Understanding the divergence: glyphosate carcinogenicity assessment
The carcinogenicity of glyphosate has been reviewed by several national and international agencies (Ibrahim 2015). The outcome of the EU assessment, the differences with the IARC evaluation (IARC 2015), and the authors’ views explaining these differences, are summarised below. Additional details are provided in the supporting information.
Human evidence
IARC (2015) offered the most up-to-date review of human epidemiological studies on glyphosate. Positive evidence regarding an association between exposure to glyphosate and non-Hodgkin lymphoma, observed in some case-control studies but not confirmed by cohort studies, was considered sufficient by IARC to conclude on “limited evidence” in humans. Limited evidence is defined as a positive association observed between exposure to the agent and cancer, for which a causal interpretation is considered to be credible, but chance, bias or confounding could not be ruled out with reasonable confidence. This definition was developed by IARC and introduced in the UN-GHS criteria (United Nations 2003) and EU Regulation (EC) No 1272/2008. EFSA re-assessed the same information; the association with non-Hodgkin lymphoma was discussed during an expert meeting. The statistically significant association was considered limited due to low power, lack of consistency, and the view that greater weight should be given to the cohort study for non-rare tumours. Considering causality, the majority of the experts concluded that the epidemiological evidence was very limited, and insufficient for classification. Although the role of the weight attributed to case–control studies versus cohort studies cannot be fully ruled out, the main reason for the divergent views could be the possibility of bias, chance results and confounding effects, as IARC concluded that the limited evidence in humans was supported by sufficient evidence of carcinogenic potential in animals and strong mechanistic evidence for genotoxicity and oxidative stress. As explained below, the EU evaluation used additional evidence regarding animal carcinogenicity and genotoxicity, and reached different conclusions.
Carcinogenicity in animals
Information sources
There is only one published study on the carcinogenicity of the active substance glyphosate in rats (Chruscielska et al. 2000), which showed no significant increase in tumour incidences in any treated group. Two additional published studies on glyphosate formulations, the first one on initiation-promotion in mice (George et al. 2010) and the second one, a study of rats (Seralini et al. 2014) that was retracted and republished creating some controversies (Fagan et al. 2015), were considered inadequate by IARC and EFSA for carcinogenicity assessment (European Food Safety Authority 2012; IARC 2015). Consequently, industry-sponsored studies, required by several jurisdictions worldwide, have constituted the basis for the assessment of animal carcinogenicity by both IARC and EFSA. As expected for a regulatory assessment, EFSA assessed the original study reports. According to their principles, IARC used unpublished studies based on secondary sources, i.e. the information on the studies as published by JMPR (2004) and US-EPA (1993). The time difference, over a decade, between the IARC monograph and the published regulatory assessments must be considered. Five new studies, not assessed by the JMPR and US-EPA, and therefore, not considered by IARC, were considered valid and included in the EU assessment. The IARC assessment is based on the re-assessment of industry-sponsored studies, two in mice and four studies in rats, plus the negative published study in rats. The EU assessment included five additional valid studies, two in mice and three in rats; one mouse study was excluded due to a likely viral infection in the experimental population and one rat study was considered inadequate due to study deficiencies. Table 4 summarises the studies used in the EU assessment; additional information is provided in Table S-2 as supplementary material, with links to the detailed summaries for each study and its assessment as published in the EFSA background document (Germany 2015). Additional information and raw data have been published as supplementary information in a recent industry-sponsored review of glyphosate carcinogenicity (Greim et al. 2015).
Table 4 Review of long-term chronic toxicity and carcinogenicity studies considered during the EU assessment
Assessment of the available evidence
In its weight of evidence, the IARC Working Group considered a statistically significant trend for renal tumours in male mice in one study (study A in Tables 4, 5) and for haemangiosarcoma in the other (study B in Tables 4, 5). No statistically significant increase in tumour incidence in females was observed in these studies. In the weight of evidence in rats, the IARC Working Group considered increases in the incidence of adenomas, with no evidence of progression to carcinomas, in pancreatic islet cells in males (studies E and F in Table 4), hepatic cells in males (study E in Table 4) and thyroid C-cell in females (study E in Table 4). No increase in tumour incidence was observed in three studies (studies G, K and M in Table 4). The EU assessment followed the weight of evidence approach required by the UN-GHS criteria (United Nations 2015) and further clarified in the ECHA guidance (European Chemicals Agency 2015). The statistical significance found in trend analysis in some studies was balanced against the lack of statistical significance in pair-wise comparison tests, lack of consistency in multiple animal studies, slightly increased incidences only at dose levels at or above the Maximum Tolerable Dose (MTD), lack of pre-neoplastic lesions and/or whether the studies fell within the relevant historical control range. A specific comparison of tumour incidences in male CD-1 mice from four carcinogenicity studies (no change in tumour incidence was observed in females) is provided in Table 5, and the detailed scientific assessment and weight of evidence for each tumour type is summarised in Table 6.
