Introduction

Declines in insect populations have received significant attention (Wepprich et al. 2019; Wagner 2020; Sánchez-Bayo and Wyckhuys 2021), but deciphering their scope and causal factors is complicated (Didham et al. 2020). Such declines appear to be especially pronounced in remnant grassland ecosystems, particularly in areas of high agricultural intensification (Swengel and Swengel 2015; Habel et al. 2019; Seibold et al. 2019; Raven and Wagner 2021). The increasingly fractured remnants of North America’s endangered tallgrass prairies (Samson and Knopf 1996; Ricketts et al. 1999; Lark et al. 2019) are under stressors from many factors, including invasive species, incompatible management schemes, isolation, and climate change. However, perhaps more than any other factor, non-target exposure to agricultural pesticides has been suggested as a key driver in the declines of grassland dependent species (Mineau et al. 2005; Gibbs et al. 2009; Gibbons et al. 2015; Forister et al. 2016; Sánchez-Bayo and Wyckhuys 2019).

Two historically widespread butterflies endemic to central North American prairies, the Poweshiek skipperling (Oarisma poweshiek Parker: Hesperiidae) and Dakota skipper (Hesperia dacotae Skinner: Hesperiidae), have declined dramatically in recent decades (Royer and Marrone 1992a, b; Swengel et al. 2011). The Poweshiek skipperling, extirpated from > 99% of its historic known locations (Belitz et al. 2018), is listed as Endangered in the United States (US Fish and Wildlife Service 2014) and Canada (COSEWIC 2014a) and Critically Endangered on the IUCN Red List (Royer 2020), ranking among the most imperiled species in the world. Similarly, the Dakota skipper has been extirpated from at least 76% of all historic locations, and is now categorized as Threatened in the United States (US Fish and Wildlife Service 2014), Endangered in Canada (COSEWIC 2014b), and globally Endangered by the IUCN (Royer 2019).

The widespread extirpation of these and other prairie dependent butterflies coincides with large-scale changes in applications of several agricultural insecticides in the north-central United States during the mid-2000s. Of particular note is the essentially concurrent (1) introduction of the new class of neonicotinoid insecticides, which are now nearly universally applied as a seed coat to corn, and (2) invasion of the economically damaging soybean aphid (Aphis glycines Matsumura) in 2000 (Ragsdale et al. 2011) against which the broad spectrum organophosphate (particularly chlorpyrifos) and pyrethroid insecticides became the primary means of control in the early 2000s. These two classes of pesticides produce different exposure routes. Neonicotinoids are generally applied proactively as systemic seed coatings and may be transported via dust during planting early in the growing season (Tapparo et al. 2012) onto non-target plants or incorporated into soils or waters (Main et al. 2014, 2015; Hladik et al. 2014; Jones et al. 2014; Bonmatin et al. 2015; Williams and Sweetman 2019). Neonicotinoids are now the world’s most widely applied class of insecticides and have been widely posited as a primary driver of declines of butterflies, other pollinators, and indeed broad swaths of wildlife (Mason et al. 2013; Goulson 2013; Fairbrother et al. 2014; Gibbons et al. 2015; Pecenka and Lundgren 2015; Gilburn et al. 2015; Forister et al. 2016; Basley and Goulson 2018; Olaya-Arenas and Kaplan 2019). Conversely, organophosphates and pyrethroids are contact insecticides that are typically applied reactively later in the summer in response to pest pressures via aerial spraying and thus may be blown or absorbed into rain (Foreman et al. 2000; Mackay et al. 2014) that falls onto non-target areas.

Given the correlated declines of imperiled prairie butterflies and changes in pesticide applications, we document the presence and quantities of pesticides and their residues on putative larval host grasses of Dakota skipper and Poweshiek skipperling from prairie remnants formally designated as Critical Habitat for them in Minnesota and South Dakota between 2014 and 2020. We sought to assess how their composition and quantities varied across these sites and across years, if their occurrence and quantities would be lower within prairie interiors than along associated agricultural margins, how pesticide prevalence related to population and extirpation status, and how this composition might affect conservation practices.

