An (Eco)Toxicity Life Cycle Impact Assessment Framework for Per- And Polyfluoroalkyl Substances

A framework for characterizing per- and polyfluoroalkyl substances (PFASs) in life cycle impact assessment (LCIA) is proposed. Thousands of PFASs are used worldwide, with special properties imparted by the fluorinated alkyl chain. Our framework makes it possible to characterize a large part of the family of PFASs by introducing transformation fractions that translate emissions of primary emitted PFASs into the highly persistent terminal degradation products: the perfluoroalkyl acids (PFAAs). Using a PFAA-adapted characterization model, human toxicity as well as marine and freshwater aquatic ecotoxicity characterization factors are calculated for three PFAAs, namely perfluorooctanoic acid (PFOA) perfluorohexanoic acid (PFHxA) and perfluorobutanesulfonic acid (PFBS). The model is evaluated to adequately capture long-term fate, where PFAAs are predicted to accumulate in open oceans. The characterization factors of the three PFAAs are ranked among the top 5% for marine ecotoxicity, when compared to 3104 chemicals in the existing USEtox results databases. Uncertainty analysis indicates potential for equally high ranks for human health impacts. Data availability constitutes an important limitation creating uncertainties. Even so, a life cycle assessment (LCA) case study illustrates practical application of our proposed framework, demonstrating that even low emissions of PFASs can have large effects on LCA results.


Term Explanation
Bill of material (BOM) List of product content. Category endpoint An attribute or aspect of the natural environment, human health or resources which is threatened by an environmental issue of concern. Characterization In LCA, impact characterization is the conversion of LCI data (flows of emissions or resources) to numerical indicators of their impacts on human health, the environment or natural resources. Characterization factors (CFs) Generally: factors for converting LCI results into LCIA indicators. Here we focus on comparative factors describing the potential impacts associated with a chemical emission into the environment on e.g. human toxicity or ecotoxicity. Cradle-to-gate A product system including life cycle stages to manufacturing gate, i.e. excluding product use and endof-life. Cradle-to-grave A product system including all life cycle stages. Effect factors (EFs) For ecotoxicity, EFs describe the PAF of aquatic species integrated over exposed water volume and time per kg bioavailable in an exposure compartment [PAF m 3 /kg bioavailable], and for human toxicity EFs describe the cumulative population risk expressed as potential disease cases per kg intake [cases/kg intake]. End-of-life End of product use, i.e. waste handling processes such as reuse, energy recovery or landfilling. Exposure factors (XFs) Exposure factors represent for human toxicity the chemical mass taken in by the human population per day per unit mass in a given environmental compartment [kg intake/d per kg in compartment] and for ecotoxicity the chemical mass fraction in a given compartment that is bioavailable [kg bioavailable/kg in compartment]. Fate factors (FFs) FFs describe chemical partitioning, dispersion, degradation and transport in and between the various environmental compartments and are expressed in increase of chemical mass in an environmental compartment per kg emitted per day into the same or any other compartment [kg in compartment per kg emitted/d], interpreted as the chemical residence time in a given compartment for an emission into the same compartment.

Functional unit
The functional unit is a quantitative representation of the function of the product under study, e.g. m 2 ×year for flooring. Impact category A class of issues of concern to which life cycle inventory results can be assigned e.g. climate change. Impact score (IS) A characterized LCA result, e.g. the product of an emission (emitted mass per functional unit) and a CF (impacts per emission unit). Indicator A quantitative representation of aggregated contributions to an impact category, e.g. as radiative forcing (for global warming) or toxic units (for ecotoxicity). Intake fractions (iFs) Human population intake via inhalation or ingestion per unit mass emitted [kg intake/kg emitted]. Life cycle assessment (LCA) LCA is an ISO-standardized decision-support tool for quantification of potential environmental and human health impacts of goods or services along their entire life cycles. Its four phases include setting of goal and scope, LCI analysis, LCIA, and interpretation. In this paper, environmental LCA is in focus (i.e. environmental and human health considerations, and not social-LCA or life cycle costing) and LCA refers to environmental-LCA. Life cycle impact assessment (LCIA) In the LCIA phase, LCI results are classified according to the potential effects of individual flows and are multiplied by substance-specific impact CFs to estimate aggregated burden indicators.

Life cycle inventory (LCI) analysis
The LCI phase requires the generation of a quantitative list of resources used and emissions released by the studied product or service life cycle. Life cycle stage Part of a product or service's life cycle: generally raw materials extraction; manufacturing; use; and end-oflife. Midpoint level impact assessment LCIA using an indicator located on the cause-effect chain well before its category endpoint(s). For example, representing climate change impacts of methane and nitrous oxide emissions in terms of tonnes of CO2equivalents rather than in terms of years of human life lost.

Time integration
Considerations of infinite fate of a pulse emission.

S3 Background to the time integrated inventory of terminal degradation products
In an LCA case where per-and polyfluoroalkyl substances (PFAS) are included the starting point is the inventory phase where all PFAS, i.e. the primary pollutants, in the product/service's life cycle are to be identified and emissions quantified. The primary pollutants can be precursors to more stable degradation products, as illustrated in S8 Figure S 1. The procedure to arrive at a long-term (time integrated) life cycle inventory (LCI) for characterization is outlined in Table S 2 and further described and justified below. Inventory of direct emissions of primary pollutants LCI: Collect emission data for the product/service life cycle on all PFAS that are primary pollutants in the system (including impurities and residuals), i.e. PFAAs but also higher derivatives and polymers and chemicals such as HFCs and HFEs that can form perfluorinated substances. Quantify direct emissions.

