Caffeine, riboflavin and curcumin amended clays for PFAS binding

Abstract

Per- and polyfluoroalkyl substances (PFAS) are usually found in mixtures with other toxic compounds. Therefore, the study and design of broad acting sorbents, such as clays, is an attractive sorption solution. We previously demonstrated that clays amended with choline and carnitine could enhance PFAS sorption properties. Here, we used computations to screen from a pool of chemical compounds, which are either supplements or generally recognized as safe, and identified particular supplements that can be amended to clay and potentially improve its sorbing capacity for PFAS in acidic conditions. Simulations were initially used as a tool to identify promising amendments to the clay, while subsequently, simulations evaluated which selected amendments could potentially bind PFAS. Our results showed that caffeine-, riboflavin- and curcumin-amended clays can, in particular instances, enhance the binding of different PFAS compared to parent clays. Experiments investigated the sorption properties of the designed systems. Notably, caffeine-amended clay significantly enhanced GenX binding when compared to parent clay, with its binding capacity being increased from 0.15 mol/kg to 1.17 mol/kg. Caffeine-amended clay also enhanced binding for PFOS by 125%, compared to the parent clay, and for PFOA to a lesser extent. Additionally, riboflavin-amended clay enhanced binding for GenX, PFOA and PFOS by 120%, 23%, and 70%, respectively, compared to the parent clay. Our studies provide atomistic details into their mechanisms of action. Both the novel computational library of chemical compound-amended clays and the approach utilized, combining computations and experiments, could enhance the future design of novel amended clays for other toxins.

1. Introduction

Per- and polyfluoroalkyl substances (PFAS) constitute a big and diverse family of compounds comprising more than 4000 partially or fully fluorinate (Sunderland et al., 2019) linear, branched (Sunderland et al., 2019; Rahman et al., 2014), or cyclic compounds (Glüge et al., 2020). PFAS have the general chemical formula: C_nF_2n+1-R (Sunderland et al., 2019). They comprise a common aliphatic carbon backbone containing 1 or more carbon atoms where hydrogen atoms are either partially (poly-fluorinated) or completely (perfluorinated) substituted by fluorine atoms (Rahman et al., 2014). This backbone is called a per- or polyfluoroalkyl moiety, and is often referred to as the molecule’s tail, while R in the chemical formula represents additional functional groups like −COOH, −SO₃ etc. (Fernandez et al., 2016) The tail is highly chemically and thermally stable with strong carbon-fluorine bonds (Rahman et al., 2014), with both hydrophobic and oleophobic properties (Fernandez et al., 2016). PFAS are highly stable (Buck et al., 2012), and are considered to be neither biologically nor physically degradable at environmental conditions (Göckener et al., 2021).

PFAS are widely used in many industrial processes and for the production of consumer goods (Sunderland et al., 2019; Rahman et al., 2014). They were first used in the 1940s, and ever since, they have been accumulating in the environment (Sunderland et al., 2019) and have been detected in drinking water around the world (Domingo and Nadal, 2019; Hanssen et al., 2013) and animals (Hanssen et al., 2013), as well as the human body (Olsen et al., 2017). Coupled with their prolific and consistent use in industry (Sunderland et al., 2019), PFAS’s lengthy environmental persistence (Fernandez et al., 2016) and high biological half-lives (Jian et al., 2018), resulted in their abundance in water, food, air (Sunderland et al., 2019) and ground (Gebbink and van Leeuwen, 2020). PFAS were also detected in aquatic environment and aquatic fauna (Quinete et al., 2009; Ahrens and Bundschuh, 2014). PFAS traces have been identified in the bloodstream (Jian et al., 2018; North Carolina Department of Health and Human Services 2018; Fenton et al., 2021) of nearly the entire population in developed countries (Fenton et al., 2021), causing various health issues (Fenton et al., 2021; Kato et al., 2011), while reports show that many different PFAS molecules have been detected in the human milk (Jian et al., 2018) and/or urine (Jian et al., 2018). PFAS exposure was detected in the serum of 95% in the population tested, including PFOA (Perfluorooctanoic acid), and PFOS (Perfluorooctane sulfonic acid) (Khalil et al., 2016). Various epidemiological studies highlight the detrimental impact of different PFAS molecules on health. The effects range from immune system (Fenton et al., 2021; Guillette et al., 2020; Kvalem et al., 2020) and thyroid malfunction (Fenton et al., 2021; Guillette et al., 2020) to liver diseases (Fenton et al., 2021; Guillette et al., 2020), kidney diseases, cancer (Fenton et al., 2021; Guillette et al., 2020), and others. Long-chain PFAS are stored in the liver, accompanied by hepatocellular adenomas and carcinomas (Fenton et al., 2021). Furthermore, reviews highlight that exposure to PFAS during childhood can increase the chance of atopic dermatitis and lower respiratory tract infections (Kvalem et al., 2020).

PFAS remediation has been the focus of many studies with different approaches. These approaches include, among others, granular activated carbon (GAC), anion exchange (AIX) resins, reverse osmosis (RO) and nanofiltration (NF) membranes (Crone et al., 2019; Tow et al., 2021; Flores et al., 2013). RO and NF are effective methods to remove PFAS from water, however the treatment or disposal of waste concentrate stream comprises a key challenge (Tow et al., 2021). GAC and AIX constitute examples of adsorption technologies, which are the most commonly used methods for PFAS remediation (Tow et al., 2021). They are often coupled with RO or NF to maximize the PFAS removal efficiency (Tow et al., 2021; Franke et al., 2021); yet, they demonstrate sorption limitations that become more profound for smaller PFAS molecules (McCleaf et al., 2017). PFAS remediation methods also include PFAS destruction technologies like electrochemical oxidation, photocatalysis, and sonolysis (Meegoda et al., 2022).

