Evaluation of the herbicide glyphosate, (aminomethyl) phosphonic acid, and glyphosate-based formulations for genotoxic activity using in vitro assays

Glyphosate, the most heavily used herbicide world-wide, is applied to plants in complex formulations that promote absorption. The National Toxicology Program reported in 1992 that glyphosate, administered to rats and mice at doses up to 50,000 ppm in feed for 13 weeks, showed little evidence of toxicity, and no induction of micronuclei was observed in the mice in this study. Subsequently, mechanistic studies of glyphosate and glyphosate-based formulations (GBFs) that have focused on DNA damage and oxidative stress suggest that glyphosate may have genotoxic potential. However, few of these studies directly compared glyphosate to GBFs, or effects among GBFs. To address these data gaps, we tested glyphosate, glyphosate isopropylamine (IPA), and (aminomethyl)phosphonic acid (AMPA, a microbial metabolite of glyphosate), 9 high-use agricultural GBFs, 4 residential-use GBFs, and additional herbicides (metolachlor, mesotrione, and diquat dibromide) present in some of the GBFs in bacterial mutagenicity tests, and in human TK6 cells using a micronucleus assay and a multiplexed DNA damage assay. Our results showed no genotoxicity or notable cytotoxicity for glyphosate or AMPA at concentrations up to 10 mM, while all GBFs and herbicides other than glyphosate were cytotoxic, and some showed genotoxic activity. An in vitro to in vivo extrapolation of results for glyphosate suggests that it is of low toxicological concern for humans. In conclusion, these results demonstrate a lack of genotoxicity for glyphosate, consistent with observa-tions in the NTP in vivo study, and suggest that toxicity associated with GBFs may be related to other components of these formulations.

and to gain insight into mode-of-action (MoA). Chromosomal damage was evaluated in human B-lymphoblastoid TK6 cells using the Multi-Flow DNA Damage Assay, which uses a multiplexed biomarker system to identify whether genotoxicants exhibit signatures of clastogenic or aneugenic activity, and an assay for detection of micronuclei (MN), which can arise from chromosome breaks or changes in chromosome number. A third in vitro assay, the bacterial reverse mutation (Ames) assay, was used to assess test articles for the potential to induce gene mutations. Lastly, an in vitro to in vivo extrapolation (IVIVE) was conducted to relate the top concentration of glyphosate tested in TK6 cells to estimates of human exposure.

| Test articles
Chemicals (Table 1) were procured via MRIGlobal (Kansas City, MO), which also confirmed the identity of the chemicals using quantitative nuclear magnetic resonance (glyphosate acid, glyphosate IPA, AMPA, diquat dibromide monohydrate) or high-performance liquid chromatography (HPLC)/ultraviolet-mass spectroscopy (UV-MS) (mesotrione, metolachlor To make stock solutions, all chemicals were weighed to the nearest 0.1 mg and dissolved in distilled water, except for metolachlor, which was dissolved in dimethyl sulfoxide (DMSO) (Sigma-Aldrich, St. Louis, MO).
Dosing solutions were prepared fresh each day of use at concentrations such that the final vehicle volume in the treated cultures was 10% (water vehicle) or 1% (DMSO vehicle). Dosing solutions were also pH-adjusted to 7.0-7.6 using 2.5 N sodium hydroxide (NaOH). Serial dilution was used to prepare the remaining dosing solutions for each assay.
GBFs were mixed in their containers by gentle inversion at least 10 times and samples were drawn from the center of the volume of the container using a serological pipette. Samples were diluted with sterile water, the pH of each solution was adjusted to 7.0-7.6 using 2.5 N NaOH, and this first dilution was adjusted to 1:10. Serial dilutions of the GBFs (1:2 for Ames assays; 1:1.22 to 1:1.41 for the Cleaved PARP Kit, MultiFlow DNA Damage Assay, and the in vitro micronucleus assay) were prepared using sterile water.