Inventory of indirect emissions of secondary pollutants
Step A-1. Identify the relevant chemicals for characterization, herein focus is on the PFAAs but an inventory can also include other highly persistent PFAS such as high (eco)toxicity intermediate degradation products. Identify their respective accumulation compartments in relation to the inventoried PFAS emissions in Step 1.
Step A-2. Assign transformation fractions ( Table 1 in the main article) to convert Step A-1 into a quantitative emissions inventory.

Total inventory
Step A-3. Summarize all direct and indirect emissions of PFAAs, or other relevant intermediates, over the product's or service's life cycle, per compartment. Step A-1: Transformation of PFAS precursors into their respective terminal degradation products (mainly perfluorionated alkylated acids; PFAAs 5 ) occur both abiotically 8,9 and biotically [10][11][12] , in air water and soils. The compartment where the chemical resides, and type of precursor will determine what transformation chemical that can be expected. Biotransformation pathways varies between experimental systems (i.e. microbial and animal models but also between e.g. soil types), as described by Butt, et al. 11 and Liu and Mejia Avendano 10 . In addition to differences between experimental systems there were also differences between similar substances in the same type of systems, such as that the major degradation product of 8:2 fluorotelomer alcohol (FTOH) in soil incubations was perfluorooctanoic acid (PFOA) (40%) while for 6:2 FTOH it was not the analogous metabolite perfluorohexanoic acid (PFHxA) (8%) but instead perfluoropentanoic acid (PFPeA) (30%). As an attempt to reduce LCI complexity it is here suggested to disregard the possibility of different types of degradation pathways for PFAA precursors of different chain lengths. Instead "related substances" (i.e. the major polymeric and non-polymeric precursors) of each of the PFAAs are to be identified. In analogy with the PFOA restriction proposal 13 PFOA related substances 1 would contribute to PFOA accumulation and PFHxA related substances (i.e., with C6F13 and C5F11 attached) to PFHxA accumulation. The sulfonic higher derivatives were assumed to yield the corresponding Cn PFSA (e.g. PFBS from the perfluoroalkane sulfonamidoethanol, EtFBSE). This approach does not capture all degradation products but aim to quantify accumulation of PFAAs generated by degradation in a simplified manner focusing on a few chemicals.
This simplification assigns PFOA as a proxy for all PFAAs generated from fluorotelomer based non-polymeric substances with a C7F15-or C8F17-moiety, PFHxA as a proxy for all PFAAs generated from fluorotelomer based non-polymeric substances with a C5F11-or C6F13-moiety and PFBS from non-polymeric PASF-based substances with a -C4F9 moiety. Although this is a deviation from the degradation patterns empirically observed it is justifiable considering data availability, as most other PFAAs have been studied to a limited extent. 14-16 PASF-based substances have been shown to generate PFSAs and PFCAs in air, 8,9,17,18 unaccounted for here. Emission compartments are suggested to be assigned to the chemicals where they are expected to be created, e.g. emission of FTOH to air that degrade to PFCA generates a PFCA emission to air. When the compartment where the chemical is generated is not known it is suggested to assign the primary emission compartment, i.e. where the mother compound was emitted.
Step A-2: The amount of chemical yielded in relation to its mother compound emissions depend on compartment and degradation pathway, as indicated in Step A-1. What type of degradation to include? Clearly the degradation of the fairly rapidly degradable non-polymeric PFAS are to be considered. Laboratory experiments can be used to 1 " Any related substance (including its salts and polymers) having a linear or branched perfluoroheptyl group with the formula C7F15-directly attached to another carbon atom, as one of the structural elements."; "Any related substance (including its salts and polymers) having a linear or branched perfluorooctyl group with the formula C8F17-as one of the structural elements." S10 derive such yields but need in this context be extrapolated to accommodate the long-time perspective (laboratory studies cover at maximum a few years while it is here the aim to capture accumulation over at least 100 years).
Summarized degradation data e.g. REF 6 S4 Background information to model set-up and adaptations made

S4.1.3 Air-water/solids partitioning
USEtox air compartments consist of a gas phase, an aerosol phase and a rainwater phase. The fraction of chemical in gas phase air (and indirectly in aerosol and rain water) is determined based on a chemical's octanol-air partitioning coefficient (KOA), air-water partitioning coefficient (KAW) and its pKa if it is an acid REF 23, eq. 8 . The KOA is based on KOW (KOA is estimated as D / (Kh.FRorig), where D is the apparent octanol/water partition coefficient at neutral pH and the Kh.FRorig is the dimensionless gas-water partitioning coefficient for the original species). For inorganics the fraction in gas phase is set to zero (via the flagging in the substance data sheet) and for several PFAS, i.e. the PFAA, this could be a viable modelling option as air concentrations are expected to be negligible. 25 Another, more robust option, is to include the aerosol-gas partitioning coefficient (KQA) as a parameter in the USEtox substance data, S15 allowing for inclusion of empirical data or alternative estimation methods to populate the model. The KQA is dimensionless (normally expressed as (mol/m 3 )/(mol/m 3 )). 27 The KQA, the partitioning coefficient between aerosol particles and air, was introduced into the Substance data sheet to allow for inclusion of data directly for this parameter instead of basing it on KOW. In the fate calculation the intermedia partitioning at urban, regional and global scale was adapted to use the KQA in the calculation of the fraction of the chemical in gas phase air for organics with such data. For organics without the KQA specified, the original calculation routine was left unchanged.
The fraction of the original species in gas phase (FRg.aU/C/G) was calculated with Eq. S 1 for chemicals with data on KQA: Where the frQ is the volume fraction of aerosol particles in air, the frW is the volume fraction of water in air, fr.orig.cldw is the mass fraction original species in cloud water, and the KQA and KAW are the partitioning coefficients for the original species, between aerosol-air and air-water, respectively. The frW was set to 2.46×10 13 as in the USEtox model Particle-gas partitioning coefficients (Kp (m 3 /µg) see e.g. Finizio, et al. 28 or Kip (m 3 /g), see e.g. Arp and Goss 29 ) can be converted to the dimensionless KQA by multiplication with the aerosol density according to Eq. S 2, where ρp is the aerosol density, and the value of 2000 kg.m-3 according to Finizio, et al. 28 , which is also the density used in SimpleBox 4.0, is recommended.

