan, Jiangsu, Jiangxi, and Sichuan). Because the Ministry’s data do not distinguish between F₁ and F₂ hybrids, we used two methods to calculate the proportion of hybrid cotton planted with each of these two hybrids. For 2004–2009, we used responses regarding planting of F₁ and F₂ hybrids from a telephone survey of 12 seed companies (SI Appendix, Table S8). For 2010–2015, we used the data from the percentage of seeds containing Cry1Ac for the 68 hybrids (SI Appendix, Tables S5 and S6) in conjunction with the data from Ministry of Agriculture on the hectares planted to each of the 68 hybrids (SI Appendix, Tables S9 and S10). For each year, we summed the hectares planted to the 23 F₁ hybrids and, separately, to the 45 F₂ hybrids. We calculated the proportion of hybrid cotton planted to F₁ (or F₂) hybrids in each year as the hectares planted to the F₁ (or F₂) hybrids divided by the total hectares planted to the F₁ plus F₂ hybrids. For 2010–2015, the mean percentage of total hectares planted to the 68 tested hybrids was 58.4% (SE = 6.3). For 1 y, 2010, we have the data on the percentage of hybrids consisting of F₂ hybrids for both methods: 52% from the survey versus 77% based on analysis of Cry1Ac in seeds and hectares planted to each hybrid. This contrast suggests that the survey may underestimate the planting of F₂ hybrids.

Percentage of Cotton Planted to Non-Bt Cotton.

To calculate the percentage of cotton hectares planted to non-Bt cotton plants in each year, we multiplied the proportion of non-Bt cotton seeds (1 minus the proportion containing Cry1Ac based on the assays described above) for each of the four types of cotton (F₁ and F₂ hybrid cotton, Bt cotton, and non-Bt cotton) by the percentage of cotton hectares planted of the corresponding type of cotton in that year (Fig. 2A), and then summed the four resulting percentages. For each of the 23 F₁ hybrids and 45 F₂ hybrids, we calculated the mean proportion of non-Bt seeds for 2010–2015. For each year from 2010 to 2015, we calculated the hectares of non-Bt cotton plants for each F₁ hybrid by multiplying the mean proportion of non-Bt cotton seeds for that hybrid by the hectares planted to that hybrid in that year. We calculated the percentage of non-Bt cotton plants for all F₁ hybrids for each year from 2010 to 2015 as 100-fold the total hectares of non-Bt cotton plants for all F₁ hybrids divided by the total hectares of all F₁ hybrids. We used the same approach to calculate the percentage of non-Bt cotton for all F₂ hybrids for each year from 2010 to 2015. For 2004–2009, when seeds were not tested for Cry1Ac, we used the mean proportion of non-Bt cotton seeds based on all assays from 2010 to 2015 for F₁ hybrids and separately for F₂ hybrids. Relative to the extensive data on Cry1Ac in seeds for F₁ hybrids (38 assays) and F₂ hybrids (82 assays), which enabled calculation of the proportion of non-Bt cotton seeds separately for each year from 2010 to 2015, we had fewer assays for varieties of Bt cotton (nine assays) and non-Bt cotton (11 assays). Accordingly, we used the mean proportion of non-Bt cotton seeds from 2010 to 2015 to estimate the percentage of non-Bt cotton plants for all years for Bt cotton varieties (mean = 0.013, SE = 0.006, n = 9 assays of six varieties; SI Appendix, Table S3) and non-Bt cotton varieties (mean = 0.77, SE = 0.02, n = 11 assays of 10 varieties; SI Appendix, Table S4).

Computer Simulations.

To assess the potential effects of F₂ hybrid cotton on the evolution of pink bollworm resistance to Bt cotton in the Yangtze River Valley of China, we used a previously described deterministic population genetic model of resistance to a Bt crop (26, 29) with some minor modifications. We chose this simple, previously described population genetic model for several reasons: to make clear the link between the model and the hypothesized increase in refuge percentage associated with planting F₂ hybrids, to enable incorporation of realistic biological parameters for pink bollworm in the Yangtze River Valley, to examine projected outcomes of different assumptions about refuges and fitness values using the same basic model, and to make the modeling results readily verifiable by readers. Furthermore, as reported previously (26), the results from this model match precisely with the results of Gould (30) and are similar to results from much more complex models, such as the results of Gustafson et al. (31). The standard values for parameters and the alternative values used in sensitivity analyses are listed in SI Appendix, Table S12.

