LL2-10 — Ch10 Bisphenol A: contested science, divergent safety evaluations#

Report: Late lessons from early warnings: science, precaution, innovation (EEA Report No 1/2013), Part A “Lessons from health hazards”. Report pages: 215–239. Main text, box, figure and tables are on pp. 215–229; references are on pp. 230–239. PDF pages: 217–241. Read: the whole text extract in order, through the last marker (PDF 241 / p. 239). Figure 10.1, Table 10.1, Table 10.2, Box 10.1 and footnote 2 were checked against a second text extraction of the PDF. Page images could not be rendered in this environment, so the bar heights in Figure 10.1 were not checked visually. [Audit: the Figure 10.1 page (PDF 220) was later rendered and checked visually; see the Figure 10.1 notes under 10.4.]

Conventions: “p.” means the printed report page. Material in square brackets marked [Check] or [Verified] is my own addition. It comes from primary sources consulted in this session (mostly PubMed abstracts via NCBI E-utilities, plus the 2003 correspondence in PMC) and is not part of the chapter.

Authors and standpoint#

Authors: Andreas Gies and Ana M. Soto (p. 215). A footnote says the chapter “is based on the scientific opinions of the authors and does not necessarily reflect the opinions or policies of the institutions they are working for” (p. 215).

• Andreas Gies: head of the Department of Environmental Hygiene at the German Federal Environment Agency (Umweltbundesamt, UBA) in Berlin, where he has worked since 1988. He also heads the WHO Collaborating Centre for Air Quality Management and Air Pollution Control at UBA. His research covers human biomonitoring, health effects of environmental pollutants and exposure analysis, and he edits Chemosphere — Environmental Toxicology and Risk Assessment (author biographies, Annex 1, p. 690).
• Ana M. Soto: professor of anatomy and cellular biology at Tufts University School of Medicine, professor of cancer development at the University of Ulster, and a member of the Centre Cavaillès (history and philosophy of science) at the École Normale Supérieure, Paris. Her research covers control of cell proliferation by sex steroids and the effects of endocrine disruptors on organogenesis, cancer and reproduction (p. 698).

Standpoint. Both authors are protagonists in the BPA controversy, on the side that holds low-dose effects to be real. This is visible in the chapter’s own reference list but never flagged in the text:

• Soto co-authored many of the works the chapter relies on:
• the Chapel Hill consensus statement (vom Saal et al., 2007; Soto is one of its 38 signatories). The chapter lists it alongside the regulatory assessments in 10.8, though it states that it “is not a risk assessment in the classical sense” (p. 222);
• Myers et al. (2009), the critique of GLP-based evidence selection behind Box 10.1;
• Vandenberg et al. (2012), the low-dose and non-monotonic dose-response review;
• the Endocrine Society statement (Diamanti-Kandarakis et al., 2009);
• Cabaton et al. (2011), Rubin et al. (2001), Ramos et al. (2001) and Markey et al. (2002);
• Soto and Sonnenschein (2010);
• Sonnenschein et al. (1989), cited as the origin of the term “inverted u-shaped dose-response curves” (p. 217).
• Gies authored Gies (2007), the source Table 10.1 is “modified from” (p. 228) and the review cited for the “at least 46” low-dose studies (p. 221). He is also first author of Gies et al. (2009), the UBA workshop report cited for EFSA’s “unproven” assumption (p. 226). The UBA position of 2010, presented among the national assessments (p. 223), comes from his own agency.

So the chapter is partly first-hand testimony by participants in the scientific and regulatory dispute. That makes it valuable as an insider account of how low-dose researchers and some regulators experienced the conflict. It is not a neutral history.

Evident stance. The chapter argues five things: - (a) low-dose, non-monotonic and developmental effects of BPA are well established in animals; - (b) EFSA’s and the FDA’s reliance on a few industry-sponsored GLP guideline studies, and their exclusion of hundreds of academic studies, “cannot be defensible” (p. 223); - (c) industry influence on assessments and on advisory bodies is documented and plausible; - (d) exposures should be cut precautionarily by ending BPA uses involving close human contact via food or the environment (p. 226); - (e) testing should be structurally separated from producer funding (p. 229).

It is explicitly advocacy as well as analysis.

Panels: none. The chapter summary says “The chapter is followed by a panel analysing the value of animal testing for identifying carcinogens” (p. 215), but no panel follows. Chapter 11 (DDT) begins directly on p. 240. The identical sentence appears in the summary of Chapter 8, Vinyl chloride (p. 179), and the panel it describes is Panel 8.2, “Value of animal testing for identifying carcinogens” by James Huff (pp. 194–196). This is almost certainly an editorial copy-paste error. Consequence: unlike some chapters, Ch10 carries no commentary by industry, a regulator or any other dissenting voice. The only counter-positions are the ones the authors choose to quote.

Section-by-section notes#

Chapter summary (p. 215)#

• Framing. BPA is among “the world’s best-selling chemicals”, used mainly for polycarbonate (baby bottles, electronics, medical devices, food-container coatings). It is an oestrogen mimic that leaches. The summary is more hedged than the body: “Studies have suggested that even exposure to low doses of BPA may cause endocrine disrupting effects”.
• Latency frame. Sensitivity peaks during development and effects appear later, so “at the time when the effects become detectable, the chemical exposure has vanished” (p. 215).
• Aims:
• map rodent and human findings;
• examine evaluation where “industry‑sponsored studies and independent scientific research seem to deviate strongly”;
• propose ways “to uncouple financial interests from scientific research and testing”;
• compare US and EU assessments, showing how “similar evidence is evaluated differently”, which “presents challenges for applying the precautionary principle”.
• Previewed lessons: the limits of monotonic dose-response; BPA is not a “weak oestrogen”; the pharmaceutical interest in SERMs (p. 215).

10.1 The first known endocrine disruptor (p. 216)#

• Calling BPA “emerging” is “somewhat euphemistic”. It was “probably the first synthetic substance known to mimic” oestrogen.
• Dodds and Lawson (work from 1934, published 1936 and 1938), seeking cheap substitutes for natural oestrogen, found it a weak oestrogen in rat tests. “It failed to make a career as a medicine.”
• The same team found the far more potent DES (1938), later a drug with “severe side effects”.
• In 1957 BPA was polymerised with phosgene to make polycarbonate: “At that time everyone thought that plastics, particularly polycarbonate, were significant advances that would improve our lives.”
• [Check: epoxy use, a third of consumption, is undated. Annex 3 of the same report (Swan, p. 730) gives 1930 for oestrogenicity and “introduction into plastic in 1940”, so the report is internally inconsistent on the early chronology.]

10.2 A growing problem (p. 216)#

• Scale: 3.8 Mt global production in 2006 (industry source). EU consumption 1.15 Mt in 2005–06, up 69% in seven years. Uses: polycarbonate 66%, epoxy 33%. Via thermal paper, BPA reaches recycled paper and then food containers.
• Documented ignorance:
• the EU RAR could not identify the use of more than 7,000 t/yr consumed in the EU;
• up to 16 phenolic impurities, about 10,000 t/yr, which “show structural features of oestrogenic chemicals but have never been toxicologically characterised” (Terasaki et al., 2004).
• Leaching: although covalently bound, “the monomer is subsequently released over time”, more so with age, alkalinity and heat. “All polycarbonate items are probably a source of BPA.” There is “a growing stock of material … in our homes and in the environment”, a “potential continuous source”.
• Other sources: dental sealants, can linings, microwave containers, ICU devices.
• Where it turns up: dust, indoor air, hand wipes, food, drinking water.
• Limits and levels: EU specific migration limit of 3 mg/kg (Directive 90/128/EEC). Canned food up to 380 μg/kg; canned drinks up to 4.5 μg/kg.
• Analytical contamination: BPA is “a real nightmare for analytical chemists” because lab plastics contaminate samples. [This bears directly on the later dispute over measured free BPA (p. 224), but the chapter never makes the connection.]

10.3 Identifying the risk was an accident, not the result of a regulatory process (p. 217)#

• Concept history: the Wingspread conference (1991) coined “endocrine disruptor”; the first paper followed (Bason and Colborn, 1992), which “suspected mainly pesticides” as the cause of hormonal effects in wildlife and humans; BPA was added to the list a year later.
• Rediscovery (1993): Krishnan et al. at Stanford traced an unknown oestrogen contaminating their assays to BPA released from autoclaved polycarbonate. [Verified: the labware was flasks used to autoclave water for yeast media, not “cell culture dishes”. The chapter’s own reference title says “released from polycarbonate flasks during autoclaving” (p. 234). BPA’s receptor affinity was about 1:2000 that of estradiol.]
• Institutional failure:
• no government or industry programme had identified BPA “although such programmes had been run in Europe since 1982” [unreferenced];
• European firms with decades of hormone research “did not play a role”;
• “Industry missed a chance to care for their products responsibly.”
• Agenda-setting: workshops in 1995; Our Stolen Future (1996, foreword by Al Gore) put endocrine disruptors on the global agenda; BPA became the leading example in the late 1990s.