Methods

Sample collections

We collected samples from four prairie remnants (“Sites”) in Minnesota (“Clay”, “Pope”, “Lincoln”, and “Pipestone”) and one in South Dakota (“Day”) (Fig. 1). All of the Minnesota sites are designated as Critical Habitat for both Poweshiek skipperling and Dakota skipper (US Fish and Wildlife Service 2015). Dakota skipper remains extant at two of the sites (Day and Clay) but is extirpated from the remaining three (Pope, Pipestone, and Lincoln). Poweshiek skipperling historically occurred at all sites but is now extirpated from the region. These sites are also home to other species of conservation concern, including regal fritillary (Argynnis idalia Drury: Nymphalidae). The exact site names and specific GPS locations of sample points are censored here due to the presence of federally protected species at some sites.

Fig. 1
figure 1

Locations of prairies in Minnesota and South Dakota from which samples were collected for pesticides analysis

From 2014 to 2020, we collected samples from between two to nine points per prairie site. Using aerial imagery in a GIS, we partitioned each prairie into 10 × 10 m grid cells, and classified cell Locations as either “edge” (within 10 m of an agricultural field) or “interior” (≥ 100 m from any edge). We then used a random number generator to select interior and edge cell coordinates for sampling. Given our interest in understanding the potential exposure to federally protected butterflies and the high-quality prairies they depend upon, we eliminated randomly selected interior locations that did not comprise suitable habitat. Fewer edge samples were collected in some years due to budgetary constraints.

Within each selected cell, we clipped > 5 g from a single, randomly selected little bluestem (Schizachyrium scoparium (Michx.) Nash: Poaceae) or clipped from a cluster of immediately adjacent little bluestems if biomass on the initially selected plant was low. If little bluestem was not available in sufficient quantities within a grid cell, we instead sampled from big bluestem (Andropogon gerardii Vitman). These grasses are indicative of intact native prairies and are among the host plants for Poweshiek skipperling, Dakota skipper, and other imperiled prairie skippers (Dana 1991; Royer et al. 2008; Rigney 2013; Seidle et al. 2018; Nordmeyer et al. 2021; Henault and Westwood 2022). We also collected paired > 25 g of soil beneath subsets of grass samples in 2014, 2015, and 2016, sieving the soil through a 1.27 mm mesh to homogenize particulate sizes for analysis. We cleaned collection supplies (scissors, collection bowl, hand shovel, mesh sieve) with 100% acetone in the field between samples. Clean nitrile gloves were changed between each sample collection. All samples were double-bagged in 1 quart plastic zip-loc bags, immediately placed on dry ice in the field, and then transferred to a -20oC freezer at the Minnesota Zoo until overnight shipment to the laboratory for analysis.

Sampling occurred in two seasons: (1) shortly after the planting season in early June 2015 and mid-May 2016, and (2) late summer from mid-August to early September in 2014 through 2020. Samples were generally collected on one date within each season for each site. These sampling periods were selected to coincide with the likelihood for the applications of different classes of pesticides. Of particular interest were the potential occurrence of neonicotinoid insecticides early in the season following seed-coated crop plantings, and the later summer aerial applications of broad-spectrum insecticides targeting soybean aphid and other agricultural pests. Due to high per sample testing cost and minimal detections in early season samples (see below), we elected not to continue early season sampling after 2016. Sites generally were sampled without information on when or where spraying may have occurred relative to sampling. However, we had the opportunity for a single time series sample collection when an airplane-applied spray event was observed adjacent to the Pipestone site within an hour of the cessation of the collection of the first set of samples in 2014. A second set of samples was then collected the following day from many of the same cells to quantify “Before” and “After” changes in pesticides levels.

Pesticides screening

Samples were screened for pesticide residues at the U.S. Department of Agriculture Agricultural Marketing Service’s National Sciences Laboratories (“NSL”; Gastonia, NC) using the QuEChERS extraction method (AOAC 2007.01; Anastassiades et al. 2003). The NSL performed all extractions from the provided substrates (either grass or soil) and then conducted Liquid Chromatography tandem mass spectrometry to determine the quantities of pesticides and some of their residues in the samples in reference to certified analyte standards for each compound. The NSL conducted all independent quality assurance and quality control standards and re-analyzed some samples as needed to satisfy control standards.