We assumed that resistance to Cry1Ac is controlled by two alleles at a single autosomal locus: a recessive allele r conferring resistance and an allele s conferring susceptibility. This assumption is based on previous analyses of pink bollworm resistance to Cry1Ac in many laboratory-selected strains from Arizona and in a strain derived from three field-selected populations in India (32–35). Data reported here for a laboratory-selected resistant strain (AQ47) of pink bollworm derived from the Yangtze River Valley also demonstrate autosomal recessive inheritance of resistance to Cry1Ac in bolls of Bt cotton, F₁ hybrid cotton, and F₂ hybrid cotton (SI Appendix, Table S18). In addition, preliminary data indicate that in several strains of pink bollworm established recently from populations in the Yangtze River Valley, resistance to Cry1Ac is conferred by mutations in the same cadherin gene that is associated with recessive resistance to Cry1Ac in pink bollworm from Arizona and India (32–36). We used the data from China reported here and the extensive, previously published empirical data on pink bollworm tested in greenhouse or field experiments to estimate the relative fitness of each genotype on Bt cotton and non-Bt cotton (25) (SI Appendix, Tables S12 and S18). To reduce the influence of differences between strains unrelated to resistance, we estimated fitness on non-Bt cotton using data only from experiments with resistant and susceptible strains that shared a common genetic background (i.e., the resistant strain was derived from the susceptible strain via laboratory selection).

We assumed random mating among all adults and projected the change in r allele frequency from one generation to the next as follows:

$$\mathrm{\Delta }p=pq\left(p\left[{\text{W}}_{\text{rr}}\mathrm{-}{\text{W}}_{\text{rs}}\right]+q\left[{\text{W}}_{\text{rs}}\mathrm{-}{\text{W}}_{\text{ss}}\right]\right)/{\text{W}}_{\text{M}},$$ [1]

where p and q are the frequencies of the r and s alleles, respectively; W_M is the population mean fitness; and W_rr, W_rs, and W_ss are the fitnesses of rr, rs, and ss, respectively. Fitness for each genotype was calculated as follows:

$${\text{W}}_{\text{rr}}={\text{BtW}}_{\text{rr}}\cdot {\text{P}}_{\text{Bt}}+{\text{RefW}}_{\text{rr}}\cdot {\text{P}}_{\text{Ref}},$$ [2]

$${\text{W}}_{\text{rs}}={\text{BtW}}_{\text{rs}}\cdot {\text{P}}_{\text{Bt}}+{\text{RefW}}_{\text{rs}}\cdot {\text{P}}_{\text{Ref}},$$ [3]

$${\text{W}}_{\text{ss}}={\text{BtW}}_{\text{ss}}\cdot {\text{P}}_{\text{Bt}}+{\text{RefW}}_{\text{ss}}\cdot {\text{P}}_{\text{Ref}},$$ [4]

where P_Bt is the proportion of Bt cotton plants and P_Ref is the proportion of refuge (non-Bt cotton plants), and fitnesses for rr, rs, and ss, respectively, are BtW_rr, BtW_rs, and BtW_ss on Bt plants and RefW_rr, RefW_rs, and RefW_ss in refuges. Population mean fitness was calculated with the standard equation (37):

$${\text{W}}_{\text{M}}=p^2{\text{W}}_{\text{rr}}+2pq{\text{W}}_{\text{rs}}+q^2{\text{W}}_{\text{ss}}.$$ [5]

We iterated Eq. 1 with a computer program to project changes in the r allele frequency over successive generations.

For each year, we simulated three generations, corresponding to the three annual generations of pink bollworm in the Yangtze River Valley (38). We modeled evolution of resistance from the end of the third generation of 2008 (end of the season) to the end of the third generation of 2015 (end of the season). We started with 2008 because 2008 is the first year that resistant individuals were detected in the Yangtze River Valley (18), which allowed us to use the monitoring data to set the percentage of resistant individuals (rr) at the end of 2008 as the observed value of 0.88% (18).

