Thyroid-Disrupting Chemicals: Interpreting Upstream Biomarkers of Adverse Outcomes

1Recent epidemiologic studies have demonstrated significant relationships between circulating levels of thyroid hormones (THs) and exposures to environmental chemicals. In controlled animal studies, environmental chemicals have been shown to cause a reduction in serum TH levels, also supporting a causal association. In this article we review the role of THs in development and adult life, the impact of xenobiotics on thyroid status, the relationships between adverse outcomes of thyroid disruption and upstream causal biomarkers, and the societal implications of perturbations in THs by xenobiotic chemicals.

2THs include both thyroxine (T4) and triiodo-thyronine (T3). The independent regulation of circulating levels of these two forms of TH is complex, but in this review we refer generally to both forms as TH. THs are evolutionarily conserved molecules present in all extant vertebrates and some invertebrates. Molecular signaling pathways regulated by these hormones affect development, energy balance, and metabolism in all taxonomic groups. For example, TH induces metamorphosis in the sand dollar, flounder, and frogs, and TH is essential for development in birds and mammals. In humans, TH is important for normal development of brain, lungs, heart, and other organs. Likewise, the mechanism(s) by which THs exert their actions through nuclear receptors that influence gene expression is highly conserved across the vertebrate taxa.

3The regulation of serum TH levels and of TH action in various tissues involves a complex interplay of physiologic processes. Thyroid function depends on iodine uptake, TH synthesis and storage in the thyroid gland, stimulated release of hormone into and transport through the circulation, hypothalamic/pituitary control of TH synthesis, cellular TH transporters, tissue-specific TH deiodination, and degradation of THs by catabolic hepatic enzymes (Figure 1). Given the key role of TH for normal development and physiologic function in all vertebrates, it is important to identify environmental factors that may adversely affect thyroid function and/or TH signaling and to evaluate their ability to adversely affect public health. In addition, because of the highly conserved nature of TH chemistry, synthesis, signaling, and regulation, environmental factors that affect thyroid function or TH signaling in one species may well affect thyroid function or TH signaling in others—including humans.

4It is becoming clear that, although somatic and brain growth retardation occur with severe TH insufficiency, moderate or even transient TH insufficiency can cause specific developmental defects in rodents and in humans. Small differences (~25%) in point-estimates of maternal T4 during the early fetal period are associated with adverse outcomes (e.g., reduced IQ scores), even though these deficits do not constitute clinical hypothyroidism. However, in a hallmark study by, the Mental Development Index of children with congenital hypothyroidism was affected by the age of onset of treatment, rather than the serum free T4 concentration after treatment. Thus, the degree of TH insufficiency is not the only variable affecting human development; the duration of the insufficiency and the developmental timing of the insufficiency are also important and may vary by species, presenting a challenge for hazard assessment.

5Experimental work in animals provides strong support for the hypothesis that moderate TH insufficiency can alter development in rodents. Integrating data over a series of studies, a decrease in serum total T4 by 50% during the critical period for cochlear development was associated with a permanent hearing loss in adult offspring. found that less than a 30% decrease in serum total T4 in dams, for only 3 days, was associated with structural abnormalities in the brains of their offspring. An average decrease in serum total T4 of only 28% in 2-week-old pups given low doses of propylthiouracil was associated with marked reduction in cell density of the corpus callosum. found that a 28% reduction in circulating levels of T4 in rat dams produced significant adverse effects on synaptic function of the adult offspring despite no detected change in serum T4 levels in the pups after birth. Thus, these experimental findings confirm what has been observed in humans: small, even transient, decreases in serum total T4 are associated with altered brain development.

