Davidson PW, Myers GJ, Cox C, Axtell C, Shamlaye C, Sloane-Reeves J, Cernichiari E, Needham L, Choi A, Wang Y, Berlin M, Clarkson TW. Effects of Prenatal and Postnatal Methylmercury Exposure From Fish Consumption on NeurodevelopmentOutcomes at 66 Months of Age in the Seychelles Child Development Study. JAMA. 1998;280(8):701-707. doi:10.1001/jama.280.8.701
From the University of Rochester School of Medicine and Dentistry, Rochester, NY (Drs Davidson, Myers, Cox, Axtell, Wang, and Clarkson, and Mss Sloane-Reeves, Cernichiari and Choi); the Republic of Seychelles Ministry of Health, Victoria, Mahé, Seychelles (Dr Shamlaye); the Centers for Disease Control and Prevention, Atlanta, Ga (Dr Needham); and The University of Lund, Lund, Sweden (Dr Berlin).
Context.— Human neurodevelopmental consequences of exposure to methylmercury (MeHg)
from eating fish remain a question of public health concern.
Objective.— To study the association between MeHg exposure and the developmental
outcomes of children in the Republic of Seychelles at 66 months of age.
Design.— A prospective longitudinal cohort study.
Participants.— A total of 711 of 779 cohort mother-child pairs initially enrolled in
the Seychelles Child Development Study in 1989.
Setting.— The Republic of Seychelles, an archipelago in the Indian Ocean where
85% of the population consumes ocean fish daily.
Main Outcome Measures.— Prenatal and postnatal MeHg exposure and 6 age-appropriate neurodevelopmental
tests: the McCarthy Scales of Children's Abilities, the Preschool Language
Scale, the Woodcock-Johnson Applied Problems and Letter and Word Recognition
Tests of Achievement, the Bender Gestalt test, and the Child Behavior Checklist.
Results.— The mean maternal hair total mercury level was 6.8 ppm and the mean
child hair total mercury level at age 66 months was 6.5 ppm. No adverse outcomes
at 66 months were associated with either prenatal or postnatal MeHg exposure.
Conclusion.— In the population studied, consumption of a diet high in ocean fish
appears to pose no threat to developmental outcomes through 66 months of age.
INORGANIC MERCURY (Hg) discharged into lakes, rivers, and oceans is
converted to methylmercury (MeHg) by microorganisms and bioaccumulated up
the aquatic food chain.1 Concern about the
potential public health threat from MeHg arose in the United States in the
early 1970s when elevated concentrations were found in fish in the Great Lakes.
Today, recreational fishing is restricted in many states and Food and Drug
Administration guidelines regulate interstate commerce of fish because of
their MeHg content.
Mass health disasters in Minamata and Niigata, Japan, caused by consumption
of fish highly contaminated with MeHg from an industrial source,2,3
and in Iraq following consumption of bread containing MeHg fungicide,4- 6 confirmed that MeHg
was neurotoxic and that the prenatal period was the most sensitive stage of
the life cycle. For example, severe exposures in Iraq (up to 674 ppm of mercury
in hair) were associated with microcephaly, seizures, mental retardation,
and cerebral palsy. The Iraq outbreak also resulted in less severe outcomes
typified by developmental delays and abnormal results of neurological examinations.
A dose-response analysis suggested that effects may occur at maternal hair
concentrations of mercury as low as 10 ppm,5,6
although there was considerable uncertainty in this estimate. This compares
with an average in the US population of 1 ppm or less.7
All fish contain MeHg. Frequent consumption of ocean fish can lead to
MeHg levels in excess of 10 ppm and as high as 50 ppm in hair.1
Epidemiological studies8- 10
on populations consuming fish where Hg was biologically methylated failed
to find clinical cases of MeHg poisoning. The possibility that prenatal MeHg
exposure from maternal consumption of a fish diet may be associated with subtle
changes in children's cognitive and neurological development has been examined
in these studies with inconclusive results.1,11
We have followed longitudinally12,13
a large inception cohort of mother-child pairs in the Republic of Seychelles,
a westernized archipelago in the middle of the Indian Ocean where 85% of the
population consumes marine fish daily.14 This
article presents the results of the neurodevelopmental examination of the
Seychelles Child Development Study (SCDS) cohort at 66 months, an age at which
neuropsychological tests may be given that are sensitive enough to assess
potential associations between developmental outcomes and MeHg dietary exposure.