= ×
Webster and Ellis 30 showed that for PFOA the KOA based estimation equation (KQA=α×KOA, where α is an empirically derived constant ) can be used if correction factors as suggested by Ahrens, et al. 31 are applied. For ionizable chemicals gas-particle partitioning that account for both the neutral and ionized form are applicable (the neutral form is in the gas phase but on the particle both the neutral and the ionized form can exist) and are normally expressed as Kp´.

S16
Alternatively, the fraction of chemical on aerosol particles could be estimated according to the method applied in the European risk assessment procedure under REACH, based on Junge 1977 32 . In this method the substance's subcooled vapor pressure and the aerosol surface area are the basis for the estimation. Xiao and Wania 33 concluded that methods to estimate particle-gas partitioning based on either KOA or sub-cooled vapor pressure can be used interchangeably. However, Webster and Ellis 30 found better support for the KOA based equation compared to the subcooled vapor pressure equation in their comparative study with PFOA. Ahrens, et al. 34 found that the relationship between the particle associated fraction and the sub-cooled vapor pressure was offset by about 4 orders of magnitude for PFAS compared to other semi-volatile organic compounds, probably due to that many PFAS partition to aqueous phases. Ahrens, et al. 34 reason that micelle formation and surface-active agents in aqueous aerosols need to be taken into account for derivation of PFAS specific models. The Junge equation was applied in the SimpleBox 3.0 model, in the SimpleBox 4.0 however the basis to arrive at a fraction of chemical in air was a KOA approach based on Götz, et al. 35 and Harner and Bidleman 36 .
In the PFAA adapted model PFCAs in the atmosphere were predicted to be almost entirely in the gas phase (96% for PFOA) and though PFCAs do absorb to both aerosols and water in the atmosphere this is to be expected due to the low fraction of both aerosols and water in the atmosphere in the model. This is also in line with the reasoning by Arp and Goss 29 , who state that negligible fractions of PFCAs will be absorbed to aerosol and cloud water, except during rain events when PFCAs will almost completely partition into the water.

S4.2.1 Bioaccumulation
Bioaccumulation parameters, bioaccumulation factors (BAF) for plants and fish, and biotic transfer factors (BTF) for meat and milk are all estimated from KOW if data are not entered into USEtox. For several PFAS special care must be taken when populating these parameters as the standard BAF or BTF may not reflect PFAS accumulation routes. 37,38 The KOW dependence is not completely removed by population of all BAF and BTF parameters in the substance data of USEtox since BAF for air-particles/gas transfer to above ground produce is estimated in the simplified plant uptake model based on KOW. 23 For PFAS expected to occur in negligible levels in the air compartment the air-plant uptake can be disregarded. To arrive at a model set-up allowing for inclusion of such data if available, the parameters LCFairgas and LCFair-particulate were added to the Substance data tab. In the Human exposure tab, the formulae were adapted to allow for inclusion of data on LCFair-gas and LCFair-particulate if such were given in the Substance data tab. Original estimation routines for organics and inorganics was left unchanged for chemicals with no data on LCFair-gas and LCFairparticulate.
The rate constant dissipation in above-ground plant tissues is estimated based on KOW in USEtox if data for this parameter is not entered. As this parameter described the rate constant for elimination by chemical transformation it can be approximated with the chemical degradation rate. S17 S5 Substance data for three PFAAs  48 Goss and Arp 47 recommend modelling approaches to include the span of possible pKa values between 0 and 4. The authors Moroi, et al. 49 could not determine an experimental pKa of PFOA but found C1-C5 3 pKas to be lower than 1 and C9-C10 pKas to be between 2-3. Vierke, et al. 50,51 studied the partitioning of PFAS in a wastewater treatment plant and the measured air-water concentration ratio was in the lower range of the modelled ratio based on pKas ranging from -0.2 to 3.8 (comment by Rayne 52 ). Based on the available literature a best estimate for the pKa of PFOA is difficult to assign and the suggested range from 0-4 should be modelled. A pKa of 0.5 was assigned to the best estimate as it was the result of most recent experimental studies.

