Environ. Sci. Technol. 2009, 43, 5830–5836
Comparative Assessment of the Global Fate and Transport Pathways of Long-Chain Perfluorocarboxylic Acids (PFCAs) and Perfluorocarboxylates (PFCs) Emitted from Direct Sources JAMES M. ARMITAGE,† MATTHEW MACLEOD,§ AND I A N T . C O U S I N S * ,† Department of Applied Environmental Science (ITM), Stockholm University, SE-10691 Stockholm, Sweden, and Institute for Chemical and Bioengineering, Swiss Federal Institute of Technology, ETH Zurich, HCI G129, CH-8093 Zu ¨ rich, Switzerland
Received March 11, 2009. Revised manuscript received June 1, 2009. Accepted June 2, 2009.
A global-scale multispecies mass balance model was used to simulate the long-term fate and transport of perfluorocarboxylic acids (PFCAs) with eight to thirteen carbons (C8-C13) and their conjugate bases, the perfluorocarboxylates (PFCs). The main purpose of this study was to assess the relative longrange transport (LRT) potential of each conjugate pair, collectively termed PFC(A)s, considering emissions from direct sources (i.e., manufacturing and use) only. Overall LRT potential (atmospheric + oceanic) varied as a function of chain length and depended on assumptions regarding pKa and mode of entry. Atmospheric transport makes a relatively higher contribution to overall LRT potential for PFC(A)s with longer chain length, which reflects the increasing trend in the air-water partition coefficient (KAW) of the neutral PFCA species with chain length. Model scenarios using estimated direct emissions of the C8, C9, and C11 PFC(A)s indicate that the mass fluxes to the Arctic marine environment associated with oceanic transport are in excess of mass fluxes from indirect sources (i.e., atmospheric transport of precursor substances such as fluorotelomer alcohols and subsequent degradation to PFCAs). Modeled concentrations of C8 and C9 in the abiotic environment are broadly consistent with available monitoring data in surface ocean waters. Furthermore, the modeled concentration ratios of C8 to C9 are reconcilable with the homologue pattern frequently observed in biota, assuming a positive correlation between bioaccumulation potential and chain length. Modeled concentration ratios of C11 to C10 are more difficult to reconcile with monitoring data in both source and remote regions. Our model results for C11 and C10 therefore imply that either (i) indirect sources are dominant or (ii) estimates of direct emission are not accurate for these homologues.
* Corresponding author e-mail:
[email protected]; phone: +46 (0)8 16 4012. † Stockholm University. § Swiss Federal Institute of Technology. 5830
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 15, 2009
Introduction Long-chain perfluorocarboxylic acids (PFCAs) and their conjugate bases, the perfluorocarboxylates (PFCs), are persistent pollutants that have been reported in wildlife at detectable concentrations in both industrialized and remote locations (1–8). The environmental loadings of these compounds, collectively referred to here as PFC(A)s, are due to direct emissions during manufacture and use (9) and degradation of precursor compounds such as fluorotelomer alcohols (FTOHs) and perfluorooctanesulfonyl fluoride (POSF)-based chemicals in the environment (10–12). According to Prevedouros et al. (9), emissions during the manufacture and subsequent use of commercial ammonium perfluorooctanoate (APFO) and perfluorononanoate (APFN) as processing aids in the production of fluoropolymers represent the most important direct sources of PFC(A)s to the environment. While commercial APFO and APFN products predominantly contain the eight (C8) and nine (C9) carbon homologue respectively, other homologues (C4-C13) are also present as impurities. For example, a representative APFN product was reported to contain approximately 74% C9, 20% C11, 5% C13, and trace amounts of C8, C10, and C12 (9). Based on smog chamber experiments, degradation of the 8:2, 10:2, and 12:2 FTOHs in the atmosphere is expected to yield PFCAs with a range of chain lengths (including C8-C13) along with other degradation products (10). Therefore both direct and indirect sources contribute to the concentrations of the major homologues reported in biological samples. It is frequently, although not universally, observed that PFC(A)s with an odd number of carbons (e.g., C9, C11) are at higher concentrations in biota than the next shortest evencarbon homologues (e.g., C8, C10) (1–8). These observations have been interpreted as supporting the hypothesis that degradation of FTOHs is the