Our results also include examination of the role of postnatal exposure from
The cohort consisted of 711 mother-child pairs living in the Republic
of Seychelles, representing 91% of the 779 pairs originally enrolled in the
SCDS main study.11 Informed consent was obtained
from the caregiver of every participating child. The protocol was approved
by human subjects review boards at the University of Rochester, Rochester,
NY, and the Ministry of Health, Victoria, Mahé, Republic of Seychelles,
before enrollment began. The sample size was sufficient to detect a 5.7-point
difference on any test with a mean (SD) of 100 (16) between low (0-3 ppm)
and high (>12 ppm) MeHg groups for a 2-sided test (α=.05 at 80% power).
Twenty-eight mother-child pairs were excluded because of medical problems
that might seriously affect development.13
An additional 16 pairs had insufficient maternal hair available to accurately
recapitulate prenatal exposure, and 24 did not return for testing at 66 months.
Demographic characteristics of the Seychelles and the cohort were reported
earlier.15 At enrollment, the mothers reported
eating an average of approximately 12 marine fish meals per week. Sea mammals
are not consumed in the Seychelles. We have previously documented that lead
levels in whole blood are less than 0.48 µmol/L (10 µg/dL) in
a representative group of Seychellois children and mothers.16
Each child was evaluated at 66 months (±6 months) at a child
development center staffed by a team of specially trained Seychellois nurses
blinded to MeHg levels and the results of testing during previous visits.
Five children were tested between 72 months and 79 months of age. All evaluations
were conducted between July 1994 and October 1995. The test battery assessed
multiple developmental domains17 and was similar
to those used to demonstrate adverse developmental effects of exposure to
lead18 and polychlorinated biphenyls (PCBs)19 and to those used in earlier studies to measure MeHg
exposure effects.9,20 The tests
are sufficiently sensitive and accurate to detect neurotoxicity in the presence
of a number of confounding factors.21
The test battery included the following 6 primary measures: (1) the
General Cognitive Index (GCI) of the McCarthy Scales of Children's Abilities22 to estimate cognitive ability; (2) the Preschool
Language Scale23 (PLS) total score to measure
both expressive and receptive language ability; (3) the Letter and Word Recognition
and (4) the Applied Problems subtests of the Woodcock-Johnson (W-J) Tests
of Achievement24 to measure reading and arithmetic
achievement; (5) the Bender Gestalt test25
to measure visual-spatial ability; and (6) the total T score from the Child
Behavior Checklist (CBCL)26 to measure the
child's social and adaptive behavior. The CBCL questionnaire was completed
by each child's primary caregiver. All tests were given in Creole, the language
spoken by 98% of Seychellois at home.
Pure tone hearing thresholds were tested using a portable audiometer.
Caregiver IQ was determined using the Raven Standard Progressive Matrices,
a nonverbal test designed to minimize the effects of culture on measurement
of IQ.27 When the children were between 42
and 56 months of age, the Home Observation for Measurement of the Environment
(HOME) Inventory for Families of Preschool Age Children28
was administered during home visits. Following procedures described earlier,17 on-site test administration reliability was assessed
by an independent scorer; percentage of disagreement ranged from 0% to 8%.
Mean intraclass correlations for interscorer reliability were 0.96 to 0.97.
Reliability of final scoring in Rochester was conducted by rescoring a sample
of tests; the mean intraclass correlation coefficient was 0.96.
Prenatal exposure was assessed by measuring the concentration of total
mercury (THg) in a segment of maternal hair representing growth during pregnancy.
