Neurobehavioral Function in School-Age Children Exposed to Manganese in Drinking Water

1Manganese is widespread in the environment. It is an essential element, required for the function of many enzymes and involved in oxidative stress protection, as well as in the formation of connective tissue and bone. However, inhaled manganese is a potent neurotoxicant with well-documented effects in workplace settings. Recent epidemiologic studies suggest that manganese exposure from drinking water is associated with poorer cognition and behavioral problems. High levels of manganese in water (MnW) are common in groundwater, because this element leaches from manganese-bearing minerals and rocks into the aquifers.

2Studies in Bangladesh among children exposed to high MnW reported associations with lower IQ scores, impaired perceptual reasoning and working memory, poorer academic achievement scores in mathematics (but not language), as well as higher scores of internalizing and externalizing problems. In addition, we reported findings from two separate investigations conducted in Quebec (Canada) at MnW levels much lower than those in Bangladesh. In our pilot study, hair manganese levels (MnH) were associated with hyperactive and oppositional behaviors in children, and in the epidemiological study that followed, MnW was associated with lower IQ scores. On average, we estimated that there was a 6.2-point difference in IQ between children in the lowest and highest MnW quintiles.

3A neurobehavioral test battery is often employed in studies aiming to identify neurotoxic effects of environmental exposures. This assessment typically results in a large number of scores used as outcomes, raising the problem of multiple testing. To avoid this issue, investigators often select a few test scores thought to be the most sensitive to detect neurotoxic effects associated with a given contaminant. This selection, however, is difficult because different studies often report slightly different findings. Although these apparent discrepancies could result from differences in the tests employed, they could also be attibutable partly to measurement errors in the neurobehavioral assessment. A better approach would be to consider test scores as measurements of an underlying neurobehavioral function, and use several scores as indicators for this function. Structural equation modeling (SEM) can be used to implement this approach, in which several different but related scores are used as indicators of a more global neurobehavioral domain, resulting in a more comprehensive assessment of the domain. In addition, this approach greatly reduces the number of outcomes (because several test scores are aggregated), therefore avoiding inference errors arising from multiple comparisons.

4We conducted a cross-sectional study among 375 children exposed to a relatively wide range of MnW levels in Quebec (Canada), where the presence of manganese is naturally elevated. We previously reported lower IQ scores with higher MnW, MnH, and manganese intake from water ingestion. In the present study, we report the relation between these same manganese exposure indicators and neurobehavioral functions (memory, attention, motor function, and hyperactivity) in these children.

5Study design and recruitment. This cross-sectional study was conducted in eight municipalities located in southern Quebec (Canada) in 2007–2009. Municipalities were considered as potential study sites if their aqueduct was supplied by groundwater, and were selected to achieve a gradient of MnW. However, we did not limit our recruitment to children living in houses connected to the aqueduct, and many participating children lived in a house with a private well (we had no information on the MnW in private wells before recruitment). Children were recruited through elementary schools in the selected municipalities, and enrollment was restricted to children who had lived in the same house for > 3 months to ensure that the measured MnW was representative of the water children had been consuming for at least this duration. A total of 375 children 6–13 years of age participated in the study. The Human Research Ethics Board of the Université du Québec à Montréal approved the study protocol, and parents signed an informed consent.

6Manganese in water and hair. We collected a water sample directly from the kitchen tap in each home. For homes that had a point-of-use filter attached to the tap, we collected one sample of filtered water, and a second sample with the filter removed. We used the following procedure to standardize tap water sampling: a) open the tap for 5 min, b) close and leave untouched for 30 min, and c) collect first draw. We added 0.15 mL nitric acid (50%) to the 50-mL water sample and stored samples at 4°C. Manganese and other metals (arsenic, copper, lead, iron, and zinc) were measured by inductively coupled plasma mass spectrometry (ICP-MS) at the Environmental Chemistry Laboratory of McGill University (Montreal, Quebec, Canada). Further details about analytical techniques and quality assurance and control procedures can be found in the study by.

7For a subset of participants (n = 30 houses), we repeated tap water sampling four times (once per season) during 1 year to examine seasonal variability of MnW. Analysis of these data indicated an intraclass correlation coefficient of 0.91, indicating very little temporal variability in MnW concentrations.