main source of exposure, based on studies showing (i) that the molar yield of each substance in an “odd/even” homologue pair is similar via atmospheric degradation (10) and (ii) that bioaccumulation potential and homologue chain length are positively correlated (13, 14). However, the same pattern of “odd/even” homologue ratios might be expected from direct sources (9) since emissions of C11 > C10, C13 > C12, and, although emissions of C8 > C9, a higher bioaccumulation potential of C9 may be amplified through food webs such that concentrations of this homologue exceed C8 in biota. Therefore, the homologue ratios observed in biota do not unequivocally support the hypothesis that degradation of FTOHs is the dominant source of PFC(A) exposure for biota. Understanding the long-range transport (LRT) potential of PFC(A)s is important from both a scientific and regulatory perspective because this information will help to elucidate the potential exposure attributable to different sources. Recent modeling studies have concluded that (i) oceanic transport of PFCs is an efficient transport pathway from source to remote regions in comparison to atmospheric transport and degradation of precursors (15, 16) and (ii) that the oceanic mass flux of the eight-carbon homologue (C8), i.e. PFO(A), into the Arctic attributable to direct emissions is well in excess of indirect sources (15–17). Atmospheric transport has previously been discounted as a likely LRT mechanism for long-chain PFC(A)s, in part because they were assumed to exist almost exclusively as anions (PFCs) in the environment. However, the pKa of PFOA (C8) was recently estimated to be approximately 3.8 at infinite dilution using potentiometric titration (18). While there is no consensus on 10.1021/es900753y CCC: $40.75
2009 American Chemical Society
Published on Web 06/17/2009
the true value of pKa for PFCAs at environmentally relevant concentrations (19–21), overall LRT potential of C8 was enhanced by atmospheric transport when pKa of 3.5 was assumed in model simulations (22). These findings raise the possibility that direct sources of PFC(A)s contribute to the contamination of both terrestrial and marine environments through atmospheric LRT. The main purpose of the study described here was to assess the fate and relative LRT potential of C8-C13 PFC(A)s emitted from direct sources. For reasons discussed earlier, we focus on the relative behavior of the following homologue pairs (i) C8 and C9, (ii) C10 and C11, and (iii) C12 and C13, and the extent to which differences in physical-chemical properties result in fractionation (i.e., preferential transport of one homologue over the other) on a global scale. In particular, we test the hypothesis that enhanced sorption to solids with increasing chain length (23) limits LRT potential. Two types of model results are reported. First, we use an indicator of overall LRT that is independent of emission rate to assess the transport potential of each PFC(A). Then, using emission estimates derived from available sources (9, 24), we attempt to account for concentrations of PFC(A)s measured in the abiotic environment and relate the modeled concentrations to observed homologue patterns reported in biota in source and remote regions.
Methods Model Description. All simulations were conducted using the BETR Global model presented in Armitage et al. (22), which is adapted from the version first introduced by MacLeod et al. (25). Briefly, BETR Global is a fugacity-based mass balance model that describes the global environment as 288 regions based on a 15 × 15° grid. Weak acids can be simulated in our version of the model by specifying the acid dissociation constant (pKa), environmental pH, and physicalchemical property values for the acid and conjugate base. The distribution coefficient approach (26) is then used to describe the overall partitioning of the acid-base conjugate pair. The model includes representations of intermittent precipitation (27), deep water formation in the northern North Atlantic, and a surface ocean flow regime derived from a global drift buoy array (29). Physical-Chemical Properties. The fate of PFC(A)s in the environment is determined by their partitioning properties and speciation. Key environmental partitioning properties typically required as model input include the air-water partition coefficient (KAW), organic carbon-water partition coefficient (KOC), and aerosol-air partition coefficient (KQA). The latter two are often estimated from the octanol-water (KOW) and octanol-air (KOA) partition