Discussion

We used data from a prospective pregnancy and birth cohort to compare the associations between fluoride exposures during different developmental windows and preschool aged children’s intellectual abilities. The GEE method advances our understanding of early-life fluoride neurotoxicity by formally comparing strength of associations across windows of exposure. The strongest association between fluoride and child IQ was observed between standardized MUF and age-normed PIQ (B = -2.36, 95% CI: -3.63, -1.08); the association was significant during infancy, B= -2.11, 95% CI: -3.45, -0.76), but negligible in childhood. Our results, which show that fetal fluoride exposure is more strongly associated with children’s intelligence than postnatal fluoride exposure, are consistent with a Chinese study examining different susceptibility windows of fluoride exposure; lower IQ was found in children whose mothers were exposed to high fluoride levels in drinking water (>1.0 mg/L) during pregnancy compared to those with high postnatal and low prenatal fluoride exposure (Xu et al., 2020). We did not identify clear differences between the effects of different trimester exposure windows on cognitive outcomes (Supplementary Table 1), and so it may be that the entire prenatal period confers susceptibility.

Critical windows of exposure may also differ by sex; animal and human literature have noted sex differences in response to fluoride exposure (Green et al., 2019; Green et al., 2020; Mullenix Debesten, Schunior, & Kernan, 1995) as well as several other environmental neurotoxicants (Comfort & Re, 2017; Torres-Rojas & Jones, 2018). When we tested sex differences across windows, our results suggested that prenatal fluoride exposure was a critical developmental window for boys for FSIQ and PIQ, whereas infancy was a critical developmental window for girls for PIQ. Specifically, boys showed a 4-point decrement in PIQ per 0.5 mg/L increase in MUF whereas girls showed a 2-point decrement in PIQ per 0.1 mg increase in IFI (effect estimates are shown based on approximate average values for MUF and IFI in our sample). While the effect of exposure in infancy was greater among girls than boys, the IFI by sex interaction for PIQ was not significant indicating that exposure in infancy is not associated with a statistical difference between boys and girls. After excluding outlying dyads, the adverse association between IFI and PIQ strengthened among girls (from B = -2.0 to B = -3.6), while this association among boys remained about the same (from B = -1.1 to B = -1.4).

Within animal research, a rat experiment similarly demonstrated an interaction between sex and fluoride exposure across developmental windows (Mullenix et al., 1995). Male rat pups were most sensitive to late prenatal exposure whereas female rats were most sensitive to exposure occurring in the postnatal (weanling) period. Exposed adult females also showed a lower threshold for behaviour deficits than exposed adult males. These findings are consistent with some (Baran-Poesine et al., 2013; Bera et al., 2007; Flace et al., 2010) but not all (Bartos et al., 2015; Jiang et al., 2014) rat studies examining sex-specific effects of prenatal exposure to fluoride. Further research is needed to examine sex-specific effects of fluoride neurotoxicity as many of the animal studies conducted to date have been identified as having a high risk of bias (NTP, 2016).

Boys and girls may respond differentially to neurotoxicants. Indeed, studies have shown that boys are often more vulnerable to early-life exposure to neurotoxicants than girls (Brubaker, Dietrich, Lanphear, & Cecil, 2010; Desrochers-Couture et al., 2018, Jedrychowski et al., 2009; Kern et al., 2017; Ris, Dietrich, Succop, & Berger, 2004; Pagalan, 2018; Singh et al., 2018; Torres-Rojas & Jones, 2018). While the biological mechanisms underlying sex-based differences of fluoride neurotoxicity are not well understood, disruption to maternal thyroid or sex hormone levels could potentially contribute to sexually dimorphic effects (Batista & Hensch, 2019). Fluoride may target the hypothalamic-pituitary-thyroid axis (Malin, Riddell, McCague, & Till, 2018; Bai, Huang, Wang, & Guo, 2020; Du et al., 2020), though we are not aware of any epidemiologic studies that have measured fluoride-induced changes in thyroid and sex steroid hormone levels in pregnancy. In addition, the timing of neurologic development of specific brain regions differs between the sexes (Perera & Herbstman, 2011; Lenroot et al., 2007), which might increase susceptibility of fluoride exposure during a particular developmental window. In the Mullenix et al. (1995) rat study, fluoride concentrations differed by sex in some brain structures (e.g. hippocampus), which could also contribute to sexually dimorphic changes in behaviour. See Green et al. (2020) for further discussion of mechanisms that may contribute to sex-based differences of fluoride neurotoxicity.

