Landscape-associated pesticide contamination shapes pollen provision microbiomes in Osmia cornifrons (Hymenoptera: Megachilidae)

Abstract

Bees live and forage across a landscape increasingly altered by human activity, collecting pollen, nectar, and an essential microbiome from the local environment as well as chemical contamination from surrounding human activities. The increasing frequency and intensity of antimicrobial chemical use in agricultural systems disrupts the environmental microbial community and has been linked to bee declines. However, it remains largely unknown how pesticides and antimicrobial agents vary across the landscape, and in turn, cascade to alter microbiome assembly in solitary bees. To evaluate landscape-level effects of pesticide contamination on the pollen microbiome of solitary bees, we surveyed Osmia cornifrons (Hymenoptera: Megachilidae Radoszkowski, 1887) pollen provisions across 3 contrasting land use types in central upstate New York: forest, urban, and orchard habitats. Pollen provisions were screened for 94 common pesticides and sequenced to characterize their bacterial community. We found pesticide concentrations and risk varied significantly across land use types and was highest at orchard sites. Microbial communities also differed across landscapes, with increased surrounding forest cover corresponding to higher abundances of Apilactobacillus, a beneficial bee-associated genus in the family Lactobacilliacea. We found a significant correlation between increased pesticide concentration and decreased pollen-associated microbial diversity. These results indicate links between land use, pesticide exposure, and the microbiome of solitary bees.

Introduction

Bees exist in a world shaped and altered by humans. Ninety-five percent of the world’s terrestrial land has been disturbed or modified by humans (Kennedy et al. 2019). In the United States, 52% of terrestrial land is used for agricultural purposes and a further 3% has been urbanized (Winters-Michaud et al. 2024). These landscape modifications can alter the resources collected by bees (Peters et al. 2022). During foraging trips, bees collect not only pollen and nectar from plants, but also a suite of microbes and contaminants, such as pesticides, that exist in the local environment. In solitary bees, these resources and contaminants are incorporated into a pollen provision, a stored food mass upon which an egg is laid before the brood cell is sealed. This pollen provision supplies all the nutrition required for the complete development of the bee larvae, as well as being the source of the larval microbiome, thereby strongly influencing larval developmental success (Danforth et al. 2019).

Pollen provision microbial communities commonly include a range of microbes from beneficials to saprophytes to pathogens (Steffan et al. 2024). The pollen microbiome of some solitary bees has been shown to confer beneficial effects, with increased larval mortality observed when pollen provisions are sterilized and their microbial communities removed (Dharampal et al. 2020). While the microbial community present in the pollen provision of solitary bees is largely shaped from the surrounding flower community (McFrederick et al. 2017), some bee species demonstrate a consistent core microbiome of bacterial genera in their pollen provisions across large geographic ranges and diverse pollen diets, suggesting a degree of microbial control exerted by the mother bee (Dew et al. 2020). For many bee species, putatively beneficial acidogenic microbes, typically in the family Lactobacillaceae, may help stabilize provisions throughout larval development and prevent dysbiosis (McFrederick et al. 2012, Hammer et al. 2023).

Bees’ pesticide exposure is highly dependent on land use surrounding a bee’s nesting site. While much research has addressed exposure and impacts of pesticides on social honeybees and bumblebees, comparatively little is known regarding pesticide exposure and impacts on solitary bees (Sgolastra et al. 2019, Raine and Rundlöf 2024). Solitary bees nesting in apple orchards often have high pesticide contamination in pollen provisions, which, in addition to direct toxicity, also have sublethal impacts on bees (Centrella et al. 2020). The widespread use of fungicides and bactericides to control plant pathogens may indirectly impair larval development by shifting local environmental microbial communities and disrupting the pollen provision microbiome, with work showing that increased fungicide use can increase pathogen prevalence in bee provisions (Krichilsky et al. 2021). This indirect impact of antimicrobials may be especially pronounced in apple orchards in the northeastern United States which undergo weekly fungicide and antibiotic sprays during bloom to control for diseases (Cooley et al. 2013), while also requiring bees for crop pollination and fruit set (Wallis et al. 2025). These spray mixes will often include streptomycin to control for the bacterial disease fireblight (Erwinia amylovora) usually mixed with various fungicides to control for apple scab and mildews (Wallis et al. 2025). While the use of antimicrobials in agriculture has been shown to impact environmental microbial communities (Batuman et al. 2024, Zhang et al. 2024, Shi et al. 2025), it remains largely unknown how the use of antimicrobial agents influence the pollen provision microbiome formation of bees foraging in human impacted landscapes.