The identities, quantities, and minimum Levels of Detection (LOD) of a total of 233 pesticides and their residues were reported for each sample, with some variation in the number of analytes between years (2014 and 2015: 174; 2016: 178; 2017 and 2018: 199; 2019 and 2020: 193) (Appendix 1). LODs varied across compounds but were internally consistent for all samples within a year. Individual compounds within a sample were reported as either (1) “Not Detected” (i.e., below their LOD), (2) “Trace” (above their LOD but lower than a level that could be quantified), or (3) quantified on a parts per billion (ppb) basis (i.e. ng/g of homogenized sample material). We treated chemicals categorized as “Not Detected” as zeroes and considered a compound to be “Present” in a sample if it was either reported as “Trace” or if amounts could be specifically quantified. Extraction protocols and some instrumentation at the NSL were upgraded between the 2016 and 2017 samples, resulting in generally lower LODs and some additional analyzed pesticides, so 2017–2020 data are likely to be more quantitatively robust. We prepared and submitted blind replicates of four samples in 2014 (Appendix 2: Su8, Su10, Su135, Su127) to verify consistency; all replicate samples varied in observed quantities only slightly.

Temporal and spatial variation

We summarize descriptive information about variation in pesticide occurrence, composition, and quantities between seasons (early vs. late) and substrates (grass vs. soil). We also tested how the number of insecticide types and the quantities of some of these insecticides varied in late season grass samples across sites and years, and if the high-quality habitat interior locations of those prairie sites had lower insecticide occurrences and quantities than locations along prairie-agricultural edges. We excluded herbicides and fungicides from these statistical tests given (1) the relative rarity of herbicides in late season samples, and (2) the technical differences in fungicide detection between the 2014–2016 and 2017–2020 samples.

We tested the ability of the fixed effects of sites, years, and location nested within site, as well as the interaction between site and year, to explain variation in the number of detected insecticides on grasses in late summer with an Analysis of Deviance specifying a Poisson distribution in R using the glm package (R Core Team 2023; Bates et al. 2015). We similarly tested how the quantities (ppb) of chlorpyrifos and cyhalothrin on late season grasses differed with these same main effects and interactions with lm Analysis of Variance in R. Other compounds were too rarely documented to facilitate analyses. For these quantity analyses, we included data from after the known 2014 Pipestone spray event (test significance did not change if the pre-spray data were instead included). Since LOD is always less than or equal to the Level of Quantification (LOQ), we estimated the ppb for trace samples to be equal to the associated LOD. We truncated the dataset for the chlorpyrifos and cyhalothrin quantity tests to include only those points where these insecticides were detected to avoid overdispersion. The remaining quantities were log10 transformed for analysis to improve normality. We also excluded from the chlorpyrifos quantity analysis a single sample (Su86; 2290 ppb, Appendix 2) from Pope in 2017 that was highly leveraged even following log transformation; significance values were similar with and without this datapoint. We tested for statistically significant groups between the main effects of site and year for all analyses using post hoc Tukey tests.

Results

We detected eight insecticides, three herbicides, 10 fungicides, and three other compounds from 226 samples, across all years, seasons, sites, and substrates. All data are presented in Appendices 2 (late season grass samples), 3 (early season grass samples), and 4 (paired soil samples), and summarized graphically in Fig. 2. Pesticides were prevalent on grasses, particularly in late season, but were rare in the paired soil samples (only 5 of 46 samples). Soil samples were significantly less likely to have at least one detectable pesticide than their paired grass samples (χ21,92 = 21.45, p < .0001). This was true for both early season (χ21,60 = 9.93, p = .0016) and late season (χ21,32 = 12.70, p = .0004) sampling periods. Four pesticides (chlorpyrifos, clothianidin, atrazine, and tebuconazole) were detected in the soil samples. Notably, these single samples of clothianidin and tebuconazole are the only detections of these pesticides in the entire dataset.

Fig. 2
figure 2

Maximum observed quantities (Top) and the percent of samples within a site by year and season with positive detections (Bottom) of 24 pesticides and other compounds observed in grass and soil samples at five prairies from 2014 through 2020

Pesticides were prevalent on grasses in late season but rare in the early season. Spanning seven insecticides, three herbicides, and nine fungicides, 121 of 142 samples of the August and September grass samples from 2014 to 2020 contained at least one pesticide. In contrast, only a single pesticide (the herbicide atrazine) was detected grass samples in 21 of 38 samples during the two early season periods in May 2015 and June 2016.

