Integrated Treatment of Per- and Polyfluoroalkyl Substances in Existing Wastewater Treatment Plants—Scoping the Potential of Foam Partitioning

Foam fractionation is becoming increasingly popular as a treatment technology for water contaminated with per- and polyfluoroalkyl substances (PFAS). At many existing wastewater treatment facilities, particularly in aerated treatment steps, foam formation is frequently observed. This study aimed to investigate if foam fractionation for the removal of PFAS could be integrated with such existing treatment processes. Influent, effluent, water under the foam, and foam were sampled from ten different wastewater treatment facilities where foam formation was observed. These samples were analyzed for the concentration of 29 PFAS, also after the total oxidizable precursor (TOP) assay. Enrichment factors were defined as the PFAS concentration in the foam divided by the PFAS concentration in the influent. Although foam partitioning did not lead to decreased ∑PFAS concentrations from influent to effluent in any of the plants, certain long-chain PFAS were removed with efficiencies up to 76%. Moreover, ∑PFAS enrichment factors in the foam ranged up to 105, and enrichment factors of individual PFAS ranged even up to 106. Moving bed biofilm reactors (MBBRs) were more effective at enriching PFAS in the foam than activated sludge processes. Altogether, these high enrichment factors demonstrate that foam partitioning in existing wastewater treatment plants is a promising option for integrated removal. Promoting foam formation and removing foam from the water surface with skimming devices may improve the removal efficiencies further. These findings have important implications for PFAS removal and sampling strategies at wastewater treatment plants.


Limits of quantification
Method limits of quantification (LOQs) for the sample extracts were calculated using Equation SI 1, with 〈  〉 and    the mean and standard deviation of the extract concentrations in the blanks (n = 12), respectively.Outlying blank concentrations (defined as being more than three standard deviations away from the mean) were removed prior to LOQ calculations, and concentrations below the instrument quantification limit (0.05 ng mL -1 ) were set to 0.05 ng mL -1 .For the calculation of the LOQ of TOP assay samples, the concentrations in the TOP blanks (n = 7) were used instead.Because the volume of foamate extracted was lower than that of water, LOQs for water samples were lower than those for foam samples.The LOQs as given in Table SI 2 were converted based on the volume of sample extracted to give the LOQ in each sample.

Equation SI 1
Table SI 1: LOQs in sample extracts and extracts of samples after the TOP assay.The LOQs in the samples varied based on the extracted sample volume.An extract LOQ of 0.05 ng mL -1 corresponds to a sample LOQ of 0.4 ng L -1 for a sample of 125 mL, 5 ng L -1 for a sample of 10 mL (the highest foamate volume extracted) and 200 ng L -1 for a sample of 0.25 mL (the lowest foamate volume extracted).

Compound
LOQ normal extracts (ng mL

Calibration curve
The calibration curve concentrations were 0.05, 0.1, 0.5, 1, 5, 10, 50, 100, 250, 500 and 900 ng mL -1 .A calibration curve was run before and after all samples from one site.Because the variation in PFAS concentrations between the different samples and the different sites was large, it was necessary to use a calibration curve that included a wide range of concentrations.For some compounds, this curve was not linear over the full concentration range.When necessary, certain points of the calibration curve were excluded or the regression was changed to quadratic, to ensure a good fit of all calibration curve points (R 2 ≥ 0.99).All compounds with excluded concentration points or a non-linear regression are shown in Table SI 2. Extract concentrations outside of the range of the calibration curve (> 900 ng mL -1 ) were extrapolated based on the curve's regression equation.This only applied to at most three compounds at four sites, with the concentrations always being acceptably close to the highest included standard concentration.Recoveries well above or below 100 % were often due to high concentrations in the unspiked foamate samples.E.g., when the concentration of an unspiked foamate extract was 250 ng mL -1 , a method variability of 10 % may have already caused a recovery of 0 % or 200 %.Recovery of 9Cl-PF3ONS in the foamate samples was consistently low, which indicates that matrix suppression decreased the signal.Since this compound was not detected in any of the samples, it was left out of the data analysis and its low recovery thus did not affect the presented results.The stock solution used to spike the 10 ng Milli-Q samples and the foam samples from Site A was probably contaminated with PFBA, leading to recoveries that were a factor two too high.A different stock solution was used for the remaining samples, in which the recovery of PFBA was always within an acceptable range.