| Bacterial mutagenicity assays
Using a testing strategy based on OECD Test Guideline 471 (OECD, 1997), test articles were evaluated for mutagenicity in Salmonella typhimurium tester strains TA100, TA1535, TA97a, TA98, and Escherichia coli WP2 uvrA pKM101 with or without 10% T A B L E 1 Chemicals. phenobarbital/benzoflavone-induced male Sprague Dawley rat liver S9 and co-factors (S9 mix) (Moltox, Boone, NC). After conducting dose-range finding studies, test articles were tested in triplicate using a preincubation protocol (Mortelmans & Zeiger, 2000;Zeiger et al., 1992). The number of revertant colonies was counted using the Sorcerer plate counter and Ames Study Manager software (InStem, Staffordshire, UK). A complete description of the DTT testing protocol for bacterial mutagenicity assays can be accessed at https://ntp.niehs. nih.gov/testing/types/genetic/index.html.
The bacterial mutagenicity assay results were concluded to be positive if a sample induced a reproducible, concentration-related increase in histidine-or tryptophan-independent (revertant) colonies.
Results were concluded to be negative if no increase in revertant colonies was observed. Results that were not concentration-related, not reproducible, or lacked sufficient magnitude to support a determination of mutagenicity were classified as equivocal.

| Cell culture
Human B-lymphoblastoid TK6 cells were obtained from the American Type Culture Collection ® (ATCC ® ) (catalog # CRL-8015) and were determined to be free of mycoplasma contamination using the LookOut ® Mycoplasma PCR Detection Kit (Sigma-Aldrich, St. Louis, MO). Cells were cultured and maintained in RPMI 1640 medium with 10% heat inactivated horse serum plus 1.0% Pluronic™ F-68, 0.5% sodium pyruvate, and antibiotics (penicillin at 20 Units/mL and streptomycin at 20 μg/mL) at 37 ± 1 C in a humidified atmosphere with 6 ± 1% CO 2 in air.
For assays using TK6 cells, the top concentration for chemical exposure was limited to 10 mM based on OECD Test Guideline No. 487 (OECD, 2016). The top concentration for GBFs was limited to a 1:100 dilution, as a 1:10 dilution was necessary for adjusting the pH of the GBFs and an additional 1:10 dilution into the culture was required to keep the final aqueous vehicle volume at 10% of the final culture volume. No changes in osmolality of the highest tested concentration were observed for any of the chemicals or GBFs after 24 h of exposure. None of the test articles caused precipitation at analyzable concentrations after 4 or 24 h of incubation in TK6 cell-based assays.

| Concentration-range finding study
approach (Bryce et al., 2017) to analyze the data. In brief, clastogenic activity is indicated by translocation of p53 to the nucleus in response to DNA damage, and by γH2AX, a marker of DNA double-strand breaks. The phosphorylation of histone H3 occurs with condensation of chromatin in mitosis, making it a marker that is unique to mitotic cells, and the accumulation of cells in mitosis and polyploidy are both indicators of aneugenic activity. The ML approach uses multinomial logistic regression, artificial neural network, and random forest models that were built by Litron Laboratories using JMP Pro software (v13, SAS Institute, Cary, NC) and trained using data generated at ILS from a reference set of 23 chemicals (clastogens, aneugens, and non-genotoxicants) to generate probability scores for clastogenic or aneugenic activity at each test concentration. Outputs from the three models were synthesized into a final ML call using the following criteria: • genotoxic, with evidence for a clastogenic MoA, required two successive concentrations to exhibit clastogen probability scores ≥80%, or one concentration to exhibit a clastogen probability score ≥ 90%; • genotoxic, with evidence for an aneugen MoA, required two successive concentrations to exhibit aneugen probability scores ≥80%, or one concentration to exhibit an aneugen probability score ≥ 90%; and • non-genotoxic was defined as the absence of two successive concentrations exhibiting clastogen or aneugen probability scores ≥80%, and no one concentration exhibiting a clastogen or aneugen probability score ≥ 90%.
