Associations between time-weighted postnatal methylmercury exposure from fish consumption and neurodevelopmental outcomes through 24 years of age in the Seychelles Child Development Study Main Cohort

Abstract

PURPOSE:

Methylmercury (MeHg) is a known neurodevelopmental toxicant in sufficient dosage and is universally found in fish. Current fish advisories for children are based on epidemiology studies examining prenatal exposure with a premise that MeHg exposure resulting from children eating fish could also be neurotoxic and have long-term consequences. However, the evidence that this assumption is true is limited. We investigated postnatal MeHg exposure from regular fish consumption using time weighted Hg measurements to determine if there are neurotoxic consequences.

METHODS:

We examined 85 neurodevelopmental outcomes measured from ages 9 to 24 years in the Seychelles Child Development Study Main Cohort (n=312–550) and examined their association with time-weighted measures of postnatal MeHg exposure in childhood and early adulthood. Postnatal MeHg exposure measured in the first cm of participants’ hair samples collected at seven evaluations were used to create two time-weighted (TW) average MeHg exposure metrics, one for childhood (TW-C) and the other for early adulthood (TW-A). TW-C was based on Hg measures at three ages between 6 months and 5.5 years, and TW-A was based on Hg measured at up to four ages between 17 and 24 years. We examined the association between each of these exposure metrics and the neurodevelopmental outcomes using linear regression with adjustment for covariates known to influence neurodevelopmental outcomes.

RESULTS:

There were 14 statistically significant associations between a postnatal metric and an endpoint. Six were associated with the TW-C and eight with the TW-A. Thirteen were adverse. Only the TW-C association at 9 years with the Bender Gestalt error score showed improvement. TW-C was adversely associated at 9 years with the Continuous Performance Task risk score, at 22 years with the Boston Naming Test (BNT) total and no cues scores, and at 24 years with the Test of Variables of Attention (TOVA) auditory response time variability and visual response time mean on the logarithmic scale. TW-A was adversely associated at 17 years with the Wisconsin Card Sorting Test % total errors, the Woodcock-Johnson passage comprehension, and the CANTAB rapid visual information processing false alarms, and at 22 years with the BNT total and no cue scores, the CANTAB rapid visual information processing false alarms and the intra-extra dimensional shift total errors and trials.

CONCLUSION:

These findings suggest that postnatal MeHg exposure may be adversely associated with neurodevelopmental outcomes in early adulthood. However, the associations are statistical and of unknown, if any, clinical significance. The results need confirmation in other cohorts.

1. Introduction:

Globally billions of people depend daily on the nutritional properties of fish. Fish also contain small amounts of methylmercury (MeHg), which is acquired naturally from the environment. Methylmercury is a known neurodevelopmental toxicant at sufficient exposure, but the dosage where adverse associations begin is presently unknown. Organic mercury poisoning from consumption of fish with very high mercury levels occurred in Japan following industrial pollution over 50 years ago (Yokoyama, 2018). During these poisonings, MeHg had its greatest neurotoxic effect on the developing brain with minimal or no adverse effects on the mother. Experimental evidence supports the impact of MeHg exposure being greatest on the developing brain (Spyker et al., 1972). Consequently, most subsequent research on low-dose MeHg exposure from fish consumption has been focused on the prenatal period. However, epidemiological studies of populations studying only fish consumption in the absence of industrial pollution and not including whales or other sea mammals in their definition of seafood have not reported adverse associations (reviewed by (Hibbeln et al., 2019)).

Current recommendations of fish consumption for pregnant mothers, women of childbearing age, and children are based on epidemiological studies examining prenatal exposure. Despite the absence of clear evidence of harm from consuming fish, regulatory agencies have issued fish consumption guidelines based on the precautionary principle (FAO/WHO, 2011; FDA/EPA, 2019). Evidence that postnatal MeHg exposure results in harm is very limited, but the human brain is known to mature slowly, and its development continues well into adolescence and early adulthood (Shonkoff et al., 2000; Tau and Peterson, 2010). Therefore, the brain may be susceptible during postnatal development. Aside from studies taking place in the Seychelles islands, only three other longitudinal cohort studies have reported associations between postnatal MeHg exposure and child neurodevelopmental outcomes at multiple ages beyond the first decade of life: the Faroese birth cohort (Debes et al., 2016; Grandjean et al., 2014; Julvez et al., 2010); the Canadian Nunavik study of school-age Inuit children (Boucher et al., 2010; Boucher et al., 2016; Despres et al., 2005; Saint-Amour et al., 2006); and the Spanish INMA (Environment and Childhood) birth cohort (Freire et al., 2010; Guxens et al., 2012; Llop et al., 2020; Lozano et al., 2021). Findings from these cohorts suggest that general cognition, fine motor speed, and behavior and attention may be adversely affected. These studies relied on more limited exposure data and a shorter follow-up period than is available in the Seychelles Child Development Study. Additionally, some of the cohorts studied consumed sea mammals which exposed them to multiple toxicants. Here we leverage the rich data from the Seychelles study to examine long-term mercury exposure in relation to outcomes measured over a 15 year period, in a cohort without co-exposure to other contaminants.

The Seychelles Child Development Study (SCDS) “Main cohort” has longitudinal characterization of postnatal MeHg exposure data measured from early childhood into young adulthood. The cohort was recruited in 1989–1990 and has been extensively evaluated for developmental outcomes on ten occasions between age 6 months and 24 years, with an ~80% retention rate of participants. At eight of the ten assessments, we determined recent postnatal exposure by measuring MeHg concentrations in the 1 cm of hair closest to the scalp (reflecting one month of exposure). Some adverse associations were seen between concurrent MeHg exposure in Seychelles and various neurodevelopmental outcomes at several time points, but the pattern of associations has not been consistent. Adverse associations were found with measures of fine motor speed and dexterity at 9 years (Grooved Pegboard) and 19 years (Finger Tapping), of general intelligence at 9 years (Wechsler Intelligence Scale for Children (WISC) III full scale IQ) and 19 years (Kaufman Brief Intelligence Test; (K-BIT) Matrices), attention at 9 years (Conners’ Teacher Rating Scale ADHD Index), of reading achievement at 17 years (Woodcock-Johnson (W-J) Test of Scholastic Achievement-II Passage Comprehension), and of various aspects of executive function at 22 years (including sustained attention and problem solving; Intra-Extra Dimensional Set Shift (IED) subtest of the Cambridge Neuropsychological Test Automated Battery (CANTAB)) (Davidson et al., 2011; Myers et al., 2003; van Wijngaarden et al., 2013; van Wijngaarden et al., 2017). These SCDS findings suggest that concurrent postnatal MeHg exposure may be associated with neurodevelopmental outcomes, and that these associations may be present into early adulthood. However, associations with concurrent exposure were not consistently observed with the same endpoints or domains over time. Furthermore, previous analyses did not take advantage of the longitudinal nature of our exposure data. This manuscript addresses both issues by focusing on neuropsychological domains that have been evaluated at several time points and are believed to mature later in childhood or adolescence, and by using novel longitudinal postnatal exposure metrics (see (Thurston et al., submitted)).

