WORKSHOP Organized by Committee on Environmental and Natural Resources (CENR) Office of Science and Technology Policy (OSTP) The White House Scientific Issues Relevant to Assessment of Health Effects from Exposure to Methylmercury November 18-20, 1998 Table of Contents Report   Individual Panel Reports Appendix I -Exposure Panel Appendix I-A -Neurobehavioral Endpoints Panel Appendix I-B -Confounders & Variables Panel Appendix I-C -Design/Statistics Panel Appendix I-D -Experimental Panel Appendix I-E Reports from Study Teams Appendix II -Iraq & Seychelles Studies Appendix II-A -Faroe Islands Studies Appendix II-B -Amazon Studies Appendix II-C Comments from Study Groups on Workshop Report and Panel Reports Appendix III -Iraq & Seychelles Appendix III-A -Faroe Islands Appendix III-B References Appendix IV Workshop Program Appendix V Attendees List Appendix VI

The Workshop on the Scientific Issues Relevant to Assessment of Health Effects from Exposure to Methylmercury was held in Raleigh, North Carolina, November 18–20, 1998. The workshop was organized by an interagency committee at the request of the Office of Science and Technology Policy (OSTP). The organizing committee was chaired by the National Institute of Environmental Health Sciences (NIEHS), and it included representatives from:

Department of Health and Human Services (DHHS)

- Office of the Assistant Secretary for Planning and Evaluation

- Centers for Disease Control and Prevention (CDC)

- Agency for Toxic Substances Disease Registry (ATSDR)

- Food and Drug Administration (FDA)

Environmental Protection Agency (EPA)

National Oceanic and Atmospheric Administration (NOAA)

Office of Science and Technology Policy (OSTP)

Office of Management and Budget (OMB)

The purpose of the workshop was to discuss and evaluate the major epidemiologic studies associating methylmercury exposure with an array of developmental measures in children.

Although the workshop did not attempt to derive a risk assessment, the product of the workshop should facilitate agreement on risk assessment issues. The major studies considered were those that have examined populations in Iraq, the Seychelles, the Faroe Islands, and the Amazon, along with the most relevant animal studies. The workshop was the result of discussions of an interagency workgroup considering approaches for utilizing the emerging data from the Faroe Islands study and the Seychellois cohort. The Methylmercury Workshop was a response to the suggestion that the emerging Seychellois and Faroese data undergo a level of scrutiny beyond journal peer review if they are to be used in policy setting.

A series of questions were developed to help focus the workshop. These questions, which are listed below, were directed to each of the studies and were meant to help structure the workshop but not unduly constrain scientific deliberation.

1. For each study, what are the relative exposures to organic or inorganic mercury?
2. What are the sources of exposure? Is the consumption of fish, shellfish, and marine mammals the dominant source? Are dental amalgams, occupational exposures, or other sources significant confounders?
3. What is the specificity and sensitivity of health (e.g., behavioral, neurological, developmental) endpoints? What are the most specific and sensitive tests of these endpoints? How are these tests impacted by cultural or behavioral practices?
4. Are the developmental tests used comparable across different studies? Across species?
5. What are the confounders that affect health endpoints positively (e.g., [selenium], omega-3 fatty acids) and negatively (e.g., PCBs; alcohol use; health conditions with neurological effects, such as diabetes)? Do they influence the interpretation of the study?
6. What is the variability within and across populations studied in mercury exposure and host factors (e.g., age, gender, nutritional status and practices, genetic predisposition)?
7. For each epidemiologic study of interest, which features of its statistical design, research protocol, or execution in the field are particularly strong or weak? To what extent do these strengths and weaknesses affect the validity and reliability of scientific inferences based on these studies?
8. What are the contributions of animal or experimental studies toward interpretation of the human studies?

Detailed presentations were made by each study team, that focused on their responses to the eight questions. Each panel addressed their issues in plenary sessions as well as in separate breakout groups in which invited observers and other observers were allowed to participate. The observers included broad representation from State and Federal agencies, industry, and public interest groups. In addition, there were 11 public comments to the panels on the second day. The workshop program and the list of attendees are found in appendices V and VI respectively.

On the final day of the workshop, each panel presented summaries of their discussions and recommendations in plenary session. Their most significant findings and recommendations are as follows.

PANEL FINDINGS AND RECOMMENDATIONS

1. Methylmercury is a developmental neurotoxin, but effects at low doses encountered by eating fish are difficult to evaluate.
2. All the studies reviewed were considered of high scientific quality and the panel recognized that each of the investigations had overcome significant obstacles to produce important scientific information. The panel also stated that continued funding of the studies in the Seychelles, Faroes, and Amazon is necessary for the full potential of those studies to be realized. This is particularly the case for the Faroes and Seychelles studies, which have assessed and are currently assessing the potential developmental neurotoxic effects of methylmercury in fish-eating populations. The developmental studies would benefit by evaluation of common endpoints using similar analytical methods. It is important to note that the Amazon study did not assess developmental endpoints but assessed effects in adults.
3. Results from the Faroes and Seychelles studies are credible and provide valuable insights into the potential health effects of methylmercury
4. Some differences are clearly present in results from the Faroes, Seychelles, and Amazon, but the panel was not able to clearly identify the sources of these differences. Among possible sources are the different effects of episodic versus continuous exposure, ethnic differences in methylmercury responses, lack of common endpoints in the Faroes and Seychelles studies, and several other confounders or modifying factors such as those found in diet and lifestyle, as well as in chemicals present in seafood, which is the source of methlymercury to these populations. The other chemical constituents of seafood that may be explanatory include those that may be beneficial to fetal neurodevelopment (i.e., omega-3 fatty acids) and those that may be harmful to fetal neurodevelopment (e.g., PCBs).
5. These studies have provided valuable new information on the potential health effects of methylmercury but significant uncertainties remain because of issues related to exposure, neurobehavioral endpoints, confounders and statistics, and design.