coefficient, respectively. Empirically derived data on partitioning and speciation for PFC(A)s remain sparse and KOW and KOA values may only be determinable for the acids (i.e., the PFCAs). Here we use what data are available in the literature and conduct model calculations for a range of possible property combinations to reflect these uncertainties. Arp et al. (30) estimated partitioning properties for C6C11 PFCAs and other highly fluorinated compounds using property estimation software and compared these values to measurements where possible. For this study, we selected the partition coefficients estimated by COSMOtherm as default values and repeated some simulations using the values estimated by SPARC. The log KAW and log KOW values for C6-C13 are reported in the Supporting Information (SI), Table S1 with property values for C12 and C13 estimated by linear extrapolation of the data in ref 30. KOA was estimated as KOW/KAW while KOC values were estimated from KOW. The anions (i.e., PFCs) were assumed to have negligible KAW whereas KOC was based on empirical partitioning data (23). The selected values are presented in Table S2. See the SI
(Section S2) for an expanded discussion of the assumptions used to estimate sorption to environmental solids. Speciation between the acid and conjugate base form of PFC(A)s and hence the distribution coefficients are calculated from pKa and environmental pH. To reflect uncertainty about the appropriate pKa values for PFCAs we consider two values here, 0 and 3.8. These values represent the range that has been reported in the literature for PFOA (18, 21). Following the arguments in Burns et al. (18) and data in Goss (21), we assume that the pKa values of the C9-C13 acids are similar to that of C8 (i.e., negligible trend in pKa with chain length for acids with g8 carbons). Also, no distinction has been made between linear and branched isomers. Data used to parametrize compartmental pH are presented in the SI (Section S3). Emission Estimation Methodology. We reported emission estimates for the C8 homologue earlier (22) and have followed the same approach here for C9-C13. Briefly, global production, use, and minimum and maximum emission rates for commercial APFO and APFN were estimated in Prevedouros et al. (9) for the periods 1951-2004 and 1975-2004, respectively. Substantial emission reductions were assumed for the period 2005-2010 to reflect industry commitments. Total APFO/APFN emissions were distributed among model regions with APFO or APFN manufacturing facilities and model regions with fluoropolymer (FP) production sites using APFO or APFN as processing aids (9, 24), according to production capacity. This approach means that emissions per unit production are assumed to be identical at all sites. Emissions of each homologue (C8-C13) were then calculated by multiplying APFO/APFN emissions by the estimated fraction of each homologue in each commercial product (9). The homologue profiles of APFO and APFN used in our calculations are presented in the SI along with the estimated historic emissions of APFO and APFN (Section S4). Mode of Entry. Emissions from FP manufacturing facilities were assumed to be released to air (23%), freshwater (65%), and land (12%) whereas emissions from commercial APFO and APFN manufacturing were assumed to be released to freshwater (95%) with minor releases to air (5%) (9). Following ref 22, we consider two scenarios regarding mode of entry. In the first scenario, emissions to air are redirected to land in the model region of origin, representing the possibility of rapid aggregation and local deposition of stack emissions. In the second scenario, emissions to air are distributed into atmospheric phases (gas/aerosol/rain) according to physical-chemical properties and environmental conditions. Assessment of Overall LRT Potential. Wania (31) introduced Arctic Contamination Potential (eACP) as a metric of contaminant transport from source regions and accumulation in the Arctic that is independent of emission rate. Here we utilize a related metric, ACPBETR, calculated as the total mass of contaminant in model regions between 60 and 90° N (M60-90) at the end of the simulation (2010) divided by the totalmassemitted(Memitted)duringthesimulation(1950-2010), i.e., ACPBETR )
M60-90 MEMITTED
(1)
This metric is not directly comparable to eACP but it conveys similar information about transport and accumulation in the Arctic. Note that the relative contributions from atmospheric and oceanic LRT are not distinguishable from this metric alone.
