Neurodevelopmental toxicity of prenatal polychlorinated biphenyls (PCBs) by chemical structure and activity: a birth cohort study

Polychlorinated biphenyls (PCBs) are ubiquitous environmental toxins. Improper disposal has been a major source of environmental contamination. Their production and use were banned in most industrialized countries in the late 1970s because of toxic effects in wildlife [1, 2].

PCBs have been shown to have toxic effects on various organs including tissues of the nervous, reproductive, and immunologic systems [3–7]. Although there is growing evidence from in vivo and in vitro studies to support the hypothesis of adverse effects of PCBs on neurodevelopment [6–9], the mechanisms are not well understood. Additionally, various epidemiological studies have found an association between PCB exposure and decreased cognitive development [10–15].

The most well-known mechanism related to adverse health effects such as immune suppression, hepatotoxicity, and thymic atrophy is aryl hydrocarbon (Ah) receptor-mediated pathways for dioxin- like PCBs [16, 17]. Since non-dioxin-like PCBs have shown low affinity for the Ah receptor [18], they have been regarded as potentially less toxic. However, neurotoxicity [19], carcinogenicity [20, 21], and changes in hormones [22] have also been described as resulting from non-dioxin-like PCBs.

Alteration of sex-steroid hormone homeostasis by PCBs has been studied, and estrogen-like activity of PCBs was reported to change brain dopamine (DA) concentration [23, 24] or aromatase activity [25]. Gonadal hormones such as estrogens can alter the structure and function of the hippocampus, which is critical for spatial and declarative learning and memory, and they modulate neural circuits by interacting with various neurotransmitters [26]. Besides playing a major role in sex differences, including sexual dimorphic behaviors, they modulate hypoxia-induced injury of neurons in the hippocampus [27] and influence synaptic patterning [28]. Estradiol protects against free radicals and regulates apoptosis [29], and these effects might depend on brain regions and timing of exposure. When bound to estrogen receptors, the complex regulates transcription[30]. PCBs have been proposed to affect brain development through an estrogen-receptor mediated pathway [31]. Specific steroidal pathways that might mediate PCB effects on cognitive and motor development have yet to be determined.

Previously, several epidemiological studies have evaluated the effects of PCBs on neurodevelopment. If only the Bayley Scales of Infant Development (BSID) assessment is considered, findings across studies are inconsistent. A Michigan study in a fishing community [32] and a study from 12 sites across the United States [33] found no association between BSID scores at 8 months and prenatal PCB exposure. In contrast, studies from North Carolina [34, 35] and the Netherlands [14] observed prenatal PCB exposure to be associated with lower psychomotor development index (PDI) scores at 3, 6, 12, and 24 months whereas a study in Germany [15] found PCBs in breast milk, used as a surrogate measure for prenatal exposure, to be associated with decreased mental development index (MDI), but not with psychomotor development index (PDI), at 7 months of age. Several previous studies that used other instruments than the Bayley scales at different ages reported that PCBs were associated with detrimental effects on children's neurodevelopment [10, 12, 13, 36, 37].

Because of the lengthy half-lives of PCBs, any measurement in serum represents an accumulation of exposure over a period of years. These compounds circulate in the serum lipids and cross the placenta, albeit less efficiently than their hydroxy-metabolites [38]; nonetheless, the consistency of cord:maternal ratios across exposure levels provides the rationale for use of maternal concentrations as a surrogate for fetal exposures. The present study investigated associations between prenatal PCB exposures, measured from maternal and cord sera, characterized by their chemical structures and hormonal activities, and neurodevelopment measured by BSID among 16-month-old infants in eastern Slovakia.

Additionally, each of the 6 major PCB congeners (#118, #138, #153, #156, #170, and #180) was tested in separate regression models, and maternal and cord concentrations of PCB #118 were most strongly associated with reduced BSID.

We observed higher concentrations of two mono-ortho PCBs in maternal sera to be associated with decrements in neurodevelopment at 16 months of age. Several lab-based studies reported alteration of dopamine (DA) levels, changes of behavior, or a reduction in long-term potentiation (LTP), regarded as the cellular and molecular mechanisms that are linked to some forms of learning and memory, by coplanar dioxin-like PCBs such as congener #77 or #126 [60–64]. However, other studies in rats reported that the coplanar dioxin-like PCB #126 was not associated with deficits in attention and behavioral performances [65, 66].

Regarding epidemiological studies, although the birth cohort study from the Netherlands found an association of coplanar and mono-ortho dioxin-like PCBs (# 77, 126, 169, 105, 118, 156) in breast-milk collected two weeks after delivery with suboptimality at birth, their findings were null at 7, and 18 months for the PDI and MDI except they found a significant association with decreased PDI at 3 months [14, 67, 68].