10.4 Bisphenol beyond Paracelsus (pp. 217–219)#

• The challenge. “BPA challenged our belief that high doses produce more serious effects than low ones.”
• Inverted-U curves (the chapter cites Sonnenschein et al., 1989, co-authored by Soto, for the phenomenon) are reported for:
• tumours in transgenic mice;
• snail clutch size;
• rat estradiol;
• pituitary calcium influx;
• housefly pupal weight and sex ratio;
• mouse reproduction.
• Figure 10.1 (p. 218), four schematic panels:
• (a) mouse reproductive success after fetal exposure (Cabaton);
• (b) tumours per mouse (Jenkins et al., 2011, up to 2,500 μg/kg);
• (c) prostate cancer cell proliferation (Wetherill et al., 2002, 0.023–23 μg/L);
• (d) liver lipogenic gene expression (Marmugi et al., 2011, 5–5,000 μg/kg).
• Caption: “small doses show small effects, intermediate doses cause the most pronounced effects while high doses cause no change or even a decrease in effect”. Also “Schematic graphs adopted from the papers”.
• [Audit, checked visually on the rendered page. The x-axis labels read: (a) Control, 0.05, 0.25, 25; (b) Control, 2.5, 25, 250, 2 500; (c) Control, 0.023, 0.23, 2.3, 23; (d) Control, 5, 50, 500, 5 000. The earlier note that panel (a)’s doses match Cabaton et al. 2011 (0.025, 0.25, 25) conflicts with the figure’s own “0.05” label. Either the figure mislabels the lowest dose or the external check was wrong. The caption’s inverted-U pattern fits only panels (c) and (d), which peak at the middle dose.
  • Panel (a), approximate pups per dam: control ~82, 0.05 ~67, 0.25 ~88, 25 ~58. That is non-monotonic, but the largest effect is at the highest dose and the middle dose sits above control, the opposite of the caption.
  • Panel (b), tumours per mouse: control ~1.0, 2.5 ~1.6, 25 ~1.5, 250 ~1.0, 2 500 ~1.0. The effect peaks at the lowest dose tested, so “small doses show small effects” does not hold.
  • The figure source cites “Cabaton et al., 2010”; the reference list has Cabaton et al. 2011.]
• Mechanisms offered:
• WHO/IPCS: “no common dose-response mechanism can be expected” (Damstra et al., 2002);
• low-dose stimulation with receptor saturation at high doses (Welshons et al., 2003);
• multiple genes;
• superposition of effects with different dose-response curves.
• Dispute acknowledged (pp. 217–218). The Endocrine Society calls such curves “very common in the action of hormones”, but their existence is “still disputed by some scientists and members of regulatory bodies (Sharpe, 2010)”. Sharpe is the only named critic on this point.
• Calabrese and Baldwin (2001): 668 curves, “37 % … non-monotonic” (p. 218). [Verified: the 668 came from only 195 of 20,285 articles (1%) meeting strict a priori entry criteria, and the 245 counted (37%) met criteria for hormesis. It is not a base rate for toxicology, and Calabrese uses the same concept elsewhere to argue for more permissive regulation.]
• Stakes (pp. 218–219). Extrapolating from high to low doses requires monotonicity. The paradigm “if a high dose of a chemical does not cause harm, then a low dose will not either” fails “at least for physiological responses to endocrine disruptors”. “After 500 years”, Paracelsus applied “in a naive way” does not protect health.
• The low-dose literature (p. 219). Colerangle and Roy (1997) found “low dose” mammary proliferation, with BPA “much more potent than expected from its oestrogen-receptor binding profile”; “Science reacted immediately”; “hundreds of papers” followed. [Verified: their “low” dose was 100 μg/kg/day, above Table 10.1’s whole range and above EFSA’s later TDI. “Low” is relative to the regulatory NOAEL, not to human exposure.] Themes of the literature:
• greatest sensitivity during development;
• delayed effects;
• conventional tests may be insensitive without in-utero dosing and later-life follow-up;
• non-monotonic curves “reflecting feedback mechanisms, receptor saturation, and multiple mechanisms of action”.
• Consensus versus industry (p. 219). “Within the scientific community, there is far-reaching agreement on these concepts and findings.” Industry replies that “no study purporting to show low-dose effects has been replicated in a second lab, despite repeated efforts to do so” (Polycarbonate/BPA Global Group, 2012). The chapter does not engage specific replication attempts.

10.5 The time makes the poison (pp. 219–220)#

• Timing.
• Irreversible developmental effects follow foetal, neonatal or juvenile exposure in animals (Palanza et al., 2008), and effects depend on life stage (Richter et al., 2007).
• In females: cyclicity, reproduction, brain sexual differentiation, behaviour, mammary development and neoplasia (Soto and Sonnenschein, 2010).
• In males: reduced sperm (vom Saal et al., 1998; Okada et al., 2008a).
• Human latency.
• The human extrapolation is conditional: “If this also happens in humans”, effects of foetal or neonatal exposure may appear only at puberty, or, for breast cancer, decades later. Prenatal DES exposure is the precedent (Hoover et al., 2011).
• “At the time when the effects become detectable the chemical exposure has vanished. This makes it extremely difficult to apply epidemiological methods to link exposure to effects in humans.”
• Children have the highest intake and possibly higher internal free BPA (Edginton and Ritter, 2009).
• “dose is only one of the factors that make a poison” (p. 219).
• The “weak oestrogen” misframing. Binding to the classical oestrogen receptors is at least 1,000-fold weaker than estradiol (Kuiper et al., 1998). But BPA also acts via:
• a membrane oestrogen receptor;
• ERR-γ;
• GPR30;
• AhR;
• androgen-receptor antagonism;
• thyroid receptors.

In some systems its potency is “equal or even stronger” than natural hormones (pp. 219–220). - The critics (p. 220). “Influential scientists” (Greim, 2004) judged low-dose findings implausible from binding strength. “Today we know that their expectations were based on inappropriate assumptions.” Receptor-knockout experiments “provide irrefutable evidence” (Soriano et al., 2012). [Verified: islet β-cells at 1 nM BPA; effects absent in ERβ-knockout cells and seen in human islets. Strong evidence in that system; “irrefutable” is overreach. Note also that this evidence shows a low-dose effect working through a classical nuclear oestrogen receptor. It refutes inferring potency from binding affinity; it is not evidence for the non-classical pathways listed just before.] - There is now “widespread agreement” that BPA is endocrine-active with multiple modes of action. In ToxCast it was “one of the most active chemicals tested” (Judson et al., 2010).

Box 10.1 Good Science and Good Laboratory Practice (p. 220)#

• Two quality systems.
• Peer review is “a very rigorous form of quality control” (more than three in four papers rejected at prestigious journals).
• For regulatory testing, “not all results and procedures need to be published and peer‑reviewed”. GLP was created “As a reaction to low quality and fraud”, is tailored to commercial labs, is required by regulators, and “regulates how to conduct protocols and report tests”. But GLP “cannot judge whether a test is appropriate to solve a problem or whether … the relevant outcomes have been studied”, and “does not indicate that good science has been performed” (Myers et al., 2009).
• Assessment. This is the conceptual core: two different quality systems, one auditing process and one judging scientific merit, which regulators may treat as substitutes. The treatment is one-sided. Peer review is idealised: it does not test reproducibility or fitness for quantitative dose-response work, and many BPA papers appeared in specialist journals. What GLP does secure (auditability and complete reporting) gets only brief credit, in the remark that it answered “low quality and fraud”. The Box does not weigh it against peer review’s weaknesses.

10.6 Concern or no concern (p. 220)#

• At introduction: non-monotonic curves and DES foetal effects were unknown; BPA was 1,000–10,000-fold weaker than estradiol; the monomer was assumed not to be released (Biles et al., 1997). Hence “no real concern”. This is a fair guard against hindsight. [Biles et al. 1997 is an FDA migration study, an odd citation for non-release.]
• Divergence: exposure complexity, the toxicological profile “and probably the high economic importance of this substance may have contributed to the fact that” assessments “differ more markedly than for any other chemical” [asserted without comparison; note the hedges “probably” and “may have”], with acceptable doses that “differ by many orders of magnitude”.

10.7 BPA reviews and risk assessments (pp. 220–221)#

• Regulatory basis. BPA is regulated as a food contaminant: by the FDA in the US, by EFSA in the EU. Both “have essentially used … GLP guideline studies as the only source of data”.
• The pivotal studies.
• Tyl et al. (2002, rat) and Tyl et al. (2008, mouse) multigenerational GLP studies showed only non-specific toxicity (“number of live pups per litter”): LOAEL 50 and NOAEL 5 mg/kg bw d.
• “At least in the rat study there were significant effects below this level (Heinze and Chahoud, 2003)”, which the authors judged “not relevant”.
• Their endpoints miss subtle hormonal effects (Myers et al., 2009).
• The chapter fairly notes that Ryan et al. (2010), “a large independent trans-generational study”, also found no low-dose effects (p. 221).
• [Verified, Tyl 2002 abstract: doses 0.001–500 mg/kg/day; systemic NOAEL 5, reproductive NOAEL 50; “no treatment-related effects in the low-dose region (0.001–5 mg/kg/day)” and “no evidence of nonmonotonic dose-response curves”; F2 female AGD “increased at some doses”, judged not relevant. It is one rat study, though the chapter says “two … in rats”. Co-authors include industry scientists: S. Z. Cagen, for example, is listed at Shell Chemical on his 1999 BPA paper.]
• [Background knowledge, not checked: Ryan et al. 2010 (authors including L. E. Gray) is a US EPA study. A government-funded null result complicates the dichotomy in Table 10.2.]
• Low-dose count. “At least 46 peer-reviewed published studies report effects at oral doses of 50 μg/kg bw d or less” (Gies, 2007), which is the EFSA ADI (p. 221).
• ANSES (2011). After quality screening it “confirmed” effects on:
• sperm production;
• ovarian cysts and endometriosis;
• advanced puberty;
• maternal and sexually dimorphic behaviour;
• lipogenesis, “immune behaviour” (sic) and breast development.