KOW:
Is not an applicable measure for PFOA 7,25,53 and was left blank in the model input data.
KOC: Is the main descriptor of PFOA soil sorption and the soil mineral fraction has negligible effect on sorption. 54 A recent study does however show that OC content may not be the only factor determining the solids-water partitioning, but that OC content, pH and clay content are important properties. 55 Campos Pereira, et al. 56 showed that KOC for 10 of the PFAS included in their study increased with decreasing pH or increasing cation concentration (the latter occurred for PFAS of intermediate chain length only). In the further work the KOC was still used as the parameter describing the solids-water partitioning, acknowledging the uncertainty added by the possibility that other parameters than organic carbon also have a large impact on the partitioning. KOC was found to be similar for PFOA and PFOA salts, 25 and since PFOA and its salts are expected to be fully dissociated in solution this is not surprising; it is the partition of the PFO anion that is measured. Armitage, et al. 25  ), but state that partitioning behavior in the field is indicated to be different, with 3 Note that these authors denote C1-10 for the number of perfluorinated carbons, i.e. C5 is the hexanoic acid S21 higher KOC (which is also the general trend 55 ). In the most recent review by Li, et al. 55 log KOC values ranged between approximately (read from graph) 1.3 to 3.2. The authors also show a prediction of KOC based on pH (citing Lee, et al. 59 ), and the plotted literature data create a fairly even scatter around the predicted line. Based on these data a log KOC of approximately 1.86 was derived (neutral and ionized forms), assuming that the predicted levelling off in the estimated KOC based on pH represents a reasonable representative value. Findings in a selection of studies not included in the reviews above:  Vierke, et al. 60 found log KOC values of 3.9 and 4.0 L.kg-1 in their experiment with a water-saturated sediment column but expected them to be slightly overestimated. The KOC obtained from the most recent review, 55   Sol25: the water solubility of PFOA is lower than that of the salts 25 . ECHA 19 state water solubility of 9.5 and 4.14 g.L-1 and 25°C and 22°C, respectively, and the same values are reported by Armitage, et al. 25 68 gives an atmospheric degradation rate. The resulting degradation rate 1.3×10 -7 s-1 was used as best estimate and the span 1.1-1.4×10 -7 s-1 was used as minimum and maximum.
kdegW, kdegSd, kdegSl: Degradation rates in different media, were approximated based on the same model assumptions as made by Armitage, et al. 25 . PFAS that are terminal degradation products, such as PFOA, are not expected to degrade under environmental conditions. Armitage, et al. 25 reviewed a number of studies and concluded that PFCAs are stable under a wide range of environmental conditions. The authors acknowledged the possibility of slow degradation processes occurring and assigned PFO a half-life in soil and water of 69-6900 years (corresponding to degradation rates of 3.2×10 -10 to 3.2×10 -12 s-1). The full span was used in the uncertainty modelling and assigned to the degradation in water, and the slower degradation rates were set as best estimate. Soil and sediment degradation rates were calculated with the division factors 1:2:9 as proposed in the USEtox manual for organic substances, 68 assigning the 69-6900 years half-lives to water degradation.
kdissP: Dissipation rate in plants, is calculated based on KOW unless empirical data are given. Since plant uptake and adsorption to solid particles are two processes likely to be different it is not certain that a KOW back-calculated from KOC is a good proxy for the dissipation rates. As dissipation rates largely reflect chemical transformation in plants for non-volatile substances the degradation rates for water was applied also here.
avlogEC50: For the chronic toxicity it is the PFO anion that is expected to exert the effects and data for PFOA as well as its salt are assessed as relevant for inclusion. 19 As the avlogEC50 is an average across species and trophic levels minimum and maximum were not calculated. The ecotoxicity factor was calculated with data extracted from the ECOTOX database (retrieved 20191203 4 ). 78 Data for in vitro studies and data on genetic and "enzyme" endpoints were removed, as many had unclear relevance for the aquatic ecotoxicity endpoint (e.g. mRNA type of effect measurements). Data from studies with reported impurity of the active ingredient less than 90% were removed. If one study had several data included for the same species the most sensitive endpoint was used (if several data point were available for the same species and same endpoint, the geometric mean was calculated), based on the extrapolated EC50chronic equivalents. When chronic data were available for three trophic levels acute data were disregarded. Extrapolated EC50-chronic equivalents were calculated based on Aurisano, et al. 79 with the exception that tests on eggs, embryos or sac-fry stages were classified as chronic, despite short test durations. 80 Changes in species names were not controlled.
Data from studies where no test duration was recorded were assessed on a case by case basis, with support from risk assessment guidance. 80 Data items with ">" or "<" assigned were removed. Each individual dataset was not evaluated for relevance. Chronic data was not available for saltwater species in three trophic levels and the freshwater and saltwater datasets were merged. The avlogEC50 was calculated to 0.85 mg.L-1 for freshwater and marine water, according to the USEtox manual. 81 Data in Table S   BAFsoil-root: The bioaccumulation factor for roots, or the BAF for unexposed produce such as root vegetables, is in USEtox calculated with internal quantitative structure-activity relationships (QSARs), but can also be entered into the model if data are available. The unit is assumed to be kg FW soiL.kg-1 FW plant (note that the unit has been inversed in the USEtox model Substance data sheet), based on statements in the manual for inorganic chemicals that conversion is made from dry matter to fresh matter (with a factor of 5 applied to the dry weight/dry weight data as presented in the original reference), the units in the inorganic substances database and the use of the BAF with the bulk density of soil and fresh weights of vegetables. Sternbeck, et al. 90 calculated BCFs for PFDA to be 0.0007 and 0.064 kg DW soil.kg-1 FW veg, for root crops and cereals, respectively. These factors were based on PFOA data as reported by Stahl, et al. 91 . As the Sternbeck, et al. 90 31 . Log KQA calculated from field data, with these correction factors, were in the range 8.3 and 9.5 (KQA 2.0×10 8 and 3.2×10 9 ). Ahrens, et al. 108 measured PFAS atmospheric concentrations in the Arctic as well as gas-particle partitioning, without corrections for the above-mentioned artefacts. The authors found that longer chain S28 PFAS (C10-12 and C14) were particle associated with 76-83% but data were not available for PFOA for Kp calculation. Yao, et al. 109 measured PFOA in industrial, urban and background sites in China during winter and summer. Based on their data Kp´s were calculated to be in the range 0.01-0.1 m3.µg-1, by application of the Kp´= (c´p/TSP)/cg) formula, using the average concentrations and assuming particle phase concentrations were given for c´ and using the site specific average TSPs (particle weight/sample volume). Recalculated to KQA, using the aerosol density of 2000 kg.m-3, the KQAs span over 2.3×10 10 -2.0×10 11 . These values were not corrected for GFF sorption.
