Data gathering and inclusion criteria.
Our guiding principle for building the Seafood Hg Database (Karimi 2012
) was to focus on fish and shellfish from sources that could reasonably be sold in the United States. Our database was developed to reflect the range of possible Hg levels for seafood items considered the top contributors to human Hg exposure in the United States because they are relatively high in Hg and/or they constitute relatively large shares of the U.S. seafood market (top 51 Hg contributors defined by Groth 2010
). Detailed taxonomic and geographic harvest information is often lacking or incorrect in the seafood marketplace (Jacquet and Pauly 2008
). Thus, our database does not model the exact composition of the U.S. seafood market. Rather it reflects the range of seafood species and seafood Hg concentrations that are available to U.S. seafood consumers.
Data were gathered from federal and state government reports and from peer-reviewed scientific literature. We obtained data from federal and state government agencies that either made their fish tissue monitoring results publically available online [e.g., State of North Carolina (2011)
; State of Virginia, Department of Environmental Quality (2009)
; U.S. Environmental Protection Agency (EPA) National Coastal Assessment (U.S. EPA 2008); U.S. FDA raw data (FDA 2011)] or provided data upon request (e.g., State of Delaware, State of Hawaii). In addition, we searched for published, peer-reviewed papers indexed in the Web of Science (Thomson Reuters, New York, NY) before 15 December 2010. We conducted literature searches for individual seafood items based on seafood varieties listed as the top 51 Hg contributors to the U.S. population (Groth 2010
). Search terms included “mercury” and the common names of these fish or shellfish (e.g., “mercury AND salmon”) [see Supplemental Material, Search Terms for Table S2
From the data gathering and search results, we included select studies or select data from studies on fish and shellfish from sources that were likely to enter the U.S. seafood market. We included data on edible portions (fillet or whole fish) of any fish or shellfish species likely to be included in the top 51 seafood varieties (e.g., “redfish” were included with “ocean perch”) based on federal commercial fisheries landings (fisheries landed and sold in the United States) and seafood import statistics (NMFS 2007b). Hg concentrations in whole fish can be lower than concentrations in fillets (Goldstein et al. 1996
), probably because Hg is primarily associated with muscle tissue. Thus, the inclusion of data based on fillets as well as whole fish, which are common in the market particularly for smaller fish such as anchovies, may underestimate fish Hg content relative to those based on fillets only. We classified seafood items as being from domestic or imported sources based on geographic locations specified in the original study. We assumed that all marine fish caught commercially from domestic waters were relevant to the U.S. market. Data for a given fish or shellfish species collected from market basket studies or direct harvest from countries outside of the United States were included only if at least 5% of all imports of that species into the U.S. fisheries market (by volume) were from that country according to NMFS import statistics as of 2010 (NMFS 2011b). In addition, imported seafood items that did not meet this criterion were included if the samples were collected from water bodies connected to other countries that meet this criterion. Highly migratory fish caught from major ocean basins (tuna, shark, and swordfish) were included regardless of country of origin.
Of the top 51 seafood varieties, < 10 are freshwater fish. For most freshwater items collected from domestic waters, we included data from the Great Lakes because the Great Lakes are the main sources of these species to the market (NMFS 2011c). We did not include salmon species from the Great Lakes because the commercial catch of salmon from the Great Lakes has been negligible for at least one decade (Baldwin et al. 2009
). For striped bass (Morone saxatilis
), we included data for wild fish only from Atlantic states because commercial fisheries do not exist for this species in the Gulf of Mexico or Pacific coast (NMFS 2007a). For catfish, carp, and perch, we included fish collected from Atlantic or Gulf Coast states that report commercial landings of these fish (NMFS 2007a), excluding samples from interior or landlocked freshwater sources. Data for farmed species of commercial freshwater fish were included if the fish were specifically raised for consumption (e.g., farmed catfish), and the fish were of market size (vs. juvenile fish from hatcheries) and were fed conventional feed (e.g., Berntssen et al. 2010
Exclusion criteria. We screened approximately 1,000 government monitoring programs and peer-reviewed academic studies for inclusion. Upon critically examining each study, we excluded entire data sets—or select data from studies—based on one or more of the following criteria:
- Data resulting from experimental Hg exposures.
- Data on fish or shellfish that are not a primary source of commercial fish to U.S. consumers, based on the geographic location of collection.
- Studies that were not written in the English language.
- Data that were repeated from another source already included the database; for example, data repeated in review papers as well as original papers, or data repeated in aggregate federal government databases (e.g., U.S. EPA National Listing of Fish Advisories) and original state data sources (e.g., State of North Carolina). Duplicate entries were routinely screened for and excluded from all calculations.
- Data for fish from locations with known point source Hg contamination or associated fisheries closures.
- Data for young-of-year fish (born within the past year). However, we included Hg values from other smaller body size fish that may be excluded from the U.S. market because of catch restrictions. Hg concentrations tend to be lower in small fish than in larger fish of the same species, thus may lead to underestimates of the true average of Hg values in U.S. commercial fish.
- Studies conducted by nongovernmental organizations, public interest groups, or news media that were not peer-reviewed or incorporated into government monitoring efforts.
- Studies that did not report the necessary Hg data (raw data, or arithmetic mean Hg or MeHg concentration and sample size). For example, we excluded studies that presented Hg concentrations in a graph or as a range, geometric mean, or median. Geometric means and medians were rarely reported in the literature. Therefore, we included only arithmetic mean Hg concentrations, or we calculated arithmetic means based on raw data when reported.
