The exposure of the Arctic ecosystem to persistent organic pollutants (POPs) was assessed through a review of literature data. Concentrations of 19 chemicals or congeneric groups were estimated for the highest levels of the Arctic food chain (Arctic cod, ringed seals, and polar bears). The ecotoxicological risk for seals, bears, and bear cubs was estimated by applying the concentration addition (CA) concept. The risk of POP mixtures was very low in seals. By contrast, the risk was 2 orders of magnitude higher than the risk threshold for adult polar bears and even more (3 orders of magnitude above the threshold) for bear cubs fed with contaminated milk. Based on the temporal trends available for many of the chemicals, the temporal trend of the mixture risk for bear cubs was calculated. Relative to the 1980s, a decrease in risk from the POP mixture is evident, mainly because of international control measures. However, the composition of the mixture substantially changes, and the contribution of new POPs (particularly perfluorooctane sulfonate) increases. These results support the effectiveness of control measures, such as those promulgated in the Stockholm Convention, as well as the urgent need for their implementation for new and emerging POPs. Environ Toxicol Chem 2016;9999:1–12. © 2016 SETAC
Persistent organic pollutants (POPs) are chemicals that remain intact in the environment for long periods, travel long distances, accumulate in living organisms, and are toxic to humans and wildlife. They are capable of transport via air, water, migratory species, and technical matrices such as products and wastes;, thus, they become ubiquitous in the environment. Their potential for long-range transport is the primary reason for regional, continental, and global distribution [1, 2]. The different physical and chemical properties of these substances (differences in persistence, water solubility, mechanisms of bioaccumulation, mechanisms of toxic effects) also present a number of challenges [3, 4].
Control and management of POPs is hindered by their complex emission patterns and releases into the environment. Some POPs were primarily applied in agriculture, meaning that they were directly released into the environment, where they contaminated abiotic compartments and living organisms. Some were produced and applied as industrial products or intermediates. Many are produced unintentionally as byproducts of various industrial and combustion processes and also by the natural transformation of primarily released substances.
The Stockholm Convention on POPs  is a global treaty focused on substances that are toxic, are resistant to degradation, and have a strong potential to accumulate in humans and other living organisms. In the first formulation, the Convention addressed a series of legacy POPs known as the dirty dozen (such as polychlorinated biphenyls [PCBs] and dioxins). Subsequently, the list has been periodically updated to include new chemicals, including legacy and emerging POPs.
The strict regulation and, in many cases, the total ban on the most harmful POPs will reduce the global impact of these chemicals. However, because of their persistence and biomagnification potential, the presence of POPs in the global environment is likely to represent a risk to wildlife and humans for decades.
At present, this risk mainly depends on a wide spectrum of sources, including former production and application processes as well as new types of environmental emissions. Environmental compartments, which have been highly contaminated for decades, serve as important sources of the current degree of contamination. These secondary sources, developed mainly during the second half of the last century, represent a significant fraction of the globally distributed expanse of chemicals and are an important source of contamination, especially in remote areas.
Wildlife and humans from polar areas are subject to a level of risk that may be substantially higher in comparison with tropical and temperate regions where major emissions took place. This is because of the long-range distribution patterns of POPs and their persistence in cold climates, as well as the diet and the high fat content of organisms at higher trophic levels. Since the late 1960s, the presence of detectable concentrations of POPs and the relevant risk for the biological community have been discovered and documented in Antarctica and in the Arctic [6-11].
As a result of distribution and fate patterns, the Arctic environment is exposed to a complex mixture of POPs that exhibits a composition quite different from that typical of emission areas and that is subject to changes over time, with a progressive decrease in legacy POPs and a possible increase in emerging contaminants. Extensive programs have been developed for studying POP contamination in the Arctic [12-16], and several reviews have described the temporal trends in biotic and abiotic matrices of these compounds before and after the Stockholm Convention [10, 16-21].
The effects of POPs on the Arctic ecosystem have also been studied, particularly in terms of biomagnification and consequences for the organisms at the top of the food chain (such as seals and polar bears) [17, 22-25]. In spite of the large amount of information on these topics, a characterization of the risk of the mixture of POPs for a simplified Arctic food chain (cod, seal, polar bear), comparing environmental concentrations with a given toxicological endpoint, has never been attempted. The purpose of the present study, based on an extensive review of monitoring data produced over the last 4 decades, was to estimate the composition of the POP mixture likely present in recent years within the Arctic environment and to assess the risk of the mixture for the Arctic animals on the top of the food chain (ringed seal, polar bear). Trends over time for the mixture composition and risk were also estimated, highlighting the changing ecotoxicological role of individual components. The work also contributes information that could be used to assess the effectiveness of control measures (in particular, the Stockholm Convention) in reducing the global risk of POPs, to estimate the time needed for a substantial reduction in the risk of legacy POPs, and to highlight future research priorities for emerging potential POPs.