This report summarizes the evaluations and recommendations made by the workshop panels regarding the effects of low-level exposures to methylmercury. Drafts of the report were reviewed by workshop panels, the interagency organizing committee, and the Seychelles and Faroes study teams. Comments were incorporated to the extent possible without altering the conclusions of the panels. The Faroes and Seychelles study teams prepared responses to the workshop report and individual panel reports (Appendix III).

Human exposures to high levels of methylmercury result in neurotoxic effects that are well documented in a number of incidents globally, including the incident in Iraq. The effects at low levels of exposure are difficult to evaluate. Methylmercury is ubiquitous and nearly everyone has some low level of exposure. Previous efforts to establish levels considered protective of public health acknowledged the need for further studies on populations with low levels of exposure. Current studies in the Seychelles and Faroe Islands are efforts to determine the minimum effects level from exposures in utero and during childhood. The Amazon studies generally concern effects in adults although studies among children are underway. Prior to the publication of the results of the 19- and 29-month data from the Seychelles study in 1995, the earlier Iraqi study provided the largest data-base for evaluating the developmental neurotoxic effects of methylmercury.

The studies presented and discussed at the workshop represent exceptional examples of observational epidemiology to test the hypothesis that relatively low level pre-natal exposures to methylmercury via maternal seafood consumption, can adversely affect neuropsychological development in young children. In each study, the investigators carefully studied the sites and communities and developed comprehensive designs for their studies that reflected deep understanding of the existing literature on methylmercury intoxication in human populations.

1. Descriptions of Major Studies

In the Iraqi incident in 1971–72, it has been well documented that the poisonings resulted primarily from consumption in rural areas of homemade bread prepared from wheat (seed wheat) treated with methylmercury fungicides. In wheat and flour samples analyzed during and after the outbreak, methylmercury was the predominant form of mercury. These samples also contained small amounts of ethylmercury, but this was never more than 8% of the methylmercury level in analyzed samples, and no ethylmercury was detected in hair samples of patients. Barley treated with methylmercury and phenylmercury fungicides and fed to farm animals was also examined as a potential source, but levels of mercury in meat and dairy products were considered too low to be significant contributors.

In the Seychelles, where deep-sea and reef fish are staples in the population diet, mercury exposure is primarily in the form of methylmercury from fish consumption. Mothers in the Seychelles Child Development Study reported consuming an average of 12 fish meals a week during pregnancy having an average methylmercury concentration <0.3ppm. Methylmercury comprised more than 90% of the total mercury content in 25 species of fish eaten in the Seychelles. Although not systematically studied, dental amalgams and occupational exposures do not appear to be significant sources, and exposure to inorganic mercury appears to be low.

In the Faroe Islands, two main dietary sources of exposure to methylmercury have been identified: pilot whale and fish. Pilot whale is a traditional food in the Faroe Islands, but availability of fresh pilot whale meat varies substantially with location and over time. Pilot whale meat, on average, contains significantly higher concentrations of methylmercury than the local fish (mainly cod), although detailed data on levels in the several species of fish commonly consumed in the Faroe Islands were not available. Dental amalgams and occupational exposures are not significant sources of mercury exposure in the Faroe Islands.

In the Amazon, the use of mercury to amalgamate gold in mining operations has led to the release of large quantities of elemental and inorganic mercury into the air and waterways. The slash and burn agricultural practices and subsequent deforestation also release mercury from the soil into the waterways, where it is converted to methylmercury. Studies reported at the workshop of several villages along the Tapajos River (at least 250 km downstream from the mining operations) indicate that the principal exposure to methylmercury is from fish consumption. However, it is difficult to exclude the possibility of some exposures to elemental and inorganic mercury. Dental amalgams do not appear to be a significant source of exposure for this population.

2. Outcomes of Seychelles, Faroes and Amazon Studies

The Seychelles studies used standardized measures of global neurobehavioral function. The test battery used in the Seychelles was predicated on the occurrence of severe clinical neurological findings based on the historical clinical evaluations to assess effects from high exposure to methylmercury in Iraq. Affected individuals in Iraq consumed 50–400 mg mercury as methylmercury in contaminated grain over a period of as long as six months. Motor retardation was seen in infants born of mothers with hair levels in the 10–20 ppm range. In the Seychelles, in contrast to the high-level, short-term exposures in Iraq, exposure to methylmercury was from regular consumption of fish. Infants were born of mothers with (arithmatic) mean hair levels of 6.8 ± 4.5 ppm hair (range of 0.5 to 27 ppm hair). The highest recorded concentration was 35 ppm hair in one child at 6 months of age. The results of the study to date indicate that none of the measures of methylmercury exposure were significantly associated with adverse developmental effects in infants or the IQ measures used in school-age children up to 66 months of age.

In the Faroes study, the test battery was based on the expectation of multifocal, domain- related effects. Maternal hair levels ranged from 0.2–39.1 ppm (geometric mean, 4.27 ppm). Mercury concentration of cord blood showed a geometric mean of 22.9 ppb (range 0.9–351 ppb). Tests of memory, attention, and language including motor delay and executive function were negatively associated with methylmercury exposure in children up to 7 years of age. Tests of motor function and visual spatial ability were less clearly associated with methylmercury exposure.

The Amazon population (adults) experienced much more pronounced neurological effects than the Seychelles and Faroes studies, even though documented methylmercury levels were only slightly higher than levels found in the Seychelles and the Faroe Islands, (hair methylmercury between 5.6 and 38.4 ppm). Different domains were assessed using different methods.