Table 5 Summary of selected tumour incidences in male CD-1 mice from four studies with glyphosate and historical control data
Table 6 Summary of the weight of evidence of the EU assessment for the different tumour types
Comparison of both weight of evidence approaches
As indicated by Portier et al. (Portier et al. 2014), individual scientific studies are rarely, if ever, conclusive. In our view, this is particularly relevant when assessing the carcinogenicity potential in humans using animal studies, and supports the need for a consistency check combining all available studies as mandated in the UN-GHS criteria.
In the absence of conclusive human evidence, and despite some views suggesting the need for re-assessing its relevance (Beyer et al. 2011; Marone et al. 2014; Osimitz et al. 2013), rodent long-term toxicity/carcinogenicity studies are used for predicting carcinogenicity in humans (Doktorova et al. 2012). False positives and false negatives should both be considered, weighing the evidence (Lutter et al. 2015; Rhomberg 2015a, b; Rhomberg et al. 2013) and assessing specifically human relevance; and linked to the MTD concept, the relevance of toxicity-induced carcinogenic effects observed in experimental animals only at very high doses. The UN-GHS, and therefore, the EU CLP approach are based on UN harmonised criteria for weighing the evidence from rodent studies. Regulatory (European Chemicals Agency 2015) and non-regulatory (McGregor et al. 2010) guidance is available for weighing the evidence in line with the UN-GHS criteria. Table 7 summarises the assessment of the different UN-GHS Weight of Evidence elements in the EU assessment, and includes a comparison with the weight provided in the IARC evaluation. It should be noted that the authors of this paper did not participate in the IARC assessment, and therefore, the IARC columns are based on the information extracted from the IARC preamble and monograph, and do not reflect the Working Group discussions except when specifically reported in the monograph. The elements detailed in Tables 5, 6 and 7, and used in the EU evaluation, are not only specific components of the regulatory guidance (European Chemicals Agency 2015), but, as described below, are also fully supported by current scientific knowledge on the assessment of animal studies.
Table 7 Summary of the UN-GHS Weight of Evidence (WoE) elements in the EU assessment and comparison with the weight provided in the IARC assessment
Due to the large number of studies, the assessment of chance results is particularly relevant. Dose–response within the study, consistency among similar studies, consistency or justified differences between sexes, and comparison with historical controls, are considered key elements for identifying chance effects. The Bradford Hill guidelines published in 1965 are still considered a reference for assessing causality (Wakeford 2015), and have been included in the IPCS framework and its respective updates (Boobis et al. 2006, 2008; Meek et al. 2014a; Sonich-Mullin et al. 2001). Although the framework focuses on the relevance of the mode of action, dose–response relationships and consistency among studies are also indicated as key elements. The statistical assessment is the first step for assessing the results of the toxicity tests, and has received significant attention from both, regulatory bodies (e.g. OECD guidelines on testing and assessments of chemicals) and academics (Hothorn 2014); nevertheless, the statistical analysis should be considered part of an overall assessment. This is particularly relevant in cases such as glyphosate, where the statistical analysis is inconsistent or inconclusive, with significant differences in the trend, but not in the pair-wise analysis. Lack of consistency at similar doses in the same species and strain and lack of dose–response relationships can be observed for malignant lymphomas in mice (Tables 5, 6) and adenomas in rat (Table 6). Kobayashi et al. (2010) reviewed the grounds for considering statistically significant changes as incidental, observing similar trends for unpublished and peer-reviewed scientific publications. Lack of dose–response is reported as the main justification for disregarding the results as incidental, followed by lack of physiological/toxicological significance of the effects and the comparison with historical controls. These studies support the concern surrounding conclusions that are based only on statistical significance of increased tumour incidences in a particular study, without considerations of the biological relevance of the finding.
Although the concurrent control group is always the most relevant comparator, the use of historical control data, also in combination with background incidental lesions (McInnes and Scudamore 2014), can be essential in cases of equivocal results to detect both, false positive and false negative situations. In addition to best practices (Greim et al. 2003; Keenan et al. 2009), graphical visualisations (Elmore and Peddada 2009) and statistical approaches (Dinse and Peddada 2011; Peddada et al. 2007) have been developed, although direct comparison with the historical control range in the test laboratory around the time of the study is the approach mostly used in the regulatory context, and preferred in the EU assessment. This approach was considered for malignant lymphomas and haemangiosarcomas in mice when the studies reported the historical range for the test laboratory.