To evaluate landscape-level effects on solitary bees, we surveyed Osmia cornifrons (Hymenoptera: Megachilidae Radoszkowski, 1887) pollen provision across 3 contrasting land use types in central upstate New York: forest, orchard, and urban habitats. Osmia cornifrons is a generalist univoltine spring bee and is the most common managed solitary bee in New York state. It occurs in both managed and introduced wild populations across the landscape. It is an effective, agriculturally important pollinator of Rosaceae crops such as apples, and is introduced by growers to increase apple pollination as its emergence coincides with apple bloom (Matsumoto et al. 2009, Vaudo et al. 2020). We screened provisions for 94 pesticides, including insecticides, herbicides, synergists, and antimicrobials, and sequenced the bacterial and fungal communities to address the following 3 questions: (i) How do pollen provision microbiomes vary among forest, orchard, and sites? (ii) How does pollen provision pesticide contamination vary among urban, forest, and orchard sites? (iii) Does landscape-driven pesticide contamination impact the pollen provision microbiome?

Methods

We sampled 3 land-use types (forest, urban, agricultural), with 5 replicate sites per category. Forest sites were in the 11,000 acre Connecticut Hill Wildlife Management Area; urban sites were residential yards in the city of Ithaca, New York; agricultural sites were apple orchards in the local Ithaca, New York region. Connecticut Hill is actively managed to control for invasive species and promote habitat, however, no herbicides or pesticides are used during management (Cunningham et al. 2017). All sites were within a 20 km radius of one another and located in the finger lakes region of New York state in the northeastern United States. All sites were at least 1 km apart from one another to insure the independence of sites as the typical foraging radius of a provisioning adult Osmia is estimated to be approximately 500 m (Cohen et al. 2020). In early spring before snow melt, we placed 1 Osmia nesting box at each site. Each nesting box consisted of a covered wooden box containing a metal can filled with empty 8-mm cardboard Osmia nesting tubes (Crown Bees, Woodinville, Washington, United States), within which, Osmia females lay a series of eggs on collected pollen provisions before sealing each in an individual chamber and ultimately sealing the nesting tube. This allows for the easy detection of completed nest tubes and the removal and collection of Osmia eggs and associated pollen provisions. Nesting boxes were mounted to trees or metal stakes, covered in 1/4 inch (0.635 cm) metal mesh to protect Osmia nests from predation. We seeded each box with 18 nesting tubes, each filled with approximately 8 healthy O. cornifrons cocoons. Osmia preferentially nest from where they emerged and seeding cocoons ensured ample nesting at our study sites (Mader et al. 2010). Cocoons had been collected in Ithaca, New York, the previous spring and overwintered in a protected shed. We X-rayed all nesting tubes to screen for bee health (Mader et al. 2010).

We checked nest boxes routinely and collected provisions at 2 time points: an early time point when the first nest tubes were finished and the entrance hole sealed, between 3 and 16 May 2023, and a late time point, when the last nest tubes were finished at the end of the nesting season, between 19 and 31 May 2023 (Supplementary Table S1). The early and late time points occurred on different dates across sites due to elevational and local climatic differences and represent the 2 extremes of the nesting season with most nesting activity occurring between the 2. For each site and time point, we haphazardly collected 8 pollen provisions across nesting tubes. Half of each provision was used for bacterial microbiome analysis and half for fungal. Due to a freezer failure, the majority of fungal samples were lost, leaving only 1 site within each land use type. Separate from microbiome provisions, at each site and time point, we also haphazardly collected 5 g of pollen provisions for pesticide analysis (roughly 25 provisions homogenized) and froze them at −80 °C.

