Impact of Glyphosate on the Honey Bee Gut Microbiota: Effects of Intensity, Duration, and Timing of Exposure

As is true of many animal species, honey bees depend on their gut microbiota for health. The bee gut microbiota has been shown to regulate the host immune system and to protect against pathogenic diseases, and disruption of the normal microbiota leads to increased mortality. Understanding these effects can give broad insights into vulnerabilities of gut communities, and, in the case of honey bees, could provide information useful for promoting the health of these economically critical insects, which provide us with crop pollination services as well as honey and other products. The bee gut microbiota is acquired early in adult life and can be compromised by antibiotics and other chemicals. The globally used weed killer glyphosate was previously found to impact the gut microbiota of honey bees following sustained exposure. In the present study, we address how this impact depends on concentration, duration of exposure, and stage of community establishment. We found that sublethal doses of glyphosate reduce the abundance of beneficial bacteria and affect microbial diversity in the guts of honey bees, regardless of whether exposure occurs during or after microbiota acquisition. We also compared the effects of glyphosate to those of tylosin, an antibiotic used in beekeeping, and observed that tylosin effects diverge from those caused by glyphosate and are greater during microbiota acquisition. Such perturbations are not immediately lethal to bees but, depending on exposure level, can decrease survivorship under laboratory conditions.

IMPORTANCE As is true of many animal species, honey bees depend on their gut microbiota for health. The bee gut microbiota has been shown to regulate the host immune system and to protect against pathogenic diseases, and disruption of the normal microbiota leads to increased mortality. Understanding these effects can give broad insights into vulnerabilities of gut communities, and, in the case of honey bees, could provide information useful for promoting the health of these economically critical insects, which provide us with crop pollination services as well as honey and other products. The bee gut microbiota is acquired early in adult life and can be compromised by antibiotics and other chemicals. The globally used weed killer glyphosate was previously found to impact the gut microbiota of honey bees following sustained exposure. In the present study, we address how this impact depends on concentration, duration of exposure, and stage of community establishment. We found that sublethal doses of glyphosate reduce the abundance of beneficial bacteria and affect microbial diversity in the guts of honey bees, regardless of whether exposure occurs during or after microbiota acquisition. We also compared the effects of glyphosate to those of tylosin, an antibiotic used in beekeeping, and observed that tylosin effects diverge from those caused by glyphosate and are greater during microbiota acquisition. Such perturbations are not immediately lethal to bees but, depending on exposure level, can decrease survivorship under laboratory conditions.
Here, we investigate the effects of glyphosate on the honey bee gut microbiome by exposing newly emerged workers, both during and after microbiota acquisition, and using different concentrations of glyphosate, ranging from residues detected in the environment (23)(24)(25) to those detected in nectar and pollen in a semi-field experiment (20). We also investigated and compared the effects of tylosin, an antibiotic commonly used in beekeeping, to the effects of glyphosate. For that, we used quantitative PCR (qPCR) and 16S rRNA gene sequence analyses to measure bacterial abundance and composition in the bee guts. Our findings show for the first time that glyphosatemediated perturbation of the bee gut microbiota, especially the effects on Snodgrassella abundance, is dose dependent and that a similar pattern of perturbation occurs regardless of the age of the bee and duration of exposure.

RESULTS
Effects of glyphosate and tylosin on honey bees during and after acquisition of the gut microbiota. Two groups of recently emerged bees, 1-day-old bees still acquiring the gut microbiota (early-exposure group) and 5-day-old bees with established gut microbiotas (late-exposure group), were each divided into seven subgroups to be continually exposed to five different concentrations of glyphosate ranging from 0.01 mM (1.691 mg/liter, or 1.375 mg/kg) to 1.0 mM (169.1 mg/liter, or 137.5 mg/kg), tylosin at 0.1 mM (91.61 mg/liter, or 74.48 mg/kg), or sucrose syrup only. Because we expected tylosin, as an antibacterial compound, to affect the community, this treatment served as a type of positive control. Bees were sampled at different times after the start of treatment (days 5, 10, 15, and 20 in the early-exposure group and days 5, 10 and 15 in the late-exposure group) (Fig. 1). RNA was extracted from whole guts of sampled bees and processed for 16S rRNA amplicon sequencing.
