Glyphosate Vs. Caffeine: Acute and Chronic Toxicity Assessments Explained


[Please consider supporting Food and Farm Discussion Lab with an  ongoing contribution of $1, $2, $3, $5 or $10 a month on Patreon. Or make a one time donation via PayPal. ]pa

GUEST AUTHOR: Alison Bernstein, PhD is a neuroscientist, who studies the role of epigenetics and environmental exposures in Parkinson’s disease. Follow her on Facebook and G+, where she writes as “Mommy PhD”, and on Twitter @mommyphd2.

Article length: 2800 words – 12-15 minute read.

Fear of chemicals in our food has taken on an outsized role in the food discussion. Too much of the debate have become … let’s say – a bit light on the facts and long on emotion. While these are important conversations to have, confusion and misinformation about these issues abound as sound bites are traded as evidence for each “side” and everyone talks past each other.Whether it’s fear of additives in processed food or pesticides used in growing crops, too many people are needlessly wondering if we are slowly poisoning ourselves. This has come to overshadow many useful innovations.

Much of this problem arises from a general misunderstanding of the science of toxicity,.  It’s been especially pronounced in the debate around the herbicide RoundUp (glyphosate).  Here I’d like to take this opportunity to sort out LD50, a measure of toxicity has been thrown around in these discussion more than is helpful. We are going to work through what it tells us and what it doesn’t tell us. Once we’ve covered LD50, the measure of acute toxicity, we’ll sort through the various metrics for assessing chronic toxicity.

What is an LD50?

Let’s get something straight about LD50 – it is a measure of ACUTE toxicity. That is, LD50 is relevant for accidents, murders or suicides.

An LD50, or the median Lethal Dose, and the related LC50 (median lethal concentration, for inhalation rather than ingestion) are measures of acute toxicity only. Acute toxicity relates to adverse effects that occur after a single exposure or multiple exposures within a day, and effects that manifest immediately or within two weeks of the exposure. The LD50 is determined experimentally, usually with rats or mice. It is single acute dose that will kill 50% of a population given that dose. If you have a test population of 100 rats, it is the dose found to be sufficient to kill 50 of them. Likewise, the LD50 for humans is the dosage of a compound estimated that would kill 50 out of 100.

LD50s tell us about risk in cases where someone is exposed to a large amount of a chemical in a short amount of time. In other words: accidents, murders or suicides.

Charts like this get bandied about more often than they should. While it will come as a surprise to most people that it takes less aspirin, Advil or even apple cider vinegar to poison themselves than RoundUp, that’s not really what people are worried about. And acute toxicity is a poor proxy for predicting chronic effects.

Most real human exposures are not acutely lethal but have other, long-term or chronic, effects that may or may not be toxic. Thus, LD50s are not very useful when considering health effects of the large majority of human exposures.

Despite their lack of usefulness in describing chronic toxicity, as noted above, charts comparing LD50s are inevitable in almost every comment thread on the internet about chronic toxicity, lately in regards to glyphosate, the active ingredient in the herbicide Roundup. People are generally not concerned about poisonings – rather they are worried about increased risk of cancer and other long term health risks. LD50 is the wrong measure for discussions of chronic toxicity.

Measures of chronic toxicity

Chronic toxicity relates to adverse effects caused by long term exposures. These are not necessarily measured in lethality, rather in non-lethal yet still adverse effects on health. Effects such as increased risk of disease, changes in body weight, behavioral changes, or reproductive effects. Instead of LD50, the proper measures to compare chronic toxicity are:

•  No Observed Adverse Effect Levels (NOAEL)
• Lowest Observed Adverse Effect Levels (LOAEL)
•  Reference Doses (RfD)

These are the numbers that the EPA uses to set tolerances (maximum residue limits) for all pesticides approved for use in the US on domestic and imported food. Keep in mind, the LOWER the number, the HIGHER the toxicity.