This is the chapter’s best answer to the replication critique, though it is not presented that way. - No-effect level unknown. The chapter concedes that “it is still not clear what is a no-effect level” for the most sensitive endpoints and that “Further research is needed”. It speculates that “we may find” effects “in the low or sub- pg/ml range, the same range as estimates of current human exposure”. Sensitive endpoints (mammary, neurobehavioural) are absent from standard tests (p. 221). [Here the pg/ml human estimates are implicitly accepted; contrast p. 224.] - The two controversies. - Are non-GLP studies reliable enough to use? Put the other way: “is the study sponsored by The Society of the Plastics Industry, Inc. (Tyl et al., 2002) and the study of Ryan et al. (2010) so reliable that nearly all other studies can be dismissed?” - Does free BPA ever reach active levels in the body? - The exclusions. EFSA (2010) and the EU RAR (2008) dismissed all low-dose studies for: only one or two doses; few animals; inadequate statistics; inconsistency with other studies. The authors reply that peer-reviewed publication “indicates that the members of the scientific community … do not agree with the criteria chosen by EFSA” (p. 221). [A weak inference. Acceptance for publication does not show that reviewers judged a study fit for quantitative risk assessment.]

10.8 EFSA and EU risk assessments (pp. 221–223)#

• EFSA. Tyl et al. (2002) is pivotal; NOAEL/100 gives 50 μg/kg bw d. (The chapter uses ADI and TDI interchangeably.)
• “Arbitrarily selected data” (p. 222).
• An AGD increase “in males but not in females” was not carried forward (Heinze and Chahoud, 2003).
• AGD is “not a validated endpoint” but marks sexual development; lower AGD predicts poorer semen quality in men (Mendiola et al., 2011). The chapter adds that “independent studies in rodents by Gupta (2000) and Somm (2009)” found similar low-dose AGD changes.
• Tyl (2009) said there were no low-dose effects, which is “not in line with the data presented”.

• [Check, important.] The text says a male increase. Table 10.1 says “anogenital distance in female F2 ↓”. Tyl’s abstract says F2 and F3 males were unaffected and “F2 female AGD was increased at some doses”. The cited source is a letters exchange (EHP 111:A382–3):

  • J. E. Heinze (Environmental Health Research Foundation) argued that Ema et al. (2001) and Tyl et al. (2002) showed no low-dose effects;
  • I. Chahoud replied citing significant falls in F2 paired ovary weight (at 1, 300 and 5,000 μg/kg) that Tyl judged not significant, and Ashby et al.’s (1999) elevated testis and epididymal weights (“an equivocal finding” per Ashby);
  • the exchange never mentions AGD.

  The general point (study authors dismissing significant low-dose differences) is supported. The specific AGD claim as worded is not. - EU RAR (p. 222). Some Nordic countries, against the majority, proposed in a footnote four neurobehavioural studies as pivotal. The lowest effect was 40 μg/kg (Adriani), which gives an authors’-calculated ADI of 0.13 μg/kg, “lower by a factor of 380” than the EU RAR’s. - NTP 2008 (p. 222). - Toxic above 5 mg/kg. Low-dose effects are “difficult to interpret in many cases” but “should not be dismissed”. The low-dose studies “provide limited evidence that human health may be affected and there is some evidence that human health may be at risk at current exposure levels” (the chapter’s summary of NTP). - Used mouse effects at 2.4 μg/kg; calculated no TDI. - The authors calculate 0.008 μg/kg, “a factor of 6 250 lower” than EFSA. - [Both “ADIs” are the authors’ hypotheticals, not official figures. The arithmetic is correct.] - Canada 2008 (p. 222). Low-dose studies suggest rodent neurodevelopmental and behavioural effects, but “the overall weight of evidence was considered limited from the perspective of rigour”. “Nevertheless, taking a precautionary approach”, Canada characterised BPA as a possible risk and listed it as toxic in 2010. The chapter notes that Canada did not say which studies were decisive. This is the clearest case of explicit precaution on evidence the regulator itself called “limited”. - Chapel Hill (2007) (p. 222). Explicitly “not a risk assessment in the classical sense”. 38 scientists, “including most of the leading scientists working on BPA”, say action is warranted when human internal exposure reaches animal-effect levels. This shows “the developing gap between scientific knowledge about BPA and the published opinions of regulatory committees”. [Soto is a signatory.] - FDA (pp. 222–223). - The 2008 draft said the NOAEL gave “an adequate margin of safety”. - A Science Board subcommittee “harshly criticised” it: it rejected excluding non-GLP studies, called for a point of departure at least ten-fold lower, and concluded that the “Margins of Safety defined by FDA as ‘adequate’ are, in fact, inadequate”. The Board adopted this. The authors call it “unprecedented”. - January 2010 (the chapter dates it “10 January 2010”): the FDA had “some concern” about effects on brain, behaviour and prostate in foetuses, infants and young children. The ADI was unchanged, but the FDA “for the first time” acknowledged “the existence and possible importance of investigator-initiated studies”. More than 800 such studies had been published. [Background, not checked: the FDA update is usually dated 15 January 2010.] - EFSA 2010 (p. 223). - EFSA “basically reiterated” its 2006 finding of safety, with “no new compelling non-GLP studies”. More than 800 studies were “each one discarded for not meeting specific guidelines”. - Guideline endpoints (reproduction, organ weights, clinical chemistry, H&E histopathology) have been used “for the past 50 years: before endocrine disruptors were known, before the developmental basis of disease … and before low-dose and non-monotonic dose responses were known”. - It is “remarkable” that agencies ignored “over 800 peer-reviewed studies that showed toxicity … below the level of human exposure”. [Overstatement: compare “at least 46”, p. 221.] - “Certainly there are data gaps”, but excluding every non-guideline study “cannot be defensible”. Evidence should be weighed study by study and overall, “across doses and times and species”. - UBA 2010 (Gies’s agency) (p. 223). “Sufficient grounds for concern”; “need for action”; precautionary restrictions. - ANSES 2011 (p. 223). “Proven effects in animals” and “suspected effects in humans” at doses “significantly lower than the reference doses”, especially in pregnancy and the pre- and post-natal periods. The chapter says this “questions parts of EFSA’s current assessments”. - Joint EFSA–ANSES report (footnote 2, p. 223). The differences arise partly from different stages (ANSES did hazard identification; EFSA did hazard characterisation in 2010 and a full assessment in 2006) and partly from different study-evaluation criteria, e.g. routes of exposure. [The most even-handed explanation in the chapter, and it sits in a footnote.]

10.9 Bisphenol A in human bodies (pp. 223–225)#

• External exposure: “little controversy”.
• EU RAR models: regional 1.49 μg/kg bw d; local 43; adults 0.008–1.5; worst-case young children 11–13.
• From urine: NHANES adults, 95th percentile 0.15–0.22; German children, 95th percentile 0.37, maximum 7 (n = 599).
• Bottle-fed infants 2× breast-fed; NICU infants about 10× children aged 6–11.
• A diet with limited packaging cut urinary BPA by 66% in three days (Rudel et al., 2011a).
• More than 90% is conjugated, even in neonates.
• Conclusion: “particularly highly exposed risk groups in vulnerable life phases” are missing from assessments. Fasting data suggest non-food exposure or accumulation; transdermal routes are also noted (pp. 223–224).
• [Audit correction. Every human intake cited is below 50 μg/kg, but most are not below Table 10.1’s lowest dose (0.2 μg/kg). Only the NHANES adult 95th percentile (0.15–0.22) and the bottom of the modelled adult range (0.008) sit at or below it. The modelled adult upper bound (1.5), regional total (1.49), German children’s 95th percentile (0.37) and maximum (7), worst-case young children (11–13) and the local model (43) all fall inside the table’s 0.2–50 range. The “overlap” claim (p. 225) therefore holds for the lowest-dose animal studies (17 of the 49 rows are at 0.2–2.5 μg/kg) against upper-percentile and worst-case human intakes, not against typical intakes. It also assumes that oral animal doses and human intakes give comparable internal doses, which the chapter argues (Taylor et al., 2011) and EFSA disputed.]
• Internal exposure: “Major differences” (p. 224).
• Measured: 4–6 ng/ml free BPA in maternal blood (Schönfelder et al., 2002a; Padmanabhan et al., 2008).
• Modelled: pharmacokinetic estimates of 0.1–10 pg/ml (Fisher et al., 2011), about 1,000 times lower.
• Völkel et al. (2002) found no free BPA above 2 ng/ml in “nine volunteers” given 5 mg, “still the basis” of EFSA’s view. [Verified: the abstract describes three males and three females, plus four males; limit of detection 10 nM ≈ 2.3 ng/ml.]
• The RAR/EFSA judgement that the high blood levels are “questionable” is met with “Again EFSA ignores consistent results”. [Attribution between RAR and EFSA is muddled.]
• Pharmacokinetic caveats (p. 224). “Caveats in this field are legion”. Similar pharmacokinetics in primates and rodents make rodents suitable models (Taylor et al., 2011). Dermal and sweat routes are unquantified.
• Normative statement: “Risk assessment is only a protocol used by the regulatory community, not science per se. Uncertainty has to be taken into account and has to be quantified” (p. 224). It continues: “The plethora of peer-reviewed research showing low-dose effects indicates that applied test protocols and regulatory procedures are not suitable for assessing endocrine disruptors.”
• Deconjugation. The placenta and other tissues can re-activate conjugated BPA (Ginsberg and Rice, 2009), so EFSA’s reliance on rapid conjugation is “far from being precautionary”. The NTP “recognises the possibility that the published values of free BPA may, in some cases, not accurately represent the ‘true’ concentrations”, but accepts them as “sufficiently reliable” because different analytical methods agree (p. 225).
• [Omission, verified: Teeguarden et al. (2011) is listed in the references (p. 238) but not discussed. In 20 adults on high-BPA diets with hourly serum samples, unconjugated BPA was at or below the limit of detection (≈0.3 ng/ml) in all samples. The authors include CDC and FDA-NCTR scientists. The result contradicts the ng/ml values the chapter favours, and the chapter’s own contamination warning (p. 216) is never applied to them.]