Results and Discussion Global Emission Estimates. Total emissions of C8-C13 (1951-2010) from direct sources are presented for each major source area (North America, Europe/Russia, Asia) in Figure VOL. 43, NO. 15, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
5831
FIGURE 1. Estimated total emissions (1951-2010, t) of C8-C13 from North American (black), European/Russian (gray), and Asian (white) sources. Maximum emissions represented by height of bars, minimum estimates indicated by the horizontal lines. 1. C8 emissions are distributed approximately 37% to North America, 26% to Europe/Russia, and 37% to Asia, whereas C9-C13 are distributed approximately 75% to North America, 10% to Europe/Russia, and 15% to Asia. The range of estimated emissions of C8 based on the minimum and maximum scenarios is 2600-5050 metric tons (t) and emissions of C9-C13 are 450-1730, 2.4-9.2, 120-460, 0.6-2.3, and 30-115 t, respectively. The ratios of total emissions of C8:C9, C11:C10, C13:C12 are thus 3-6:1, 50:1, and 50:1. As implied by the geographical distribution of emissions, the ratio of C8:C9 emissions in North America is lower, ranging from 1.5-3:1 and higher in Europe/Russia and Asia (7-15:1). Production and use of commercial APFN dominates the estimated emissions of C9-C13 whereas emissions of C8 are dominated by production and use of commercial APFO. Commercial APFN is manufactured using a process that yields exclusively linear isomers whereas the electrochemical fluorination (ECF) process used historically to manufacture APFO yields ∼20% branched isomers. Hence, isomer profiles related to direct sources of C9-C13 differ substantially from C8 in terms of the proportion of branched isomers. Given that atmospheric degradation of the FTOHs produced in substantial quantities is expected to yield 100% linear isomers for all homologues (9), linear:branched isomer ratios in abiotic samples could potentially be used to distinguish between direct and indirect sources of C8. This approach may not be possible for C9-C13 however. Trends in Overall LRT Potential with Homologue Chain Length. Modeled ACPBETR values for C8-C13 calculated using the default partition coefficients (COSMOtherm) and the two pKa values are shown in Figure 2. Regardless of assumptions about mode of entry, the ACPBETR of all homologues is higher when pKa ) 3.8 because of a higher rate of volatilization of PFCAs from soil and water to the atmosphere as well as decreased transfer from the atmosphere to surface compartments (i.e., higher KAW leads to lower wet deposition). For a given pKa, ACPBETR values are higher when emissions to air are modeled, again demonstrating the influence of atmospheric LRT. When pKa is assumed to be 0, the overall LRT potential declines with increasing chain length regardless of mode of entry. This trend reflects the increasing importance of irreversible physical sinks in the terrestrial environment, i.e., vertical transport of PFC(A)s associated with solids to inaccessible depths in sediment and soils, with higher KOC values. Because the longer chain length homologues have higher affinity for solid phases under our model assumptions, they are more efficiently removed by these loss processes than the lower chain length homologues. In contrast, particulate settling to deep ocean has little influence on overall fate because the bulk of PFC(A) mass (g99%) in the water column is present as PFCs and is in the dissolved phase. Hence, global transport of PFC(A)s is not limited by processes 5832
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 15, 2009
FIGURE 2. ACPBETR values calculated using model output based on default physical-chemical property values assuming pKa ) 3.8 (white) and pKa ) 0 (black) with emissions to air (circles) and no emissions to air (squares). in the ocean but instead by processes in the terrestrial environment. Our model results thus indicate that the fate of higher chain length PFC(A)s in sediment and soil layers in source regions should be investigated further. When pKa is assumed to be 3.8, the trend in ACPBETR with homologue chain length depends on the mode of entry. For both emission scenarios, the increased presence of neutral PFCAs enhances volatilization from surfaces, counteracting the terrestrial loss processes to some extent. With emissions redirected to land, ACPBETR values for C8 to C10 are similar, reflecting a balance between increasing volatility of the neutral PFCAs with chain length and increasing removal by physical terrestrial sinks due to enhanced sorption to solids in sediments and soils. For C11-C13, the influence of higher KOC becomes dominant, causing a decline in overall LRT potential with chain length. In these simulations, increasing losses related to higher affinity for organic carbon with chain length are less effective at dampening the influence of atmospheric LRT, which is enhanced by the mode of entry in addition to increased volatilization from surfaces and decreased wet deposition. The variability in the calculated ACPBETR values as a result of different assumptions about pKa and mode of entry is substantially higher for the longer chain length homologues. For example, the calculated ACPBETR for C8 range from 4.0 to 5.2% considering all scenarios, whereas the ACPBETR for C13 range from 0.1 to 2.9%. ACPBETR values of longer chain length homologues are more sensitive to mode of entry because, as discussed previously, they are much less mobile