8Duplicate hair samples of approximately 20 mg were taken from the occiput of each child, cutting as close as possible to the root. We used the 2 cm closest to the scalp to measure manganese and other metals by ICP-MS in the laboratory of co-author S. Sauvé at Université de Montréal. Children were excluded from analysis when they reported using hair dye in the preceding 5 months because it could influence manganese hair content. When manganese concentrations for certified hair material were outside of the designated concentrations, we excluded the measures from the analyses; we therefore had valid MnH measures for 313 children. When there was only one available measure (n = 45), due to contamination or insufficient hair for duplicates, the sole measure was retained for the statistical analyses; for the others, the mean of duplicates was used (n = 268). Further methodological details can be found in Supplemental Material, “Measurements of manganese hair concentration.”

9Manganese intake from water ingestion. During the home visit, we orally administered a semiquantitative food frequency questionnaire to the parent and the child to assess manganese intake from water consumption, including direct water ingestion and water incorporated in food preparations (e.g., juices made from concentrate, soups) during the preceding month. We estimated water consumption from different sources—bottled, tap, tap filtered with a pitcher, and tap with an attached filter. For each water source, the amount consumed was multiplied by the measured or estimated concentration of manganese. We then divided the estimated intake by the weight of the child to derive an estimate of monthly intake in micrograms per kilogram. For untreated tap water and tap water treated with a carbon filter attached to the tap, the measured concentrations were used in the calculation. We conducted experiments to assess manganese removal efficacy for water filtered by a pitcher with activated carbon and an ion-exchange resin filter, showing an average manganese removal efficacy of 74%, which we used in our calculations. For bottled water, we attributed a concentration of 0 μg/L based on a survey of bottled water in Canada that reported a median < 1 μg/L for manganese concentration.

10Assessment of neurobehavioral function. Each child completed the following test battery: California Verbal Learning Test–Children’s Version (CVLT-C), Conners’ Continuous Performance Test II Version 5 (CPT II), Digit Span, Santa Ana Test, and manual Fingertapping. The Wechsler Abbreviated Scale of Intelligence was also administered, and the relation with manganese exposure was published elsewhere. The Conners’ Rating Scales were completed by a teacher (CRS-T) and a parent (CRS-P).

11Potential confounders. We collected information from the mother on socioeconomic status indicators and other variables (e.g., alcohol and tobacco consumption during pregnancy), and we assessed maternal nonverbal intelligence with the Raven’s Progressive Matrices Test and maternal symptoms of depression with the Beck Depression Inventory-II (BDI-II). Potential confounders were chosen a priori based on established associations and/or plausible relations with the outcomes, and included child’s sex, age (years, continuous), maternal education (not completed high school/completed high school/some college/some university), nonverbal maternal intelligence (Raven score, continuous), family income (nine categories with Can$10,000 increment between two categories, entered as an ordinal variable), maternal depressive symptoms (Beck-II score, continuous), and lead concentration in tap water (log10 transformed). In addition, we used sensitivity analyses to assess the influence of additional adjustment for tap water arsenic concentration (log10 transformed), water source (aqueduct or private well), child’s test administrator, birth weight, and maternal smoking and alcohol consumption during pregnancy (entered one at the time in the models).

12Statistical analysis. We used SEM to examine relationships between exposure to manganese and children’s neurobehavioral functions. We built three models using different variables to quantify manganese exposure (Figure 1). First, we constructed a latent manganese exposure variable that is an underlying construct of measured MnH. In this model, we postulated that internal manganese loading is manifested by MnH and that this also depends on manganese intake from water ingestion. Second, given that there is no consensus that hair is a good biomarker of exposure to manganese, we used MnW as the latent variable for manganese exposure. For the third model, we used the estimated intake of manganese from water ingestion as the indicator for the exposure latent variable to account for variation in the amount of water consumed by each child and the child’s weight.

13Children’s scores on the neurobehavioral tests were used as indicators of four latent variables representative of underlying neurobehavioral domains: memory, attention, motor function, and hyperactivity (Figure 1). We constructed these latent variables using a confirmatory factor analysis allowing good discriminant validity between the factors. We also took into account the response modalities of outcome scores in accordance with SEM guidelines, especially for normality and the number of unique values. To correct for local dependence when the correlation between the indicators could not be fully explained by the underlying latent variable, we allowed measurement errors of several outcomes indicating the same neurobehavioral function to correlate. We tested goodness of model fit using several indices (see Supplemental Material, Table S1).