10.10 Spheres of influence (pp. 225–226)#

• The summary claim (p. 225): “a large body of scientific literature obviously indicates deleterious effects in rodents at low doses. The effective doses in these studies overlap the doses of current human intake.” Most authorities say BPA is safe, and “Industries rely on these risk assessments”.
• Market response. “Massive pressure from consumers and politicians” forced US baby-bottle makers and a European aluminium-bottle maker to withdraw products, with costs to image, brand and earnings. This raises the question of earlier industry influence; the tobacco precedent is cited (Grüning et al., 2006).
• The Weinberg Group (p. 225).
• Described as “active for the tobacco industry” and “successfully hired by the BPA industry to influence the European assessment, in particular the classification and labelling (C&L)”.
• The company’s website (accessed 9 August 2005) claimed:
  • five years of work on polycarbonate/BPA;
  • “identification of opponent’s likely arguments, and formation of responses to counter these arguments”;
  • contribution of “its academic and regulatory network”;
  • that the C&L working group chose “the more benign Category 3” over the Rapporteur Member State’s Category 2;
  • ongoing support against “persistent NGO attacks”.
• Category 2 would have meant a skull-and-crossbones label and, under the new legislation, authorisation for every use.
• [A self-promotional claim. It shows intent, not causation. “Every use” simplifies the REACH authorisation route. An archived copy of the page could not be retrieved in this session.]
• Advisory-panel conflicts of interest.
• “Science is vulnerable. It is based on the independence of scientists and of science itself.”
• Nine of 21 EFSA AFC panel members stated “in the conflict of interest statements they supplied to the agency” that they had links to industry or to industry-funded bodies (Greenfacts, ILSI Europe); one was paid by industry to write a review (Dekant and Völkel, 2008). [The chapter names the members’ own declarations as the basis but cites no document or date, and does not name the member. The references include an MEP’s written question on EFSA conflicts over BPA (Breyer, 2009), which is not cited in the text, and a Corporate Europe Observatory report, which is cited only for proposed remedies (p. 229).]
• EFSA has since tightened its independence rules, since “the value of its scientific advice is directly linked to the level of trust held in it by the public” (EFSA, 2012).
• The chapter concedes that “working for the chemical industry and its organisations or other NGOs is a job like any other”, then adds: “Whether it is wise to give people who are directly or indirectly paid by industry the task of controlling industry may be questioned” (p. 225).
• Funding and outcome. Most studies of the question find an association (Lesser et al., 2007; Moses et al., 2005; Blumenthal, 2003). Table 10.2 (p. 228; the table’s source line says “Hughes and vom Saal 2005”, the reference list “vom Saal and Hughes, 2005”; studies published to 2004):

[Verified abstract: 115 in vivo studies to December 2004, 94 positive. The same paper attributes some industry nulls to ignored positive controls and an oestrogen-insensitive rat strain, i.e. to design as well as funding. Caveats: n = 11 industry studies; the compiler is a protagonist; “harm” means any significant effect; academic publication bias is not considered.] - “Doubts … whether EFSA’s decision was unbiased” (pp. 225–226). After EFSA raised the TDI fivefold in 2006, the chapter lists: - at least ten more rodent studies with effects below the TDI; - metabolic and obesity effects (Somm, 2009; Rubin et al., 2001); - children’s biomonitored doses at rodent-effect levels (Betts, 2010); - ICU subgroups; - new sources (pacifiers, warm-water tubes); - human associations: - maternal BPA and daughters’ behaviour at age 2 (Braun et al., 2009) and age 3 (Braun et al., 2011); - IVF oocyte and embryo quality; - workers’ sexual function; - obesity (Carwile et al., 2011); - birth weight (Miao et al., 2011); - EFSA’s human-versus-rodent internal-dose assumption being “unproven” (Gies et al., 2009); - the 2011 EU baby-bottle ban.

The causation caveat is attached only to the Braun et al. (2009) item: “Like other cross-sectional studies, these associations are not a proof of causation but should be regarded as additional warning signs” (p. 226). The other human associations (IVF, workers, obesity, birth weight, Braun 2011) are listed without a caveat, and Braun 2011 is described causally (“affected”). The list also includes “Numerous other in vivo and in vitro studies” without citation. [Braun et al. (2009) is called “cross-sectional” but, verified, is a prospective birth cohort: 249 pairs, prenatal urine, behaviour at two, with an association “only among females”. The list mixes evidence, a self-citation and a regulatory act.]

10.11 Lessons to be learned (pp. 226, 229)#

• The “same old story”: “The competing urgency of public health and economic stakes puts the scientific process under enormous pressure”. The case resembles asbestos, PCBs and DES (p. 226).
• Best science and transparency: “dare to start again with the risk assessment”, “conducted by the scientists authoring the papers with high scientific impact in this field”; stakeholder conferences to expose the interests of “industry and other NGOs” (p. 226).
• Precaution: as an interim step (“Until final decisions are made”), lower exposure “well below” rodent-effect and human-behavioural levels, “terminating those uses of BPA involving close contact with humans via food or the environment” (p. 226).
• Independent science (pp. 226, 229).
• REACH relies on industry data.
• “Independent science and regulatory toxicology seem to speak different languages.” Industry studies “need doses orders of magnitude higher to produce any effects”.
• Academic labs specialise and “may be better qualified” to detect subtle changes; contract labs are “per se not economically independent”.
• “Independent science is interested in finding the effects of a substance and publishing these findings” (p. 229). [A directional incentive the chapter does not treat as a possible bias.]
• Structural remedies (p. 229).
• “Uncoupling of financial interests and scientific and regulatory research and testing seems to be necessary.”
• Labs should be paid by “a fund that is financed by the industry, over which industry has no control and which is managed by governments”.
• Contract-lab results “must not outweigh” academic ones.
• Test procedures should be updated.
• ILSI ties “may be regarded as incompatible” with advisory independence; conflict-of-interest documentation is incomplete.
• Experts should be paid adequately, with costs covered by industry fees; academic workloads should be reduced to allow service.

10.12 Lessons learned (p. 229)#

• Test strategies are “slowly” improving: NTP-CERHR (2007) and ANSES (2011) used both single- and multiple-dose studies (Arnich et al., 2011). This directly answers one of the EFSA/RAR exclusion criteria (“only one or two doses tested”, p. 221), though the chapter does not make the link.
• BPA is not simply a weak oestrogen (as SERM research shows), “although some investigators persist” in calling it one.
• The OECD is adding endpoints (timing of vaginal opening, AGD).
• The chapter ends here, with no conclusion and no panel.

Table 10.1: studies with oral effect levels at or below 50 μg/kg bw d (pp. 227–228)#

• Contents. 49 rows, 0.2–50 μg/kg bw d, in rats, mice and a gerbil. Endpoints are very varied: reproductive organ size and AGD, sperm, mammary development and carcinogenicity, prostate hyperplasia and neoplasia, aneuploidy, immune function, maternal and sexual behaviour, anxiety, memory, metabolism. Source: “Taylor et al., 2008, modified from Gies, 2007”.
• Route note. Studies by other routes “should also not be dismissed” because internal exposures are similar. [That was shown in neonatal mice; route-dependence in adults is exactly what regulators disputed.]
• Observations.
• (a) Two rows are studies whose authors concluded “no effect”: Tyl et al. (2002), flagged “not regarded as relevant by the authors”, and Ashby et al. (1999). [Verified: Ashby et al. is an AstraZeneca study titled “Lack of effects for low dose levels of bisphenol A … on the prostate”, a failed replication of vom Saal/Nagel. The “effect” listed follows Chahoud’s 2003 reading.] [Background, not checked: Ema et al. (2001), the first row, is also a two-generation rat study whose authors reported no low-dose effects. Per the earlier note on the Heinze letter, Heinze cited it alongside Tyl 2002 as negative, so it may be a third such row.] This mirrors the “arbitrarily selected data” charge made against EFSA.
• (b) Rows count studies with some effect across heterogeneous endpoints. Few are independent replications of the same endpoint at the same dose.
• (c) Seven of the ten 40 μg/kg rows (Dessì-Fulgheri, Farabollini, Aloisi, Adriani, Porrini, Della Seta, Ceccarelli) come from overlapping Italian groups who frequently co-publish.
• (d) The text says “at least 46”; the table has 49 rows. Fifteen rows are dated 2007–2011, consistent with the “at least ten” post-2006 studies claimed (p. 226).
• (e) Minor citation mismatches between table and references: “Ryan et al., 2006” is Ryan and Vandenbergh 2006; “Jones et al., 2010” is 2011 in the references; the text’s “Adriani et al., 2005” (p. 222) is the 2003 paper with a 2005 erratum. The table title reads “50 µ/kg” for 50 µg/kg.

Case timeline#

Lag analysis (my calculations; the chapter computes none) - Oestrogenic activity known (1936) to the first EU ban on a use (2011): about 75 years. (This is not the first EU measure: Directive 90/128/EEC set a specific migration limit in 1990, p. 216, and a Category 3 reproductive-toxicant classification, undated in the chapter, preceded 2011.) The 1930s knowledge, though, was of a weak pharmacological activity, not of harm at exposure levels. The chapter’s point is that it was never connected to the decision to put BPA in consumer materials. - Leaching rediscovered (1993) to EU baby bottle ban (2011): 18 years. - First low-dose reports (1997) to FDA “some concern” (2010): 13 years; to the EU bottle ban (2011): 14 years. - In the chapter’s account, the only use restriction is the EU infant-bottle ban (2011), alongside Canada’s toxic listing (2010) and the general EU migration limit for food contact (1990). No measure it reports targets can linings, other food-contact polycarbonate or thermal paper. [Background, not checked: France legislated a wider food-contact ban in December 2012, before publication; the chapter does not mention it.]

What was known when - By 1993 the hazard property (oestrogenicity) and a leaching route were both documented. - From the late 1990s rodent low-dose developmental effects were being reported. The industry-sponsored GLP studies (Tyl et al., 2002, 2008) reported none, and neither did the later “independent” Ryan et al. (2010) (p. 221). - By 2008–2010 several expert bodies (NTP, Canada, the FDA Science Board, UBA, and ANSES in 2011) had moved towards concern while EFSA had not. - Human evidence as of 2013 was associational (cross-sectional and cohort), with no demonstrated causal harm. The chapter says this explicitly only for the Braun et al. (2009) item (“not a proof of causation”, p. 226), not for the other human associations it lists.