PFAS contamination reports show that usually affected areas are not contaminated by a single PFAS, rather many PFAS compounds are present (Sunderland et al., 2019; McCord and Strynar, 2019; Solo-Gabriele et al., 2020; Susmann et al., 2019). Furthermore, exposure to PFAS occurs simultaneously with exposure to other toxic substances. According to research on toxic compounds present in the drinking water in the U.S., PFAS were found in mixtures with other toxic substances such as arsenic, lead, and others (Levin et al., 2024). Additionally, co-contamination of breast milk with PFAS and Polychlorinated biphenyls (PCBs) (Fromme et al., 2022), as well as co-contamination of soil with PFAS, and heavy metals (Cai et al., 2023) have been recorded. In addition, PFAS and microplastics (MP) have been found as surface-water contaminants (Chen and Hua, 2024), and PFAS with polycyclic aromatic hydrocarbons and heavy metals in soil (Zhang et al., 2024). Therefore, it is crucial to consider augmenting broad acting sorbents’ binding properties for PFAS. Clay and clay composites comprise broad acting sorbents with the capacity to bind many toxic substances (Maged et al., 2023; Ewis et al., 2022). Studies, including previously published work from our labs, showed that clays can bind heavy metals (Wang et al., 2021a; Mitchell et al., 2014), glyphosate (Khoury et al., 2010), as well as glyphosate and paraquat (Wang et al., 2019), aflatoxin (Phillips et al., 2019; Maki et al., 2017; Pollock et al., 2016; Oladele et al., 2025a), dieldrin (Hearon et al., 2020), and others (Orr et al., 2021). Additionally, our studies showed that clay sorption properties for particular toxins can be improved through the use of chemical compounds or proteins as amendments on clay surfaces. For example, β-lactoglobulin enhances the binding of cadmium and lead (Lilly et al., 2024), carnitine enhance the binding of bisphenols (Orr et al., 2020), chlorophyll enhances the binding of benzene (Rivenbark et al., 2022; Rivenbark et al., 2024), nano-plastics (Wang et al., 2024a), and ochratoxin (Oladele et al., 2024). Furthermore, our studies showed that amendments can elevate their functionality to target PFAS in acidic conditions. More specifically, choline and carnitine amended clays showed increased binding of PFAS when compared to the parent (unamended) clay (Wang et al., 2021b; Hearon et al., 2022).

Motivated by our findings in previous studies, we aimed to augment PFAS binding to calcium montmorillonite clays by using computations to design amended clays at acidic conditions, simulating stomach conditions (Wang et al., 2021b). Studies have shown the capacity of MD simulations for the investigation of adsorption phenomena on clays (Fashina et al., 2024) and other surfaces (Liu et al., 2020). We selected to study four PFAS molecules: GenX (Perfluoro-2-methyl-3-oxahexanoic acid), PFOA, PFOS, and PFBS (Perfluorobutane sulfonic acid). PFOA and PFOS are longer-chain PFAS which have long half-lives in the human body and were associated with multiple adverse health effects. PFBS and GenX are shorter-chain PFAS that were later introduced to industry as a substitute for PFOA and PFOS, however they have also been detected in the environment (Olsen et al., 2007). Acidic conditions, and particularly pH 2, were chosen to simulate conditions found in the stomach, as in our previous experimental and computational studies (Wang et al., 2021b). Computations were used to predict potential amendments that could improve PFAS binding to calcium montmorillonite clay and be validated experimentally. Both the novel computational library of chemical compound-amended clays and the computational approach utilized could pave the way to enhance the design of novel amended clays for other toxins in the future.

2. Methods

2.1. In silico screening of chemical compound-amendments to clays

In this study, we selected seventeen chemical compounds comprising FDA-Generally Recognized as Safe (GRAS) compounds and/or supplements (Jodra et al., 2020; Panknin et al., 2023; Tavares et al., 2009; Sharifi-Zahabi et al., 2023; Kansakar et al., 2023; Shi et al., 2021; Schwarz et al., 2018; Sawicka et al., 2020; Ju et al., 2017) which are presented in Table S1. The selection was based on additional simulations of a larger set of compounds which initially screened out compounds with considerably low binding capacity. We investigated the ability of the selected compounds to be amended on clay at acidic conditions, initially using computational methods. As our studies focused on acidic conditions, emphasis was placed on determining their protonation state at pH 2. Table S1 additionally shows the chemical compounds’ PubChem ID and their charge at pH 2. The protonation state for riboflavin was taken from (Khan and Mohan, 1973; Quick et al., 2013) with the two studies guiding the exact position of the protonated group. For the rest of the compounds, their protonation state was determined using graph-convolutional neural network model-based calculations by MolGpKa (Pan et al., 2021), in line with their protonation state at pH 2 based on their pK_a (e.g., caffeine (Wang et al., 2022), curcumin (Martínez-Guerra et al., 2019), EGC, EC, EGCG, ECG (Muzolf-Panek et al., 2012), sucrose (Lara-Cruz and Jaramillo-Botero, 2022), choline (Perrin, 1972)). The model structures for the uncharged chemical compounds were taken from PubChem and were used as initial structures for the simulations. Additionally, the model structures of charged choline and carnitine zwitterion were also taken from PubChem and were used as the initial structure for the simulations. To construct the initial structure of the positively charged chemical compounds (caffeine, riboflavin, and theobromine), CHARMM-GUI Ligand Reader and Modeler (Jo et al., 2008; Brooks et al., 2009; Lee et al., 2016; Kim et al., 2017) was used, starting from their corresponding uncharged structures from PubChem and adding a hydrogen to the group predicted according to MolGpKa (Pan et al., 2021), and others (Khan and Mohan, 1973; Quick et al., 2013). Their topology and parameter files were obtained using CgenFF (Vanommeslaeghe and MacKerell, 2012) version 3.0. As in our previous studies (Oladele et al., 2025a; Lilly et al., 2024; Oladele et al., 2024), the initial calcium montmorillonite structure at acidic conditions was initially constructed using CHARMM-GUI (Jo et al., 2008; Brooks et al., 2009; Lee et al., 2016; Choi et al., 2022; Heinz et al., 2013) according to INTERFACE-FF (Heinz et al., 2013). A model of two layers of clay with chemical composition (Si₄)^I-V(Al_1.67Mg_0.33)^VIO₁₀(OH)₂, dimension 50 × 50 Ų, miller indices 001, and a ratio of defect of 0.33333 was obtained, as in our previous studies (Oladele et al., 2025a; Lilly et al., 2024; Rivenbark et al., 2022; Rivenbark et al., 2024; Wang et al., 2024a; Oladele et al., 2024; Oladele et al., 2025b). The two layers were subsequently separated at a distance of 21 Å (Greenland and Quirk, 1960). INTERFACE-FF (Heinz et al., 2013) force field was used for the parameterization of clay, in conjunction with CHARMM-GUI.