For a test article to have a clastogenic and/or aneugenic signature in the ML approach, a majority vote ensemble (2 out of 3 models indicating the same mechanism) was used to synthesize the results of the three ML models. The polyploidy endpoint is not included in the +S9 condition as aneugens typically have been found to not require metabolic activation (Bernacki et al., 2016).
The GEF approach uses cutoff values for significant fold increases for each biomarker. The cutoff values were derived from an interlaboratory training set generated by several laboratories (Bryce et al., 2017). This approach was used to identify responses that are robust but are not recognized as patterns by the ML models. GEFs were also used to evaluate positive control data for quality control.
Clastogenic signatures using the GEF approach were identified by fold increases in two consecutive concentrations that meet or exceed cutoffs for at least two of the following clastogenic responses: • ≥1.51-fold 4 h γH2AX • ≥2.11-fold 24 h γH2AX • ≥1.40-fold 4 h nuclear p53 • ≥1.45-fold 24 h nuclear p53 Aneugenic signatures using the GEF approach were identified by fold increases in two consecutive concentrations that meet or exceed cutoffs for at least two of the following aneugenic responses: • ≥1.71-fold 4 h P-H3 • ≥1.52-fold 24 h P-H3 • ≥5.86-fold 24 h polyploidy • ≥1.45-fold 24 h nuclear p53 The results from the two approaches (ML models and GEFs) were evaluated separately. If either method identified a test article as having a clastogenic or aneugenic signature, the test article was considered to be genotoxic.

| In vitro micronucleus assay
Micronuclei were evaluated using a testing strategy based on OECD Test Guideline 487 (OECD, 2016). Logarithmically growing TK6 cells were plated into 96-well plates at a density of 2.0 ± 0.25 Â 10 5 cells/ mL and exposed to 12 concentrations of test articles (based on con- conducted first to obtain C max following a dose of 1 mg/kg/day. Then the EAD corresponding to 10 mM was calculated using linear extrapolation shown as following: EAD = 10,000 (μM) Â 1 (mg/kg/day)/ C max (μM).
The same approach as described in the Supporting Information file "IVIVE-PBPK Analyses for Herbicides and AMPA.xlxs" was used to conduct human toxicokinetic analyses of glyphosate IPA, AMPA, diquat dibromide, metolachlor, and mesotrione. Similar to glyphosate, AMPA was analyzed using an in vitro concentration of 10 mM, since no cytotoxic or genotoxic effects were observed up to this top concentration that was used for testing. Activity concentration at cutoff (ACC) values based on in vitro MN data was used for glyphosate IPA and the other three herbicides.

| Quality assurance and CEBS database
The data presented in this manuscript underwent quality assur- 3 | RESULTS

| Bacterial mutagenicity assays
Glyphosate, glyphosate IPA, and AMPA were not mutagenic in any of the five strains used in this assay when tested at concentrations up to 6000 μg/plate, ±S9 (Tables 3-5, respectively). Diquat dibromide (Table 6) and metolachlor (Table 7) also were negative in all tester strains, as were the 13 GBFs, when tested to the limit of cytotoxicity, ±S9. Mesotrione, however, was mutagenic in TA100, TA1535, TA97a, and TA98 and non-mutagenic in E. coli, when tested at concentrations up to 6000 μg/plate, ±S9 (

| Concentration-range finding studies with TK6 cells
Test articles were first assessed for cytotoxicity and activation of apoptosis in TK6 cells using the Cleaved PARP Kit to identify concentrations of chemicals and dilutions of GBFs to use in the MultiFlow and micronucleus assays. Because poly (ADP-ribose) polymerase (PARP) is a target of caspases 3 and 7, the presence of cleaved PARP indicates initiation of apoptosis. Cells were exposed to chemicals for 24 h at 20 concentrations ranging from 0.01 to 10 mM. Glyphosate and AMPA were not cytotoxic and did not activate apoptosis-dependent cleavage of PARP at any concentration ( Figure S1). Glyphosate IPA reduced cell survival by about 30% at concentrations of 7 and 10 mM, accompanied by very small increases in cleaved PARP at those concentrations ( Figure S1). In contrast, diquat dibromide, metolachlor, and mesotrione were cytotoxic to TK6 cells, and reduced cell survival was accompanied by concentration-dependent increased cleavage of PARP for all three herbicides ( Figure S2).