Reports from Germany indicated reduced MDI at 7, 18 and 30 months in relation to prenatal PCBs (the sum of PCB #138, #153, and #180), although the 18-month scores were somewhat less precise (p = 0.06) as compared with the other time points [13, 15]. Our study didn't show associations with either non-dioxin like PCBs or the sum of PCB #138, #153, and #180 even though the concentration of the sum of PCB #138, #153, and #180 was higher than that of German study (1.16 vs 0.55 ng/ml) [15]. It is not certain if our inconsistent findings with the German study were from differences between the two studies in other nonquantitated PCBs or in other chemical compounds that correlated with the measured non-dioxin like PCBs, or for other methodologic reasons.

A study with 6-month-old children in Japan [69] reported no statistically significant association between prenatal non-dioxin like PCBs (#170, #180) and the Bayley scores while some dioxin-like PCBs (#118,#126, #157, and #189) showed suggestive associations with reduced PDI (p < 0.1) but not MDI. Compared to our levels of maternal PCBs, median concentrations in this study of PCB #118, #156, #170, and #180 were lower by a range of approximately 2 to 13 fold. Some research on non-dioxin-like PCBs [24, 70] suggests that they tend to show more potency for neurotoxicity than dioxin-like PCBs through various mechanisms: interference with protein kinase C(PKC) [70, 71], changes in DA levels [23], and apoptosis [72]. In contrast to some other epidemiological or lab-based studies [13, 15, 24, 70], we found the evidence was weaker for a relationship between neurodevelopment and prenatal exposures to di-ortho-substituted PCBs, although a suggestive association was observed with PDI (β = -.195, p = 0.09) from these compounds measured in cord, but not in maternal samples (Table 3).

Both mono-ortho and di-ortho substituted PCBs can affect nervous system tissues, but these effects are possibly mediated by different pathways [73]. In a related publication, we evaluated the hydroxylated metabolites of PCBs in relation to the scores on the Bayley Scales and reported that 4-OH-PCB107 was the only metabolite that showed any relationship to neurodevelopment at 16 months [74]. Both maternal and cord serum concentrations of 4-OH-PCB107 were associated with deficits on the MDI, and cord levels were also associated with deficits on the PDI. As 4-OH-PCB107 is the metabolite of the two congeners PCB #105 and PCB #118, both being mono-ortho substituted, the two investigations present a consistent signal: increased prenatal exposures to both the parent mono-ortho PCB compounds and the hydroxylated metabolites of those mono-ortho PCBs occur in association with lower neurobehavioral scores. The question therefore arises: through what mechanism might one or more of these compounds, but not the di-ortho substituted congeners and corresponding metabolites measured in our cohort, act?

Mechanisms hypothesized to underlie PCB neurotoxicity have included thyroid disruption [75], interference with sex steroids either as agonists or antagonists [64, 76], and Ah-receptor activity. Thyroid hormones are critical to neuronal proliferation, migration, synaptogenesis, and brain myelination [77]. It has long been hypothesized that one mechanism for neurodevelopmental toxicity of PCBs is through disruption of thyroid hormones [78], and more recently, hydroxylated metabolites have been implicated in this mechanism because of greater transplacental transfer than the parent compounds [79], high affinity for transthyretin, a transport protein, and accumulation in fetal tissues, particularly 4-OH-CB107 [80–82]. A recent in vitro study showed that PCB 105 and/or 118 act as thyroid hormone receptor agonists, but only in the presence of PCB 126 induction of CYP1A1 [83]. The lack of a similar effect on thyroid receptors from PCB #138 and #153 conforms with our results, supporting plausibility of this mechanism for 4-OH-CB107, the common metabolite of two mono-ortho PCB #105 and #118, in impairment of neurodevelopment. Disruption of thyroid hormones may itself operate via altered cell signaling and/or neurotransmitters [78].

We did not find significant associations of neurodevelopmental scores with anti-estrogenic PCBs (sum of #138, 156, 170 and 180), but did not evaluate possible roles of PCB metabolites with regard to estrogenic and/or anti-estrogenic activities in this paper. As mentioned above, in our previous work examining six hydroxylated metabolites [74], only 4-OH-CB-107 was significantly associated with decreased Bayley scores. Interestingly, an in vitro study that examined inhibition of an estrogen inactivator by OH-PCBs and subsequent increases in estrogen levels at the target tissue, reported 4-OH-CB-107 to be one of strongest inhibitors of the human estrogen sulfotransferase (hEST), an enzyme that catalyzes the sulfation and inactivation of estrogens. The hEST activity was 1.5 to 2 orders of magnitude greater than that observed for 4-OH-CB-146 or 4'-OH-CB-172 [84], metabolites of PCB 138, 153, and PCB 170, 180 respectively. Thus, despite previous mixed reports of anti-estrogenicity [85] or estrogenicity [86] for 4-OH-CB-107, (as well as similar observations for estrogenic [51] and anti-estrogenic [87] activity for the parent PCB 118), a possible mechanism of neurodevelopmental toxicity is through hEST induction of estrogenic activity, rather than binding directly to the estrogen receptor. This could provide a coherent explanation for our observations linking deficits in neurodevelopment to PCB 118 and/or its metabolite, 4-OH-CB-107, but not to the di-ortho congeners or their OH-metabolites.