- Data from areas with no commercial fishing activity, such as no-take marine reserves and national parks (e.g., Rencz et al. 2003; Wyn et al. 2009).
We extracted mean Hg concentrations (parts per million, wet weight), sample size, and geographic location for each seafood item reported in each study. Approximately 40% of the included sources reported SDs or SEs. Thus, analyses requiring SDs or SEs would exclude the bulk of the data set. Therefore, we focused on examining mean Hg concentrations in the interest of including the range of Hg concentrations for each seafood item using the largest possible data set. We extracted total Hg values whenever possible, but we used MeHg values when they were reported instead of total Hg. Approximately 95% of total Hg in fish muscle tissue occurs in the form of MeHg (Bloom 1992
). Therefore, we assumed that MeHg concentrations are similar to total Hg concentrations. Nevertheless, because MeHg concentrations are lower than total Hg, our calculated, grand mean Hg concentrations for certain seafood items may be slightly lower than if they were based solely on total Hg concentrations. Hg values reported as dry weight concentrations were converted to wet weight concentrations according to moisture content, if reported, or by assuming 80% water content. When Hg concentrations were reported as nondetects (approximately < 10% of all database entries), we entered values as one-half the detection limit from the study (Clarke 1998
); when detection limits were not reported, these values were excluded.
When a study reported multiple mean Hg values for a given seafood item (e.g., Hall et al. 1978
), we calculated a weighted mean using sample size for the mean as the weight. When a study reported multiple Hg values for a given seafood item but did not provide sample sizes for individual values (e.g., Cossa et al. 1992
; Deshpande et al. 2009
; Jackson 1991
), we assumed sample sizes were equivalent across values. Thus, overall means calculated from these studies were not weighted.
Data analysis. We calculated an aggregate, grand weighted mean (H-g-w) for each seafood item based on means weighted by sample size across studies
where Hgi is the ith reported mean and wi is the weight (reported sample size) of the ith observation. We estimated variability of Hg in seafood items by calculating a weighted grand SD, corresponding to the grand mean. The Seafood Hg Database comprises mean Hg values reported by individual studies as observations, as opposed to raw Hg data values. By definition, the SD of sample means is the SE of the global distribution of Hg values. Therefore we estimated the weighted SE (SEw) of the distribution underlying the grand mean using the formula for the weighted SD,
where N is the number of studies from which mean Hg values were collected. To obtain the weighted SD (SDw) of the global distribution, we multiplied SEw by the square root of average sample size across studies for each seafood item, yielding the formula
Monte Carlo simulations tested for potential bias of Equation 3
using hypothetical data approximating the composition of the database. Specifically, we simulated a true standard deviation of the global distribution using random numbers drawn from normal and lognormal distributions, where wi
ranged from 2 to 100 and N
ranged from 50 to 300. Tests of 10,000 replicates demonstrated that Equation 3
was an unbiased estimator of the true standard deviation of the global distribution and was insensitive to both the type of distribution used and variation in sample size (data not shown).
We calculated the a) grand mean, b) grand SD, c) range (minimum and maximum reported means), d) coefficient of variation (CV), and e) total number of samples across all studies for each seafood item name searched (e.g., salmon), as well as for seafood items with higher taxonomic resolution within the search results (e.g., Atlantic salmon) and for broader taxonomic categories for specific analyses. Thus, results are presented for a larger number of seafood groups than the original top 51 seafood items from the search. We compared our findings with summarized Hg data accessed from the FDA-MP (2011) on 15 September 2011 for seafood items for which direct comparisons were possible given available data (58 seafood items). In some cases, seafood items were grouped together into larger seafood categories, which often included multiple taxa. For example, for direct comparison with Hg concentrations for “crab” reported by the FDA-MP, we grouped together blue crab, king crab, and snow crab data. Formal parametric statistical comparisons, such as analysis of variance, were not possible for our analyses because the database is composed of aggregate mean Hg values instead of raw data. Thus, unknown distributions of the underlying Hg data, together with unequal sample sizes for the comparisons of interest, made statistical comparisons inappropriate for our study. Finally, we calculated the percentage of studies reporting a mean Hg concentration exceeding the FDA action level (1 ppm) and the U.S. EPA human health criterion (0.3 ppm) for seafood items with relatively higher taxonomic resolution when possible in order to yield more detailed results than those from broader seafood categories. The FDA action level for MeHg of 1.0 ppm represents the threshold above which the agency can take legal action (e.g., removing the product from the marketplace) (FDA 2007). The U.S. EPA MeHg criterion of 0.3 ppm represents the fish tissue concentration that should not be exceeded for safe consumption of sport-caught fish in local waters based on average consumption (U.S. EPA 2001).
To compare farmed items to wild-caught items within the same seafood category, we focused on species with established or emerging, rather than nascent, farming or ranching industries. For some seafood categories, the species composition of farmed and wild items was not identical. For example, wild-caught catfish include channel catfish, blue catfish, and brown bullhead, whereas farmed catfish include channel catfish and striped catfish. We designated individual data as farmed or wild according to information from original studies. When farmed or wild status was not reported, as with some market basket studies, we made assumptions based on FAO fisheries statistics for individual species (FAO 2011). Specifically, we assumed that lake trout were wild-caught and rainbow trout were farmed. For eel species from market studies, we assumed that Japanese eel (Anguilla japonica) were farmed and European conger eel were wild caught. Finally, we assumed that Atlantic salmon from market studies in North America and Europe were farmed unless otherwise specified, given the endangered status of wild Atlantic salmon.