Anion Exchange Resin and Inorganic Anion Parameter Determination for Model Validation and Evaluation of Unintended Consequences during PFAS Treatment

Abstract

When implementing anion exchange (AEX) for per- and polyfluoroalkyl substances treatment, temporal drinking water quality changes from concurrent inorganic anion (IA) removal can create unintended consequences (e.g., corrosion control impacts). To understand potential effects, four drinking water-relevant IAs (bicarbonate, chloride, sulfate, and nitrate) and three gel-type, strong-base AEX resins were evaluated. Batch binary isotherm experiments provided estimates of IA selectivity with respect to chloride ($K_{\mathrm{x}∕\mathrm{C}}$) for IA/resin combinations where bicarbonate < sulfate ≤ nitrate at studied conditions. A multi-IA batch experiment demonstrated that binary isotherm-determined $K_{\mathrm{x}∕\mathrm{C}}$ values predicted competitive behavior. Subsequent column experiments with and without natural organic matter (NOM) allowed for the validation of a new ion exchange column model (IEX-CM; https://github.com/USEPA/Water_Treatment_Models). IA breakthrough was well-simulated using binary isotherm-determined $K_{\mathrm{x}∕\mathrm{C}}$ values and was minimally impacted by NOM. Initial AEX effluent water quality changes with corrosion implications included increased chloride and decreased sulfate and bicarbonate concentrations, resulting in elevated chloride-to-sulfate mass ratios (CSMRs) and Larson ratios (LRs) and depressed pH until the complete breakthrough of the relevant IA(s). IEX-CM utility was further illustrated by simulating the treatment of low-IA source water and a change in the source water to understand the resulting duration of changes in IAs and water quality parameters.

Graphical Abstract

Graphical Abstract

INTRODUCTION

Per- and polyfluoroalkyl substances (PFAS) are a class of fluorinated, synthetic chemicals used in consumer and industrial applications due to their desirable chemical characteristics (e.g., chemical and thermal stability, surface activity, and hydro- and lipophobicity).^1-3 However, PFAS are associated with negative health effects (e.g., developmental and immune disorders, infertility, thyroid disease, organ toxicity, and cancer).^4,5 Because ingestion represents a primary PFAS exposure pathway, PFAS-impacted drinking waters are a concern.⁴ Therefore, in 2021, the United States Environmental Protection Agency announced its intention to regulate perfluorooctanoic acid (PFOA) and perfluorooctane sulfonic acid (PFOS) in drinking water,⁶ and in 2022, issued updated, interim lifetime drinking water health advisories (HAs) for PFOA (0.004 ng/L) and PFOS (0.02 ng/L) and final HAs for perfluorobutane sulfonic acid (PFBS, 2000 ng/L) and hexafluoropropylene oxide dimer acid (HFPO-DA) and its ammonium salt (“GenX chemicals,” 10 ng/L).⁷

Drinking water PFAS removal by anion exchange (AEX) has been demonstrated and often employs single-use, chloride-form, gel-type, strong-base AEX (SB-AEX) resins functionalized with quaternary ammonium.^8-12 For new resins, and depending on the PFAS and treatment goals, PFAS removal occurs over months to years prior to resin changeout.^13,14 SB-AEX resins can also exchange inorganic anions (IAs; e.g., chloride, bicarbonate, sulfate, and nitrate) and natural organic matter (NOM),^12,15-19 but this coremoval is shorter-term (e.g., hours to days)¹⁶ and is expected to be PFAS-independent because resins are PFAS-selective⁸ and PFAS concentrations are orders of magnitude lower (i.e., nequiv/L vs. mequiv/L).^20,21 During AEX treatment, IA removal is expected to cause temporal effluent water quality changes including decreased pH and alkalinity (i.e., bicarbonate removal), decreased sulfate, and increased chloride.²² Changes in source water quality brought about gradually (e.g., seasonal changes) or rapidly (e.g., alternating source inputs) could likewise lead to temporal effluent water quality changes. Water quality shifts may impact the distribution system where (i) stable alkalinity and pH are typically desired^23-25 and (ii) chloride, sulfate, and bicarbonate concentrations affect corrosion potential as indicated by the Larson ratio (LR)²⁶ for iron-based pipes (e.g., LR >0.8 is shown to increase iron release)^26-31 and the chloride-to-sulfate mass ratio (CSMR) for galvanic corrosion (e.g., CSMR >0.5 is shown to increase lead release).^22,32-36 Therefore, it is important to understand transient water quality to minimize potential unintended consequences when implementing SB-AEX for PFAS removal, and the development of an ion exchange column model (IEX-CM) to simulate effluent IA and PFAS concentrations would be beneficial.

Haupert et al.³⁷ discussed IEX-CM theory for SB-AEX resin removing IA and PFAS, noting (i) required IEX-CM parameters (i.e., selectivity and mass transfer parameters) were scarce, (ii) common drinking water IAs were often omitted, not characterized or controlled in PFAS studies, and (iii) no publicly available IEX-CM existed to simulate IA and PFAS removal. Given the large and growing number of PFAS and varied source water conditions, an IEX-CM would allow SB-AEX simulation and optimization for diverse treatment scenarios and observation of transient water quality to anticipate unintended consequences. Also, an IEX-CM would provide potential insights when traditional AEX theory fails to simulate experimental results as reported for longer chain PFAS (e.g., PFOS), nonstoichiometric AEX, and deviations from theoretical AEX capacity.^11,20,38,39 Finally, an IEX-CM would be applicable for species other than PFAS (e.g., perchlorate or nitrate removal applications).

The current work represents a first step toward IEX-CM development and validation for IA and PFAS, focusing on IAs. Therefore, the study objectives included using drinking water-relevant IAs (i.e., chloride, bicarbonate, sulfate, and nitrate) known to follow AEX theory to (i) determine IA/chloride selectivity from binary batch kinetic and isotherm experiments; (ii) validate a competitive batch AEX model with results from an IA competitive experiment; (iii) validate an IEX-CM with results from IA AEX column experiments conducted with and without NOM; and (iv) simulate practical scenarios with the validated IEX-CM, illustrating transient AEX effluent water quality.

EXPERIMENTAL METHODS

Materials.