MATERIALS AND METHODS
The selected chemicals
All of the original POPs, most new POPs, and POP candidates listed in the most recent iterations of the Stockholm Convention were considered (Supplemental Data, Table S1). Some chemicals were excluded (e.g., chlordecone, pentachlorobenzene) because detailed information on the concentrations in the various matrices of the Arctic ecosystem was lacking.
The list of the compounds considered (Table 1) includes some individual chemicals often present in the environment together with some analogous compounds (congeners, isomers, metabolites) and some large groups of congeners or similar compounds. A complete description of the chemicals selected is given in the Supplemental Data (Section 2).
|Polar cod (ng/g lipid wt)c||Ringed seal (ng/g lipid wt)c||Polar bear (ng/g lipid wt)c||Bear milk (calculated) (ng/g wet wt)||Bear milkb (measured) (ng/g wet wt)||Seal/cod||Bear/seal|
Study area and food chain
The data collected cover a large sector of Arctic and sub-Arctic regions. Most data refer to the area between the Svalbard Islands and Alaska. Only the Russian Arctic is poorly covered. The distribution of sampling areas and the quantitative coverage of different Arctic regions is shown in Supplemental Data, Figure S1.
To describe the behavior of POP mixtures in the Arctic ecosystem, a simple food chain has been considered. Data on POP concentrations have been collected for fish (Arctic cod, Boreogadus saida), ringed seal (Pusa hispida), and polar bear (Ursus maritimus) as one of the most representative Arctic top predators, classified by the International Union for Conservation of Nature  as Vulnerable. The Arctic cod–ringed seal–polar bear food chain has been very well defined, and is typical of a low-biodiversity ecosystem such as the Arctic, with simple predator–prey relationships .
The diet of polar bears has been studied extensively. Polar bears predominantly eat the blubber and meat of ringed seals (Phoca hispida) or other seal species, as well as other marine mammals . The effects of climate changes may affect the diet of polar bears , and diet composition may affect the contaminant burden. However, the consequences of global changes have not been considered in the present study, and the main assumption is that polar bear feed only on ringed seal. A similar diet simplification was assumed recently by Pavlova et al.  in their modeling exercise.
Sources of variability and uncertainty in the risk assessment
The fundamental hypothesis assumed is that in remote areas, far from emission sites, where only long-range transport may occur, environmental and geographical characteristics are the main factors affecting the environmental concentrations of POPs. It may be supposed that, because of many characteristics (relatively uniform cold temperature, absence of dry land, etc.), the Arctic region, as defined by the Arctic Monitoring and Assessment Programme (AMAP) , is a relatively consistent area, at least for the purposes of a general assessment such as the present study. Therefore, POP concentrations may be assumed to be relatively homogeneous, and data from different literature sources may be considered as comparable.
These assumptions must be regarded with care and considered to refer to the objective and scale of the present study. Meteorology and transport of POPs via air, ocean currents, and rivers are different in various Arctic subregions. This may produce differences in water concentrations in the different Arctic basins. Moreover, in top predators such as polar bears, concentrations may be affected even by regional dietary differences . Additional sources of variability derive from the origin of the literature data (different survey programs, sampling procedures, laboratories, etc.) as well as other confounding factors such as sex and age.
McKinney et al.  showed that in 11 polar bear populations, distributed from Svalbard to Alaska, the variability of concentrations of PCBs and polybrominated diphenyl ethers (PBDEs) is within a factor of approximately 3, excluding only the 2 very southern populations of Hudson Bay located outside the Arctic Circle. Even including these populations, the variability is within 1 order of magnitude. In ringed seals and polar bears from the North American Arctic (Canada and Alaska) Braune et al.  observed a moderate variability, usually within a factor of 3, for many legacy POPs. They observed that the effects of sex and age may be more relevant than geographic differences.
We are aware that this variability is not negligible. However, the objective of the present study is not a precise description of POP distribution in the Arctic, but rather an evaluation of the ecotoxicological risk determined by POP concentrations that might realistically occur in the Arctic environment. For a large-scale assessment (in space and time) of ecotoxicological risk from POP mixtures, like those performed in the present study, a certain level of variability may be assumed to marginally affect the general assessment and conclusions. Considering the sources of uncertainty in this kind of assessment (toxicity data, application factors, mixture assessment, etc.), a geographic variability within a factor of 2 or 3 may be considered acceptable.
All of the data we collected confirm the hypothesis of a moderate spatial variability. The variability of concentrations of the same chemical in samples collected in different Arctic sampling areas over sampling periods of 5 yr is relatively small, usually not higher than a factor of 3. This confirms the hypothesis that in the Arctic, the variability of POP pollution, determined by long-range transport, is not comparable to levels observable at emission sites. It also confirms the suitability of the data collected data for the objectives of the present study. Details of the assessment of variability within the dataset are given in the Supplemental Data (Section 4, Table S3).