Total Hg in maternal hair during pregnancy correlates well with blood levels
of MeHg1 and with THg levels in fetal brain.29 Methylmercury accounts for over 80% of the THg in
hair samples collected from fish-eating populations.16,29
Maternal hair levels of THg have been the biological indicator of choice in
nearly all previous epidemiological studies of fetal exposure to MeHg. There
is considerable variation in the relationship between hair and blood THg in
different individuals. However, the key relationship, that of hair levels
and brain levels, may not show the same variability.29
Postnatal exposure was determined by measuring THg at 66 months of age from
a 1-cm segment of the child's hair nearest the scalp. This age was chosen
because it was coincident with the age of testing and all children were postweaning
and eating a fish diet. In 25 children, hair samples taken during the HOME
administration (at 48 months of age) were used to determine postnatal MeHg
exposure because their hair sample at 66 months of age was insufficient for
Total Hg was analyzed in 5 or more samples of species of fish caught
and consumed in the Seychelles, including yellow fin tuna (Thunnus albacares), Indian mackerel (Rastrelleger
kanagurta), brown spotted grouper (Epinephelus chlorustigma), green jobfish (Aprion virescens), bonito
(Euthynnus affinis), bludger (Carangoides gymnostethus), and spangled emperor (Lethrinus nebulosus).
The analysis of hair samples and fish homogenates for THg and inorganic
Hg was done by cold vapor atomic absorption spectrometry. The analysis technique
and quality control procedure are given elsewhere.16
Polychlorinated biphenyls are known to be present in some ocean fish
and may be associated with developmental delays in children. Levels of PCBs
in serum of 49 of the children at 66 months of age were analyzed at the laboratories
of the US Centers for Disease Control and Prevention, Atlanta, Ga. The analytic
method for measuring the PCBs involved deproteinization of the serum with
formic acid, elution through a column containing octadodecyl (C18) packing
material, elution through a column containing Florisil packing material, concentration
of the organic eluents, and analysis by dual capillary column–gas chromatography
with electron capture detection.30
The effect of prenatal and postnatal MeHg on each outcome variable was
adjusted for covariates, specified as part of the study design, and selected
because of their potential to bias the assessment of the association between
Hg and outcome.17 Covariates associated with
the child included birth weight, birth order, sex, history of breast-feeding,
hearing status, and the child's medical history. Covariates associated with
the mother and family included maternal age, maternal smoking during pregnancy,
maternal alcohol consumption during pregnancy, maternal medical history, caregiver
intelligence, language spoken in the home, Hollingshead socioeconomic status
(SES), and HOME score. Two multiple linear regression analyses (with 2-tailed
significance tests using a significance level of P≤.05)
including both prenatal and postnatal THg were performed for each of the 6
primary measures. The first involved all covariates (full model) and the second
included only covariates believed to most likely influence child development
in the Seychelles (reduced model), including sex, birth weight, child's medical
history, maternal age, HOME score, caregiver IQ, SES, and hearing status.
Each full and reduced model was run both with and without THg by sex interaction
terms to test the hypothesis that males and females have different THg slopes.
All models were examined for statistical outliers and influential points.31 Each model was run first with outliers, then without
outliers, and the results were compared. All of the results were essentially
the same with or without outliers. The regression analyses for all 6 primary
measures were also repeated without influential points to determine whether
the original results were dependent on such points.31
The final analysis included influential points that were not also outliers.
The results without influential points were consistent with the original analysis.
Secondary analyses tested the hypothesis that associations between developmental
outcomes and THg exposure might be nonlinear. All regression analyses were
repeated, first using the log of the prenatal and postnatal THg values, then
classifying THg variables into 5 groups each for prenatal and postnatal exposure.
A total of 350 samples of fish were analyzed. The median THg for each
of the 25 species ranged from 0.004 ppm to 0.75 ppm, with most medians in
the range of 0.05 to 0.25 ppm. These levels are comparable with fish in the
US market. The lowest levels occurred in reef fish. Methylmercury accounted
for over 90% of the THg in 34 fish homogenates analyzed by gas chromatography–atomic
fluorescent detection. This finding confirms many previous observations.32
Twenty-eight PCB congeners ranging from congener 28 to 206 were measured
in each serum sample. All samples had no detectable levels of any PCB congeners.
