Our Stolen Futurea book by Theo Colborn, Dianne Dumanoski, and John Peterson Myers



vom Saal, F and W Welshons. 2006. Large effects from small exposures. II. The importance of positive controls in low-dose research on bisphenol A. Environmental Research 100: 50-76.

Part 1 of this series
More on bisphenol A

A review of BPA's history, basic chemistry and current uses
A review of the prediction that BPA would have biological impacts at low exposures
Low-dose BPA results challenge the assumptions used in chemical risk assessment
Evidence of bias in industry-funded research on BPA

Latest news about bisphenol A

More news about

While chemical manufacturers claim in public statements that "high-quality repeat studies in independent laboratories have not been able to replicate" claims that bisphenol A causes adverse effects at low levels (for example on industry's NoAB319 website), a careful analysis of the 'high quality' studies they reference reveals deep flaws, and that industry-financed research on the health effects of bisphenol A is strongly biased against finding adverse effects. These flaws result from the way the studies have been designed and executed.

Read about industry's AB319 claims

According to vom Saal and Welshon's analysis in this paper, of the 130 studies of low dose effects of BPA published as of summer 2005, 119 have been funded by governments and 11 by industry. 92% of the government funded studies report adverse effects. None of the industry studies do. In this paper, vom Saal and Welshons examine in detail the government and industry funded studies that found no effects, and show several reasons why poor study design destined them to failure.

Not only is industry's own research flawed, but they have repeatedly chosen to ignore all but a handful of the studies on low level effects of bisphenol A that have been published in the peer-reviewed scientific literature. This is a classic ploy, perfected by representatives of tobacco companies, to undermine progress toward better health standards. They focus on a small piece of the overall picture, focus on creating the impression that that piece is flawed, and then argue that because of that flaw the entire body of literature is invalid.


For bisphenol, this tactic requires them to ignore over 100 studies that have been published in the peer reviewed literature by independent scientists and to rely instead upon a small number of demonstrable flawed studies conducted by industry labs.

What did vom Saal and Welshons do?

Section 1: A review of BPA's history, basic chemistry and current uses.

  • The ester bond that links BPA molecules into polycarbonate and resins is subject to hydrolysis, which results in leaching of BPA into water even from new polycarbonate at room temperature. The leaching rate increases over 1000-fold as it begins to show signs of wear. Older studies that reported no leaching used very insensitive methods. Hence the FDA's assertion that "with baby bottles, we haven't been able to detect BPA" is outdated.
  • Until recently, BPA has been considered to be a weak estrogen because in some assays it can be 10,000 to 100,000-fold weaker than estradiol, based on binding affinity to the nuclear estrogen receptor. New research contradicts, this however. BPA can be just as powerful as estradiol, with detectable effects at the lowest doses tested, approximately 0.23 parts per trillion.

Section 2: A review of the original prediction that BPA would have biological impacts at very low levels of exposure, published in 1997.

  • Previous studies had used only high doses, and in light of the relative low binding affinity to the nuclear hormone receptor, this seemed reasonable. But then Nagel et al. discovered that while serum binding proteins bind with most of the estradiol in serum, rendering it biologically inactive, they bind with little of the BPA. This led them to calculate a correction factor for the relative potency of BPA compared to estradiol, and to calculate, based on known effects of estradiol, that 20 µg/kg/day of BPA fed to a pregnant female should stimulate an increase in the prostate size, decrease sperm count and alter other reproductive organs in male offspring. This is in fact what they found.
  • These findings have been replicated by an independent laboratory and by vom Saal's lab.
  • In this section vom Saal and Welshons also introduce the first criticism of one of the failed attempts by industry labs to replicate their findings. This study, by Tyl et al. used a strain of rats known to be insensitive to any exogenous estrogen, including potent estrogenic drugs. Tyl et al. used this strain because it was their standard experimental animal in their laboratory. In reviewing the work, the National Toxicology Program concluded: "animal selection should be based on responsiveness to endocrine active agents of concern, not on convenience and familiarity." Unfortunately, Tyl et al. did not include a positive control in their work. Based on the overall insensitivity of the strain they used, it is likely that the positive control would have shown no reaction, proving the lack of validity of the study. Without data on a positive control, Tyl et al. cannot conclude that bisphenol A does not have low dose activity.

Section 3: Low-dose BPA results challenge the assumptions used in chemical risk assessment.


Findings by vom Saal's lab, and by many others, that low level doses of BPA lead to diverse and permanent changes in reproductive function and behavior are at odds with the current health standards established by the US EPA and FDA.