Late season grass insecticides

Chlorpyrifos was by far the most frequently detected pesticide in the entire dataset, in 108 of the 142 (76.1%) late season grass samples. This organophosphate insecticide was detected in every year of the study and in every sample from 2018 to 2020. Most chlorpyrifos quantities ranged between 1 and 50 ppb (78 of the 88 quantifiable samples; global average: 49.9 ppb, median: 11.6), but a single sample contained of 2290 ppb (Appendix 2). Two other insecticides, cyhalothrin and bifenthrin, were each present in about a quarter of all late season samples (27.5% and 23.9%, respectively). The four other late summer insecticides (carbofuran, cypermethrin, diflubenzuron, esfenvalerate) were generally rare. Notably, all usage tolerances of carbofuran were revoked in the United States in 2009 (U.S. Environmental Protection Agency 2009), so the four observations at Pope in 2017 represent a prohibited application(s).

Late season fungicides

At least one of the 10 fungicides was detected in 64 of the 142 late season grass samples. Azoxystrobin was the most frequent, in 46 samples. Notably, however, all fungicide-positive samples were collected from 2017 to 2020 (n = 75), while fungicides were not detected in any of the late season samples (n = 67) from 2014 to 2016. As noted above, this temporal variation is likely due to upgrades in extraction and analysis protocols at the NSL. Thus, the fungicide data from the 2017–2020 samples are more likely representative of annual deposition patterns than the 2014–2016 data.

Late season herbicides

Herbicides were rare on grasses in late summer, with generally only Trace detections of atrazine, hexazinone, and metolachor. Most summer herbicide detections were in 2019, which was an extremely wet year that likely prompted additional or later herbicide applications within and adjacent to agricultural fields.

Other compounds

Three compounds were detected for which there are no known local agricultural applications: diphenylamine, thymol, and DEET. Diphenylamine was nearly ubiquitous in Trace or low quantities (1–22 ppb) in 2017 and 2020, but there is no apparent regional agricultural usage, and no diphenylamine was sold in Minnesota in any year for which there is documentation (2018–2020) (Minnesota Department of Agriculture https://www.mda.state.mn.us/agricultural-pesticide-sales-use-reports-statewide/). Thymol was found in a single sample at Pope in 2017. Typically used as a miticide within European honey bee (Aphis mellifera Linneaus: Apidae) hives and as an animal repellent, thymol is also a naturally occurring compound in some native plants (such as Monarda fistulosa Linnaeus: Lamiaceae). DEET was detected in three samples. Applied to clothing or skin as an insect deterrent, this compound was never used by the researchers that collected our samples in the field, so the detections are likely the result of other people who had applied DEET and coincidentally walked through our sampling locations.

Before and after a known spray

Immediately following completion of sampling at Pipestone on August 19, 2014, an airplane was observed spraying a soybean field ~ 350 m northwest of the nearest previously sampled point. Light northwest prevailing winds during and immediately following the late afternoon spray at the nearest weather station (Pipestone, MN; ~19 km SW of the Pipestone site; https://mesonet.agron.iastate.edu/) would be expected to blow aerosolized droplets containing pesticides from the spray in the direction of the Pipestone site and the previously sampled points (Su1-Su7, Appendix 2). Before and after the spray, only the three primary insecticides applied against soybean aphids (chlorpyrifos, bifenthrin, and cyhalothrin) were detected across the Pipestone site. Chlorpyrifos and cyhalothrin quantities always rose after the spray (Su8-Su12, Appendix 2). Mean chlorpyrifos quantities rose from 19.3 ppb (range 9.5–21.1) to 135.7 ppb (range 51.9–278.0). Similarly, mean cyhalothrin quantities rose from 4.0 ppb (range 2.7–7.2) to 24.8 ppb (range 8.4–83.3). In contrast, bifenthrin mean quantities fell from 31.3 ppb (range 8.3–74.5) to 24.8 ppb (range 9.5–68.9) the day after the spray. It therefore seems likely that the aerial spray was a mix of chlorpyrifos and cyhalothrin. The presence of bifenthrin in the samples was likely the result of a previous spray event or events.

