Invited Commentary: Maternal Plasma Polybrominated Diphenyl Ethers and Thyroid Hormones—Challenges and Opportunities

Abstract

Thyroid hormones play a fundamental role in fetal and child development. While iodine deficiency-related maternal and child hypothyroidism may cause severe mental retardation, recent evidence suggests that milder forms of maternal hypothyroidism and hypothyroxinemia during pregnancy are also associated with altered neurodevelopment. On the other hand, hyperthyroidism during pregnancy has been associated with adverse fetal outcomes. Findings published by Abdelouahab et al. in the American Journal of Epidemiology (Am J Epidemiol. 2013;178(5):701–713) suggest that plasma concentrations of maternal polybrominated diphenyl ethers (PBDEs), which were used as flame retardants until recently and are detected in the tissues of virtually every North American, are associated with umbilical cord and maternal thyroid hormone levels during pregnancy. Although PBDEs have been consistently shown to reduce levels of free and total thyroxine in experimental animal studies, the direction of associations in human studies has been inconsistent. In this commentary, I discuss challenges beyond the factors often cited in the epidemiologic literature to explain inconsistent findings which more specifically apply to the study of PBDEs and thyroid hormones. These include the determination of iodine intake status, the method used to adjust for blood lipid concentrations, the measurement of free thyroid hormone levels, the possible effect of PBDE metabolites, and the potential for reverse causality.

Thyroid hormones play a fundamental role in fetal and child development. The importance of adequate thyroid function is most evidently demonstrated by the severely stunted growth and cognitive impairment observed in persons affected by cretinism, a long-recognized condition caused by iodine deficiency-related maternal and child hypothyroidism (1). More recent evidence suggests that even milder forms of maternal hypothyroidism (2) and hypothyroxinemia (3–5) during pregnancy are associated with impaired cognition, attention, expressive language, and motor performance in children. A large body of evidence suggests that polychlorinated biphenyls, highly persistent synthetic chemicals that were used in electrical transformers, inks, plastics, and other consumer products until the 1970s, may interfere with thyroid hormone levels and function during pregnancy (6–11). Polybrominated diphenyl ethers (PBDEs), which were widely used as flame retardants until 2004, when they were banned in the United States and Canada for most uses, have chemical structures and properties that closely resemble those of polychlorinated biphenyls. The study by Abdelouahab et al. (12), published in the current issue of the American Journal of Epidemiology, adds to the steadily accumulating evidence that PBDEs may also disrupt thyroid function during pregnancy (13–17) and contributes a unique feature to this body of literature. While Zota et al. (15) previously measured thyroid hormone levels around the time of fetal thyroid function onset in a small study (n = 25), the study by Abdelouahab et al. is the first to have measured maternal thyroid hormone levels prior to this important milestone in a majority of participants (mean gestational age = 10.8 weeks (standard deviation, 2.7); range, 3.3–20.0 weeks). Disruption of maternal thyroid hormone levels before the onset of fetal thyroid function, which occurs between 18 and 22 weeks' gestation (18), is of particular significance, since during this period the fetus depends entirely on maternal thyroid hormones for normal brain development, including neuron proliferation and migration, and thyroid hormone-related cortical gene expression (18–20).

In their study, Abdelouahab et al. reported positive associations between lipid-standardized (i.e., expressed in ng/g lipids) plasma concentrations of PBDEs and levels of both free triiodothyronine (T3) and free thyroxine (T4) but an inverse association with total T3 and total T4 before 20 weeks' gestation (12). Inverse associations with free T3 and total T4 were found at delivery, and inverse associations with free T4 and total T4 were found in umbilical cord blood. No significant associations were found with thyroid-stimulating hormone. The study was well-designed, and Abdelouahab et al. considered a number of important potential confounders that were not consistently measured in previous studies, including urinary iodine levels and blood levels of selenium, mercury, and thyroperoxidase antibodies (12). Results obtained by Abdelouahab et al. are somewhat consistent with those of Stapleton et al. (14), who found a positive association between the concentration of pentabromodiphenyl ether and free T4 during pregnancy, and Herbstman et al. (16), who reported an inverse association between bromodiphenyl ethers 100 and 153 and total T4 in cord blood. However, they contrast with much of the rest of the literature on pregnant women, which is notably inconsistent. Beyond the general factors often cited in the epidemiologic literature to explain inconsistent findings (e.g., uncontrolled confounding, collider stratification bias, selection bias, and measurement error), a few challenges are more specific to the study of the thyroid hormone disruption potential of PBDEs and may explain the discrepancies within the literature. These include the determination of iodine intake status, the method used to adjust for blood lipid concentrations, the measurement of free thyroid hormone levels, the possible effect of PBDE metabolites, and the potential for reverse causality. In this commentary, I discuss these challenges and aim to encourage a conversation on the most appropriate strategies for overcoming them, either through the application of existing methods or through further research.