GBFs were tested using dilutions that ranged from 0.0000138 (1/72,407) to 0.01 (1/100) (Table S2). In general, loss of cell viability from exposure to GBFs occurred at dilutions ranging from 0.00011 (1/9051) to 0.00125 (1/800) ( Figure S3). Loss of cell viability was  Figure S4).   Note: Data are presented as revertants/plate (mean ± SE) from three plates.  Of the 13 GBFs, Roundup Custom ( Figure 15) and Remuda Full Strength ( Figure 16) were judged to be weakly positive and the remaining 11 GBFs were judged to be equivocal (4), negative (6), or not determined (Halex GT) in the micronucleus assay ( When using the httk.PBTK model for forward dosimetry, a single 7 mg/kg/day daily dose of glyphosate was simulated for 5 days (using "solve_pbtk" function) to reveal a C max of 42.9 μM using fu and Clint predicted from ADMET predictor, and a C max of 51.6 μM using fu and Clint values from OPERA. The EAD corresponding to the 10 mM concentration was calculated assuming a linear relationship between dose and C max . The EAD were 1633.2 and 1357.0 mg/kg/day, respectively.

| In vitro micronucleus assay
The EAD estimated using GastroPlus model were 11.8-18.5-fold higher than those using the httk. also in accordance with OECD guidance. Negative results for glyphosate in the Ames assay were consistent with testing previously conducted by the DTT (Chan & Mahler, 1992). Glyphosate was also negative in a micronucleus assay in which male and female B6C3F1 mice were exposed up to 10,800 or 12,000 mg/kg/day glyphosate, respectively, via feed for 13 weeks (Chan & Mahler, 1992). Glyphosate IPA produced a positive result in the 24-h exposure condition for the micronucleus assay but was identified as non-genotoxic in the MultiFlow assay when tested at the same concentrations. Despite reaching statistical criteria for a positive response in the micronucleus assay, increases in micronuclei occurred at concentrations ($4 to 5 mM) that approached the 10 mM testing limit recommended by the OECD (OECD, 2016). It may be possible that the positive result for glyphosate IPA in the micronucleus assay was due to IPA, which has not been tested for genotoxicity, or due to IPA improving the solubility of glyphosate, as the solubility of glyphosate IPA in water is 100-fold greater than that of glyphosate. It should be noted that several GBFs that contained glyphosate IPA were not identified as genotoxic in any of the in vitro assays used in this study. Some GBFs exhibited genotoxic activity, and herbicides other than glyphosate were active in at least one assay. A summary of the results for all of the test articles is provided in Table 10.
Estimates for human exposure to glyphosate range from 0.47 to 7 mg/kg/day (USEPA, 2017). The highest estimate, 7 mg/kg/day, was F I G U R E 4 MultiFlow DNA Damage assay results for diquat dibromide in the absence (a, b) or presence of S9 (c, d) are shown in radar charts, accompanied by relative cell survival curves. Each radar chart shows the fold increase over the vehicle control for each biomarker and time point for the five highest consecutive concentrations before meeting cytotoxicity exclusion criteria. Concentrations flagged for genotoxic characteristics: green circles = random forest algorithm, stars = random forest, neural network, and logistic regression algorithms; gray circles = concentrations that were excluded from analysis due to cytotoxicity (b). Concentrations flagged for genotoxic activity: red circles = neural network algorithm, two-tone squares = random forest and neural network algorithms, two-tone circles = neural network and logistic regression algorithms, stars = random forest, neural network, and logistic regression algorithms; gray circles = concentrations that were excluded from analysis due to cytotoxicity (d).
calculated for occupational handlers who mix and load GBFs without the use of personal protective equipment, and this calculation also uses the maximum application rate for high acreage agricultural crops.