The Seychelles assessed methylmercury effects with measures of global IQ, whereas the Faroe study employed domain-specific testing. It is plausible that prenatal exposure to toxic substances might result in no effect on overall IQ, but might cause an effect in domain-specific findings such as memory deficits, motor delay, or effects on the complex domain involved in formulating behavior called executive function. Indeed, the Faroes study observed such domain-specific deficits, including losses in memory, motor delay, and executive function. Thus, it might be that the effects of methylmercury at lower doses are domain-specific and only detectable by domain-specific tests used in the Faroe Study, but not with the more general tests used in the Seychelles Study. However, the Seychelles investigators had also determined the domain-related McCarthy subscales (subtests) and made these results available to the panel. From this group of subtests there was no evidence of mercury-related effects in the Seychelles.

The adverse effects shown by tests used on the adult population in the Amazon study indicated that those tests were apparently responsive to mercury toxicity. However, it is not possible at this time to compare the results of this study to either the Seychelles or Faroes studies because the Amazon studies tested for different functions in adults.Studies on children are in progress.

There are many possible reasons for the apparent differences in outcomes between the Faroe Islands and Seychelles studies. Factors that may have influenced the outcomes include differences in exposures or exposure measures, differences in the neurobehavioral tests used or their administration or interpretation, influences of confounders and covariates, and biostatistical issues involved in the analysis of the data. It is also possible that the different populations simply produced different results.

3. Exposure Issues

The objective of both the Faroe Island and Seychelles studies was to evaluate the offspring of mothers exposed to methylmercury during pregnancy The epidemiological studies presented at this workshop are unusually rich in exposure data. Even in occupational epidemiology, only rarely does the epidemiologist have comprehensive biological monitoring data on the study subjects themselves.

In the Faroe Islands study, umbilical cord blood was the preferred biomarker of exposure, although maternal hair was also collected and analyzed. In the Seychelles study, maternal hair was used as the measure of fetal exposure. Each of these surrogate measures has its advantages and disadvantages.

Cord blood is as near to a direct, noninvasive measure of fetal methylmercury blood levels as can be obtained, but it is only available at a single point in time (parturition), andis a composite value for a range of exposures over mainly the third trimester. In the Faroes study, cord blood and hair levels of mercury were qualitatively similar predictors of performance in neurobehavioral tests, but regression coefficients for cord blood were generally larger. For both exposure measures, changes were observed in most of the same tests although there was some indication that hair measurements could provide a better correlation with tests of motor function. Blood is a convenient matrix for measurement of other potentially relevant exposures (e.g., selenium, lead, some chlorinated organics, markers of nutritional status). Cord tissue may be a better matrix for measurement of lipophilic substances such as polychlorinated biphenyls (PCBs).

Hair is a well-established matrix for measuring exposures of an individual to methylmercury, facilitating intercomparison of studies. Analysis of small segments of hair sequentially allows one to get information on the pattern of maternal (and therefore fetal) exposure to methylmercury over the course of the pregnancy. The correlation of levels of methylmercury in maternal hair and newborn infant brain has been reported. There are a number of variables (e.g., hair growth rate, density, and color; external sources of contamination) that must be taken into account or controlled when using hair measurements of mercury to predict exposure to methylmercury. From the analytical point of view, one must take into account that the concentration of methylmercury in hair is at least two orders of magnitude greater than the corresponding levels in blood.

Differences in timing of exposure among the exposed population could result in different patterns of neurobehavioral deficits. We know that the fetus develops functionality at specific times over the course of pregnancy. Exposure assessments related to behavior should optimally be evaluated within these specific time periods. Reliance on a single measure or an average measure of exposure necessarily precludes temporal interpretations of observed patterns in behavior attributable to in-utero exposure. If possible, analyses of trimester-specific exposure would be a valuable addition to what has already been completed. There may be differences in patterns of exposure between the Faroese and Seychellois cohorts that contribute to the apparent disparity in reported effects. Neither study attempted to gather detailed information on diet. However, the simple observations that are available are consistent with a greater degree of episodic exposure in the Faroe Islands:

Episodic exposure could possibly reduce or enhance the likelihood of detecting a consistent "neurobehavioral signature injury" specific to methylmercury, and may account for different observations in children with the same average exposure.

Temporal exposure patterns may be important for several reasons. There may be windows (periods) of enhanced sensitivity of the developing fetus to the neurotoxic effects of methylmercury, so that the time course of maternal/fetal exposures may be a critical factor. Similarly, neurotoxic effects could be more a function of episodic higher-level (peak) exposures than of average (continual) exposures over the course of a pregnancy. The pattern of consumption of pilot whale (in the Faroe Islands) and fish may affect not only methylmercury intake but also other potentially relevant exposures.

The Amazon studies described at the workshop provided an interesting and credible rationale for the seasonal variation in methylmercury exposures (based on segmental analysis of hair samples) that parallels the seasonal variation in consumption of fish with high levels (piscivorous or omnivorous) or low levels (herbivorous) of methylmercury.

Pattern of exposure, such as peaks versus more continual exposure, is an important parameter that has not been fully examined as yet in any of the studies. On the other hand, it is by no means certain that the differences in outcome of the studies will be solely, or even partially, attributable to differences in this parameter. Thus, an estimated biological half-life of about 45 days for methylmercury in the human body would tend to even out short-term differences in exposure. Nevertheless, it seems prudent to recognize the possible significance of this parameter in comparisons of results across other studies.

Exposure in these studies was measured in terms of markers of internal dose, i.e., hair or blood mercury levels. It is assumed in all the studies that the vehicle of exposure to methylmercury was through fish and/or whale consumption, but actual dietary intake was not assessed. While biomarkers of internal dose are often preferred for exposure assessment, it is not known which biomarker, if any, offers significant advantage, i.e., which biomarker relates most closely to the toxic dose or concentration at the site(s) of toxic action, presumably within the fetal brain. Few studies exist that allow for correlations among biomarkers. It should be noted that, although both the Seychelles and Faroes studies collected maternal hair samples, these samples did not cover the exact same portion of gestation. Both measures (hair and blood) may be important for the full evaluation of neurobehavioral endpoints.

