Article pubs.acs.org/est
Cite This: Environ. Sci. Technol. 2018, 52, 2314−2322
Insecticide Resistance Signals Negative Consequences of Widespread Neonicotinoid Use on Multiple Field Crops in the U.S. Cotton Belt Anders S. Huseth,*,† Thomas M. Chappell,‡ Anitha Chitturi,§ Alana L. Jacobson,§ and George G. Kennedy† †
Department of Entomology and Plant Pathology, North Carolina State University, Campus Box 7630, Raleigh, North Carolina 27695, United States ‡ Department of Plant Pathology and Microbiology, Texas A&M University, 435 Nagle Street, College Station, Texas 77843, United States § Department of Entomology and Plant Pathology, Auburn University, 301 Funchess Hall, Auburn, Alabama 36849, United States S Supporting Information *
ABSTRACT: The intensification of industrial agriculture has been enabled by improved crop varieties, genetically engineered crops, fertilizers, and pesticides. Over the past 15 years, neonicotinoid seed treatments have been adopted worldwide and are used on a large proportion of U.S. field crops. Although neonicotinoids are used widely, little is known about how large-scale deployment affects pest populations over long periods. Here, we report a positive relationship between the deployment of neonicotinoid seed-dressings on multiple crops and the emergence of insecticide resistance in tobacco thrips (Frankliniella f usca), a polyphagous insect herbivore that is an important pest of seedling cotton but not soybean or maize. Using a geospatial approach, we studied the relationship between neonicotinoid resistance measured in 301 F. f usca populations to landscape-scale crop production patterns across nine states in the southeastern U.S. cotton production region, in which soybean, maize and cotton are the dominant crops. Our research linked the spatiotemporal abundance of cotton and soybean production to neonicotinoid resistance in F. f usca that is leading to a dramatic increase in insecticide use in cotton. Results demonstrate that cross-crop resistance selection has important effects on pests and, in turn, drives pesticide use and increases environmental impacts associated with their use.
1. INTRODUCTION The intensification of agricultural production and associated increases in farm size have characterized industrial agriculture in much of the world since the Green Revolution. Productivity increases have been enabled by rapid technological advances in conventional crop breeding, genetically engineered crops, fertilizer use, and pesticides;1−4 this transition has resulted in significant improvement in farm-level productivity and operational efficiency. Among crop protection technologies targeting pests, neonicotinoid seed treatments (NSTs) and genetically engineered crops expressing insecticidal traits have been adopted on a global scale, especially in field crops (i.e., maize, cotton, soybean).5,6 When applied to seeds, neonicotinoids (e.g., imidacloprid, thiamethoxam, clothianidin) provide systemic insecticidal protection from pests of field crop seedlings.7,8 Because crop seeds are treated with these neonicotinoids before planting, farmers value the improved efficiency and accuracy of NST pest protection in field crop production systems.9 A comprehensive analysis of NST use in the US found that the frequency and extent of use has rapidly increased in field crops since 2003.10 In a 14-state survey of US © 2018 American Chemical Society
soybean producers, an estimated 87% of soybean area planted in those states were treated with NSTs.11 A similar pattern of adoption was documented in cotton, where NST use was estimated to be 90% of planted acreage in the Mid-South and Southeast portions of the US Cotton Belt in 2017.12,13 When applied at broad spatial scales,14 NSTs and GM crops have provided farmers with effective and efficient means to manage risk of losses to pests, and their long-term use at broad spatial scales has suppressed some target pest populations and reduced the quantity of broad-spectrum insecticide applied.15−23 While many studies have examined the indirect effects of neonicotinoid insecticides on nontarget species (pollinators, birds) and the environment,24−32 few have connected the vast spatial scale of cross-crop neonicotinoid seed dressing deployment to any direct effects that their long-term use has on insect pest populations over time.15 Received: Revised: Accepted: Published: 2314
November 22, 2017 January 5, 2018 January 23, 2018 January 23, 2018 DOI: 10.1021/acs.est.7b06015 Environ. Sci. Technol. 2018, 52, 2314−2322
Article
Environmental Science & Technology
quantifying crop production surrounding resistant and susceptible F. f usca populations, we show the importance of both cotton and soybean in selection for resistance in one common polyphagous pest species. Accordingly, we were able to use the quantification of crop presence to geographically project estimates of consistency of selection for neonicotinoid resistance, constituting a resistance likelihood map useful in describing one impact of widespread NST deployment in multiple field crops. This study provides some of the first evidence documenting a direct linkage between the scale of NST use and a measurable, long-term biological impact on an arthropod species at a regional scale.