We initially employed short molecular dynamics simulations in explicit solvent, for 16 copies of each chemical compound in complex with the clay. Input files to set up and simulate the compound-clay systems were initially generated using CHARMM-GUI (Jo et al., 2008; Brooks et al., 2009; Lee et al., 2016; Choi et al., 2022; Heinz et al., 2013), with particular adjustments and modifications performed. Initially, the two clay layers were centered in a cubic (90 Å) periodic boundary conditions box and solvated by explicit water molecules in CHARMM (Brooks et al., 2009). Na⁺provided by default by CHARMM-GUI (Jo et al., 2008; Brooks et al., 2009; Lee et al., 2016; Choi et al., 2022; Heinz et al., 2013) for the clay model were manually deleted and 20 Ca²⁺ (1 every two Na⁺) were placed randomly and allowed to freely move in the water box. Additional Ca²⁺ were added to neutralize the constructed systems, based on the number of molecules added and their corresponding charge in each case. The 16 chemical compounds in each case were randomly placed and oriented in the box. Prior to the simulations water molecules in close proximity to clay were removed, and SD and ABNR energy minimization was conducted in CHARMM (Brooks et al., 2009). Before the production, a 200 ps equilibration (NVT) run at 300 K was performed. Subsequently, a 30 ns production (NPT) run was performed at 300 K, 1 atm and using an isotropic barostat. During both the equilibration and the production steps magnesium and aluminum atoms of clay were constrained with a force constant of 400 kJ/(mol nm²). Both equilibration and production steps were performed using OpenMM (Eastman et al., 2017).

Upon simulations’ completion, the ten chemical compounds with the highest binding probability to clay were selected for subsequent investigation, and are shown highlighted in Table S1. The binding percentage probability was calculated by counting the total number of interactions of each chemical compound to clay directly or indirectly (chemical compound aggregate bound to clay) for all analyzed snapshots and it was then normalized by the total number of chemical compound copies (16) in the simulation and the total number of snapshots analyzed. Two entities were considered to interact with each other, when the distance between any pair of their atoms (including hydrogens) was ≤ 3.5 Å. For these calculations in-house FORTRAN programs were utilized to analyze the simulation trajectories every 1 ns (30 snapshots were analyzed per simulations). The structures of the selected chemical compound-amended clays were extracted from the last snapshot of the corresponding simulation. These were used as initial structures in subsequent simulations to investigate PFAS binding to the particular amended clays. The library of selected chemical compounds was investigated below in complex with PFAS.

2.2. Simulations investigating selected chemical compound-amended clays for PFAS binding

The ten selected chemical compound-amended clays were investigated in complex with four different PFAS molecules, independently: PFOA, PFOS, PFBS, and GenX, shown in Table S2. Table S2 also shows the PFAS’ PubChem ID and their charge at pH 2. All four PFAS molecules were investigated at their monoanionic states, guided by graph-convolutional neural network model-based calculations by MolGpKa (Pan et al., 2021) performed by us in this study. The calculations depicted pKa values considerably lower than 2; i.e., −2.9 for PFOS, −2.9 for PFBS, 0.1 for PFOA and 0.0 GenX. Notably, pK_a values lower than 2 were reported for PFBS and PFOS (Lasáková and Jandera, 2009; Steinle-Darling and Reinhard, 2008), and a most recent study for PFOA (Vierke et al., 2013) reported a value of 0.5 for PFOA, which support the investigation of a monoanionic state at pH 2. To construct the initial structure of the monoanionic states of the PFAS molecules, CHARMM-GUI Ligand Reader and Modeler (Jo et al., 2008; Brooks et al., 2009; Lee et al., 2016; Kim et al., 2017) was used, starting from their corresponding uncharged structures from PubChem and removing the hydrogen from their functional groups. Afterwards, the topology and parameter files for the uncharged molecules were obtained using CGenFF (Vanommeslaeghe and MacKerell, 2012) version 3.0, while the topology and parameter files for the monoanionic molecules were obtained through CGenFF via CHARMM-GUI (Jo et al., 2008; Brooks et al., 2009; Lee et al., 2016). In Table S2 the compounds are shown along with the corresponding PubChem IDs and their charged state at acidic conditions.

To investigate the ability of the ten selected chemical compound-amended clays to bind PFAS molecules, we first performed a single explicit solvent molecular dynamics simulation for each selected chemical compound-PFAS combination. Particularly, for each chemical compound-amended clay, four different combinations were investigated, in acidic conditions, with GenX, PFOA, PFOS, and PFBS, independently. The initial structures of the PFAS molecules were obtained/constructed as explained above, while the initial structure of the chemical compound-amended clays were extracted by the simulations as described in Section 2.1. Overall, we performed 40 independent simulations of 100 ns for all combinations. For each simulation 16 copies of the PFAS molecules were investigated in conjunction with the different chemical compound-amended clays. Input files to set up the simulations were initially generated using CHARMM-GUI (Jo et al., 2008; Brooks et al., 2009; Lee et al., 2016; Choi et al., 2022; Heinz et al., 2013), with particular adjustments and modifications performed. Each simulation started with the chemical compound-amended clay initially being adjusted in a cubic (90 Å) periodic boundary conditions box, such that all amendment molecules exist within the box and are solvated by explicit water molecules in CHARMM (Brooks et al., 2009). For the control runs, the clay was centered in the cubic (90 Å) periodic boundary conditions box and solvated by explicit water molecules in CHARMM (Brooks et al., 2009). The 16 copies of PFAS molecules in the corresponding state were randomly placed and oriented in the box, such that they were not in contact with the clay or the amendments. Prior to the simulations water molecules in close proximity to the clay, the amendments, or PFAS were removed, and SD and ABNR energy minimization was conducted in CHARMM (Brooks et al., 2009). Before the production, a 200 ps equilibration (NVT) run at 300 K was performed. Subsequently, a 100 ns production (NPT) run was performed at 300 K, 1 atm and using an isotropic barostat. During both the equilibration and the production steps magnesium and aluminum atoms of clay were constrained with a force constant of 400 kJ/(mol nm²). Both equilibration and production steps were performed using OpenMM (Eastman et al., 2017).

Upon completion of all simulations, in-house FORTRAN programs were utilized to analyze the simulation trajectories and quantify the binding capacity of each PFAS to the ten selected chemical compound-amended clays, as well as the parent clay. Simulations were analyzed with snapshots extracted every 1 ns, thereby 100 snapshots were analyzed per simulation. We determined the binding percentage probability of each PFAS to each one of the 10 selected chemical compound-amended clays. The binding percentage probability was calculated by counting the total number of interactions of each PFAS molecule to the chemical compound-amended clay (or parent clay in the case of control simulations) and it was normalized by the total number of copies of PFAS molecules (16) in the simulations and the total number of snapshots analyzed (100). Indirect interactions involving cases that a PFAS (or PFAS cluster) interacts with a chemical compound (or cluster of chemical compounds) bound to the clay were also taken into consideration in the counting. Two entities were considered to interact with each other, when the distance between any pair of their atoms (including hydrogens) was ≤ 3.5 Å.