To perform a risk assessment of a POP mixture, a realistic quantitative composition of the mixture that includes all 19 chemicals (or congeneric groups) selected must be defined. The number of POPs analyzed varies substantially by publication. Therefore, finding data on all of the 19 selected chemicals analyzed in the same sample was impossible. Thus, different studies reporting data on different POPs in a given environmental matrix, referred to a given temporal window, were used for assessment of the mixture composition. This result may be assumed to be a realistic mixture for the Arctic environment, even if data do not refer strictly to the same geographic area. This procedure is also supported by the moderate spatial variability, as described above.
Literature survey. Published data on POP concentrations in Arctic biota were collected from scientific journals or technical reports available online. The sampling period considered for data selection included more than 40 yr (from the late 1960s to 2011). For assessing the risk to biota, the average of a 5-yr period (2006–2010) was used. For this period, data were available for most of the chemicals, at least for seals and polar bears. If not available, data from the previous 5-yr period (2001–2005) were used. More recent data are rare in the literature.
For lipophilic chemicals, whole-body concentrations in B. saida and in fat of P. hispida and of U. maritimus were considered. The results were normalized to lipid content; data given in the original study on a dry or wet weight basis were converted to a wet weight lipid basis. A different approach was used for perfluorinated compounds (perfluorooctanoic acid and perfluorooctanesulfonic acid [PFOS]), which are water-soluble and accumulate in proteins instead of lipids. For these chemicals, data are expressed as wet weight concentration in whole body for fish and in liver for seal and bear.
Age and sex of the organisms analyzed were not considered as a discriminating factor. Particularly in older studies, these details were not reported. Additional support for this choice is given in the Discussion section.
If a given study reported multiple observations for a single chemical, the geometric mean of all values was calculated. In some studies, the geometric mean was directly reported by the authors. All the geometric means of single studies, within the temporal window selected (2006–2010), were collected to calculate the final geometric mean for the period, weighted as a function of the number of observations. The concentration values obtained, reported in Table 2, may be considered the realistic concentrations in biota in the Arctic for the selected temporal window.
|ADI (mg/kg body wt/d)||Seal||Bear (adults)||Bear (cubs)|
Calculated data. Some missing data have been estimated through calculation. In particular, concentrations of lipophilic chemicals in polar bear milk have been calculated from concentrations in bear fat, assuming an equilibrium between body and milk lipids and a lipid concentration in milk of 33% 
For perfluorinaled chemicals, which accumulate in proteins, concentrations in milk were calculated from concentrations in bear liver assuming an equilibrium between liver and milk proteins and protein concentrations in milk and liver of 10% and 20%, respectively [32, 33]
The most appropriate endpoint suitable for assessing the risk from such a complex mixture should be selected, particularly taking into account that it is a multicomponent mixture with individual components present at low or very low concentrations or doses. The different classes of POPs may have completely different toxicological modes of action. Many of them are known to be endocrine disruptors. Many other effects are known, at least for humans and mammals (neurotoxicity, immunotoxicity, liver toxicity, etc.). However, precise endpoints—which, in human toxicology, often refer to a specific target organ—are meaningless in ecotoxicology because of the different objectives of environmental protection (protecting the structure and functions of the biological community) and of human health protection (protecting individuals). Moreover, in ecotoxicology, knowledge of the toxicological modes of action on all the different types of organisms that may be present in an ecosystem is largely incomplete. Finally, in spite of the recognized vulnerability and fragility of the polar ecosystem, very few specific data are available on toxic effects of contaminants to polar biota [22, 23].
To describe the toxicological behavior of a mixture of chemical substances, 2 different approaches may be used: the concentration addition (CA) or the independent action (IA) model. The 2 models are applicable to chemicals with the same mode of action or with a different mode of action.
However, the current status of our knowledge may justify the general use of CA as a pragmatic default approach to the predictive hazard assessment of chemical mixtures . The use of CA as a reasonable worst-case approach to the predictive hazard assessment of chemical mixtures has been supported in a recent European Commission document . A more detailed justification of the use of the CA approach is given in the Supplemental Data (Section 5).
Therefore, in the present study, the risk of mixtures has been estimated using CA as a worst-case approach. Risk was calculated at 3 different levels: seals eating fish; adult bears eating seals; and bear cubs consuming milk. The risk to fish was not calculated because of the lack of comparable long-term toxicity data for all the chemicals considered.
For all types of effects, to apply the CA approach, the same toxicological endpoint must be used for all chemicals considered. Moreover, as a result of the problem of long-term exposure to relatively low doses, long-term toxicity data should be preferred. Finally, because most POPs are endocrine-disrupting chemicals, endpoints dealing with reproduction and development should also be preferred.
The requirements listed are optimal; however, one must be aware that the most relevant drawback in selecting a suitable endpoint is the availability of reliable toxicity data for the same endpoint for all components of the mixture. Therefore, the only possibility of performing at least a preliminary assessment is accepting rough compromises, using the data that are available and relatively homogeneous for all the chemicals examined.