In simplified agroecosystems, where cropland is a dominant component of the landscape, the magnitude of insecticide use positively relates to the proportion of available cropland in space and time.33,34 This stable trend in spatiotemporal pesticide use scales with available cropland and indicates a logical link between the intensity of agricultural production and the cascading effects of recurring insecticide use on insect populations.33,34 However, this broad relationship does not reveal specific interactions between individual insect species and crop-specific insecticide use among crop hosts; information that is necessary to understand the full impact of widespread insecticide use in agricultural landscapes. In this study, we connect an emerging problem of neonicotinoid resistance in tobacco thrips (Frankliniella f usca, Hinds) to the large-scale deployment of NSTs in cotton and soybean. Frankliniella f usca is an important polyphagous insect pest of cotton that also infests but is rarely damaging to soybean; it also infests several other crops, and many abundant non-crop plants.35−39 On the basis of nearly all the acreage being planted with cotton and soybeans that are treated with a neonicotinoid seed treatment, we used the abundance of these two crops in space and time to describe variation in resistance over multiple years in the Cotton Belt region across nine southeastern US states (AL, AR, GA, LA, MS, NC, SC, TN, VA). To do this, we measured neonicotinoid resistance in F. f usca populations. We then developed a measure of cotton and soybean presence that takes into account both spatial and temporal influence of crop presence and related that measure to the occurrence of neonicotinoid resistant F. f usca. Because NSTs are used on nearly all F. f usca host crop area in the Cotton Belt (i.e., cotton, on which it is a consistent and important pest, and soybean, on which it is not),10,11,40 the abundance of these crops in space and time represents the consistency of potential selection for resistance at the landscape level. We also assessed the potential role of other crop species where neonicotinoids are commonly used (e.g., maize, peanut, sorghum, tobacco). Of the remaining annual crops, maize represents the only widespread field crop where NSTs are a recurring part of insect pest management (>80% of US maize receives an NST).10 In the Cotton Belt, peak dispersal of F. f usca typically occurs in late May when maize is at midvegetative development (greater than V4).41,42 A recent study of temporal NST concentration in maize, showed that the concentration of active NST active ingredient declines significantly soon after emergence.43 This indicates that for most maize grown in the region NST concentration has peaked well before major F. f usca dispersal events. Together, the asynchrony of crop-pest phenology and absence of documented reproduction by F. f usca on the NST-treated maize suggests that this crop species is not a major selection patch in the landscape. Despite the limited biological linkage between F. f usca resistance selection and these alternative host crop species, we tested for a relationship with all other neonicotinoid-treated crops that could potentially select for F. f usca resistance. To do this, we assessed the contribution of annual agricultural crops excluding cotton or soybean that could represent potential NST resistance selection patches in the landscape. Specifically, our questions were as follows: (i) How does spatiotemporal intensity of cotton production relate to neonicotinoid resistance levels in F. f usca? (ii) Is understanding of resistance patterns enhanced by accounting for spatiotemporal soybean production intensity in this system? By
2. MATERIALS AND METHODS 2.1. Field Sites and Neonicotinoid Resistance Bioassays. From 2014 to 2016, we sampled 301 F. f usca populations from cotton production agroecosystems in nine different southern states (Figure 1). Over three study years, the
Figure 1. Spatial distribution of F. f usca collection locations (N = 301 unique sites). Color shading denotes the average annual area of cotton production by county reported to the US Farm Service Agency, 2007− 2016.