6It is important to recognize that TH concentrations are correlated with adverse effects in organ systems other than the nervous system, including the cardiovascular system and control of serum lipids, pulmonary system, and kidney. Total cholesterol, low-density lipoproteins (LDL), non-high-density lipoproteins (non-HDL), and triglycerides increased linearly with increasing thyroid-stimulating hormone (TSH), and HDL decreased consistently with increasing TSH across normal reference ranges without evidence of any threshold effect. Similar trends in lipid profiles were identified across clinical categories from hypothyroid to euthyroid to hyperthyroid individuals. Within the reference ranges for TSH, there was a linear positive association between TSH and both systolic and diastolic blood pressure (Figure 2). Intimal medial thickness, a measure of atherosclerosis and predictive of coronary vascular disease and stroke, was inversely related to free T4 after controlling for lipids, clinical factors, and thyroid autoantibodies. Some of these adverse effects were ameliorated by treatment with T4. Not surprisingly, deficits in thyroid homeostasis were associated with cardiovascular risk in multiple epidemiologic studies. A meta-analysis of 14 epidemiologic studies found an overall increase in risk of coronary heart disease of > 65% in those with subclinical hypothyroidism (elevation in TSH with normal T4). A higher relative risk was noted in those studies that adjusted for most cardiovascular risk factors, suggesting that confounding was not responsible for these effects. Treatment with L-T4 of patients with subclinical hypothyroidism resulted in improvements in cardiovascular risk factors, including total cholesterol and endothelial function (flow-mediated dilatation). found that treatment with T4 of hypercholes-terolemic individuals who have “high normal” TSH values significantly reduced both total and LDL cholesterol, additionally supporting a causal association. In addition, environmental exposure to the thyroid-disrupting chemical (TDC) polychlorinated biphenyls (PCBs) had an inverse association with T3 in men and was associated with both unfavorable lipid profiles and self-reported cardiovascular disease in men and women. Therefore, epidemiologic as well as mechanistic and therapeutic evidence substantiates the concern that TDCs may adversely affect cardiovascular risk in humans by reducing serum T4.

7TDCs are broadly defined as xenobiotics that interfere with TH signaling. These can include chemicals that alter the structure or function of the thyroid gland (e.g., perchlorate and methimazole), alter binding of hormones to thyroid receptors (e.g., bisphenol A, PCBs, and polybrominated diphenyl ethers), or alter regulatory enzymes associated with TH synthesis (e.g., propylthiouracil). A number of extrathyroidal mechanisms affect TH levels by altering binding to hormone transport proteins (e.g., hydroxyl-PCBs), hepatic clearance (e.g., PCBs, triclosan), inhibition of deiodination to T3 (e.g., FD&C red dye number 3), and receptor agonism/antagonism (e.g., tetrabromobisphenol A). The downstream consequences of these effects are to alter TH-directed transcription either directly or via changes in circulating or tissue concentrations of THs. Several uncertainties complicate basic risk assessment approaches when assessing the hazards of TDCs. These include defining the biomarkers used for assessing hazard, defining the magnitude of change in the biomarker(s) that reliably predict downstream adverse outcomes, intraspecies extrapolation that is hampered by a lack of mechanistic and dose response data, and predicting the effects of real life exposures to low-level mixtures of xenobiotics that contain components that individually have vastly different kinetic and dynamic properties.

8Several specific chemicals were shown to bind to TH receptors (TRs). This has important implications because there is good evidence that different effects of TH in the developing brain are mediated by different TR isoforms. There are two different classes of TRs (TRα and TRβ), and different chemicals can selectively interact with various isoforms. Thus, these chemicals will likely produce a mosaic of effects on TH signaling in the developing brain and may do so without affecting circulating levels of TH. It also may be challenging to develop high-throughput in vitro screens for TR binding because many of these screens use only the ligand-binding domain of the receptor, and there is some evidence that environmental chemicals can bind to an allosteric site on the DNA binding domain of the TR.