14The distributions of manganese exposure indicators (MnH, MnW, manganese intake from water consumption) and water lead concentrations were skewed, and were therefore log10 transformed. The results are presented as the adjusted change in the latent variables for neurobehavioral functions (β) associated with a 10-fold increase in manganese exposure indicators. Because β has no unit, we also present standardized association estimates, expressed as the change in percent of standard deviation (SD) of neurobehavioral function scores associated with a 1-SD increase in log10-transformed manganese exposure indicators. Some test scores were missing for some children: CPT II (n = 9), Digit Span (n = 14), Santa Ana (n = 2), CRS-P (n = 1), and CRS-T (n = 53). We imputed missing data using the full information maximum-likelihood method. In sensitivity analyses, we restricted our models to children with no missing data, and we included additional covariates in the models (listed in “Potential Confounders”). Finally, we explored effect modification by sex using multigroup SEM analyses.

15Because SEM models data linearly, we also used generalized additive models (GAMs) with penalized smoothing regression splines to detect possible nonlinear relations. We extracted the scores from the factor-analytic component of the SEM and used them as outcome variables, adjusting for covariates. Manganese exposure indicators were introduced as spline functions. We examined departure from linearity using analysis of variance to assess the difference between the model with manganese levels introduced as a spline function and the model with manganese levels introduced as a linear term. To estimate thresholds for effect, we looked for the manganese level above which the estimated adjusted change in function was lower than the 95% confidence interval (CI) lower bound of the intercept (i.e., neurobehavioral score at the lowest manganese concentration).

16All the outcomes variables were continuous; however, because some of them were not normally distributed, we used maximum-likelihood estimation with robust (Huber-White) standard errors. The threshold for statistical significance was set to 0.05 (two-sided tests). We used the lavaan and mgcv packages in R (R Foundation for Statistical Computing, Vienna, Austria).

17Finally, to assess the consistency of the SEM approach with a more traditional approach, we used general linear models to estimate the associations between manganese exposure indicators and individual test scores, adjusting for the same set of confounders.

18Descriptive statistics. Table 1 shows the characteristics of children in the study, as well as levels of manganese exposure indicators. Almost all the children were white (99%) (data not shown), and 76% of mothers had at least some college education. Thirty-five percent of children resided in their current residence from birth, and most children (66%) had resided for > 5 years in their present home, based on residential history collected from parent. About half of participating children lived in a house connected to the aqueduct, and the other half in a house with a private well (51% and 49%, respectively). Tap water MnW ranged from 1 to 2,701 μg/L, with an arithmetic mean of 99 μg/L and a geometric mean (GM) of 20 μg/L, whereas estimated manganese intake from water ingestion ranged from 0 to 1,059 μg/kg/month with a GM of 5.5 μg/kg/month. There were 4.3% of homes with water levels of manganese above the U.S. health reference value of 300 μg/L, and 43.7% exceeding the aesthetic level of 50 μg/L for manganese in drinking water, above which the taste, smell, or color of water might be impaired. Water levels of arsenic and lead were elevated (≥ 10 μg/L) for 4% and 0.5% of children, respectively (data not shown). The Pearson correlation of MnW with other elements was 0.68 (iron), 0.26 (zinc), 0.11 (copper), 0.06 (arsenic), and –0.02 (lead).

19In univariate analyses, there was no significant difference in MnW with respect to children’s sex and age, family income, maternal education, maternal depression, and maternal intelligence (Table 1). However, MnW was significantly higher in children living in homes connected to the aqueduct than in children living in homes with a private well (geometric mean of 54.0 vs. 8.2 μg/L, p < 0.001). Univariate associations with estimated manganese intake from water ingestion were consistent with those for MnW (data not shown).

20Children’s MnH ranged from 0.1 to 20.7 μg/g with an arithmetic mean of 1.4 μg/g and a GM of 0.8 μg/g. In univariate analyses, MnH levels did not vary significantly with respect to children’s sex or age, or with maternal education or depressive symptoms (Table 1). However, MnH was higher in children living in homes connected to the aqueduct than in those living in homes with a private well, and MnH was also higher in children with family’s income ≤ Can$50,000 than in those with a higher income (both at p = 0.02). In addition, MnH differed significantly between maternal intelligence score categories (p = 0.04) (Table 1), but this association was no longer significant after additional adjustment for mother’s test administrator (p = 0.17) (data not shown).

21SEM. Table 2 describes the factor loadings and estimated correlation of measured variables to each of the latent variables for neurobehavioral functions when MnH was the exposure (model 1). All test scores had good factor loadings and were adequate as indicators of latent variables, and this result was similar when exposure was based on MnW and estimated Mn intake from water ingestion (models 2 and 3, respectively; data not shown). Higher scores for memory, attention, and motor functions indicate better performance, but higher scores for hyperactivity suggest more problems related to hyperactivity.