average distance separating sample locations was 14.8 km (±16.1 SD, min. 0, max 104.8). Fourteen samples were collected from sites separated by less than 0.5 km among three years and represented 4.7% of the total sample number. The average distance separating sample locations within any year was 22.0 km (±29.3 SD, min. 0.3, max 239.5). Frankliniella f usca were sampled from one of five different agricultural crops species or from non-crop, host-plant species within cotton production agroecosystems. All samples were collected in spring between April and Mid-June in each year; prior to or during the period when F. f usca are actively colonizing and infesting seedling cotton. In crop fields, F. f usca were sampled from alfalfa, cotton, peanut, wheat, or soybean. Because insecticide-treated crop plants could eliminate susceptible individuals from F. fusca populations prior to bioassays, we requested pesticide treatment history from cooperating growers. In total, 12 populations (4%) were collected from neonicotinoid-treated crop plants. We observed no detectable effect of collection from neonicotinoid-treated crop plants on measured neonicotinoid sensitivity of these populations in our diet-based bioassay (see below) when compared to neighboring collections. To limit potential confounding effects of host plant treatment, we collected 188 2315
DOI: 10.1021/acs.est.7b06015 Environ. Sci. Technol. 2018, 52, 2314−2322
Article
Environmental Science & Technology
from publicly available, remotely sensed USDA National Agricultural Statistics Service-Cropland Data Layer (CDL) that has a spatial resolution of 30 × 30 m.45 Classification accuracy estimates for CDL crop and non-crop land use categories for each state can be found in CDL metadata resources.45 Annual CDL crop data were reclassified into general land use categories (annual or perennial crops, noncrop) with subcategories grouping similar land use types (Table S2). Because cotton and soybean are both hosts capable of supporting large populations of F. f usca and are treated with neonicotinoids, we hypothesized the spatiotemporal abundance of these crops would be related to reduced sensitivity in F. f usca. Maize was also a dominant annual agricultural component in the Cotton Belt agricultural landscape (Table S3). Because the rate of NST use in maize is high,10 this crop could also be a potential driver of F. f usca neonicotinoid resistance. Similarly, small grain, tobacco, and peanut were abundant in some landscapes, although the area of cultivated land occupied by these crops was smaller than either cotton or soybean (Table S3). The widespread use of neonicotinoid insecticides on these crops may impose resistance selection at some locations where there is temporal overlap in F. f usca occurrence with presence of seedlings. To account for locally specific selection in alternate neonicotinoid treated hosts, we grouped all other crops excluding cotton and soybean to account for potential selection in minor crops or maize. Annual production history in each surrounding collection location was determined for three crop types (cotton, soybean, and annual agriculture excluding cotton and soybean) for each of the six sequential years prior to each collection. The year prior to each bioassay year was used to characterize the abundance of all other land use types in the landscape. Landcover data were reclassified into a binary layer (e.g., focal crop or not) and iteratively extracted for each bioassay location and prior crop year combination using Python and the integrated ArcPy site package (Table S4).46 2.3. Regional Crop Composition Potential for Resistance Selection. The emergence of resistance results from an increase in frequency of resistance alleles in response to selection over time. To capture both the spatial and temporal aspects of selection resulting from exposure to NST-treated crops at a landscape scale, we derived a variable, “weighted hectare years” (WHY), from CDL crop presence data. WHY was conceived to be a measure of spatiotemporal crop presence that could be used to characterize potential selection at a given place and time. In developing the WHY term, we assumed that nearby neonicotinoid use is more influential on the affected population than is far neonicotinoid use. We also assumed that neonicotinoid used in the distant past has less influence on a given population than that used recently. The form of each of these relationships is the basis for the weighting in WHY. The units of data used are denominated in area per year, so that a given instance of crop production is a hectare year. The farther away from an affected thrips population that hectare is, and the greater the elapsed time since neonicotinoid was used on that hectare, the less the weight of that hectare year in the calculation of WHY. We chose simple functions for the spatiotemporal weighting. For space, the inverse square law is a reasonable interpretation of decreasing influence with distance. The combined effects of organismal dispersal, allelic migration, and other unknown