4. Study Design Issues for Assessment of Neurobehavioral Endpoints.

There were no fatal flaws in either the Faroes or Seychelles studies. Overall, both studies are commendable in terms of design, analytic strategy, and consideration of a wide range of confounders. There are, however, issues that may have influenced outcomes as discussed below.

The confounders and variables panel raised the potential issue of selection bias, particularly in the Seychelles study, where subject recruitment was limited to individuals for whom severe debilitating conditions were not present. However, exclusion criteria for the Seychelles study have been published (Shamlaye et al., Neuro Toxicology 16:597–611, 1995). Of the 779 mother–infant pairs, 39 met exclusion criteria, leaving a cohort of 740 for analysis. There were only 18 subjects excluded because they met a priority medical exclusion criteria. While this type of restriction can lead to an underestimation of the impact of the exposure, especially when the true nature of the dose–effect curve is unknown, it is unlikely that this small number had a significant outcome on the cohort analysis.

The domains most clearly affected by methylmercury in the whole data set of subjects in the Faroes study at approximately 7 years of age are memory, attention, and language. However, there were differences in different subsets of the cohort. Decrease in psychomotor speed was observed in association with hair–mercury concentration and in a nested case-control design based on high (10–20 ppm) versus low (<3 ppm) maternal hair concentrations. Visuospatial performance seemed to be affected by postnatal mercury exposures, while no association was apparent for any of the other domains. Therefore, no specific "neurobehavioral signature of injury" from methylmercury could be found in the data presented in these studies. This is not surprising because, to date, there is not a specific pattern of outcomes for other neurotoxins such as lead and PCBs.

The tests used in the Seychelles were translated, from English to Creole or French. Likewise, the tests used in the Faroes were adapted from English or Scandinavian tests. Nonetheless, the tests appear to be culturally appropriate, and, in general, within the limits of the expertise of the panel. Cultural sensitivity appears to be a hallmark of both studies.

The assessment of neurodevelopmental outcomes depends greatly upon the appropriate uses of the test for the domain being assessed and the age of the subjects being assessed. Despite similar average exposures, statistically significant associations between methymercury exposure and tests of memory, attention, reaction time (psychomotor speed), and language were seen in the Faroes among 7-year-olds but not in the Seychelles among 66-month-olds.

One concern with the test battery administered in the Seychelles study is the selection of ages for assessing the children. Generally speaking, developmental assessments are likely to be less sensitive in detecting subtle neurotoxic effects when they are administered during a period of rapid developmental change because of the high variability in the control population. The period covering ages 60–72 months is one such period, during which marked individual differences in the rate of cognitive maturation are likely to eclipse subtle differences in function attributable to a teratogenic exposure. This may, in part, be responsible for the absence of observed differences reported for the Seychelles cohort at ages 29 months and 66 months.

With regard to the Bayley Scales, studies at multiple ages of early exposure to alcohol, PCBs, and other substances have repeatedly failed to detect effects at 18 months, probably because it, too, is a period of rapid cognitive maturation involving the development of language abilities. Also, 29 months is likely to be an insensitive testing point for the Bayley Scales because it comes at the end of the age range for which the Bayley was standardized, and there is a substantial risk of "ceiling effects," i.e., too many children receiving the highest possible scores on several items. The failure of the Seychelles study to detect methylmercury effects to date could, therefore, be due in part to the ages of the children when the assessments were made. It is fortunate that the next round of testing will be administered at age 8 years, a point in development that should be optimal for detecting neurodevelopmental effects, as has been shown in studies of lead.

The Seychelles study controlled for age by testing all children at the same age and by converting the raw test scores to age-corrected standard scores using conversion tables based on U.S. norms, whereas the Faroes study analyzed the raw scores, adjusting statistically for the childs age (in days). The latter may be the preferred approach. The U.S. norms are known very precisely but there is wide variance in the U.S. norms, consistently leading to the overestimate or underestimate performance of normal children. Results based on fewer numbers will be less precise than the U.S. norms but age correction of the data as used in the Seychelles should not lead to bias.

The order in which the tests are administered may affect the outcome. In the Seychelles and Faroes studies, different approaches were used. In the Seychelles study, the children were administered all the assessments in the same order. In the Faroes, each child was tested in one of four predetermined orders. When different orders are used, the investigators need to check for order effects routinely by running regressions with mercury by order interaction terms. Although mercury exposures were reported not to differ,the median scores on the tests by time of day (morning versus afternoon) should be compared. Also, the scores of the children who had to travel some distance should be compared with those who did not. These types of influences may not be significantandthey can be controlled statistically in the regression analyses. Reanalysis might be a satisfactory solution to this issue.

A question is whether the high IQ scores and low rate of referral for mental retardation of the Seychelles children account for the differences in the studies. One major difference between the two cohorts is in the early developmental trajectories of study subjects. In the Seychelles, the very low number of abnormal Denver Developmental Screening Test designations, the mean Psychomotor Development Index of 126, and the low rate of referral for mental retardation raise the suspicion of selection bias or some characteristic of data gathering or handling that might exclude affected children. However, for cognitive measures (e.g., MDI, GCI), means and standard deviations were similar to U.S. norms. Thus, it is unlikely that differences between the two prospective studies are due to the developmental robustness of the Seychelles children. Furthermore, the Denver Developmental Screening Test is specific but insensitive, especially when administered to 6-month-olds, as in the Seychelles study.