Three single-use, gel-type, polystyrene, chloride-form, strong-base AEX resins were used (Table 1): (i) Purofine PFA694E (Purolite), (ii) DOWEX PSR2 Plus Cl (Dupont; product renamed to Amberlite PSR2 Plus in September 2021), and (iii) CalRes 2304 (Calgon Carbon). Resin conditioning and characterization are described in Supporting Section 1.1 and Tables S1 and S2. Chemical and NOM information is provided in Supporting Section 1.2.

Table 1.

Resin Properties (Per Manufacturer Unless Otherwise Noted)^a

property	unit	resin no. 1	resin no. 2	resin no. 3
manufacturer		Calgon	Purolite	DuPont
resin		CalRes 2304	Purofine PFA694E	DOWEX PSR2 Plus Clb
material		Styrene DVB	Styrene DVB	Styrene DVB
type		SB-AEX	SB-AEX	SB-AEX
type I or II		type I	na	na
matrix structure		gel	gel	gel
functional group		100% N-tri-butylamine	complex amino	tri-N-butylamine
ionic form		Cl−	Cl−	Cl−
anion exchange capacity (AEC)	Qb (mequiv/mL packed bed), ±SD	0.65 (min)	0.5–0.9c	0.7 (min)
		0.77 ± 0.01d	0.86 ± 0.00d	0.76 ± 0.00d
		0.72 ± 0.01e	0.82 ± 0.00e	0.71 ± 0.01e
		0.74 ± 0.03f	0.84 ± 0.02f	0.73 ± 0.03f
	Qm (mequiv/g dry resin), ±SD	1.93 ± 0.02d	2.18 ± 0.01d	1.64 ± 0.01d
		1.78 ± 0.02e	2.08 ± 0.01e	1.54 ± 0.02e
		1.85 ± 0.08f	2.13 ± 0.06f	1.59 ± 0.06f
	QT (mequiv/L swollen resin), ±SD	1237 ± 13d	1498 ± 6d	1239 ± 7d
		1143 ± 15e	1429 ± 6e	1164 ± 15e
		1190 ± 53f	1464 ± 38f	1201 ± 42f
specific volume (Vsp)	g dry resin/mL packed bed, ±SD	0.35–0.45g	0.33–0.43g	0.45–0.52g
		0.40 ± 0.00h	0.40 ± 0.01h	0.46 ± 0.00h
water retention capacity (WRC)	%, ±SD	37–47	38–50	25–35
		40.1 ± 0.4h	38.8 ± 0.2h	28.6 ± 0.8h
bed porosity (εb)		na	0.33–0.38g	0.36g
		0.37 ± 0.01i	0.42 ± 0.01i	0.39 ± 0.01i
bulk density (ρb)	g swollen resin/mL packed bed	0.66–0.72	0.65–0.70	0.69
		0.67j	0.65j	0.64j
particle density (ρp)	g swollen resin/mL swollen resin	na	1.05	1.07
		1.07j	1.12j	1.06j
apparent resin density (ρa)	g dry resin/mL swollen resin	na	0.49–0.69g	0.7–0.81g
		0.64j	0.69j	0.75j
whole bead	% (min)	95	na	na
large fraction	% (max)	5 (>1.25 mm)	na	na
small fraction	% (max)	1 (<0.32 mm)	na	1 (<0.3 mm)
uniformity coefficient		1.7	1.3	1.1
mean diameter (dp)	μm	750k	675 ± 75	700 ± 50
maximum temperature	C	na	100 (Cl− form)	na

^aDVB: divinyl benzene; SB-AEX: strong-base anion; na: not available; SD: standard deviation.

^bProduct renamed to Amberlite PSR2 Plus in September 2021.

^cPersonal communication with Purolite.

^dCurrent research, sulfate elution basis (Tripp et al.²¹).

^eCurrent research, nitrate elution basis (modification to Tripp et al.²¹).

^fCurrent research, average of all sulfate (n = 3) and nitrate (n = 3) results.

^gCalculated from manufacturer-provided data (Table S1).

^hCurrent research (Tripp et al.²¹).

ⁱCurrent research.

^jCalculated from current research data (Table S1).

^kCalculated from data obtained from personal communication with Calgon (Table S2).

Analytics.

Chloride (Cl⁻), nitrate (NO₃⁻), and sulfate (SO₄²⁻) concentrations were determined by ion chromatography (IC) per EPA Method 300.1.⁴⁰ Temperature and pH were determined per Standard Methods 2550 and 4500-H+ B, respectively.⁴¹ Total inorganic carbon (TIC) was determined using a Shimadzu TOC-VCPH Total Organic Carbon Analyzer per manufacturer’s instructions,⁴² using the Inorganic Carbon (IC) method in the TOC-Control V software.⁴² Bicarbonate (HCO₃⁻) concentrations were calculated using pH, TIC concentration, and carbonate system pK_a values (at 25 °C: pK_a1 = 6.352; pK_a2 = 10.329).⁴³ Total organic carbon concentration (TOC; mg C/L), UV₂₅₄ absorbance (UVA₂₅₄; cm⁻¹), and specific UV-absorbance (SUVA; L/mg·M) of NOM-containing samples were determined per EPA Method 415.3.⁴⁴ Duplicates were analyzed for 10% of aliquoted samples.

Batch Experiments.

Two types of batch experiments (BEs) were conducted (Table 2): screening kinetic experiments (KEs) and detailed isotherm experiments (IEs). Experiments for nitrate, sulfate, and bicarbonate are denoted by “N,” “S,” and “B,” respectively. Binary KEs (i.e., anion and chloride; KE-N, KE-S, and KE-B) established a minimum time-to-equilibrium per anion and provided initial anion selectivity estimates for the IE design. Binary IEs (IE-N, IE-S-1, IE-S-2, IE-S-3, and IE-B) were then conducted to obtain anion selectivity estimates for each resin. In addition, a competitive, multisolute IE (IE-ALL; Table S3) was conducted to observe competitive interactions among the anions and evaluate the ability of the binary IE-determined selectivity to simulate the competitive IE results. Detailed information for the BEs is provided in Supporting Section 1.3.

Table 2.