The difference in magnitudes and divergence in the direction of some of the associations between verbal and non-verbal intellectual abilities may have several progenitors that reflect these distinct types of cognitive ability. While we would not expect higher fluoride intake in infancy to be beneficial to VIQ, we would expect it to be detrimental to non-verbal (PIQ) intelligence. Fluid (i.e. non-verbal) abilities are more biologically determined whereas crystallized intelligence (i.e. VIQ) is more likely to be shaped by experience (Asbury et al., 2005; Luster & Dubow, 1992). Past studies have suggested that prenatal and early-life exposure to some neurotoxicants, like lead, is more strongly associated with non-verbal intelligence than verbal intelligence in young children (Bellinger et al. 1991; Dietrich et al. 1991, 1993; Factor-Litvak et al. 1999; Jusko et al., 2008; Wasserman et al. 1997). Consistent with this pattern, our findings showed a decrement of IFI to PIQ (statistically significant decrease of 1.6 points per 0.1 mg/day), but not VIQ (non-significant increase of 1.0 points per 0.1 mg/day).

Our current results are consistent with and extend our previous findings. The effect of MUF on FSIQ was significant for boys (2.45-point decrement in FSIQ per 0.5 mg/L increase in MUF; Table 2), reproducing our prior work (Green et al., 2019) in which we found a 2.2-point decrement in FSIQ per 0.5 mg/L increase in MUF. We note that the current analysis did not include city in the analysis because fluoride intake from formula (i.e. IFI) is a function of residential water fluoride concentration and was therefore deemed redundant. Our finding of a 2.1-point decrement in PIQ per 0.5 mg/L increase in IFI (B = -2.11. 95% CI: -3.45, -0.76) was also consistent with our prior finding that infancy is a critical period for non-verbal intelligence in boys and girls (Till et al., 2020). Our current results extend our prior work by showing that regardless of child sex or the exclusion of influential dyads, the associations of fluoride on PIQ differs across exposure windows. However, exposures do not significantly associate with IQ outcomes once city is controlled and FDR is applied.

A 2- to 4-point decrement in PIQ may seem like a small difference at the individual level. However, a small shift in the mean of IQ scores at the population level translates to millions of lost IQ points given the ubiquity of fluoride exposure. The impact of such a shift has a disproportionate effect among vulnerable populations who are at the lower end of the population IQ distribution because the loss in productivity per IQ point is not the same across the entire IQ distribution (i.e. a drop in IQ from 80 to 77 is not the same as 120 to 117) (Rose, 1985). Finally, previous benchmark dose analyses for testing lead and fluoride neurotoxicity have selected 1 IQ point as the benchmark response because of the significant societal and economic burdens of reduced IQ (Budtz-Jorgensen et al., 2004).

Strengths of the present study include the relatively large sample with repeated exposure measures during pregnancy, infancy, and early childhood that resulted in precise estimates of effects, as reflected by narrow confidence intervals. We used the FDR method to guard against false positive conclusions due to multiple comparisons, even though multiplicity control is rarely imposed when evaluating multiple predictors in regression-based models (Cribbie, 2017). We adjusted for numerous potential confounders and avoided problems with collinearity among critical windows of fluoride toxicity by using GEE. Although several epidemiological studies have applied GEE to test critical windows of environmental contaminants on neurobehavioral outcomes (Jackson-Browne et al., 2018; Stacy et al., 2017; Vuong et al., 2017; Zhang et al., 2017), this is the first study to use GEE to model critical windows of fluoride toxicity.