Mean PFAS concentrations for each site
Table SI 6: Mean PFAS concentrations (ng L -1 ) in influent, effluent, water under the foam and foam for each site (A-J) included in the study.W When all triplicates had concentrations below the LOQ, the concentration is reported as < LOQ.When at least one triplicate had a concentration above the LOQ, the other triplicates' concentrations were set to half the LOQ, and the average of the three values was reported.1, labels of the subplots correspond to the site identifiers), with concentrations below the LOQ set to the LOQ.MBBR = moving bed biofilm reactor, AS = activated sludge, EC = electrocoagulation, Ozone = ozonation.

Quantification of required foam fraction for increased long-chain PFAA removal
When ignoring sorption to sludge and reactive transformation of precursors into target PFAS, the mass balance over a foam-forming process can be written as follows, provided that the foam would be removed from the top of the reactor: Here, V is the volume of the reactor (m 3 ), Cbulk, CIn, CEf and CFoam the PFAS concentration in the reactor, influent, effluent and foam, respectively (all mol m -3 ), and   ,   and   the influent, effluent and foam flow rate (all m 3 hr -1 ), respectively.At steady state and constant reactor volume, this means that: −     −     = 0, and: )   Then: The next step is to relate the foam concentration to the effluent concentration.Using the analysis by Stevenson and Li (2017) 1 , CFoam can be related to CEf as below, with  the surface excess concentration of PFAS (mol m -2 ) and r32 the Sauter mean bubble radius of the foam (m): The equilibrium relation between the surface excess and the effluent concentration is given by an adsorption isotherm.For simplicity, a Henry's law isotherm is used, which is relatively realistic at low concentrations, with Henry's constant KH (m): There are weaknesses in this analysis that must be pointed out.As aforementioned, the analysis ignores adsorption to sludge and reactive transformation of precursors.Secondly, a Henry adsorption isotherm is only realistic at low concentrations and frequently the more accurate, but more complex, Langmuir isotherm is used.Thirdly, in reality, r32 is a variable that will often change when the foam fraction increases.Changing the foam fraction without changing the bubble radius is difficult, since the wetness of the foam is a function of bubble size, so K is only an independent constant if the foam fraction is increased without increasing the foam wetness.Finally, as pointed out in the main text, the foam is not removed from the water surface in any of the plants under investigation in this study, and the retention time of the foam was likely higher than that of the water in most plants.
Despite these limitation, Equation S1 may be used to roughly estimate the required foam fraction that would be necessary to achieve certain levels of long-chain PFAA removal.For this calculation, the removal and enrichment of Σlong-chain PFAA were used, since these compounds are removable with foam fractionation, and using summed concentrations moderates the effects of non-detect concentrations.As visible from Figure SI 6b, at sites D and E, an approximately seven-fold increase in volumetric foam formation may already result in a Σlong-chain PFAA removal of 80 %.Since the calculated foam fractions at these two sites were currently both < 0.5 % (Figure SI 6a), this may be achievable.Furthermore, at site J, a removal of > 99 % would require a foam fraction of only 3 % (Figure SI 6a).However, it should be stressed that these calculations are approximations only and that artificially increasing the foam formation while keeping the bubble size (and thus foam wetness and relative surface area) constant may not be possible.
(Equation S1, Foam Fraction)  1.These plots are rough approximations only, since the calculations ignore sorption to sludge and reactive transformation of PFAS and assume the relative surface area of the foam to be constant independent of foam fraction, which is not realistic.1.