At this stage, we aimed to identify the top three chemical compound amended clays for subsequent experimental and computational investigation. For this purpose, we ranked each chemical compound-amended clay based on its consensus binding percentage probability for all of the four PFAS systems: GenX, PFOA, PFOS, PFBS, which led us identify the top three predicted amendments: riboflavin, caffeine, and curcumin, which are supplements. Table S1 shows in highlighted format (yellow or green) all the chemical compound-amended clays that were simulated in complex with PFAS, and particularly highlights in green caffeine, curcumin and riboflavin. We considered important to perform two additional 100 ns simulations for each combination of the three-amended clays in complex with the four PFAS molecules. In addition, we investigated four additional systems of the PFAS in complex with parent (unamended) clay, which served as our control. Thus, triplicate simulations were performed for control runs (PFAS molecules with parent clay) and the PFAS molecules with the top three supplement-amended clays. Upon completion of the simulations, analysis was performed with snapshots extracted every 1 ns, thereby 100 snapshots were analyzed per simulation. The average binding percentage probability of each PFAS each one of the top three supplement-amended clays was calculated over the triplicate runs and was compared to the parent clay, for which the average binding percentage probability was calculated the same way.

2.3. Simulation analysis of caffeine, curcumin and riboflavin-amended clays with PFAS

We performed in-depth analysis of the simulation trajectories to investigate in atomistic detail the interactions between the PFAS molecules and the top three supplement-amended clays. Using in-house FORTRAN programs we studied (I) interactions of the three supplements with parent clay, (II) interactions of the PFAS with the three supplement-amended as well as parent clays. We aimed at obtaining atomistic insights into the PFAS binding to the three amended clay systems, in comparison to parent clays. Thus, we decomposed each PFAS and each supplement into chemical groups, as shown in Figs. S1 and S2, respectively. Subsequently, we decomposed interactions into: (ai) direct-assisted interactions, which are interactions of PFAS with the clay and amendment molecules (bound to the clay) simultaneously; (aii) direct-helped interactions, which are interactions of PFAS with the clay and amendment molecules (not bound to clay), simultaneously; and (aiii) indirect-assisted interactions, which are interactions of PFAS only with amendment molecules which are either bound directly to clay or they are part of a cluster of amendment molecules indirectly bound to clay. PFAS molecules that were part of PFAS aggregates of which at least one molecule was interacting with the amendment molecules via (ai)-(aiii) were also counted. In few cases, clusters of PFAS molecules were identified bound to the clays, and in these cases, only the directly bound PFAS molecule was considered as part of the direct interactions. Snapshots of simulations were obtained and visualized using VMD (Humphrey et al., 1996).

2.4. In vitro experimental validation

2.4.1. Chemicals and materials

PFAS (PFBS, GenX, PFOS, and PFOA) analytical standards were procured from Sigma Aldrich (St. Louis, MO, USA). pH buffers (4.0, 7.0, and 10.0) and high-performance liquid chromatography (HPLC)-grade acetonitrile were obtained from VWR (Atlanta, GA, USA). Parent montmorillonite, obtained from BASF (Lampertheim, Germany), had the generic formula (Na,Ca)_0.3(Al,Mg)₂Si₄O₁₀(OH)₂•nH₂O, an external surface area of 70 m²/g, a total surface area of about 850 m²/g, a zeta potential of −31 mV, and a cation exchange capacity of 97 cmol/kg (Kumar et al., 2020). Using previously reported protocols for amendment of clays, supplement-amended clays were created by interposing curcumin, riboflavin, or caffeine between interlayer surfaces with 100% cation exchange capacity in an acidic environment (Wang et al., 2017).

2.4.2. Chemical analysis

PFAS analysis was carried out using the protocol as in our previous studies (Wang et al., 2021b) with slight modification. The analysis of the four PFAS compounds (PFBS, GenX, PFOS, and PFOA) was conducted using a Waters Acquity ultraperformance liquid chromatography/-tandem mass spectrometer (LC/MS-MS) fitted with a triple quadrupole. The detection and quantification were carried out using an Acquity BEH C18 column (2.1 × 50 mm, 1.7 μm), which was maintained at 40 °C in the column oven. At a flow rate of 0.30 mL/min, a gradient elution was performed for 13 min using 20 mM ammonium acetate (eluent A) and acetonitrile (eluent B). For every analysis, the injection volume was 50 μL. The mass spectrometer was operated with an electrospray ionization interface (ESI) set in a negative ion mode and the spray voltage was kept at 4.5 kV. The source temperature was kept at 450 °C. The monitored precursor and product ions (m/z) for PFBS, GenX, PFOS, and PFOA were 298.9 to 80, 285 to 168.9, 499 to 80, and 413 to 369 respectively. The cone voltage (kV) for PFBS, GenX, PFOS, and PFOA was 35, 20, 40, and 25, respectively. The mass spectrometer was operated under multiple reaction monitoring (MRM) mode. The nebulizer and heater gases were argon gas, whereas the collision and curtain gases were nitrogen gas. Empower analyst software was used to control the LC/MS-MS system and acquire data.

2.4.3. Sorption isotherms

Individual PFAS solutions, containing PFBS, GenX, PFOS, and PFOA, were prepared from pure crystals at a concentration of 10 ppm (μg/mL) in pH 2 distilled water, which is the average pH of the stomach. Then, each parent and amended clay at 0.0005% was applied to a concentration gradient ranging from 5% to 100% of 1 mL of PFAS solution. One milliliter of clays suspension, PFAS solution, or blank solution (pH 2 water) were used as controls. A 2-hour vibration was applied to all samples using an IKA^® electric shaker (VIBRAX VXR basic, Werke, Germany) set to 1000 rpm and 37 °C. After centrifuging the sorbent/PFAS complex for 20 min at 2000 g, the supernatant was analyzed using LC/MS-MS.

