Mechanistic Insights into the Selectivity for Arsenic over Phosphate Adsorption by Fe³⁺-Crosslinked Chitosan using DFT

Abstract

Fe³⁺-crosslinked chitosan exhibits the potential for selectively adsorbing arsenic (As) over competing species, such as phosphate, for water remediation. However, the effective binding mechanisms, bond nature, and controlling factor(s) of the selectivity are poorly understood. This study employs ab initio calculations to examine the competitive binding of As(V), P(V), and As(III) to neat chitosan and Fe³⁺-chitosan. Neat chitosan fails to selectively bind As oxyanions, as all three oxyanions bind similarly via weak hydrogen bonds with preferences of P(V) = As(V) > As(III). Conversely, Fe³⁺-chitosan selectively binds As(V) over As(III) and P(V) with binding energies of −1.9 eV, −1 eV, and −1.8 eV for As(V), As(III), and P(V), respectively. The preferences are due to varying Fe³⁺/oxyanion donor-acceptor characteristics, forming covalent bonds with distinct strengths (Fe-O bond ICOHP values: −4.9 eV/bond for As(V), −4.7 eV/bond for P(V), and −3.5 eV/bond for As(III)). Differences in pKa between As(V)/P(V) and As(III) preclude any preference for As(III) under typical environmental pH conditions. Furthermore, our calculations suggest that the binding selectivity of Fe³⁺-chitosan exhibits pH dependence. These findings enhance our understanding of the Fe³⁺-oxyanion interaction crucial for preferential oxyanion binding using Fe³⁺-chitosan and provide a lens for further exploration into alternative transition metal-chitosan combinations and coordination chemistries for applications in selective separations.

Graphical Abstract

1. Introduction

The efficient removal of arsenic (As) from competitive water matrices remains a pressing challenge, especially in distributed drinking water systems. As is highly toxic, non-biodegradable, and mobile even at low concentrations¹. Excessive ingestion of As can cause ulcers, skin cancer, liver, kidney, and lung deterioration^{1, 2}, and potentially longer-term neurodegenerative disorders. Therefore, the World Health Organization (WHO) recommends a maximum concentration of 10 μg of As/L in drinking water³. As is a metalloid typically present in water as pentavalent As(V) (arsenate) or trivalent As (III) (arsenite) oxyanions, where their specific protonation state and ionic charge at is controlled by the pH (the environmentally relevant pKa’s are 6.9 and 9.2 for As(V) and As(III), respectively)^4–8.

As removal technologies include oxidative processes, precipitation, ion exchange, membrane processes (reverse osmosis, nanofiltration, electrodialysis), and adsorption^{4, 9–16}. Adsorption is an attractive, inexpensive technology with great promise for under-the-sink water remediation applications, particularly in decentralized drinking water systems. Unlike some state-of-the-art technologies available for municipal scale use, adsorption requires no chemical input, is reasonably fouling resistant, and can attain high removal efficiencies^16–19. However, in most contaminated waters, the performance and longevity of conventional adsorbents are stymied by the presence of other, often less toxic, species such as phosphate, silicate, sulfate, nitrate, and bicarbonate^{16, 20, 21}. These competing species are usually chemically and structurally similar to As oxyanions (As(III) and As(V)) and may occur at significantly higher concentrations (up to 10–100 times)²²; thus, they compete with As oxyanions for non-specific sites of unselective sorbent materials resulting in decreased treated water volumes before As breakthrough^{4, 23, 24}. Selective sorbent materials could enhance As removal efficiencies and capacity²⁵, and thus lowering treatment costs by decreasing sorbent replacement frequency.

Chitosan (CS) is a biopolymer with appealing properties for a sustainable sorbent, namely its biodegradability, versatility, biocompatibility, and metal oxyanion binding capabilities^{26, 27}. The adsorptive capability of CS is primarily due to the abundance of functional groups, including amine and hydroxyl groups, on its molecular structure, which serves as binding sites for target pollutants in water²⁸. For example, previous studies revealed that CS binds anionic compounds, such as As(V), at protonated amine sites via electrostatic attraction^{29, 30}. However, despite binding oxyanions, CS is non-selective in complex water matrices, i.e., those containing a mixture of pollutants, inhibiting its toxin removal capacity³⁰.