22In the first model, which was based on MnH, a 10-fold increase in MnH was significantly associated with differences of –3.6 points (95% CI: –5.2, –2.0) on the memory function and –4.2 points (95% CI: –6.7, –1.7) on the attention function (both at p < 0.01), after adjustment for confounders (Table 3). In standardized results, a 1-SD increase in log10-MnH was associated with lower memory (–24% SD; 95% CI: –36, –12%) and attention (–25% SD; 95% CI: –41, –9%) functions. No significant association was found between MnH and motor function (p = 0.57). The scores for hyperactivity tended to be lower (indicating fewer hyperactivity problems) for higher MnH with a difference of –2.0 points (95% CI: –4.6, 0.6; p = 0.13) for a 10-fold increase in MnH (Table 3).

23In the second model, where MnW was the exposure indicator, a 10-fold increase in MnW was associated with a significant difference of –1.0 point (95% CI: –1.6, –0.4; p < 0.01) of the memory function and a nonsignificant difference of –1.2 points (95% CI: –2.7, 0.3; p = 0.11) of the motor function. In standardized results, a 1-SD increase in log10-MnW was associated with memory function lower by –14% SD (95% CI: –24, –4%). No significant association was found between MnW and attention or hyperactivity functions.

24The third model showed a significant association between estimated manganese intake from water consumption and motor function. A 10-fold increase in manganese intake was significantly associated with a difference of –1.3 points (95% CI: –2.4, –0.2; p = 0.02) on the motor function. In standardized results, a 1-SD increase in log10 manganese intake was associated with motor function lower by –11% SD (95% CI: –21, –0.4%). The scores for memory tended to be lower for higher manganese intake with a difference of –0.4 points (95% CI: –0.9, 0.1; p = 0.13) for a 10-fold increase in manganese intake from water ingestion (Table 3). We found no significant association between manganese intake from water consumption and and attention function or hyperactivity.

25We assessed whether sex modifies the associations between manganese exposure indicators and neurobehavioral functions, and found that the association estimates for boys and girls were very similar (data not shown).

26Separate models for each neurobehavioral function resulted in estimates similar to those of the model taking into account all four functions at the same time (data not shown). Likewise, the model including only children without missing data and the model with additional adjustment for water arsenic concentrations, birth weight, maternal smoking, and alcohol consumption during pregnancy resulted in very similar effect estimates (data not shown).

27GAMS to examine shape of the exposure–response relations. We explored the shape of the exposure–response relation with GAMs for the constructed neurobehavioral functions, as shown in Figure 2 for MnH, Figure 3 for MnW, and Figure 4 for estimated manganese intake from water ingestion. MnH was significantly associated with poorer memory and attention (p < 0.001) with no significant departure from linearity (p = 0.1 for memory and attention; Figure 2A and 2B, respectively). For motor function, the association was significant (p = 0.02) and departed significantly from linearity (p < 0.01), with a slight increase at concentrations between 0.3 and 0.8 μg/g, and an apparent decrease in scores at MnH > 10 μg/g, but there were very few observations with such high levels (Figure 2C). The association for hyperactivity was not significant (p = 0.22) (Figure 2D).

28In the GAM, children had significantly lower scores of memory with higher MnW (p < 0.001), and the relationship departed significantly from linearity (p = 0.045), with a steeper slope at levels > 100 μg/L than at lower concentrations (Figure 3A). No significant association was found between MnW and attention (p = 0.1) (Figure 3B). For motor function, the association with MnW was significant (p < 0.001), and departed significantly from linearity (p < 0.001), with a threshold indicating that scores decreased more steeply at concentrations above 180 μg/L (Figure 3C). No significant association was observed for hyperactivity (p = 0.47) (Figure 3D).

29With GAM, estimated manganese intake from water ingestion was significantly associated with memory function (p = 0.038) in the GAM, and the function departed significantly from linearity (p = 0.04) (Figure 4A). No significant association was found for attention function (p = 0.88) (Figure 4B). For motor function, there was a significant association with manganese intake from water ingestion (p < 0.001) with no significant departure from linearity (p = 0.99) (Figure 4C). Finally, there was no significant association between manganese intake from water ingestion and hyperactivity (p = 0.44) (Figure 4D).

30Supplemental Material, Table S2, shows estimated associations from general linear models of individual neurobehavioral test scores. The results were very consistent with the findings from SEM and GAM analyses.