Excessive toxicity, for instance toxicity at doses exceeding the MTD, can cause effects such as cell death (necrosis) with associated regenerative hyperplasia, which in turn can lead to tumour development as a secondary effect, unrelated to the intrinsic potential of the substance itself to cause tumours at lower and less toxic doses (European Chemicals Agency 2015; Knight et al. 2006). Also in the assessment of cell proliferation as mode of action for non-genotoxic carcinogens, systemic toxicity and overt cytotoxicity in the target tissue should be avoided (Wood et al. 2015). It has been suggested that almost all chemicals, including those non-genotoxic and without structural alerts for carcinogenicity, would produce statistically significant trends if tested at or above the MTD in a sufficiently large number of animals (Gaylor 2005). Significant trends for tumour induction were observed in two mouse studies but only at very high doses, well above the proposed top dose for carcinogenicity studies (OECD 2012) of 1000 mg/kg bw per day; clear indications of toxicity were observed at these high doses, such as reduced body weight, histopathological changes in the bladder and liver, and other toxic signs; consequently, the tumour induction trends were considered confounding effects due to excessive toxicity.
Mechanistic assessment
The relevance of the mode of action for humans constitutes the basis of the IPCS framework (Boobis et al. 2006, 2008; Meek et al. 2014a; Sonich-Mullin et al. 2001). Mode of action is defined as a biologically plausible series of key events leading to an effect (Sonich-Mullin et al. 2001) and involves interdependent networks of events with feedback loops. Differences in networks between and within human and animal populations account, in part, for interspecies differences and human variability (Meek et al. 2014a). Current approaches explore the applicability of the Adverse Outcome Pathway approach (Collier et al. 2016; Edwards et al. 2016; Zhou 2015) as a framework for linking the initial molecular interactions with the tumour promotion though plausible key events (Becker et al. 2015; Downes and Foster 2015). As the EU evaluation concluded that the incidences were due to chance and bias and the evidence does not indicate that glyphosate is an animal carcinogen, no further assessment of relevance for humans was required.
IARC, with a different focus, not targeted to individual chemicals but to a broad range of agents, has recently developed a new weight of evidence scheme, by extracting the “key characteristics” from the physical/chemical/biological/behavioural agents classified by IARC in category 1 (Smith et al. 2016). These key characteristics are defined as common properties, not to be considered mechanisms of Adverse Outcome Pathways, although are postulated as a method to synthesize information and develop adverse outcome networks. The ten characteristics are the abilities of an agent to: (1) act as an electrophile either directly or after metabolic activation; (2) be genotoxic; (3) alter DNA repair or cause genomic instability; (4) induce epigenetic alterations; (5) induce oxidative stress; (6) induce chronic inflammation; (7) be immunosuppressive; (8) modulate receptor-mediated effects; (9) cause immortalization; and (10) alter cell proliferation, cell death, or nutrient supply. It should be noted that this new approach has been applied to the recent IARC monographs, including the assessment of glyphosate.
Genotoxicity
The EU evaluation considers in vitro genotoxicity tests and in vivo studies performed in mammals, as those are considered to be more relevant for the assessment of the risk to humans (Yauk et al. 2015). Sixteen in vivo studies in somatic cells and two in vivo studies on germ cells were reported on rodents orally treated with dose levels up to 5000 mg/kg bw, or via intraperitoneal injections. All studies were conducted according to internationally validated guidelines; some non-GLP published studies gave negative results, while two non-GLP studies were positive in mice treated intraperitoneally with dose levels in the range of the intraperitoneal LD50 for mice, one study presenting major flaws. No genotoxic effects on germ cells were detected in rats or mice treated orally at dose levels up to 2000 mg/kg bw. The induction of DNA strand breaks observed in mice treated intraperitoneally with doses close to or in excess of the LD50 has been associated to secondary effects of cytotoxicity (JMPR 2006; Kier 2015). Modes of action associated with secondary cytotoxicity should be excluded from the assessment of the intrinsic genotoxicity potential (Bryce et al. 2014; Kitamoto et al. 2015).