Microbiome Sequencing

We homogenized individual pollen samples and extracted DNA using the Qiagen Plant Pro extraction kit (Qiagen, Germantown, Maryland, United States) followed by 16S and ITS rRNA amplicon sequencing. Library preparation and sequencing via Illumina MiSeq were performed at the University of California, Davis Genome Center DNA Technologies Core (see Supplementary Methods S1 for full sample preparation and sequencing methods).

To process sequences, we removed primers from paired FASTQ files using Cutadapt (Martin 2011) before cleaning and assigning taxonomy using the DADA2 pipeline (Callahan et al. 2016). We trimmed 16S forward and reverse reads at 175 and 145 bases, respectively, and trimmed ITS based on quality score following standard DADA2 ITS filtering parameters. We removed chimeras and assigned bacterial taxonomy using the Silva v138.1 database (Quast et al. 2013, Yilmaz et al. 2014) and fungal taxonomy using the UNITE v10 database (Abarenkov et al. 2024). We filtered out potential contaminates in the bacterial samples using Decontam (Davis et al. 2018); however, our fungal samples lacked sequence controls and were not filtered using Decontam. We compared ITS sequences from samples to previous ITS sequencing runs in the lab and found no contaminants that had been identified in previous runs. Finally, we removed mitochondrial and chloroplast sequences. One ITS sample did not sequence properly and was removed from the dataset (urban site, early time point). We dropped samples that had less than 300 reads total (3 16S samples; 2 early orchard samples, and 1 late orchard sample, which all came from the same site). We filtered ASVs that were not present in at least 2 samples and that were under 0.01% of the total relative abundance.

Pesticide Quantification

Frozen pollen provision samples were extracted by modified versions of the QuEChERS procedure (Anastassiades et al. 2003) at the Cornell Chemical Ecology Core Facility (Ithaca, New York, United States) and screened for 94 pesticides, including insecticides, herbicides, synergists, and antimicrobials, by liquid chromatography tandem mass spectrometry (LC-MS/MS) as in Graham et al. (2021) and Mueller et al. (2024). Briefly, 5 g of each sample was extracted and the supernatant injected into a Vanquish Flex UHPLC system (Dionex Softron GmbH, Germering, Germany) coupled with a TSQ Quantis mass spectrometer (Thermo Scientific, San Jose, California, United States). MS/MS detection was carried out using the selected reaction monitoring (SRM) mode. Two transitions were monitored for each compound: one for quantification and the other for confirmation (see Supplementary Methods S2 for full pesticide analysis methods and SRM parameters).

Risk Assessment

To account for the different toxicity of various pesticides, we calculated a Risk quotient, also known as a Hazard Quotient (Stoner and Eitzer 2013), or toxicity-weighted concentration (Knapp et al. 2023). The risk quotient is calculated by dividing the pesticide contamination present in each pollen sample by the respective toxicity (LD₅₀) of each pesticide, then summing for all pesticides present in a pollen sample. The risk quotient formula, however, assumes that toxicity is additive and ignores pesticide synergism which can increase toxicity (Iverson et al. 2019). We obtained LD₅₀ values from the British Crop Protection Council Pesticide Manual (MacBean 2012), the ECOTOX database of the US Environmental Protection Agency (Olker et al. 2022), and the AgriTox Database of the French government (Maniere et al. 2011). We made several assumptions to apply existing toxicity data to the understudied larval stage of O. cornifrons. Since little pesticide toxicity data exist for O. cornifrons, oral LD₅₀ values from honeybees were used (Thompson and Pamminger 2019). Furthermore, due to sparse toxicity data for larval bees, here we assume consistent LD₅₀ across life stages and apply adult LD₅₀ values to the larval stage. It should be noted however, that studies on honeybees show larvae can have differing susceptibility to pesticides from adult bees with larvae being more or less susceptible depending on the pesticide (Zhu et al. 2014, Tavares et al. 2015, Dai et al. 2017).

