Was it the GMOs or the BPA that did in those rats?

**Updated 9/30/2012, at the end of the post.

It's the study that compromised journalistic integrity, leading some journalists to agree not to obtain outside comment on the paper before an embargo lifted. It's the study that featured shocking images of tumors bursting out all over hapless rats, images reproduced in various stories online in all of their tumoristic, gross morphological horror. It's the study whose authors left themselves open to criticisms from all sides--from science writers and scientists--primarily focused on their strangely lopsided presentation of results--find me the untreated control tumor images in that paper, for example--and their lack of some pretty obvious statistical analyses.

The study in question took a rat strain that's notorious for developing tumors under regular rat-life conditions, fed the rats genetically modified corn, the herbicide Roundup, or genetically modified corn diets possibly laced with Roundup, and evaluated the various groups for tumor burden, liver and kidney outcomes, and mortality. A total of 180 rats received some treatment in their water or diet, while another 20 just lived their regular rat lives eating a regular rat diet. The open-access paper is available here [PDF].

The authors, including anti-GMO activist/scientist Gilles-Eric Seralini, executed statistical analyses that took an almost global hammering on the Web. The special irony here is this paper by Seralini et al., complaining about statistics in toxicology trials Monsanto conducted, particularly related to power analyses and estimations of effect size. They wrote the paper in response to a 2007 expert panel decision rejecting their own analyses of a GM corn as having any effects related to “treatment” with the corn.

In their current paper, the authors conclude that their data demonstrate an effect of a diet containing GM corn, specifically a corn known as NK603, with or without the addition of Roundup, on tumor outcomes and some other endpoints in their two-year rat study. But they may have overlooked some other factors that influenced their results.

Were the diets even different? Maybe not 
For diet to be the culprit here, the diets themselves would have had to be different. One report suggests that the lab chow the authors used itself might have contained GM corn. The authors stated as much in a previous study of NK603 and two other GM corns, observing that the study offered “no data … to demonstrate that the diets fed to the control and reference groups were indeed free of GM feed.” In their current work, they don’t mention this comparison at all; instead, after chemical analyses, they say that “for the different corns and diets, the study of the standard chemical composition revealed no statistical difference” and describe them as having been classified as “substantially equivalent.”

The two factors the authors report as differing among the three diets they describe--regular lab chow, GM chow, or GM+Roundup chow--are caffeic acid and ferulic acid. They assert that these compounds are “not always assayed,” but Monsanto has reported in a peer-reviewed paper on ferulic acid content in NK603 versus non-GM corn and found no difference.

In their discussion, Seralini and colleagues rely on a proposed mechanism for what they found in the rats on ferulic acid in particular. This compound exists in a huge number of plants, including in plants that make up the 84% cereal grain content of the lab chow they used in the untreated rats. Standard lab chows tend to be variable in chemical content. The authors describe having made a laboratory rat chow based on the standard feed they used, except to contain varying percentages of GM corn (that they were aware of) raised with and without Roundup. They also made a “control” chow with non-GM corn at the highest percentage.

When they measured ferulic acid content in their chows, the investigators found that it was lower in their GM and GM+Roundup formulations, with the difference varying by a large range of 16 to 30%. They made up these diets themselves based on a chow with a huge cereal content and chemical variability, so it’s hard to say how relevant these large variations are.

I’m talking about ferulic acid because it’s the compound the authors use to try to explain a proposed mechanism for the presumed effects of the GM corn. They argue that ferulic acid has protective effects against carcinogenesis and mammalian tumors, citing two papers, one that is 28 years old and rather narrow in focus and another from 2010. The 2010 paper reports a positive effect of ferulic acid against chemically induced mammary tumors in Sprague-Dawley rats fed a walloping dose of the stuff. 

They then say that “these phenolic compounds and in particular ferulic acid (sic) may modulate estrogen receptors or the estrogenic pathway in mammalian cells” and cite this paper by Chang et al. Unfortunately for Seralini et al., the Chang paper says that the said “modulation” consists of ferulic acid causing "human breast cancer cell proliferation by up-regulation of HER2 and ERalpha expression.” Based on their own citation, the argument fails that dietary variation in ferulic acid levels--which were substantial in the diets regardless--would have resulted in the results they report. In fact, were GM corn to have lower ferulic acid levels, based on the Chang paper, the GM corn would have been protective.

