Comments on Seralini et al (2012) and on the criticisms of that paper
Dr Ulrich E Loening, retired Reader in Dept of Zoology, and former Director of Centre for Human Ecology, University of Edinburgh
27 October 2012

The two most frequent and substantive criticisms of Seralini's paper were that the numbers of rats were too small, with too few control rats, and that the high incidence of spontaneous tumours in Sprague Dawley (SD) rats made them not suitable for these long term experiments.

I show here how neither of these criticisms invalidate the results as a whole and suggest some other interpretations.

In some ways, a feeding trial like this one does not need a control sample at all, because SD rats have been studied worldwide for several decades and their health and the appearance of tumours over their lifetimes has been well documented, as quoted in the paper (refs. Chandra et al., 1992 and Brix et al., 2005; one could add an old study that also showed how increased fat in the diet raised the incidence of mammary tumours, (R. K. Davis, G. T. Stevenson and K. A. Busch, Tumor Incidence in Normal Sprague-Dawley Female Rats, Cancer Res 1956;16:194-197.) The main need for an internal control is to confirm normal behaviour of the rats and to check against any unforeseen local anomaly, which could distort the findings or raise unexpected matters.  In fact in these experiments the control rats lived closely as expected, with lifetimes similar to any other SD rats, of 680+/-21 days for males and 757+/-20 days for females.

Of course the controls become important in the older rats when their pathology increases; in other words when the signal-to-noise ratio becomes too low, hence the criticisms. This problem can be largely overcome by limiting the evaluations to periods of high signal-to-noise ratio, which can be done from the published data.  The experimental animals appeared to acquire tumours sooner and in larger numbers than the controls.  The difference is shown in the paper in Figure 2 as the area bounded between the control and experimental lines.  The units for this area can be called "accumulated tumour days".    These can be easily summed from the graphs in Figure 2, to put numbers to otherwise intuitive visualisation. One can then express the results as the ratio of tumour days of the experimental to control rats, for any specified period.  For example, I added the tumour days from Fig 2 FEMALES GMO  (top right) summing together all three feeding doses of 11, 22, and 33% GMO (divided by 3 to apply per 10 rats) since there seems no regular difference between doses.  In this way all 30 female rats are compared to 10 control rats.  The ratios of tumour days of treated to control are:
from 100 to 550 days        3.04    (91)
from 100 to 650 days        2.58    (192)
from 550 to 650 days        1.81    (103)
from 650 to 750 days        1.23    (94)
(the parentheses indicate the total number of tumours)
Of course a ratio of 1 would indicate no difference from control; this is approached in the last 100 days, showing numerically what can be seen in the Fig.2.

The fall in this ratio with time then provides as a measure of internal control, a high ratio suggesting a period of meaningful data. This example shows that while 100-650 days still shows about 2 ½ times more tumours in the experimental than control and so may seem a useful valid period, in fact the last 100 days of this (550-650) already show a much reduced ratio, suggesting to limit significant results to 550 days, rather than 650.  The large ratio of 3.04 up to 550 days suggests that the results are significant.

Of course, a single very early tumour could distort these figures; outliers must be removed - playing tricks with numbers is no substitute for common sense!

I conclude that the criticism of this paper that SD rats get tumours anyway, far from negating the paper, in fact confirms its validity.  Unwittingly, Seralini et al., who did not set out or expect to find tumourogenesis, chose a most appropriate strain of rats. By monitoring tumours over time they created a sensitive assay for some (unknown) biological factor in the rat chow.  This would not have been discovered in any shorter term experiments - there were no effects at 90 days - nor quite possibly by using other strains of rat that are less liable to tumours. Standard trials of potential poisons or carcinogens over 90 days as in internationally agreed protocols are employed for a different question, and comparisons with them are irrelevant to this paper. The difference is of course that standard protocols are designed to be tests of safety. To confirm the absence of disease requires much more stringent conditions and more animals; an infrequent disease might otherwise be missed. In contrast a few diseased animals suffice to confirm a lack of safety.

It is important to stress that the paper is concerned with many other cellular changes and not primarily with tumorigenesis, which appeared as an unexpected finding.  My comments are addressed mainly at the latter, since this has raised so much debate, whereas the organ and cellular findings have raised little comment.

This work confirmed how agreed standard safety tests are too short and missed effects, much as use of only small numbers of animals would be inadequate.

Nevertheless, the results are surprising in many ways and deserve close scrutiny. The two experimental treatments, of Roundup and a GM corn, have nothing in common.  That each, in very low doses, should lead to broadly comparable cellular change and tumours is extraordinary.  One must question, what was it in the feed that has given these results? One critic considered the possibility of mycotoxins; the careful monitored drying of the corn makes this unlikely.  A more extreme question would be: why do SD rats, like humans, acquire tumours frequently in later life? Is there something in normal feed that acts like the two agents in these experiments?  Could one avoid this with a more 'ultimate' control?  The findings could open new approaches to sensitive nutritional and environmental assays of safety.

I propose not only that this work be expanded and checked by replication, but also propose a 'repeat in reverse,' in which the commercial rat chow becomes the experimental treatment, and the control would be a suitably designed diet equivalent to or better than the UK Government's recommended "five-a-day" of fresh fruit and vegetables for humans. Rats are notoriously more resistant to dietary and environmental abuse than humans, but maybe SD rats are suitably sensitive and compare to humans in these respects.  Seralini et al., might have set up an experimental animal model to investigate the "diseases of civilisation."

I add some comment about the criticisms:  I have read many of these, including much of the dismissal by the EFSA, the Science Media Centre’s number of critics and the UK "NHS Choices".  Almost all the comments are generalisations made from the stance of safety tests, like the main one that SD rats get tumours anyway, and they lack technical substance that would apply to this study. A few are clearly technical mistakes from misreading the paper.

None seem to take account of the basic difference between this paper and standard protocols: an absence of disease is difficult to confirm and requires a different set of stricter criteria than confirmation of the presence of disease. A low incidence of disease could be easily missed if too few animals are used. But similarly a slow incidence of disease would be missed if too short a time is allowed.  Seralini et al. show how this is the case with the standard protocols and justifies the criticism of their short duration.

Many of the criticisms of the paper are from distinguished biologists, some of whom are my friends and colleagues.  I find it disturbing to see how rapidly this work was attacked by a blanket of condemnation, much of which, as indicated above, does not apply to the issues at stake. This condemnation may come to be counter-productive for the science of biotechnology. Distrust and disillusion may result, rather than better public understanding and education. Some of the criticism compares in quality with some of the blanket objections from more extreme anti-GMO campaigners, some of whom also do not serve public understanding even when backed by good science.

Of course, as in any other research, someone may yet demonstrate some substantive flaw in the work which could explain the results in other ways.  Until then, it would be wise for the EFSA, the EU and Governments take this finding seriously.

Finally, I declare an interest: I have for nearly 70 years been an amateur grower practising and promoting what we now call "organic" horticulture.  This, plus my career in molecular biology, certainly influences my approach and has sharpened my scientific critical imagination.