9The variety of mechanisms by which TDCs alter TH signaling (Table 1) provide a number of biomarkers that could be used in assessing hazard. These include molecular targets, which could be chemical-class specific, and downstream consequences, such as serum TH concentrations, brain morphology or biochemistry, or behavior. These changes may be either directly or indirectly related to TH action (Figure 3). Accurately and thoroughly assessing the health risks of thyroid disruption by environmental xenobiotics will require an improved understanding of how divergent mechanisms alter the relationship between serum THs and consequent adverse impacts on health.

10The most commonly used biomarker of effect for TDC exposure is serum total T4 concentrations. Although TSH is a well-accepted biomarker for hypothyroidism, a number of xenobiotics alter circulating TH levels but do not change TSH. Therefore, it is central to risk assessment to understand the relationship between perturbations in circulating concentrations of T4 and adverse effects. In addition, it is important to test the hypothesis that changes in circulating concentrations of T4 represent a common pathway by which adverse outcomes are produced. This hypothesis is consistent with the accepted role of circulating concentrations of T4 in defining thyroid disease. Many kinds of adverse effects are associated with either TH excess or insufficiency, depending on the timing, severity, and duration of the perturbation. Although the pattern of effects may differ, changes in serum TH are predictive of downstream adverse outcomes.

11Upstream biomarkers of TDC exposure are predictive of adverse effects if the mechanisms of action are well characterized. Mechanism 1 in Figure 4 illustrates this point: alterations in circulating THs during development are predictive of adverse neurodevelopmental outcomes. This concept has been known for decades and is the basis for newborn TH screening. These adverse consequences are well documented in animals for xenobiotics that alter circulating levels of TH.

12Although interspecies extrapolation of adverse effects of TDCs requires careful consideration, there are many situations in which the effects of a chemical in one species are similar to those in another, including in humans. For example, perchlorate competitively inhibits iodine uptake into the thyroid gland, with subsequent decreases in TH synthesis and declines in circulating TH concentrations. The kinetics for perchlorate inhibition of iodine uptake in humans and rats are extremely similar, indicating the homologous nature of the initial toxic event. However, species differences in the relationship between changes in serum total T4 and downstream adverse effects, perhaps mediated by differences in kinetics such as tissue TH concentrations and the sensitivity of specific developmental outcomes to low T4, cannot be ruled out at this time.

13For some TDCs, there may be little data to support cross-species extrapolation. Both in vivo and in vitro studies suggest that PCBs activate the pregnane X receptor (PXR) in rodents, which leads to up-regulation of hepatic catabolic enzymes and subsequent declines in circulating concentrations of T4. The steroid X receptor (SXR) is the human equivalent for rodent PXR, and there are species differences between PXR and SXR: Rodent PXR is activated by pregnenolone-16α-carbonitrile (PCN), but not by rifampicin, whereas human SXR is activated by rifampicin but not by PCN. In addition, in vitro data suggest that high concentrations of PCB-153 act as an antagonist at the human SXR. As well, species differences in circulatory transport proteins (e.g., transthyretin and thyroid-binding globulin) complicate extrapolation from animals to humans. Thus, species differences in the expression or structure of specific functional proteins (e.g., receptors and enzymes) may at times affect the toxicity of specific compounds in different species.

14Evaluating the potential for additive or synergistic (i.e., greater than additive) effects resulting from exposure to mixtures or environmental xenobiotics presents challenges for the assessment of endocrine disruptors. Additivity for mixtures of chemicals with a similar target is now a default assumption for some classes of chemicals. A variety of predictive models are available for use with mixtures of similarly acting chemicals. For example, the toxic equivalents methodology predicts the cumulative effects of aryl hydrocarbon receptor (AhR) agonists using dose addition. However, these models may not predict effects of mixtures containing chemicals with multiple mechanisms of action (e.g., synthesis inhibitors, low dietary iodine, hepatic catabolism). The small number of studies reporting effects of mixtures of TDCs lack, either by study design or statistical approach, the ability to test for additivity. The use of rigorous statistical models is critical for testing hypotheses of effect or dose addition and determining whether antagonism or synergism exists.