2. Methods:

2.1. Study population

The Seychelles Child Development Study (SCDS) recruited the Main Cohort of 779 mother-child pairs in 1989–1990 at 6 months (+/− 2 weeks) postpartum. All mothers were asked to provide a hair sample and Hg was measured in the 9 cm representing hair grown during pregnancy. Exclusion criteria included twins, insufficient hair to recapitulate prenatal Hg exposure, and illnesses or injuries known to affect neurodevelopment. The latter included severe perinatal illness, prematurity, encephalitis, meningitis, and closed head trauma with loss of consciousness. The Main Cohort children have completed ten test batteries at ages: 6, 19, and 29 months, and 5.5, 9, 10.5, 17, 19, 22, and 24 years. Cohort children who experienced exclusion criteria prior to testing age were excluded from analysis at that and later ages. Children’s hair was collected at all evaluations apart from those at 29-months and 10.5-years.

2.2. Neurological and behavioral outcomes

For this analysis, we included 85 neurodevelopmental outcomes measured at ages 9, 10.5, 17, 19, 22, and 24 years. The outcomes analyzed here represent five domains with the greatest development postnatally and, potentially, well into the third decade of life: executive function, attention, cognition, fine motor speed and dexterity, and language ability. We selected ages 9 and older because in our prior cross-sectional analyses, adverse associations with recent postnatal exposure were only observed at 9 years of age and older (Davidson et al., 2011; Myers et al., 2003; van Wijngaarden et al., 2013; van Wijngaarden et al., 2017). In addition, the literature on associations with postnatal exposure in later childhood and adolescence is limited. At older ages the tests are more sensitive to small neurodevelopmental differences. Furthermore, some domains can be assessed more accurately and thus subtle associations with environmental exposures are more likely to be detected. We examined the following neurodevelopmental endpoints:

Executive function measures included:

• Trail making A and B at age 9 years,

• Wisconsin Card Sorting Test (WCST) at age 17 years,

• Subscales from the Cambridge Neuropsychological Test Automated Battery (CANTAB) Intra-Extra Dimensional Shift (IED) and Spatial Working Memory (SWM) at age 17 and 22 years and Stockings of Cambridge (SOC) at age 22 years, and

• Stroop test at age 24 years.

Attention measures included:

• Conners Continuous Performance Test (CPT) and Conners’ Teachers Rating Scale (CTRS) hyperactivity index at 9 years of age,

• CANTAB Reaction Time (RT) and Rapid visual information processing (RVP) tests at ages 17 and 22 years,

• Test of Variables of Attention (TOVA) measuring visual and auditory outcomes at age 24 years, and

• Barkley adult ADHD rating scale (BAARS) at 24 years of age.

Cognitive measures examined were:

• Wechsler Intelligence Scale for Children version III (WISC) at age 9 years, and

• Kaufman Brief Intelligence Test (K-BIT) at age 19 years.

Fine motor speed and dexterity measures included:

• Finger tapping at ages 9, 19 and 24 years, and

• Grooved pegboard at age 9

Achievement, verbal learning and memory measures included:

• Woodcock Johnson (WJ) Tests of Achievement at age 9 and 17 years, and

• California Verbal Learning Test (CVLT) at age 9 and 17 years.

Additionally, we included a measure of expressive language ability, the Boston Naming Test (BNT) at age 9 and 22 years, and a measure of visual-motor coordination and perception, the Bender Visual Motor Gestalt test with Gottingen scoring at age 10.5 years.

2.3. Postnatal Mercury Exposure

Mercury was measured in the 1 cm of child’s hair closest to the scalp representing approximately 1 month of concurrent / recent exposure by cold vapor atomic absorption as described previously (van Wijngaarden et al. 2017). The accompanying manuscript (Thurston et al., submitted) describes in detail the postnatal mercury exposure data available and the creation of postnatal time-weighted exposure metrics. Briefly, our method for computing TW average exposure accommodates two challenges inherent in creating summary metrics based on repeated exposure measures: (1) how to handle missing values of individual exposures, and (2) how to combine multiple exposure measurements to create a person-specific summary measure. Missing values were imputed based on observed Hg values for the same participant at other ages, and also based on information about sex-specific Hg means among other participants at the age for which the Hg value was missing. At each age this imputation resulted in a final “observed-imputed” log(Hg) value that is the observed log(Hg) when Hg was actually measured, and is the imputed log(Hg) otherwise.

After imputing missing values, we developed two overall metrics of time-weighted (TW) postnatal average Hg exposure, one in childhood (TW-C) and another in early adulthood (TW-A). These metrics are weighted geometric means of exposure over the relevant age range (time window), using measured exposure when available and imputed exposure otherwise. Briefly a geometric mean, i.e. the mean of log(Hg), is calculated for each adjacent pair of exposure measures in the time window of interest. The weight for a particular geometric mean in the TW metric is the proportion of time spanned by the corresponding time window, relative to the total time spanned by the TW metric. TW-C was based on Hg measured at 6 months, 19 months, and 5.5 years. TW-A was based on Hg values from hair collected at 17, 19, 22, and 24 years of age. The early adult age range is further divided into three time periods (17–19 years, 19–22 years, and 22–24 years), depending on the age at which neurodevelopmental testing occurred. Note that the 17-year TW-A metric is similar to the concurrent postnatal Hg metric reported previously (Davidson et al., 2011) because it only considers the one exposure measurement at that age.

2.4. Covariates

Covariate adjustment was chosen to be as consistent as possible with our previous analyses of these endpoints. To make the results comparable across ages and domains, we adjusted all models for the same set of covariates known to be associated with the outcomes. For the child these included sex and age at testing. For the mother they included maternal IQ measured by the K-BIT, prenatal Hg exposure measured in maternal hair growing during gestation, and maternal socio-economic status (SES) using the Hollingshead Social Status Index modified for use in the Seychelles and measured at enrollment. Additionally, we included the Caldwell-Bradley preschool version of the Home Observation for Measurement of the Environment (HOME) administered when the children were 42–56 months of age (Myers et al., 2003). We did not impute missing covariate values, so models were based on participants with complete data on these covariates and a measure of the outcome.