5. Confounders

In current epidemiologic use, a confounder is a factor that produces an apparent but not real biological association between some outcomes and some exposure (i.e., methylmercury). It does so by being associated with both exposure and outcome, and by being incompletely or not controlled for in data analyses. A factor related only to outcome or exposure but not to both cannot produce such an artifactual association. By this definition, coexposures to PCBs (and chlorinated pesticides), selenium, and omega-3 fatty acids might influence outcomes of these studies.

PCBs and other organochlorine compounds are contaminants of (inter alia) pilot whale blubber in the Faroe Islands studies. The average PCB concentration in pilot whale blubber, which was episodically consumed, is about 30 ppm. PCBs may act as modifiers of the neurological effects of methylmercury. The appropriate measure of PCB exposure should capture prenatal exposure, such as maternal blood, breast milk, etc. The concentration in the child after birth is more closely related to transfer through early breast milk, but so far the major associations between PCB exposure and development effects have been with some measure of maternal body burden at term. PCB levels in cord blood, maternal blood, early breast milk, and adipose tissue strongly correlate with each other if adjusted for lipid content. PCB levels in umbilical cord blood are expected to correlate with these other measures but this has not actually been shown.

In the Faroe Islands study, PCBs were analyzed in samples of cord tissue, rather than the more traditional matrices of serum or breast milk. In this study, the most common of the 208 congeners were measured. The total PCB concentration was estimated from the three congeners that contribute approximately 50% of the total amount present in human milk in the Faroe Islands. The relative potential of different congeners for causing neurological effects is not known. It was reported that a follow-up study is underway to determine the partitioning of PCBs between mother, cord tissue, and fetus. However, one panel stated that it did not believe that the effects seen in the Faroes were due to uncontrolled confounding by PCBs, whereas another panel suggested that the potential effects of PCBs might modify neurobehavioral response to methylmercury.

In the Seychelles study 28 PCB congeners ranging from congener 28 to 206 were measured in each serum sample obtained from 49 of the study children at 5 years of age. All samples had non-detectable level of any of the PCB congeners.

Also, it was speculated that exposures to organochlorine pesticides may be higher in the Faroe Islands than in the Seychelles because they tend to be higher in pilot whale blubber than in fish. For completeness it would be helpful to have comparable information on exposures to a number of organochlorine compounds in the two populations.

Selenium was measured in cord tissue in the Faroe Islands study and evaluated as a possible confounder, but failed to meet the criterion for inclusion in the data analysis. No data were available from the Seychelles study. It may be useful to develop comparative data on selenium levels for the two studies, if possible. The nature and direction of the effect of selenium on methymercury neurotoxicity, if any, is not well understood in experimental animals or presumably, in humans.

Omega-3 fatty acids have been shown to have beneficial effects on brain development in infancy so that the expected effect of omega-3 fatty acids is attenuation of methylmercury effects (Figure 1). Differences in such levels within or between populations might help to explain the disparity in the outcomes. Plasma levels of omega-3 fatty acids parallel fish intake. Both fatty fish and pilot whale blubber are especially rich in essential fatty acids, and in the Faroe Islands study the frequency of whale blubber meals during pregnancy was a significant predictor of increased serum levels of essential fatty acids. Also, higher mercury exposures predicted higher concentrations of essential fatty acids. It may not be possible to compare the two cohorts directly, but it may be of interest to compare the available data for the two populations as a whole to see if there are significant differences in omega-3 fatty acids. It should be noted that Faroese mothers who consume pilot whale meat may not necessarily consume pilot whale blubber.

The study of methylmercury almost always involves dietary exposures, via seafood or fish consumption. Since exposures generally increase with fish/seafood intake, this presents problems in detecting effects of the toxicant in the context of varying diet. Moreover, methylmercury absorption and toxicity may be affected by specific nutrients in the diet. Absorption and toxicity of other metals are affected by interactions with nutrients but there is less information regarding methylmercury. The Seychelles, Faroes and Amazon studies acknowledge the absence of such detailed dietary/nutritional information. The potential relationship between nutrients and the effect measured are generalized in Figure 1. Nutrient deficiencies and, although less common, toxicity alter growth and neural development. There are many examples of the effects of nutrient deficiencies on the developing central nervous system in infants and young children, including deficiencies in iron, niacin, copper, and vitamins B12, C and E (per: Confounders and Variables Panel report).

For all of the studies, the estimates of intake of fish (and whale meat as appropriate) should be used with information on methylmercury levels in the food to estimate possible methylmercury intake by the pregnant women, young children, and adults. If possible, attempts to validate estimates of intake using the relationship between hair mercury and diet intake from experimental studies should be made.

Information to assist in interpretation of potential postnatal exposure is needed. As shown in the Faroes study, methylmercury in breast milk was related to mothers fish/whale intakes, yet early postnatal mercury exposure did not seem to be associated with any deficits in the examinations conducted.

The collection of dietary data and anthropometric and biochemical measures of nutrient status in studies on neurotoxic effects of environmental agents allows for consideration of the possibility that the nutritional status of the population studied or other aspects of the diet either attenuate or exacerbate the effects of the agent under study. This kind of information, if available, could play a key role in explaining apparent differences in results between the Faroes and Seychelles studies.

7. Influence of Covariates

A comparison of the covariates used in the analyses of the various studies is presented in Table 1. It can be seen that more than 30 covariates were used in one study or another. Only four covariates were used in both primary studies.