To make robust comparisons between predicted and observed outcomes, we used the means for parameters in the model (SI Appendix, Table S12) and for the observed percentage of resistant individuals in field populations based on survival at the diagnostic concentration of Cry1Ac (SI Appendix, Table S1). We also conducted sensitivity analyses with the model to assess the potential effects on resistance evolution of variation in the refuge percentage and variation in the fitness of rr on Bt cotton and non-Bt cotton, which yields variation in incomplete resistance and fitness cost (39), respectively. In sensitivity analyses, we modeled four assumptions about which cotton plants act as refuges: (i) hypothesis (a): all non-Bt cotton plants, including plants in fields of F₂ hybrids, F₁ hybrids, non-Bt cotton, and Bt cotton; (ii) hypothesis (b): only non-Bt cotton plants in fields of non-Bt cotton varieties; (iii) hypothesis (c): only non-Bt cotton plants in fields of F₂ hybrid cotton; and (iv) hypothesis (d): non-Bt cotton plants in fields of F₁ hybrids, non-Bt cotton varieties, and Bt cotton varieties (but not F₂ hybrids) (Fig. 3 and SI Appendix, Figs. S3 and S4). Based on experimental results (25), the standard fitness of rr was 0.16 on Bt cotton and 0.46 on non-Bt cotton. In sensitivity analyses, we also multiplied the fitness of rr by 1.25 on Bt cotton (0.16 × 1.25 = 0.20) and on non-Bt cotton (0.46 × 1.25 = 0.58), which tends to make resistance evolve faster by reducing the negative effect of incomplete resistance on Bt cotton and by reducing the fitness cost on non-Bt cotton. In a complementary sensitivity analysis, we divided the fitness of rr by 1.25 on Bt cotton (0.16/1.25 = 0.13) and on non-Bt cotton (0.46/1.25 = 0.37), which tends to slow evolution of resistance by increasing the negative effect of incomplete resistance on Bt cotton and by increasing the fitness cost on non-Bt cotton.

Host Plant Experiments: Larval Feeding in the Laboratory.

We conducted larval feeding experiments to test a non-Bt variety of cotton (Simian3) and six other crops common in the Yangtze River Valley as potential pink bollworm host plants: okra, tomato, eggplant, cucumber, kidney bean, and apple. The following procedures were followed for all crops except apple. The crops were planted at the Wuhan experimental farm of the Hubei Academy of Agricultural Sciences in mid-April 2014. No insecticides were used throughout the growing season. In mid-August, for each crop, we collected >50 fruits (i.e., bolls, pods, other fruits), with one to three fruits collected per plant. We bought apples from a supermarket and soaked them in water for 2 h to reduce potential pesticide residues. For the fruits used in experiments, the approximate diameter was 2 cm for cotton, and it was 7 cm for tomato and apple; the approximate length was 12 cm for okra, 10 cm for eggplant, 25 cm for cucumber, and 15 cm for kidney beans. All fruits were rinsed with water in the laboratory, air-dried, and placed in plastic boxes (43 × 32 × 22 cm). The number of fruits per box was 34 for cotton and okra; 54 for tomato, eggplant, cucumber, and apple; and 108 for kidney bean. For each of the seven crops, we tested three replicates with one replicate per box.

We used a soft fine brush to put neonates (less than 1 h old) of the susceptible QJ-S strain of pink bollworm on the fruits. The number of neonates per fruit was five for cotton, okra, and kidney bean and 10 for the others. After inoculation, the boxes were closed and held at 29 ± 1 °C, 60 ± 10% relative humidity (RH), and 13 h light/11 h dark. One day later, we counted the number of entrance holes under a stereoscopic microscope. Five days after inoculation, we put soft paper into each box to facilitate pupation of larvae emerging from fruit. Larvae emerged and pupated over a period of more than 2 wk. We stopped tracking pupation after no larvae had emerged for 3 consecutive days from any fruit.

Host Plant Experiments: Larval Infestation in the Field.