31We report that higher levels of exposure to manganese are associated with poorer performance of memory, attention, and motor functions, but not hyperactivity, in children. We estimated log-linear relationships between MnH and memory and attention functions and between manganese intake from water ingestion and motor function. The relationships for MnW exhibited a nonlinear shape, with steeper decreases in memory and motor functions at MnW levels > 100 and > 180 μg/L, respectively. These findings complement the previously reported association between the same three manganese exposure indicators (MnW, MnH, and manganese intake from water ingestion) with lower IQ scores in this same group of children. Furthermore, they are consistent with recent studies reporting adverse associations of manganese exposure with scores on tests of cognitive and motor functions. We did not observe a significant association with hyperactivity as previously reported, but MnW levels in the present study were considerably lower [GM for present study: 20 μg/L, compared with ≈ 300 μg/L, and ≈ 900 μg/L].

32Several mechanisms might underlie the association between manganese exposure and neurobehavioral outcomes. Studies have shown that manganese accumulates in the basal ganglia, white matter, and cortical structures. In animal models, manganese exposure disrupts the dopaminergic, glutamatergic, and serotonin systems that are essential to optimal cognitive functioning. Changes in gene expression (i.e., amyloid beta precursor-like protein) and markers of neurodegeneration in the frontal cortex (i.e., copper homeostasis dysregulation) induced by chronic manganese exposure may also underlie neurobehavioral changes.

33The present study has several strengths. We thoroughly assessed potential confounders, adjusting for several socioeconomic status indicators, as well as maternal intelligence and depression symptoms, and water lead concentrations. Furthermore, our results were robust to additional adjustment for water arsenic concentration, water source, child’s test administrator, birth weight, and maternal smoking and alcohol consumption during pregnancy. SEM allowed the simultaneous use of several indicators to better assess neurobehavioral functions, as well as integration of different exposure indicators. In addition, the SEM approach addresses issues arising from multiple testing and missing data that may not be adequately considered by standard regression analyses. The SEM findings were compared with the more traditional approach of analyzing each test score separately, and the findings were consistent.

34There is no consensus on the best biomarker to assess exposure to manganese. Blood and urine manganese content have been shown to correlate poorly with manganese exposure. In population studies, MnH concentration has often been used as a biomarker of exposure. Significant correlations between exposure to manganese and MnH concentration have been reported in several studies including ours, supporting the contention that hair is a valid biomarker of manganese exposure. As the hair grows slowly, the composition reflects a time-weighted exposure over several months, which is appropriate since long-term exposure is postulated to cause neurotoxic effects. However, the use of hair as a biomarker has been criticized because of potential external contamination that is not removable by washing the samples before analysis, which in the present case could result from bathing or showering in manganese-contaminated water.

35There are several limitations to this study. The reported associations could be attributable to unmeasured confounders, but MnW did not vary with socioeconomic factors, thus reducing the potential for confounding. Nonetheless, possible residual or additional unmeasured confounding by other factors cannot be ruled out. However, our findings are not likely to be explained by anthropogenic contaminants, because in our study area, the contamination of water by manganese results from natural processes associated with the bedrock geology, not human activities. There are no industrial sources of manganese emission in the study area, and the gasoline additive methylcyclopentadienyl manganese tricarbonyl (MMT) has not been used in Canada since 2004. showed very low atmospheric manganese concentrations in rural areas of Quebec, with a mean of 0.005 μg/m3; this is 10 times lower than the U.S. EPA inhalation reference concentration of 0.05 μg/m3.

36The cross-sectional design of the study limits the ability to draw strong causal inferences, although reverse causality is improbable. It is not known whether exposure during a critical developmental period is responsible for our observations. The majority (66%) of participating children had been living in the same home for at least 5 years; therefore, exposure levels are representative of several years. Finally, our assessment of children’s exposure to manganese from drinking water was based on a single measurement; however, repeated measurements in the same houses showed very little time variability in MnW levels, suggesting that the value measured on one occasion is representative of long-term exposure.

37The World Health Organization (WHO) has recently discontinued its 400-μg/L guideline for manganese in drinking water, arguing that “this health based value is well above concentrations of manganese normally found in drinking water”. In fact, drinking-water supplies with manganese concentrations exceeding this level are found in many countries, exposing tens of millions of people worldwide. Coupled with our previous report of MnW-associated IQ deficits, the present findings suggest the potential for harmful effects at levels commonly encountered in groundwater.