For mammals, several types of short-term and long-term data were available. However, comparing methods and endpoints was quite difficult. In the absence of suitable data for the same relevant endpoint for all chemicals, we decided to use the hazard index approach [36, 37]. Hazard quotients for individual chemicals were calculated using the acceptable daily intake (ADI) proposed for protecting human heath by international organizations (the World Health Organization, the Food and Agriculture Organization of the United Nations, the European Food Safety Authority, and the US Environmental Protection Agency) as the reference value
where HImix is the hazard index of the mixture, HQi is the hazard quotient of the individual chemical, TDIi is the total daily intake of the individual chemical, and ADIi is the acceptable daily intake of the individual chemical.
The following hazard quotients were considered: seal hazard quotients calculated as the ratio between TDI (fish eating) and mammal ADI; bear hazard quotients calculated as the ratio between TDI (seal eating) and mammal ADI; bear cub hazard quotients calculated as the ratio between TDI (milk eating) and mammal ADI.
The TDI is calculated according to the following equation
where DFI is the daily food intake (fish, seal fat and milk for seals, bears and bear cubs, respectively) and CF is the concentration in food. For TDI calculations, the following assumptions were made.
where Cfish is the concentration of the chemical in fish (µg/kg lipid wt).
TDI for bears. The assumptions were as follows: polar bears eat mainly seal fat , and in the present assessment, it was assumed that the whole food requirement is covered by fat; daily food (seal fat) intake = 2% of body weight/d = 0.02 kg wet weight/kg body weight .
where Cseal is the concentration of the chemical in seals (µg/kg lipid wt).
TDI for bear cubs. The assumption was that daily food (milk) intake = 20% of body weight/d = 0.2 kg wet weight/kg body weight .
where Cmilk is the concentration of the chemical in bear milk (µg/kg wet wt).
We are aware that the approach described presents some critical issues. First, in particular, the ADI developed for humans was used as a reference endpoint for Arctic mammal toxicity. We assumed that it was a possibility for using a comparable reference endpoint for all chemicals considered. Toxicity data of different POPs for seals and bears are rare and hardly comparable. Therefore, extrapolating a protective value developed for a mammal species (human) to other mammalian species (seal and bear) was assumed to be the best solution for getting a reasonable indicative value for a protective endpoint. Second, the objective of human risk assessment (protecting individuals) is different from the objective of ecotoxicological risk assessment (protecting structure and functions of ecosystems). However, for a threatened species, such as the polar bear, the protection of individuals may also be relevant from an environmental point of view. Moreover, seals and bears are K-strategic species that are keystone species for the Arctic ecosystem. Therefore, changes in their populations may produce substantial alterations in the structure and functioning of the system.
For those chemicals with enough data available for a long temporal window, a time trend has been described. Linear regression is clearly unsuitable to model temporal trends of log-concentrations of contaminants because of nonlinearity in the true pattern. Moreover, quantifying the ecological risk that such contaminants will exceed reference levels is a fundamental issue. Therefore, a suitable statistical method that is reliable from the predictive perspective is needed. Nonparametric regression  has recently been used to model environmental time series [42, 43]. In particular, the conditional expectation of the response variable Yi (contaminant log-concentration) is assumed to be equal to a smooth function g(xi) of the covariate xi (i.e., time), i=1,..,n. A widespread approach to estimation of such models is based on the LOESS procedure [44, 45], which is a local polynomial regression with variable bandwidth based on nearest neighbor regression.
In the present study, we prefer to focus on penalized regression smoothers based on splines [46, 47], because of their appealing theoretical properties. More precisely, the function g is represented as a linear combination of completely known basis functions, so that only the coefficients of the combination need to be estimated (typically by minimizing a least squares fitting objective). The cubic spline basis is particularly well suited because it can be shown to have good approximation theoretic properties. Such a spline is a curve made up of sections of cubic polynomials joined together so that they are continuous in value as well as first and second derivatives. The points at which the sections join are known as the knots and they are typically chosen in an evenly spaced way through the range of the observed covariate. The degree of smoothing is controlled by adding a roughness penalty to the objective function, so that a modified least squares criterion  should be minimized
where γ is the smoothing parameter. This represents the extent to which roughness is penalized and, therefore, it allows one to control the trade-off between model fit and model smoothness, providing flexibility in the presence of fast or slowly changing trends. Note that the cubic spline arises naturally from the specification of the above least squares criterion, as it can be shown to minimize it among all functions that are continuous on the range of the covariate and have absolutely continuous first derivatives.
The fundamental choice of the smoothing parameter value can be accomplished via cross-validation, which minimizes an estimate of the mean squared error in predicting a new variable. In particular, we prefer to adopt generalized rather than ordinary cross-validation because of computational gains as well as invariance properties. This allows one to choose the value of the smoothing parameter, which represents the best option from a predictive perspective.