So what was different if not the diet?
Because I am obsessive, I took the data from this paper and broke it down in the way I’d’ve liked to have seen. This rat strain, as noted, is highly prone to tumors. Each individual rat has an individual propensity to develop tumors, particularly those observed in this study. The authors give us the number of tumors they observed in each group of 10 rats with the number of rats bearing these tumors in parentheses (Table 2 of the paper, which I’ve entered as ratios into the first table given below). Because each individual rat has a high tendency to develop tumors, I think--and feel free to argue this--that the ratio of tumors to number of affected rats in each group might give a better indication of the tumorogenicity of each treatment versus control and clean up the data a little. So, I made those conversions. The second table below contains the original raw data in case the ratios don't suit you.

Then, I did another thing that I wish these authors had done: I compared how these ratios (or raw values) looked within a treatment type (e.g., various doses of GM only, GM and Roundup, various doses of Roundup only). I also looked at between treatments with overlapping variables (e.g., comparing GM11% to GM11%+Roundup), which the authors did for tumors and mortality. These kinds of comparisons should be able to help tease out a little bit which factor--GM or Roundup--might drive differences, if any.

Ratio of tumor/pathology number to rats bearing them for each endpoint. % relates to
%GM corn in diet. R1, R2, and R3=increasing Roundup concentration.

The original data, as presented in the paper (any errors mine). Values are number of observe anomalies per group.
Values in parentheses are number of rats in each group of 10 bearing those anomalies. 
% relates to
%GM corn in diet. R1, R2 (given accidentally as R3 here), and R3=increasing Roundup concentration.
The results are all over the place, whether you look at ratios or absolute number of affected rats or absolute number of tumors or histopathological endpoints. No distinct pattern emerges with GM percentage in the diet. No pattern is there for the alleged presence of Roundup with each diet. I’ve placed the tables showing the graphs at the end of this post for anyone who’s interested. But the graph that interested me--and that reflects the patterns that leapt out at me in Figures 1 and 2 of the paper--is this one below, showing what happens to each parameter with Roundup treatment alone. Remember that these findings reflect ratios of tumor or pathological finding per affected rat, which is most applicable to tumors.

Blue=control; red=lowest Roundup concentration;
green=mid Roundup concentration;
purple=highest Roundup concentration.
The data are messy, but in general, GM vs GM+Roundup seemed to yield quite similar or conflicting results, and control values sometimes overlap or even exceed values from the higher and highest treatment doses. That statement applies whether you view the data as ratios or in the raw data table just above.

But during my first read of this paper, something caught my eye. You can see it in the above figure and in the data in either table, and it also shows up in the mortality and tumor graphs in the paper itself. It’s the inverted U-shaped dose response curve. Was that curve a way to resolve the chaos of these data?

How about those dose-response curves?
As many critiques noted, scarcely any of the data in this paper fit an expected dose response for a carcinogen study--the rates of tumors don’t increase with increasing dose of GM corn or Roundup or both. That kind of linear relationship is typically expected for many toxicology studies--except for those related to endocrine endpoints.

In several cases in this study, at high and low and middle doses of GM corn+Roundup or Roundup, the outcome does not differ from controls. What we do see, and you can see it best in the above figure showing the Roundup treatment data, are inverted U-shaped dose response curves. These curves are classic endocrine response curves, showing a relatively low effect of an endocrine-active compound at high and low doses but a heightened effect at a mid-range dose. Indeed, this kind of curve is almost expected in studies of endocrine-disrupting compounds. The fancy word for these kinds of nonlinear curves, which also can be J-shaped, is hormesis.

Roundup is already known for its endocrine-disrupting capacities, in part thanks to studies like this one from this same author group. But how to explain the scattered nature of the data, including the fact that sometimes, the values for endpoints in the controls exceeded or equaled those for the highest dose of Roundup (a really high dose) or GM corn-possibly-combined-with-Roundup diets?