For downstream analysis, we assumed all exposure to pesticides occurs through feeding and therefore we only calculated risk using oral LD₅₀ values while ignoring contact exposure. In actuality, Osmia larvae are constantly in contact with a contaminated pollen provision and both oral and contact exposure would be happening simultaneously; however, currently there is no established method to assess risk from oral and contact exposure simultaneously. Lastly, we assume since the bee is consuming its whole pollen provision, there is a 100% transfer rate of the active ingredient into the bee. LD₅₀ studies typically test the impacts of pesticides over a period of 24 to 48 h of feeding. In our experiment, pesticide exposure continued for the entirety of the bees’ development and eventually led to all pesticides being consumed, over a time period of approximately 2 weeks (Lee et al. 2016). These assumptions were made to adapt the existing data and risk framework to our field collected solitary bee provisions. In light of the lack of toxicity endpoints for larval solitary bees, we assert using honeybee LD₅₀ data are valuable in allowing for a comparison between samples while accounting for the relative toxicity of different pesticides.

Landscape Analysis

We calculated the landscape composition surrounding each site in a 500-m radius around each nest box and quantified the proportion of the surrounding natural habitat using the United States Department of Agriculture CropScape data layer (United States Department of Agriculture 2026, see Supplementary Table S2 for landscape categories that were aggregated into “natural” cover). Five hundred meters is thought to be the typical flight radius of provisioning mothers and matches previous work that studied the impacts of landcover on Osmia adult microbiomes (Cohen et al. 2020).

Statistical Analyses

All statistical analyses were performed in Rstudio using R v4.5.2 (R Core Team 2025) and the packages “ANCOMBC” (Lin and Peddada 2020, Lin et al. 2022), emmeans (Lenth and Piaskowski 2025), glmmTMB (Brooks et al. 2017, McGillycuddy et al. 2025), “Microbiome” (Lahti and Shetty 2012), “OmicFlow” (Gusinac et al. 2025), “phyloseq”(McMurdie and Holmes 2013), and “Vegan” (Oksanen et al. 2025). Packages for data manipulation and graphing include DHARMa (Hartig 2024), DT (Xie et al. 2025), “ggordiplots” (Quensen et al. 2024), ggpubr (Kassambara 2025), microViz (Barnett et al. 2021), “MiscMetabar” (Taudière 2025), “multcomp” (Hothorn et al. 2008), performance (Lüdecke et al. 2021), “remotes” (Csárdi et al. 2024), and “tidyverse” (Wickham et al. 2019).

Microbiome Statistical Analysis

We calculated alpha diversity metrics Chao-1, Shannon, and Simpson’s for each land use type for both early and late samples. We modeled each diversity metric as a response variable independently across sites with linear models. The models of Chao-1 and Shannon used land use type, sample time, and their interaction as fixed effects and site as a random effect. The model of Simpson’s used the same model formula but was run as a beta regression with a logit link to account for the bounded and skewed structure of the data. We followed the models by calculating the estimated marginal means for land use type, grouped by sample time.

To compare the bacterial community composition across land use type, we ran a nested PERMANOVA, using Bray–Curtis dissimilarity and nesting at site and time. Because dropping samples with under 300 reads led to an unequal sample size between groups, we randomly subsampled each group to an equal size, running 999 separate PERMANOVAs (999 permutations each). We then calculated the mean and variance of the resulting test statistics and P-values across the 999 subsampled analyses. To calculate the log fold change of microbes between land use types, we ran ANCOM-BC2 twice, once with land use type as a predictor and once with percent natural cover as a predictor. Both were grouped at the genus level, with Holm–Bonferroni corrections, and a prevalence filter of 10%.