Acute toxicity: adverse effects that occur after a single exposure or multiple exposures within a day, usually within two weeks of the exposure

Chronic toxicity: adverse effects that occur as a result of long term exposure, includes non-lethal but adverse health effects

Adverse health effects: a change in bodily function that may contribute to illness or death

LD50 (Median Lethal Dose): the experimentally determined single acute dose that kills 50% :of a population given that dose

NOAEL (No Observed Adverse Effect Level): the highest level at which no increase in the frequency or severity of adverse effects is observed

LOAEL (Lowest Observed Adverse Effect Level): the lowest dose that has been tested or observed to have an adverse effect

RfD (Reference Dose): an estimate of the daily exposure to humans that is likely to be without appreciable risk of deleterious effects throughout the entire lifetime.

Tolerance: the maximum amount of pesticide residue safely allowed in or on human food or animal feed and is determined with a particular focus on protecting vulnerable populations, such as infants and children. This is typically set at the RfD


The first step in determining the Reference Dose is to identify the NOAEL (No Observed Adverse Effect Level). All existing data regarding the toxicity of the pesticide are reviewed and evaluated to identify the NOAEL. Sometimes, data is incomplete and a NOAEL has not been established experimentally or epidemiologically. In such a case, the LOAEL is used.  

Calculating LD50 means only one of you is going home. Photo by Jason Snyder | Flickr | Creative Commons license 2.0

If LOAEL is used, the second step is an extrapolation to estimate the NOAEL.The extrapolation step is based on the toxicity of the pesticide and its breakdown products, how much and how often the pesticide is applied, and how much of the pesticide remains in the food by the time it is marketed and prepared.

Next, the NOAEL and LOAEL are divided by uncertainty factors (UF) to determine the reference dose. The UFs are not arbitrary and are deliberately chosen based on the strength and quality of the data. Projecting the health impacts for humans based on animal studies creates some uncertainty. That uncertainty is accounted for by taking the dose observed to produce an adverse effect in the animal model and dividing it further by an appropriate factor. Other variables that insert uncertainty are accounted for through multiplication of UFs rather than addition in order to arrive at a cautious estimate based on the available evidence. When taking account of two 10-fold UFs, they are multiplied 10 x 10 to produce a total UF of 100.  Factors considered in calculated UFs include:

Variability within the human population: a 10-fold UF is used when extrapolating from studies of prolonged exposure in healthy humans
Extrapolation from animal studies: a 10-fold UF is used when extrapolating from results on experimental animals
Extrapolating from non-chronic data: an additional 10-fold UF is applied when data is for less than chronic exposures

Additional modifying factors (MFs) can be applied based on scientific uncertainties based on factors other than the three factors described above, including any missing data, the number of species tested, the number of subjects, etc. These MFs are at the discretion of the scientists performing the risk assessment and are often the source of differences between tolerances set by different agencies and often arise from different views on risk tolerance.

[Please consider supporting Food and Farm Discussion Lab with an  ongoing contribution of $1, $2, $3, $5 or $10 a month on Patreon. Or make a one time donation via PayPal. ]

Let’s consider this sample calculation from the EPA.

Say that we have a NOAEL established in a 90-day subchronic oral study of 5 mg/kg/day in 250 rats per dose group. That is, rats fed 5 mg per kg per day for 90 show no effect, but a higher dose would show a harmful effect.

Calculate the UF by multiplying: 10 (extrapolating to human from animal) x 10 (accounting for human variability) x 10 (subchronic dosing) = 1,000

Add a modifying factor (0.8) to account for the large number of animals per dose group. Note that this factor makes the UF smaller because a large study enables scientists to identify smaller effects.

UF x MF: 1,000 x 0.8 = 800.

Reference dose = NOAEL/(UF x MF) = 5 mg/kg/day ÷ 800 = 0.006 mg/kg/day

If the scientists observe no effect in rats at 5 mg/kg/day, we calculate a reference dose (RfD) of 0.006 mg/kg/day for humans. That’s the highest dose at which scientists are confident that we would see zero effect on humans for daily consumption over a lifetime. If 5 mg produces no effect in rats, it’s reasonable that taking that amount and dividing by 800 will give us a dose that should pose no problems for humans.