Applying the correction factors, as was done by Webster and Ellis 30 results in KQA 6.6×10 8 -5.8×10 9 . Tian, et al. 110 derived Kpa values (m3air.m-3.dep-1) that were re-calculated to KQA by application of the aerosol particle density (resulting in an average KQA of 9.2×10 8 ), acknowledging that these values may be underestimations as PFOAA on fine particulate matters may not have been measured. Of the data presented above the Arp and Goss 29 KQAs are the lowest and the Ahrens, et al. 34 the highest. The Arp and Goss 29 KQAs could still be erroneous despite the ambition to correct for overestimation of PFCA particle associated fraction 105 . However, that should result in overestimated KQAs.
It was not possible to identify any data to be more reliable than other data and the best estimate was set to the average of the KQA (2.3×10 9 ) and the min and max were set to the full span of values in the dataset.
LCFair gas/particle-leaf: The only study found was a field study by Tian, et al. 110 68 gives an atmospheric degradation rate.
kdegW, kdegSd, kdegSl: The PFCAs are believed to be equally persistent and the same degradation rates as for PFOA were applied.
Kdiss: The PFCAs are believed to be equally persistent and the same degradation rates as for PFOA were applied. The ECOTOX database data were curated as described for PFOA above. Articles were retrieved from SCOPUS with a broad search string 6 generating 258 document results, and relevant articles were picked based on relevance (chemical tested and endpoints relevant to aquatic ecotoxicity). The SCOPUS search did not generate any additional data that could be included in the EF calculation. Available data did not allow calculation of separate freshwater and saltwater EFs. The avlogEC50 was calculated to 2.5 mg/l based on data in Table S 7.  114 , not included in the ECHA 2018 review derived an average RCF 0.2 g soil.g root-1 (dry weights, OC normalized) for wheat (2.1 g soil.g root-1, on a soil basis, converted as above). Bizkarguenaga, et al. 94 , not included in the ECHA 2018 review measured PFHxA in soil and carrots in their study of uptake from compost amended soil with average BAFs to be 0.7 g soil.g root-1 (wet weights, converted as above and BAFs weighted assuming 10% peel and 90% core). No other BAF study with results expressed on the basis of soil concentrations were found. The carrot study is the most relevant to use for the BAFsoil-root, as wheat roots are not eaten, and that BAF was used as best estimate. Though not visible in the data, as fewer data were available compared to PFOA, it is likely that the BAFs are very uncertain.
BAFsoil-leaf: PFHxA enrich in above ground plants 111 Doucette, et al. 92 include the study by Yoo, et al. 115 , in which the average BAF (grass/soil accumulation factor) was 3.4 g soil.g grass-1 (dry weights). In addition to the Yoo, et al. 115 study ECHA 111 lists Wen, et al. 97 and from their data an average BAF for wheat grain of 0.9 kg soil.kg-1grain (wet weight, converted assuming 90% dry fraction in grain and 30% dry fraction in soil) was calculated, and Krippner, et al. 98 and from their data an average BAF for maize kernels of 0.5 kg soil.kg-1 grain (wet weight, converted assuming 90% dry fraction in grain and 30% dry fraction in soil) was calculated. Other data on BAF via soil exposure were not considered as grains were prioritized as the relevant crop [116][117][118] . The average over the wheat grain and maize kernels, 0.7 kg soil.kg-1grain (wet weights).
BTFmeat and BTFmilk: ECHA 111 state that PFHxA has a strong binding potential to proteins but that effective renal clearance in mammals limits bioaccumulation. Vestergren, et al. 99 included PFHxA in their bioaccumulation study in dairy cows but could only detect the substance in water samples and therefore could not derive any BTFs. Numata, et al. 102 showed fast excretion of PFHxA also in pigs and derived a BMF of 0.08 for the pig meat. The BMF was recalculated to a BTF of 0.04 days.kg-1 meat, using the data from the study (feeding rate 2 kg/day and animal, average feed concentration 48 µg/kg, feed dry weight 90%). It should be noted that the PFOA BTF derived from this study (2.7 days.kg-1) is higher than the BTF for cows derived by Vestergren, et al. 99  BAFfish: Holmquist, et al. 20 reviewed bioaccumulation data for PFHxA in fish and found that in laboratory experiments bioaccumulation was low, field BAFs were higher but still the classification of bioaccumulation potential was set to low. ECHA 111 came to the same conclusion. In their laboratory experiment Martin, et al. 119 PFHxA could not be measured in the fish tissues. Assuming that the LOD was approximately 0.003 µg.g-1 (based on lowest concentrations measured, Fig 2) a BCF estimation (conc. in fish tissue over conc. in water) resulting in a BCF of 1.8 L.kg-1 was made. This value is highly uncertain and based on the above-mentioned conclusions that the bioaccumulation in fish is low the BAF was set to 1.
KQA: Arp and Goss 29 measured KIPs to 0.49-4.18 m3.g-1 (at 50% RH), with an average of 2.2 m3.g-1. Converted to a dimensionless KQA of 4.5×10 6 . Ahrens, et al. 34 did not present Kps for PFHxA as the gas-particle distribution was 0. Yao, et al. 109 could not measure PFHxA in gas and/or particles in several of their samplings, data on both gas phase and particle phase average concentrations were only available from two locations and only summer samples. The average dimensionless KQA was 1.8×10 9 . Field data from Tian, et al. 110 were recalculated to an average KQA of 5.8×10 8 .
It was not possible to identify any data to be more reliable than other data and the best estimate was set to the average of the KQA average values (7.9×10 8 ) and the min and max were set to the full span of values in the dataset.
LCFair-leaf: as for PFOA, air-plant uptake is assumed to be low, due to the low PFHxA prevalence in ambient air.
Lacking data on this parameter a low value of 0.01 was assumed.