2.4.4. Data calculations and curve fitting

The amount of free PFAS in the solution was determined using the PFAS found by LC/MS-MS method. The amount of bound PFAS in the adsorption investigation was determined using the concentration difference between the test and control groups, and it was plotted as mol/kg on the isotherms. Plotting these data and determining the values for the variable parameters were done using Table-Curve 2D Based on the equation that fit the data with the highest correlation coefficients and the randomness of the residuals from triplicate studies, the adsorption isotherm was shown using well-known Langmuir or Freundlich models (Grant and Phillips, 1998). Monolayer adsorption onto a surface with a finite number of identical sites and homogeneous adsorption energies is described by the Langmuir isotherm. The Langmuir equation and functions:

$$\mathit{\text{Langmuirmodelq}}=Q_{max}\frac{{\text{K}}_{\text{d}}{\text{C}}_{\text{w}}}{1+{\text{K}}_{\text{d}}{\text{C}}_{\text{w}}}$$ (1)

where C_w = equilibrium concentration of OTA (mol L⁻¹), K_d = Langmuir distribution constant, Q_max = maximum binding capacity (mol kg⁻¹), and q = the amount of OTA adsorbed (mol kg⁻¹).

The Freundlich isotherm is used to describe the adsorption characteristics for a heterogeneous surface. The Freundlich model is represented by the following equation:

$$\mathit{\text{Freundlich}}\mathit{\text{model}}q=K_f{C}_w^{1/n}$$ (2)

Kf = Freundlich distribution constant, 1/n = degree of heterogenicity.

3. Results and discussion

3.1. Simulations screening chemical compounds in complex with clay and simulations of PFAS in complex with top three selected supplement-amended clays

Upon completion of the simulations screening the seventeen chemical compounds in complex with clay, a subset library of ten selected chemical compounds was created based on the percentage binding probability of the amendments to the clay: riboflavin, curcumin, caffeine, theobromine, choline, epigallocatechin gallate, propylparaben, propyl gallate, triethyl citrate, and salidroside.

The corresponding ten selected chemical compound-amended clays were simulated in complex with the PFAS molecules. Fig. 1 shows the percentage binding probabilities calculated for each PFAS molecule to each one of the ten selected chemical compound-amended clays in acidic conditions. PFBS showed the lowest binding for all ten selected chemical compound-amended clays, with only riboflavin-amended clay showing binding greater than ~30%. GenX also showed reduced binding when compared to the other PFAS molecules for all ten selected chemical compound-amended clays, with only riboflavin-amended clay showing binding greater than ~40%. Contrary, PFOA and PFOS showed improved binding, exceeding the ~50% for more than half of the ten selected chemical compound-amended clays. This higher binding with long-chain PFAS can be possibly attributed to the increased hydrophobic interaction with the chemical compound-amended clays as we also discuss in detail below. Based on the percentage binding probabilities of PFAS to the chemical compound-amended clays we ranked the ten selected amended clays in order of most binding to least binding and the results showed that the top three chemical compound-amended clays are riboflavin-amended clay (CMribo), caffeine-amended clay (CMcaff) and curcumin-amended clay (CMcurc), which all comprise supplement-amended clays. Since our goal was to design broad-acting sorbents, we deemed important to proceed with these three supplement-amended clays and perform experiments on them as well as computationally investigate their mechanistic properties in complex with PFAS molecules.

Fig. 1.

Panels A-D show the percentage binding probability of GenX, PFOA, PFOS, and PFBS, respectively, to different chemical compound-amended clays from a single simulation at pH 2. Triplicate runs were performed only for parent clay as a control. Average values are shown for the control which were calculated from triplicate runs. Error bars denote standard deviation values calculated from triplicate runs.

Fig. 2 shows the average binding percentage probability calculated for each PFAS to each one of the top three supplement-amended clay in acidic conditions, as calculated from triplicate simulations runs. PFOA and PFOS bind to CM with an average binding percentage probability of ~40%. GenX and PFBS show minimal binding to CM, lower than 20%. Overall, our simulations predicted that these amendments can potentially enhance binding of particular PFAS to clays, and experiments were used for further investigation.

Fig. 2.

Panels A-D show the average percentage binding probability of GenX, PFOA, PFOS, and PFBS, respectively, to different supplement-amended clays at pH 2. In each case, the average percentage binding probability of PFAS molecules to the parent clay (CM) is also shown. Direct, direct-assisted, direct-helped, and indirect-assisted interactions are shown in dark blue, orange, green, and cyan, respectively. The average values are calculated from triplicate runs. Error bars denote standard deviation values for the total binding probability calculated from triplicate runs.

3.2. Adsorption isotherms

Adsorption isotherms for the four PFAS used in this study were conducted at pH 2 to simulate gastrointestinal-relevant conditions (Fig. 3). Three of the isotherms (PFOS, PFOA and GenX) fit the Langmuir equation while all PFBS isotherms fit Freundlich model except CMcurc. The models were used to determine degree of heterogenicity (1/n), binding affinity (K_d) and binding capacity (Q_max) values for PFAS binding to the parent (CM) and the supplement-amended clays (CMcurc, CMcaff, and CMribo) (Table 1). The resulting Q_max and K_d values along with the curved shape of the Langmuir plot indicated saturable binding of GenX, PFOA and PFOS to active surfaces of the parent and the supplement-amended clays.

Fig. 3.

Panels A-D show the adsorption isotherm of GenX, PFOA, PFOS, and PFBS, respectively, on parent and supplement-amended clays at pH 2. The solid lines represent the adsorption isotherm plots based on the Langmuir/Freundlich model, while the dashed lines represent the 95% confidence band of Langmuir/Freundlich model.

Table 1.

Parameters and correlation coefficients of adsorption at pH 2.

	Qmax	Kd	ΔG	r2
GenX
CM	0.15	3.47 × 106	−28.92	0.90
CMcaff	1.17	7.80 × 105	−26.27	0.92
CMcurc	0.35	2.04 × 106	−24.25	0.90
CMribo	0.33	1.04 × 106	−23.96	0.88
PFOA
CM	0.35	2.07 × 105	−24.29	0.97
CMcaff	0.40	2.10 × 105	−25.12	0.92
CMcurc	0.34	6.24 × 105	−24.07	0.87
CMribo	0.43	4.04 × 105	−25.55	0.88
PFOS
CM	0.20	2.64 × 105	−26.90	0.84
CMcaff	0.45	1.14 × 105	−25.09	0.91
CMcurc	0.16	2.02 × 105	−23.01	0.85
CMribo	0.34	1.87 × 105	−24.71	0.92
PFBS
CMcurc	0.19	2.12 × 105	−21.58	0.88
	1/n	Kf	ΔG	r2
CM	0.94	3.97 × 104	−25.77	0.88
CMcaff	0.81	2.88 × 105	−26.85	0.85
CMribo	0.91	8.17 × 104	−26.62	0.93

Q_max: binding capacity (mol/kg); r²: square correlation coefficients; ΔG: Gibbs free energy (kJ/mol); K_d: binding affinity; K_f: Freundlich distribution constant; 1/n: degree of heterogenicity.