For more details, read about how the EPA assesses risk and calculates RfDs and tolerances.

The final piece of a risk assessment is determining or estimating the level of exposure and comparing that to the RfDs. This is what dictates the kind of regulatory response required for a given chemical. For example, a chemical where the expected exposure of an average person is many orders of magnitude below the RfD would trigger a very different response than for a chemical where exposure is above the RfD.

There are current questions in risk assessment: what level of risk should trigger a regulatory response and if the RfDs are conservative enough (meaning should we divide by larger uncertainty factors), for example.

To illustrate this method and the types of research used to arrive at these conclusions, I’d like to walk through calculations of references doses for caffeine and glyphosate.

Oral RfD for caffeine

This review of the toxicity and exposure data on caffeine, carried out by scientists at Health Canada in 2003, determined the LOAEL of caffeine for various groups of people (Nawrot 2003). The study cites over 700 references to previous studies on the biological effects of caffeine consumption in humans and animals, in chronic and subchronic models, and with a variety of endpoints including general toxicity, cardiovascular effects, effects on bones and calcium homeostasis, mutagenicity, reproductive health and effects on human behavior. When inconsistencies are reported from the body of literature, the review addresses possible sources of those discrepancies and incorporates these inconsistencies into estimating a conservative LOAEL.

The LOAELs reported in this review for various groups of people were:   

Group LOAEL Daily intake
Healthy adults 6 mg/kg/day ~400 mg for a 65 kg/~140 lb person
Pregnant women or women who may become pregnant 4.6 mg/kg/day ~300 mg for a 65 kg/~140 lb person
Children 2.5 mg/kg/day 125 mg of a 25 kg/~50 lb child

In this 2016 paper, written by a scientific consulting firm that assists municipal, private and federal agencies, oral reference doses for caffeine were calculated based on the above LOAELs (Sorell 2016). The RfDs were calculated by dividing the lowest LOAEL (for kids) by a UF of 1000. The UFs included were: 10 to extrapolate from acute to chronic, 10 to extrapolate from subchronic to chronic and 10 to account for variability in sensitivity within the population.

2.5 ÷ 1000 = 0.0025 mg/kg/day

How much is 0.0025 milligrams/kilogram/day? This is 2.5 MICROgrams of caffeine per kilogram of your own body weight, daily. What does this mean in real life?

If you drink a tall coffee from Starbucks, you take in about 260 mg of caffeine. If you are brewing coffee at home, it might be as low as 50 mg. A good rule of thumb is 100 mg of caffeine per 8 ounces (a small at most places is 12 ounces though). You can look up how much caffeine is in your favorite drink on this Mayo Clinic website.

I typically have 2 cups of coffee day. If the kids woke up in the middle of the night, I have a deadline at work, or I spent too much time in a discussion on my Facebook page, that number will be more. More than I’d care to admit. This means that per day, I am taking in at least 100 mg of caffeine. I weigh about 60 kg. Using the most conservative RfD, my daily intake should be no higher than 0.0025 x 60 = 0.15 mg.

My actual intake of 100 mg is at least 666 times higher than the most conservative RfD estimate. Even if I use the least conservative number available, the LOAEL for “healthy adults”, my intake is still at least 16 times higher. This is not an atypical level of caffeine exposure, yet most of us are not concerned about this level of exposure. I, and most people I know, consider this a tolerable risk.

Oral RfD for glyphosate

RfDs (analogous to ADIs for acceptable daily intake in the EU) for glyphosate have been established in 2013 by the Environmental Protection Agency (EPA) in the US and in 2015 by European Food Safety Authority (EFSA) in the EU. All data used for the EFSA assessment has been made publicly available.

The paper, entitled “Conclusion on the peer review of the pesticide risk assessment of the active substance glyphosate”, cites over 30 other summaries and reports developed by EFSA, other EU regulatory agencies, the International Agency for Research on Cancer (IARC) and UN committees. The assessment also considered studies submitted by the applicants as is required by law, as well as all published studies for a total of over 6000 pages of documents. The final paper even points out that: “The glyphosate dossier consists of an exceptionally large database, therefore the toxicological evaluation adopted by the RMS (rapporteur Member State) and agreed during the peer review rely on a magnitude of valid studies rather than on one ‘key study’ for each endpoint.”