Initial Condition Summary for Batch Kinetic (KE) and Isotherm (IE) Experiments^a

			design initial concentrations		measured pH
experiment	anion(s)	length (days)	chloride (mM)	anion(s)	initial	final	design resin mass (g)
KE-N	nitrate	7	5.0	8 mg N/L	6.6–6.9	5.2–6.7	1.1
KE-S	sulfate	7	2.0	27 mg SO42−/L	6.6–7.0	5.3–6.7	1.0
KE-B	bicarbonate	7	1.0	16 mg C/L	6.1–8.3	4.8–7.6	1.0
IE-N	nitrate	3	5.0	5–15 mg N/L	6.6–6.9	4.8–6.9	0.22–2.0
IE-S-1	sulfate	4	2.0	27–54 mg SO42−/L	5.9–7.5	3.7–7.8	0.35–2.0
IE-S-2	sulfate	4	5.0	27–54 mg SO42−/L	6.3–7.3	3.7–7.3	0.5–5.0
IE-S-3	sulfate	7	2.0	175–275 mg SO42−/L	6.3–6.7	3.3–4.4	0.5–2.8
IE-B	bicarbonate	5	1.0	1.9–39.1 mg C/L	5.9–8.4	5.7–7.7	2.0
	bicarbonate			45 mg C/L
IE-ALL	nitrate	≥7	2.0 or 5.0	6 mg N/L	6.7–8.4	2.6–8.4	1.0–4.0
	sulfate			150 mg SO42−/L

^aAll experiments were conducted at 25 °C in a controlled temperature chamber where bottles were rotated at 30 rpm.

Column Experiments.

Lead-lag column trains (two columns in series) with overall nominal empty bed contact times (EBCTs) of 2 and 4 min (2-min EBCT per column) were operated for column experiments (CEs). Four trains were constructed, with duplicate trains (“replicate 1” and “replicate 2”) dedicated for use with and without NOM. Figure S1 provides a schematic of a CE train, while Table S4 provides detailed CE operational information. The CE sampling scheme allowed for three sets of column performance data to be obtained per train: “column 1” (column 1 influent and effluent), “column 2” (column 1 effluent/column 2 influent and column 2 effluent), and “columns 1 + 2” (column 1 influent and column 2 effluent). Four different CE influent conditions were investigated (Table S5): low (CE-Low), medium (CE-Med), and high (CE-High) anion concentrations and medium anion concentration with NOM addition (CE-Med+NOM). A single resin, Purofine PFA694E, was selected for CEs based on its narrow range of particle size and representative behavior in the BEs. Additional details for the CEs are provided in Supporting Section 1.4.

Selectivity Determination.

For the exchange of two ionic species, A and B, with charges ${\zeta }_{{}_{\mathrm{A}}}$ and ${\zeta }_{{}_{\mathrm{B}}}$, respectively, equilibrium between the resin (denoted by overbar) and liquid phases can be described by the law of mass action (eq 1).⁴⁵

$${\zeta }_{{}_{\mathrm{B}}}\overline{\mathrm{A}}+{\zeta }_{\mathrm{A}}\mathrm{B}\rightleftharpoons {\zeta }_{{}_{\mathrm{B}}}\mathrm{A}+{\zeta }_{{}_{\mathrm{A}}}\overline{\mathrm{B}}$$ (1)

Applying eq 1 and assuming chemical activities are equivalent to concentrations (a common assumption for low ionic strength matrices)³⁹ results in the definition for selectivity of ion B with respect to ion A ($K_{{}_{\mathrm{B}∕\mathrm{A}}}$; eq 2). In eq 2, $q$ is the ionic equivalent concentration per resin-phase volume (mequiv ion/L of swollen (hydrated) resin) and C is the ionic equivalent concentration per liquid phase volume (mequiv ion/L of water).⁴⁶

$$K_{{}_{\mathrm{B}∕\mathrm{A}}}={\left(\frac{C_{{}_{\mathrm{A}}}}{q_{{}_{\mathrm{A}}}}\right)}^{{\zeta }_{{}_{\mathrm{B}}}}{\left(\frac{q_{{}_{\mathrm{B}}}}{C_{{}_{\mathrm{B}}}}\right)}^{{\zeta }_{{}_{\mathrm{A}}}}$$ (2)

Using eq 2 and mass balances on the resin phase, liquid phase, and replacing anion, three batch ion exchange equilibrium equations were derived: (i) a monovalent ion (nitrate) replacing a monovalent ion (chloride), (ii) a monovalent ion with acid–base properties (bicarbonate) replacing a monovalent ion (chloride), and (iii) a divalent ion (sulfate) replacing a monovalent ion (chloride). Details of the derived equations for BEs are provided in Supporting Section 1.5 and Tables S6-S8. In the current work, $K_{{}_{\mathrm{B}∕\mathrm{A}}}=K_{{}_{\mathrm{x}∕\mathrm{C}}}$ where x represents the replacing anion (N for nitrate, S for sulfate, or B for bicarbonate) and C represents chloride.

The only unknown in each batch ion exchange equilibrium equation was $K_{\mathrm{x}∕\mathrm{C}}$; therefore, BE data allowed $K_{\mathrm{x}∕\mathrm{C}}$ estimation for each anion. $K_{\mathrm{x}∕\mathrm{C}}$ estimation was accomplished in R, a free statistical computing and graphics software environment.⁴⁷ In R, the function optimize() was used to estimate $K_{\mathrm{x}∕\mathrm{C}}$ by minimizing the weighted residual sum of squares (WRSS) between measured and simulated anion liquid concentrations, weighting by the inverse square of the measured anion liquid concentrations, resulting in a dimensionless WRSS.⁴⁸ This procedure was used to prevent greater anion liquid concentrations from biasing parameter estimates.⁴⁹ To estimate $K_{\mathrm{x}∕\mathrm{C}}$ uncertainty, bootstrapping with sample replacement (n = 10 000) was used.

Homogeneous Surface Diffusion Model.

The homogeneous surface diffusion model (HSDM) includes both film and surface diffusion mass transfer limitations, and an HSDM batch implementation in R and the IEX-CM (https://github.com/USEPA/Water_Treatment_Models) were used to interpret KEs and CEs, respectively.^37,47 Supporting Section 1.6 details HSDM derivation and implementation.