DETERMINATION OF IODINE STATUS

Iodine is an essential component of T3 and T4, and although Canada and the United States are considered iodine-sufficient areas, recent data (2005–2008) from the US National Health and Nutrition Examination Survey suggest that as many as 57% of pregnant US women have low urinary iodine concentrations, indicating insufficient intake (21). However, few studies of PBDEs and thyroid hormone levels have collected data on iodine intake, which may act as both a confounder and an effect modifier (22). Iodine has a short half-life, and its excretion exhibits large within-person variability (23). Because of this, although single iodine measurements are considered adequate to evaluate iodine status in populations, it has been reported that even 10 serial urine samples could only estimate iodine status at the individual level with 20% precision (23). The cost of a sufficiently large number of measurements would likely be prohibitive. Some studies have instead used food frequency questionnaires to estimate iodine intake, but whether data generated by these instruments provide better intake estimates in individuals than urinary iodine from spot samples would require further investigation. Abdelouahab et al. must be commended for conducting the first study of PBDEs and thyroid hormone levels in pregnancy to determine urinary iodine levels in participants, but because iodine status is expected to be a strong determinant of thyroid hormone levels, residual confounding may remain in this and prior analyses if an association (whether structural or empirical) is observed between iodine intake and PBDE blood concentrations.

CORRECTION FOR BLOOD LIPID LEVELS

Of particular interest is the observation that results reported by Abdelouahab et al. differed according to the method used to adjust for the concentration of blood lipids. Because of their high lipophilicity (octanol:water partition coefficients = 5.7–8.3) (24), PBDEs are expected to be primarily (though not entirely) sequestered in blood lipids, and the concentration of PBDEs on a blood volume basis is generally significantly correlated with total blood lipid levels (unpublished observation), which vary greatly within and between persons (25, 26). Thus, some form of adjustment for lipid levels appears to be necessary, but as Abdelouahab et al. point out (12), there is controversy regarding the most appropriate method for doing so.

Some investigators do not favor lipid standardization in studies of the health effects of lipophilic chemicals (27) such as PBDEs, but the alternative method—adjustment for blood lipids by including the variable in multiple regression models (or lipid stratification)—also has its limitations and has not been widely adopted. One possible explanation for the limited use of lipid stratification is that blood lipids do not qualify as a confounder, since this variable is not expected to cause a change in thyroid hormone levels or to be located on a backdoor path in a causal diagram. However, blood lipid concentrations are expected to affect PBDE concentrations on a blood volume basis. Adjusting for such variables (i.e., ones that cause the exposure but not the outcome), sometimes referred to as instrumental variables, has been shown in theoretical papers, simulations, and empirical examples to adversely affect precision and to induce bias by amplifying uncontrolled confounding (28–32). Hence, rather than confounding, the issue at hand may be one of measurement error, where the concentrations of PBDEs on a blood volume basis need correction.

Many investigators operationalize this correction by way of lipid standardization, assuming that PBDEs equilibrate within body lipids and that PBDE concentrations in blood lipids and the lipid fraction of target organs are correlated. Indeed, some investigators have reported strong correlations between the lipid-standardized concentrations of PBDEs in serum, adipose tissue, and the liver in humans (33, 34). Therefore, lipid standardization may remain the better method, but challenges remain irrespective of the lipid adjustment method. Among other things, blood lipid levels increase postprandially, which may temporarily dilute the lipid concentration of PBDEs. The collection of fasting samples may thus be more appropriate, but obtaining such samples in the context of epidemiologic studies is often difficult. Total lipids are also rarely measured directly, and the Phillips et al. formula (35) that is customarily used to estimate them may introduce additional variability and is probably not appropriate for pregnant women, whose lipid profiles differ from those of the nonpregnant adult population from which the formula was derived. Finally, the distribution of PBDEs may vary between lipoproteins and lipids with different polarities, as has been shown for other lipophilic environmental chemicals (36, 37), so the use of total lipids for correction may not be appropriate. Clearly, as Abdelouahab et al. duly note (12), more research on the topic of lipid correction is needed.

MEASUREMENT OF FREE THYROID HORMONE LEVELS

Another challenge concerns the measurement of free thyroid hormone levels. Immunoassays, which are generally used to measure free thyroid hormone levels, have been reported to be influenced by the concentration of T4-bound transport proteins. In a series of studies investigating the performance of different immunoassays, Nelson and collaborators prepared gravimetrically calibrated serum-based solutions with varying concentrations of protein-bound T4 (but constant free T4) and other solutions with varying levels of free T4 without binding proteins (38–40). They found that increases in the concentration of protein-bound T4 positively biased the free T4 readings generated by most immunoassays and reported that different assays varied widely in performance. Measurement of free T3 would be expected to be similarly biased. On the other hand, equilibrium dialysis, which as a first step physically separates free T4 from protein-bound T4, generated accurate results in samples with normal or elevated free T4 concentrations. Thus, the respective performance of the different immunoassays used in the studies of PBDEs and free thyroid hormones may partly explain the inconsistent results found in the literature.