Importantly, the supplement-amended clays CMcaff and CMribo had notable higher binding capacity for PFOS with 0.45 mol/kg and 0.34 mol/kg respectively than CM with 0.20 mol/kg (p < 0.05). This indicated that clay amendment with riboflavin and caffeine enhanced the binding capacity of the clay by 70% and 125% respectively. Similarly, CMcaff and CMribo demonstrated higher binding capacity for PFOA than CM. The Q_max of CMcaff and CMribo is 0.40 mol/kg and 0.43 mol/kg respectively while the CM is 0.35 mol/kg. This data indicated that amendment of caffeine improved the binding capacity for PFOA by 14% while riboflavin improved the binding capacity of clay for PFOA by 23%. These results corroborate with our in silico studies. Also, binding to parent clay is higher in PFOA, followed by PFOS and then GenX according to Q_max values, also in agreement with in silico studies.

Remarkably, all the three supplement-amended clays tested in this study demonstrated significantly higher binding capacity for GenX than the parent clay, also in line with our in silico studies. The result showed that Q_max of CMcaff, CMcurc and CMribo is 1.17 mol/kg, 0.35 mol/kg and 0.33 mol/kg respectively while the CM is 0.15 mol/kg. This result revealed that caffeine, riboflavin, and curcumin modified the binding surfaces of the parent montmorillonite clay which could allow for attraction of more lipophilic moieties of the PFAS and their functional groups. The mechanisms are discussed in detail below. The findings in this study are in tandem with previous reported studies that have shown the improved binding capacity of clays following different amendments. For instance, different studies from our group have reported enhanced binding capacity of clays for PFASs following clay amendment with nutrients such as carnitine and choline (Wang et al., 2021b), as well as green amendments using chlorophyll (Wang et al., 2024b).

Furthermore, the Gibbs free energies of the binding interactions between the four PFAS substances and the clays had a high absolute value. This indicates that the reactions between the clays and the PFAS are thermodynamically favorable.

3.3. Mechanistic computational investigation of the top three supplement-amended clays

We performed an in-depth analysis to shed light on the binding mechanism of PFAS to the supplement-amended clays. Curcumin interacted with PFAS primarily through indirect-assisted interactions, while caffeine and riboflavin primarily through direct-assisted interactions (Fig. 2). Example cases corresponding to the last snapshots extracted from a particular simulation of the amendment molecules and the clay in the absence of PFAS are presented in Fig. 4. Caffeine was mostly lying flat on clay (~89%), as shown in Fig. 4A with solid arrows, making it very likely for PFAS to interact with both the clay and caffeine simultaneously in direct-assisted interactions. Curcumin showed a tendency to form large clamps (~93%) that were predominantly in the exterior surface of clay, as shown in Fig. 4B with solid arrows, enabling PFAS molecules interacting mainly with curcumin amendments that were indirectly bound to clay. Curcumin’s amending properties are somehow analogous to the ones of chlorophyll with respect to the fact that both molecules can form larger aggregates bound to the clay (Rivenbark et al., 2022; Rivenbark et al., 2024; Wang et al., 2024a). Riboflavin also formed clumps (~65%) which were smaller in size compared to curcumin, and they could also be present in the interlayer of clay, as shown in Fig. 4C with solid arrows. Riboflavin interacted with clay both via its rings, as shown in Fig. 4C with dotted arrows, and its polar tail, as shown in Fig. 4C with dashed arrows. Single riboflavin molecules interacted in a few cases with the clay primarily through their polar tail, without necessarily being part of small aggregates.

Fig. 4.

Panels A-C show the binding of caffeine, curcumin, and riboflavin molecules, respectively, directly to clay. These correspond to the last snapshots extracted from the simulations of the molecules in the absence of PFAS and to the initial structure of the supplement-amended clays when simulated with PFAS. Solid, dotted and dashed arrows are indicating molecules and/or binding modes discussed in the main text. Clay and amendments are shown in vdW representation. Atoms are colored by atom type. Hydrogen atoms of CM were omitted for clarity (Humphrey et al., 1996).

Example cases corresponding to the last snapshots extracted from a particular simulation of PFAS and the clay in the absence of amendments are presented in Fig. 5. In the absence of supplement amendments, GenX had lower binding probability to CM compared to PFOA and PFOS (Fig. 2). GenX interacted with the clay predominantly through its perfluoroalkyl tail, with its functional group (−COO⁻) facing outwards to the solvent (~90%), as shown in Fig. 5A with solid arrows. In a few cases GenX interacted with the clay both with its perfluoroalkyl tail and its functional group (~10%), while GenX-GenX interactions were seldom observed. In the absence of supplement amendments, PFOA had the highest binding probability compared to PFOS and GenX (Fig. 2). PFOA interacted with the clay predominantly through its perfluoroalkyl tail, with its functional group (−COO⁻) facing outwards to the solvent (~75%), as shown in Fig. 5B with solid arrows. In the remaining other cases PFOA interacted with the clay both with its perfluoroalkyl tail and its functional group (~25 %). In addition, PFOA showed little tendency (~7%) to lay on top of other PFOA molecules bound to clay and interact via their perfluoroalkyl tails. In the absence of supplement amendments, PFOS interacted with the clay mainly through its perfluoroalkyl tail, with its functional group ($-{\text{SO}}_3^{-}$) facing outwards to the solvent (~55%), as shown in Fig. 5C with solid arrows. Also, PFOS interacted with the clay both with its perfluoroalkyl tail and its functional group (~45%), as shown in Fig. 5C with dotted arrows. In a few cases (~8%) PFOS laid on top of other PFOS molecules bound to clay via their perfluoroalkyl tails, as shown in Fig. 5C with dashed arrows. The tendency of PFOA and PFOS to form these multiple-layer aggregates could be attributed to the fact that both these molecules have the longest perfluoroalkyl tail enabling extended hydrophobic interactions between −CF₂ or −CF₃ groups. According to both computations and experiments, PFBS was depicted to bind minimally either to CM or supplement-amended clays. Fig. 5D shows an example case of PFBS interacting with CM in the absence of amendments, and as shown in Fig. 2, its binding is minimal.

Fig. 5.