Limitations of our study include modeling marginal effects of fluoride exposures without controlling the effects from other exposure windows or assessing cumulative fluoride exposure, which may be more etiologically relevant. However, it would not be realistic to estimate partial effects that vary one exposure window while fixing other exposures. Another limitation is not having MUF, IFI, or CUF levels on all study participants, although we were able to incorporate cases with incomplete data in the GEE analyses. In any research on single neurotoxicants, simultaneous exposure to other environmental contaminants may confound effect estimates. For instance, trace amounts of aluminum can bind fluoride and affect cellular processes (Li, 2003). Moreover, there is always the possibility of residual confounding. We considered many potential confounders in prior research conducted in the same sample examining the association between MUF and child IQ (Green et al., 2019) and they did not influence our findings. We also controlled for several other chemicals in our prior analyses including lead, mercury, PFOA, arsenic, manganese, and second-hand smoke exposure. Controlling for these chemicals did not affect our estimates appreciably. The demographic characteristics of our sample also constrained our ability to test potential fluoride susceptibility in different subpopulations. For example, fewer than 3% of women smoked in the first trimester of pregnancy and 89% of the sample was Caucasian, which limited our ability to assess effect modification by smoking or race. Further, MUF concentration averaged across three trimesters was the strongest predictor of IQ scores among boys and was more reliable than IFI and CUF. Fluoride concentrations measured in single spot urine samples (i.e. trimester-specific MUF and CUF concentrations) suffer from measurement error due to the rapid elimination kinetics of fluoride (half-life in urine < 6 hours; Ekstrand, 1983) and lack of control for water/beverage consumption and dental product use prior to urine sampling. IFI may also suffer from measurement error due to using mother’s self-report of infant water intake and breastfeeding duration, and our reliance on water fluoride measurements made at water treatment plants as opposed to measuring fluoride directly in household tap water. While we did not have specific information on the type of water used to reconstitute formula (i.e. bottled/filtered versus tap water), we derived IFI only for children of women who reported drinking tap water. However, these possible sources of measurement error are more likely to produce negatively biased effect estimates than positively biased estimates (Budtz-Jorgensen, Keiding, & Grandjean, 2004).

Our findings raise the question of whether a decrease in children’s cognitive abilities is worth the benefit that fluoride ingestion provides. To answer this question, we need to consider how and when fluoride works for the developing child and pregnant woman. Fluoride prevents dental decay by being present in the mouth when a decay-inducing acid attack occurs, by precluding minerals from leaving the dental enamel during the attack (prevention of demineralization) and by incorporating into the enamel after the acid attack (promotion of remineralization). These processes only occur after teeth have erupted (CDC, 2001; Ten Cate and Buzalaf, 2019) Fluoride incorporated into enamel before eruption has a minimal effect on the prevention of dental decay (CDC, 2001; Takahashi et al., 2017). In contrast, there is potential risk of reduced IQ associated with fluoride exposure during fetal and infant development. Consistent with this conclusion, the Center for Disease Control and Prevention does not recommend the use of fluoride supplements during pregnancy (CDC, 2001). If a pregnant woman chooses to decrease her ingestion of fluoridated water, which accounts for 75% of her fluoride intake (CDC, 2001) or not drinking tea or eating foods high in fluoride, common alternatives for minimizing risk of dental decay in pregnancy include reducing sugar intake and using topical fluorides, such as fluoridated toothpastes and varnishes.

Given a heightened sensitivity of the developing brain to environmental toxicants, identifying critical windows of vulnerability to fluoride exposure is essential for promoting child health. Our results suggest the associations of prenatal and postnatal fluoride exposure with cognitive development may be modified by sex, though further replication of this finding is needed. These results indicate that it is important to balance the risks of fluoride exposure during early brain development with its potential to prevent caries, especially for pregnant women and infants.