Pesticide Statistical Analysis

To compare the pesticide risk across land use type, we ran a linear model calculating the impact of land use type and sample time, and their interaction, on the log risk quotient, with a random effect for site. We followed the model by calculating the estimated marginal means for land use type, grouped by sample time. To compare the pesticide residue across land use type, we ran a PERMANOVA (9,999 permutations), nested at the site level, using Bray–Curtis dissimilarity.

Pesticide–Microbe Correlation Analysis

To determine if total pesticide concentration or cumulative pesticide risk had an impact on any alpha diversity metrics, we ran linear models for each alpha diversity metric. These models used total pesticide concentration, sample time, and their interaction, percent natural area at 500 m, and the cumulative pesticide risk as fixed effects. The interaction of cumulative pesticide risk with sample time was removed to increase parsimony as it was nonsignificant. Simpsons alpha diversity was fitted as a beta regression with a logit link to account for the skewed and bounded nature of the data. To assess if there was a link between the composition of pesticide contamination and microbiome composition, we used a Mantel test (9,999 permutations) with Bray–Curtis dissimilarity, and Spearman correlation. To construct consistent dissimilarity matrices for bacteria and pesticides required by the Mantel test, we first aggregated bacteria via mean at the site level within each time point.

Results

Microbiome

When comparing bacterial alpha diversity between sites, there was no significant effect of land use type on Chao-1; however, there was an effect of sample time (Wald X²₁ = 9.553, P = 0.002, Fig. 1). For both Shannon and Simpsons, there was a strong significant effect of land use type (Wald $X_2^2$ =14.996, P < 0.001 and Wald $X_2^2$ = 14.124, P < 0.001, respectively) as well as sample time (Wald $X_1^2$ = 6.505, P = 0.011 and Wald $X_1^2$ = 4.755, P = 0.029) and their interaction (Wald $X_2^2$ = 15.308, P < 0.001 and Wald $X_2^2$ = 11.352, P = 0.003).

Fig. 1.

Shannon and Simpson alpha diversity was significantly different between landscape types, while Chao-1 alpha diversity was not significantly different in the pollen provisions of Osmia cornifrons. Boxplots show a line at the median value and extend from the first to third quantile. Whiskers extend to the largest and smallest value within 1.5 times the interquartile range. Stars indicate significant pairwise differences (n.s., no significance, *P < 0.05, **P < 0.005, ***P < 0.0005).

The composition of the bacterial community was significantly different between land use types (F = 10.643, df = 2,177, P = 0.005) (Fig. 2 and Supplementary Fig. S1). Across land use type, the bacterial genus with the greatest log fold change (LFC) in abundance was Apilactobacillus (Lactobacillaceae), being highly abundant in forest sites compared to orchard sites (LFC = −4.31) and urban sites (LFC = −4.37) (Supplementary Table S3). For each percentage point increase in natural habitat coverage within 500 m of a site, the abundance of Apilactobacillus increased by an LFC of 0.054 ± 0.003 SE (Fig. 3). The genus with the second highest increase in LFC with increasing forest cover was Acinetobacter (Moraxellaceae) (LFC = 0.030 ± 0.004).

Fig. 2.

The relative abundance of the top 15 bacterial families in each pollen provision sample. All other taxa are grouped into “Other.” Taxa that could not be identified to the family level are labeled at their lowest identified taxonomic level. See Supplementary Fig. S2 for fungal relative abundance.

Fig. 3.

Log fold change (LFC) in bacterial abundance per percentage-point increase in natural land cover (500-m buffer). Only bacterial genera with robust changes identified through sensitivity analysis are included.

Pesticide Contamination and Risk

The mean number of unique pesticides in pollen provisions across land use type within time ranged from 3.2 ± 0.2, mean ± SE (late forest) to 10.8 ± 1.2 (late season orchard) (Supplementary Table S4), while the mean concentration of pesticides ranged from 0.144 ± 0.046 ppb (late season forest) to 17.177 ± 9.464 ppb (late season orchard). The composition of pesticide present at each site was significantly different across land use type (F = 4.8, df = 2, 27, P = 0.005, Fig. 4).