Table 1: Covariates Used in the Various Studies

Covariates	IRAQ	SEYCHELLES	FAROE	AMAZON
Birth Weight		X
Raven Score		X	X
Age		Age adj. Scores	X	X
Sex		X	X	X
Gestational Age		X
BMI
Smoking (during pregnancy)		X	X	X
Duration Breast Feeding		X	X	X
Alcohol Use		X	X	X
PCB		X	X
Education M & P			X
Employment P/Income		X	X
Obstetric Care			X
Daycare		X	X
Computer Acquaintance			X
Examiner			X	NR
Birth Order		X
Maternal Age		X
Child’s Medical History*		X
Language at Home		X	NR
Maternal Medical History		X
Postnatal Experience
Hair Lead		X (whole preg.)	X (at parturition preg. – 45 days)
Mining Exposure				X
Malaria History				X
Parasite History				X

*Intra-uterine growth retardation or low head circumference

The analysis of alcohol as a covariate appeared to differ from the general strategy of handling covariates by the Faroe Islands group. In the summary of the answers prepared for the conference, the authors of the Faroes study state that alcohol was associated with lower mercury levels and with improved performance on some tests. If a variable is associated with both exposure and outcome (regardless of the direction of the association), it is a potential confounder and has to be accounted for. Despite their stated strategy, the authors decided to drop alcohol from their models. The reason given was that maternal alcohol is known to cause neurotoxicity in the offspring and not the opposite as suggested by the data. Alcohol may impact the toxicokinetics of mercury in the body. In an early paper, the authors describe how women in their cohort who consumed alcohol on a few occasions during pregnancy (none consumed alcohol regularly) had lower mercury levels in their cord blood, but it is not clear whether this means the fetal brain was exposed to more or less mercury.

The Seychelles study team and the Faroe Islands study team used different approaches in the way they handled potential confounding variables and intermediate variables in their analyses. This makes it difficult to compare the results of the two studies. Specifically, the Seychelles team tended to use many variables that may have been interrelated in their models. This raises the possible concern that they may have "overcontrolled" their statistical models, and thus, could have masked subtle effects in their data. Conversely, the Faroes team tended to use simpler models that may have omitted important variables associated with exposure and outcome, and that possibly may have created a spurious relationship between exposure and neurobehavioral outcome in their data. However, the strategy used in the Faroe Islands where the same battery of confounders were used in all regression analyses may have resulted in "overadjustment." Supplemental analyses provided by the Faroe Island group addressed several additional potential confounders.However, none of the panels identified a credible candidate confounder. A detailed analytic strategy should be worked out that could be applied to all methylmercury studies, although cultural differences may be a considerable obstacle.

8. Questions Regarding Exclusion or Inclusion of Data

The Faroe Islands study team excluded Year 2 results because of a problem with test administration. Exclusion of data on the grounds of improper test administration is appropriate, but it is important to show the reader concrete data demonstrating lack of validity, e.g., a comparison of the mean and standard deviation scores for Years 1 and 2 and/or correlation data demonstrating that the Year 2 scores fail to correlate with other test scores and sociodemographic background characteristics that they would be expected to relate to if they were valid. Optimally, such exclusion should be made before the data are analyzed. The failure of an outcome to relate to exposure is never grounds for the exclusion of data. This is also recognized by the Faroes team.

9. Concerns about Measurement Error Resulting in Underestimate of Risk

Measurement error can impact significantly on both the estimated levels of effect and the decision on the level of exposure at which an effect is detected because of potential for misclassification. However, the data presented in the workshop suggest that precision of measurements of methylmercury in hair or cord blood is very good. Contamination of hair is a potential problem but the exposure panel felt that this was not a major problem.

10. Problems in Consideration of Covariates in the Models

Variables for inclusion in the modifiers and variables models are supposed to reduce variance and control confounding. There are no universally accepted methods for selecting such variables. Some investigators choose variables solely on biological grounds; others use highly mechanized data-driven procedures to do this. Both of these research groups took a serious and plausible approach, but they did not do identical analyses.

For the Faroe Islands study, the publications from this group were not always clear about the temporal relationship between cord mercury, gestational age, and birth weight. The average birth weight in the Faroes study was about 3,600 g. Also, almost no children in the Faroese cohort were preterm. In their reduced model, the Seychelles studies group controlled for sex, birthweight, child’s medical history, maternal age, HOME score, caregiver IQ, SES, and hearing status. If methylmercury affected birth weight or child’s medical history, which in turn affected child development, overcontrol may have occurred because these variables would have been on the causal pathway.

Maternal smoking was left out of the models even though about 40% of the women in the Faroes cohort smoked during pregnancy. Again, it would have been helpful to see the actual data suggesting why the variable was dropped. During the workshop, it was pointed out that smoking is often associated with some of the electrophysiological measures used.

In the Seychelles study, the inclusion of birth weight in the models might obscure an exposure effect relationship if mercury or the fish that contains it has any biological effect on birth weight and on the causal pathway. Similarly, the Seychelles study includes duration of breast-feeding into the statistical models. Because length of breast-feeding is associated with both exposure and outcome, controlling it could obscure a possible effect.

For the Faroes study, town versus rural residence should have been used as a control variable for both studies (full cohort and cohort subset). Raven’s scores were higher in the town than in the rural areas, suggesting that parents in the town may have a cultural advantage for providing more optimal stimulation for their children. If so, a dichotomous town/rural variable could be used to provide additional statistical control for parental input. In contrast to most contemporary human behavioral teratology studies, the Faroes study decided not to administer the HOME Inventory to control for quality of parental intellectual stimulation. This decision is justifiable in terms of the uncertain applicability of the HOME to Faroese culture. Other variables that reflect quality of the intellectual environment that are collected in most studies (parental education and occupation and maternal intelligence as indicated by the Raven’s test) were also collected in the Faroes study. However, it is not clear that without the HOME these variables are sufficient to control for parental input.

11. Contributions from Experimental Studies

Experimental studies in rodents and nonhuman primates have provided background for the interpretation of human epidemiological studies. Methylmercury toxicity in laboratory animals is primarily manifested by deleterious effects on the nervous system, and the developing brain in laboratory animals is most susceptible to the effects of the chemical.