Computing confidence intervals (both for the model parameters and for the smooth terms) and performing hypothesis testing require quantification of the uncertainty of the estimators. This can be accomplished on the basis of a frequentist approach to inference—that is, the classical approach based on the repeated sampling principle—as the estimators can be shown to be asymptotically unbiased and normally distributed. Unfortunately, it is well known that, when smoothing parameters have been estimated, the p values are typically lower than they should be, meaning that the test rejects the null hypothesis too readily.
This is why we preferred a Bayesian approach, which results in a posterior distribution for the parameter estimators and for quantities derived from them. (Note that, for non-normal data, posterior normality of the estimators is again an approximation justified by large sample results.) Therefore, p values for smooth terms have been based on a test statistic motivated by an analysis of frequentist properties of Bayesian confidence intervals , which show better frequentist performance than the alternative strictly frequentist approximation.
To perform inference (i.e., estimation and prediction) on spline regression smoothers, it is convenient to exploit the fact that they can be viewed as particular generalized additive models (GAMs) [49, 50]. Indeed, such smoothers are GAMs with only 1 covariate and link function between mean response variable and linear predictor equal to identity, provided that such a response variable is generated from a dispersion-exponential family  (in the present study, we assumed the normal one). Therefore, the implementation has been performed by means of the mgcv package of R software , which is devoted to GAM estimation and prediction.
Finally, some normality tests (i.e., Anderson–Darling, Lilliefors, and Shapiro–Wilk) on the response variable confirmed the reliability of the inferential results even for moderate sample sizes in the majority of cases (normality hypothesis acceptance rates higher than 95% for bears and higher than 50% for seals).
A synthesis of recent data (for the 5-yr period 2006–2010) available for different environmental matrices characterizing the Arctic food chain is shown in Table 1. For fish, data were not readily available for some chemicals, while for seals and bears the information was complete. This complete dataset is given in the Supplemental Data (Tables S10–S12).
For large chemical classes, like PBDEs and PCBs, the total sum reported in the different studies is often calculated on a nonconstant number of congeners. However, selected congeners (e.g., Σ4 PBDE and Σ10 PCB) represent a very high percentage of the total sum of analyzed congeners. Therefore, this selection represents a more reliable and comparable figure. Particularly relevant, especially for biota, are PBDE47 and PCB153 . The percentage of PBDE47 is approximately 47% and 70% of the total PBDE concentration in fish and mammals, respectively. For PCB153, the values of approximately 10% to 15% of the total for fish, 20% for seal, and 42% for bear are in good agreement with those reported by Muir et al. . Finally, it is interesting to note that the calculated concentrations in bear milk were in good agreement with the few data available in the literature .
Concentration values were used to calculate biomagnifications factors (BMFs) in marine mammals (Table 1). Some apparently surprising data, such as the very high value for aldrin of the bear/seal BMF, may be justified by the relatively few data available for 1 or both animals, for some chemicals, in the period considered. More relevant is the low DDT value for bear/seal BMF, indicating no biomagnification at the top level of the food chain. In this case, the value is supported by a huge amount of data from several studies in different Arctic locations indicating that, for all the cases, the concentrations in seals were higher than in bears. Possible explanations are discussed in the section Toxicity data and mixture risk characterization.
Toxicity data and mixture risk characterization
The ADIs for humans, assumed as a reference value for seal and bear, were selected from the literature. The selected ADIs and the hazard quotients for seal and bear are given in Table 2. Bear datasets (adults eating seals and cubs drinking milk) are complete, whereas for seals, some experimental data on fish was lacking (Table 1). Therefore, estimated fish concentrations were used as approximated exposure data (see details of the estimation procedures in Supplemental Data, Section 7). All approximated values were calculated using worst-case approaches, so the results may be overestimated.
The following observations can be made from the results shown in Table 2.
For seals, the hazard index is higher than the threshold of 1. It must be considered that the assessment is based on the very conservative ADI developed for human health and that all the assumptions used to cover the uncertainties were worst-case assumptions. Nevertheless, the hazard index value and the level of uncertainty indicate at least a situation of potential concern.
For bears, the hazard index is extremely high—2 and 3 orders of magnitude higher than the threshold of 1 for adults and cubs, respectively. Even considering the very conservative approach used, the probability of toxicological risk for bears is high. In particular, because most POPs may have endocrine-disrupting effects, the growth and development of bear offspring may be endangered.
From the fingerprint of chemical risk in adult bears and cubs (Figures 1 and 2), some relevant observations can be made on individual chemicals. It is worth noting that, besides the most important complex groups of POPs (PCBs, polychlorinated dibenzodioxins/dibenzofurans [PCDD/Fs]) some individual chemicals (or small groups), such as chlordanes, aldrin, and dieldrin, also reach a very high hazard quotient level. In particular, for bear cubs some hazard quotient values are close to or even higher than 100. Therefore, even the risk from individual chemicals is high. The contribution of DDTs and its metabolites to the total potency of the mixture is relatively low. This is mainly because of the higher ADI of DDTs and, for bear cubs, because of the low BMF between seals and bears. For seals (Supplemental Data, Figure S2) the highest hazard quotients correspond to PCDD/Fs. Only PCDD shows an individual hazard quotient higher than 1.