Fig. 4.

Pesticide contamination of pollen provisions was significantly greater in orchard sites compared to urban or forest sites. Each column represents 1 sample. Rows are grouped by A) fungicides, B) herbicides, C) insecticides, and D) synergists. Squares are colored by the pesticide concentration in parts per billion (ppb) with gray indicating the sample had no pesticide or the residue was below the level of detection. The “total” row shows the additive concentration of all pesticide contamination in that sample.

When adjusting to pesticide risk, the total risk present was significantly driven by the land use type (Wald $X_2^2$ = 9.012, P = 0.011), the sampling time (Wald $X_1^2$ = 11.413, P < 0.001), and their interaction (Wald $X_2^2$ = 30.772, P < 0.001). There was no significant difference in the total risk between site types in the early samples; however, sites were significantly different in the late time point with late season orchard samples having significantly more pesticide risk compared to late season forest sites (df = inf, estimate = 8.473, SE = 2.1, Z ratio = 4.032, P < 0.001) and late season urban sites (df = inf, estimate = 5.865, SE = 2.1, Z ratio = 2.791, P = 0.015, Fig. 5).

Fig. 5.

Pesticide risk (risk quotient) was highest in orchard sites compared to urban and forest sites. Each point represents 1 sampling location. The dashed lines indicate the US Environmental Protection Agency (EPA) and European Food Safety Authority (EFSA) levels of concern for exposure to honeybees. Risk quotient shows the additive risk of all pesticides found within the sample and may underestimate risk caused by pesticide synergism. Letters indicate significant differences between landscape types within a time point.

Pesticide–Microbe Correlation

Bacterial alpha diversity was significantly driven by the concentration of pesticides present at a site with the strength of the impact often modulated by the time of sampling (Fig. 6). Chao-1 alpha diversity was significantly impacted by the total concentration of pesticides present at a site (df = 1,230, F ratio = 16.486, P < 0.001), as well as the time of sampling (df = 1,230, F ratio = 13.938, P < 0.001). Shannon diversity was significantly impacted most by the total pesticide concentration (df = 1,230, F ratio = 75.553, P < 0.001), as well as the time of sampling (df = 1,230, F ratio = 14.280, P < 0.001), their interaction (df = 1,230, F ratio = 5.895, P = 0.016), and the percent of surrounding natural habitat (df = 1,230, F ratio = 10.848, P = 0.001). Simpsons alpha diversity was significantly impacted most by the total pesticide concentration (df = 1, inf, F ratio = 67.251, P < 0.001), as well as the time of sampling (df = 1, inf, F ratio = 12.129, P < 0.001), their interaction (df = 1, inf, F ratio = 9.002, P = 0.003), and the percent of surrounding natural habitat (df = 1, inf, F ratio = 5.800, P = 0.016). Pollen provision bacterial alpha diversity was not significantly impacted by the total toxicity risk of pesticides present, and we found no correlation between the specific composition of the pesticides present at a site and the composition of bacteria present (r = 0.097, P = 0.177).

Fig. 6.

All alpha diversity metrics for bacterial communities in pollen provisions were negatively correlated with total pesticide concentration. Lines show linear model predictions with shaded areas representing 95% confidence intervals. The X-axis is log transformed to aid in readability, resulting in curved best-fit lines. Points are shaded according to percent natural area within 500 m from each site.