Temporal and spatial variation

Insecticides were detected in late season at all sites, but the number of insecticide types varied across sites and years (Fig. 2). Site and Year interacted significantly, largely driven by increases in the number of observed insecticides at Day and Pope between 2014 and 2015 with contrasting concurrent declines at Pipestone and Clay (Appendix 5 A). The number of insecticides observed increased across all sites between 2016 and 2019 and then decreased 2020. Averaged across years, fewer insecticide types were found at Clay than at Pipestone and Day (Fig. 3A). Notably, the last known extant Dakota skipper population in Minnesota is at Clay. Fewer pesticides were detected in 2016 than in 2019, averaged across all sites (Fig. 3B). The number of insecticides detected along prairie-agricultural edges did not differ from the number detected within prairie interior locations, as indicated by a non-significant interaction between Site and nested Location (Table 1).

Fig. 3
figure 3

Mean number of insecticides (± 95% CI) detected in late season samples across sites (A) and across years (B). Letters represent significant post hoc contrasts at p = .05

Table 1 Analysis of Deviance of the fixed effects of site, year, and location (i.e. Edge vs. Interior) within a prairie on the number of detected insecticides (left), and Analyses of Variance of the quantities of chlorpyrifos (center) and cyhalothrin (right) on grasses from five prairies in late summer 2014–2020

Similar to the pattern with the number of insecticide types noted above, Site and Year interacted significantly for both chlorpyrifos and cyhalothrin quantities, but in a manner that still allows for some interpretation of main effects (Table 1, Appendix 5B and 5 C). Crop types, pest loads, climatic effects, and associated pesticide application rates are not expected to be uniform within years across the large geographic extent of this study, and overall patterns of average exposure at a site across years are important. In terms of main effects, chlorpyrifos quantities varied significantly across both sites and years when present, while cyhalothrin quantities only varied across years (Fig. 4). Quantities of both compounds did not differ between prairie interiors and agricultural edge locations (as evidenced by non-significant interactions between Site and the nested Location for each compound). It is notable that in a post hoc Tukey contrast, chlorpyrifos quantities were significantly lower at the two sites that retain extant Dakota skipper populations (Day and Clay) than at those where Dakota skippers have been recently extirpated (Pope, Lincoln, and Pipestone).

Fig. 4
figure 4

Quantities (log mean parts per billion ± 95% CI) of the insecticides chlorpyrifos (A and B) and cyhalothrin (C and D) across sites and years from five prairies in Minnesota and South Dakota. Statistically significant post hoc groupings following Tukey test comparisons are indicated with letters

Discussion

To our knowledge, this study represents the longest time series monitoring of pesticide occurrence across the upland portions of multiple prairie remnants that are of significant conservation interest. Most non-target prairie deposition studies have focused on wetland systems (ex: Donald et al. 1999; Messing et al. 2011; Belden et al. 2012; Main et al. 2015; Mimbs et al. 2016; McMurry et al. 2016; Evelsizer and Skopec 2018). Relatively few terrestrial pesticide occurrence studies have targeted specific compounds or situations (Main et al. 2020; Goebel et al. 2022; Zioga et al. 2023), with particular focus in North America on potential risks to monarchs and contamination of their host milkweeds (Pecenka and Lundgren 2015; Halsch et al. 2020; Krishnan et al. 2020; Hall et al. 2022; Grant et al. 2022).

Contrary to frequently cited concerns, we did not find a significant exposure signal from neonicotinoid insecticides, only detected in the soil of one of our spring samples, and none later in the summer. This does not indicate that neonicotinoids may not pose a seasonal risk or that they are not a risk in other prairie strata, but the exposure potential appears to be lower for the grasses and the soils in the hilly upland gravel prairies that we studied. In contrast, broad-spectrum insecticides, particularly organophosphates (chlorpyrifos) and pyrethroids (primarily cyhalothrin and bifenthrin) likely targeting soybean aphids in the second half of summer, were prevalent in their upland grass larval hosts. The landscape of soybean aphid insecticide applications is changing though due to (1) evolving resistance by soybean aphids to pyrethroids like cyhalothrin and bifenthrin with decreases in control effectiveness noted in Minnesota beginning in 2014 (Hanson et al. 2017; Koch et al. 2018; Menger et al. 2022) and (2) the revocation of all agricultural usage authorizations for chlorpyrifos in the United States in 2022 (U.S. Environmental Protection Agency 2022). This combination of pyrethroid resistance and chlorpyrifos regulation may have at least a short-term effect of reducing insecticide exposure to non-target organisms in these prairies.