The most harmful individual chemical for bear cubs is PFOS, which has a substantially lower risk in adult bears. This is because of the very high BMF of this chemical from seal to bear (BMF = 34). To confirm the reliability of this value, data from the same geographic area, for the same period, and reported in the same study  show a difference of 2 orders of magnitude between seal and bear concentrations. The reasons for this particular pattern should be further investigated.
Temporal trend of risk
The dataset presented in Table 1 was enlarged by including all data available in the literature, covering the period lasting from the late 1960s to 2011. The criteria for the selection of data and their elaboration are the same as reported before. The complete dataset is given in the Supplemental Data (Tables S10–S12).
Sufficient data are available for many chemicals to reliably characterize the temporal trends in seals and bears. In some cases, reliable data allow the trend to be evaluated for more than 40 yr.
The complete temporal trends, together with the statistical data of the models obtained and an estimation of the half-lives of the different chemicals, are given in the Supplemental Data (Section 8, Figure S3, and Table S9).
Temporal trends for legacy and emerging POPs are described in several recent reports and studies in the literature, either for the Arctic as a whole [7, 10, 16-19, 56-59] or for specific areas, such as Greenland or the Canadian Arctic [20, 21, 60-63]. The results obtained are in good agreement with the literature data showing a general decrease in legacy POPs beginning in the 1980s. In contrast, PBDEs and perfluorinated compounds (particularly PFOS) continuously increased, and only very recent data (after 2005) seem to indicate a decrease [64-70].
On the basis of the trend models obtained, a temporal trend of the risk for bear cubs has been calculated, for chemical mix tures for which enough data were available to calculate a reliable temporal trend model. Although it is not complete, it represents approximately 80% of the total risk calculated for the period 2006 to 2010. Among the excluded chemicals, only PCDD/Fs represent a relevant percentage of the risk (approximately 16%). The trend of the risk for bear cubs, from 1985 to 2010, is shown in Figure 3, together with the percentages of the 5 chemicals (or groups of chemicals) most relevant in the composition of the mixture. The following comments on these results may be made. First, the total risk is slowly decreasing, with a reduction in hazard index of approximately 30% over 25 yr, but the composition of the mixture has changed substantially. Second, the percentage of risk determined by some legacy POPs (chlordanes, dieldrin) has constantly decreased, whereas for PCBs the decrease is delayed by approximately 10 yr; this is probably because of greater difficulties in controlling PCB emissions, compared with pesticidal POPs. Moreover, the temporal trend was calculated for PCB153 only, because of its higher biomagnification capacity . This property could increase its persistence in biota and makes this congener particularly relevant for risk assessment. Third, and in contrast, the percentage of risk determined by PFOS is strongly increasing; in 2010, it was approximately 50% of the total hazard index.
It is important to stress that our approach is approximate and is based on a number of assumptions, mainly as part of the estimation of mixture response. Applying the CA model to hazard quotients calculated using estimated ADIs for humans is an extreme simplification, particularly for a complex class of chemicals such as POPs, with extremely different toxicological modes of action. However, the method represents a possibility for estimating an approximated response when incomplete information is available.
Another source of uncertainty and approximation is the possible variability in POP content as a function of age and sex. This could be particularly relevant in relation to the transfer of POPs to offspring during reproduction and lactation. Aguilar and Borrell  estimated the reproductive transfer of organochlorine pollutants in the offspring of fin whales (Balaenoptera physalis). They observed a decrease in DDTs and PCBs in adult females, probably because of excretion during reproduction and lactation. Thus they calculated that the total intake to the offspring through lactation was approximately 1 g of PCBs and 1.5 g of DDTs for primiparous females, decreasing to 0.2 g of PCBs and 0.3 g of DDTs for old females. This kind of detail was not the objective of the present study. However, for some chemicals (e.g., chlordane, DDTs, PCBs) and some temporal intervals, it was possible to find separate data for male, female, and subadult bears (see the database in the Supplemental Data). The observed differences were, in any case, lower than a factor of 2, which was assumed to be irrelevant for an approximated assessment.
Aguilar and Borrell  concluded that, considering the size of the fin whale, the toxicological risk was low. Indeed, the body weight of a newborn fin whale is approximately 2 metric tons, the lactation time-span is at least 6 mo, and the weight at weaning is more than 10 metric tons . For a polar bear cub in the first 40 wk of life (with body weight increasing from ∼0.6 kg to ∼80 kg), a comparable calculation indicates a total intake of approximately 4 g of PCBs with an average body weight of approximately 38 kg for the 40-wk interval (see details of the calculation in Supplemental Data, Section 7, Table S8). This shows the enormous difference in risk between whales and polar bears, mainly because of their different positions in the food web.