We measured infestation of non-Bt cotton (Simian3), okra, and tomato by pink bollworm larvae in field plots at the Wuhan experimental farm of the Hubei Academy of Agricultural Sciences in 2014 and 2015. We planted the crops in a total of six fields, consisting of the following three pairs of crops in adjacent fields: cotton and okra, cotton and tomato, and okra and tomato. Within each pair, each crop was planted on 1.8 × 36 m and separated from the other crop in the pair by 0.5 m. The distance between paired fields was 2 m. Crops were planted in mid-April, and fruits were sampled six times: early, middle, and late August and September. Each sample consisted of 100 randomly collected fruits of each crop. The diameter was ∼2 cm for cotton bolls and 10 cm for tomatoes, and okra length was ∼12 cm. All samples were stored in plastic boxes (43 × 32 × 22 cm) covered with 100-mesh gauze, and held in the laboratory at 29 ± 1 °C, 60 ± 10% RH, and 13 h light/11 h dark. After 10 d, we checked the fruits for emergence holes, and then opened the fruits to check for live larvae. We calculated the total number of pink bollworm larvae per fruit as the number of emergence holes plus the number of live larvae per fruit.

Host Plant Experiments: Adult Males Caught in Pheromone Traps.

We used traps baited with pink bollworm female sex pheromone (Pherobio Technology Co.) to catch male pink bollworm moths at field sites in four counties (provinces): Anqing (Anhui), Anxiang (Hunan), Qianjiang (Hubei), and Wuxue (Hubei). Cotton, okra, and tomato were planted in each county, with roughly 50 km separating fields of the three different crops within a county (four counties × three field sites per county = total of 12 field sites). At each of the 12 sites, three fields of the same crop were planted with one trap in each field (total of three traps per site × 12 sites = 36 traps). Males caught in each trap were counted daily from June through September. For data analysis, the moths caught were totaled for each of the 12 traps per crop for each of the 4 mo of the field study.

Pink Bollworm Population Density.

To evaluate pink bollworm population density, we used the number of first-generation pink bollworm eggs per 100 cotton plants. A total of 150 cotton plants per county were sampled every 4 d in 12–24 counties of six provinces of the Yangtze River Valley (Anhui, Hubei, Hunan, Jiangsu, Jiangxi, and Sichuan) during June and July each year from 1995 to 2015. For each sampled plant, a thorough visual whole-plant survey was conducted to count the pink bollworm eggs, and each egg found was removed. Pink bollworm females do not distinguish between Bt and non-Bt cotton for oviposition (40), and the type of cotton plant (Bt or non-Bt) was not recorded.

Insecticide Sprays Targeting Pink Bollworm.

We obtained the data on insecticide sprays targeting pink bollworm from the National Agricultural Technology Extension and Service Center, which has a network of plant protection stations that report these data. The number of sprays targeting pink bollworm in each of 14–16 counties per year is a mean based on approximately 10 cotton fields per county.

Dominance of Resistance to Cotton Bolls Producing Cry1Ac in a Pink Bollworm Strain from the Yangtze River Valley.

We determined the dominance of pink bollworm resistance to Cry1Ac-producing bolls by testing four types of larvae derived from the Yangtze River Valley: resistant (AQ47 strain), susceptible (QJ-S strain), F₁ progeny from resistant females × susceptible males, and F₁ progeny from susceptible females × resistant males (additional details are provided in SI Appendix). We used survival on bolls containing Cry1Ac of the resistant strain, the susceptible strain, and their F₁ progeny to calculate dominance (h) of resistance (which ranges from 0 for completely recessive to 1 for completely dominant) as described previously (41).

Data Analysis.

We analyzed diet bioassay data with probit regression (42) to determine LC₅₀ values and their 95% fiducial limits, as well as slopes of the concentration-mortality lines. We calculated the resistance ratio as the LC₅₀ for a strain divided by the LC₅₀ for the susceptible QJ-S strain tested in the same year. We used Fisher’s exact test to determine if the proportion of populations with one or more survivors at the diagnostic concentration differed between 2008–2010 and 2011–2015. We used the Mann–Whitney U test to determine if percentage survival at the diagnostic concentration differed between 2008–2010 and 2011–2015.