2.5. Statistical analyses

Consistent with our approach for prior analyses of these endpoints, we used covariate-adjusted linear regression to examine the associations between the TW Hg measures and each outcome in separate models. Model assumptions were checked using standard methods (Weisberg 2005) and when needed outcomes were transformed to better meet the assumptions of linearity, constant variances, and approximate normality of the residuals. We evaluated model assumptions for each outcome by examining the Box-Cox plot to determine what outcome transformation most closely resulted in a normal distribution of the model residuals (Box and Cox, 1964). We also examined residual plots for the model using the untransformed outcome and, when indicated, for the model using the transformed outcome. We checked for unusual observations, including those that were identified in model fitting as extreme outliers or that had an unusually large Cook’s distance. In a small number of cases, invalid or extreme values were identified and set to missing.

In many cases, the outcome transformations we used were the same as those used in earlier papers that reported on outcomes at a single age. However, to facilitate comparison of model results across many outcomes, we report all model results using either the untransformed outcome, or the natural logarithm of the outcome. When a logarithmic transformation was warranted, if 0 was a possible outcome value, we added 1 to all values of that outcome prior to taking the logarithm so that original values of 0 could be included. We use the traditional two tailed p < 0.05 as a measure of statistical significance, recognizing that the large number of comparisons we make in this paper does not prevent a Type I error. Although we comment on statistical significance, our focus is on the overall pattern and consistency of associations across domains, time points, and exposure metrics.

3. Results

3.1. Descriptive analyses

Tables 1, 2, and 3 provide descriptive statistics for the outcomes, TW exposures and covariates.

Table 1:

Summary statistics for 9- and 10.5-year outcome and covariates among the 682 participants not excluded at 9 years who have at least one outcome at 9 or 10.5 years. Column labeled ‘n’ denotes the number of participants with values of the relevant variable. Arrows show direction of better performance on test.

	n	mean	SD	Median
Neurodevelopmental Outcomes at Age 9 Years
Cognition
Weschler Intelligence Scale for Children version III
Full IQ ↑	643	80.60	11.61	81.00
Verbal IQ ↑	643	79.09	10.13	79.00
Performance IQ ↑	643	85.71	14.20	84.00
Similarities (included in verbal IQ) ↑	643	5.21	2.66	6.00
Digit Span ↑	643	7.53	2.56	7.00
Block Design (included in performance IQ) ↑	643	7.78	3.58	8.00
Fine Motor Speed and Dexterity
Finger tapping (time in seconds)
Dominant hand ↑	642	34.01	5.73	33.60
Non-dominant hand ↑	642	30.04	4.57	29.80
Grooved Pegboard
Dominant hand ↓	643	91.81	20.48	88.00
Non-dominant hand ↓	641	104.21	25.33	100.00
Executive Function
Trailmaking Test (time in seconds)
A ↓	621	32.87	15.63	29.00
B ↓	610	80.14	47.99	64.00
Attention
Continuous Performance Task
Hit ↑	533	31.58	14.06	33.42
Attention d’ ↑	533	57.52	9.73	57.51
Risk-taking β ↓	532	74.10	20.20	72.22
Conners’ Teacher Rating Scale hyperactivity index ↓	584	55.43	12.78	51.00
Achievement, and Verbal Learning and Memory
Woodcock-Johnson
Applied Problems ↑	608	95.39	15.51	93.00
Letter Word Identification↑	603	131.73	40.31	142.00
California Verbal Learning Test
Short delay recall ↑	642	−0.01	1.04	0.00
Long delay recall ↑	603	−0.09	1.03	0.00
Other
Boston Naming Test
Total ↑	637	26.51	4.77	26.00
Cues ↑	637	4.52	2.15	5.00
No cues ↑	637	21.98	4.19	22.00
Neurodevelopmental Outcomes at Age 10.5 Years
Other
Bender Visual Motor Gestalt test with Gottingen scoring
Errors ↓	357	21.61	4.95	21.00
Recognition ↑	357	5.18	1.45	5.00
Postnatal MeHg Exposure
TW-C Hg	682	5.34	2.47	4.90
Covariates
Girl	682	0.51	0.50	1.00
Maternal Hg	682	6.78	4.49	5.87
HOME score	651	33.30	5.46	33.00
Maternal Enroll SES	681	22.14	12.66	14.00
Maternal KBIT	599	80.21	15.28	78.00
Age at 9Y test	682	8.99	0.34	8.99
Age at 10.5Y test	357	10.70	0.20	10.72

Table 2:

Summary statistics for 17- and 19-year outcomes and covariates among the 602 participants not excluded at 17 years who have at least one outcome at 17 or 19 years. Column labeled ‘n’ denotes the number of participants with values of the relevant variable. Arrows show direction of better performance on test.

	n	mean	SD	Median
Neurodevelopmental Outcomes at Age 17 Years
Executive Function
CANTAB Intra-Extra Dimensional Shift
Completed stages errors ↓	593	16.88	10.43	14.00
Pre-ED errors ↓	593	10.16	6.50	8.00
Total errors ↓	593	38.78	26.69	35.00
Total trials ↓	593	123.72	47.34	120.00
Wisconsin Card Sorting Test % error ↓	552	29.18	26.76	19.00
CANTAB Spatial Working Memory
Strategy ↓	593	33.96	3.88	34.00
Total errors ↓	593	28.46	17.77	27.00
Within errors ↓	593	2.14	3.79	1.00
Attention
CANTAB Reaction Time
Five choice ↓	592	339.51	55.23	333.46
Simple ↓	592	316.72	67.66	305.12
CANTAB Rapid Visual Processing
Mean latency ↓	587	454.31	117.71	427.29
False alarms ↓	593	2.96	5.34	2.00
Total misses ↓	593	14.11	5.20	14.00
Achievement, and Verbal Learning and Memory
Woodcock-Johnson
Applied Problems ↑	587	85.40	14.36	85.00
Letter Word Identification↑	543	99.67	26.85	104.00
Passage Comprehension ↑	523	75.68	20.62	78.00
Math Fluency ↑	594	73.36	12.04	73.00
Calculation ↑	592	84.20	16.99	86.00
California Verbal Learning Test
Short delay recall	592	−0.70	1.25	−0.50
Long delay recall	592	−0.79	1.27	−1.00
Neurodevelopmental Outcomes at Age 19 Years
Cognition
Kaufman Brief Intelligence Test
Matrices ↑	528	102.18	18.29	100.00
Verbal Knowledge ↑	530	12.80	2.48	13.00
Fine Motor Speed and Dexterity
Finger tapping
Dominant ↑	534	46.22	6.19	46.40
Non-dominant ↑	534	41.11	5.75	41.00
Postnatal MeHg Exposure
TW-C Hg	602	5.36	2.48	4.95
TW-A Hg (17Y)	598	8.03	4.63	7.39
TW-A Hg (17–19Y)	598	8.81	4.49	8.06
Covariates
Girl	602	0.53	0.50	1.00
Maternal Hg	602	6.81	4.48	5.86
HOME score	576	33.43	5.35	33.00
Maternal Enroll SES	601	22.22	12.53	17.00
Maternal KBIT	540	80.64	14.95	78.00
Age at 17Y test	594	17.22	0.41	17.10
Age at 19Y test	536	19.51	0.33	19.47