Harms and costs - The chapter does not quantify human health harm or economic cost. - The only costs named are reputational and earnings losses for firms forced to withdraw products (p. 225). - Benefits get only a sentence or two on p. 216 (the “plastics revolution”; transparent, low-weight materials). Specific functions such as can-lining protection, and the costs or risks of substitutes, are not discussed.

The authors’ own lessons and conclusions#

Lessons derived from the evidence (analytical) 1. The classic dose paradigm fails for hormone-like agents. Non-monotonic curves undermine high-to-low dose extrapolation; “if a high dose … does not cause harm, then a low dose will not either” does not hold for endocrine disruptors (pp. 218–219, 229). 2. Timing is a dimension of toxicity. “The time makes the poison”: developmental windows, latency, irreversibility; tests without in-utero dosing and later-life follow-up can be blind (pp. 219, 215). 3. Potency inferred from one mechanism misleads. BPA is not simply a “weak oestrogen”; its multiple receptor pathways make receptor-binding potency a poor predictor (pp. 219–220, 229). 4. GLP compliance is not scientific adequacy (Box 10.1, p. 220). 5. Evidence-selection rules drive divergent outcomes. Similar evidence gives safe levels that differ by orders of magnitude; differences in study-quality criteria and assessment stage partly explain EFSA–ANSES divergence (pp. 220–223, fn 2). 6. Regulatory endpoints are outdated. Guideline studies use 50-year-old endpoints blind to developmental and epigenetic effects (p. 223). 7. Funding correlates with findings (p. 225, Table 10.2). 8. Assessments missed highly exposed vulnerable groups: ICU neonates, bottle-fed infants (pp. 224, 226). 9. Formal systems did not find the hazard; accident did. Industry’s in-house hormone expertise was not applied to its own products (p. 217). 10. Test strategies are slowly improving: NTP and ANSES include single-dose studies; the OECD is adding endpoints (p. 229).

Framing lesson (interpretive) - The BPA story is the “same old story”, like asbestos, PCBs and DES: “putting a chemical into widespread use without understanding its health implications”, then trying to resolve public-health questions “while facing the intense pressure of serious economic consequences” (p. 226). The analogy is framed around process, which can hold whatever the outcome. But it invites an inference of harm that the analogues had and BPA, on the human evidence the chapter presents, had not demonstrated as of 2013.

Recommendations and advocacy - Restart the EU risk assessment, transparently, “conducted by the scientists authoring the papers with high scientific impact in this field”; use stakeholder conferences to expose interests (p. 226). - Take precautionary measures now to lower exposure “well below” rodent effect levels, i.e. end BPA uses with close human contact via food or the environment (p. 226). - Decouple testing from producer funding through an industry-financed, government-managed fund; no direct industry contracting of labs (p. 229). - Contract-lab results “must not outweigh” academic results; update guideline tests (p. 229). - Strengthen adviser independence; treat close ILSI ties as potentially incompatible; improve conflict-of-interest documentation (p. 229). - Pay experts adequately, finance through industry fees, and relieve academic workloads (p. 229).

Mechanisms and dynamics#

1. How the warning arose: knowledge that existed but was not connected#

• Oestrogenicity was established in 1930s pharmacology (p. 216). The substance then moved into a different domain (industrial materials) where that property was apparently not treated as relevant, because of a chain of reassuring assumptions: weak potency, no release of the monomer, and ignorance of foetal sensitivity (p. 220).
• Rediscovery came by accident, from basic endocrinology whose assays were contaminated. It did not come from regulatory screening or from industry’s own hormone research groups (p. 217).
• Knowledge was siloed across disciplines (pharmacology and materials chemistry), across institutions (firms’ hormone researchers and their product stewards) and across time.

2. Mental models of those who deployed and assessed the substance#

The chapter attributes the following assumptions to early deployers, classical toxicologists and regulators: - (a) “the dose makes the poison” with monotonic curves (pp. 217–219); - (b) potency proportional to receptor-binding affinity, hence “weak oestrogen”, 1,000–10,000 times weaker than estradiol (pp. 219–220); - (c) covalent polymer binding means no release (p. 220); - (d) rapid conjugation makes internal exposure to active BPA negligible (p. 224; Völkel et al., 2002); - (e) GLP guideline studies are the reliable evidence base (pp. 220–221).

These models produced confidence (“no real concern”, p. 220) and were then used to judge new findings implausible (Greim, 2004, p. 220). The chapter’s central dynamic is paradigm defence: anomalous results were dismissed because the prevailing model said they could not be true.

The chapter’s competing model: - hormonal action with feedback and receptor saturation; - multiple receptors; - windows of susceptibility; - latency; - deconjugation in tissues.

It presents this model as settled (“far-reaching agreement”, p. 219), though it also concedes continuing dispute (p. 218).

3. Evidence-selection rules as the arena of conflict#

• The fight was less about individual data than about which data count. EFSA and the RAR excluded studies for few doses, small n, statistics, or inconsistency with other studies (p. 221). The NTP included them with caveats (p. 222). The FDA’s advisers rejected blanket exclusion (pp. 222–223). ANSES used different quality criteria and a different assessment stage (p. 223, fn 2).
• A largely shared evidence base yields TDI/ADI values of 50, 0.13 or 0.008 μg/kg bw d depending on which study is “pivotal” (pp. 221–222). Only the 50 is an official figure; the other two are the authors’ own calculations from the Nordic and NTP choices.
• Standards of proof differ by institution:
• Canada acted on evidence it called “limited … from the perspective of rigour” (p. 222);
• UBA acted “despite uncertainties and gaps” (p. 223);
• EFSA required guideline-quality reliability;
• the FDA voiced “some concern” but kept its number (p. 223).
• Validation circularity: AGD is “not a validated endpoint for regulatory studies” (p. 222). Evidence from new endpoints is discounted until the endpoints are validated, and validation proceeds slowly through the OECD (p. 229).

4. Lock-in of test methods and of prior regulatory positions#

• Guideline endpoints unchanged for 50 years (p. 223) show methodological path dependence.
• EFSA’s 2010 reiteration of 2006 (p. 223) and the FDA’s unchanged ADI despite new concern (p. 223) suggest institutional reluctance to revise. The chapter reads this as bias; an alternative is genuine disagreement about evidence quality.

5. Time lags, latency and irreversibility#

• “At the time when the effects become detectable the chemical exposure has vanished” (pp. 215, 219). This structurally disadvantages the side that must prove human harm: epidemiology is weak when exposure is ubiquitous (no unexposed controls, a point implicit rather than stated), transient and early-life.
• Irreversible developmental effects (p. 219) raise the stakes of waiting.
• DES is the evidential precedent that such long latency is real (p. 219).

6. Exposure dynamics and hidden stocks#

• Exposure is ubiquitous and multi-source, from food containers, thermal and recycled paper, dust, dental sealants and medical devices (p. 216). Durable products form a growing in-home stock (p. 216).
• There is ignorance about uses (over 7,000 tonnes unaccounted for) and impurities (10,000 tonnes uncharacterised) (p. 216).
• Vulnerable subgroups (ICU neonates, bottle-fed infants) sit at the high end of exposure and sensitivity (pp. 219, 224).
• Non-food and dermal exposure were discovered late (p. 224).

7. Interests, influence and funding#

• The substance’s “high economic importance” is suggested as a factor in divergent assessments (p. 220).
• A consultancy with tobacco-industry history, by its own account, organised a European advocacy effort focused on a procedural chokepoint: hazard classification, which triggers labelling and possibly authorisation obligations (p. 225).
• Adviser conflicts of interest (nine of 21 panel members, p. 225) are attributed to the members’ own declarations, but no document is cited.
• Outcomes correlate with funding (Table 10.2, p. 228).
• The industry framed critics as “persistent NGO attacks” (Weinberg quote, p. 225) and challenged replication (p. 219).
• Structural dependence: REACH relies on industry-generated data (p. 226), and contract labs are “per se not economically independent” (p. 229).

8. Markets and publics as de facto regulators#

• Consumer and political pressure forced product withdrawals ahead of or against official safety verdicts (p. 225).
• This shows a gap between formal risk assessment and social legitimacy. The costs of that gap fell on firms’ reputations and earnings (p. 225).
• The chapter uses the episode to raise the question of earlier industry influence on assessments (p. 225). It does not analyse the gap between formal assessment and public legitimacy as a governance problem in its own right.

9. Institutional behaviour and correction#

• Internal correction mechanisms worked partially:
• the FDA’s own Science Board rebuked the agency (pp. 222–223);
• EFSA reformed its independence rules (p. 225);
• EFSA and ANSES produced a joint reconciliation of their differences (p. 223);
• the OECD began revising guidelines (p. 229).
• Divergence across jurisdictions (US federal agencies with each other, EU against member-state agencies, Canada, Nordic states against the majority) shows how institutional culture, mandate and assessment stage shape conclusions from shared evidence.

10. Framing and language#

• Terms the chapter contests: “emerging” substance (“euphemistic”, p. 216); “weak oestrogen” (“erroneously viewed”, p. 219); “safe” (p. 225); “pivotal” study (p. 221); effects “regarded as not relevant by the authors” (pp. 221, 227).
• Industry framing: “persistent NGO attacks”, “more benign Category 3” (p. 225).
• The authors’ own rhetoric: “same old story” (p. 226); “Science is vulnerable” (p. 225); “irrefutable evidence” (p. 220); “unprecedented” (p. 223); independent science and regulatory toxicology “speak different languages” (p. 229); “The time makes the poison” (p. 219).
• Shifting meaning of “low dose”: 100 μg/kg/day in Colerangle and Roy (p. 219; dose from external check, not stated in the chapter) versus 0.2–50 μg/kg in Table 10.1. What counts as “low” is relative to the regulatory NOAEL, not to human exposure.