RESULTS AND DISCUSSION

Resin Characterization.

Required resin properties were determined (Table 1) to design and analyze KEs, IEs, and CEs. Measured properties agreed with manufacturer-reported properties, except where an 11% greater bed porosity (${\epsilon }_{\mathrm{b}}$) was measured for PFA694E (0.42 vs. 0.33–0.38 manufacturer-reported). For a given resin, nitrate- and sulfate-based elution anion exchange capacities (AECs) on a mass basis (Q_m) were similar (4.7–7.6% relative percent difference (RPD)); therefore, the average of the nitrate and sulfate elution AECs was used. The Q_m (mequiv/g dry resin) trend was PSR2 Plus Cl (1.59) < CalRes 2304 (1.85) < PFA694E (2.13).

CEs and associated IEX-CM modeling required additional parameters. The specific volume (V_sp) was measured and used to obtain average AECs on a packed bed basis (Q_b, mequiv/mL packed bed) that slightly changed the trend seen with Q_m such that Q_b for PSR2 Plus Cl (0.73) ≈ CalRes 2304 (0.74) < PFA694E (0.84), and all Q_b exceeded manufacturer-reported minimums (Table 1). The observed trend in water retention capacity (WRC) was PSR2 Plus Cl (28.6) < PFA694E (38.8) < CalRes 2304 (40.1), and there was more variability in WRC than other measured properties (3.3–33% RPD). Apparent resin density (ρ_a) and ε_b were also determined, and both exhibited trends of CalRes 2304 < PSR2 Plus Cl < PFA694E (Table 1).

Batch Binary Kinetic Experiments.

From the KEs, liquid concentration profiles were used to assess when equilibrium was reached (Figure S2). Equilibrium was obtained by the first sample point (1 day) for sulfate and bicarbonate and the second sample point (2 days) for nitrate, as evidenced by the constant liquid concentrations for the remainder of the KE (i.e., 7 days). IE runtimes were set to meet or exceed the determined time-to-equilibrium (i.e., 2 days, Table 2). Initial $K_{\mathrm{x}∕\mathrm{C}}$ estimates and their uncertainty (Table 3 and Figure S3) were obtained from KEs to design IEs. Further discussion of $K_{\mathrm{x}∕\mathrm{C}}$ estimates is provided along with those determined from IEs in the following section.

Table 3.

Anion Selectivity with Respect to Chloride (${\mathit{K}}_{\mathbf{x}∕\mathbf{C}}$) Determined from Batch Binary Kinetic (KE) and Isotherm (IE) Experiments^a

		anion (x) selectivity with respect to chloride (C), $K_{\mathrm{x}∕\mathrm{C}}$b
		PSR2 Plus Cl		Purofine PFA694E		CalRes 2304
anion	experiment	$K_{\mathrm{x}∕\mathrm{C}}\pm \text{SD}$	n	$K_{\mathrm{x}∕\mathrm{C}}\pm \text{SD}$	n	$K_{\mathrm{x}∕\mathrm{C}}\pm \text{SD}$	n
nitrate	KE-N	17 ± 0.1	9	13 ± 0.1	9	13 ± 0.1	9
	IE-N	16 ± 0.1	26	13 ± 0.2	27	13 ± 0.2	26
sulfate	KE-S	0.017 ± 0.001	12	0.027 ± 0.001	12	0.037 ± 0.001	12
	IE-Sc	0.018 ± 0.001	80	0.025 ± 0.001	77	0.033 ± 0.001	80
bicarbonate	KE-B	0.28 ± 0.008	9	0.32 ± 0.005	9	0.32 ± 0.003	9
	IE-B	0.31 ± 0.006	25	0.35 ± 0.007	26	0.31 ± 0.005	27

^an = number of observations; SD = standard deviation.

^b$K_{\mathrm{x}∕\mathrm{C}}$ is dimensionless for monovalent ions (e.g., nitrate and bicarbonate) and has units of L swollen resin/L liquid for divalent ions (e.g., sulfate).

^cOverall selectivity estimate for all sulfate IEs.

Batch Binary Isotherm Experiments.

IEs were conducted for nitrate, sulfate, and bicarbonate with chloride as the background anion (Table 2). A single $K_{\mathrm{x}∕\mathrm{C}}$ (and associated uncertainty) was determined for each anion and resin combination ($K_{\mathrm{x}∕\mathrm{C}}$, Table 3 and Figure S4). Figure 1A shows measured vs. simulated anion liquid concentrations against a 1:1 line, demonstrating that a single $K_{\mathrm{x}∕\mathrm{C}}$ represented data for a given resin and anion pair. The agreement (Figure 1A) supports that the law of mass action was able to describe AEX behavior and a single $K_{\mathrm{x}∕\mathrm{C}}$ is appropriate and applicable throughout the anion concentration range investigated and equilibrium was achieved. Figure 1B summarizes IE results as anion liquid phase vs. resin-phase equivalent fractions for each resin and anion combination. Based on the convex upward shape of the equivalent fraction curves, which represents greater resin loading with regard to liquid concentration, nitrate and sulfate are favorable isotherms; the concave downward shape of the bicarbonate equivalent fraction curve is, conversely, an unfavorable isotherm.⁵⁰

Figure 1.

Batch binary isotherm experiment comparison of (A) measured vs. simulated equilibrium liquid concentrations with the 1:1 dashed correlation line and (B) equivalent fraction plots.

KE- and IE-determined $K_{\mathrm{x}∕\mathrm{C}}$ estimates were similar (absolute percent differences (APD) <12%, Tables 3 and S9), but t-tests revealed significant differences (p < 0.05) between KE- and IE-determined $K_{\mathrm{x}∕\mathrm{C}}$ for many of the anion/resin combinations (Table S9). Overall, initial $K_{\mathrm{x}∕\mathrm{C}}$ estimates obtained from KEs were useful in designing IEs, but detailed IEs were still necessary to cover a broader range of IA concentrations. Tukey’s multiple comparison test⁵¹ was used to compare $K_{\mathrm{x}∕\mathrm{C}}$ among resins for the IAs, and $K_{\mathrm{x}∕\mathrm{C}}$ were significantly different among all but two resin pairs (i.e., $K_{{}_{\mathrm{N}∕\mathrm{C}}}$ PFA694E:CalRes 2304 and $K_{{}_{\mathrm{B}∕\mathrm{C}}}$ PSR2 Plus Cl:PFA694E; α-value = 0.05; Table S10).