Because of the focus on prenatal effects of methylmercury in the human epidemiology studies, it is necessary to recognize the importance of the time of development when the exposure occurred, the level of intensity of the exposure, and possible mechanisms for the methylmercury toxicity. The importance of normal sequence of neuronal migration to subsequent development was emphasized by the experimental panel. Short-term higher rates of exposure (for example, a few hours or a few days) may interfere with the process of normal neuronal migration. This emphasizes the importance of peak exposures during significant developmental windows. By integrating this information, it is possible to predict the sublethal exposure to methylmercury during the first trimester that is most likely to affect neuronal proliferation and migration. Exposure during the second trimester will also affect migration and proliferation, but only of late-developing brain structures. The potential impact of methylmercury can be evaluated by knowledge of what brain regions are undergoing development during peak exposure and utilizing an evaluation strategy (test) to evaluate the functional modality served by these brain regions. Basic information to apply this strategy is incomplete. However, animal studies do provide some insights as to the nature of effects that might be expected in humans from methymercury exposure at different periods of development. Since the strategy will require extrapolation from animals to humans, comparative time frames for both brain development and methylmercury exposure need to be considered.

Patterns of neurobehavioral damage produced by developmental methylmercury exposure in animals resemble those found in humans and include sensory system effects, motor or sensorimotor system effects, and cognitive effects.

Sensory System Effects: Exposure of monkeys from birth to young adulthood produced lasting deficits in visual contrast sensitivity, elevated thresholds for pure tones, and elevated thresholds of detection of vibration with fingertip. Changes in visual evoked potentials have been observed in rodents in utero with no effect on auditory thresholds.

Motor (or Sensorimotor) Effects: Monkeys exposed to methylmercury from birth to young adulthood developed somatosensory impairment during middle age. Also in rodents, the most salient feature of methylmercury poisoning is impairment in motor function.

Cognitive Effects: The evidence for developmental cognitive deficits in experimental animals from methylmercury exposure is weaker than for sensory and sensorimotor effects. In this way, methylmercury differs from lead. Probably the strongest evidence of a cognitive effect is the observation of retarded object permanence and deficits in the Fagan test of memory during infancy in monkeys exposed in utero. Doses that produce overt toxicity may produce deficits on learning tasks. At lower doses, effects on cognitive endpoints have been largely negative in experimental animals. Such findings emphasize the importance of studies on domain-specific effects.

There is substantial evidence from experimental studies that exposure to PCBs during development may result in cognitive deficits. Effects in rodents include complex visual discrimination, delayed alteration performance, and a radial arm maze memory task. In monkeys, exposure in utero to postnatal age of 4 months to Aroclor 1248 or 1016 resulted in differences in performance on a number of visual discrimination reversal problems and in delayed alteration performance. Exposure from birth to 20 weeks of age to a congener mixture similar to that found in human breast milk resulted in an impairment on a number of tasks that may be characterized as learning deficit, and inability to inhibit inappropriate responding. Blood and fat levels of PCBs in treated animals were typical of levels in persons in industrialized countries.

It should be emphasized that at this time it is not possible to differentiate the effects of PCBs on neurodevelopment from effects of methylmercury because of the lack of mechanistic information for both toxins.

12. Generalizations from Workshop Discussions

The validity of the studies reviewed by the panels appears to be very good in that many of the expected relationships between covariates and outcomes were found. Whether the results of all three studies can be generalized to other populations is unclear for several reasons.

Both the Faroe Islands and Seychelles studies used different neurodevelopmental endpoints to assess possible effects of methylmercury exposures. While there is considerable evidence that methylmercury is a developmental neurotoxin, experimental studies suggest that it is difficult to differentiate potential effects of methylmercury from effects of other neurotoxins such as PCBs because of lack of mechanistic information.

There are many differences in the characteristics of the cohorts that make interpretation and comparisons between cohorts difficult. Both island populations are distinct in origin, genetic backgroundand in sociocultural conditions, and both differ substantially from the U.S. population. In many respects, the children of the Faroes study group and the Seychelles are very advantaged, as compared to many poor children in the United States, especially in terms of the quality of health care and education. The Amazon population, while clearly different from the other cohorts, offers information on the effects of methylmercury in a less advantaged population, but only adults have been studied to date.

In spite of the stated weaknesses and uncertainties, the finding of adverse effects in the Faroes study and in the Amazon study raises some concern that risks of lower exposure to methylmercury may exist. Of particular concern are exposures to women of childbearing age or pregnant women. Dietary stresses and co-exposures to other chemicals could plausibly enhance or alter risk, but there are inadequate data on this to draw meaningful conclusions at this time.

13. Summary of Panel Findings and Recommendations

Methylmercury is a developmental neurotoxin, but effects at low doses encountered by eating fish are difficult to evaluate.

All the studies reviewed were considered of high scientific quality and the panel recognized that each of the investigations had overcome significant obstacles to produce important scientific information. The panel also stated that continued funding of the studies in the Seychelles, Faroes, and Amazon is necessary for the full potential of those studies to be realized in assessing developmental neurotoxic effects of methylmercury. Studies would benefit by evaluation of common endpoints using similar analytical methods.

Results from the Faroes and Seychelles studies are credible and provide valuable insights into the potential health effects of methylmercury.

Some differences are clearly present in the results from the Faroes and Seychelles studies, the two studies that examined the possible effects of methylmercury in children, but the panels were not able to identify clearly the sources of these differences. Among possible sources are the different effects of episodic versus continuous exposure, ethnic differences in methylmercury responses, lack of common endpoints in the Faroes and Seychelles studies, and several other confounders or modifying factors such as those found in diet and lifestyle, and in chemicals present in seafood, which is the source of methylmercury to these populations. The other chemical constituents of seafood that may be explanatory include those that may be beneficial to fetal neurodevelopment (i.e., omega-3 fatty acids) and those that may be harmful to fetal neurodevelpment (e.g., PCBs).