An exposure level of 7 mg/kg/day translates to a C max of 0.0067-0.01 mM or 0.043-0.052 mM for a 70 kg male human, as calculated using GastroPlus software or the EPA HTTK R-package, respectively.
The estimates calculated by these approaches differ by $8-fold due in part to a calculated absorbed fraction of 36% by the GastroPlus software, which is likely closer to the actual fraction absorbed in humans, given that absorption of glyphosate from feed is $30%-40% in rats (ATSDR, 2020;Chan & Mahler, 1992), whereas a more conservative assumption of 100% bioavailability is made by the HTTK R-package.
Considering both estimates of C max , TK6 cells were exposed to a top concentration of glyphosate that was 192-to 233-fold (HTTK R- IVIVE analyses were conducted to relate the top concentration of 10 mM glyphosate used in our assays with TK6 cells to human exposure. The EAD estimated using the httk.PBTK model was 11.8-to 18.5-fold lower than those using GastroPlus model when using the same set of fu and Clint values, suggesting IVIVE using the httk.PBTK F I G U R E 5 MultiFlow DNA Damage assay results for metolachlor in the absence (a, b) or presence of S9 (c, d) are shown in radar charts, accompanied by relative cell survival curves. Each radar chart shows the fold increase over the vehicle control for each biomarker and time point for the five highest consecutive concentrations before meeting cytotoxicity exclusion criteria. Concentrations flagged for genotoxic activity: green circles = random forest algorithm, stars = random forest, neural network, and logistic regression algorithms, two-tone circles = neural network and logistic regression algorithms; gray circles indicate concentrations that were excluded from analysis due to cytotoxicity (b). Concentrations flagged for genotoxic activity: red circle = neural network algorithm; gray circles = concentrations that were excluded from analysis due to cytotoxicity (d).
model provides a more conservative approach for risk evaluation. AMPA, the major microbial metabolite of glyphosate, was not genotoxic or cytotoxic to TK6 cells when tested at concentrations up to 10 mM and was not mutagenic or cytotoxic in five bacterial tester strains when tested up to 6000 μg/plate, ±S9. The detection of trace amounts of AMPA in the colon and serum of rats exposed to oral administration of glyphosate (Anad on et al., 2009;Brewster et al., 1991), the conversion of glyphosate in a GBF to AMPA in a synthetic porcine microbiome (Fritz-Wallace et al., 2020), and the detection of AMPA in the serum of humans who have been poisoned by ingestion of GBFs (Han et al., 2016;Hori et al., 2003;Motojyuku et al., 2008;Zouaoui et al., 2013)  F I G U R E 6 MultiFlow DNA Damage assay results for mesotrione in the absence (a, b) or presence of S9 (c, d) are shown in radar charts, accompanied by relative cell survival curves. Each radar chart shows the fold increase over the vehicle control for each biomarker and time point for the five highest consecutive concentrations before meeting cytotoxicity exclusion criteria. Concentrations flagged for genotoxic activity: green circles = random forest algorithm; gray circle = concentrations that were excluded from analysis due to cytotoxicity (b).
Mixed results have been reported in the literature for in vitro genotoxicity studies of glyphosate and AMPA. When the existing literature is limited to quality studies with sufficient detail for independent evaluation (Eastmond, 2017), our findings of negative results for glyphosate and AMPA agree with the critical and extensive analyses of the literature conducted by others who have classified these chemicals as lacking genotoxic activity both in vitro and in vivo (ATSDR, 2020;Brusick et al., 2016;FAO & WHO, 2016;USEPA, 2020;Williams et al., 2000). Most in vitro studies of glyphosate, glyphosate salts, and AMPA in the literature were performed using the comet assay, which is a potential limitation of our study for comparison with previous findings. However, whereas the in vitro comet assay is sensitive and useful for hazard identification, it is an indicator test for DNA damage that potentially could be repaired (OECD, 2015). For this study, we tested high concentrations of these chemicals using in vitro assays (micronucleus and Ames) that detect irreversible damage to DNA and have OECD acceptance, which are factors that are given stronger consideration for regulatory decision making (Eastmond, 2017;OECD, 2015).