When comparing individual anions, $K_{\mathrm{x}∕\mathrm{C}}$ are not directly comparable between monovalent (nitrate and bicarbonate) and divalent (sulfate) anions because anion charges lead to a difference in $K_{\mathrm{x}∕\mathrm{C}}$ units (dimensionless and L swollen resin/L liquid, respectively). For a direct comparison of sulfate to nitrate and bicarbonate, a separation factor for sulfate (${\beta }_{{}_{\mathrm{S}∕\mathrm{C}}}$) can be calculated (eq 3) from $K_{{}_{\mathrm{S}∕\mathrm{C}}}$, chloride resin loading ($q_{{}_{\mathrm{C}}}$), and chloride liquid concentration ($C_{{}_{\mathrm{C}}}$).

$${\beta }_{{}_{\mathrm{S}∕\mathrm{C}}}=K_{{}_{\mathrm{S}∕\mathrm{C}}}\frac{q_{{}_{\mathrm{C}}}}{C_{{}_{\mathrm{C}}}}$$ (3)

Because ${\beta }_{{}_{\mathrm{S}∕\mathrm{C}}}$ depends on $C_{{}_{\mathrm{C}}}$ and $q_{{}_{\mathrm{C}}}$, ${\beta }_{{}_{\mathrm{S}∕\mathrm{C}}}$ is not constant and can only be determined for specific conditions. Using the studied conditions, a ${\beta }_{{}_{\mathrm{S}∕\mathrm{C}}}$ range of 2.6–13 was calculated for all experiments and resins. For monovalent anions, ${\beta }_{\mathrm{x}∕\mathrm{C}}=K_{\mathrm{x}∕\mathrm{C}}$. Reasonable agreement—within 1 order of magnitude for all IAs—was observed with separation factors previously reported for a gel-based, chloride-form, SB-AEX polystyrene resin (${\beta }_{{}_{\mathrm{N}∕\mathrm{C}}}=2$, ${\beta }_{{}_{\mathrm{S}∕\mathrm{C}}}=10$, ${\beta }_{{}_{\mathrm{S}∕\mathrm{C}}}=0.5$).²¹ To summarize, $K_{{}_{\mathrm{N}∕\mathrm{C}}}>{\beta }_{{}_{\mathrm{S}∕\mathrm{C}}}>K_{{}_{\mathrm{B}∕\mathrm{C}}}$ for $q_{{}_{\mathrm{C}}}∕C_{{}_{\mathrm{C}}}$ less than 394 (CalRes 2304), 520 (PFA694E), or 889 (PSR2 Plus Cl). The greater determined selectivity of nitrate compared to sulfate is consistent with previous studies^52-54 on nitrate-selective resins where increasing functional group chain length (i.e., triethylamine > trimethylamine), and therefore greater distance between exchange sites, resulted in a preference for monovalent vs. divalent ions. The tributylamine resins used in this study could be expected to continue this trend, with an even stronger preference for monovalent ions (including many PFAS).

Batch Competitive Isotherm Experiment.

A competitive IE (IE-ALL) was conducted under a variety of initial conditions (Tables 2 and S3) to validate that using binary IE estimated $K_{\mathrm{x}∕\mathrm{C}}$ (Table 3) in a competitive isotherm model (Table S8) could simulate equilibrium anion concentrations. Figure 2A summarizes each anion’s initial concentration, equilibrium measured concentration, and competitive isotherm model-simulated concentration. Boxplots of RPDs between measured and model-simulated equilibrium anion concentrations show that the interquartile range is within ±20% for the anions on all three resins and within ±15% for CalRes 2304 and PFA694E (Figure 2B). Overall, the competitive isotherm solution (Table S8) reasonably simulated anion behavior on each resin in a multisolute system, supporting the use of $K_{\mathrm{x}∕\mathrm{C}}$ determined from binary IEs in multisolute systems.

Figure 2.

Batch competitive isotherm simulated vs. measured equilibrium anion liquid concentrations for each of the 15 initial conditions (A) and boxplot summary of the relative percent difference (RPD) between measured and simulated anion liquid concentrations (B). “TIC” = total inorganic carbon.

Column Experiments.

CEs were conducted to evaluate IA breakthrough behavior from packed bed columns, quantify associated water quality changes, investigate potential NOM impacts on IA breakthrough, and provide data for IEX-CM validation. As previously described, the CE setup (Figure S1) allowed for three data sets from a single-column train: column 1, column 2, and columns 1 + 2. Effluent concentrations were normalized by the mean influent concentration (Table S5), generating influent-normalized effluent concentrations ($C∕C_{{}_0}$). The breakthrough behavior of bicarbonate, sulfate, and nitrate is summarized as bed volumes (BVs) treated at 0.1, 0.5, and 0.9 $C∕C_{{}_0}$ (Tables 4, S11, and S12). The observed order of breakthrough was bicarbonate, sulfate, and nitrate, in agreement with the relative IE-determined $K_{\mathrm{x}∕\mathrm{C}}$ (or ${\beta }_{\mathrm{x}∕\mathrm{C}}$).

Table 4.