Exploratory principal-component analysis (PCA) of the relative abundance of the main bee gut bacterial taxa suggested a continuum in divergence between gut communities of control and glyphosate-treated bees in both early-and late-exposure groups, with higher glyphosate concentrations and later sampling dates being associated with greater divergence (Fig. 2). The taxa contributing most to this divergence were Snodgrassella and Gilliamella, which were negatively correlated with glyphosatetreated bees in both exposure groups, and Bifidobacterium, Lactobacillus Firm-4, and Lactobacillus Firm-5, which were positively correlated with glyphosate-treated bees, especially in the early-exposure group. Noncore environmental bacteria were also positively associated with glyphosate-treated samples in the early-exposure group.
Communities in tylosin-treated bees diverged from those in both control and glyphosate-treated bees throughout the experiment in the early-exposure group and to a lesser extent in the late-exposure group (Fig. 2). They showed shifts opposite to those in glyphosate-treated bees in relative abundances of bacterial taxa. Moreover, Commensalibacter abundance was positively correlated with tylosin-treated bees at specific sampling times in both early-and late-exposure groups. Some of these trends from the PCAs correspond to significant changes in bacterial relative abundance between control and treated samples ( Fig. S1; Table S1A). We also estimated absolute bacterial abundance using quantitative PCR (qPCR). Glyphosate treatment reduced total 16S rRNA copy numbers in the guts of bees exposed to 1.0 mM glyphosate at day 20 in the early-exposure group, while absolute abundances were not significantly affected for lower concentrations, earlier sampling times, or late exposure ( Fig. 3; Table S1B). Taking into consideration both relative and absolute abundances, glyphosate treatment inhibited growth of Snodgrassella, for which 16S rRNA copy numbers were reduced in all bees treated with 0.07 to 1.0 mM glyphosate throughout the experiment in both early-and late-exposure groups ( Fig. 4; Table S1C). In most cases, the 0.04 mM treatment also resulted in a significant decrease in Snodgrassella. A dose-dependent response of Snodgrassella abundance to glyphosate treatment was observed for both early-and late-exposure groups (Fig. 5). The 50% effective doses (ED 50 s), i.e., the half-maximal effective concentrations of glyphosate at which Snodgrassella growth is reduced by 50%, ranged from 0.02 mM to 0.06 mM glyphosate (Table S2). Regarding other core bee gut bacteria, Gilliamella absolute numbers were reduced in bees treated with 0.07 mM or 1.0 mM glyphosate at specific sampling times in the early-exposure group ( Fig. S2; Table S1C).
Tylosin treatment broadly impacted the gut microbiome in the early-exposure group, decreasing total bacterial abundance ( Fig. 3A; Table S1B) and specifically lowering abundance of the Gram-negative bacteria Snodgrassella (days 5, 10, and 20 [ Fig. 4A; Table S1C]) and Gilliamella (day 5 [ Fig. S2A; Table S1C]), as well as the Gram-positive bacteria Lactobacillus Firm-5, Lactobacillus Firm-4, and Bifidobacterium at almost all sampling times ( Fig. S2; Table S1C). However, in the late-exposure group, tylosin treatment reduced total bacterial numbers only at day 10 ( Fig. 3B; Table S1B), suggesting that exposure has less impact on the established microbiota. In the lateexposure group, tylosin did not affect Snodgrassella abundance ( Fig. 4B; Table S1C) but did cause some changes in 16S rRNA copy numbers for other bacteria, such as reductions for Lactobacillus Firm-4 and Lactobacillus Firm-5 at specific sampling times and an increase for Commensalibacter at day 15 ( Fig. S2; Table S1C). Therefore, the higher concentrations of glyphosate impact the abundance of specific bacterial members, especially Snodgrassella, in both developing and established gut communities. In contrast, tylosin effects extend to other core bacterial members and appear to differ depending on whether exposure is early or late, with effects being greater for developing communities.
Survivorship of worker honey bees exposed to glyphosate or tylosin. The life spans of young worker bees chronically treated with different concentrations of glyphosate and with tylosin were monitored at two stages, during and after gut  (Table S1B). microbiota acquisition, and during two seasons, fall 2018 and summer 2019. In the fall 2018 experiment, bees were also sampled every 5 days to assay changes in gut microbial abundance and composition.