Table 3:

Summary statistics for 22- and 24-year outcomes and covariates among the 570 participants not excluded at 22 years who have at least one outcome at 22 or 24 years. Column labeled ‘n’ denotes the number of participants with values of the relevant variable. Arrows show direction of better performance on test.

	n	mean	SD	Median
Neurodevelopmental Outcomes at Age 22 Years
Executive Function
CANTAB Intra-Extra Dimensional Shift
Completed stages errors ↓	562	14.42	10.43	11.00
Pre-ED errors ↓	564	7.79	5.76	6.00
Total errors ↓	564	42.31	26.76	49.00
Total trials ↓	564	125.36	46.05	138.00
CANTAB Spatial Working Memory
Strategy ↓	564	35.33	3.92	35.00
Total errors ↓	564	32.86	21.29	28.00
Within errors ↓	564	1.67	3.18	1.00
CANTAB Stockings of Cambridge
Mean moves for 3 move problem ↓	564	3.39	0.58	3.00
Mean moves for 4 move problem ↓	564	5.62	1.04	5.50
Mean moves for 5 move problem ↓	563	7.16	1.47	7.00
Attention
CANTAB Reaction Time
Five choice ↓	564	370.77	70.24	360.33
Simple ↓	564	372.09	91.97	353.27
CANTAB Rapid Visual Processing
Mean latency ↓	562	462.83	149.39	423.04
False alarms ↓	564	8.10	20.62	2.00
Total misses ↓	564	12.37	5.28	13.00
Other
Boston Naming Test
Total ↑	567	40.66	7.20	41.00
Cues ↑	567	2.91	2.75	2.00
No cues ↑	567	37.75	7.01	38.00
Neurodevelopmental Outcomes at Age 24 Years
Executive Function
Stroop Interference ↑	534	−26.62	11.39	−26.00
Attention
Test of Variables of Attention auditory
D-prime ↑	548	5.10	1.53	4.71
Response Time mean ↓	549	431.21	97.76	420.70
Response Time variance ↓	549	141.05	54.71	131.49
Omission errors ↓	549	2.67	6.89	0.93
Commission errors ↓	548	1.78	2.35	1.23
Test of Variables of Attention visual
D-prime ↑	548	5.01	1.26	4.75
Response Time mean ↓	549	358.22	62.07	349.32
Response Time variance ↓	549	91.11	32.06	83.42
Omission errors ↓	549	1.79	5.77	0.31
Commission errors ↓	548	2.93	2.56	2.16
Barkley Adult ADHD Rating Scale
Hyperactivity ↓	535	7.28	2.20	7.00
Impulsivity ↓	535	5.38	1.67	5.00
Inattention ↓	535	12.61	3.10	12.00
Sluggish cognitive ↓	535	13.21	3.63	13.00
Total ↓	535	25.27	5.52	24.00
Fine Motor Speed and Dexterity
Finger tapping
Dominant hand ↑	546	52.76	7.18	53.00
Non-dominant hand ↑	545	47.20	6.57	47.20
Postnatal MeHg Exposure
TW-C Hg	570	5.38	2.53	4.89
TW-A Hg (17–22Y)	565	7.75	4.02	7.01
TW-A Hg (17–24Y)	565	7.13	3.77	6.45
Covariates
Girl	570	0.53	0.50	1.00
Maternal Hg	570	6.83	4.51	6.04
HOME score	545	33.12	5.20	33.00
Maternal Enroll SES	569	21.96	12.33	14.00
Maternal KBIT	509	79.96	15.15	77.00
Age at 22Y test	570	23.31	0.38	23.27
Age at 24Y test	549	24.96	0.30	24.97

Summary statistics for outcomes measured at age 9 and 10.5 years, TW-C Hg, and covariates are shown in Table 1. Among these 682 participants who had at least one outcome measurement at 9 or 10.5 years, all had a measure of TW-C Hg. Nine-year outcomes were missing for a varying number of participants with 39 missing WISC IQ outcomes, and 149–150 participants missing the CPT. Trailmaking A and B were set to missing for the 22 participants who could not read at 17 years of age or later (reading ability was first queried at age 17). The distribution of Trailmaking B, and to a lesser extent Trailmaking A, were positively skewed with the mean substantially larger than the median. A few participants had extremely large scores for Trailmaking A and/or B, indicating very poor performance (data not shown). The 10.5-year Bender outcomes were only measured for 357 participants (approximately half the cohort). The maternal K-BIT was missing for 83 participants.

Ability to see color is essential for all CANTAB and Wisconsin card sorting tests. Seven participants are colorblind, and all CANTAB and WCST outcomes for these seven people were set to missing. Summary statistics for the 17- and 19-year outcomes, TW-C, TW-A, and covariates are presented in Table 2. Of the 602 participants who had at least one outcome measure at 17 or 19 years, all had a TW-C Hg measure, while 598 had a TW-A Hg measure at 17 years (which is the same as the 17-year observed / imputed Hg) and a TW-A Hg at 19 years. The 17-year CANTAB outcomes (IED, SWM, RTI, and RVP) were measured on 587–593 participants. Somewhat fewer participants had measurements of WCST, WJ, and CVLT at 17 years. For all CANTAB outcomes measured at 17 years of age, larger scores represented decreased performance. For most 17-year CANTAB outcomes the largest value was much farther from the median than the smallest value (data not shown), indicating that one or more participants had extremely poor performance. At 19 years, KBIT matrices was measured on 528 participants, and KBIT verbal knowledge on 530 participants, with means of 102.2 and 12.8 respectively. FT was measured on 534 participants.