Basing these decisions on many studies, as was described for caffeine, helps to provide confidence in the calculations, especially when the studies are in agreement. Such a large body of research also helps identify where gaps in knowledge exist and this review found specific gaps for certain ecological measures, but not for human toxicity.


We cannot discuss this process without discussing the elephant in the room. This dossier on glyphosate does include studies from Monsanto. There are a few important points regarding this. Funding by industry and studies carried out by industry have inherent conflicts of interest. However, as I have written about previously, conflicts of interest themselves are not a form of research misconduct.

Conflicts of interests may lead to bias that produces research misconduct. This is why disclosures of conflicts of interest and scrutiny of the data, experimental design and scientific process is so important. We can look at the science itself to see if research misconduct has occurred or if bias has influenced the interpretation of the results. When the committee reviewed the data, they found that the studies from Monsanto are largely in agreement with studies from other sources, demonstrating that, in this case, there does not appear to be any research misconduct.

The EFSA review established an overall long term NOAEL in animals of 100 mg/kg/day for chronic exposures, found no genotoxic potential and no evidence of carcinogenicity. The only evidence for carcinogenicity in animal studies was at doses that exceeded the limit dose of 1000 mg/kg/day (this is a high upper limit of doses considered in risk assessments). *


It bears mentioning here that IARC concluded something different than EFSA and nearly every other scientific agency on this. IARC conducted a hazard assessment rather than a risk assessment. That is, they looked at whether it was possible for glyphosate to cause cancer at some potential level of exposure rather than at real world exposure levels.

This discrepancy has been extensively dissected elsewhere (start here, here and here and the watch this video from the University of Michigan to explain the difference between hazard and risk). In addition, this calculation is just one aspect of the questions that the public has regarding glyphosate. Many of these questions about glyphosate have addressed in detail here.

As with caffeine, the RfD is calculated from the lowest determined NOAEL. For glyphosate, EFSA based the RfD calculation on the lower NOAEL for maternal and developmental toxicity. Here, the NOAEL of 50 mg/kg/day is divided by an uncertainty factor of 100 to calculate an RfD of 0.5 mg/kg/day**. As a comparison, the EPA has set an RfD of 0.1 mg/kg/day.

We can do the same type of calculation we did above for caffeine to see how this RfD relates to actual exposure data. Agricultural exposures are the highest real world exposures that arise from normal use (i.e. not accidents, murders or suicides).

If we look at the Farm Family Exposure Study data summary (part of the Agricultural Health Study), 60% of pesticide applicators in the study had detectable levels of glyphosate in their urine and the average urine level for was 3.2 parts per billion (ppb). Average urine levels for spouses and children of applicators was less than 1 part per billion with only 4% and 12%, respectively, of each group having detectable exposures.

Actual exposure levels were estimated from these urine levels in this peer reviewed paper. The average exposure of 3.2 ppb for applicators corresponds to a dose of 0.1% of the RfD. Even the highest urine level of 223 ppb reported in the Farm Family Exposure Study corresponds to be only 4% of the RfD. This is equivalent to 0.004 mg/kg/day rather than the RfD of 0.1 mg/kg/day for the most highly exposed individual who didn’t take appropriate safety precautions.

Other reports, such as this one, have measured urine levels in people whose only exposures are dietary, in an effort to gauge dietary exposure levels in non-pesticide applicators. The average urine levels in this study were 0.2 ppb, with a maximum level of 1.82 ppb. Despite methodological concerns and the fact that this report has not been published in a peer-reviewed scientific journal, the levels can be informative. Using the highest level detected, the estimated dietary exposure is only 0.1% of the EFSA RfD (equivalent to 0.5% of the EPA RfD).