T A B L E 9 Summary of Multiflow DNA damage assay results.  (Benigni et al., 1979;Levin et al., 1982). Similar to other reports, diquat dibromide was markedly cytotoxic to S. typhimurium and E. coli WP2 strains. Our findings were consistent with the USEPA's Reregistration Eligibility Decision (RED) for diquat dibromide, which reported a positive result in a chromosomal aberration test using human blood lymphocytes and negative results in bacterial mutagenicity tests (USEPA, 1995a). Additionally, the USEPA reported that diquat dibromide was negative in a mouse bone marrow F I G U R E 7 MultiFlow DNA Damage assay results for Roundup Custom in the absence (a, b) or presence of S9 (c, d) are shown in radar charts, accompanied by relative cell survival curves. Each radar chart shows the fold increase over the vehicle control for each biomarker and time point for the five highest consecutive concentrations before meeting cytotoxicity exclusion criteria. Concentrations flagged for genotoxic activity: red circle = neural network algorithm, stars = random forest, neural network, and logistic regression algorithms; gray circles = concentrations that were excluded from analysis due to cytotoxicity (b). Concentrations flagged for genotoxic activity: red circles = neural network algorithm, twotone square = random forest and neural network algorithms (d).
micronucleus test, indicating that the clastogenic effects observed in vitro were not observed in vivo (USEPA, 1995a). Diquat dibromide was classified as "not likely to be carcinogenic to humans" (Group E) by the USEPA due a lack of carcinogenic activity in rats and mice (USEPA, 1995a).
While metolachlor was negative in the Ames test battery, it was identified as a clastogen in the MultiFlow assay and produced a positive result in the 24-h exposure condition for the micronucleus assay at concentrations of 0.16 and 0.20 mM. Metolachlor was classified as "possibly carcinogenic to humans" (Group C) by the USEPA due to increased liver tumors in female rats. However, the USEPA RED attributed this carcinogenic response to a non-genotoxic mechanism as metolachlor was not mutagenic in several assays submitted for regulatory review, including an Ames test, the mouse lymphoma L5178Y Tk +/À assay, a micronucleus assay conducted using Chinese hamsters, and a mouse dominant lethal assay (USEPA, 1995b).
Mesotrione, which has an aromatic nitro group, was mutagenic in tester strains TA97a, TA98, TA100 (all ±S9), and TA1535 (ÀS9), but not in E. coli WP2 (Table 8). The mutagenicity of nitrosubstituted compounds in the Ames assay is often due to activation by bacterial nitroreductases (Josephy et al., 1997;Zenno et al., 1996). It remains to be determined whether the mutagenicity of mesotrione is dependent on nitroreductase activity in S. typhimurium. Although mesotrione was clearly mutagenic in the Ames assay in this study, it was reported in a regulatory submission to the USEPA as negative in strains of S. typhimurium and in E. coli F I G U R E 8 MultiFlow DNA Damage assay results for Halex GT in the absence (a, b) or presence of S9 (c, d) are shown in radar charts, accompanied by relative cell survival curves. Each radar chart shows the fold increase over the vehicle control for each biomarker and time point for the five highest consecutive concentrations before meeting cytotoxicity exclusion criteria. Concentrations flagged for genotoxic activity: green circles = random forest algorithm, two-tone circles = neural network and logistic regression algorithms, star = random forest, neural network, and logistic regression algorithms; gray circles = concentrations that were excluded from analysis due to cytotoxicity (b). Concentrations flagged for genotoxic activity: red circle = neural network algorithm; gray circles = concentrations that were excluded from analysis due to cytotoxicity (d).