Conservation implications

The rapid decline and complete extirpation of Poweshiek skipperling from the vast majority of its historic range across Minnesota, Iowa, North Dakota, and South Dakota was concurrent with changes in pesticide usage and the invasion of the soybean aphid. Sympatric Dakota skipper populations declined, and many also disappeared, shortly thereafter. These extirpation events occurred before this pesticide sampling began, so determining the exact linkage between the pesticides that we have observed, and the butterfly declines is necessarily speculative. However, it is highly noteworthy that the sites where Dakota skipper populations remain extant today had both fewer types of insecticide on average, as well as lower quantities of chlorpyrifos, than other sites where Dakota skippers have been extirpated.

Poweshiek skipperling and Dakota skipper are univoltine with the adult flight in late June and early July. Eggs are laid in July, and 1st through 3rd instar larvae of both species feed through the end of summer until winter hibernation on a wide range of prairie graminoids, including the little bluestem primarily sampled in this study. Thus, both species are still foraging as small larvae on their grass hosts during the late season aerial spraying of broad-spectrum insecticides that we have documented. They resume feeding in spring (when they would theoretically be exposed to early season pesticides) and pupate in June. Poweshiek skipperling and Dakota skipper may be at increased exposure risk relative to regal fritillary which, despite being another historically co-occurring prairie endemic of conservation concern, is still rather common at the same sites that we sampled but whose larvae do not feed until spring and may be thus more buffered from late season insecticide applications. Poweshiek skipperling may be even more vulnerable to environmental stressors, like pesticides exposure, than Dakota skipper and many other grass-feeding skippers because Poweshiek skipperling larvae do not construct shelters and remain higher on their host grass. While correlative, this larval foraging difference may therefore have contributed to the steeper and more dramatic disappearance of the once more common Poweshiek skipperling from the same prairies where Dakota skippers and other prairie butterflies remain.

These retrospective views of potential causes and effects can be cautiously useful, but of greater interest is understanding the current landscape of risk to advance conservation goals. Rare butterflies are not necessarily more likely to persist in protected sites than in unprotected areas (Warren 1993; Schlicht et al. 2009). While the formal U.S. designations of the sites from which we sampled as Critical Habitat for Poweshiek skipperling and Dakota skipper do afford increased oversight protecting key features necessary for their conservation, that oversight only relates to activities that involve a federal permit, license, or funding. Critical Habitat is a tool to guide federal agency actions to fulfill their responsibilities to protect endangered species and the ecological resources upon which they depend, but the activities of private landowners are not affected if there is no federal funding or authorization “nexus”. The results of this study suggest that some degree of non-target pesticide exposure in these Critical Habitats are common, even in their interiors, and should probably be expected in prairie remnants across much of the historic known range of these protected butterflies. Efforts should be made to minimize risks when possible, basing them on the best available evidence to develop protection and recovery plans (e.g., Montgomery et al. 2009).

Unlike more vagile species (Ries and Debinski 2001), Poweshiek skipperling and Dakota skipper are unlikely to colonize significantly dispersed prairie remnants. Lack of connectivity among remnants presents significant risks to low-dispersing prairie dependent species (Leach and Givnish 1996; Attwood et al. 2008; Koper et al. 2010; Nowicki et al. 2014; Wimberly et al. 2018; Crone and Schultz 2019). The U.S. Fish and Wildlife Service identifies that the re-establishment of dozens of “healthy” populations for each species via reintroductions is needed to satisfy downlisting criteria (U.S. Fish and Wildlife Service 2021, 2022), a foundational effort that the Minnesota Zoo began initiating in 2017. A healthy population is formally defined as “one that is demographically, genetically, and physically robust and occupies large areas of high-quality remnant prairie habitat” (U.S. Fish and Wildlife Service 2022). To be “robust”, a population must be “comprised of individuals with good body condition and with pesticide and pathogen loads that are below levels that could cause meaningful loss of reproductive capacity”. Thus, the risks associated with pesticide exposure must therefore be one of the formal factors considered when assessing locations for potential reintroductions (e.g., Delphey et al. 2017). The persistence of these low dispersal species will depend on the maintenance of large, high-quality prairies, for which pesticide load is an important metric.