Discussion

We found a significant effect of land use and pesticide contamination on the bacterial community in the pollen provisions of O. cornifrons. The pollen microbiome at forest sites had a significantly different composition from urban and orchard sites, and an increase in natural cover was correlated with a significant increase in the abundance of the bee-associated bacteria Apilactobacillus (Lactobacillaceae). Apilactobacillus is commonly detected in solitary bee pollen provisions (Hammer et al. 2023) and bee gastrointestinal tracts (Sless et al. 2026) and is highly beneficial across many bee species, providing pathogen resistance and increased larval survival (Todorov et al. 2024, Nguyen and Rehan 2025). We also found a strong negative impact of total pesticide contamination, and to a lesser degree decreased natural cover, on pollen provision bacterial alpha diversity. While we did not monitor bee health in this study, decreased pollen microbiome richness has been shown to lower developmental success in the closely related Osmia lignaria (Westreich et al. 2023). We did not find a significant effect of hazard quotient on pollen provision bacterial diversity likely because many antimicrobial pesticides have a very low direct toxicity (high LD₅₀). This highlights that even pesticides with low toxicity can have significant sublethal impacts on bees by shifting their pollen microbiome and decreasing its diversity.

While we found a similar number of bacterial species present regardless of land use type, there were significant differences in the degree of evenness across sites and time. Orchard sites had significantly less evenness than urban sites in early time points and significantly lower evenness than both other land use types in the late time point as natural sites increased in diversity, likely driven by an increase in diverse floral resources as spring progressed. On the other end of the spectrum, urban sites had comparatively higher levels of Shannon and Simpson alpha diversity, especially in the early time points, perhaps due to the extreme diversity of foraging resources found in urban environments at the start of the spring (Seitz et al. 2022). This suggests that the disrupted nature of orchard sites, where monoculture plantings and continuous antimicrobial sprays are common, might depopulate and skew the microbial community and allow for the increased dominance of certain taxa.

The strong impact of pesticide contamination on bacterial diversity could be driven directly by the detected pesticides, with many insecticides, herbicides, and fungicides having been shown to have off-target impacts on the bacterial microbiomes of bees (Daisley et al. 2022, Helander et al. 2023). Osmia cornifrons larvae fed on specific fungicide contaminated pollen were shown to have significantly decreased bacterial microbiome diversity and increased mortality (Porras et al. 2024). Alternatively, as our multiresidue pesticide screening method using HPLC-MS-MS is not well suited for capturing highly polar molecules such as many antibiotics which ionize poorly, it is also possible that the total pesticide concentration detected may act as an indicator for the level of undetected antibiotics which could impact the bacterial diversity as well.

The significant decrease in Apilactobacillus, a bacterium commonly associated with bee pollen provisions, together with the overall decline in bacterial diversity, suggest that agricultural development and human land management alter the environmental microbial assemblage and, consequently, the microbial communities colonizing bee pollen provisions. Previous work has demonstrated that pollen provision microbiomes are dynamic and are modified during larval development, with larvae selectively enriching certain bacterial taxa while reducing others, potentially influencing nutrient availability (Kueneman et al. 2022). However, little is known about the exact roles that different taxa play in the solitary bee pollen provision. It is likely, however, that anthropogenic changes to the environmental microbial community may have downstream consequences for pollen provision microbiomes and larval development.

We did not find a direct correlation between the specific pesticides detected at a site and the composition of the pollen provision bacterial microbial community. It is possible that this is due to the loss of our fungal samples and the predominance of fungicides that we detected. It is also possible that the impacts of antimicrobial sprays do not induce a consistent shift in the microbiome membership but instead induce broader patterns by suppressing diversity and allowing fewer taxa to dominate.

We detected pesticide residues in every sample across land use types with cumulative estimated toxicity highest in orchard sites and lowest in forest sites. High levels of herbicides were found in all samples. While herbicides are thought of as low toxicity and many have no recorded honeybee LD₅₀, sublethal effects can occur. Atrazine, which was detected in nearly all our samples has been shown to decrease food intake and increase susceptibility to pathogens in honeybees (Wang et al. 2023). Fungicides were the predominant pesticides detected in orchard sites, and to a lesser degree, urban sites and recent work suggests that some fungicides have high levels of direct toxicity (Ladurner et al. 2005, Zhu et al. 2014, De Souza et al. 2024) as well as often having synergistic effects with other pesticides (Iverson et al. 2019). A high frequency of fungicides and herbicide contamination has been previously found in wild bees along agricultural fields (Hladik et al. 2016), although as far as we know, no studies have quantified the pesticide contamination of solitary bees in forests.