Panels A-D show GenX, PFOA, PFOS, and PFBS, respectively, directly binding to clay in the absence of amendments. These correspond to the last snapshots extracted from a particular simulation of the corresponding control system. Solid, dotted and dashed arrows are indicating molecules and/or binding modes discussed in the main text. Clay is shown in vdW representation, while PFAS are shown in licorice representation. Atoms are colored by atom type. Hydrogen atoms of CM were omitted for clarity (Humphrey et al., 1996).

According to experiments, CMcurc and CMribo enhanced GenX binding, while CMcaff enhanced the binding capacity of GenX significantly. Example cases corresponding to selected snapshots extracted from a particular simulation of GenX and the supplement-amended clays are presented in Fig. 6. The significant enhancement of GenX binding to CMcaff could be attributed to the molecule laying on clay through its perfluoroalkyl tail and simultaneously interacting with caffeine’s non-polar and polar groups, while its functional group faced toward the solvent (~66 %), as shown in Fig. 6A with solid arrows. In the remaining cases, GenX either interacted with the exposed surface of flat caffeine molecules, as shown in Fig. 6A with dotted arrows, or in rare occasions it interacted with caffeine molecules that were bound to clay but were not laid flat. In these instances, the interactions were predominantly formed between GenX tail and caffeine’s non-polar or polar groups. These additional interactions with caffeine amplified the overall binding capacity between GenX molecules and CMcaff, primarily through direct and direct-assisted interactions. Additionally, GenX had somehow similar average binding probability directly to clay as in the control runs in the absence of amendments, as shown in Fig. 2. The primary binding mechanism of direct binding remained the same; GenX laid on clay with perfluoroalkyl tail and its functional group faced outwards to the solvent, as shown in Fig. 6A with a red solid arrow. The enhancement of GenX binding to CMcurc could be attributed to the fact that GenX was attracted by the big curcumin clump where it could interact with curcumin molecules and less frequently with other GenX molecules; the majority of these interactions were indirect-assisted interactions (~83%). Interestingly, GenX molecules interacted more with curcumin amendments through indirect-assisted interactions rather than with clay through direct-assisted and direct interactions combined. In nearly all cases (~88%) GenX perfluoroalkyl tail interacted with curcumin non-polar groups, as shown in Fig. 6B with solid arrows, and GenX functional group faced toward the solvent. GenX and curcumin interactions were primarily formed between GenX perfluoroalkyl tail and curcumin’s non-polar groups. Moreover, the enhancement of GenX binding to CMribo could be attributed to the fact that GenX interacted with the clay through its perfluoroalkyl tail and simultaneously interacted with riboflavin’s non-polar and polar groups, through direct-assisted interactions (~80%). Overall, considering all direct- and indirect-assisted interactions, in most cases (~70%), GenX perfluoroalkyl tail participated in non-polar interactions with riboflavin rings, while its functional group faced towards the solvent, as shown in Fig. 6C with solid arrows. In the remaining cases (~30%), GenX functional group and/or perfluoroalkyl tail interacted with riboflavin’s polar tail (~58% of ~30%), including instances of hydrogen bonds, as shown in Fig. 6C with dotted arrows. In some cases, GenX molecules interacted with each other simultaneously as they interacted with riboflavin and the clay, as shown in Fig. 6C with brackets.

Fig. 6.

Panels A-C show the interactions of GenX with caffeine-, curcumin-, and riboflavin-amended clay, respectively. These correspond to selected snapshots extracted from a particular simulation of the corresponding systems. Solid, dotted and dashed arrows and brackets are indicating molecules and/or binding modes discussed in the main text. Clay and amendments are shown in vdW representation, while GenX is shown in licorice representation. Atoms are colored by atom type. Hydrogen atoms of CM were omitted for clarity (Humphrey et al., 1996).

According to experiments, CMcaff and CMribo enhanced PFOS binding, while CMcurc did not. Example cases corresponding to selected snapshots extracted from a particular simulation of PFOS and the supplement-amended clays are presented in Fig. 6. The enhancement of PFOS binding to CMcaff could be attributed to the fact that PFOS laid on clay through its perfluoroalkyl tail and/or functional group and simultaneously interacted with its perfluoroalkyl tail with caffeine’s non-polar and polar groups (~50 %), as shown on Fig. 7A with solid arrows. The enhancement could also be attributed to the fact that PFOS remained on top of laid flat caffeine molecules (~50 %), as shown in Fig. 7A with dotted arrows. The former interactions were considered as direct-assisted, while the latter were considered indirect-assisted interactions. Notably, PFOS showed a tendency to form multi-layered aggregates that interacted with clay or amendments (Fig. 7A) similar to the control runs in the absence of amendments. In the case of CMcurc, PFOS appeared to be a constituent inner part of the aggregates, rather than binding externally to the aggregates’ surface, i.e., PFOS participated in mediated interactions between curcumin aggregates, as shown in Fig. 7B. Such mediated interactions could occur to a lesser extent in experiments at which curcumin amendments are well preformed. This could possibly explain the lower binding observed in experiments compared to simulations. Most of the interactions taking place between PFOS and CMcurc were indirect-assisted interactions (~73%). Overall, considering all direct and indirect assisted interactions, in most cases (~73%) PFOS perfluoroalkyl tail was interacting primarily with curcumin’s non-polar groups while PFOS functional group faced the solvent. Furthermore, the enhancement of PFOS binding to CMribo could be partly attributed to direct-assisted interactions (~55%), at which PFOS laid on clay through its perfluoroalkyl tail and simultaneously, interacting with riboflavin’s non-polar and polar groups with just its perfluoroalkyl tail (~50%), as shown in Fig. 7C with solid arrows, or both its tail and functional group (~50%), as shown in Fig. 7C with dotted arrows. The remaining ~45% of the interactions of PFOS with riboflavin corresponded to indirect-assisted interactions. Overall, considering all direct and indirect assisted interactions, in CMribo, we observed that in ~67% of the total binding instances hydrogen bonds were observed between PFOS functional group ($-{\text{SO}}_3^{-}$) or perfluoroalkyl tail and riboflavin’s polar tail. Finally, the binding mechanisms of PFOA to the supplement-amended clays were similar to the ones of PFOS. Interestingly, as shown in Fig. 2, direct and direct-assisted interactions have the highest contribution in the binding of PFOA to supplement-amended clays. This could potentially explain the slightly lower effect of the supplement-amendments, CMcaff and CMribo, in PFOA compared to PFOS, since the molecule has intrinsically high tendency to bind to CM.

Fig. 7.