of 301 (62%) populations from non-crop weeds growing in unmanaged crop boundaries and road verges. In total, 243 of 301 (81%) F. f usca populations were collected from host plants, where no recent insecticide applications were made (Table S1). Two different methods, sweep net and destructive plant samples, were used to collect adult female F. f usca at field sites depending on the type of host plant. For non-crop weeds and small grain sites (e.g., wheat, rye, oats), female F. f usca were collected using a sweep net and then aspirated to sample containers for transportation. For the crop plant samples, 300− 400 plants were collected at random from production fields and were layered into buckets. The spatial location, host plant identity, and insecticide treatment history (if available) of the host plant were recorded for each sample location. Filled sample buckets were cooled with ice packs for transport. Upon arrival at the laboratory, adult F. f usca females were identified then transferred to thrips-proof cages. Individual populations were maintained at 27 °C on insecticide-free white cabbage (Brassica oleracea var. capitata L.) until entered into bioassays. Over three study seasons, the period between collection and bioassay ranged from 0 to 49 days, with a mean of 7 days (±8 SD). A diet-based diagnostic dose bioassay was used estimate neonicotinoid sensitivity in field-collected F. f usca populations. Each assay administered formulated imidacloprid (Gaucho 600FS, Bayer CropScience, Research Triangle Park, NC USA) or thiamethoxam (Cruiser 5FS, Syngenta Crop Protection, Greensboro, NC USA) in a 3% sucrose aqueous solution at a rate of 360 mg of formulated insecticide per L of sucrose diet (0.175 g imidacloprid L−1 sucrose diet; 0.168 g thiamethoxam L−1 sucrose diet).44 Insecticide diets were prepared with green food dye (0.5% v/v) to improve insect feeding. All insecticide diets were made with formulated insecticide to replicate common NSTs used in crop protection.44 These formulations contain additional ingredients that improve application to seeds (e.g., glycerin, propylene glycol). Preliminary testing of the formulation without insecticide active ingredients in diet revealed no significant mortality effect when compared to a sucrose control diet alone.44 Snap cap 1.5 mL microcentrifuge tubes were used to contain groups of thrips throughout the bioassay. A 155 μL aliquot of diet solution containing insecticide or untreated diet alone was pipetted into the interior surface of each detached microcentrifuge cap. A 2 cm square of parafilm was then stretched across the solution and wrapped around the cap to provide a membrane for thrips to feed through. Five to eight adult female F. f usca were aspirated into individual bioassay tubes, and then caps with diet and feeding membranes were replaced. Between 10 and 20 replicate tubes per formulated insecticide were administered to each population. Five tubes of insects received untreated sucrose diet as a negative control treatment. Insects were held in a temperature-controlled chamber at 27 °C for 48 h postenclosure. After the exposure period, insects were classified as alive, moribund, or dead. Moribund insects (those unable to coordinate movement greater than one body length when agitated with a fine brush), were considered incapable of recovery and classified as dead for analysis. 2.2. Agricultural Landscape Analysis. To determine the agroecosystem composition surrounding sampled F. f usca populations, collection points were buffered in concentric annular rings with outer edge distances of 5, 10, 15, 20, 25, or 30 km. Composition of managed agricultural and seminatural habitats surrounding sampled fields in each season were derived 2316
DOI: 10.1021/acs.est.7b06015 Environ. Sci. Technol. 2018, 52, 2314−2322
Article
Environmental Science & Technology
hexagons were populated with an attribute variable identifying independent polygons with the Identity Tool in ArcGIS.46 2.4. Statistical Analysis. To relate WHY to bioassay results of individual populations, we analyzed the log-odds of survival during bioassays and regressed these data against site-wise WHY terms (i.e., WHYcotton, WHYsoybean, WHYcotton or soybean, or WHYannual agriculture excluding cotton and soybean). The GLM procedure of the SAS System version 9.4 was used (SAS Institute, Cary, NC). To relate WHY to neonicotinoid resistance estimates from F. f usca bioassays within hexagonal tiles, averages of logodds of survival during bioassays were calculated for samples taken within each sampling neighborhood. Neighborhoods having fewer than five bioassay data points were not included. These neighborhood-level values were modeled as WHY nested within the year in which the bioassayed thrips population was collected, with random intercepts for each of the three collection years.
spatial processes are all expected to affect a given distant point with decreasing intensity or probability as the distance increases. Because neonicotinoids degrade over time and organisms and alleles migrate from sites of use over time, exponential decay is a reasonable interpretation of decreasing influence over time. We assumed a neonicotinoid-inf luence halflife of one year. This is an estimate of the rate at which neonicotinoid use decreases in influence on the resistance of a thrips population over time; reflecting the combined effects of the half-life of neonicotinoid effects, the half-life of resistance in a thrips population and other known and unknown processes. The weighting function leading to WHY is shown in eq 1 6
WHY =
6
⎡
∑ ∑ ⎢CotSoytd(e−λt ) t=1 d=1
⎣
1 ⎤ ⎥ (5d)a ⎦
(1)