FIGURE 3. Modeled C8:C9 concentration ratios in surface ocean waters in 2005 assuming the maximum emission scenarios, pKa ) 0 and emissions to air. from the soil compartment than the shorter chain homologues. Model assumptions that tend to decrease the proportion of PFC(A) mass in the soil compartment are therefore more influential. ACPBETR values of longer chain homologues are more sensitive to assumptions about pKa for a related reason; volatilization of the neutral species is the most important pathway for mobilization of longer chain homologues from soil under our model assumptions. Model output using SPARC physical-chemical properties is presented in the SI (Figure S3). ACPBETR values are higher than default model output based on COSMOtherm properties (Table S8) despite higher KOC values for the neutral form. This model output is explained by the counteracting influence of the higher KAW values estimated by SPARC (i.e., atmospheric LRT is sufficiently enhanced to outweigh increased association with solids). As discussed in ref 22, the relationship between KQA and atmospheric LRT also depends on the assumed pKa. For pKa ) 0, atmospheric LRT increases with increasing KQA (i.e., contaminant mass associated with aerosol) because rain scavenging of the gas phase species is more efficient than scavenging of aerosol-associated PFC(A). For pKa ) 3.8, rain scavenging of gas-phase PFCAs is less efficient and atmospheric LRT decreases with increasing KQA because wet and dry deposition of aerosol-associated PFC(A) is more efficient than total deposition of gas-phase PFCA. Despite the large uncertainty in the true values, the pKa’s of all homologues are expected to be similar (18, 21), therefore we do not expect assumptions regarding aerosol-air partitioning to have a large impact on the modeled fractionation between homologue pairs. However, for quantitative comparisons to monitoring data (e.g., deposition fluxes, concentrations in precipitation), uncertainties in both pKa and KQA need to be reduced. This is particularly true for PFC(A)s with greater than 8 carbons as overall LRT becomes increasingly sensitive to model assumptions determining atmospheric transport. Fractionation and Modeled Concentration Ratios in Surface Ocean Waters. Modeled concentrations of C8-C13 in surface ocean waters reflect the geographical and temporal distribution of emission estimates, circulation patterns of surface ocean water, and overall LRT potential. The potential for fractionation between homologue pairs can be assessed by the ratio of the respective ACPBETR values for each model scenario. For C9:C8 and C11:C10, fractionation due to abiotic fate and transport processes is relatively low regardless of model assumptions (i.e., a factor of 2 or less). For C13:C12, LRT of C13 is approximately five times less efficient assuming
pKa ) 0 and no emissions to air; in all other scenarios, fractionation is less than a factor of 2. The fractionation between homologue pairs using SPARC physical-chemical properties spans a similar range (see Table S8) indicating that the relative behavior of the homologue pairs is not highly sensitive to our assumptions made regarding physicalchemical property values. The global pattern of modeled C8:C9 concentration ratios assuming maximum emission scenarios is shown in Figure 3 for the year 2005, assuming pKa ) 0 and emissions to air. As illustrated in this figure, there are marked regional differences which are primarily due to the spatial distribution of emission estimates. For example, the modeled C8:C9 ratios in the mid to north Atlantic Ocean are e3, whereas ratios in most regions of the Pacific Ocean are 6-10 or higher. These ratios reflect emissions from North American and Asian sources, respectively (see Figure 1). The homologue ratio signal from Europe can be seen to dilute the North American signal in the eastern Arctic Ocean while both atmospheric inputs and some limited oceanic exchange between Arctic basins influences the Asian signal flowing into the Arctic Ocean via the Bering Strait. This Pacific inflow signal is more strongly evident in the Western Arctic regions (regions 1-5, 25-31) assuming no emissions to air (see SI, Figure S3) but more strongly diluted assuming pKa ) 3.8 (due to enhanced atmospheric LRT and hence dispersion). For C10-C13, the ratio of emissions between the homologue pairs is identical (50:1) in all source regions and time periods. By 2005, modeled surface water concentrations of C11 in nonsource regions are approximately 25-45 times higher than C10 while modeled concentrations of C13 are approximately 10-35 times higher than C12 assuming the default parameter values in 2005. Comparison to Monitoring Data. Modeled concentrations of C8-C13 in surface ocean waters are presented in the SI (Figures S4-S9). Assuming maximum emissions, modeled concentrations of C11 and C13 are