Summary statistics for 22- and 24-year outcomes are presented in Table 3. All 570 of these participants had values for TW-C Hg, and 565 had values for both TW-A Hg (17–22 years, and 17–24 years). Nearly all 570 participants had measures of 22-year outcomes, and all but 21–36 had measures of 24-year outcomes. As was the case at 17 years, the distribution of most 22-year CANTAB outcomes were skewed with more extreme large values than small values (data not shown). The distribution of other 22-year outcomes was approximately symmetric. At 24 years, the distribution of TOVA omission errors (both auditory and visual) were highly skewed with the mean much larger than the median. Several other outcomes were moderately skewed, including TOVA commission errors and response time variance (both auditory and visual). Consistent with previous analyses (van Wijngaarden et al., 2017), we considered BAARS outcomes to be valid only if at most one item was missing; non-valid BAARS outcomes were set to missing. One participant had an unusually large value for TOVA auditory commission errors, and another had an unusually large value for TOVA visual commission errors. Both extreme values were greater than 95, while the next-largest values were 33.4 and 16.0 respectively. These two values were considered extreme outliers and were set to missing. One participant had a negative value on the TOVA visual D-prime and another had a negative value on the TOVA auditory D-prime. These were considered invalid scores and were set to missing.

3.2. Summarized findings for TW-C and TW-A metric by domain

Presentation of the detailed regression results for the associations between the postnatal TW-C and TW-A metrics and neurodevelopmental outcomes are organized by domains. The associations between the two metrics and executive function outcomes are shown in Table 4, the associations with attention outcomes are shown in Table 5, and associations with cognition, fine motor speed and dexterity, as well as the BNT and Bender outcomes are shown in Table 6.

Table 4:

Executive function outcomes for time-weighted childhood (TW-C) and early adulthood (TW-A) Hg. Slopes, p-values and 95 percent confidence intervals adjusted for test age and five covariates. Entries in column labeled ‘log’ indicate whether the outcome was log-transformed (‘N’ indicates it was not, ‘Y’ indicates it was, and ‘Y+1’ indicates a 1 was added prior to the log-transformation). Arrows show direction of better performance on test.

		TW-C Hg				TW-A Hg
	log	n	slope	p	95% CI	n	slope	p	95% CI
Age 9 years
Trailmaking
A ↓	Y	532	0.007	0.276	(−0.006, 0.02)	-	-	-	-
B ↓	Y	525	−0.007	0.413	(−0.024, 0.01)	-	-	-	-
Age 17 years
CANTAB Intra-Extra Dimensional Shift
Completed stages errors ↓	Y+1	509	0.01	0.358	(−0.011, 0.03)	506	0.004	0.486	(−0.007, 0.015)
Pre-ED errors ↓	Y	509	0.007	0.463	(−0.011, 0.024)	506	0.004	0.474	(−0.006, 0.013)
Total errors ↓	Y	509	−0.015	0.231	(−0.039, 0.01)	506	−0.001	0.854	(−0.015, 0.012)
Total trials ↓	Y	509	−0.008	0.194	(−0.02, 0.004)	506	−0.001	0.837	(−0.007, 0.006)
Wisconsin Card Sorting Test % error ↓	N	472	0.153	0.753	(−0.805, 1.111)	469	0.533	0.048	(0.005, 1.061)
CANTAB Spatial Working Memory
Strategy ↓	N	509	−0.001	0.991	(−0.136, 0.135)	506	−0.038	0.318	(−0.112, 0.037)
Total errors ↓	N	509	−0.048	0.879	(−0.662, 0.566)	506	−0.044	0.798	(−0.383, 0.294)
Within errors ↓	Y+1	509	0.015	0.316	(−0.014, 0.044)	506	0.002	0.802	(−0.014, 0.018)
Age 22 years
CANTAB Intra-Extra Dimensional Shift
Completed stages errors ↓	Y	484	−0.024	0.068	(−0.05, 0.002)	481	−0.014	0.128	(−0.031, 0.004)
Pre-ED errors ↓	Y	485	−0.004	0.649	(−0.022, 0.014)	482	0.009	0.157	(−0.003, 0.021)
Total errors ↓	Y	485	0.02	0.107	(−0.004, 0.045)	482	0.022	0.01	(0.005, 0.039)
Total trials ↓	Y	485	0.011	0.072	(−0.001, 0.023)	482	0.011	0.005	(0.003, 0.019)
CANTAB Spatial Working Memory
Strategy ↓	N	485	0.014	0.851	(−0.128, 0.155)	482	0.009	0.859	(−0.087, 0.105)
Total errors ↓	N	485	0.086	0.821	(−0.665, 0.837)	482	−0.083	0.748	(−0.593, 0.426)
Within errors ↓	Y+1	485	−0.019	0.157	(−0.046, 0.007)	482	−0.009	0.337	(−0.027, 0.009)
CANTAB Stockings of Cambridge
Mean moves for 3 move problem ↓	N	485	0.003	0.753	(−0.018, 0.025)	482	0.005	0.537	(−0.01, 0.019)
Mean moves for 4 move problem ↓	N	485	−0.03	0.126	(−0.068, 0.008)	482	−0.006	0.66	(−0.032, 0.02)
Mean moves for 5 move problem ↓	N	484	−0.02	0.467	(−0.073, 0.033)	481	−0.012	0.507	(−0.048, 0.024)
Age 24 years
Stroop Interference ↑	N	473	−0.006	0.979	(−0.438, 0.427)	471	0.164	0.311	(−0.154, 0.481)

Table 5:

Attention outcomes for time-weighted childhood (TW-C) and early adulthood (TW-A) Hg. Slopes, p-values and 95 percent confidence intervals adjusted for test age and five covariates. Entries in column labeled ‘log’ indicate whether the outcome was log-transformed (‘N’ indicates it was not, ‘Y’ indicates it was, and ‘Y+1’ indicates a 1 was added prior to the log-transformation). Arrows show direction of better performance on test.