15tested a mixture of 18 TDCs (dioxins, dibenzofurans, and PCBs) for effects on serum T4. These chemicals were each known to decrease circulating concentrations of T4. The mechanisms by which these chemicals alter THs involve up-regulation of hepatic catabolic enzymes (e.g., uridine diphosphate glucuronosyltransferases). 2,3,7,8-Tetrachlorodibenzo-p-dioxin (TCDD), dibenzofurans, and dioxin-like PCBs activate a network of phase II and III proteins via binding of the AhR. The non-dioxin-like PCBs activate a slightly different set of enzymes (and possibly transporters) via binding to PXR and the constitutive androstane receptor (CAR). These differences in mechanisms of action (i.e., AhR agonists and CAR/PXR agonists) suggest that dose addition theory would not predict the effects of the mixture. A “flexible single-chemical-required” method demonstrated no deviation from dose additivity at the lowest doses of the mixture but a greater-than-additive effect at the highest mixtures doses (Figure 5). At high doses the dose-additivity model underpredicted the empirical effects by 2- to 3-fold but worked well at lower doses typical of environmental exposures.

16Future work is needed to improve the ability of mixtures models to account for the homeostatic processes that are activated by changes in both tissue and serum TH concentrations. The paucity of data in this area makes it difficult to determine whether these models will accurately predict changes in common downstream adverse outcomes after exposure to complex mixtures of chemicals that act on multiple upstream targets. Indeed, the effects of the complex mixtures will likely depend on the interaction of both kinetic and dynamic factors. Increasingly, it may become possible to identify interactions of chemicals in population-based biomonitoring databases. For example, sizable subpopulations for whom the relationship between perchlorate exposure and serum T4 concentrations are modified by coexposure to thiocyanate, nutrition (iodide consumption), and behavior (smoking) have been identified using the National Health and Nutrition Examination Survey database. Because additivity or synergy of TDCs with different mechanisms of action has been demonstrated, as noted above, a broad approach to cumulative risk that would account for these interactions seems appropriate. This is particularly true considering the limitations of current modeling methodologies.

17A critical issue affecting the interpretation of upstream events is the relationship between biomarkers captured in clinical or animal studies and specific adverse outcomes. Studies involving upstream biomarkers are most useful when these biomarkers have been causally linked to downstream adverse outcomes. For example, interpreting studies of perchlorate and T4 are relatively straightforward because the only known toxic effect of perchlorate is interference with thyroid function; thus, any effects of perchlorate on the nervous system are necessarily interpreted to be subsequent to a reduction in serum THs.

18Difficulties can arise when attempting to predict changes in upstream biomarkers based on adverse outcomes. For example, if the adverse outcome(s) of a specific toxicant or mixture is caused by more than one mechanism, then individual downstream outcomes (i.e., “effects”) are not diagnostic of upstream events, and causative links between a known exposure and outcome are difficult to discern. Figure 4 illustrates this by the alternative mechanisms activated by chemical X that may cause similar adverse outcomes. Indeed, some of these adverse outcomes may be caused by exposure to other chemicals (chemical Z). A key to using adverse outcomes in these cases is the use of patterns of outcomes that may be diagnostic.

19PCBs offer a good example of the problems associated with inferring upstream changes in THs as the causative agent of downstream neurotoxic outcomes. PCBs produce changes in a number of behavioral domains in humans and animals. They also affect multiple neurochemical pathways in addition to TH. Although changes in THs during development predict specific behavioral changes, effects of PCBs on some specific tasks in animals or outcomes in epidemiologic studies may not necessarily be attributable to changes in THs.