IARC combines information on glyphosate and glyphosate-based formulations, compiling studies on humans, other mammals, other vertebrates, invertebrates, and plants. Regarding in vivo mammalian studies, IARC reports positive effects for 5 out of 11 studies; four negative studies on micronucleus formation and dominant lethal mutation reported by JMPR (2006) are not included in the IARC evaluation. Positive effects are described only for intraperitoneal administrations at doses of 300 mg/kg bw. Although these effects had been previously postulated as secondary to (cyto)toxicity (Heydens et al. 2008; JMPR 2006), the role of (cyto)toxicity is not discussed in the IARC monograph. Positive effects are mostly observed in the liver, an organ that is considered inappropriate for assessing in vivo genotoxic effects after intraperitoneal administration (JMPR 2006).
A recent meta-analysis on micronuclei frequency (Ghisi et al. 2016) has confirmed that positive effects are limited to intraperitoneal administrations, and that the response is much higher for glyphosate-based formulations than for the active substance. Cytotoxicity of the surfactants added to the formulations is presented as a plausible explanation, while the cytotoxicity of glyphosate in intraperitoneal administrations at high doses is not discussed. Significant differences are observed for males but not for females, a general difference is reported in the comparison of mammalian and non-mammalian systems, although similar responses are observed for mice and crocodilians (Ghisi et al. 2016).
Non-genotoxic modes of action
Non-genotoxic modes of action for carcinogenicity are assumed for about 9% of IARC classifications (Hernandez et al. 2009) and include endocrine disruption, tumour promotion, tissue-specific toxicity and inflammation, cytotoxicity and immune suppression, inhibition of gap-junction intercellular communications (GJICs), and other mechanisms (Benigni et al. 2013; Hernandez et al. 2009).
In the EU evaluation, the lack of evidence for carcinogenic potential of glyphosate meant that no further thought regarding the mode of action was considered necessary. IARC assessed the “key characteristics of human carcinogens” (Smith et al. 2016), concluding that there is weak evidence for receptor-mediated effects, cell proliferation or death, and immune effects, and strong evidence of oxidative stress.
Role of surfactants and other co-formulants
The EU assessment focuses on glyphosate, aiming to establish the properties of the active substance to be considered in the assessment of each formulation by individual Member States. IARC has a different approach, addressing both glyphosate and its formulations. The potential role of the co-formulants, which differ among formulations, is not assessed; however, the IARC monograph reports a large number of mechanistic studies with negative results for glyphosate but positive results for glyphosate-based formulations, as well as differences between formulations containing similar concentrations of glyphosate, indicating that other ingredients could lead to the effects observed when testing formulations (Coalova et al. 2014; Cox and Surgan 2006). Similar results are observed for other pesticides and particularly for herbicides (Cavas 2011); this is not surprising, as the mode of action leading the herbicidal activity is usually not linked to the toxicological profile in mammals.
Surfactants are frequently used in herbicide formulations, including glyphosate. Polyethoxylated tallowamines are several orders of magnitude more cytotoxic than glyphosate (Mesnage et al. 2013); the mode of action is cell death with inhibition of the mitochondrial succinate dehydrogenase activity and membrane damage leading to necrosis. This mode of action is different from glyphosate, while similar to that observed for glyphosate-based formulations (Benachour and Seralini 2009). These tallowamines also produce oxidative and DNA damage (Nobels et al. 2011), and increase the apoptotic potential of glyphosate (Kim et al. 2013). Other surfactants as well as solvents used in pesticides formulations are cytotoxic and, possibly, genotoxic (Nobels et al. 2011).
The cytotoxicity and potential genotoxicity of other ingredients should be considered before assuming that the effects observed for a formulated product are linked to the active substance. Secondary genotoxic effects produced by cytotoxicity should also be distinguished from true genotoxic potential (Bryce et al. 2014; Kitamoto et al. 2015). In fact, the UN and EU guidance recommends carcinogenicity and genotoxicity studies to be conducted on individual chemicals, limiting testing of mixtures/formulations to cases where synergistic effects are expected (United Nations 2015).
From hazard assessment to public health risk assessment
While IARC focuses exclusively on the hazard identification, regulatory assessments also include the estimation of the toxicological potency of the substance and the setting of toxicological reference values to be used in the human health risk assessment. The toxicological reference values offer quantitative indications of the toxicity of a chemical, indicating the levels of human exposure that, according to the current scientific knowledge, are considered acceptable from a regulatory perspective. The recent EFSA evaluation has changed significantly the toxicological profile of glyphosate, compared to the previous EU assessment (Table 8).