		TW-C Hg				TW-A Hg
	log	N	slope	p	95% CI	n	slope	p	95% CI
Age 9 years
Continuous Performance Task
Hit ↑	N	454	−0.272	0.305	(−0.793, 0.248)	-	-	-	-
Attention d’ ↑	N	454	0.161	0.376	(−0.196, 0.517)	-	-	-	-
Risk-taking β ↓	N	453	0.753	0.041	(0.032, 1.474)	-	-	-	-
Conners’ Teacher Rating Scale hyperactivity index ↓	Y	482	0.003	0.441	(−0.005, 0.011)	-	-	-	-
Age 17 years
CANTAB Reaction Time
Five choice ↓	Y	508	0.001	0.697	(−0.004, 0.007)	505	0	0.882	(−0.003, 0.003)
Simple ↓	Y	508	0	0.975	(−0.007, 0.007)	505	0.001	0.76	(−0.003, 0.004)
CANTAB Rapid Visual Processing
Mean latency ↓	Y	504	−0.002	0.68	(−0.01, 0.006)	501	0	0.984	(−0.005, 0.004)
False alarms ↓	Y+1	509	0.003	0.835	(−0.024, 0.03)	506	0.017	0.026	(0.002, 0.031)
Total misses ↓	N	509	0.031	0.74	(−0.152, 0.214)	506	−0.014	0.786	(−0.115, 0.087)
Age 22 years
CANTAB Reaction Time
Five choice ↓	Y	485	0.003	0.382	(−0.004, 0.009)	482	0.002	0.471	(−0.003, 0.006)
Simple ↓	Y	485	0.004	0.302	(−0.004, 0.012)	482	0	0.861	(−0.005, 0.006)
CANTAB Rapid Visual Processing
Mean latency ↓	Y	483	0.001	0.902	(−0.009, 0.01)	480	−0.004	0.282	(−0.01, 0.003)
False alarms ↓	Y+1	485	0.008	0.672	(−0.031, 0.048)	482	0.001	0.914	(−0.025, 0.028)
Total misses ↓	N	485	0.078	0.421	(−0.113, 0.269)	482	−0.002	0.974	(−0.132, 0.127)
Age 24 years
Test of variables of Attention auditory
D-prime ↑	Y+1	485	−0.004	0.391	(−0.013, 0.005)	482	0.004	0.253	(−0.003, 0.01)
Response Time mean ↓	Y	486	0.007	0.088	(−0.001, 0.015)	483	0.005	0.121	(−0.001, 0.011)
Response Time variance ↓	Y	486	0.014	0.049	(0, 0.029)	483	0.004	0.41	(−0.006, 0.015)
Omission errors ↓	Y+1	486	0.018	0.235	(−0.012, 0.047)	483	−0.004	0.694	(−0.025, 0.017)
Commission errors ↓	Y+1	485	0	0.978	(−0.021, 0.021)	482	−0.011	0.141	(−0.026, 0.004)
Test of variables of Attention visual
D-prime ↑	Y	485	−0.007	0.131	(−0.017, 0.002)	482	0	0.967	(−0.007, 0.007)
Response Time mean ↓	Y	486	0.006	0.049	(0, 0.012)	483	−0.001	0.803	(−0.005, 0.004)
Response Time variance ↓	Y	486	0.015	0.011	(0.003, 0.026)	483	0.002	0.595	(−0.006, 0.011)
Omission errors ↓	Y+1	486	0.02	0.147	(−0.007, 0.046)	483	0.005	0.578	(−0.014, 0.025)
Commission errors ↓	Y+1	485	−0.009	0.414	(−0.031, 0.013)	482	0.002	0.836	(−0.014, 0.017)
Barkley Adult ADHD Rating Scale
Hyperactivity ↓	Y	474	0	0.955	(−0.01, 0.01)	472	0.003	0.422	(−0.004, 0.01)
Impulsivity ↓	Y	474	−0.009	0.078	(−0.019, 0.001)	472	0.002	0.656	(−0.006, 0.009)
Inattention ↓	Y	474	−0.001	0.759	(−0.01, 0.007)	472	−0.001	0.847	(−0.007, 0.005)
Sluggish cognitive ↓	Y	474	−0.004	0.353	(−0.014, 0.005)	472	0.002	0.571	(−0.005, 0.009)
Total ↓	Y	474	−0.002	0.523	(−0.01, 0.005)	472	0.001	0.759	(−0.005, 0.006)

Table 6:

Other outcome associations with TW-C and TW-A Hg. Slopes, p-values and 95 percent confidence intervals are adjusted for test age and five covariates. Entries in column labeled ‘log’ indicate whether the outcome was log-transformed (‘N’ indicates it was not, ‘Y’ indicates it was, and ‘Y+1’ indicates a 1 was added prior to the log-transformation). Arrows show direction of better performance on test.

		TW-C Hg				TW-A Hg
	log	N	slope	p	95% CI	n	slope	p	95% CI
Cognition
Age 9 years
Weschler intelligence scale for children version III
Full IQ ↑	N	550	−0.11	0.54	(−0.461, 0.242)	-	-	-	-
Verbal IQ ↑	N	550	−0.088	0.58	(−0.401, 0.225)	-	-	-	-
Performance IQ ↑	N	550	−0.11	0.629	(−0.555, 0.336)	-	-	-	-
Similar ↑	N	550	−0.017	0.7	(−0.104, 0.07)	-	-	-	-
Digit ↑	N	550	0.042	0.333	(−0.043, 0.128)	-	-	-	-
Block ↑	N	550	−0.043	0.488	(−0.164, 0.078)	-	-	-	-
Age 19 years
Kaufman Brief Intelligence Test
Matrices ↑	N	450	0.011	0.976	(−0.67, 0.691)	450	−0.159	0.444	(−0.567, 0.249)
Verbal Knowledge ↑	N	451	−0.015	0.751	(−0.107, 0.077)	450	−0.004	0.881	(−0.06, 0.052)
Fine Motor Speed and Dexterity
Age 9 years
Finger tapping
Dominant hand ↑	N	549	−0.054	0.568	(−0.238, 0.13)	-	-	-	-
Non-dominant hand ↑	N	549	−0.107	0.148	(−0.253, 0.038)	-	-	-	-
Grooved Pegboard
Dominant hand ↓	Y	550	−0.003	0.362	(−0.01, 0.004)	-	-	-	-
Non-dominant hand ↓	Y	549	0.003	0.321	(−0.003, 0.01)	-	-	-	-
Age 19 Years
Finger tapping
Dominant ↑	N	455	0.148	0.201	(−0.079, 0.374)	454	−0.042	0.541	(−0.179, 0.094)
Non-dominant ↑	N	455	−0.024	0.81	(−0.223, 0.174)	454	−0.039	0.517	(−0.159, 0.08)
Age 24 years
Finger tapping
Dominant ↑	N	484	0.087	0.485	(−0.158, 0.333)	481	0.045	0.623	(−0.134, 0.223)
Non-dominant ↑	N	484	0.058	0.622	(−0.173, 0.289)	481	0.037	0.662	(−0.129, 0.203)
Achievement
Age 9 years
Woodcock-Johnson
Applied Problems ↑	N	520	−0.053	0.84	(−0.574, 0.467)	-	-	-	-
Letter Word ↑	N	520	0.18	0.791	(−1.158, 1.519)	-	-	-	-
California Verbal Learning Test
Short delay recall ↑	N	549	0.013	0.482	(−0.023, 0.048)	-	-	-	-
Long delay recall ↑	N	513	0.013	0.497	(−0.024, 0.049)	-	-	-	-
Age 17 years
Woodcock-Johnson
Applied Problems ↑	N	502	0.119	0.626	(−0.361, 0.6)	499	−0.151	0.245	(−0.407, 0.104)
Letter Word ↑	N	465	−0.444	0.359	(−1.394, 0.506)	462	−0.405	0.104	(−0.894, 0.083)
Passage comp ↑	N	447	−0.51	0.158	(−1.218, 0.199)	444	−0.419	0.032	(−0.801, −0.037)
Math fluency ↑	N	509	0.037	0.85	(−0.352, 0.426)	506	−0.016	0.885	(−0.228, 0.197)
Calculation ↑	N	506	0.282	0.299	(−0.25, 0.814)	503	−0.116	0.434	(−0.407, 0.175)
California Verbal Learning Test
Short delay recall ↑	N	507	−0.014	0.511	(−0.056, 0.028)	504	−0.01	0.373	(−0.033, 0.013)
Long delay recall ↑	N	507	−0.012	0.574	(−0.054, 0.03)	504	−0.009	0.467	(−0.032, 0.015)
Other
Age 9 years
Boston Naming Test
Total ↑	N	545	−0.051	0.526	(−0.207, 0.106)	-	-	-	-
Cues ↑	N	545	0.002	0.949	(−0.073, 0.077)	-	-	-	-
No cues ↑	N	545	−0.053	0.442	(−0.189, 0.082)	-	-	-	-
Age 10.5 years
Bender Visual Motor Gestalt test with Gottingen scoring
Errors ↓	N	313	−0.235	0.029	(−0.446, −0.025)	-	-	-	-
Recognition ↑	N	313	0.001	0.966	(−0.067, 0.07)	-	-	-	-
Age 22 years
Boston Naming Test
Total ↑	N	486	−0.253	0.048	(−0.504, −0.002)	483	−0.222	0.01	(−0.39, −0.054)
Cues ↑	N	486	0.03	0.568	(−0.072, 0.131)	483	−0.001	0.988	(−0.069, 0.068)
No cues ↑	N	486	−0.283	0.021	(−0.523, −0.043)	483	−0.222	0.007	(−0.383, −0.061)