when tested up to 5000 μg/plate (±S9). However, it was also reported as negative for gene mutations in the mouse lymphoma L5178Y Tk +/À assay when tested up to 1000 μg/mL ( In general, evaluating the genotoxicity of GBFs was challenging due to the steep cytotoxicity curves produced by these test articles in TK6 cell cultures (e.g., Figures S3, S5, and S9). For the micronucleus assay in particular, despite very close spacing of dilutions, it was difficult to acquire multiple data points that ranged from low levels of cytotoxicity to the cutoff for the assay. Interpretation of results for GBFs were further complicated as many data sets showed a nearly flat concentration-response for %MN until the last, analyzable dilution for the test, which was followed by a dramatic reduction in cell survival for subsequent dilutions. A qualitative comparison of the cleaved PARP data (apoptosis) and 24-h EMApositive events (apoptosis and necrosis) in the micronucleus assay suggests that for GBFs, cell death likely was due to necrosis arising from the cell-membrane disrupting effects of the surfactants and detergents present in formulations. However, three of the GBFs (Roundup Custom, Halex GT, and Hi-Yield KILLZALL II) also clearly induced apoptosis (cleaved PARP assay, Figure S4), suggestive of a mechanistic process for cell death, which is consistent with induction of DNA damage (increased γH2AX) by Roundup Custom T A B L E 1 0 Summary of in vitro genetic toxicity testing for glyphosate, related chemicals, and GBFs.  Figure S12b).

| CONCLUSIONS
Using an in vitro screening approach, which covered chromosomal damage in human TK6 cells with MoA information (clastogenicity versus aneugenicity), and gene mutations in bacterial mutagenicity tests, our data do not support a genotoxic mechanism of action for glyphosate or AMPA. Although some studies suggest that these chemicals may have genotoxic activity, it is not clear how these chemicals, given their lack of structural alerts associated with DNA reactivity (Brusick et al., 2016), could directly interact with or damage DNA, and mechanistic experiments have not been conducted to determine whether these chemicals are capable of interfering with, or overwhelming, DNA repair processes. Our data indicate that the genotoxic and cytotoxic effects of GBFs are not due to glyphosate and may be due to other components of formulations, such as herbicides other than glyphosate that are present in some GBFs, or one or more of the many ingredients in GBFs that are classified as inert. Notably, each of the herbicides other than glyphosate that were present in GBFsdiquat dibromide, metolachlor, and mesotrione-showed genotoxic activity in at least one of our assays, although according to regulatory submissions to the USEPA, all three were reported as negative for induction of micronuclei in vivo, and only metolachor was categorized by the USEPA to be a rodent carcinogen with a non-genotoxic MoA (USEPA, 1995a(USEPA, , 1995b(USEPA, , 2001. In vivo testing of GBFs that showed in vitro genotoxic activity in this study are needed to better understand risk for human exposure (Eastmond, 2017). The cytotoxicity of GBFs is likely due to the presence of surfactants and detergents, which compromise cell membranes leading to necrosis; however, three of the GBFs clearly induced apoptosis, suggesting they contain ingredients that induce controlled, programmed cell death. Lastly, an IVIVE analysis of glyphosate indicated that an adult, 70 kg human would need to ingest large amounts of glyphosate (e.g., 95-115 g glyphosate, or about half a liter of a formulation that is 41% glyphosate) to achieve the same amount of glyphosate in blood as the top concentration of glyphosate (10 mM) that was tested in human TK6 cells and found to be non-genotoxic. Ingesting approximately half a liter of glyphosate formulation is lethal to humans, with toxicity most likely due to the surfactants in glyphosate formulations (Roberts et al., 2010;Tominack et al., 1991). These findings suggest that glyphosate does not pose a genotoxic hazard to humans.