Ah-receptor activity as shown for non-ortho substituted PCBs, is another plausible mechanism for the associations we observed. On the one hand, dioxin-like activity of mono-ortho PCBs occurs only at much higher concentrations than that of coplanar PCBs or of 2,3,7,8-TCDD; on the other, congeners #118 and #156 are present, in virtually every human population studied, at levels from 4 to 5 orders of magnitude greater than 2,3,7,8-TCDD itself (e.g., in U.S., European, and Japanese studies [88–91]). Because of the exceedingly low levels of the coplanar PCBs, furans, 2,3,7,8-TCDD and other dioxins and the high cost to quantitate them, we were unable to measure these compounds in cord and maternal sera. However, in adult women aged 20-39 from the same population, i.e., the same two Slovak towns, we have calculated TEQs using the WHO 2005 revised TEFs for mono-ortho PCBs (0.00003) for the sum of PCB #118 and #156 (geometric mean TEQ, 1.36 pg/g lipid), and they are over four times higher than the TEQs for 2,3,7,8-TCDD itself (0.32 pg/g lipid). In other words, the lower toxicity was more than compensated by the far higher abundance, supporting relevance of dioxin-like activity from mono-ortho PCBs in this population.

A cautionary note is in order, however: TEFs are calculated based on intake and are not considered applicable to circulating levels in human tissues [17, 92]. The WHO 2005 report [17] calls for studies that would provide TEFs for tissue concentrations, but at present this research has yet to be conducted. Whether the actual TEQ in this study is higher or lower than we calculated remains unclear, but the shorter half-life of PCB #118 in comparison with that of 2,3,7,8-TCDD indicates relatively greater intake would be required to maintain blood levels, while the half-life of PCB #156 may be similar to that of 2,3,7,8-TCDD [93]. The TEQs for intake of the two mono-ortho PCBs together would therefore likely be larger than what we have calculated based on circulating concentrations. Also notable is a recent effort to develop a Neurotoxic Equivalence Scheme that would take into account multiple mechanisms [94]. Another concern about the calculation of dioxin-like potencies in epidemiologic studies relates to the difficulty in extrapolating from rat to human, given that large differences in AhR-mediated gene expression and EROD activity are observed even comparing rats with mice [95].

Further concerns have focused on the presence of impurities in test samples, specifically 2,3,7,8-substituted polychlorinated dibenzo-dioxins (PCDDs), -furans (PCDFs), or PCB 126 [17]. Our inability to measure trace quantities of these compounds precludes drawing inferences about the role of impurities, although other data (personal communication with Dr. Kocan, 2009) from women of roughly the same age range in these two towns demonstrate low correlations of the sum of PCB #118+ #156 with 2,3,7,8-TCDD (0.016, p = 0.94) and with coplanar PCBs (0.26, p = 0.23) and negative correlations with PCDDs (-0.31, p = 0.15) and PCDFs (-0.29, p = 0.17). Thus, the associations we observed for neurocognitive developmental deficits with the sum of the two most abundant mono-ortho PCBs are unlikely to be a result of confounding from these other compounds.

The subset with cord PCBs measured had similar sociodemographic characteristics to the larger group with maternal PCBs, and sensitivity analyses did not indicate selection bias. Since the mono-ortho-substituted PCBs were present at relatively lower concentrations in cord than maternal serum, whereas di-ortho-substituted PCBs were at similar levels, these findings may indicate that PCBs measured in the cord are more potent, and/or a better marker of what reaches the target CNS tissue in the fetus.

A large refusal rate (47 percent) for participation and attrition (29 percent) of the follow-up represented a study limitation. The eligibility of those who refused is unknown, but given our exclusions based on residence in the district, parity, and age, a substantial proportion of the refusals may have been ineligible. Comparing those lost to follow-up vs. those who stayed in the study, lipid adjusted levels of 4 major PCBs (#138, #153, #170, and #180) were not different (0.57, and 0.59 ng/mg lipids, respectively) implying that the study losses were not related with exposure levels. On the other hand, there were some differences in regard to socio-demographics (data not shown): those lost to follow-up included more Romani (28.0 vs. 18.4 percent), more Svidnik residents (37.0 vs. 26.7 percent), more participants with basic schooling only (28.0 vs. 18.6 percent), and more women of low parity (49.2 vs. 38.3 percent) compared to those who returned for the 16 month follow-up visit. BSID scores for those lost to follow-up were, by definition, not obtained in our study. Nevertheless, our final model included adjustments for HOME and Raven's scores, which relate to the infant's nurturing environment at home and the mother's IQ respectively. If missingness was at random within any given level of the adjustment variables, bias from non-random loss of subjects would be less likely.