This has implications for the study by Abdelouahab et al. In their paper, the authors suggest that their finding of a positive association between lipid-adjusted PBDE concentrations and free T3 and T4 before 20 weeks’ gestation may be due to measurement error introduced by standard immunoassays and to the fact that thyroid-binding protein levels are elevated in early pregnancy (12). However, for immunoassays to have biased associations upwards in the present study, PBDEs would need to be positively associated with protein-bound T4 concentrations, which presumably would result in a positive relationship between PBDEs and total T4 rather than the inverse association observed. Furthermore, different concentrations of thyroid-binding proteins eliciting different associations between free T3 or T4 and PBDEs would imply that this variable acts as an effect modifier. There is little evidence supporting this. An alternative explanation is that PBDEs may indeed be positively associated with free T3 and T4. Although free T3 has seldom been measured, several previous studies have found positive associations (both statistically significant and nonsignificant trends) between PBDEs and free T4 and inverse associations with thyroid-stimulating hormone, suggesting a hyperthyroidic effect of PBDEs (11, 13, 14, 41–44). Positive associations with free T3 or T4 would have particular public health significance because maternal hyperthyroidism during pregnancy has been associated with multiple adverse fetal outcomes, including premature birth, low birth weight, intrauterine growth restriction, and fetal loss (45–49).

PBDE METABOLITES

Results from in vitro studies suggest that hydroxylated PBDEs (formed by phase I metabolism), in particular, may be associated with thyroid hormone disruption. Hydroxylated PBDEs, whose chemical structures more closely resemble those of T3 and T4 than their parent compounds, have indeed been shown to displace T4 from the binding protein transthyretin (50), to compete with binding of T3 to human thyroid receptors α and β (51), and to up-regulate type I deiodinase (52), which is involved in the deiodination of T4 to T3 and reverse T3. There is little evidence that PBDE parent compounds have such effects, suggesting that hydroxylated PBDEs may be more potent in disrupting thyroid hormone homeostasis. If verified, this hypothesis could also help explain discrepancies not only among human studies that have measured PBDE parent compounds but also between human and animal studies.

REVERSE CAUSALITY

A final issue in studies investigating associations between exposure to PBDEs and thyroid hormone levels concerns reverse causality. Thyroid hormones regulate lipid metabolism. Thus, persons with hypothyroidism tend to have higher concentrations of blood lipids and greater fat mass, which can potentially dilute lipid-standardized PBDE values while increasing concentrations on a blood volume basis; opposite trends apply to persons with hyperthyroidism. In addition, thyroid hormones affect the activity of some cytochrome P-450 enzymes which are involved in PBDE biotransformation (53, 54). Thyroid hormones may thus affect the concentration of PBDEs, and, as a consequence, discrepancies in the literature may be partly explained by varying degrees of reverse causality. It is noteworthy that for ongoing exposures such as PBDEs, even longitudinal studies are subject to reverse causality, since the time ordering of exposure and outcome cannot always be firmly established. This is the case for studies of exposure to PBDEs and thyroid hormone levels, where, because of its effect on lipid metabolism, thyroid hormone status may have already influenced PBDE concentrations at the time of measurement.

Despite the challenges facing epidemiologic investigators studying associations between blood PBDE concentrations and thyroid hormone levels during pregnancy, most studies conducted to date have found associations with at least one thyroid hormone. The direction of associations has not been consistent, which may be due to the factors identified above, nonlinear dose-responses, concomitant exposure to other environmental chemicals, or different distributions of variables affecting susceptibility between study populations (e.g., age, iodine intake status, thyroid autoimmunity, and other environmental exposures). Further research on the best ways to address these issues is warranted. The weight of the evidence is strengthened, however, by the consistent results that have been reported in numerous experimental studies conducted in rodents and other mammals (55–57). Results from mechanistic studies also support an effect of PBDEs (primarily hydroxylated PBDEs) on thyroid hormone homeostasis. The potential consequences of altered thyroid hormone homeostasis by PBDEs include adverse neurodevelopment, as reported in several studies (17, 58, 59), but may also include reduced fertility (60) and low birth weight (61). While PBDEs have been withdrawn from the market, exposure is expected to continue since PBDEs leach from household items that are often kept for long periods of time, such as furniture, motor vehicles, and draperies, as well as from construction materials, all of which are likely to act as long-term reservoirs. Importantly, Technical Bulletin 117, the unique California regulation believed to be largely responsible for the ubiquitous use of flame retardants in consumer products manufactured for sale in North America, is still in effect (62). Until an updated version of this regulation is adopted (one has recently been proposed by the California Department of Consumer Affairs), PBDEs are being replaced by other flame retardants whose potential health effects are not well understood but may be similar to those of PBDEs. Indeed, recent research has suggested that replacement flame retardants such as hexabromocyclododecanes interfere with thyroid hormone action in the developing brain in rats (63), and house-dust levels of the organophosphate flame retardant tris-dichloropropyl phosphate were inversely associated with free T4 levels among men recruited from an infertility clinic (64). These results speak to a need to reverse the burden of proof from regulatory agencies and society to those who profit financially from the sale of these products. The study by Abdelouahab et al. constitutes a valuable contribution to this argument.