Figure SI 3 :
Figure SI 3: PFAS concentrations after the TOP assay in the influent (In), effluent (Ef), water under the foam (UW) and foamate (Foam) for all treatment plants included in the study (see main text Table 1, labels of the subplots correspond to the site identifiers).Concentrations below the LOQwere set to 0.5•LOQ.Foamate concentrations are presented on the y-axis on the right.Titles give the enrichment factors (EF) calculated based on the concentrations after TOP.The text above the bars gives the molar percentage of PFAS compared to the target measurement, i.e. percentages above 100 % indicate an increased ΣPFAS concentration due to precursor degradation.Concentrations below 100 % are probably due to measurement uncertainties.The standard deviations are based only on the variability in target concentrations (n = 3), since TOP assays were done without replicates.MBBR = moving bed biofilm reactor, AS = activated sludge, EC = electrocoagulation, Ozone = ozonation.

Figure SI 4 :
Figure SI 4: PFAS concentrations after the TOP assay in the influent (In), effluent (Ef), water under the foam (UW) and foamate (Foam) for all treatment plants included in the study (see main text Table1, labels of the subplots correspond to the site identifiers), with concentrations below the LOQ set to zero.MBBR = moving bed biofilm reactor, AS = activated sludge, EC = electrocoagulation, Ozone = ozonation.

Figure SI 5 :
Figure SI 5: PFAS concentrations after the TOP assay in the influent (In), effluent (Ef), water under the foam (UW) and foamate (Foam) for all treatment plants included in the study (see main text Table1, labels of the subplots correspond to the site identifiers), with concentrations below the LOQ set to the LOQ.MBBR = moving bed biofilm reactor, AS = activated sludge, EC = electrocoagulation, Ozone = ozonation.

Figure SI 6 :
Figure SI 6: Foam fraction (a) and increase in foam fraction (b) required to reach a certain Σlong-chain PFAA removal.Asterisks in a) represent the Σlong-chain PFAA removal and calculated foam fraction as found from the concentrations obtained in this study, i.e. during normal plant operation.Only sites for which the measured Σlong-chain PFAA removal was positive were included, since for the remaining sites the calculated foam fractions would be negative.The letters in the legend correspond to the site identifiers given in main text Table1.These plots are rough approximations only, since the calculations ignore sorption to sludge and reactive transformation of PFAS and assume the relative surface area of the foam to be constant independent of foam fraction, which is not realistic.

2. 4
Figure SI 7: ΣPFAS EF grouped by a) treatment process and b) water type, with concentrations below the LOQ set to zero.MBBR = moving bed biofilm reactor, AS = activated sludge, EC = electrocoagulation, Ozon = ozonation, LL = landfill leachate, WW = wastewater, PW = process water, SW = stormwater runoff from landfill bottom ash collection site.Error bars represent the standard deviation (sd) within the EF for each plant (n = 3 for foamate as well as influent concentrations), but are difficult to see for all plants except H, because the sd was relatively small.The letters in the legend correspond to the site identifiers given in main text Table1.

Figure SI 8 :
Figure SI 8: ΣPFAS EF grouped by a) treatment process and b) water type, with concentrations below the LOQ set to the LOQ.MBBR = moving bed biofilm reactor, AS = activated sludge, EC = electrocoagulation, Ozon = ozonation, LL = landfill leachate, WW = wastewater, PW = process water, SW = stormwater runoff from landfill bottom ash collection site.Error bars represent the standard deviation (sd) within the EF for each plant (n = 3 for foamate as well as influent concentrations), but are difficult to see for all plants except H, because the sd was relatively small.The letters in the legend correspond to the site identifiers given in main text Table 1.

Table SI 2
: Compounds with changed calibration curve regression methods.For compounds not included in this table, all concentration points were included and a linear regression was used.
◊ PFBA: the 900 ng mL -1 point was included for site D, because PFBA concentrations were >500 ng mL -1 in some of the sample extracts of this site.○ 4:2 FTSA and NaDONA: R 2 was 0.98 for the calibration curve regression of one site (I and A, respectively) for each of these compounds.

Table SI 5
: Recoveries (%, as mean (min -max)) of Milli-Q spiked with 2.5 and 10 ng of each PFAS, and foamate samples spiked with 25 ng of each PFAS.

Table SI 7
: General chemistry of the influent for all sites.