Manganese Exposure from Drinking Water and Children’s Academic Achievement

2.1 Overview

This cross-sectional study is a component of an ongoing, prospective elementary school-based intervention study for lowering arsenic exposure from drinking water in Araihazar, Bangladesh. The study area is adjacent to a previously described study area for a larger cohort study of adults (Ahsan et al. 2006) consisting of three unions of Araihazar upazilla located about 25 km southeast of the capital city Dhaka. Araihazar has an area of 183km² and contains 12 unions, the smallest administrative units in Bangladesh, which each consists of 10–15 villages. Every family in Araihazar typically lives in a house made of tin, mud, hay and in some cases concrete. Several houses are clustered together to form a bari representing a small segment of the community (sometimes an extended family). We report on 840 children enrolled in an ongoing school intervention study at 14 elementary schools.

2.2 Selection of Schools and Participants

We initially identified 27 schools in three unions (Haizadi, Uchitpur and Khagkanda) of Araihazar within reasonable travel distance to the field clinic where our project offices are located. Schools were selected for participation based on three eligibility criteria: 1) ten or more age-appropriate children (8–11 years) in each classroom, 2) all teachers agreed to participate and 3) schools agreed to provide us with the academic performance records of the participating children. Children were recruited from all 14 schools in the three unions that met the selection criteria.

Our field staff visited each of the 14 schools and obtained a list of all 1925 enrolled students with their addresses. We randomly selected 9–12 children per village, and conducted home visits to begin enrollment. Homes of 952 potential participants were visited, continuing until our targeted sample size of roughly 800 children were enrolled. Inclusion criteria restricted enrollment to children aged 8–11 years who attended school in an age-appropriate grade, had no known physical disability or chronic illness, and were not twins. Of the 952 children whose families were visited for eligibility review, we were unable to locate 75, either because the family had moved (n=12) or because no one was available at the time of the visit (n=63). Twenty-seven children were older than the specified age, and 10 attended school only infrequently. Altogether, 840 children agreed to participate. By design, measures of urinary As (UAs) were available on half the sample (n=420), whereas measures of water As and Mn (WAs, WMn) from the home well were available on all.

2.3 Procedure

Prior to conducting this study, we secured approval from Institutional Review Boards at Columbia University Medical Center and the Bangladesh Medical Research Council, and obtained written informed consent from parents and school-teachers, as well as child assent. Once parental consent and child assent were obtained, the field team collected socio-demographic information during a structured interview and observation in home visits, at which time well water and urine samples were collected. Children’s height, weight, head and arm circumferences were measured during the visit to our health clinic. The field team visited the schools and subsequently met with teachers to obtain the performance records of each child in the most recent annual school-wide district tests in both language (Bangla and English) and mathematics. Teachers were blind to children’s household well As and Mn status.

2.3 Measures

2.4 Outcome Assessment: Academic Achievement

Elementary school children are evaluated in three disciplines in Bangladesh: Bangla, English as a second language, and mathematics. We obtained the annual scores from the academic achievement records of the schools. These scores are based on national uniform tests given to students in participating public and private schools. Scores are reported as percent correct and range from 0 to 100. Language tests evaluate children’s memory skills; for example, “Write the first four sentences of rhyme ‘X’”, “Translate the following sentences into English”, and “When was Mr. Y (a famous writer) born?” In contrast, mathematics test questions evaluate analytical skills and problem solving abilities, for example, calculations relying on addition and subtraction, or word problems.

2.5 Statistical Analyses

Preliminary analyses found the distributions of both WMn and WAs to be skewed and the logarithmic transformations were used in the analyses. Summary statistics were calculated to describe the sample. Spearman correlation coefficients were used to estimate bivariate association between markers of exposures. To account for correlations in children’s academic achievement scores when they were with the same primary class teacher, we employed linear models using repeated measures. Models were estimated both with and without adjustment for potential confounders. Children’s academic achievement test scores with the same primary class teacher tended to be correlated and the models accounted for within teacher correlations. To identify the control variables, we first examined associations between socio-demographic and maternal variables with the outcome variables. Variables were considered potential confounders if they were associated with either Mn exposure or outcome. Potential confounders were included in the final model if their exclusion changed the estimated regression coefficient between Mn exposure and outcome by half its standard error.

WMn was also categorized into five categories, one for below WHO standard (i.e. 400 μg/L) and four for higher exposure levels with approximately equal numbers of subjects in each. We repeated the models using dummy variables to describe the categorization. When the coefficients of adjacent categories were similar, the categories were collapsed to yield parsimonious models. We also tested for WAs by WMn interaction in the full model. Finally, we estimated a piece-wise linear function with an a priori knot at WMn of 400 μg/l to examine the associations below and above the safe water standard.

3.1 Sample Characteristics

Characteristics of included children are described in Table 1. More female students than male students were included. On average, children had attended school for approximately a lifetime total of 30 months, although they were currently attending classes as second through fifth graders. About half the parents had no formal education. Participants were exposed to expectably high levels of As and Mn in drinking water: mean WAs and WMn concentrations were 13 and 3 times the WHO standards of 10 and 400 μg/l, respectively. On average, teachers had 10 years of teaching experience. The average test percentage scores for each of the three academic areas were approximately 50%.

3.2 Relationships among measures of exposure and outcomes

Measures of As exposure in drinking water (WAs) and urine (UAs) were positively and significantly correlated (Spearman correlation coefficient r = 0.68, p<0.0001). WAs was also significantly correlated with WMn (r=0.22, p<0.0001). The three measures of academic achievement were also positively correlated (r’s between 0.70 and 0.75, all p <0.0001).