In the fall 2018 experiment, 1-day-old bees treated with 0.1 mM or 1.0 mM glyphosate for 20 days (early exposure) exhibited increased mortality compared to control bees ( Fig. 6A; Table S3). Similar effects were observed for 5-day-old bees treated with 0.1 mM or 1.0 mM glyphosate for 15 days (late exposure) ( Fig. 6B; Table S3). Surprisingly, 5-day-old bees chronically treated with 0.01 mM glyphosate exhibited higher survival rates than control bees ( Fig. 6B; Table S3). Treatment with 0.1 mM tylosin Groups with different letters are statistically significantly different (P Ͻ 0.05, Kruskal-Wallis test followed by Dunn's multiplecomparison test) (Table S1C).
increased mortality of bees only when exposure started immediately after emergence ( Fig. 6A; Table S3).
In summer 2019, bees in the early-exposure group treated with 0.1 mM or 1.0 mM glyphosate exhibited increased mortality compared to control bees after 40 days of exposure ( Fig. 6C; Table S3). This was also true for bees in the late-exposure group after 35 days of treatment ( Fig. 6D; Table S3). However, we did not observe increased mortality of bees chronically treated with 0.1 mM tylosin ( Fig. 6C and D; Table S3). Overall, bees in the late-exposure group experienced elevated mortality rates compared to bees in the early-exposure group, probably due to the extra handling and CO 2 exposure at day 5, when they were reassigned to new cup cages.
Potentially, the decreased survivorship of honey bees chronically exposed to 1.0 mM glyphosate is due to effects on bee physiology other than perturbations of the gut microbiota. To investigate this, we performed a laboratory experiment in fall of 2019 in which we exposed 1-day and 5-day-old bees to sterile sucrose syrup with and without 1.0 mM glyphosate throughout their life span (bees also had free access to sterile pollen). These bees were deprived of their normal microbiota, an abnormal state that is possible only under laboratory conditions. For both age groups, glyphosate-exposed bees exhibited higher mortality than control bees ( Fig. S3; Table S3). Thus, the decreased survivorship of honey bees exposed to 1.0 mM glyphosate under laboratory conditions may involve effects other than those on the microbiome.
Microbial diversity analysis. To investigate potential changes in microbial community diversity between control and treatment groups, we used the relative abundance profiles obtained by 16S rRNA amplicon sequencing to estimate dissimilarity matrices using Bray-Curtis dissimilarity, which reflects community composition, and weighted UniFrac distance, which takes into account phylogenetic relationships among members of the bacterial communities and amplicon sequence variant (ASV) abundance (29). In addition, we calculated alpha diversity by means of Shannon's H index, a commonly used metric that accounts for both taxon richness and evenness of ASVs in each sample.
Based on pairwise permutational multivariate analysis of variance (PERMANOVA) using Bray-Curtis dissimilarity indices, and visualized using principal-coordinate analysis (PCoA), gut community compositions were usually significantly affected by glyphosate doses higher than 0.04 mM in the early-exposure group or 0.07 mM in the late-exposure  (Table S2). group ( Fig. 7; Table S4). On the other hand, pairwise PERMANOVA using weighted UniFrac indices detected effects of only the higher dose of glyphosate (1.0 mM) and only after 10 days of exposure ( Fig. S4; Table S4). Gut community compositions of tylosin-treated bees were significantly different from those of both control and glyphosate-treated bees throughout the experiments, regardless of exposure group, sampling time, and dissimilarity matrix used (Fig. 7, Fig. S4, and Table S4). Regarding alpha diversity, Shannon's H index was significantly lower in tylosintreated bees than in control bees at most sampling times in the early-exposure group but only at the last sampling time in the late-exposure group (Fig. S5; Table S5). No significant changes were observed between control and glyphosate-treated bees ( Fig. S5; Table S5).