Executive domain:

The logarithmic transformation was used for most executive domain outcomes to better satisfy regression assumptions. TW-C was not significantly associated with any outcome in the executive domain (Table 4). There were adverse associations between TW-A and three executive domain outcomes: the WCST % total errors at 17 years (slope = 0.53; 95% confidence interval (CI) = 0.005, 1.06; p = 0.048) and two components of the CANTAB IED at 22 years on the logarithmic scale: total errors (slope = 0.022; CI = 0.005, 0.04; p = 0.01) and total trials (slope = 0.011; CI = 0.003, 0.02; p = 0.005).

Attention domain:

With the exception of the 9-year CPT outcomes and RVP total misses at 17 and 22 years, the remaining outcomes were log-transformed for model fitting. Three outcomes in the attention domain had adverse associations with TW-C (Table 5): the CPT Risk-taking β at age 9 years (slope = 0.753; CI = 0.032, 1.474, p=0.04), the logarithm of the TOVA auditory response time variance at 24 years (slope = 0.014; CI = 0, 0.029; p = 0.049) and the 24-year TOVA visual response time mean (slope = 0.006, CI = 0, 0.012, p=0.049) and response time variance (slope = 0.015; CI = 0.003, 0.026; p = 0.01), both on the logarithmic scale. Except for the 24-year BAARS, the sign of the association between TW-C and nearly all remaining attention outcomes was in the adverse direction. TW-A was significantly associated with one attention outcome: the logarithm of the CANTAB RVP false alarms at 17 years (slope = 0.017; CI = 0.002, 0.031; p = 0.026), also an adverse association.

Cognition, fine motor, and achievement domains:

All cognitive outcomes, all fine motor outcomes except the 9-year grooved pegboard, and all achievement outcomes were included in models without any transformation. There were no significant associations between either TW-C or TW-A and any outcomes in the cognition or fine motor speed and dexterity domain (Table 6). In the achievement domain, TW-A was significantly associated with the WJ passage comprehension (slope = −0.42; CI = −0.80, −0.037; p = 0.032), an adverse association.

Other domains:

Neither the Boston naming test nor the Bender tests required log transformations for model fitting. The Bender test errors outcome was significantly associated with TW-C (slope = −0.235, CI = −0.446, −0.025, p = 0.03), a beneficial association. The 22-year BNT was adversely associated with both TW-C and TW-A metrics. The slope relating TW-C and BNT total was −0.253 (CI = −0.504, −0.002; p = 0.048) and the slope for TW-C and BNT no cues was −0.283 (CI = −0.523, −0.043, p=0.02). For TW-A the associations with BNT total and with BNT no cues were slope=−0.222 (CI = −0.39, −0.054; p=0.01) and slope = −0.222 (CI = −0.383, −0.061; p=0.007) respectively.

Figure 1 shows the relationships between three outcomes and both TW-C and TW-A. The three outcomes shown are the 22-year Boston naming test no cues, and the logarithm of both the 24-year TOVA visual response time mean and TOVA visual response time variance. Superimposed on each plot is the covariate-adjusted regression line with 95% confidence interval.

Figure 1.

Relationship between 22-year Boston Naming Test and the logarithm of the 24-year TOVA visual Response Time mean and visual Response Time variance versus the TW-C and TW-A mercury metrics among the SCDS Main Cohort participants. Superimposed are the slopes (solid lines) and 95% confidence intervals for the mean response (dashed lines) from the covariate-adjusted regression models.

4. Discussion

We examined the association of our long-term time weighted mercury metrics (TW-C & TW-A) with neurodevelopmental outcomes at multiple ages and observed 14 statistically significant associations (Tables 4, 5, and 6). The association of TW-C with the 10.5-year Bender visual motor Gestalt test with Gottingen scoring was in the beneficial direction, while the remaining 13 associations were adverse. Adverse associations with TW Hg were seen at ages 9, 17, 22, and 24 years, across executive, attention, and achievement domains, and for the Boston Naming Test.

At age 9 years, the only adverse association was between TW-C and CPT risk-taking beta. At 17 years of age, adverse associations were seen between TW-A and three outcomes: WCST % error, RVP false alarms, and WJ passage comprehension. At 19 years of age only four outcomes were measured, and none were associated with either TW metric. At age 22 years, the TW-A metric was adversely associated with four outcomes: IED total errors and total trials, and the BNT no cues and total score. The BNT associations were also present with TW-C, but the 22-year IED outcomes were not significantly associated with the TW-C metric. At age 24 years, adverse associations were present with TW-C Hg and three TOVA outcomes: the auditory response time mean, the visual response time mean and the visual response time variance.