Summary of Measured vs. IEX-CM-Simulated Breakthrough and NOM Effects from Column 1 and Columns 1 + 2 for Medium Condition Column Experiments (CE-Med and CE-Med+NOM)^a

		column 1							columns 1 + 2
		experimental BV ± SD		IEX-CM BV ± SD		RPD (%)			experimental BV ± SD		IEX-CM BV ± SD		RPD (%)
anion	$C∕C_{{}_0}$	−NOM	+NOM	−NOM	+NOM	−NOM vs +NOM	−NOM IEX-CM vs exp.	+NOM IEX-CM vs exp.	−NOM	+NOM	−NOM	+NOM	−NOM vs +NOM	−NOM IEX-CM vs exp.	+NOM IEX-CM vs exp.
bicarbonate	0.1	10 ± 0.36	11 ± 0.068	13 ± 0.53	12 ± 0.068	6	22	7.6	13 ± 0.26	13 ± 0.93	17 ± 0.68	16 ± 0.054	2.2	32	24
	0.5	23 ± 1.3	24 ± 0.36	30 ± 0.98	29 ± 0.15	2.3	26	19	22 ± 0.39	22 ± 0.41	32 ± 1.1	31 ± 0.1	−0.4	38	35
	0.9	39 ± 3.2	41 ± 0.77	53 ± 1.1	53 ± 0.45	4	30	25	33 ± 1.2	35 ± 0.2	49 ± 1.4	49 ± 0.31	5.2	38	32
sulfate	0.1	120 ± 6.1	120 ± 3.3	130 ± 8	130 ± 0.48	−1.5	14	9	110 ± 14	110 ± 1.3	150 ± 8.5	140 ± 0.46	−5.6	26	27
	0.5	140 ± 4.6	150 ± 1.6	160 ± 8.4	150 ± 0.44	4.4	11	1.8	130 ± 5.8	130 ± 1.8	160 ± 8.6	150 ± 0.46	0.63	20	15
	0.9	160 ± 7.9	170 ± 1.2	180 ± 8.4	170 ± 0.56	2.2	7.9	1.8	140 ± 4.7	140 ± 0.93	170 ± 8.7	160 ± 0.41	2.1	19	13
nitrate	0.1	590 ± 17	560 ± 33	480 ± 15	460 ± 0.56	−4	−21	−21	630 ± 19	630 ± 12	540 ± 19	530 ± 1.5	−0.27	−15	−17
	0.5	740 ± 19	750 ± 3.2	650 ± 24	630 ± 1.6	2.3	−13	−18	700 ± 6.2	700 ± 4.3	650 ± 25	640 ± 1.3	0.063	−7.1	−9.5
	0.9	880 ± 46	910 ± 23	830 ± 40	810 ± 4.6	2.4	−6.6	−11	790 ± 20	790 ± 19	750 ± 31	740 ± 4.7	0.39	−4.4	−6.5

^aBV ± SD = bed volumes ± standard deviation. −NOM = no natural organic matter. +NOM = with natural organic matter. RPD = relative percent difference between means $(x_1\text{and}{x}_2)=(x_1-x_2)∕[(x_1+x_2)∕2]\times 100$.

For the CEs, breakthrough curves were generated for $C∕C_{{}_0}$ vs. BVs treated to visualize the IA breakthrough (Figures 3 and S5-S7). Using column 1 from CE-Med as an example (Figure 3), the following trends in chromatographic effects—i.e., changes in effluent concentration with respect to influent concentration—were observed and were consistent among CEs: (i) bicarbonate removal released chloride (BVs ~10–39); (ii) sulfate removal released chloride and produced a bicarbonate chromatographic roll-off effect ($C∕C_{{}_0}>1.0$; BVs ~120 to 160); (iii) nitrate breakthrough showed a longer initial lag phase and more gradual breakthrough curve than sulfate or bicarbonate, and full breakthrough coincided with sulfate plateauing at its influent concentration after a prolonged roll-off period (BVs ~590 to 880); and (iv) chloride effluent concentrations were initially elevated (2.0–2.6-fold) and then descended toward its influent concentration with the breakthrough of each subsequent IA (BVs ~0 to 250).

Figure 3.

Medium condition column experiment (CE-Med) anion breakthrough curves.

A breakthrough occurred earlier with greater influent anion concentrations, and the longest time to initial breakthrough (defined as $C∕C_{{}_0}\ge 1.0$) was observed from CE-Low nitrate (~1400 BVs, Table S11). By comparison, PFAS are present at concentration orders of magnitude lower than IAs and are much more selective (e.g., $K_{{}_{\text{PFOA}∕\mathrm{C}}}={10}^{4.36}$),³⁹ so it is expected that PFAS of interest (such as PFOA) will break through much later than IAs. In fact, studies have reported a breakthrough in the 10⁶ BVs range for PFOA in groundwater on PFA694E.⁵⁵

NOM AEX behavior was observed from CE-Med+NOM results and was characterized by UVA₂₅₄, TOC, and SUVA breakthrough data (Figure S8). UVA₂₅₄ breakthrough reached 0.83–0.84 $C∕C_{{}_0}$ for column 1 and 0.76–0.86 $C∕C_{{}_0}$ for columns 1 + 2 by the experiment end (1383 and 993 BVs, respectively). TOC breakthrough reached 0.75–0.82 $C∕C_{{}_0}$ for column 1 and 0.63–0.69 $C∕C_{{}_0}$ for columns 1 + 2 by the end of the experiment. The majority of UVA₂₅₄ and TOC breakthroughs occurred within the first ~250 BVs, after which the rate slowed (Figure S8). SUVA breakthrough peaked at ~250 BVs (Figure S8) and then stabilized at 1.2–1.3 $C∕C_{{}_0}$ for column 1 and 1.3–1.4 $C∕C_{{}_0}$ for columns 1 + 2. The differences in the removal of UVA₂₅₄ vs. TOC as well as in rates of removal over the course of treatment could be attributed to the heterogeneous nature of NOM, with different fractions exhibiting distinct properties (e.g., hydrophobicity, charge density, and molecular weight) affecting removal by AEX resin.^15,56-59 The UVA₂₅₄-absorbing NOM fraction is attributed to acidic, aromatic substances and is associated with disinfection byproduct (DBP) formation.^58,60 The relatively greater UVA₂₅₄ breakthrough observed indicates that this fraction was poorly removed compared to overall TOC, corresponding with increased SUVA.

Comparison of CE-Med and CE-Med+NOM results (Table 4 and Figures 3 and S7) allowed evaluation of NOM effects on IA breakthrough. IA breakthrough RPD between CE-Med and CE-Med+NOM was within 10% for all IAs (Table 4), indicating that NOM minimally impacted IA breakthrough and may not need to be considered in related modeling efforts involving IA concentrations and NOM compositions similar to those used in this study. PFAS, however, may be affected by NOM quite differently due to their greater molecular size and much longer anticipated breakthrough times as compared to IAs.