Table 8 Summary of the recent EU toxicological assessment of glyphosate and derivation of reference doses of risk assessment
The Acute Reference Dose (ARfD) and Acceptable Daily Intake (ADI) represent oral doses that should not be exceeded in a single event (or repeated within 24 h) or daily in long term exposures, respectively. The Acceptable Operator Exposure Level (AOEL) represents a systemic daily dose that should not be exceeded in non-dietary exposures. Figure 1 visualises the current and previous EU toxicological reference values for glyphosate, compared with those established for the entire group of herbicides assessed in the EU. The ranking and percentile within the distribution of ca. 150 herbicides assessed in the EU (data extracted from the EU pesticides database http://ec.europa.eu/food/plant/pesticides/eu-pesticides-database/public/?event=homepage&language=EN) gives an indication of the relative toxicity of glyphosate to humans compared to the other herbicides. In contrast with previous evaluations, effects produced after acute exposures were considered relevant, requiring an ARfD and an acute risk assessment (European Food Safety Authority 2015b). The human, animal and mechanistic evidence indicates that glyphosate cannot be considered as a potent DNA reactive tumour-initiating chemical, and that a risk assessment based on threshold toxicological reference values is scientifically valid (SCOEL 2013). The data summarised in Tables 4, 5 and 6 confirms that the proposed reference values (Table 8) provide sufficient protection for all effects observed in the carcinogenicity and long-term toxicity studies, including the trends for tumour induction considered as sufficient evidence by IARC.
Glyphosate has a relative low long-term dietary toxicity, being within the 10% of herbicides with higher ADI. Regarding short-term dietary exposure, the EU assessment proposed an ARfD which ranks glyphosate as slightly more toxic (45th percentile) than the average for herbicides. This new toxicological profile requires the re-assessment of health risks, which had only considered chronic exposure until now (Shao-Wen and Chun-Hong 2015). The need for personal protective equipment for glyphosate applicators is identified in the EFSA Conclusion. The need for an ARfD triggers also new considerations regarding the role of sporadic AOEL exceedance when addressing the risk of short-term inhalation and dermal exposures during application, including bystander and resident exposure in aerial applications, which are standard practice outside the EU in forest (Rolando et al. 2013) and for the control of illegal crops (Benner et al. 2016). Exposure estimations for children entering the area after application (Solomon et al. 2007) are higher than the proposed toxicological threshold.
Regarding residues in food, a comprehensive update of the dietary risk assessment will be performed in the EU, following the decision on the approval of glyphosate, covering all EU uses and the residues expected on imported food. Meanwhile, Niemann et al. (2015) have compiled information on human biomonitoring data, and concluded that current exposures are well below the toxicological references values; exposure of European citizens seems to be lower than that of Americans. To complement these estimations, an indicative consumer exposure assessment based on EU monitoring data for glyphosate residues in food generated by competent authorities in the EU Member States is described below. The assessment covers over 10,000 samples of different types of food analysed for glyphosate residues between 2012 and 2014 (Fig. 2). Member States focussed the control activities for glyphosate mainly on crops relevant for human consumption, where the presence of glyphosate was expected, such as cereals (almost 4000 samples), followed by fruits, vegetables, pulses and oilseeds; it should be noted that only limited information is available on feed products such as soya beans (only nine soya beans samples were analysed). Overall glyphosate was detected in 6.3% of the samples, mostly in cereals (11.7% of the samples analysed contained residues above the Limit of Quantification), but also in lentils, linseed and table grapes, mostly from outside the EU. The legal limits were exceeded in 0.2% of the samples analysed for glyphosate. A very conservative risk assessment screening has been conducted with the EFSA PRIMO model (European Food Safety Authority 2007), using conservative assumptions. Table 9 summarises the residue levels measured in food items which were identified as main contributors in the risk assessment using European food consumption data. The data have been extracted from the EU pesticides residues monitoring programme (European Food Safety Authority 2016). Detailed information is provided in the supporting information.
Table 9 Glyphosate residue levels reported for the food items contributing with over 0.1% of the ADI or 2% of the ARfD in the European consumers’ risk assessment (EFSA 2016)
The acute risk assessment used the maximum reported result. The chronic risk assessment used mean residue concentrations, assuming that residues below the Limit of Quantification (LOQ) actually occurred in concentrations equivalent to the LOQ; considering that over 94% of the samples analysed did not contain residues above the LOQ, this assumption contributes to the conservatism of the estimated exposure. The chronic exposure was well below the ADI (0.5% for unprocessed products and 0.6% of the ADI when processed foods are included). In the acute risk assessment, the highest exposure was calculated for lentils (23.4% of the ARfD), followed by beans (14.6%) and wheat (11.6%). Pending on the on-going EFSA assessment, these estimations further support the conclusion that glyphosate residues in food do not represent a public health concern for European citizens.