DISCUSSION
Anthropogenic chemicals potentially affect gut microbiotas, with consequences for hosts themselves. Previous evidence of impacts of glyphosate, the main herbicide used globally in agriculture, on the honey bee gut microbiota prompted us to investigate the effects of dose and of timing of exposure. Honey bees acquire their gut microbiotas during the first few days after emergence from the pupal stage, and these symbionts Bees were sampled from each group every 5 days after the start of treatment, and dead bees were counted and removed from cup cages in a daily census. (C and D) The experiment was replicated in the summer of 2019, when 1-day-old bees (C) and 5-day-old bees (D) were divided into seven groups, composed of 4 cup cages each with 38 to 40 or 30 to 32 bees, respectively, which were fed 0.01, 0.04, 0.07, 0.1, or 1.0 mM glyphosate in sterile sucrose syrup, 0.1 mM tylosin in sterile sucrose syrup, or sterile sucrose syrup for 40 and 35 days, respectively. Bees were not sampled during treatment; dead bees were removed from cup cages in a daily census. The Cox proportional hazards model was implemented in the survival package in R (Table S3). *, P Ͻ 0.05; **, P Ͻ 0.01; ***, P Ͻ 0.001. have multiple functions in bee development, behavior, nutrition, and immunity (2,3,5,17,19,30,31). Changes in microbial abundance or composition due to chemical exposure can interfere with these functions (9,10,13). Newly emerged worker bees perform crucial tasks in the hive, such as feeding and taking care of the brood (2), and they exhibit high physiological and behavioral plasticity (32), potentially making them more susceptible to perturbations of the gut microbiota. Exposure of young bees can  (Table S4).
result from direct applications of chemicals in colonies or from residues in pollen and nectar that foragers bring to colonies. While some effects, such as those due to insecticide exposure, are well known (33)(34)(35)(36), the effects of other agrochemicals, in particular antibiotics (10), fungicides (37), and herbicides (13), on the honey bee gut microbiota and health are starting to gain more attention.
Here, we demonstrate that laboratory-raised worker bees chronically treated with glyphosate have abnormal microbiotas. Bees, both at the stage of microbiota acquisition and with established microbiotas, were fed glyphosate in concentrations ranging from those detected in environmental samples (0.01 mM) (23)(24)(25) to those measured in nectar and pollen collected by foragers from plants recently sprayed with glyphosate (1.0 mM) (20). These concentrations are far lower than concentrations sprayed in the field from glyphosate-based formulations; they represent levels documented to occur in at least some circumstances (20,(23)(24)(25). Bees were sampled throughout the experiment to assess potential changes in microbial abundance and composition. Effects of glyphosate on the microbiota, especially on Snodgrassella, one of five main core bacterial species in the bee gut, were detected throughout the sampling times (Fig. 4). Although we expected that early exposure, during initial colonization by the microbiota, might have a greater impact, we found instead that impact was similar regardless of whether treatment started before or after microbiota acquisition. For both early and late exposure, these effects increased from lower to higher concentrations of glyphosate, exhibiting a dose-response relationship (Fig. 5). Because we sampled bees destructively, we could not follow changes within a single gut community over time, and this limitation was compounded by the large variation in community composition among individual bees. Despite variation, experiments of this study nonetheless revealed strong effects that are consistent with those of previous studies (11,13).
The species consistently most affected by glyphosate was Snodgrassella alvi, a Gram-negative bacterium that primarily colonizes the hindgut, where it forms a biofilm on the ileum wall. This biofilm may facilitate colonization by other beneficial bacteria and/or act as a mechanical barrier against pathogen invasion (2). Previous findings have suggested that Snodgrassella strains vary in susceptibility to glyphosate. Motta et al. (13) reported two strains of Snodgrassella that were tolerant to glyphosate when cultured in vitro, whereas all the other strains tested were susceptible. However, the overall abundance of Snodgrassella decreased in every experiment in which honey bees were exposed to sufficient concentrations of glyphosate, as shown in this study and others (11,13). Thus, tolerant Snodgrassella strains either are atypical or are not resistant in the context of an intact gut community.