Panels A-C show the interactions of PFOS with caffeine-, curcumin-, and riboflavin-amended clay, respectively. These correspond to selected snapshots extracted from a particular simulation of the corresponding systems. Solid and dotted arrows are indicating molecules and/or binding modes discussed in the main text. Clay and amendments are shown in vdW representation, while PFOS is shown in licorice representation. Atoms are colored by atom type. Hydrogen atoms of CM were omitted for clarity (Humphrey et al., 1996).

Notably, our computationally designed amended clays for PFAS were not biased by previous studies demonstrating modified and unmodified clay’s capacity to bind caffeine, riboflavin, and curcumin. A series of experimental studies investigated the adsorption of caffeine to modified or unmodified montmorillonite clay. Previous experimental demonstrated that caffeine adsorption to the clay follows a Langmuir model (Okada et al., 2015; Yamamoto et al., 2017a, 2017b; Shiono et al., 2017; Yamamoto et al., 2018; Sakuma et al., 2020; Fakioğlu and Kalpaklı, 2022; Quintero-Jaramillo et al., 2024). Additional studies highlight that the adsorption is more efficient in acidic conditions (Ravi et al., 2020). Studies by Yamamoto et al. highlighted the key role of ions and interlayer spacing influencing adsorption of caffeine onto montmorillonite clay (Yamamoto et al., 2017b). Our studies showed that caffeine predominantly laid flat (~89%) on the clay surfaces, complying with both experiments by Yamamoto et al. suggesting caffeine molecules to be probably oriented parallel to the montmorillonite layers (Yamamoto et al., 2018), as well as computations by Sakuma et al. (Sakuma et al., 2020), suggesting that caffeine molecules may be aligned parallel to the clay surfaces. In addition, Yamamoto et al. (2018), suggested that caffeine molecules were adsorbed both into the interlayer space and onto the surface and interacted with Si–OH and siloxane, in particular, and emphasized on the fact that both the conditions of the interlayer space of and the surface of montmorillonite are considered to be highly important in the interaction between montmorillonite and caffeine. Also, Sakuma et al., used simulations and highlighted key factors associated with caffeine binding to clay: (i) the strength of the attractive electrostatic interaction attractive strength between polar caffeine molecules and the interlayer cations, (ii) the attractive electrostatic interaction strength between the interlayer cations and the silicate layers, (iii) the amount of space available that water can be exchanged with caffeine molecules (Sakuma et al., 2020). In our computational study, the spacing between interlayers allowed calcium ions to move freely in the water box and not be bound to the clay surface, in line with our previous studies (Wang et al., 2019; Oladele et al., 2025a; Orr et al., 2021; Lilly et al., 2024; Orr et al., 2020; Rivenbark et al., 2022; Wang et al., 2024a; Oladele et al., 2024; Wang et al., 2021b; Hearon et al., 2022; Oladele et al., 2025b). This has enabled protonated caffeine molecules to bind the clay surface, with cations not necessarily mediating these interactions. Additionally, in our computational study, clay layers were constrained and thus the interlayer spacing was maintained throughout the simulation, investigating idealistic clay layers at a particular separation distance. Our results demonstrated caffeine binding was predominantly in the interlayer, while the binding of caffeine was minimal at the edges (~7–8%; see depiction of edge atoms at Fig. S3). While studying in-detail the mechanism of caffeine adsorption to the clay was not the goal of our study, we suggest that further studies are warranted to investigate this, considering different interplanar spacings for different ions (Yamamoto et al., 2017b). Caffeine’s predominant mode, as depicted by our simulations, to lay flat allows Si-O siloxane dipoles on clay’s exterior surface, with partial positive and negative charge respectively, to interact favorably with the positively charged group of caffeine, and additionally with C—H, C—O and C—N dipoles of caffeine with partial positive and negative charge (Fig. S4). This conforms with studies suggesting that the presence of an oxygen-rich surface helps the adsorption of polar compounds (like caffeine) by increasing the electrostatic attraction between the molecule and the surface (Fakioğlu and Kalpaklı, 2022; Bedia et al., 2018; Ahmad et al., 2014). Interestingly, previous studies showed that at environmentally-relevant PFOS concentrations, pyrolyzed spent coffee grounds showed minimal PFOS removal, while the activated material removed 9.9%–99.6% of PFOS depending on activation conditions, suggesting a link between caffeine and PFOS removal (Steigerwald and Ray, 2021). Furthermore, previous studies depicted that curcumin could adsorb to montmorillonite clay and other composite/modified materials including montmorillonite clay, both experimentally (Madusanka et al., 2015; Sreekanth Reddy et al., 2021; Liu et al., 2024; Gonçalves et al., 2017; Araujo et al., 2023; Ruggeri et al., 2022), and computationally (Karataş et al., 2017). Moreover, previous studies investigated the ability of riboflavin to adsorb to montmorillonite clay. In vitro experiments showed that riboflavin adsorbs to montmorillonite clay at ~67% (Kihal et al., 2020), or 0.48 mmol/g (Mortland and Lawless, 1983), and additionally other studies have also shown that other clay minerals, like sepiolite (Mateos et al., 2018), or modified montmorillonite (Kaygusuz et al., 2015) are also good adsorbents for riboflavin. According to Mortland et al., the adsorption of riboflavin to clay follows a Langmuir model (Mortland and Lawless, 1983).

4. Conclusion

The goal of our study was to design amended clays augmenting PFAS binding at acidic conditions found in the stomach. MD simulations allowed for the selection of compounds with a high binding probability to clay. Subsequent simulations focused on evaluating which of these compounds could potentially bind PFAS as amendments and led us to the identification of the three top predicted supplement-amended clays: CMcaff, CMcurc, and CMribo. Experiments and simulations performed for PFAS in complex with parent clays depicted high agreement between the experimentally determined Q_max values and computationally predicted binding capacities, showing a higher binding for PFOA, followed by PFOS and then GenX. The agreement provided support for the computational setup and simulations. Experiments showed that CMcaff significantly enhanced binding for GenX and PFOS, CMribo enhanced binding particularly for GenX and PFOS, and CMcurc enhanced binding for GenX. Computational structural analysis of the simulations identified the key modes through which the particular amendments contributed to enhanced binding properties. We consider that both the novel computational library of chemical compound-amended clays and the computational approach utilized could enhance the future design of novel amended clays for other toxins. Following analogous approaches, computational and experimental studies can be used to design optimum amendments for one or combinations of toxins.

Supplementary Material

Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.compchemeng.2025.109215.

Data availability

Data will be made available on request.