The CEs allowed assessment of transient CSMR and LR, commonly interpreted as corrosion indicators, and pH, which affects corrosion, DBP formation, and chloramine chemistry (Table S13 and Figures S9 and S10).^60,61 CSMR and LR calculations are described in Supporting Section 2.4. CSMR was elevated at AEX column startup prior to returning to influent values (Figure S9), corroborating previous research.^22,34 With increasing influent IA concentrations, CSMR stabilization time decreased (e.g., column 1: 20 h CE-Low and 3.4 h CE-High). NOM was not observed to impact CSMR stabilization time (e.g., column 1: 6.5 h for CE-Med and CE-Med-NOM). LR changes post-AEX startup appear as three distinct regions in Figure S9: (i) LR elevated above influent values resulting from decreased alkalinity during bicarbonate removal (e.g., CE-Low column 1: <110 BVs, <3.7 h); (ii) LR less than influent values due to decreasing chloride concentrations, sulfate removal, and bicarbonate post-breakthrough (e.g., CE-Low column 1: 110–570 BVs, 3.7–20 h); (iii) LR at influent values as sulfate and chloride concentrations (and therefore CSMR) stabilized (e.g., CE-Low column 1: >570 BVs, >20 h). LR stabilization times were not determined because effluent TIC sampling concluded upon full bicarbonate breakthrough. Effluent pH was initially depressed below influent pH (Table S13 and Figure S10), and the pH impact was inversely related to influent bicarbonate concentrations for both observed pH percent decrease (e.g., 33.1% CE-Low and 7.1–16.3% CE-High) and pH decrease duration (e.g., column 1: 8 h CE-Low and 0.55 h CE-High). As expected, pH trends directly corresponded with bicarbonate (i.e., TIC) removal (Figure S10). The importance of understanding the magnitude and duration of changes in water quality constituents (IAs) is discussed further under Practical Implications.

Column Model Validation.

IEX-CM validation used experimental IA concentrations. Supporting Section 2.5, Tables S14-S17, and Figures S11 and S12 detail IEX-CM inputs used to generate simulated IA concentrations. Nitrate resin-phase diffusion (D_s; 2.0 × 10⁻⁷ cm²/s) and film transfer (k_L; 4.52 × 10⁻³–4.62 × 10⁻³ cm/s) coefficient estimates were used for simulations (Supporting Sections 2.5.1 and 2.5.2, respectively). Simulated effluent IA concentrations were normalized by the CE mean influent IA concentration (Table S5) to obtain $C∕C_{{}_0}$, and simulated results are overlaid with experimental data in Figures 3 and S5-S7. Breakthrough curve visual evaluation revealed a high level of agreement between experimental data and IEX-CM simulations; therefore, no parameter optimization was performed. For each CE, Tables 4, S11, and S12 compare the RPD for BVs-to-breakthrough between experimental data and IEX-CM simulations. BVs at specific $C∕C_{{}_0}$ (0.1, 0.5, and 0.9) were calculated via linear interpolation between adjacent points.

IEX-CM simulations tended to deviate more from experimental data for lower $C∕C_{{}_0}$ (e.g., CE-Med column 1: sulfate RPD 14% at 0.1 $C∕C_{{}_0}$ and 7.9% at 0.9 $C∕C_{{}_0}$). Compared to the CE data, the IEX-CM tended to simulate a later breakthrough for bicarbonate and sulfate and an earlier breakthrough for nitrate. Across CEs, model agreement ranged from 1.0–30% RPD for sulfate and −35 to 10% RPD for nitrate. Bicarbonate had the greatest deviation from the CE data, with RPDs up to 46% (CE-Low columns 1 + 2). A current IEX-CM limitation is that it does not currently account for pH changes and bicarbonate is an acid/base species; however, the IEX-CM is still useful for simulating bicarbonate breakthrough as the absolute differences between measured and simulated BVs (≤31 BVs) translated to relatively short durations for the studied conditions (≤1.0–2.2 h). With few exceptions, a better agreement between data and IEX-CM simulations was obtained by simulating each column independently (i.e., column 1 or column 2) rather than as a combined column (i.e., columns 1 + 2; e.g., CE-Low nitrate: independently −11 to −1.1% RPD vs. combined −27 to −9.5% RPD across $C∕C_{{}_0}$ range, see Tables 4 and S12). IEX-CM simulations improved for some conditions with increasing influent concentrations (e.g., columns 1 + 2 nitrate RPDs at 0.1 $C∕C_{{}_0}:-11\%$ (CE-Low), −6.2% (CE-Med), and −1.0% (CE-High)), although no clear pattern was observed. Similar RPDs were observed for CE-Med and CE-Med+NOM data vs. IEX-CM simulations (CE-Med −21 to 38% RPD and CE-Med +NOM −21 to 35% RPD; Table 4).

IEX-CM simulation sensitivity was assessed by varying important input parameters from their baseline values (Figure S13): $K_{{}_{\mathrm{N}∕\mathrm{C}}}(\pm 3)$, $D_{{}_{\mathrm{s}}}(\pm 10\times )$, and $k_{{}_{\mathrm{L}}}(\pm 5\times )$. A $K_{{}_{\mathrm{N}∕\mathrm{C}}}$ decrease shifted the simulated nitrate breakthrough earlier, with a $K_{{}_{\mathrm{N}∕\mathrm{C}}}$ increase producing the opposite effect. For both $D_{{}_{\mathrm{s}}}$ and $k_{{}_{\mathrm{L}}}$, an increase created a steeper breakthrough curve slope, whereas a decrease created a more gradual slope. The IEX-CM was sensitive to decreases in either $k_{{}_{\mathrm{L}}}$ or $D_{{}_{\mathrm{s}}}$ but was not sensitive to increases, suggesting a mixed diffusion regime for nitrate that was supported by calculated CE Biot numbers (the ratio of the liquid film phase to intraparticle mass transfer rates)⁶² between 0.8 and 3.8 (see Supporting Section 2.5.3and Table S18).⁴⁶ Overall, the IEX-CM performed as intended for simulating IA concentrations in a multisolute system without including considerations for NOM.

Practical Implications.