Previous studies have shown that sublethal doses of glyphosate affect the microbiotas of young and adult worker bees under laboratory conditions, but no doseresponse relationship was determined. Dai et al. (12) studied the midgut microbial composition of newly emerged bees exposed to glyphosate (0.8, 4, and 20 mg/liter) as larvae. They found changes in microbial diversity between control and treatment bees, especially in bees fed 20 mg/liter glyphosate at the larval stage. Besides the effects on the microbiota, they observed lower survival rates for larvae fed 4.0 and 20 mg/liter glyphosate. In addition, we previously found that recently emerged bees treated with glyphosate (16.91 and 169.1 mg/liter) have altered hindgut microbiota with reduced populations of beneficial bacteria (13). We also detected reduced microbial abundance in adult workers released in their hives after being fed 5 or 10 mg/liter glyphosate (13). Blot et al. (11) found that glyphosate exposure at 253.6 or 1,268.2 mg/liter affects the gut microbiota of adult worker bees under laboratory conditions. Together, these results corroborate the findings described in this study, in which microbiotas of young bees exposed to glyphosate were affected, both during and after establishment of the gut community.
We compared effects of glyphosate to effects of tylosin, an antibiotic commonly used in beekeeping for the control of American and European foulbrood disease (38)(39)(40). Tylosin mainly affected total bacterial abundance and Snodgrassella abundance when treatment started immediately following worker emergence, unlike glyphosate, which decreased Snodgrassella abundance regardless of treatment timing. Tylosin strongly reduced the abundance of Gram-positive bacteria, which was not the case for glyphosate. This corroborates previous findings on the effects of other antibiotics, such as tetracycline (10,41) and penicillin-streptomycin (9), on the bee gut microbiota. The lack of effects of glyphosate on Gram-positive bacteria in the bee gut under laboratory conditions, especially on Lactobacillus, may be explained by the fact that these bacteria lack a functional shikimate pathway, which contains the enzyme EPSPS, the target of glyphosate (13). However, occasional decreases in abundance for these bacteria may reflect their reliance on the uptake of essential nutrients produced by glyphosate-affected bacteria. Because essential amino acids are absorbed by the host midgut and low in the hindgut (18), the hindgut community must synthesize these, using ammonia from host waste that enters at the midgut-hindgut juncture. Thus, some species rely on cross-feeding, at least for amino acids, and possibly for other nutrients, and disruption of glyphosate-sensitive species can have ripple effects throughout the community.
A critical question is how observed shifts in microbiota affect bee survivorship. Our survivorship experiments used bees caged in the laboratory (Fig. 6) and thus failed to capture sources of mortality that affect bees in colonies. Indeed, previous experiments with a single glyphosate dose showed that exposure lowered the ability of caged bees to survive following challenge with pathogenic Serratia marcescens but did not affect survivorship otherwise (13). Similarly, tetracycline exposure increased mortality of bees in colonies and of caged laboratory bees challenged with S. marcescens but did not affect survivorship of caged control bees (10). We found that, under laboratory conditions and without pathogen challenge, only the higher doses of glyphosate (0.1 mM and 1.0 mM) reduced the life span of young worker bees. In studies that did not examine the role of gut microbiota, glyphosate was shown to directly compromise larval development and adult behavior (12,(42)(43)(44)(45), although some other studies failed to detect effects on survivorship (20,46,47). Tylosin, which had large effects on the gut microbiota, increased mortality in only one of the four trials with caged laboratory bees (Fig. 6). Further experiments tracking exposed bees in field colonies could determine whether the observed microbial perturbations are detrimental for bees in the hive environment, where they are more exposed to opportunistic pathogens and other environmental stressors.
The higher doses of glyphosate used in our study are similar to the ones quantified in nectar and pollen from plants recently treated with glyphosate-based formulations (20). Although foraging bees may encounter glyphosate at these concentrations, glyphosate-susceptible plants die a few days after exposure and glyphosate concentrations decrease with time on those plants (20). Therefore, it is unlikely that young worker bees would be chronically exposed to these concentrations of glyphosate. The lower dose of glyphosate tested, 0.01 mM (1.691 ppm), did not affect the gut microbiota or the survival rates of honey bees (except in one trial in which survivorship of control bees was lower). This lower dose resembles concentrations previously measured in the environment in water sources (23)(24)(25) and in nectar samples taken from colonies in a semi-field experiment (20) but is still higher than concentrations documented in honey, the most obvious source of chronic exposure (21,22,48,49).