As I was reading the paper, in addition to noting the hormetic dose-response curves, I noted something else: These rats lived for two years in polycarbonate cages. The protocol does not mention providing fresh cages, but the cages were handled at least twice a week for replacement of litter and contained two rats each. Why do I mention polycarbonate? Because I think it’s possible that one way to explain these whacky-looking data could be the presence of another estrogenic compound, even for controls, that might have influenced outcomes, and that compound is bisphenol A (BPA).

Polycarbonate cages (and possibly water bottles) contain--and leach--bisphenol A
The rats in this long-term study resided in polycarbonate cages for two years. A report from 2003 found that the BPA in these cages can leach out, even at room temperature, even when cages are new. But the leaching kicks up with length of use, with scratches and other marks enhancing the process. This leached BPA affects the animals housed in cages that contain it who also are drinking water from polycarbonate water bottles, according to the 2003 study--and after only one week. The authors concluded that “laboratory animals maintained in polycarbonate and polysulfone cages are exposed to BPA via leaching, with exposure reaching the highest levels in old cages.” 

Another study of an accidental exposure of a female mouse control group to BPA from damaged polycarbonate cages showed that the exposure can result in meiotic abnormalities. One thing that remains unclear about the Seralini protocol is whether or not they used glass or polycarbonate water bottles while the rats also resided in the polycarbonate cages; if the latter, the BPA exposure would have been even greater. This video shows the lab where the rats were kept.

Water bottles or not, rats can take up BPA through the skin. In fact, Sprague-Dawley rats, the kind used in this study, have a skin permeability to BPA that is 12 times that of our own skin. In addition, carcinogenicity studies with BPA show that among the blood-related cancers it might be associated with, it also was associated with … mammary tumors in male rats. This link is of particular interest because, based on a search of relevant terms in PubMed, Roundup doesn’t seem to have been associated with rat mammary tumors. Indeed, the authors point out that it might act to inhibit aromatase, an enzyme that converts androgens to estrogens, and aromatase inhibitors are used to treat breast cancer.

Seralini and co-authors state in their paper that, “As expected, mammary tumors in males occurred far less frequently than in females.” According to a supplier of the rats used in this study, female Sprague-Dawley rats develop exactly the type of mammary tumors seen in this study, with a "high incidence (76%) of mammary gland tumors (predominantly fibroadenomas) (that) resulted in unscheduled sacrifices of many female (s)." Such mammary tumors in these and other male rats, however, appear to occur at a rate of zero. Except when these males are exposed to BPA.
Additional estrogenic disruptor effects from exposing rodents to bisphenol A in these bioassays include an increasing trend for tumors of the mammary glands in male rats (an unusual tumor for males)…. The data for the mammary gland tumors in male rats were 0/50 controls, 0/50 in low dose group, and 4/50 (8%) in the top dose group. …  in my opinion these endocrine tumors should be considered as related to the administration of bisphenol A.
The study cited above involved rats (not Sprague-Dawley) exposed to BPA in their diet for two years. Huff, the author of the linked commentary, concluded, “overall, it appears that BPA exposure via the diet for two years should be considered associated with tumors of the hematopoietic system in rats and mice, and of the testes and of the mammary glands in male rats.”

So, we have probable exposure to compound, BPA, via cages where these rats lived their entire lives. That compound is known specifically to cause mammary tumors in male rats while the treatment in the study is not, and male Sprague-Dawley rats seem to never spontaneously develop these tumors. And we’ve got some pretty obvious hormetic dose response curves that include some relatively high control values, suggesting some underlying endocrine activity. In other words, perhaps we’ve got the wrong three-letter acronym and what we’re seeing here is not the GMOs, it’s the EDCs (endocrine-disrupting compounds).

But that’s not all: What about the soy?
Nothing about the endocrine system is simple, as the curves related to what happens to that system can attest. A part of the diet these rats consumed appears to have contained another exposure to EDCs, in this case plant phytoestrogens, namely those in soy. The diet was about 8% soy and yeast, based on the data from the manufacturer. The concentrations of plant phytoestrogens in rodent diets correlate directly with how much soy they contain, but the concentrations can vary considerably, “and dietary phytoestrogens have the potential to alter results of studies of estrogenicity.” To quote one paper, “Commercial rodent diets are a major source of inadvertent estrogen exposure for laboratory animals.” Diets lacking these compounds are available. 