To demonstrate IEX-CM utility, two practical scenarios were simulated to investigate source water IA concentration impacts on AEX-treated effluent water quality: (i) initial AEX startup for a low-IA source water and extent of impacts to corrosion parameters (i.e., CSMR, LR, and pH) and (ii) a sudden shift in influent sulfate concentrations post-IA breakthrough from an abrupt change in the water source and its potential influence on nitrate maximum contaminant level (MCL, 10 mg N/L) exceedance. An EBCT of 4 min was used for both scenarios.

For scenario one, the U.S. Geological Survey (USGS) source water IA concentration data (1/9/2017–3/1/2022)⁶³ was obtained and averaged for the Susquehanna River at Conowingo, MD, to provide IEX-CM influent concentrations (Table S19). The Susquehanna River was chosen for its relatively low source water IA concentrations (0.57 ± 0.18 mequiv/L chloride, 0.97 ± 0.23 mequiv/L bicarbonate, 0.48 ± 0.14 mequiv/L sulfate, and 0.08 ± 0.02 mequiv/L nitrate from USGS parameter codes 00940, 00453, 00945, and 00618, respectively). Additional required IEX-CM input parameters were $K_{\mathrm{x}∕\mathrm{C}}$ (Table 3) and system specifications (CE-Med columns 1 + 2 values, Tables S4 and S16). Concentrations and breakthrough times are summarized in Table S19 and visualized in Figure S14. Effluent concentrations were simulated to reach 0.9 $C∕C_{{}_0}$ at 18, 82, and 166 h (i.e., 280, 1200, and 2500 BVs) for bicarbonate, sulfate, and nitrate, respectively, which may serve as approximate timeframes for the effects of decreased pH (280 BVs) and increased CSMR (>1200 BVs) and LR (~0 to 300 and ~1200 to 2500 BVs) from changes in relevant IA concentrations (Figure 4).

Figure 4.

IEX-CM-simulated effluent chloride-to-sulfate mass ratio (CSMR) and Larson Ratio (LR) for Susquehanna River at Conowingo, MD, source water, 4-min EBCT.

For scenario two, the IEX-CM simulated a sudden change in source water quality (Figure S15). Baseline influent conditions were 7.05 mequiv/L chloride, 6.24 mequiv/L bicarbonate, 0.71 mequiv/L nitrate (i.e., MCL),⁶⁴ and 1.04 mequiv/L sulfate. Additional IEX-CM input parameters were $K_{\mathrm{x}∕\mathrm{C}}$ (Table 3) and system specifications (CE-Med columns 1 + 2 values, Tables S4 and S16). All IAs were allowed to stabilize (full breakthrough, ~1000 BVs), at which point influent sulfate concentration was doubled to 2.08 mequiv/L, resulting in increased effluent nitrate concentrations (i.e., >MCL, maximum 0.81 mequiv/L) for ~400 BVs.

Given that the ultimate application of the studied AEX resins could be for PFAS removal and that PFAS-dictated resin changeout will likely be on the order of months to years understanding the magnitude and duration of transient water quality changes associated with IA removal may inform the timeframe for necessary mitigation strategies, such as wasting treated effluent, blending of treated and untreated effluents, or strategies for staged changeout of parallel or series AEX treatment trains. Buffered resins (for example, resins initially loaded with bicarbonate) are another newly available approach for targeting stable effluent pH and CSMR.^52,65 The two scenarios illustrated how the validated IEX-CM, along with $K_{\mathrm{x}∕\mathrm{C}}$ determined in the current research, provide a useful tool to evaluate the extent of potential unintended consequences to water quality during AEX treatment.

CONCLUSIONS

The current research developed required input parameters for and validated the open-source IEX-CM to simulate drinking water IA removal, allowing future extension to PFAS removal. Initially, three AEX resins (CalRes 2304, Purofine PFA694E, and PSR2 Plus Cl) were characterized and required resin properties were determined. Next, binary batch IEs determined $K_{\mathrm{x}∕\mathrm{C}}$ for bicarbonate, sulfate, and nitrate. For a given anion, $K_{\mathrm{x}∕\mathrm{C}}$ were similar between resins and trended $K_{{}_{\mathrm{B}∕\mathrm{C}}}<K_{{}_{\mathrm{S}∕\mathrm{C}}}({\beta }_{{}_{\mathrm{S}∕\mathrm{C}}}\text{at studied conditions})$ < $K_{{}_{\mathrm{N}∕\mathrm{C}}}$. Subsequently, a competitive batch IE demonstrated that binary IE-determined $K_{\mathrm{x}∕\mathrm{C}}$ was applicable for competitive system simulation by the law of mass action.

CEs were then conducted with PFA694E that (i) corroborated IE-determined $K_{\mathrm{x}∕\mathrm{C}}$ as the anion breakthrough order was bicarbonate, sulfate, and then nitrate and (ii) illustrated minimal NOM impact on IA breakthrough for the specific NOM source and IA concentration ranges studied. Observed transient water quality (pH decreased until complete bicarbonate breakthrough and CSMR and LR were elevated until the breakthrough of relevant IAs) has potential corrosion control-related unintended consequences but will likely be short-lived (i.e., hours to days) compared to operation times anticipated for PFAS removal (i.e., months to years). Using IE-determined $K_{\mathrm{x}∕\mathrm{C}}$ the IEX-CM effectively simulated experimental CEs with and without NOM; therefore, it was not possible to develop or test corrections for NOM in the current IEX-CM. However, NOM impacts on the PFAS breakthrough and associated IEX-CM modifications remain a future research area, along with providing D_s estimates and revising the IEX-CM to simulate pH. Finally, IEX-CM utility was further illustrated by simulating the treatment of low-IA source water and a change in the source water to understand the resulting duration of changes in IAs and associated water quality parameters.

In summary, determined resin properties and $K_{\mathrm{x}∕\mathrm{C}}$ are applicable to future research with PFAS and provide fundamental information to the drinking water treatment industry for engineering AEX treatment for IA removal. For general IA removal (e.g., nitrate), the developed IEX-CM could be applied in a pilot testing design, adapting resin installation or replacement procedures for minimizing water quality disruptions and simulating the performance of a particular resin.