Remaining questions

Gains have been made in our understanding of the prevalence and occurrence of pesticides in these remnant prairies, but three substantial questions remain about the landscape of risk at both small and large scales. First, how closely do the observed pesticide quantities and composition match their peak values at these sites? The data presented here are likely underestimates the extent of maximum exposures. With the exception of a single date, we do not know the specific timing or locations of the pesticide applications in the vicinity of the sampled prairie remnants (unlike say, Goebel et al. 2022). Pesticides like chlorpyrifos decay under natural conditions but may persist at low levels in the environment weeks after application (Kamrin 1997; Christensen et al. 2009; Das et al. 2020). Pesticide deposition is also expected to decline with distance from application location (Teske et al. 2002; Goebel et al. 2022), but chlorpyrifos and many other pesticides also are known to be transported long distances once volatilized through air (Harnly et al. 2005; Felsot et al. 2011; Giesy et al. 2014; Mackay et al. 2014). Thus, the observed quantities likely represent fractional remnants of the original amounts following initial deposition, and are best thought of as an estimate of the typical distributions of ppb on the landscape in late summer. We also do not fully understand the extent of pesticide composition and occurrence in mid-summer, between our early and late sampling periods. Pesticide applications are probably less likely to occur in other portions of the growing season, particularly during late June and July when Poweshiek skipperling and Dakota skipper are adults, but we cannot be certain that we are not missing an exposure window.

Second, what is the actual risk posed by the observed pesticide composition and their quantities? Risk is the combination of exposure and toxicity. We have documented widespread exposure to many pesticides, but whether or not their observed quantities generate mortality and/or sub-lethal effects to skippers and other prairie insects is central. There are few suitable studies with surrogate species for non-model systems (like the imperiled grass-feeding skippers) against which we can make useful comparisons on the effects of pesticides. What does exist often varies in method, exposure route (oral, contact, etc.), and/or metrics (ng/bee, kg/ha, or % of active ingredient, etc.) (US Environmental Protection Agency 1999). For chlorpyrifos, Krishnan et al. (2020) report laboratory contact LD50 toxicity of 79 µg/g (= 79,000 ppb) for first instar monarch larvae (Danaus plexippus Linneaus: Nymphalidae), and a dietary LD50 of 9.9–33 µg/g (= 9,900–33,000 ppb) of leaf material. These values are well above those observed in this study, but Krishnan et al. (2020) still observed significant mortality in companion standard field spray assays following similar aerial application protocols to what would be expected adjacent to the prairie remnants in this study. The reported contact LD50 for chlorpyrifos for honey bee is much lower though (0.059 µg/bee = 59 ppb; Arena and Sgolastra 2014), a value that was exceeded in seven samples in this dataset. Similarly, a reported honey bee contact LD50 for bifenthrin (0.01462 µg/bee = 14.6 ppb; Tomlin 2000) was exceeded 13 times in our dataset.

Experiments testing responses to a wide range of the pesticides observed are needed across multiple related species, particularly given the imperiled state of Dakota skipper and Poweshiek skipperling. Multiple forms of exposure studies, ranging from direct contact studies under artificial conditions that remove as many variables as possible to studies that capture as much natural history as possible, would provide useful inferences of the “real world” risks. Pesticides exposure studies must test both lethal and non-lethal responses to individual pesticides but also responses to the synergistic effects of exposure to multiple pesticides. It cannot be expected that each of the nearly two dozen pesticides observed in this study will individually affect prairie skippers, arthropods, and other wildlife in a vacuum. For example, Hladik et al. (2016) recorded nineteen pesticides and degradates on wild native bees. Indeed, compounding and non-additive effects of simultaneous exposures to multiple pesticides should be expected at both individual species and species assemblage levels (Barmentlo et al. 2018). As noted by Schulz et al. (2021), the total applied toxicity of pesticides to pollinators has risen in recent decades, a feature largely driven by increases in the toxicity of pesticide classes like pyrethroids and neonicotinoids and not necessarily the total mass of pesticides applied.

Finally, what landscape factors are driving this non-target deposition, and are these sampled prairies representative of others? How well do adjacent land-use patterns predict pesticide composition and quantities into prairie remnants? Are these sites unusual? A deeper exploration of site-specific and regional influences, including crop rotations, pest cycles, weather, topography, and other factors that shape deposition is warranted, within and across both years and sites. A more exhaustive regional inventory of pesticide exposures at additional prairie remnants coupled with predictive analyses of the factors that drive non-target exposure would be beneficial for conservation and risk minimization efforts, for Poweshiek skipperling, for Dakota skipper, and beyond.