S5.3 Data selection and justification: 375-73-5 Perfluorobutanesulfonic acid (PFBS)
Target class for pesticides/Chemical class for pesticides: Not applicable as PFBS is an industrial chemical and a degradation product.  68 gives an atmospheric degradation rate.
kdegW, kdegSd, kdegSl: The PFSAs are believed to be equally persistent and the same degradation rates as for PFOA were applied.

S33
Kdiss: The PFCAs are believed to be equally persistent and the same degradation rates as for PFOA were applied.
avlogEC50: For the chronic toxicity it is the PFBS anion that is expected to exert the effects and data for PFBS as well as its salt are assessed as relevant for inclusion (based on the same line of reasoning as for PFOA). Data were retrieved from two recent reviews, 20,83 complemented with data from the ECOTOX database (https://cfpub.epa.gov/ecotox/) 7 and from studies reported in articles retrieved from SCOPUS (www.scopus.com). The ECOTOX database data were curated as described for PFOA above. Articles were retrieved from SCOPUS with a broad search string (see PFHxA) generating 258 document results, and relevant articles were picked based on relevance (chemical tested and endpoints relevant to aquatic ecotoxicity). The SCOPUS search added data for two species. Available data did not allow calculation of separate freshwater and saltwater EFs. The avlogEC50 was calculated to 1.9 mg/l based on data in Table   S 8. as above). None of these data are directly applicable to edible root vegetables but an average of 7 g soil.g root-1 (fresh weights) was used as the best estimate, using the two available values as minimum and maximum.
BAFsoil-leaf: Krippner, et al. 98 measured PFBS in soil and maize kernels and from those data an average BAF was calculated (n=2) to 0.02 g soil.g kernel-1 (fresh weights, derived as above). Wen, et al. 97  KQA: Based on the results presented by Ahrens, et al. 34 a KQA was calculated (as for PFOA above) to 1×10 10 , with the range 5.4×10 9 -1.9×10 10 . Yao, et al. 109 could not measure PFBS in gas and/or particles in several of their samplings, data was only available on particle concentrations and a KQA could not be calculated. Field data from Tian, et al. 110 were recalculated to an average KQA of 1×10 7 (data were only available for 3 out of 5 locations). It was not possible to identify any data to be more reliable than other data and the best estimate was set to the average of the KQA average values (5×10 9 ) and the min and max were set to the full span of values in the dataset.
LCFair-leaf: as for PFOA, air-plant uptake is assumed to be low, due to the low PFBS prevalence in ambient air.
Lacking data on this parameter a low value of 0.01 was assumed.

S5.4
Toxicokinetic extrapolation of non-cancer toxicity data EFSA 86, 87 have in their last two risk assessments on PFOA made use of human epidemiological data, arriving at TWIs orders of magnitude lower than before, and thus indicating high potential toxicity. To explore implications of this high potential toxicity the epidemiological data from the 2018 risk assessment were used to roughly extrapolate a noncancer human lifetime equivalent ED50 value for PFOA, and further extrapolate this value to PFHxA and PFBS (currently lacking comparable epidemiological data).
For the endpoint of liver enlargement, it has been shown that if internal concentrations are compared at observed effect levels, short-and long-chain PFCAs have similar toxic potencies. 125 We assume that the PFAAs in focus here also have similar potencies with regard to their effects on cholesterol levels and the immune system, i.e. the most sensitive endpoints for PFOA. 86, 87 For serum cholesterol increase, the lower bound benchmark dose based on a 5% serum cholesterol increase (BMDL5), was 9.2-9.4 ng/mL internal concentration, corresponding to 0.8 ng/kg bw per S35 day. 86 The cholesterol threshold was extrapolated to PFHxA and PFBS based on their human elimination half-life relative to that of PFOA and PFOS (3% 126, 127 and 2% 127, 128 , respectively, see Table S 9). Though it is not known if the potencies are similar across PFAAs for this effect, it was assessed as a relevant precautionary assumption due to the absence of epidemiological data for this sensitive endpoint for PFHxA and PFBS.
The PFOA non-cancer ED50 was calculated to 1.3×10 -5 kg.lifetime-1, based on the BMDL5 of 0.8 ng.kg bw-1.day-1 for increase of cholesterol levels. 86 The benchmark response of 5%, which is probably close to a NOAEL for this continuous response, 129 was used and treated as a NOAEL when calculating the ED50 following the USEtox manual.
According to EFSA an increase of serum cholesterol levels by 5% would increase the prevalence of individuals with high cholesterol levels (>240 mg/dL) by more than 5%, justifying the use of this continuous variable. A 1:1 relationship between increased cholesterol levels and disease was assumed in the following calculations herein, though individual risk depends on other factors such as age and blood pressure (the explained variance in the regression model by Steenland, et al. 130 for example was 14%). The BMDL5 value was multiplied with the factor 9 to arrive at an ED50, and converted to the person and lifetime unit by using a lifetime age of 70 years and body weight of 70 kg. The PFOA non-cancer ED50 was extrapolated to PFHxA and PFBS by multiplication with the quotient of the PFOA and PFOS elimination half-life and that of the respective substance (Table S 9). Extrapolated ED50 values were 5.1×10 -4 and 2.0×10 -3 kg/person and lifetime for PFHxA and PFBS, respectively. The sensitivity of the results to the data on elimination half-lives was tested by using a range (2.3-8.5 years) of halflives for PFOA. 132 The extrapolation of ED50 was based on a half-life of 3.5 years for PFOA. 127 The impact of data selection for the PFOA half-life was tested by application of the half-life range of 2.3-8.5 years, 132 resulting in ED50 values for PFHxA of 66-240% of the original value. While EFs of PFHxA and PFBS were shown to be sensitive to this data, the half-life of 3.5 years was assessed as a relevant best estimate, as the higher end of the range was assessed to not accurately reflect half-life in exposed communities. 132 Since the 2020 risk assessment is still preliminary, data were not used to calculate ED50 values for EF derivation. A NOAEC of 31.9 ng/mL internal serum concentration for the sum of four PFAAs, including PFOA and PFOS, was derived by EFSA 87 for the 1-year old child. The endpoint was reduction in antibody titres against haemophilus influenzae type b (Hib) (inverse associations has been shown also for diphtheria and tetanus). Based on physiologically based pharmacokinetic (PBPK) modelling EFSA derived an equivalent NOEC for intake by the mother (exposing the S36 child during pregnancy and via breastfeeding) of 1.16 ng.kg bw-1.day-1 (for the sum of four PFAAs). As EFSA assumed equal potency of the four PFAAs, ¼ of this NOAEC would then be possible to associate with PFOA, i.e. a level similar to that associated with increased cholesterol levels.