Questions of additivity or synergy and EDCs
You might be wondering why, if Roundup and BPA and soys and whatever else was present for these rats why the Roundup itself in increasing doses via drinking water didn’t result in rats that were basically one entire tumor at the highest dose. Anyone who’s done research in the field of endocrine disruptors wonders the same, but they can also tell you that these reductive effects with mixtures are more the rule than the exception. My very first study of mixtures taught me that lesson. 

The possible explanations are legion, but with several different kinds of estrogen receptors with different actions in different tissues, compounds that block a receptor at one concentration but activate it at another, compounds that interact with different kinds of hormone receptors in different ways, and differential effects in different species--it’s no wonder the results with mixtures are themselves so mixed. The one thing that doesn’t leap out here as being involved, among a sea of likely possibilities, is the GM corn itself.

The authors were aware of some of this
In their discussion, they note:
As is often the case for hormonal diseases, most observed effects in this study were not proportional to the dose of the treatment (GM maize with and without R application; R alone), non-monotonic, and with a threshold effect.
They don’t mention overlap of many of these endpoints with control results. 

Finally, their findings suggest that sex steroids are also modiļ¬ed in treated rats. That modification implies endocrine-active compounds at work here, and the candidates, to me, at least, are pretty obvious, and they're not corn. A search for endocrine activity related to genetically modified corn turned up one relevant hit, a study that found no effect, in mice using a Bt corn.

Rather than go for the obvious, though (perhaps having overlooked the BPA possibility), the authors sought to construct a fragile argument around one of the two sole differences they found in the GM corn versus the non-GM corn in caffeic and ferulic acid levels. As I noted above, their speculation of GM implication falls apart based on their own citation of a study suggesting that ferulic acid enhances proliferation of breast cancer cells. They also appear to have evaluated not the GM corn itself but the diet containing other ferulic acid sources as a whole for these values, and their finding of a difference in ferulic acid content counters that of a peer-reviewed Monsanto study finding equivalent values in a direct corn-to-corn comparison.

Thus, their efforts to implicate GM corn in their findings in this study rely on a failed speculation. They might have better spent their time considering the various confounders their experimental protocol introduced that could have directly affected their experimental endpoints. At least that might have pulled some signal from all the noise of their data.

ETA: I take a look at the water bottles the group used for their rats for the two-year study here. Please note that anything about BPA is simply speculation, as no one measured BPA exposure, and there's certainly enough wrong with this study design, analysis, and presentation that's right in front of us without having to turn to BPA to explain why their controls seemed to have been as affected in some endpoints as their treatments. I emailed both the contact email for the paper and the supplier about the bottles; neither has responded.

**Finally, in an email exchange with an EDC researcher, I've been told that the BPA exposure routes via polycarbonate cages and/or bottles is of specific concern to BPA-related studies only. I don't understand why that is, given the confounding potential of BPA or any other EDC compound in a study involving EDCs, but that's what this investigator said. All that time we spent looking for glassware instead of polycarbonate products, wasted! Not really. I still argue that in such studies, we need to be as careful as possible to exclude potentially confounding exposures.

More tables graphs, just for fun.

1=controls; 2=11%GM corn in diet; 3=11%GM corn+Roundup
used on corn when grown (concentrations unknown).
Four is a creator error. Values are ratio of tumors
to number of rats bearing tumors.

1=controls; 2=22%GM corn in diet; 3=22%GM corn+Roundup
used on corn when grown (concentrations unknown). 
Values are ratio of  tumors/histo findings
to number of rats bearing tumors

1=controls; 2=33%GM corn in diet; 3=33%GM corn+Roundup
used on corn when grown (concentrations unknown). 
Values are ratio of tumors/histo findings
to number of rats bearing tumors

Comparison of GM percentage in diets and effects on tumors/histopathology.
Blue=control; red=11%; green=22%; purple=33%.
Values are ratio of
 tumors/histo findings
to number of rats bearing tumors.

Comparison of percentage of GM corn exposed to Roundup in diets
and effects on tumors/histopathology. Blue=control; red=11%GM corn+Roundup;
GM corn+Rounduppurple=33% GM corn+Roundup. Values are ratio of tumors/histo findings to number of rats bearing tumors.