11. Distribution of benefits, risks and costs#

• Benefits (material properties: transparent, light and hard plastics; protective coatings) go to producers, product makers and consumers. The chapter barely mentions them (p. 216).
• Potential risks fall disproportionately on foetuses, infants, young children, ICU patients and workers (pp. 219, 224, 226), none of whom chose the exposure.
• Costs of precautionary action fell on firms through market pressure (p. 225). The costs of substitution and the risks of substitutes are not analysed.

12. Innovation effects#

• There is almost no discussion of alternatives or innovation in materials.
• Innovation is discussed only as scientific and test-method innovation: ToxCast high-throughput screening (p. 220), new endpoints at the OECD (p. 229), SERMs as a pharmacological insight (pp. 215, 229).
• The recommendation to terminate food-contact uses (p. 226) is not weighed against what replaces them.

13. Complexity#

• Multiple receptors, feedback, receptor saturation, superposition of dose responses (p. 217), and life-stage-dependent effects (p. 219) make the system poorly suited to single-endpoint, single-mechanism, linear-extrapolation assessment.
• The chapter’s deeper claim is that the assessment apparatus assumed a simpler system than the one it was assessing.

Transferable insights (technology-neutral)#

1. A known hazardous property can fail to travel with the substance into new applications. Knowledge generated in one domain may not be carried into the decision to deploy the same thing in another. Assessment then starts from comfortable assumptions rather than known properties. - Evidence: oestrogenicity known from the 1930s (p. 216); deployment in consumer plastics with “no real concern” (p. 220); firms’ hormone expertise not applied (p. 217). - Strength: moderate. The chronology is documented, but the chapter shows nothing of what the 1950s deployers actually knew or weighed.

2. Formal screening systems can miss what curiosity-driven research stumbles on. Hazard identification may depend on scientists outside the regulatory system encountering anomalies in their own work. - Evidence: accidental rediscovery in 1993; no government or industry programme flagged BPA (p. 217). - Strength: moderate. One well-documented instance; the claim about programmes since 1982 is unreferenced.

3. Assessment methods embed assumptions about how harm scales. When the phenomenon behaves differently (non-linearly, with thresholds, or with timing-dependent effects), the methods can return confident “no effect” results that are artefacts of design. - Evidence: monotonicity and high-to-low extrapolation (pp. 218–219); endpoints blind to developmental effects (pp. 221, 223); the Tyl and Ryan null results versus the Table 10.1 findings (pp. 221, 227–228). - Strength: moderate. The general point about hormone-like action is widely accepted (Endocrine Society, p. 217). How far it applies to BPA at human exposures was, and remains, contested. The supporting base-rate figure (37%) comes from a heavily pre-selected sample.

4. Potency judged through a single mechanism can badly understate impact when an agent acts through multiple pathways. - Evidence: “weak oestrogen” by classical receptor binding versus action via membrane receptors, ERR-γ, GPR30, AhR, androgen and thyroid receptors (pp. 219–220); Soriano et al. (2012). - Strength: moderate. The mechanistic evidence is substantial; its translation to in vivo human risk is contested.

5. In contested risk decisions, the real decision is often made in the rules about which evidence counts. Eligibility and quality criteria, the choice of “pivotal” study and the assessment stage can move a “safe” level by orders of magnitude on a largely shared evidence base. - Evidence: EFSA 50 versus the implied 0.13 and 0.008 μg/kg bw d (pp. 221–222); exclusion criteria (p. 221); EFSA–ANSES footnote on different stages and criteria (p. 223). - Strength: strong for the pattern of divergence, which is documented across several bodies. The largest ratios are the authors’ own hypothetical calculations, not official figures.

6. Process-assurance standards (protocol compliance, auditability) and scientific adequacy (the right question, sensitive endpoints) are different things. Treating one as a proxy for the other lets well-documented but insensitive studies outweigh informative but less standardised ones, and the reverse error is possible too. - Evidence: Box 10.1 (p. 220); reliance on guideline studies (pp. 220–221, 223). - Strength: moderate. Conceptually sound and influential. The chapter idealises peer review and does not credit what process standards protect against.

7. Standardised tests ossify, and validation requirements create an interim period in which more sensitive, newer measures are discounted because they are not yet validated. - Evidence: AGD “not a validated endpoint” (p. 222); 50-year-old endpoints (p. 223); OECD only “currently modifying” guidelines (p. 229). - Strength: moderate. The pattern is documented. The chapter gives no timeline for how long validation takes.

8. When effects appear long after exposure and exposure is transient, observational detection in the exposed population is structurally hard. Demanding direct proof of harm in that population then amounts in practice to waiting until harm is irreversible and widespread. - Evidence: latency to puberty or middle age; “the chemical exposure has vanished” (pp. 215, 219); the DES precedent (p. 219). - Strength: strong for the latency mechanism itself (the DES precedent, p. 219; the chapter says epidemiology is “extremely difficult”). Moderate for the normative corollary in the second sentence. That is the note-taker’s inference: the chapter does not say that demanding proof means waiting for “widespread” harm. For BPA itself, human harm remained undemonstrated in 2013.

9. Who pays for evidence correlates with what it finds. Where the producer funds the studies regulators treat as decisive, a structural conflict of interest exists whatever individual integrity. - Evidence: Table 10.2 (p. 228); the funding-outcome literature (p. 225); REACH’s reliance on industry data (p. 226); contract labs “not economically independent” (p. 229). - Strength: moderate. The split is stark (0/11 versus 94/104) and fits wider literature. But the industry sample is small, the compilation is by a protagonist, design differences (strain, positive controls) are confounded with funding, and publication bias in academic work is not considered.

10. Interested parties can target procedural chokepoints, such as a classification decision that triggers downstream obligations, rather than contesting the science head-on.

    • Evidence: the Weinberg Group’s claimed role in the Category 3 versus Category 2 outcome and its downstream consequences (p. 225).
    • Strength: suggestive. It rests on a consultancy’s self-promotional claim. Causation is not independently established and I could not verify the web page.

11. The credibility of expert advice depends on the perceived independence of advisers. Conflict-of-interest rules may be tightened only after controversy.

    • Evidence: nine of 21 panel members with industry links; EFSA’s subsequent independence rules, justified by the link between its advice and public “trust” (p. 225); recommendations on ILSI (p. 229).
    • Strength: suggestive to moderate. EFSA’s response is documented. The panel count is attributed to members’ declarations but no document is cited. The chapter places the reforms after the controversy (“Meanwhile”) but does not show that the controversy caused them; “only after” is an inference from one case.

12. Markets and public pressure can outrun formal regulators. Product withdrawals under consumer and political pressure while official assessments still said “safe” expose a gap between technical risk assessment and social legitimacy, and push costs onto firms.

    • Evidence: baby bottle and drinking bottle withdrawals (p. 225).
    • Strength: moderate. The events are real but only briefly evidenced here.

13. Exposure can accumulate through durable stocks and poorly tracked diffuse uses. Ignorance of where something goes and what comes with it (unknown uses, uncharacterised by-products) is itself a risk factor.

    • Evidence: the growing in-home polycarbonate stock; over 7,000 tonnes of unidentified use; 10,000 tonnes of uncharacterised impurities (p. 216).
    • Strength: suggestive to moderate. The figures come from the EU RAR and a single impurity study. Consequences are inferred, not shown.

14. The most exposed are often the most vulnerable and the least visible to assessments built on average exposures.

    • Evidence: ICU neonates about ten times more exposed; bottle-fed infants twice; young children with “the highest rate of daily ingestion” and different metabolic capacity (pp. 219, 224, 226).
    • Strength: moderate. Biomonitoring data are cited. Whether these exposures are harmful remains contested.

15. Different institutions reach different conclusions from shared evidence because of differences in mandate, assessment stage and culture. The spread of responses (precautionary listing on “limited” evidence, reaffirmed safety, advisory rebuke, concern without numerical change) is a natural experiment in how standards of proof are set.

    • Evidence: Canada, EFSA, FDA and its Science Board, NTP, UBA, ANSES, the Nordic footnote (pp. 222–223).
    • Strength: strong for the divergence itself. The positions are documented, though characterised by the chapter’s authors. Moderate for the explanation. Only assessment stage and study-evaluation criteria are evidenced, in the EFSA–ANSES footnote (p. 223). “Mandate” and “culture” are the note-taker’s glosses, and the chapter itself leans towards method conservatism and industry influence.

16. Independent advisory review can act as an internal corrective to an agency’s commitment to its prior position, but it may shift rhetoric more than numbers.

    • Evidence: the FDA Science Board subcommittee; the FDA’s “some concern” with an unchanged ADI (pp. 222–223).
    • Strength: moderate. One instance.

17. Correcting one form of bias by handing authority to another interested group relocates the bias; it does not remove it. [My analytical inference from the chapter, not the authors’ claim.] The authors propose that the reassessment be “conducted by the scientists authoring the papers with high scientific impact in this field” (p. 226), and they note that independent science “is interested in finding the effects of a substance and publishing these findings” (p. 229).

    • Strength: asserted/analytical. It shows the chapter’s own blind spot rather than a lesson the chapter evidences.

Limitations, contestation and bias check#

Standpoint and absence of dissent#

• The authors are protagonists (see Authors and standpoint). Soto’s co-authored work and Gies’s own agency and workshop are cited in support (pp. 221, 222, 223, 226) without the text noting the authors’ involvement. The only disclosure is the generic footnote that the chapter reflects the authors’ opinions (p. 215).
• No panel or response is included; the promised panel is an editorial error (p. 215 versus p. 179). No regulator (EFSA, FDA), industry scientist or sceptical academic (for example Sharpe, Tyl, Ashby) is given space. Their views appear only as brief quotes or citations framed by the authors.