The present study indicates a very high potential for toxic effects at least at the top levels of the Arctic food web, particularly for a top predator such as the polar bear and for its offspring. Moreover, it must be noted that, in addition to the effect of the mixture, for 4 chemicals (or group of chemicals) the risk for the most endangered organisms of the food web (bear cubs) is 2 orders of magnitude higher than 1, which is assumed as the threshold of risk (Table 2).
We are aware of the approximation and uncertainty of the assessment as a result of a number of factors (geographic variability, sex, age and diet differences, etc.). Nevertheless, a risk that is orders of magnitude higher than a safety threshold should prevail over these uncertainties. Therefore, the results represent a serious warning about the risk from POPs for the Arctic ecosystem, particularly in terms of endocrine disruption. Endocrine disorders in polar bears have been described by Wiig et al. , who observed cases of pseudo-hermaphroditism in females sampled at Svalbard in 1996, possibly as a result of POP contamination.
Extensive reviews [74, 75] highlight the occurrence of several health effects in Arctic top predators, particularly polar bears, that may be attributed to POPs. Some examples are reported in Table 3, but the list is far from exhaustive. Considering the estimated risk of POP mixtures, particularly for bear cubs, such evidence of deleterious health effects is not surprising.
|Observed effect||Chemicals involved||Ref.|
|Activity of cytochrome P-450 and associated enzymes||Several groups of POPs (e.g., PCBs, PCDD/Fs)||[87, 88]|
|Alteration to the endocrine system||Several groups of POPs (e.g., PCBs, PBDEs, HCB, HCHs, and DDTs)||[89-96]|
|Malfunctioning of reproductive organs||Several groups of POPs||[97-101]|
|Liver alterations||Several groups of POPs||[102, 103]|
|Neurological damages||Several groups of POPs|||
Differences among individual chemicals
Individual chemicals play different roles in the composition and fingerprint of the mixtures, both as concentrations and as toxic effects on different organisms of the food chain (Tables 1 and 2, and Figures 1 and 2; Supplemental Data, Tables S5–S7 and Figure S2). It can be seen that a large percentage of the total potency of the mixture is composed of a few chemicals. In particular, for adult bears, approximately 90% of the total mixture potency is explained by 5 chemicals or chemical classes: PCDD > PCDF > toxaphene > dieldrin > chlordanes. In contrast, for bear cubs, approximately 90% of the total mixture potency is explained by 5 other chemicals or chemical classes: PFOS > PCB > chlordanes > PCDD > dieldrin. In particular, PFOS alone is responsible for approximately 50% of the toxic potency. Indeed, PFOS concentrations are relatively low in seals and dramatically increase in bears and, as a consequence, in bear milk. Moreover, PFOS is considered very toxic for mammals, with a very low ADI . However, the sequence must be considered with care, as the exposure to some of the chemicals was calculated with an approximated procedure.
Among chlorinated insecticides, DDT plays a relatively lesser role in the mixture, in spite of the extremely high global emissions, estimated to be between 1.2 million metric tons and 4.1 million metric tons . This is partly because of the relatively low toxicity for mammals in comparison with other chlorinated insecticides (ADI 2 orders of magnitude higher than those for aldrin, dieldrin, and chlordane). Moreover, the relatively low concentrations in adult bears, substantially lower than in seals, indicates that no biomagnification has occurred in bears. This justifies the low hazard quotient for bear cubs.
The low values of DDTs in polar bear could be explained by a capacity of bears to metabolize DDTs more efficiently that other POPs. This capacity should be typical for bears and not for seals. This kind of metabolic capability has been observed by Norstrom et al.  and Letcher et al. . Polischuck et al.  observed that that polar bears are able to metabolize DDTs more readily than most organochlorine compounds and that they appear to be unique in the animal world in their capability to metabolize p,p′- dichlorodiphenyldichloroethylene. Letcher et al.  measured BMFs between seals and bears of 0.6 and 15.1 for DDTs and PCBs, respectively. These values are in reasonable agreement with the 0.5 and 24 values, respectively, calculated in our dataset (Table 1).
Risk trend and effectiveness of control measures
For some decades, the global emissions of legacy POPs have decreased dramatically because of several international agreements such as the Basel Convention on the Control of Transboundary Movements of Hazardous Wastes and Their Disposal signed in 1989 ; the Aarus Protocol on Long Range Transboundary Air Pollution (LRTAP) of POPs, signed in 1998 ; and the Stockholm Convention, which was signed in 2001 and has been in force since 2004 . In particular, the last 2 agreements are based on an original list of chemicals to be controlled. This list is periodically amended, and new chemicals are added. The list of chemicals controlled to date by the Aarus Protocol and the Stockholm Convention is given in the Supplemental Data (Table S1).
