Since we can assume that glyphosate use will not go on forever, we need a robust assessment of the risks of the next generation of ​herbicides.

The early summary of Volume 112 appeared in The Lancet Oncology in May. The most striking feature was that laboratory carcinogenicity of glyphosate was referenced to the website for proceedings of a 2004 WHO meeting on pesticides, which linked to a 1991 EPA review of glyphosate data, which quoted results from unpublished 1981 bioassays conducted by a contractor to Monsanto. An earlier EPA review of these data had classified glyphosate as a “Group C” carcinogen. The 1991 re-review flushed this warning back to no evidence. IARC has posted the full glyphosate monograph, which includes additional links to “gray” literature. The references to, and analysis of, data in regulatory submissions raises a concern for my toxicology and risk assessment students at Hunter College. A standard systematic review starts by searching peer-reviewed publications via Medline and similar databases. This is the method I teach for writing an assigned paper. Unfortunately, even a Google Scholar search mostly misses NTP bioassays and IARC monographs. IARC considers certain government documents, especially NTP bioassay reports, which are peer reviewed but not all published in journals, if the documents are publicly available. IARC working groups consider only original research reports with adequate detail, not reviews. “Meta-analyses,” which are based on calculations from individual studies, are considered original research. An IARC monograph includes sections on exposure (usually not controversial), cancer in humans (epidemiology), cancer in laboratory animals, and other relevant data (genetic bioassays and mechanistic studies). For human and laboratory studies, the working group determines whether evidence is sufficient, limited, or inadequate. Inadequate could mean no relevant studies, null studies, or equivocal studies. The overall assessment of carcinogenicity, voted by the working group, is known, probable, possible, or not classifiable. THE GLYPHOSATE MONOGRAPH Glyphosate was classified as “probably” carcinogenic to humans based on “limited” evidence in humans and “sufficient” evidence in laboratory animals, supported by mechanistic data. The working group’s conclusions on human effects were based on the National Cancer Institute’s Agricultural Health Study (57,000 pesticide applicators), multiple other studies, and a pooled meta-analysis of six studies. Excesses were observed most consistently for non-Hodgkin lymphoma: the meta-analysis showed risk ratios between 1.3 and 1.5 with upper confidence limits of 1.65 and 2.0. The working group concluded these various studies were “limited” evidence. (A published 2012 review of epidemiologic data paid for by Monsanto concluded there was “no consistent pattern of positive associations” and no plausible mechanism.) The working group’s conclusions on laboratory animal effects were based on seven studies, of which six were unpublished industry studies submitted to EPA and WHO. Elevations in rare kidney tumors were found in mice and pancreatic tumors in rats. The working group considered these observations “sufficient” evidence. (A published 2015 review of bioassays by an industry coalition included the IARC-reviewed studies plus an additional seven studies not previously reviewed. The industry authors concluded that “there was no evidence of a carcinogenic effect” and no plausible mechanism.) The working group also cited “strong” mechanistic evidence, including observation of chromosomal damage in people in sprayed areas, a large number of experiments showing genotoxicity in mammalian systems, and studies showing oxidative stress. Genotoxicity and oxidative stress are thought to be pathways to carcinogenesis. LESSONS I want to turn from the specifics of glyphosate risk, which are being contested, to broader issues in risk assessment.

1. Back to the title, “Fifty Shades of Gray”: the inclusion of NTP bioassays in IARC reviews is a no-brainer. These reports publish much more detail and data than would be found in a journal article, and are reviewed in a public forum with public comment, rather than the typical two anonymous reviewers who approve most journal articles. I’d call these whiter than white, although not easily found in a Medline-based search. Substantially darker gray were the six or seven management-sponsored bioassay reports on glyphosate submitted to regulatory agencies. These reports were posted to an EPA website but are hard to find if you don’t already know they’re there. Without these data, there would be little to review. For other agents there may be TSCA 8(e) reports and possibly FDA reports that may be important. I find the TSCA database quite opaque. Some of this gray literature may be so hard to access that it fades to black.
2. Chronic bioassays in laboratory animals (rats and mice), especially those claimed to be null, must be scrutinized against the standard of NTP bioassays. Null results should be given little weight from studies of less than two years’ duration, without testing at the maximum tolerated dose, without mortality-adjusted statistics, without historical control data for rare tumors, without statistical analysis of non-neoplastic lesions, and without consistent criteria for levels of evidence for each species-gender experiment. The IARC working group did this in an appropriate manner.
3. For epidemiology, elevated relative risks at a tumor site should be considered evidence for carcinogenicity, even if these risks don’t achieve statistical significance in the individual study. (For glyphosate, results for non-Hodgkin lymphoma did achieve significance.)
4. Management has long demanded assessment of mechanism in carcinogen classification, usually to advance hypotheses that reduce the certainty of risk to people. The monograph working group was appropriately responding to this demand in using data to enhance concerns.
5. IARC classifications are risk communication phrases about the certainty of evidence of the toxic potential of an agent. The glyphosate conclusion of “probably” carcinogenic would equate to the NTP ROC classification of “reasonably anticipated” to be carcinogenic to humans.

MOVING FORWARD The immediate public health need is to assess exposure levels and reassess controls for pesticide applicators and for communities adjacent to those applications. Likely these issues are more salient in the developing world. Ignoring the epidemiological evidence, as does the industry-sponsored review mentioned above, would be “throwing precaution to the winds.” The epidemiological data could be used to estimate the bounds of human risk rates by trying to quantify the semi-quantitative exposure levels. Another priority would be exposure levels in consumer use. These steps should proceed while the debate swirls about risks of glyphosate to the general population and the environment. Since we can assume that glyphosate use will not go on forever, as the weeds evolve resistance, we need a robust assessment of the risks of the next generation of herbicides. For now, we can assume that if a crop is GMO, it’s been treated with glyphosate.