S6 List of PFAS estimation models
In addition to the data collection for the three terminal degradation products in focus, a literature review was made to identify relevant in silico models that can be applied in case additional CFs need to be calculated for PFASs that are lacking empirical data. In USEtox, EPISuite is currently the recommended data source for physico-chemical properties, degradation rates and bioaccumulation factors (priority given to experimental data). It has been shown that for PFASs, EPISuite does not perform well, 133 though new versions of EPISuite (from v. 4) have increased performance for some PFASs substance groups. 69 Recent in silico efforts use other generic models, such as SPARC and COSMOtherm but also these have limitations for PFASs. 134 For PFASs terminal degradation products, which are only expected to degrade negligibly under environmental conditions, very low degradation rates must be assigned to obtain realistic steady-state mass balances and data derived by expert judgement may be preferable over estimated data. For example, Armitage, et al. 25 set the degradation to 0.01-1% per year for PFOA (these data were used in the present study). There have been efforts to develop PFASs-specific prediction tools for physico-chemical properties, such as water solubility 135 , vapor pressure [135][136][137] and pKa 42, 120, 138 as well as environmental partitioning 139,140 . Grisoni, et al. 141 compared QSAR models, based on KOW, to predict fish BCF and found that for PFASs the BCF was generally underestimated by these types of models. This is not a complete inventory of in silico methods for PFAS but provide an overview of methods published in the peer-reviewed literature, with focus of PFCAs and PFSAs.
One method to predict chemical partitioning and bioaccumulation, making use of a relatively simple structure based estimation method is the application of molecular connectivity index (MCI) see e.g. REFs 139,142,143 , making use of descriptors of a chemical's relative degree of branching and, in the case of polar chemicals, polar groups. However, an MCI method derived based on branching of hydrogenated alkanes probably doesn't capture the properties of a perfluorinated alkane chain and correction factors will be needed. MCI based methods were therefore here identified as a category of methods that might be applicable for PFAS to derive data for relevant parameters in CF calculation, but was not looked into further as further method development will be needed.
List of in silico methods, specific for PFAS, per substance data parameter in USEtox. The methods have not been assessed for relevance other than that they cover PFAS and have not been compared with each other.
pKa.gain, pKa.loss: With empirical pKa values still being debated for some of the most well studied PFAS (see S1.1 ) estimation methods are bound to be associated with large uncertainty. The physical-chemical property estimation softwares SPARC and COSMOtherm do calculate pKa that have been assessed as relevant for PFAS. In addition to these general models Rayne and Forest 120 further developed a computational method to estimate pKa for C1 through C8 PFSAs. Rayne, et al. 138 found that estimation methods for pKa of PFCAs produce underestimations for longer-S37 chain compounds. Rayne 119 used the perfluoroalkyl chain length as a measure of hydrophobicity but did also correlate the experimentally derived BCFs with the critical micelle concentration (CMC), which has been proposed as a relevant parameter describing bioconcentration of surfactants. 152 The authors found increasing bioaccumulation potential with decreasing CMC for carboxylates but also that CMC did not explain the higher accumulation of PFSAs compared to PFCAs. Eq. S 3

= ⃗
As the model solves the mass-balance for steady-state, model evaluation needs to be made with steady-state conditions in mind. In some compartments, such as air and freshwater bodies, steady-state is obtained fairly quickly for PFAAs, that is within a few years, while in others, such as the global ocean, it can take much longer. 153 As PFAA concentrations in the environment are currently either increasing or decreasing, depending on the media studied, [154][155][156]  Since the authors consider the higher bound of the emission estimates to be most relevant, 6 this was used as model input for evaluation, i.e. inserted as E in Eq. S 3. Lacking information about the emission compartment, a division was made setting 50% emission to continental freshwater and 50% to continental air, assuming that the main direct emissions would be water emissions from fluoropolymer manufacturing and the main indirect emissions would be directed to air via precursors.
Data on environmental concentrations of PFOA for aquatic and air compartments were collected from recent publications. 58,155,157,158 Model evaluation was performed by comparing measured environmental concentrations with predicted continental concentrations.
A modification giving an option to re-circulate chemicals leached from the soil compartment was introduced to the model to explore the importance of the groundwater as a sink. The chemical fraction in soil (natural soil and agricultural soil) that in the original model settings is predicted to leach to groundwater was in this modification further

S7.2 Results
The results of the model evaluation and Figure S 2 is commented in the main text.

S41
S8 Results of the sensitivity analysis Table S 10: Variation of input parameters for PFOA from the interim parametrization with ±50% and the recorded change in characterization factor (%). Changes outside the span 95-105% are marked in blue.

Parameter
Var.

S9 Results of the uncertainty analysis
Results of the uncertainty assessment are shown in Table S 11, Table S 12 and   Table S 13. Parameters were varied in clusters. As a calculation example, the combined parameter uncertainty was calculated for PFOA marine ecotoxicity effects for emissions to soil, using values for each parameter contributing to a low CF. The lowest CF possible, with available data, was 0.7% of the best estimate and thus approximately within two orders of magnitude.