Another issue in laboratory experiments is that captive honey bees usually do not defecate, causing DNA from dead bacterial cells to accumulate within the gut. Therefore, DNA samples used to profile microbial communities may give a biased picture of abundances of living bacterial cells. Indeed, for soil microbial communities, studies have demonstrated that relic DNA can obscure estimates of microbial diversity (50). Most experiments documenting effects of glyphosate on the bee gut microbiota have been performed under laboratory conditions using DNA extracted from dissected guts. All such experiments have revealed a strong reduction in the abundance of Snodgrassella, regardless of potential issues with dead bacterial cells (11,13). To reduce artifacts resulted from relic DNA, we extracted RNA and compared relative rRNA abundances; the effects of glyphosate were similar to those observed in studies based on genomic DNA. However, this may not be true for all chemical stressors. In a previous study, we extracted both DNA and RNA from individual bee guts, and only RNA samples showed significant effects of tylosin treatment for some bacterial taxa (13). Therefore, the use of RNA samples, or removing relic DNA before DNA extraction from viable cells, could provide more accurate measures of the perturbations to living bacterial populations.
While glyphosate exposure had substantial and consistent effects on abundance of individual bacterial taxa, these were not reflected in Shannon's H estimates of alpha diversity ( Fig. S5; Table S5) or in total bacterial abundance (Fig. 3). Thus, standard measures of diversity fail to capture changes in community composition that have large functional consequences, given that individual members of the community have very different metabolic capabilities and occupy different locations in the gut (13,14). We also note that the diversity measures in our study were based on profiles of amplicons within 16S rRNA genes and thus would not reveal changes in strain-level composition. Potentially, using more variable marker loci would reveal reductions in strain diversity, as observed in a study of community shifts following tetracycline exposure (41). For example, a few strains of Snodgrassella appear to be glyphosate resistant (13), and these might increase in frequency following exposure.
In summary, glyphosate concentrations ranging from 0.07 mM to 1.0 mM reduced the abundance of beneficial bee gut bacteria, particularly Snodgrassella, in a dosedependent way. Such effects occurred regardless of the stage of microbiota establishment and regardless of duration of exposure. Although the concentrations used in this study are higher than the ones commonly detected in honey samples, they are in the range of concentrations found in nectar and pollen of recently exposed plants that may be used as a source of food by foraging bees. These concentrations are far lower than the concentrations sprayed in fields, where foraging bees can be topically exposed. Although a chronic exposure would be unlikely in this last case, direct spraying on foragers is a potential route through which hives might be contaminated.

MATERIALS AND METHODS
Chemicals and solutions. Glyphosate standard was purchased from Research Products International, USA (lot 32612-38399). Tylosin tartrate was purchased from GoldBio, USA (lot 2313.081915A). For all the experiments, glyphosate and tylosin were initially dissolved in distilled water and then diluted to the final concentration with filter-sterilized 0.5 M sucrose syrup. Glyphosate concentrations were expressed in molarity (millimolar units) but can be converted to milligrams per liter by multiplying by the molecular weight of glyphosate, which is 169.1 g/mol, or to milligrams per kilogram by multiplying by the molecular weight of glyphosate and then dividing by the density of sucrose syrup, considered here as 1.23 g/ml. For example, the lower concentration of glyphosate used in our experiments was 0.01 mM, which is 1.691 (0.01 ϫ 169.1) mg/liter, or 1.375 (1.691/1.23) mg/kg. The same can be done for tylosin, whose molecular weight is 916.1 g/mol.
First chronic exposure of honey bees to glyphosate and tylosin. Approximately 2,000 late-stage pupae, with pigmented eyes but lacking movement, were removed from two brood frames from a hive of the European Apis mellifera kept at the University of Texas-Austin in fall of 2018. Pupae were transferred to clean plastic bins and placed in an incubator at 35°C and ϳ60% relative humidity to simulate hive conditions until they emerged as adults. Healthy newly emerged workers (NEWs) were transferred in groups of 26 to 30 to cup cages containing sterile sucrose syrup and bee bread (total of 56 cup cages). The bee bread was mixed with a bee gut homogenate so the bees could acquire their normal microbiota. In brief, guts were aseptically pulled out from healthy workers from the same hive and homogenized with equal proportions of 1ϫ phosphate-buffered saline (PBS) and sterile sucrose syrup. Gut homogenate (200 l) was transferred to the bee bread provided to the bees in each cup cage. Then, bees were divided into two main groups.