Report-level framing versus chapter evidence#

• The report’s Introduction classifies BPA among the nine “false negatives” in Part A (p. 10), alongside lead, vinyl chloride and others with demonstrated harm. It says three themes emerge from them: “more than sufficient evidence for much earlier action”, obstructive business behaviour, and “the value of independent scientific research and risk assessments” (p. 10).
• The chapter itself concedes, at least for the cross-sectional associations, that they “are not a proof of causation” (p. 226), and its harm evidence is in rodents.
• Chapter 2 of the same report acknowledges that “some scientists argue that precautionary regulation of … bisphenol-A (BPA) in items that come into contact with food may represent ‘over-regulation’” (p. 33).
• As of 2013, BPA is better described as a contested or unresolved case than a demonstrated false negative. The “same old story” analogy to asbestos, PCBs and DES (p. 226) is framed around process (use before understanding, economic pressure), but it borrows the moral weight of cases with proven harm.

Overstatements and internal inconsistencies#

• “Far-reaching agreement” (p. 219) and “widespread agreement” (p. 220) sit uneasily with “still disputed by some scientists and members of regulatory bodies” (p. 218).
• “Irrefutable evidence” from one study (p. 220; an ex vivo islet study per the external check).
• Figure 10.1 (p. 218): the caption’s inverted-U description (“small doses show small effects, intermediate doses cause the most pronounced effects”) fits panels (c) and (d) only. Panel (a) shows its largest effect at the highest dose, and panel (b) peaks at the lowest dose. Panel (a)’s lowest-dose label (0.05) may also not match the cited paper (see 10.4).
• “Over 800 peer-reviewed studies that showed toxicity of BPA at exposure levels below the level of human exposure” (p. 223) versus “at least 46” studies at ≤50 μg/kg (p. 221). The 800 is a count of investigator-initiated studies on BPA toxicity, not of positive low-dose findings.
• “Differ more markedly than for any other chemical” (p. 220) and “unprecedented” (p. 223) are asserted without comparison.
• The Tyl AGD claim: male increase in the text (p. 222), female decrease in Table 10.1 (p. 227), female increase in Tyl’s own abstract; the cited Chahoud letter discusses ovary weights, not AGD.
• Smaller factual slips:
• Völkel et al. is said to have “nine” volunteers (p. 224); the abstract gives 3 + 3 + 4;
• Braun et al. (2009) is called “cross-sectional” (p. 226) but is a prospective cohort;
• Krishnan et al. used flasks, not “cell culture dishes” (p. 217);
• the chapter’s early chronology conflicts with Annex 3 (pp. 216 versus 730).
• On p. 221 the chapter speculates that effects may occur in the pg/ml range, “the same range as estimates of current human exposure”. On p. 224 it favours the ng/ml measurements over those model estimates. It uses whichever figure supports concern in each place.

Selective engagement with counter-evidence#

• The replication critique is presented only as an industry claim (p. 219) and answered with an assertion of consensus.
• Specific failed replications are not engaged: Ashby et al. (1999) appears in Table 10.1 as a positive study; Cagen et al. (1999, Shell) is not cited. [Verified: Cagen et al. “purposely duplicated” the Nagel/vom Saal protocol and found no effects.]
• Ryan et al. (2010) is acknowledged in one sentence (p. 221); Sharpe (2010) in half a sentence (p. 218).
• Teeguarden et al. (2011), with undetectable free BPA in all serum samples, is in the reference list (p. 238) but not discussed.
• The chapter’s own warning about laboratory contamination (p. 216) is not applied to the contested high measured serum values.
• Table 10.1 reads positive findings into studies whose authors concluded otherwise (Tyl, Ashby). That is the mirror image of the “arbitrarily selected data” charge against EFSA (p. 222).

Weak inferences#

• Peer-reviewed publication is taken to show the scientific community “do[es] not agree with the criteria chosen by EFSA” (p. 221). A weak inference: acceptance for publication is not endorsement for use in quantitative risk assessment.
• The “overlap” of animal effect doses with human intake (p. 225) holds only at the margins. The lowest animal doses (0.2–2.5 μg/kg, 17 of the 49 rows in Table 10.1) overlap upper-percentile and worst-case human intakes: modelled adults up to 1.5, German children up to 7, worst-case young children 11–13 μg/kg. Typical intakes are lower (NHANES adult 95th percentile 0.15–0.22), and most table rows are at 10–50 μg/kg. It also relies on the contested assumption of comparable internal dose. [Audit: corrected from an earlier statement that cited intakes “mostly sit below” the table range.]
• The Calabrese and Baldwin 37% (p. 218) is offered as evidence that non-monotonic curves are not “rare in toxicology”. It comes from the 1% of articles meeting strict entry criteria (external check), so it is not a base rate.

Asymmetric treatment of interests#

• Financial conflicts of industry-linked scientists are scrutinised (pp. 225, 229).
• Intellectual, career and advocacy interests of academic protagonists are not. The chapter recommends that those very scientists conduct the reassessment (p. 226) and openly says independent science is oriented to finding and publishing effects (p. 229).
• No document is cited for the nine-of-21 panel count, which the chapter attributes to members’ own declarations. The reference list’s related sources are political or advocacy documents: a parliamentary question (Breyer, 2009, not cited in the text) and a Corporate Europe Observatory report (cited only for proposed remedies, p. 229).

Omissions#

• No discussion of benefits of BPA uses, costs and feasibility of alternatives, or the risk of regrettable substitution (for example other bisphenols). This matters because the core recommendation is to terminate food-contact uses (p. 226).
• No lag or cost quantification.
• Little on the mechanics of the C&L decision beyond the consultancy’s claim.
• No explicit treatment of the precautionary principle as a decision standard. The Commission Communication on the precautionary principle (EU, 2000) is in the reference list (p. 232) but never cited or discussed. Precaution appears only as Canada’s “precautionary approach”, UBA’s “precautionary action”, the charge that EFSA’s conjugation assumption is “far from being precautionary”, and the interim recommendation (p. 226). The chapter never says what weight of evidence should trigger action, although its summary promises to address “challenges for applying the precautionary principle” (p. 215).

Hindsight bias#

• Handled reasonably for the 1950s: the chapter explicitly says non-monotonic curves and DES effects were unknown then (p. 220).
• Less so for the 2000s regulators, whose caution about reproducibility and internal dose is attributed mainly to method conservatism and industry influence rather than to real epistemic disagreement. The joint EFSA–ANSES footnote (p. 223) is the one place a more neutral explanation appears.

Fairness in the other direction#

The chapter documents real and verifiable institutional events: - the FDA’s own advisers declared its safety margins “inadequate” (p. 223); - Canada listed BPA as toxic (p. 222); - ANSES “confirmed” multiple low-dose animal effects (p. 221); - the NTP expressed concern (p. 222); - the EU banned BPA baby bottles (p. 226); - EFSA itself reformed its independence rules (p. 225); - the funding split in the vom Saal and Hughes review is as reported (verified).

Its central methodological critique (Box 10.1; exclusion of non-guideline studies) was later widely taken up. Its call to “start again with the risk assessment” in Europe was followed in substance. See the pointers below.

Post-2013 pointers (for the hindsight strand; partial)#

Verified against primary sources in this session - EFSA 2023 (EFSA Journal 21(4):e06857, PMID 37089179): - notes a 2015 temporary TDI of 4 μg/kg bw/day; - established a TDI of 0.2 ng/kg bw/day, about 250,000 times lower than the 50 μg/kg TDI criticised in the chapter (20,000 times lower than the 2015 value); - critical effect: Th17 immune cells in mice; - mean and 95th-percentile dietary exposures “exceeded the TDI by two to three orders of magnitude”; conclusion: “there is a health concern from dietary BPA exposure”. - The re-evaluation used a pre-established, publicly consulted protocol that included academic studies, broadly the direction the chapter urged. - BfR dissent: the German Federal Institute for Risk Assessment “opposed EFSA’s revision”, according to vom Saal et al. (2024, EHP 132:45001). That is a protagonist commentary co-authored by Soto; the BfR’s own documents were not checked. - CLARITY-BPA (FDA–NIEHS consortium joining a guideline study with academic grantee studies): - the FDA/NCTR GLP core study found “No BPA-related effects … in the in-life and non-histopathology data”, with possible effects only at 25,000 μg/kg/day (Camacho et al., 2019, Food Chem Toxicol 132:110728); - the integrated academic studies reported effects in brain, prostate, urinary tract, ovary, mammary gland and heart, “many … at the lowest dose tested, 2.5μg/kg/day”, many non-monotonic (Heindel et al., 2020, Reprod Toxicol 98:29; Soto co-author); - an industry-consultancy analysis found little evidence of non-monotonic responses in the core study (Badding et al., 2019, Exponent). - The chapter’s “different languages” divergence therefore persisted even inside a jointly designed programme.

Unverified here (from background knowledge; check against primary legal texts) - EU Regulation (EU) 2024/3190 banning BPA and certain other hazardous bisphenols in food-contact materials, with transition periods. - Harmonised EU classification of BPA as toxic to reproduction Cat. 1B (2016). - Identification as a substance of very high concern for reproductive toxicity (2017) and endocrine-disrupting properties (2017 human health; 2018 environment). The Category 3 outcome the Weinberg Group claimed credit for was eventually superseded. - France’s national ban on BPA in food-contact materials (law of December 2012, effective 2015). - US FDA removal of BPA uses in baby bottles, sippy cups (2012) and infant formula packaging (2013) on grounds of industry abandonment, with FDA statements (2014, 2018) maintaining that BPA is safe at current exposure levels. - A US/EU regulatory divergence therefore persists more than a decade after the chapter.