While many studies have looked at the pesticide exposure of bees, fewer have focused on solitary bees or their pollen provisions. Here, we sample the pollen provisions of the easily managed O. cornifrons which may act as a proxy or sentinel species for difficult to sample cooccurring bee species. As a generalist, Osmia can have a broad foraging breadth (Vaudo et al. 2020) and may have potential for collecting data on environmental pesticide contamination, similar to studies using passive silicone bands (Graham et al. 2025). However, no studies have tested how the pesticides collected in Osmia pollen correlate to the local pesticide spray regime, as spray records are often proprietary secrets and pesticide exposure in different bee species can differ drastically, even when collected in the same apple orchards (Mueller et al. 2024). Additionally, it is unknown how these shifts in pollen microbiome induced by land use change and pesticide exposure will impact O. cornifrons, or other solitary bees, at the population level. Previous work suggests increased fungicide use can increase the prevalence of the highly lethal fungal pathogen Ascosphaera, although it was also found that increased forest cover can moderate that effect (Krichilsky et al. 2020). This moderating effect may be due to the more diverse pollen microbiome we found to be present in natural sites which could provide a defensive mutualism and prevent the fungal pathogen from taking over the pollen microbiome. It should be noted that these data were collected in an area with comparatively small-scale agriculture. The average size of apple orchards in our study was approximately 21 acres (ranging from ∼10 to ∼43 acres) and often had a mix of land uses between, and surrounding, fields. Natural areas surrounding agricultural sites can buffer bees from the impact of pesticides (Park et al. 2015) and it is likely the observed impacts would be even stronger in more disrupted agricultural areas where there is less natural cover and increased pesticide use. However, despite the small scale of agriculture in our study, we still found significant impacts of land use and pesticides on bee pollen provision.

Our findings demonstrate that even local-scale differences in land use can substantially influence both pesticide contamination and the pollen provision bacterial communities of O. cornifrons. Rather than acting solely through direct toxicity, agricultural development and pesticide exposure may also affect solitary bees indirectly by altering the microbial communities associated with larval food resources. The positive association between natural habitat, increased microbial diversity, and greater abundance of Apilactobacillus further suggests that surrounding natural areas may help maintain microbial communities important for brood development. Collectively, these findings identify pollen provision microbiomes as a previously underappreciated link between landscape change and bee health, and highlight the need to incorporate microbial-mediated pathways into future assessments of pollinator conservation and pesticide risk.

Author Contributions

Tobias Mueller (Conceptualization [equal], Formal analysis [equal], Investigation [equal], Methodology [equal], Writing—original draft [equal], Writing—review & editing [equal]), Jordan Kueneman (Conceptualization [equal], Investigation [equal], Methodology [equal], Writing—review & editing [equal]), Rachel Fordyce (Conceptualization [equal], Investigation [equal], Methodology [equal]), Scott McArt (Methodology [equal], Supervision [equal], Writing—review & editing [equal]), and Bryan N. Danforth (Conceptualization [equal], Funding acquisition [equal], Methodology [equal], Supervision [equal], Writing—review & editing [equal])

Supplementary Material

Supplementary material is available at Journal of Insect Science online.

Funding

This work was supported by National Science Foundation grant NSF-DEB 1929499 “Collaborative Research: The brood cell microbiome of solitary bees: origin, diversity, function, and vulnerability” (PI: Bryan Danforth, Co-PIs: Shawn Steffan, Rachel Vannette, and Quinn McFrederick) as well as United States Department of Agriculture Multistate grant (NC1173) administered by the Cornell University Agricultural Experiment Station, the Cornell Atkinson Center for Sustainability Academic Venture Fund, and United States Department of Agriculture NIFA (2018-08601). TM was funded through NSF GRFP #DGE-2139899.

Conflicts of Interest

The authors declare no conflict of interest.