20Another example of the difficulty in linking serum TH to adverse outcomes is provided by the recent observation in humans of an abnormal TH profile in boys with a genetic mutation in the T3-specific transporter mono-carboxylate anion transporter 8 (MCT8). In all cases, serum T3 is elevated, but serum T4, free T4, and TSH may be low, normal, or elevated. Thus, the elevated serum T3 appears to be a biomarker of the MCT8 mutation among the patients evaluated, although it is not the only mechanism by which T3 can become elevated. In addition, all of the boys evaluated presented with severe psychomotor deficits, but it is unlikely that the elevated serum T3 itself was the root cause of their condition. Thus, environmental factors that influence T3 transport through MCT8 may represent a situation in which the profile of serum TH hormones is perturbed in ways that are not immediately recognizable as due to an endocrine disruptor, but may signal that adverse effects occur through a mechanism that interferes with TH signaling.

21Recognition of the role of “critical windows of exposure” in characterizing causal relationships between toxicant effects on serum THs and downstream adverse effects is critical. Specifically, the role of TH in brain development changes as development proceeds. Therefore, to establish a causal role of toxicant-induced low TH in the mechanism of neurotoxicity, it is important to show that T4 replacement can reverse the effects of toxicant. However, it is important to be cognizant of the relevant “windows” of vulnerability in the design of these experiments. For example, the impact of TH disruption on the development of auditory function in rats correlates well with circulating T4 levels during the second postnatal week. This is entirely consistent with the known role of THs in auditory development, the critical postnatal ontogeny of auditory function, and the pharmacokinetics of the chemicals tested. In addition, this correlation establishes a prognostic power of early postnatal T4 for adverse consequence of developmental exposure to TDCs in rats. An understanding of the role of THs in development, coupled with hormone level measurement during the critical window, allows the establishment of a developmental mode of action that assigns a key causative role to TH disruption in the adverse outcome (Figure 4).

22Studies designed to test for associations between toxicant exposures and circulating levels of TH in humans require careful consideration of confounding variables. For example, blood levels of TH vary among individuals, which will affect the number of samples required for such a study to be sufficiently powered to identify associations of interest. In the case of newborn TH levels, a number of maternal, infant, and delivery factors influence TH levels in cord blood and in infant serum, and these must be carefully considered when attempting to identify associations between toxicant exposures and serum TH levels. A good recent example is that of, who showed that PCB measures in cord blood were associated with circulating levels of TH only in those babies born via an unassisted vaginal delivery. Thus, these confounding variables may explain the studies in which PCB body burden has not been found to be associated with THs.

23There may be individuals within the general population who are more at risk than others (i.e., sensitive subpopulations). For example, because pregnancy causes an increased demand on the thyroid gland, pregnant women may be particularly sensitive to specific kinds of toxicants that produce an additional burden on the thyroid gland, such as perchlorate, or chemicals that activate liver metabolism of T4. Women in general appear to be more sensitive to the adverse effects of perchlorate, although it is not clear why. An estimated 7.3% of the U.S. population either have self-reported hypothyroidism or take thyroid medication, and three-quarters of these are women. More than 17% of those > 12 years of age report taking medications known to alter TH levels (e.g., estrogen, lithium, androgens). Those 50–79 and ≥ 80 years of age have a 2-fold and 5-fold increased risk of hypothyroidism, respectively, compared with those 12–49 years of age. These are examples of large sub-populations at risk with any additional exposures that affect thyroid homeostasis.

24The set-point around which THs are regulated is very individualistic, and differences between individuals in their set-point is largely determined by genetics. Epidemiologic studies have identified elevated risk of cardiovascular disease in patients with subclinical hypothyroidism, characterized by elevated TSH with normal T4. Many studies identify that TDCs are associated with decreases in T4 but not elevations in TSH. However, the low level of interference with thyroid homeostasis seen in subclinical hypothyroidism and with TDCs may be equivalent, suggesting that elevated risk of cardiovascular disease should be considered possible from exposure to TDCs. The variance in serum T3, T4, and TSH in individuals is about half of the range of population variance, known as the “reference range,” as shown for T4 in Figure 6. Therefore, a value within standard “normals” is not necessarily normal for the individual, and an elevated TSH (which responds with a logarithmically amplified variation to minor changes in T3 and T4) should be interpreted as indicating that serum T3 and T4 levels are not normal for the individual. Thus, it is highly likely that unidentified subpopulations exist that have particular sensitivity to thyroid disruption. The ability of epidemiologic studies to identify associations between thyroid disruptors and cardiovascular (or other) outcomes may be diminished as a result of failure to recognize risk in individuals who may have T4 levels in the normal population range but below their own normal individual range. Therefore, any exposure that would result in altered TH homeostasis in a population should be considered an adverse effect.