Notable quotes#

1. “It was by accident that the risks associated with it were re-discovered.” (p. 217)
2. “Industry missed a chance to care for their products responsibly.” (p. 217)
3. “BPA challenged our belief that high doses produce more serious effects than low ones.” (p. 217)
4. “After 500 years it has become clear that Paracelsus’s paradigms do not contribute to the protection of human health and environment if they are applied to risk assessments in a naive way.” (p. 219)
5. “At the time when the effects become detectable the chemical exposure has vanished.” (p. 219)
6. “GLP does not indicate that good science has been performed or that the scientific results are adequate and sufficient to protect human health and the environment” (Box 10.1, p. 220)
7. FDA Science Board subcommittee: “the Margins of Safety defined by FDA as ‘adequate’ are, in fact, inadequate.” (quoted p. 223)
8. “Risk assessment is only a protocol used by the regulatory community, not science per se.” (p. 224)
9. Weinberg Group (quoted): “This approach proved very effective, as ultimately the C&L working group did not follow the recommendation of the Rapporteur Member State to classify BPA as a Category 2 reproductive toxicant” (p. 225)
10. “Independent science and regulatory toxicology seem to speak different languages.” (p. 229)

Open questions#

1. What exactly did Tyl et al. (2002) find on AGD and ovary weight at low doses, and how did EFSA and the FDA document their reasons for disregarding those differences? This needs the full paper and the EFSA 2006/2010 opinions.
2. What was the provenance of the 4–6 ng/ml free-BPA measurements, and how did the contamination debate resolve? What does the post-2013 literature on serum measurement (including Teeguarden et al. 2011 and later cross-lab validation) say about which side was right on internal dose?
3. What were the “programmes … since 1982” that failed to flag BPA (p. 217), and did any of them consider endocrine endpoints at all? A primary-source check would test the chapter’s claim of institutional failure.
4. The C&L decision: what reasons did the working group record for Category 3 over the Rapporteur’s Category 2, and is there independent evidence (minutes, documents) of the advocacy’s influence beyond the consultancy’s own marketing?
5. Which declarations of interest (date and panel composition) underlie the “nine of 21” EFSA AFC panel count? The chapter attributes it to members’ own declarations but cites no document. And which member is the one paid for the Dekant and Völkel (2008) review?
6. How reproducible were the Table 10.1 findings when the same endpoint was retested by independent labs? Which of them did ANSES, EFSA (2023) and the CLARITY academic studies confirm or fail to confirm?
7. What replaced BPA in the uses restricted after 2011, and did substitutes (for example other bisphenols) carry similar hazards? The chapter is silent, yet this determines whether its central recommendation was net-protective.
8. Did an industry-financed, government-managed testing fund (p. 229) ever get implemented anywhere, and with what effect on evidence quality?
9. How did EFSA’s 2023 systematic-review protocol handle the chapter’s core dispute over inclusion of non-guideline studies? And did the BfR’s and others’ objections reproduce the earlier split along similar lines?
10. Is BPA properly a “false negative” in the report’s typology, or a case of unresolved scientific conflict in which the governance lesson concerns how institutions decide under persistent disagreement? This bears on how the case should be used as an analytical lens.

Audit log#

Independent fact-check against the full chunk text (PDF 217–241), the rendered Figure 10.1 page, and other report pages (pp. 10, 33, 179, 194–196, 240, 690, 698, 730). Only report-internal claims were re-checked. External “[Verified]” items from the earlier pass were not re-checked, apart from those the chapter’s own reference titles confirm (Krishnan’s “flasks”; Taylor 2008 in neonatal mice; Ashby 1999 “Lack of effects”).

• Confirmed against the PDF: the panel copy-paste error (p. 215 against p. 179; Panel 8.2 by Huff, pp. 194–196; Ch11 starts on p. 240); the author bios; p. 10 “false negatives”; the p. 33 “over-regulation” quote; Annex 3 at p. 730 (by Swan).
• Read line: noted that Figure 10.1 was later rendered and checked visually.
• Gies bio: gave the full name of the WHO Collaborating Centre (“… and Air Pollution Control”).
• Standpoint: corrected the claim that the chapter “presents” Chapel Hill as an assessment; it says the statement “is not a risk assessment in the classical sense”.
• Standpoint: added Gies (2007) as the review cited for the “at least 46” studies; “led” changed to “first author of”.
• Evident stance (b): the misquote “not defensible” corrected to the verbatim “cannot be defensible”.
• Summary: added the promised “challenges for applying the precautionary principle”.
• 10.2: added that the impurities “show structural features of oestrogenic chemicals”.
• 10.3: added that Bason and Colborn “suspected mainly pesticides”; noted that the chapter’s own Krishnan reference title says “flasks”.
• 10.4: “term attributed to Sonnenschein” softened to “cited for the phenomenon”.
• Figure 10.1: corrected panel (a) doses (label reads 0.05, not 0.025) and recorded visual bar values. The caption fits only panels (c) and (d); panel (a) has its largest effect at the highest dose and (b) peaks at the lowest. Noted the Cabaton 2010/2011 mismatch.
• 10.4: added Colerangle and Roy’s “much more potent than expected from its oestrogen-receptor binding profile”.
• 10.5: restored the conditional “If this also happens in humans”; made the epidemiology quote verbatim.
• 10.5: added that Soriano et al. show a nuclear-ER-mediated effect, which refutes binding-affinity reasoning rather than supporting the non-classical pathways.
• Box 10.1: added that regulatory tests need not be published or peer-reviewed; softened “GLP’s value is not credited” to “gets only brief credit”.
• 10.6: restored the hedges (“probably”, “may have contributed”); “mean” removed.
• 10.7: added “number of live pups per litter”; ANSES “immune behaviour” (sic) restored.
• 10.7: added the no-effect-level concession and “Further research is needed”; softened “non sequitur” to “weak inference”.
• 10.8: added the Gupta (2000) and Somm (2009) AGD corroboration; clarified the Nordic factor of 380 is relative to the EU RAR.
• 10.8: added NTP’s “difficult to interpret” and “limited evidence” wording; Canada “weak” changed to the verbatim “limited”, plus the note that decisive studies were unstated.
• 10.8: Chapel Hill “not a risk assessment” caveat added; FDA “for the first time” acknowledgement added, with the 15 January 2010 date flagged as unchecked background; “Certainly there are data gaps” concession added; ANSES “questions parts of EFSA’s” added.
• 10.9: corrected an error. The notes said most cited human intakes are below Table 10.1’s lowest dose. Most are not; the overlap holds for the 17 rows at 0.2–2.5 μg/kg against upper-percentile and worst-case intakes.
• 10.9: added the “not suitable for assessing endocrine disruptors” sentence and NTP’s caveat that free-BPA values may not represent “true” concentrations.
• 10.10: corrected “no in-text source” for the nine-of-21 count (the chapter attributes it to members’ declarations); clarified the Breyer and CEO references.
• 10.10: added the “a job like any other” concession; noted the Table 10.2 author-order mismatch.
• 10.10: corrected the scope of the causation caveat (attached only to Braun 2009; the other associations are uncaveated; Braun 2011 is described as “affected”).
• 10.11: added the interim framing “Until final decisions are made”.
• 10.12: linked the single- and multiple-dose inclusion to EFSA’s “one or two doses” exclusion criterion.
• Table 10.1: flagged Ema et al. (2001) as a possible third “no effect” study (background, unchecked); listed minor citation mismatches.
• Timeline: flagged externally sourced details (100 μg/kg; “failed replication”; EPA); marked the Weinberg dates as inferred; added missing page refs (218, 226); named the Annex 3 author.
• Lag analysis: “first EU restriction (2011)” corrected to “first EU use ban”, noting the 1990 migration limit and the Category 3 classification.
• Lag analysis: corrected “no restriction covered main routes / action limited to bottles” to what the chapter reports, with France 2012 flagged as unchecked background.
• What was known when: corrected “GLP studies from industry and government [in the late 1990s]” (Tyl is 2002/2008, Ryan 2010); corrected the causation-caveat scope.
• Harms and costs: “benefits not discussed” softened to “only a sentence or two on p. 216”.
• Framing lesson and Limitations: rebalanced the “same old story” critique, since the analogy is framed around process.
• Mechanisms: removed “first-pass” (not in the source); flagged the 0.13 and 0.008 as the authors’ calculations and “same evidence” as “largely shared”.
• Mechanisms: corrected the nine-of-21 sourcing; replaced “treats as vindication” (not in the source) with what the chapter actually does; “not biologically relevant” misquote corrected to “regarded as not relevant by the authors”.
• Insight 5: “no change in the underlying evidence” changed to “largely shared evidence base”.
• Insight 8: split the rating (strong for latency, moderate for the normative corollary).
• Insights 11, 14 and 15: softened 11’s “only after controversy” and corrected its sourcing; removed the unsupported “per kg” from 14; split 15’s rating (strong for divergence, moderate for causes).
• Limitations: “appear as independent corroboration” softened to “cited in support”; the Figure 10.1 caption mismatch added; the pg/ml inconsistency restated precisely.
• Limitations: overlap bullet corrected; Calabrese “as if typical” softened; conflict-of-interest sourcing clarified.
• Limitations: new omission added (no explicit precautionary-principle standard; EU 2000 Communication in the references but uncited).
• Open question 5: reworded in line with the chapter’s own attribution of the count.
• Digest: “too weak” changed to “unsuitable”; “found none” changed to “reported none”, with Ryan 2010 and the Tyl dismissal added; exclusion criteria attributed to EFSA and the EU RAR.
• Digest: politicians added to market pressure; GLP lesson and interim recommendation quoted accurately; insights 7 and 12 ratings split; Chapel Hill caveat fixed.
• Digest: caveats for “overlap”, precaution standard and Figure 10.1 added; causation-caveat scope corrected; “vindicates” softened.
• Scanned both files for references to contemporary technologies or companies not in the source: none found.