How was the current BPA standard derived? That current standard, called the reference dose, is set at 50 µg/kg/day. The reference dose is the level considered to be safe for human exposure. It was determined from data published in 1982 showing that 50 mg/kg/day caused weight loss in rodents. That exposure (50 mg/kg/day) was the lowest exposure used in the experiments and was deemed the 'lowest-observed-effect level, or LOEL. No effort was made in those experiments to test lower doses. Hence these experiments never determined a level at which no effect was seen. Instead, according to the Society of the Plastics Industry, it was assumed that the "no effects level" was "probably not far below the LOEL of 50 mg/kg/day. The EPA then applied a 1000-fold safety factor to the LOEL to predict that 50 µg/kg/day would be safe. No attempt was made to verify that experimentally.

The study by Nagel et al. published in 1997 was the first experimental effort to determine whether effects were caused at lower levels. They found effects at levels 25,000 times beneath the 'LOEL' identified in 1982. Other researchers have subsequently reported effects up to 2,000,000 times beneath that level in rodent studies, and even lower in snails and cells.

Clearly, the assumption that 50 mg/kg/day was near the LOEL was erroneous. Yet those data are still the basis for current EPA standards for BPA.

This conclusion is important not just for BPA. Most toxicological studies examine the impact of high-level doses and then assume that beneath the identified 'no effects level' there are no adverse effects. They never examine the effects at low doses, environmentally-relevant doses, because the assumption is that if an effect can't be seen at high doses, then it doesn't happen at low doses. For compounds that interact with hormone receptors, like BPA, this assumption can't be taken for granted. Scientists working with these systems are discovering that low dose effects can be very different from what happens at high doses, and most important, not predictable from those experiments. More...

We have no assurance that many other contaminants don't behave similarly. Many examples have now been published in the scientific literature showing patterns of dose-response curves that fit this model, called 'non-monotonic dose response curves.' Current methods used to establish public health standards assume that dose-response curves are monotonic. This is the assumption that allows, for example, a 10-fold safety factor to be added to calculate a reference dose from the 'lowest-observed-adverse-effect level. The implication is that many compounds will need to be retested, examining the effects of dosages far beneath the current high-dose standards.

According to vom Saal and Welshons: "It is well established in endocrinology that high doses of hormones and hormonally active drugs and chemicals can exert inhibitory effects on processes that are stimulated at much lower doses" resulting in non-monotonic curves. In other words, when given at high levels, chemicals can suppress (or kill) but at low levels they can stimulate.


Section 4: Evidence of bias in industry-funded research on BPA


As of July 2005, there are 40 published studies in animals showing a wide range of adverse effects of BPA at and below doses of 50 µg/kg/day, and 130 examining effects beneath 50 mg/kg/day.

Of those 130, 84% found adverse effects. None of the industry funded studies reported effects. (Table to right, from Vom Saal and Welshons 2006).  
Biases in industry studies of BPA

As of July 2005, there are 40 published studies in animals showing a wide range of adverse effects of BPA at and below doses of 50 µg/kg/day, and 130 examining effects beneath 50 mg/kg/day.

Why do a small number of studies find no effect?

Industry has focused their argument and their research on one specific endpoint, reported by Nagel et al. in 1967, replicated by Gupta in 2000 and extended by Timms et al. in 2005. That endpoint is enlargement of the prostate following exposure to low levels of BPA in the womb.

Vom Saal and Welshons present a detailed analysis of the experiments that failed and find 3 patterns:

1. The experiment didn't fail. One of the studies widely cited by industry was conducted at the Chemical Industry Institute of Toxicology. This study was reviewed for the US EPA by a panel of 36 independent scientists convened by the National Toxicology Program, expressly to review the low-dose issue. Their review was conducted prior to publication of the study. The authors of the study concluded there was no effect. The NTP panel of independent experts, in contrast, reanalyzed the study's data and found that this conclusion was "flawed," "illogical," and "misleading." The NTP statistical review of the original data found adverse effects. Even though these conclusions were made known to the authors prior to publication of the article, the authors ignored the review and kept their original conclusion. The American Chemistry Council in public comments to the EPA about the NTP panel reiterated the finding of 'no effect,' citing this study. According to vom Saal and Welshons "The authors ... thus misrepresented their findings in the published article and the ACC misrepresented the actual findings in their letter to the US EPA."

This pattern of conclusions contradicted by data within the same industry publication has been seen in other chemical controversies, including atrazine.


2. The test system was contaminated with an estrogen. In several of the papers reporting negative results, the control animals have prostates already almost as large as the exposed animals in the studies that found effects. This indicates that the animals used in the experiments had already been exposed to an estrogenic contaminant. With even higher exposures, prostate size begins to decrease, so if animals (control and experimental) in the experiment were contaminated by an estrogenic substance, it would mask the effect of the BPA.

