The precise classification of suboptimal thyroid status requires measurement of more than one hormone. For example, low free T4 with normal TSH reflects hypothyroxinemia . Elevated TSH with normal free T4 reflects subclinical hypothyroidism. In the present study, where we used TSH by itself as a screening measure, we saw no indication that PFOS was associated with abnormally high TSH. Rather, the data suggested an association within the normal range of TSH, and in the absence of a measure of free T4, the clinical importance is unclear. Because we saw no association with abnormally high TSH nor with the presence of a history of thyroid disease, our generally null results were reassuring and consistent with a subtle association that may reflect only a biochemical curiosity. However, the possibility exists that our study did not have enough statistical power to detect a small association with abnormally high TSH, and that there is a small effect on public health (increased hypothyroidism).
In a study of 44 Korean pregnant women with serum measures of PFAS and several thyroid hormones in cord blood , inverse associations were found between maternal PFOS and neonatal total T3, and between maternal perfluorotridecanoic acid and neonatal total T3 and T4. In addition, maternal PFOA was associated with increased cord TSH (n = 31). However, the median maternal PFOS concentration (2.93 [IQR: 2.08, 4.36] ng/mL) was much lower than that in our study. In a case–control study of hypothyroxinemia among Canadian pregnant women (96 cases, 175 controls), the authors found no association with serum concentrations of PFOA, PFOS and PFHxS. In that study, the geometric mean of PFOS (7.39 ng/mL)  was lower than among women in the present study (12.77 ng/mL).
In animal experiments, treatment with PFAS generally does not affect serum TSH levels [14, 34]. When effects on TSH are seen, they tend to be transient decreases after dosing . PFASs appear to displace thyroxine from transthyretin, causing a transient increase in free T4 . The end effect may be a reduced total T4 with adequate free T4 . Data from animal models also indicate other potential mechanisms by which PFAS may affect thyroid hormone metabolism . Thyroid metabolism in rodents, however, is different than in humans, partly because of transthyretin being a more important transport protein in rodents. Furthermore, during human pregnancy, human chorionic gonadotropin and estrogen affect the TSH feedback mechanism , further complicating comparison across species.
Compared with women from other western countries, the median concentrations of PFOS and PFOA in this study were slightly lower [21, 37, 38]. For TSH levels, the median in our study was higher than that in other studies of women in the second trimester [39, 40]. The difference may explained by the use of different immuno-metric assays. TSH levels can differ by as much as 1.0 μIU/mL because different monoclonal antibodies may recognize different TSH isoforms circulating in the blood [36, 41]. Iodine deficiency also affects TSH levels, but is rare in Norway . The prevalence of self-reported thyroid disease in the present study (2.2%) was a little higher than that among pregnant women in another recent study from Norway .
We found concentrations of PFOA and PFHxS were both correlated with PFOS, suggesting they might come from common sources. However, they were not associated with TSH levels statistically and PFOA showed a different direction of association. These substances were usually present lower at levels than PFOS, and have different binding capacities with transthyretin due to different structures .
The observed association between PFOS and TSH was not present in the subfecund women. Although thyroid disorders are related to reduced fertility , we did not find a significant difference in TSH levels between subfecund women and controls. Subfecundity may mask the association, if any, between PFASs and TSH levels.
Our study had several limitations. Unfortunately, we did not have access to sufficient plasma volume to perform any other measurements of thyroid hormones, such as total T4, free T4 and T3. TSH, however, is a sensitive marker of thyroid status and reflects the physiologic log-linear relationship of TSH to free T4 , and is often used by itself in screening tests. Plasma TSH levels can reflect mild thyroid functional impairment even when plasma T4 and T3 concentrations are within normal ranges . Second, because our study was cross-sectional study, reverse causation could affect the results. For example, if low thyroid function led to reduced glomerular filtration rate , this could increase circulating PFAS concentrations . Third, as TSH levels change during pregnancy, a single measurement may not adequately characterize thyroid homeostasis during pregnancy. Fourth, although we considered some known potential confounders, there may be unmeasured confounders not taken into account. For example, the effects of other chemicals, such as polychlorinated biphenyls (PCBs), on TSH, can not be ruled out, given that fish intake is a source of both PFASs and PCBs. Finally, the participation rate in MoBa was low, raising the possibility of selection bias. Nilsen et al. compared general characteristics, exposures, pregnancy complications, and birth outcomes between MoBa participants and all women giving birth in Norway and found differences in prevalence estimates in most variables (Additional file 1: Table S1) . For the subjects in the present study we tabulated their characteristics using the same format that Nilsen et al. used (Additional file 1: Table S1). The present subjects were generally similar to all MoBa participants, though some minor differences were apparent, as would be expected due to the subgroup sampled because of a long time-to-pregnancy. In spite of the differences between all women giving birth in Norway and MoBa participants, Nilsen et al. found estimates of exposure-outcome associations were similar in the whole population and the MoBa participants , and we expect this generalization holds for the present analysis, in which inverse-sampling weights were used.