(i) Early-exposure group. Right after the gut homogenate treatment, 28 cup cages were divided into 7 subgroups which were fed 0.01, 0.04, 0.07, 0.1, or 1.0 mM glyphosate in sterile sucrose syrup, 0.1 mM tylosin in sterile sucrose syrup, or sterile sucrose syrup for 20 days. Fifteen bees were sampled from each group at days 5, 10, 15, and 20 after the start of the chemical feeding, placed in 5-ml Falcon tubes, and stored at Ϫ80°C. Bees were quickly immobilized with CO 2 before each sampling. Survival rates were monitored, and dead bees were removed in a daily census.
(ii) Late-exposure group. After 5 days of gut homogenate treatment, which is sufficient time for establishment of the gut microbiota (51), the remaining cup cages were divided into 7 subgroups which were fed similarly to the early-exposure group. Fifteen bees were sampled from each group at days 5, 10, and 15 after the start of the chemical feeding, placed in 5-ml Falcon tubes, and stored at Ϫ80°C. Bees were quickly immobilized with CO 2 before each sampling. Survival rates were also monitored, and dead bees were removed in a daily census.
The second pooled library consisted of samples from the late-exposure group (5-day-old bees treated with chemicals for 15 days; total of 258 samples). Each library was loaded onto an Illumina iSeq cartridge according to the manufacturer's instructions and subjected to Illumina sequencing on the iSeq platform (2 ϫ 150 sequencing run; instrument model number FS10000184). PhiX (5%) was used to increase library diversity.
Processing of 16S rRNA amplicon data. Illumina sequence reads were demultiplexed on the basis of the barcode sequences by the iSeq software, then processed according to exposure group in QIIME 2 version 2019. 10 (52). Due to the lack of sufficient overlap between forward and reverse reads, downstream analyses were performed with forward reads only. Primer sequences were removed using the cutadapt plugin (53). Then, reads were truncated to a length of 120, filtered, and denoised, and chimeric reads were removed using the DADA2 plugin (54). Taxonomy was assigned to amplicon sequence variants (ASVs) using the SILVA database in the feature-classifier plugin (55). Reads with less than 0.1% abundance were removed from the data set using the feature-table plugin, as were unassigned, mitochondrial, and chloroplast reads using the taxa filter-table plugin. Microbial diversity analyses were performed using the diversity plugin with a sampling depth of 1,000 reads per sample.
Statistical analyses. Statistical analyses were performed in R version 3.5.2 (56) or QIIME 2 version 2019.10 (52). Principal-component analyses (PCAs) were performed with the prcomp function in the R package stats to determine the similarity of the bacterial communities between control and treatment groups. For that, the taxonomic classification based on the SILVA database was complemented by assigning the ASVs to the major bacterial taxa of the bee gut microbiota based on a BLAST search against the NCBI nucleotide database. Then, the relative abundances of the phylotypes were transformed by using log 10 (x ϩ 1) and used to plot the PCAs. Comparisons of changes in bacterial abundance or alpha diversity (Shannon's H index) between control and treatment groups were performed using the nonparametric Kruskal-Wallis test followed by Dunn's multiple-comparison test, if significant, in R. Principalcoordinate analyses based on Bray-Curtis dissimilarity or weighted UniFrac were plotted using the R package phyloseq (57), and statistical tests were performed using pairwise permutational multivariate analysis of variance (PERMANOVA) tests with 999 permutations in QIIME 2. Dose-response models were fitted using the drm and LL.4 functions to fit and define the structure of the regression model, and the modelFit function was used to obtain a lack-of-fit test, all in the R package drc (58). Comparisons of survival rates between control and treatment groups were performed using Kaplan-Meier survival curves and the Cox proportional hazards model implemented in the R package survival (59).
Data availability. All sequence data are available in NCBI BioProject, no. PRJNA578403. The other data generated during this study are included in this article and the supplemental material.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only.