In two of the studies heavily cited by industry, there is strong evidence for this contamination. Both these studies, (Ashby et al. 1999; Cagen et al. 1999) included positive controls in their studies (see column to right). In both these experiments, a known estrogen used for the positive control failed to produce a result.

The graph below compares the 1997 data from Nagel et al. (Missouri) with that of Ashby. Ashby's (negative) controls have prostates as heavy as Missouri's treated animals (BPA and DES). And his positive control (DES) has no effect. Missouri's BPA and DES groups differed significantly from the Missouri's negative control group.

The failure of Asbhy's positive control

adapted from vom Saal and Welshons 2006


What are positive controls?

Positive controls are a crucial element of experimental design. A well-designed experiment testing the impact of a contaminant whose impact is unknown will contain three basic treatment groups

  • Experimental. The animals that receive the contaminant.
  • Negative control. The animals that are treated with only the 'vehicle,' e.g., water, used to deliver the contaminant to the experimental group
  • Positive control. Animals that receive an agent known to cause an effect similar to the one being tested for in the contaminant.

The main comparison is between the experimental group and the negative control. This comparison reveals the effect of treatment.

If, however, there is no difference, the positive control becomes crucial. If the positive control also shows no difference, then the experiment failed, not because the contaminant didn't produce an effect but because the controls were already contaminated by something that made them the same as the positive control group. When the positive control fails, no conclusion can be reached about the effect of the contaminant because the experiment has failed.

More on positive controls...

Despite the failure of the positive control the scientists who conducted the research concluded bisphenol A had no effect, and chemical industry representatives continue to cite these papers as evidence supporting their position.

It is also important to note that the Ashby and Cagen studies were presented by the chemical industry as exact replicas of the Missouri work. That is factually incorrect, as they used different types of feed and these feed types could be the source of the contamination evident above.

A similar difference in controls was evident comparing Ashby et al.'s efforts to replicate a study by Sakaue et al. showing low-dose effects of BPA on sperm production. As with the Missouri comparison, above, Ashby et al.'s controls were similar to the animals treated by Sakaue et al. with bisphenol A, indicating contamination in Ashby's experiments.

3. The experiments used an insensitive strain of animal. Two studies frequently cited by industry (Ema et al. 2001, Tyl et al. 2002) used a strain of inbred rat, the Charles River Sprague-Dawley strain, that has become highly insensitive to any form of estrogen. The experimental design did not include a positive control; without one, it is impossible to conclude whether the lack of a BPA effect was because BPA had no effect, or because the rat strain was unable to respond to any estrogen (see above re positive controls). Hence their conclusion that bisphenol A did not have an effect cannot be accepted. None of the 6 government-funded studies with this rat strain found effects of bisphenol A, either. Nonetheless, the chemical industry holds up the Tyl study as strong evidence that BPA has no effect. vom Saal and Hughes 2005 examine the strain issue in greater depth.


What do vom Saal and Welshons conclude?

To recap, current US government standards are based upon experiments conducted in the 1980s. No 'no-adverse-observed-effect' level was ever determined experimentally during those early experiments; instead it was assumed the NAOEL was near the lowest level used, which had an effect.

Over 100 studies now show adverse effects at levels beneath the 'lowest observed effect' level found in those experiments, some 2,000,000 times beneath that level.

Industry maintains that low dose effects cannot be replicated. It does this by (1) focusing on one experiment showing effects and insisting that it cannot be replicated; (2) ignoring replications of that work in the peer reviewed literature; (3) promoting studies with clear biases and flaws; and (4) maintaining that human exposure are "millions of times lower than levels shown to be safe in experiments" with animals. The wealth of data from many independent laboratories, along with biomonitoring data from the CDC, show clearly that most Americans are exposed to BPA at levels well within the range that is biologically active.

vom Saal and Welshons conclude by stating: "The weight of evidence based on examination of the published studies concerning low-dose effects of BPA in experimental animals demonstrates the need for a re-evaluation of the prior estimate of the acceptable limit of daily human exposure to BPA, which is currently 50 µg/kg/day in the US."






OSF Home
 About this website
Book Basics
  Synopsis & excerpts
  The bottom line
  Key points
  The big challenge
  Chemicals implicated
  The controversy
New Science
  Broad trends
  Basic mechanisms
  Brain & behavior
  Disease resistance
  Human impacts
  Low dose effects
  Mixtures and synergy
  Ubiquity of exposure
  Natural vs. synthetic
  New exposures
  Wildlife impacts
Recent Important    Results
Myths vs. Reality
Useful Links
Important Events
Important Books
Other Sources
Other Languages
About the Authors
Talk to us: email