Recent attention has focused on inducing selectivity in CS sorbents. For example, Yamani et al. revealed that including Cu²⁺ as a crosslinking agent in CS enables selective As(V) adsorption over phosphate (P(V))^{31, 32} despite the well-known similarity in the oxyanion structure, chemical nature, and binding behaviors^33–38. They attributed the selectivity to mild differences in molecular size, partial charge densities, and polarizabilities between the oxyanions, which cause more favorable electrostatic interactions between Cu²⁺-chitosan and As(V) than phosphate P(V)^{31, 32}. However, Cu²⁺-chitosan only selectively binds the more challenging As(III) after pre-photooxidation, which increases the complexity of point-of-use remediation systems^{25, 32}. Pincus et al. examined other crosslinker ions and reported that while Fe³⁺, Ni²⁺, and Zn²⁺-chitosan systems outperformed Cu²⁺ and Al³⁺ in terms of the quantity of As removed, only Fe³⁺ achieved selectivity for both As species over P(V) due to differences in inner-complexation behaviors³⁹.

Density functional theory (DFT) has been applied to predict the molecular and electronic structures of chitosan oligomers of various lengths^{40, 41}. Rahangdale et al. studied the binding of organic molecules by molecularly imprinted chitosan⁴². The binding of heavy metal cations to neat chitosan has also been examined^43–49. Although there are several computational studies of metal cation binding to chitosan, there has been much less attention to heavy metal oxyanion binding to neat and modified chitosan. Mishima et al. studied chromate binding on epichlorohydrin and glutaraldehyde crosslinked chitosan⁵⁰. Despite their experimental promise, transition-metal crosslinked chitosan (TMC) has not been extensively examined with computational models for their ability to bind toxic oxyanions selectively. To our knowledge, only Pincus et al. have attempted to examine these systems using DFT. In their study, they used DFT to predict the geometric structure of As(III), As(V), and P(V) binding to Cu²⁺, Ni²⁺, and Fe³⁺ as a reference point to explain Extended X-ray Absorption Fine Structure (EXAFS) results. While this study successfully showed that the oxyanions bind in an inner sphere configuration to TMCs in agreement with the experimental results, the relative energetics, binding preferences, underlying electronic behavior, and pH effects were not reported or explicitly considered³⁹. These properties are crucially important due to the sensitivity of binding to pH effects and considering the use of pH swings for adsorption/desorption cycles⁵¹. A better understanding of the underlying mechanism that leads to differences between As and P binding could enable directed design of these materials to improve selectivity by increasing desirable electronic and geometric structures.

Therefore, in this study, we closely examine the competitive binding of As(V), As(III), and P(V) to Fe³⁺-chitosan using DFT calculations to identify the underlying causes for their relative selectivity. The findings of this work elucidate the role of the transition-metal ion in strong and selective oxyanion binding and provide new knowledge that can be applied to separate similar oxyanion species based on mild differences in oxyanion ligand characters.

2. Computational details

2.1. First-principles calculations

DFT calculations were performed using the Vienna Ab Initio Simulation Package (VASP)^52,53, with periodic boundary conditions. Calculations were completed with the HSE06 hybrid meta-generalized gradient approximation (GGA) functional⁵⁴ because the pure GGA Perdew-Burke-Ernzerhof (PBE) functional⁵⁵ overestimated the binding energies of the species by up to 0.4 eV compared to HSE06, see Fig. S1. The Projector Augmented Wave (PAW) pseudopotential approximation⁵⁶ lowered computational costs by only explicitly including valence electrons. PAWs explicitly calculated transition metal’s 4s and 3d orbitals, while non-metals (C, N, and O) describe the 2s and 2p orbitals, and H, the 1s orbital. A summation of plane waves with kinetic energies up to 600 eV composed the wave functions. Including higher energy waves (700eV) resulted in only a 0.05 eV and 0.03 eV improvement in total binding energies and binding energy differences, respectively. The Brillouin zone was sampled at the Γ-point for all calculations because there was no significant improvement in calculated binding energies with higher k-point meshes, e.g., going from a $1 \times 1 \times 1$ to a $2 \times 1 \times 1$ k-mesh for the crosslinked polymer model resulted in only a 0.01 eV decrease in binding energy. All calculations were spin-polarized, while Van der Waals dispersion forces were accounted for via the empirical DFT-D3 method of Grimme et al.^{57, 58}. Implicit water solvent was implemented in the system via the polarizable continuum model (PCM)⁵⁹ (implemented in VASP) to account for solvation effects on the binding of species^59–61.