The age and distance of each NASS-CDL pixel designated as either cotton or soybean is used to calculate that pixel’s contribution to the summation of all pixels for each of six years (t) and six distances (d, where each distance is the midpoint between the near and far rings bounding a band). According to the earlier description of chosen functions, the parameter (lambda) for decay is set to solve the eq 1 = Ln(2/λ), approximately 0.693, and the parameter (alpha) is set to 2 as the inverse square. To understand the recurrence of NST crop production in space, we calculated WHY for locations across the entire Southeast region. To generate a sample extent, we used the US Department of Agriculture Natural Resources Conservation Service Land Resource Regions coverage.47 Using population collection locations as a spatial selection criteria, we extracted four Land Resource Regions that overlapped the sample points. These included: Atlantic and Gulf Coast Lowland Forest and Crop Region, East and Central Farming and Forest Region, Mississippi Delta Cotton and Feed Grains Region, and South Atlantic and Gulf Slope Cash Crop Region, Forest, and Livestock Region. Regions were dissolved into a single polygon and 3800 random points were distributed within this extent using the Generate Random Point tool in ArcGIS. Crop abundance for each point was calculated using the same methods as described above. From these data, WHY values were calculated. To generate a visual interpretation of the spatial distribution of WHY values, a spline interpolation of logtransformed WHY values was used. To determine the interrelatedness among bioassay locations, the spatial threshold clustering of sample bioassay responses was determined with incremental spatial autocorrelation for log odds of F. f usca bioassay survival using the Incremental Spatial Autocorrelation Tool in ArcGIS. The peak value (∼80 km) indicates a distance where spatial processes promoting clustering are most pronounced. The Create Hexagons tool was used to generate a lattice of 227 80 km diameter hexagons across the dissolved Land Resource Region layer.47 The hexagon lattice and participation of 20 random WHY points (n = 4540 total points) within each independent hexagon was not optimized to maximize the number of site-wise observations per tile. The average distance between random locations was 8.6 km (±4.4 SD). We calculated cotton and soybean WHY production intensity over 6 y (2010−2015) for each random location and used these values to visualize hexagon tile-level production intensity across the region. All sample population collection points within discrete sampling
3. RESULTS 3.1. Neonicotinoid Resistance Survey. From 2014 to 2016, we collected 301 F. f usca populations from 118 counties in nine different states located in the US Cotton Belt (Figure S1). The majority (96%) of F. f usca populations were sampled from cotton producing counties where intensive management of F. f usca with NSTs occurs annually. Because most populations were collected in the spring of the year before resistance selection could occur, we used the prior growing seasons to characterize patterns in crop production and land use. Landscapes surrounding F. f usca collection sites were dominated by seminatural habitat with smaller amounts of annual and perennial cropland (Table S3). Of the available cropland used to grow annual crops, 39.9 ± 5.4% of the area was used for either cotton or soybean production (±SD, min. 20.9%, max 47.3.%, Figure S1). Excluding cotton and soybean area, 63.5 ± 26.9% of the remaining annual agricultural land was used for corn production (min. 8.1%, max 99.9%). Together, 87.8 ± 13.8% of the annual crop area was used for either corn, cotton or soybean production (±SD, min. 37.5%, max 99.9%, Figure S2). While the annual abundance of neonicotinoid-treated crops is important, temporal abundance (frequency of exposure) of these crops also matters for resistance selection. Across all sites, the average cotton and soybean abundance over 6 years was 12.7 ± 12.2 thousand ha and 24.9 ± 24.3 thousand ha (mean ± SD), respectively. Significant variation in the total area of hostcrop species grown on annual cropland and their proportional abundance on that land suggests that areas of strong landscapelevel resistance selection were present across F. f usca population collections (Figure S1). Our bioassays tested each population’s response to imidacloprid and thiamethoxam at an equivalent dose of 360 mg formulated insecticide L−1 sucrose diet. Populations with greater than 12% of individuals surviving were considered suspect for neonicotinoid resistance.44 The North Carolina State laboratory F. f usca colony was used to generate baseline susceptibility information. Average survival of the laboratory population was 2.0% for imidacloprid (±1.6 SD, min. 0, max 4.8) and 1.7% for thiamethoxam (±1.9, min. 0, max 5.3) across 16 independent assays among three years. Across all fieldcollected populations, average bioassay survival was 16.5% for imidacloprid (±12.5 SD, min. 0, max 60.9), and 18.9% for thiamethoxam (±15.4, min. 0, max 71.2). Imidacloprid and thiamethoxam survival was positively correlated for populations 2317
DOI: 10.1021/acs.est.7b06015 Environ. Sci. Technol. 2018, 52, 2314−2322
Article
Environmental Science & Technology
Table 1. ANOVA and Coefficient of Determination for Each of Three Models of Log-Odds of Bioassay Survival Averaged within Each of 40 Populated Hexagonal Neighborhoods from Those Shown in Figure 3 Crop(s) included in WHY calculation cotton soybean cotton, soybean annual crops excluding cotton and soybean
variable WHY year WHY year WHY year WHY year
within year within year within year within year
DF
type III SS
F value
Pr > F
model R2
3 2 3 2 3 2 3 2
42.429 10.836 49.825 10.788 80.066 10.419 14.20 7.43
3.66 1.40 4.55 1.48 9.67 1.89 1.01 0.79
0.0219 0.2602 0.0087 0.2424