The studies have provided valuable new information on the potential health effects of methylmercury but significant uncertainties remain because of issues related to exposure, neurobehavioral endpoints, confounders and statistics, and design.

The interagency organizing committee unanimously agreed that the deliberations of the panels and the workshop report will be a key factor in subsequent public health policy actions taken by each of the participating agencies.

Patterns of Exposure

• Analyze hair samples segmentally or continuously.

Although some segmental analyses have been done for both of the large cohorts, the number of cases is insufficient to examine the question of possible correlation of outcomes with patterns of exposure (e.g., peak level, frequency of excursions above a certain level, peaks within a hypothetical sensitive "window," etc.). Segmental analyses of hair, when performed, have measured approximately monthly averages (ca. 1.1 cm segments), generally by a cold-vapor atomic absorption technique. In some cases X-ray spectrometry, with a resolution of about 2 mm, has been used to provide "continuous" single strand mercury analysis and to validate the cold-vapor atomic absorption data. Continuous single strand mercury analyses by x-ray spectrometry might be used to determine the time course of exposures more precisely.

• Conduct more complete dietary surveys to characterize intake of fish/pilot whale over time.

More detailed data are needed to characterize the patterns of exposure and their possible correlation with neurobehavioral test outcomes. New studies should include data regarding measures of fish consumption (frequency, amount, species) in new studies. Population-based surveys could be helpful in evaluating possible significance of peak exposures, variability as a function of type of fish consumed, and exposures to other potentially relevant substances.

• Examine the "spiking effect" of whale dinners in the Faroe Islands.

Pilot whale, on average, has substantially higher concentrations of methylmercury. Traditional Faroese whale dinners may result in a "spike" (short-term peak) in blood methylmercury levels. Special studies could be designed to measure the magnitude of this effect.

• Compare full exposure distributions for the populations of interest in the Faroe Islands and the Seychelles.

Population-based data concerning the characterization and comparison of the distributions of methylmercury exposures for both populations could help in modeling these exposures retrospectively and in developing realistic distributions for comparisons

• Determine historical exposures in the Amazon populations.

The Amazon study described at the workshop provides an interesting and credible rationale for the seasonal variation in methylmercury exposures (based on segmental analysis of hair samples), because of the parallel seasonal variation in consumption of fish with high levels (piscivorous or omnivorous) or low levels (herbivorous) of methylmercury. There should be follow-up studies of historical exposures in the region (e.g., from sediment core samples, other populations) in view of the unique processes thought to result in these exposures, and the possible significance of this parameter should be considered to the extent that it is feasible.

Other Possible Exposures

• PCBs–Extend quantitative assessment of exposures.

Study should be continued to determine partitioning of PCBs between mother, cord tissue and fetus. The possibility that PCBs may modify the effects of methylmercury needs to be thoroughly examined.

• Selenium–Compare levels in Faroe Islands and Seychelles.

Selenium was measured in cord tissue in the Faroe Islnds and considered, but not included, as a potential confounder in data analysis. No data were available from the Seychelles study. It may be useful to obtain comparable data from both studies.

• Omega-3 fatty acids–Compare existing data for the two populations

Omega-3-fatty acids have been shown to be beneficial for brain development. Any difference in omega-3 fatty acids intake between the children in the two studies might help explain disparities in outcomes.

Neurobehavioral Endpoints

• Introduce more complex testing as cohorts age.

The two study cohorts are aging, future studies should begin to include complex and challenging measures of executive functions, information processing, and social behavior. Also, as subjects reach preadolescence, the studies should consider introducing measures of somatosensory function such as vibrotactile thresholds, olfaction, and hearing, since these are the domains commonly affected by mercury exposures in adults.

• Studies should use common assessments for direct comparisons.

The Neurobehavioral Endoints panel recommends that the two prospective studies adopt similar methods of statistical analysis and presentation. While each study presents different challenges to the analysis (e.g., differing confounders, covariates, covariance stuctures), the studies should strive to pursue methods that will allow meaningful comparison of findings. The two current prospective studies should cooperate in the endeavor.

• Prospective studies include multiple assessments of single domains in their neuropsychologic batteries.

Domains that should be tested include language, attention, executive functions, psychomotor speed, neuromotor coordination, and social behavior. IQ is an outcome variable that is useful in comparing studies.

• Introduce challenging test conditions.

Experimental and clinical literature has shown that the introduction of challenging test conditions such as irrelevant, intrusive stimuli reveals deficits that are not seen in the absence of such perturbations.

• Analyze individual child performances.

Analysis of patterns of performance tests using the individual child as the level of analysis may be helpful in identifying whether there is a specific neurobehavioral signature of injury from methylmercury exposure.

Statistics and Design Recommendations

• Studies of cohorts should be comparable in order to allow qualitative and quantitative research synthesis.

There should be a systematic and detailed review of similarities and differences in target populations, study populations, exposure measurements, outcomes, and confounders in the studies.

Quantitative research synthesis will not be possible until 8-year assessments. There should be substantial overlap in assessment instruments. The studies should involve the following covariables: hair mercury, age at testing (as a linear term, categorized into years), gender, prenatal experience, maternal Raven is score (entered in a linear fashion, entered as categories), and breastfeeding data.

Recommendations from Experimental Studies

Consider chronology of development of human nervous system in relation to peak concentrations of mercury in the maternal–fetal unit.

Tests of sensory and motor function should be included in test battery.

Future testing of cohorts should be comparable and involve testing of specific domains.

Social behavior should be assessed.

Growth data already collected should be analyzed, and the determination of height and weight should continue through puberty.

Investigators are encouraged to stay current with emerging PCB literature in humans (e.g., the Oswego study) and animals.