S10 Partial model results
CFs and partial model results for emissions to continental freshwater, air and agricultural soil are presented in Table  S 14. For results for other emission compartments see S11.
S47 Table S 14: Summary table of characterization factors and partial model results for perfluorooctanoic acid (PFOA), perfluorohexanoic acid (PFHxA) and perfluorobutanesulfonic acid (PFBS). PAF=potentially affected fraction. Human health effect factors were not differentiated for ingestion and inhalation and ecotoxicity effect factors were not differentiated for freshwater and marine waters. Human health effect factors based on rodent data.

Midpoint characterization factor
Fate factor Intake fraction Exposure factor Effect factor Human health [

S11
The PFAA adapted LCIA model The PFAA adapted LCIA model is made available as an electronic supplementary information. S12 Case study data and results

S12.1 LCI
Case study data for the cradle to gate LCA are listed in Table S 15. The DWR treated drape was set up to be a 1-layer non-woven polyester (PES) and the plastic-coated drape was set up to be a 2-layer textile with one outer-layer of nonwoven PES and one inner-layer of liquid proofing PE-film (approx. 40 µm thick). To arrive at the functional unit (FU) only the liquid repellency barrier was modelled. The composition of the DWR product was simplified to a content of only a C6 side-chain fluorinated acrylate co-polymer, disregarding the fact that commercial DWR products typically contain co-solvents, dispersing agents and water. (In addition, the DWR finishing product often needs to be complemented with additional chemical products, such as cross linkers, wetting agents and catalysts.) The used amount of C6 co-polymer was back-calculated from an assumption of a polymer deposition of 1 weight-% on the outer fabric (Personal communication, Steffen Schellenberger, ACES Stockholm University). The DWR finishing step was assumed to be a wet-treatment and drying in a stenter frame, and described using data published by European Commission 161 . The plastic film coating was assumed to have a surface weight of 40 g/m 2 (based on unpublished data) and to be laminated by melt extrusion. The C-6 fluorinated acrylic copolymer production dataset did not include PFAS emissions, so those were added following the framework proposed here, to arrive at a long-term inventory of terminal degradation products. The PFAS primary pollutants emission inventory was constructed based on the following assumptions and data:  The C6 co-polymer was assumed to contain 60% fluorine, based on documentation in the C-6 fluorinated acrylic copolymer production dataset, corresponding to approximately 75% C6F13 by weight (this is a high F-fraction and commercial DWR would in general contain less F).

S54
 The content of PFHxA (or any other PFAA) impurities was assumed to be negligible, based on Wang, et al. 6 .
 The content of residuals was approximated to one single species, 6:2 FTOH, and the concentration in the polymeric product was assumed to be 4%×0.05, based on Wang, et al. 6 .
 Polymer emissions were assigned to water. An emission factor (EmF) of 0.06 was assigned to the chemical manufacturing step based on the relevant ECHA 32 Environmental Release Category (ERC1: "Manufacture of the substance"). An EmF of 0.3, based on ECHA 32 (ERC8c: "Widespread use leading to inclusion into/onto article (indoor)"), was assigned to the textile finishing step. To account for wastewater treatment emission were further reduced by 90%.
 Residual emissions were assigned to air and an EmF of 0.002 was assigned to the chemical manufacturing step and to the textile finishing step based on Hischier, et al. 165 .

S12.2 Calculation of PFAS IS
PFAS ISs were calculated according to Eq 1 and 2. in the main article. The PFAS compounds (x) considered here were the C6 co-polymer and 6:2 FTOH. Polymer emissions to water (Ewater, POL) and 6:2 FTOH emission to air (Eair, FTOH) were calculated as: Eq S1. Ewater, POL = ∑PROC (UPOL, PROC × EmFwater, PROC × 10%) Eq S2. Eair, FTOH = ∑PROC (UPOL, PROC × 4% × 0.05 × EmFair, PROC) Where PROC is the respective process, here chemical manufacture and textile finishing. UPOL is the amount of polymer used per FU (corrected for emission losses), the EmF is used as stated above and the 10% is the reduction due to wastewater treatment. By application of Table 1  Where 75% is the fraction PFHxA in the polymer, 50% is the transformation fraction for polymer degradation and 60% is the transformation fraction for the non-polymeric fluorotelomer based substances (applied to FTOH but also to the non-polymeric derivative created at polymer degradation). The molecular weight of PFHxA is 314.05 and of 6:2 FTOH 265.1, and their fraction was used to calculate the mass fraction PFHxA that is created from FTOH transformation.

S12.3 Results
Results of the case study are shown in Figure S 12. PFAA emissions were modelled in three scenarios (based on   the translation table, Table 1 in the main text): Basic: Polymer degradation with a yield of non-polymeric PFAS of 50%. Degradation of non-polymeric PFAS with a PFAA yield of 60%.
Low: No polymer degradation. Degradation of non-polymeric PFAS with a PFAA yield of 30%.
High: Polymer degradation with a yield of non-polymeric PFAS of 100%. Degradation of non-polymeric PFAS with a PFAA yield of 90%.
The first three set of results from the DWR-drape ( Figure S 12, to the left) are ISs calculated with PFAA-CFs where the EFs were based on rodent-data, and the fourth set of results are also for the basic scenario but with CFs for human toxicity non-cancer effects, where the EFs were based on roughly extrapolated human epidemiological data (see Figure 2 in the main text). In the fifth set of results CFs based on EFs based on roughly extrapolated human epidemiological data were used but with the low emission scenario. Using the high emission scenario (not shown in the graph) indicator results for the DWR-drape, for human toxicity non-cancer, would be more than 2000 times that of the coated drape.