To evaluate the effectiveness of the Stockholm Convention, a complex strategy has been developed based both on a global monitoring plan, national reports, and the measures taken to implement the Convention, and on other initiatives under the control of the Stockholm Convention Conference of the Parties . A full evaluation of the effectiveness of the Convention is planned for 2017.
The effectiveness of the control measures is supported by the evidence of a decreasing trend in the Arctic ecosystem beginning in the late 1980s. However, the decrease is very slow. For individual compounds, half-lives in the order of years to decades are reported in the literature . Comparable values have been calculated with the models developed in the present study (half-lives ranging from 4 yr to 96 yr for the different chemicals; see Supplemental Data, Table S9).
A relationship between the trends in biota (seals and bears) and the trends in the surrounding environment (water) is not easy to find because water data are scattered, and comparability among data from different surveys is often poor. However, Choi and Wania  have theoretically demonstrated the possibility of a very slow environmental reversibility for some classes of POPs in conditions of relatively long air and water half-lives, which likely to occur in polar environments. This reversibility may last far beyond the middle of this century.
Moreover, the concern is growing for other chemicals that are already listed as POPs under the Stockholm Convention since 2009, but for which a decrease began recently or has not yet occurred, such as PBDEs and, particularly, PFOS. The concentrations of PFOS in polar bears are surprisingly high (2 orders of magnitude more than in seals) and are a source of growing concern. At present, for polar bear cubs, PFOS represents the most harmful chemical in the mixture (Figure 2).
The need to improve our knowledge of temporal trends of legacy and new POPs in biotic and abiotic matrices of the Arctic is highlighted by Muir and de Wit , who also consider possible effects of climate change on ecological characteristics of the system and on fate patterns of the chemicals.
Climate change may alter contaminant pathways and concentration patterns as well as the vulnerability of fragile ecosystems; the processes involved are far from completely understood or predictable .
The number of POPs circulating to date in the global environment is quite controversial. Brown and Wania  examined a dataset of more than 100 000 chemicals and identified 120 high–production-volume chemicals with properties suggesting they are potential Arctic contaminants. Their list includes several current-use pesticides as well as halogenated chemicals that have not been measured previously in the Arctic. Indeed, Morris et al.  recently reported evidence of the presence of current-use pesticides in the Arctic food chain. Scheringer et al.  examined a dataset of more than 90 000 compounds and found 510 chemicals exceeding the criteria to be considered potential POPs. Considering the uncertainty of the screening exercise, these authors concluded that at least 190 chemicals may be considered potential POPs and that several 10s of potential POPs may need to be evaluated in the future under the Stockholm Convention. Currently, the Stockholm Convention includes 23 chemicals (or congeneric groups) and 6 candidates.
Another relevant issue concerns chemicals that are unintentionally produced as byproducts of industrial processes (e.g., PCDD/Fs). For these chemicals a complete phasing out is virtually impossible, and only a reduction in emissions is realistically achievable, at least over the short term.
In the last few decades, a great number of studies and technical reports have been published on the presence and temporal trends of contaminants, particularly POPs, in the Arctic environment. Many of them focus on specific groups of chemicals, specific matrices or animal species, or specific locations. Other are more general, providing a wider picture of Arctic contamination. In a few cases, attempts have been made to assess the general toxicological impacts of global emissions of contaminants of human origin . However, even if the occurrence of adverse effects is highlighted, no attempts have been made to characterize the risk quantitatively.
The objective of the present study was not to repeat information already in the literature, but rather to use the bulk of information available to develop a quantitative characterization of the ecological risk of a mixture of high-volume POPs that is likely to occur in the Arctic environment as a whole.
The results of the present review are based on a series of assumptions and approximations that were necessary because of the lack of complete and detailed knowledge of many aspects of the process of risk characterization. In particular, the complexity of the toxicological modes of action of the chemicals considered led us to assume, as a toxicological endpoint, not a specific effect but a conservative reference value such as the ADI. For the same reason, the only option to estimate the response to the chemical mixture has been the use of the CA model. All these assumptions led to a worst-case characterization and to a possible overestimation of the actual risk. Nevertheless, the extremely high level of the risk and the very slow decrease estimated to have occurred over the last 20 yr (in spite of stringent control measures) indicate that POPs represent a serious environmental concern on the planetary level. Moreover, the observed changes in the composition of the mixture highlight the fact that more attention must be paid to emerging contaminants. The present results also provide insights for international stakeholders into the need for further implementation of mitigation measures, such as the Stockholm Convention, for legacy as well as for additional, not yet controlled, POPs, to avoid global pollution problems that could only be solved over generations.
The Supplemental Data are available on the Wiley Online Library at DOI: 10.1002/etc.3671.
The present study was supported by the United Nations Environment Programme (UNEP)/POPS/POPRC (Persistent Organic Pollutants Review Committee). The authors are grateful to T. Harner and R. Letcher (Environment Canada) for comments and suggestions.
Data are available on request from the corresponding author (firstname.lastname@example.org).