3.3 Associations between sociodemographic factors and academic achievement

Associations between sociodemographic factors (i.e., school-grade, head circumference, maternal education and paternal education) and all academic outcomes were in the expected directions (Table 2). Fourth and fifth grade students had significantly lower academic scores than second graders. Achievement test scores increased as head circumference increased. Children whose mothers and fathers attended middle school or beyond had significantly higher achievement test scores than those whose parents had no formal schooling. No other sociodemographic variables were associated with the outcomes when added to the models.

3.4 Associations between exposure markers and academic achievement

The covariate and WAs adjusted mean test scores by WMn categories were similar for WMn>400 μg/L, suggesting a threshold effect (table 3). High exposure categories of WMn were therefore combined. Covariates and WMn adjusted mean test scores in all WAs categories were similar indicating that WAs was not related to any of the three test scores. The results also suggested a dichotomization of the exposure variables on the basis of the WHO standards. Table 4 presents the associations between the exposures and the children’s academic achievement test scores with dichotomized WAs and WMn.

4.1 Mn neurotoxicity and working memory

Animal studies suggest that certain neurotransmitters have particular impact on memory and learning, including glutamate, γ-amino butyric acid (GABA), dopamine, acetylcholine, serotonin and norepinephrine (Myhrer 2003). Animal models suggest that Mn affects neurotransmitters in the dopaminergic and glutamatergic systems (Tran et al. 2002) as well as serotonin binding (Velez-Pardo et al. 1995). For example, animals dosed with various concentrations of Mn postnatally commit more errors and exhibit learning deficiencies in radial arm mazes compared to control animals (Kern et al. 2010). Animal studies also support the view that brains of young animals have the ability to achieve higher Mn concentrations than those of adults (Dorman et al. 2000; Moreno et al. 2009).

Animal studies have identified brain compartments, such as basal ganglia, prefrontal cortex, cerebellum and hippocampus where Mn primarily accumulates (Bock et al. 2008; Guilarte et al. 2006; Rose et al. 1999; Schneider et al. 2009; Yamada et al. 1986) resulting in detrimental effects on working memory processes and learning. Other research with non-human primates has documented similar impact of Mn on both spatial and non-spatial working memory (Schneider et al. 2006; Schneider et al. 2009).

Importantly, Mn exposure results in similar types of deficits in occupational epidemiological studies of welders, where multiple investigations consistently demonstrate detrimental effects on immediate, short-term and long-term memory functions (Bowler et al. 2007; Chang et al. 2009; Chang et al. 2010; Lucchini et al. 1995; Lucchini et al. 1999). Working memory contributes to children’s learning capacity. Compared to IQ measures, working memory better predicts academic achievement and overall learning (Alloway and Alloway 2010). In a longitudinal study, poorer working memory skill in kindergarten children was associated with lower academic attainment in reading, spelling, mathematical reasoning and number operations at age 10–11 years (Alloway and Alloway 2010). Recently, in a separate cohort of Bangladeshi children, we observed decreased WISC-IV working memory scores in 10 year old children drinking from household wells with elevated levels of WMn (Wasserman et al. 2011).

4.2 As and Academic Achievement

We did not observe associations between measures of As exposure and academic achievement. Toxicity of As occurs through oxidative stress leading to neuronal injury (Larochette et al., 1999), changes of neurotransmitter levels by affecting basal ganglia (Rodriguez et al., 2003), decreases of superoxide dismutase (Modi and Flora, 2007) and glutathione-related enzyme activities (Kannan and Flora, 2004). Recent work has also proposed detrimental effects of As through disruption of neuron cytoskeletal network (Giasson et al., 2002; Kunz et al., 2004; Vahidnia et al., 2008). Our own research group at Columbia University has also reported lower scores on intelligence tests in children in relation to As exposure (Wasserman et al., 2004; Wasserman et al., 2007). However, evidence of the effect of As on working memory of children, a stronger predictor of academic achievement at later stage in life is not convincing in the literature. Our group has recently found significant effect of As on working memory of the Bangladesh children in statistical models before adjustment for important sociodemographic variables known to be related to intelligence scores; this association became non-statistically significant after control (Wasserman et al. 2011). Thus, we anticipated that the effect of As on a functional outcome like academic achievement is less likely to be observed at this age group even though As is considered as a potential neurotoxicant. In addition, to the best of our knowledge, no systematic laboratory study has been done to show the effect of As on different brain compartments that are associated with learning. We therefore propose that that As alters cognitive and neurological functions but has minimal impact on learning capacity and academic achievement. Alternatively, As toxicity may show latent effects on learning.

4.3 Limitations of the Study

Our study results may have encountered limited “geographic generalizability” since the study population may represent only comparable communities with similar sociodemographic characteristics. Our findings may not be generalizable to children living in urban communities. The cross-sectional design of this study can be considered as another limitation as it can hinder cause and effect inferences. The lack of tests for working memory is another limitation with respect to interpreting the results.

The design of the prospective educational intervention study from which this cross-sectional study evolved allowed collection of urine and water samples for measuring exposures among participants. Therefore, no Mn biomarker of dose except WMn was available. Blood manganese (BMn) reflects only recent exposure and therefore, may not reliably indicate total body burden of Mn (Bouchard et al. 2011). Although two recent studies reported associations between hair manganese (HMn) and child intelligence (Bouchard et al. 2011; Menezes-Filho et al. 2011) a recent study found significant association in girls only (Riojas-Rodriguez et al. 2010). Thus literature indicates that defining optimal Mn biomarker of dose remain an open question.