The outcomes that had the most significant adverse associations with a TW metric each measure different aspects of neurodevelopment. The BNT test is considered a measure of language ability, while the IED outcomes and WCST are measures of executive function, and CTRS hyperactivity and TOVA outcomes are measures of activity and attention respectively. The number of statistically significant adverse associations was small relative to the number of exposure-outcomes assessed. Had the Bonferroni correction been applied to adjust for the large number of outcomes measured at each age, none of the associations would have been considered significant. It is interesting to note that attention outcomes appear to be associated with the TW-C metric whereas some executive function outcomes were associated with the TW-A metric. These findings raise the possibility that consistent prolonged postnatal exposure to Hg from fish consumption may have adverse consequences on some aspects of neurodevelopment, but that these associations may vary depending on the time window of exposure. These findings differ from those of our previous analyses with concurrent exposure where we did not observe clear patterns over time or within neuropsychological domains. This may be due to several considerations, including use of the new TW metric, differences in the set of covariates used here and in previous analyses, and inclusion of a larger sample size in the current analysis due to the imputation approach used to create the new TW metrics.

Associations between postnatal MeHg and neurodevelopmental outcomes at multiple ages in childhood were examined in the Faroese birth cohort, a study from Nunavik, Canada, and the Spanish INMA (Environment and Childhood) birth cohorts. The exposure data from these studies were not as comprehensive as that from the Seychelles study, and concurrent Hg was used as the postnatal metric.

In the Faroese study at 7 years of age (n=694–923), the authors concluded that concurrent postnatal MeHg exposure in cord blood was adversely associated with visuospatial processing and memory (Grandjean et al., 1997). In a later study the researchers state “The results for the Bender reproduction score suggest that postnatal methylmercury exposure may affect visuospatial memory, and that the effect in regard to this function may be stronger for exposures at late preschool age than for prenatal exposures.” (Grandjean et al., 2014). However statistically significant beneficial associations were seen between concurrent Hg exposure and both finger tapping in the non-preferred hand, and CVLT learning (Grandjean et al., 2014). At 14 years of age in this cohort, concurrent Hg exposure was significantly adversely associated with the finger tapping test (Debes et al., 2006) and demonstrated an association on peak V of the evoked potentials (Murata et al., 2004), suggesting that recent postnatal exposure could result in a decrease in motor speed even at exposure levels similar to EPA’s reference dose (Debes et al., 2006).

In the Nunavik study (n=265), prenatal Hg exposure was not significantly associated with any of the outcomes studied. However, concurrent blood Hg concentration at the time of testing was associated with higher action tremor amplitude during pointing movements at age 4–6 years (Despres et al., 2005). Concurrent blood Hg concentrations were also associated with shorter latencies on the N75 and P100 components of visual evoked potentials measured at this age (Saint-Amour et al., 2006). In a follow-up study at 11 years of age, event-related potential measures of information processing were not associated with child blood Hg measured at the time of examination (Boucher et al., 2010) but concurrent Hg was associated with slower fine motor speed (Boucher et al., 2016). The Faroese study reported similar findings at 14 years (Debes et al., 2006), but opposite findings at 7 years (Grandjean et al., 2014).

The Spanish INMA project consists of a network of birth cohorts recruited in different locations at different times, including three original cohorts (Ribera d’Ebre; Granada; Menorca) and four additional cohorts (Valencia, Sabadell, Asturias, and Gipuzkoa) with a shared protocol (Guxens et al., 2012). Freire and colleagues in the Granada cohort reported on the association between postnatal MeHg exposure and performance on the McCarthy Scales of Children’s Abilities (MSCA) at 4 years of age which assesses quantitative, verbal, memory, perceptual-performance, and motor domains (N=72 with complete data) (Freire et al., 2010). Total hair Hg levels greater than 1 μg/g were associated with lower scores in the general cognitive, memory and verbal domains (Freire et al., 2010). Among 1,252 children 4–5 years of age with complete data in three other INMA cohorts (Gipuzkoa, Sabadell, and Valencia), postnatal Hg exposure was associated with better MSCA scores, but these associations were attenuated after accounting for fish intake (Llop et al., 2020). Lozano et al. in the Valencia cohort reported that increased hair Hg concentrations at 9 years were associated with worse scores on the Child Behavior Checklist (CBCL) internalizing and total problems scales and the ADHD index of the Conners’ Parent Rating Scales-Revised: Short Form (CPRS), especially among boys and in subgroups of polymorphisms in the GSTP1, BDNF and APOE genes (Lozano et al., 2021). An adverse association was also found with the Attention Network Test (ATN) but only after adjustment for prenatal Hg exposure.

Findings from these three cohorts suggest that general cognition, fine motor speed, and behavior and attention may be adversely associated with postnatal Hg exposure, although some beneficial associations with fine motor speed were also noted. We did not find associations between postnatal Hg exposure and any measures of general cognition or fine motor speed in outcomes extending into early adulthood. Varying results across studies could be due to a number of factors including differences in the child’s ages at which samples for mercury were collected, the ages at which outcomes were measured, and control for confounders. Use of concurrent rather than average or cumulative exposure and follow-up only into early adolescence in earlier studies made it difficult to assess the potential long-term effects of postnatal exposure.

This study has a number of strengths. We were able to use all available Hg measures over multiple ages to obtain TW Hg metrics that represent long-term postnatal exposure. The cohort is relatively large and follow up has been ~80% over 24 years. The participants and evaluators have been blinded to exposure throughout the study. We have examined a large number of outcomes measured in multiple domains over multiple years.

The study also has limitations. Although hair was collected at multiple ages, these evaluations and samples ages were not evenly spaced over time. No Hg measures were available between ages 9 and 17 years. Consequently, our two measures of TW Hg do not capture an approximate measure of TW Hg over the lifespan of the participants. Further, our analyses included a large number of comparisons, and the number of statistically significant associations was relatively small, the associations were limited to a few developmental domains, and with the exception of the Boston Naming Test, the associations were not consistent between the two metrics. Therefore, the role of chance in explaining these associations cannot be excluded.

In conclusion, our findings suggest that long-term postnatal MeHg exposure may be adversely associated with some neurodevelopmental outcomes in early adulthood, in particular attention and language ability. However, the associations are of unknown clinical significance and need confirmation in other cohorts.