The detection limit for the PCB analysis was 0.2 ng/mL. These results are
typical for persons with no known exposure to PCBs.33
The mean (SD) maternal hair level of THg during pregnancy was 6.8 (4.5)
ppm (n=711), and the mean child hair level at 66 months was 6.5 (3.3) ppm
(n=708). The ranges (maternal hair, 0.5-26.7 ppm; child hair, 0.9-25.8 ppm)
were sufficient to test for exposure effects using regression analysis. Maternal
and child THg concentrations were not highly associated Pearson r=0.15, n=708, P<.001) as observed by others.20 The exposure levels found in the Seychelles are typical
of populations that depend on fish as a major dietary source of protein and
Table 1 and Table 2 show test score means and SDs for prenatal and postnatal
exposure levels. These results are similar to what would be expected from
a healthy, well-developing US population. No test indicated a deleterious
effect of MeHg exposure. Four of the 6 measures showed better scores in the
highest MeHg groups compared with lower groups for both prenatal and postnatal
Primary Analyses. The models with all covariates (full models) and limited covariates
(reduced models) were both significant (ie, each model was able to describe
the data) and yielded similar results for every measure. The THg by sex interaction
test for prenatal exposure was not significant in any regression model. The
THg by sex interaction was significant for postnatal exposure for the Bender
Gestalt test; hence, we report the results of the reduced model with both
interaction terms included. For all other analyses, we report the results
for reduced models without THg by sex interactions.
The regression coefficients for all variables in the 6 sets of analyses
are shown in Table 3. Figure 1 shows partial residual plots (end
points adjusted for covariates) for prenatal and postnatal exposure for the
McCarthy GCI, the PLS total score, and the W-J Applied Problems test score.
For the McCarthy GCI analysis, the model (F [15, 628]=4.41, P<.001, R2 = 0.10) indicated
that slopes for both THg exposures did not differ from 0 (P=.59 and .06 for prenatal and postnatal exposure, respectively). For
the PLS analysis, the model (F [15, 590]=6.25, P<.001, R2 = 0.14) showed effects of both prenatal and
postnatal THg exposure (P=.02 for both), but the
effects were very small and in a direction of enhanced performance. The total
increase in scores across the entire range of THg exposures was less than
4.5 points. The W-J Applied Problems model (F [15, 625]=5.38, P<.001, R2 = 0.11) indicated
that the slope for prenatal exposure did not differ significantly from 0 (P=.41). There was a significant beneficial postnatal exposure
effect (P=.05), but no evidence for an adverse effect.
The postnatal exposure slope for the W-J Applied Problems test was 0.36 ppm,
representing a 9.7-point increase over the full exposure range, or a 10% improvement
Figure 2 shows partial residual
plots for the Bender Gestalt test, separating male from female subjects. The
model (F [17, 613]=4.53, P<.001, R2 = 0.11] showed no significant association with prenatal
exposure. The interaction of postnatal THg with sex was significant (P=.004). The slightly positive slope for female errors
was not significant (P=.14). The regression coefficient
of −0.16 ppm for male subjects was significant (P=.009), resulting in a reduction of 4.3 errors over the entire exposure
range (a 40% performance improvement given that the average score for the
lowest-exposure group was about 10 errors).
Although the models for the W-J Letter and Word Recognition achievement
score and the CBCL yielded significant overall F statistics, none of these
was significantly associated with either prenatal or postnatal exposure.
The data for covariate effects shown in Table 3 indicate that test scores were frequently influenced by
sex (female subjects scored higher than male subjects), and were directly
related to SES, quality of home stimulation, and caregiver IQ, as would be
expected in westernized cultures. These data also indicate that performance
by Seychellois children on these tests was similar to what would be expected
of US children.
Table 4 gives the partial R2 values for the effects of prenatal and postnatal
THg exposure on each developmental measure and the 95% confidence intervals
for the effect of a 10-ppm increase in hair THg concentration. The small partial R2 values shown in Table 4 indicate that THg exposure accounted for little of the variance
associated with each test. For the McCarthy GCI and the CBCL, the magnitude
of the negative lower confidence limit might be of clinical significance,
at least over a greater range of hair THg levels. However, the inclusion of
0 in the confidence intervals shows that the effects are not statistically
Secondary Analysis. Log transformations of the 2 THg variables did not alter the direction
of any effects. In the categorical analyses, test scores for children with
an MeHg exposure level greater than 12 ppm were not significantly different
than for children with exposure levels of 3 ppm or less.