25The burden to society of even small changes in function should not be dismissed or underestimated. The consequences of developmental lead exposure provide an informative example of the effects of a small shift in the IQ of a population. Lead exposure has been widespread in the United States, although blood lead concentrations decreased from a mean toddler blood lead of 15 μg/dL to < 2 μg/dL over the past four decades with the introduction of nonleaded gasoline and other measures. A mean toddler blood lead of 15 μg/dL would be expected to decrease population IQ by ≥ 5 points. Although the consequences of a 5-point decrease in an individual’s IQ may be difficult to discern, the impact of this 5% shift at the tails results in a 57% national increase in those classified as mentally retarded (IQ < 70) and a concomitant decrease in individuals considered gifted (IQ > 130).

26Small decrements in maternal T4 or free T4 during the first trimester are associated with impaired neuropsychological development in the child. However, children born to women with moderately low TH identified in these studies largely fall within the lower portion of the normal range for measures of neuropsychological function. Although they have lower IQ as a population, their individual IQ is in the normal range.

27The cardiovascular consequences of disruption of thyroid homeostasis also potentially affect a large portion of the adult population. As noted above, there is a linear association between TSH (including through the normal reference range) and both blood pressure and cholesterol. The magnitude of these changes associated with changes in THs would be considered to confer minimal risk to an individual, even though the individual risk of myocardial infarction (MI) and death from MI increases linearly for increased systolic and diastolic blood pressure and serum cholesterol (Figure 7). There is an important distinction that needs to be recognized, however: the difference between individual (relative) risk and population-attributable risk. Typically, the medical community assigns specific values for blood pressure and cholesterol as “high” or “borderline” to advise individuals on individual health risk. However, as illustrated in Figure 7, most of the morbidity in the population as a whole is associated with lower rather than higher levels, because a higher percentage of the population falls within the low to moderate range.

28The population-attributable risk can be used to monetize the societal burden of exposure to chemicals that affect thyroid function. For example, the U.S. EPA estimated the effects of lead, which is associated with increases in both systolic and diastolic blood pressure, on cardiovascular function. The monetary burden of lost IQ associated with lead or methylmercury exposure has also been estimated at billions of dollars per year. Similar estimations could be made for the burden of exposure to chemicals that decrease THs and result in IQ deficits or increased incidence of cardiovascular disease. It is important to recognize that these outcomes are not only relevant if “abnormal” (e.g., mental retardation, clinically defined high blood pressure, or high cholesterol) but also relevant to outcomes in the “normal” range. Therefore, it is extremely important not to confuse the goal of minimizing population risk with arguments focused on individual relative risk.

29Two conclusions follow from the recognition that thyroid dysfunction affects multiple end points and that population-attributable risk is greater at levels associated with lower individual risk. First, from fetal life through old age, people are potentially vulnerable to adverse health effects as a consequence of exposure to TDCs. Second, any degree of thyroid disruption that lowers TH levels on a population basis should be considered a biomarker of increased risk of adverse outcomes. Because TH insufficiency in both humans and experimental animals results in serious neurodevelopmental and cardiovascular effects with large societal costs, chemicals with the ability to affect thyroid homeostasis should be carefully evaluated for potential population impacts. Finally, considering the complexity of the regulatory mechanisms affecting TH signaling and the variety of known TDCs that affect the thyroid system at different points of regulation, it will be essential to incorporate new information in human risk assessment strategies as it becomes available.