Using even the highest estimates of exposures, pesticide applicators are exposed to levels of glyphosate that are a very small percentage of the safe limits. Consumers are exposed to much lower levels. Measures across other studies of agriculture and dietary exposures were consistent with these results.


We started this discussion with the premise that RfDs, not LD50s, are the appropriate comparison for chronic toxicity. Using RfDs, how do caffeine and glyphosate compare? Keeping in mind, the LOWER the number, the HIGHER the toxicity. 

EPA RfD for glyphosate:  0.1 mg/kg/day
EPA RfD for caffeine: 0.0025 mg/kg/day
0.1  ÷ 0.0025 = 40

This means that caffeine is 40 times more toxic than glyphosate. However, this is only a useful number if we know our typical exposures. The exposure numbers above show that we don’t give a second thought to consuming caffeine at levels hundreds of times higher than the oral RfD, but are simultaneously worried about exposures to glyphosate that are 100 times lower than the RfD. In discussions of toxicity, we must use the correct data to back up our points to step outside the cycle of misinformation.

* They also established a NOAEL of 50 mg/kg/day for maternal and developmental toxicity. Glyphosate is readily degraded by soil microbes to aminomethylphosphonic acid or AMPA, so the toxicity of AMPA was also considered. The panel concluded that AMPA has a similar toxicological profile.

** Note: The EFSA uses the term acceptable daily intake, or ADI, instead of RfD, but these are essentially the same thing. Both represent the daily exposure to humans that is likely to be without appreciable risk of deleterious effects throughout the entire lifetime.

ADDENDUM (September 5, 2018)
While I love to see that this piece has been widely read since it was published, I have seen it used to support the very arguments that it was meant to critique. Because of this observation, I’d like to clarify a few points.

• This piece began from a frustration at the confusion over chronic and acute toxicity metrics being used interchangeably. When consumers express concern over chronic toxicity, science advocates often cite acute toxicity metrics. On the other side, when science advocates talk about chronic toxicity, anti-chemical activists challenge them to drink a glass of [insert chemical here], which would be an experiment in acute toxicity, but tells us nothing about chronic toxicity.

• The main purpose of this piece is to define measures of acute and chronic toxicity, explain how those are calculated and used, and explain how regulatory agencies currently use these metrics to determine “safe” levels. (Safe is in quotes because safety and risk are always relative.) It is not an exploration of what level of risk should trigger regulation, which is a related, but separate, issue.

• There are some major differences between glyphosate and caffeine that make comparing their toxicity an apples-to-oranges comparison. The adverse effects for caffeine include things that some would classify as desired effects (you take caffeine to stay awake and thus report sleeplessness). A wider range of very subtle effects are used, which drives these metrics down for caffeine. Generally, when you drink caffeine, you WANT these effects but here they are considered to be toxic effects. These types of effects are not part of the regulatory assessments for pesticides (which mainly consider carcinogenicity and neurodevelopmental effects).

• The purpose of the comparing glyphosate to caffeine in this article is to explore how these toxicity metrics are calculated and we perceive risk differently for chemicals that we are completely comfortable.

• An important take home from this piece is that our risk tolerance is often affected by things unrelated to actual risks. Is exposure to the hazard within our control? Does the risk involve our children? Our perception of risk from things we cannot control and from things that involve our children is greatly exaggerated.

• Another important take home is that toxicity metrics (acute or chronic) are not measures of risk. The are a benchmark of what level of exposure is deemed to be minimally risky. Risk assessment involves comparing actual (or potential) exposures to those benchmarks to determine if a risk exists.

• For more on hazards, risks and risk perception, I’d encourage you to read the Risk in Perspective series that I co-wrote with Iida Ruishalme. You can find it on SciMoms and Thoughtscapism.

[Please consider supporting Food and Farm Discussion Lab with an  ongoing contribution of $1, $2, $3, $5 or $10 a month on Patreon. Or make a one time donation via PayPal. ]


Please consider supporting by ongoing contribution of $1, $2, $3, $5 or $10 a month on Patreon.
Become a patron at Patreon!