Results from this study are relevant for the United States and other
countries with similar dietary intake of fish. The major source of MeHg in
the Seychelles is ocean fish and the average fish levels are similar to those
on the US market. Seychellois MeHg levels are 10 to 20 times higher than in
the United States because the Seychellois consume more fish, not because they
eat a few fish with abnormally high MeHg levels. Thus, any potential adverse
effects of MeHg in fish should be detected in the Seychelles long before such
effects would be seen in the United States. Our findings confirm our earlier
report in which no adverse developmental effects were found in toddlers following
prenatal MeHg exposure.12 Our data extend preliminary
results from the SCDS Pilot Study,35 involving
a less statistically powerful, less well-controlled developmental evaluation
of 217 Seychellois children at 66 months using most of the same measures but
lacking many of the covariates used here.
We applied multiple regression analysis to our data as has been done
in other studies on the effects of mercury,9- 13
lead,18,36 or PCBs19
on child development. Although our models showed no negative associations
between MeHg and outcome scores, our test procedures did detect other factors
known to be associated with child development. The HOME test indicated that
the quality of the home environment had a substantial impact on child performance,
significantly affecting the results of all tests. The SES of the family, the
caregiver's IQ, and the child's sex were all found to have an influence on
performance scores, in keeping with the literature on child development.37,38 These results increase confidence
in the sensitivity of our tests to child development functions. Postnatal
exposure to MeHg at 66 months of age was associated with a small but statistically
significant increase on several developmental outcomes. Our hypothesis did
not predict positive effects, since there are no reasons to suppose that such
effects are associated with exposure to MeHg. However, MeHg levels in hair
are known to correlate closely with fish intake, and other factors or agents
associated with fish, such as omega-3 fatty acids, may have beneficial effects.
A large cohort study under way in the Faroe Islands found enhancement of developmental
milestones in suckling infants exposed to MeHg in breast milk.39
They suggested that MeHg levels in the infant were a surrogate for the length
of breast-feeding, which is reported to have a positive association with developmental
In contrast with the conclusions from this and our earlier studies of
the main Seychellois cohort,12,13
the Faroe Islands study found evidence of cognitive deficits associated with
prenatal exposure to THg when children were tested at 7 years of age.20 Important differences between the 2 populations may
explain the divergent outcomes (eg, nutritional practices, housing, and lifestyle).
However, the major difference between our study and the Faroe Islands study
is the source of exposure. Ocean fish are the source of MeHg in the Seychelles,
whereas pilot whales are the predominant source in the Faroe Islands.20 The average MeHg level in the meat of pilot whales
sampled in the Faroe Islands was 1.6 ppm,41
approximately 10 times higher than the average level in fish consumed in the
Seychelles. Approximately the same level of inorganic Hg is also present in
whale meat.41 In addition, whale blubber is
also consumed by the Faroese41 and is heavily
contaminated with fat-soluble pollutants.42
The average PCB concentration in pilot whale blubber from Faroese waters is
elevated (about 30 ppm).43 In general, fatty
tissues of marine mammals in the North Atlantic also contain elevated levels
of persistent organochlorine compounds including dibenzofurans and dioxins,
DDT and its metabolites, and other pesticides. It is difficult to determine
the relative toxicological impact of individual compounds. Some of these contaminants
are believed to affect child development.44
The Faroese study may be relevant to populations consuming large, perhaps
episodic, amounts of marine mammals, but its relevance to people consuming
ocean fish remains to be established.
A Swedish expert group conducted the first extensive evaluation of human
health risks from MeHg in fish in 1971.45 They
concluded the lowest toxic level in hair was 50 ppm in adults. The World Health
Organization (WHO) expert group46 subsequently
reaffirmed the Swedish conclusion and applied a safety factor of 10 to cover
risks to the most sensitive subgroup of the population, assumed to be those
who are prenatally exposed. Thus, 5 ppm in hair was adopted as the international
standard for the upper tolerable level of Hg in hair. Subsequent epidemiological
studies of human populations prenatally exposed to MeHg from fish have given
strong support to the WHO guideline.8,10,12,13
Our results add further support to the validity of this long-standing guideline.
In summary, the results of extensive performance tests conducted with
cohort children at 66 months of age strongly support our findings reported
at younger ages. The development of these children is proceeding well without
any detectable adverse influence of MeHg. Our results support Egeland and
Middaugh's observation47 that it would be inadvisable
to forgo the health benefits of fish consumption to protect against a small
risk of adverse effect at the levels of MeHg found in ocean fish on the US