The means of MDI for the two psychologists were similar (93.1 for Michalovce, n = 559; 95.7 for Svidnik, n = 201), but lower than the mean scores of 100 in the populations for which they are standardized. Because the BSID has not been normed in the Slovak language or culture, some cross-cultural or language differences might have played a role in the lower MDI scores in our study [96]. Our research objective was to examine possible associations between prenatal PCBs and BSID scores within this study cohort. Therefore, while comparisons of BSID scores with other populations might not be valid, the absolute scores are not the focus of our investigation.

For reliability of the BSID scores, good agreement on the MDI (ICC: 0.84) between raters was obtained. However, we were not able to evaluate reliability on the PDI because the child frequently moved out of range of the video camera during the test. Also, we observed a large difference in mean scores of PDI, approximately 10 points, between two districts. This suggests the possibility of poor reliability across raters, and/or some other unmeasured variables related to district that might have resulted in the substantial difference in PDI. Therefore, we also examined the effects of maternal dioxin-like-PCBs in each district separately (results not shown). In Michalovce, where the psychologist is more experienced, we still observed significant associations, with similar magnitude as for the entire sample, between mono-ortho substituted PCBs and the Bayley scores; in Svidnik, the district with a less experienced psychologist, we did not find any association with maternal mono-ortho substituted PCBs. Overall, PCB exposure distributions substantially overlapped for these two districts, however, mean PCB levels were significantly higher in Michalovce (which had a longer right tail) than in Svidnik (Figure 1). Adjustment for district is required because it was a design variable, but would not have constituted over-adjustment for PCB concentrations.

As mercury and lead are well-known neurotoxins, we considered the possibility of confounding. However, in a sub-sample of the cohort, correlations between these heavy metals and dioxin-like PCBs didn't exist (correlation coefficients were either negative or less than 0.03, and none of them were significant). Although fish can be a major source of methyl mercury, in other work from our Slovak birth cohort, mothers' fish consumption was very low, contributing a negligible amount of fat to the diet, and hence would not be expected to be an important source of PCBs [97].

Multiple comparisons can be a problem in the study because we tested 36 various combinations of PCBs and outcomes. While some of our results could have been due to chance, it is not possible statistically to determine which ones are just random variations and which represent true underlying associations. P-values are based on the assumption of the universal null hypothesis being correct, whereas a priori knowledge motivated several of the hypotheses we examined, specifically those related to different congener classes; in this situation, multiple comparisons can be justifiable as part of a careful and thorough evaluation of patterns of associations [98]. Notably, our significant results were not sporadic, as would be expected if they were chance findings, but rather showed a pattern whereby dioxin-like congeners showed the strongest associations in all analyses: cord or maternal serum, MDI or PDI. Furthermore, the pattern is upheld by our analysis of hydroxy-metabolites of PCBs: of the six that we evaluated, the only one associated with neurobehavioral deficits was 4-OH-PCB107, the metabolite for PCB #118 and another mono-ortho congener [74].

Several strengths are worth emphasizing: the large sample size; direct exposure assessments of prenatal PCBs rather than using a surrogate such as breast-milk or fish consumption during pregnancy; good quality control of laboratory measurements. Unfortunately, not many previous epidemiological studies have examined functionally-defined subsets of PCB congeners, and fewer still have done so in relation to neurodevelopment. Also, we were able to measure only a limited roster and some were present at very low levels relative to our limits for quantitation, which prevented a more definitive assessment of associations with regard to neurodevelopment in children.

A change in a few points on the Bayley scales may seem clinically unimportant for an individual child. However, beginning with studies from the population level, it can be predicted that a slight shift in the mean performance in the adverse direction can result in a substantial increase in the number of children who are in the subnormal range in the clinical setting [99].

Based on the results herein, the high levels of mono-ortho PCBs, and the literature on dioxin-like activity in relation to neurodevelopment, we conclude that the associations of prenatal dioxin-like mono-ortho PCBs, in either maternal or cord serum, with decreased performance in both the cognitive and motor domains based on BSID scores could plausibly be causal. However, a suggestive association of di-ortho-substituted PCBs with decreased motor development was found in cord but not maternal serum, and we did not find statistically significant associations with anti-estrogenic PCBs.