CS was modeled as a periodic chain of tetramers (~21.15Å) linked to its periodic images via a β-oxygen linkage as shown in Fig. S2. Fe³⁺ crosslinking of the Fe³⁺-chitosan (Fe³⁺-CS) system occurred between proximal amine and hydroxyl groups of a CS tetramer and a dimer. We used a tetramer chitosan model for our study because increasing the model beyond this size did not significantly alter the local metal-oxyanion interactions of interest at the metal site, either chemically or sterically. Increasing from the tetramer to a hexamer model resulted in only a 0.05 eV difference in the predicted selectivity between As(V) and P(V). Hence, the tetramer model was used for all calculations in this study. Two orientations of chitosan chains relative to each other were sampled for the system: parallel (ϴ = 180°) and perpendicular (ϴ = 90°). The configurations are shown in Fig. 1a, b, while the oxyanions are shown in Fig. 1c. These orientations were chosen because they are extreme cases for the crosslink between the chitosan oligomers at the minimum energy amine sites of chitosan. We note that there exists a continuum of possible rotation angles and configurations at the crosslink site, which is prohibitively computationally expensive to sample using DFT methods. Therefore, we have selected the extreme cases for analysis. A more detailed sampling of the TMC mesoscale structure for other possible crosslink configurations using molecular mechanics of larger system sizes is being conducted as part of follow-up work. In the parallel orientation, the axis of the CS dimer is ~180° rotated from the axis of the tetramer chain, and the crosslink site consists of a square planar coordination complex between Fe³⁺ and the CS molecules’ neighboring N and O atoms. In the perpendicular configuration, the CS dimer is rotated by 90° along the cell y-axis, resulting in a smaller N-Fe-N bond angle at the crosslink site. The parallel orientation was lower in energy than the perpendicular by −0.1 eV and was used for all subsequent binding calculations. A vacuum space > 15Å was applied in the direction normal to the crosslinking plane for all systems to avoid interactions across the periodic images of the simulation cells in that direction, while the length of the CS chain ensures similar separation in the other directions. Molecular structures were relaxed until atomic forces were less than 0.001eV/Å.

Figure 1.

Molecular models of Fe³⁺-CS in the (a) parallel, and (b) perpendicular configurations, (c) Molecular models for As(V), As(III) and P(V), respectively. H, O, N, C, P, Fe, and As are represented by white, red, light blue, dark brown, purple, light brown, and green spheres, respectively.

Oxyanion binding on neat CS was studied at an amine binding site due to more favorable binding interactions than a hydroxyl site. As(V) and P(V) bind more favorably at the amine site than at a hydroxyl site by 0.1 eV. These results are consistent with previous experimental findings^{30, 62}. Different configurations for oxyanion binding involving Fe³⁺ and nearby hydroxyl sites were considered for the Fe³⁺-CS system; we only report on the lowest energy configuration for each oxyanion.

Binding energies for all systems were calculated using equation 1.

$$\Delta E_{\mathit{\text{bind}}}= E_{CS/F{e}^{3+}*oxo}- E_{CS/F{e}^{3+}+oxo}$$ 1

where $\Delta E_{\mathit{\text{bind}}}$ is the binding energy for oxyanion, $E_{CS/F{e}^{3+}+oxo}$ is the total energy of the system with oxyanion (As(V), As(III) or P(V)) removed from CS/$F{e}^{3+}$ in the same simulation box and $E_{CS/F{e}^{3+}*oxo}$ is the system’s total energy with the oxyanion bound at a CS site or Fe³⁺ for Fe³⁺-CS. This procedure was necessary to achieve charge consistency across all simulation cells and eliminate the potential discrepancies that may arise from using a background Jellium charge. A negative value for the binding energy represents an exothermic process, while a positive one corresponds to an endothermic process.

2.2. Chemical bonding analysis

Crystal Orbital Hamilton Population (COHP) analysis was performed using the Lobster v3.1⁶³ and Pymatgen packages^{64, 65}. COHP weights orbital energies from the Density of States (DOS) by the corresponding elements of the Hamiltonian and thus partitions the band-structure energy into bonding, nonbonding, and antibonding contributions. Thus, the Integrated Crystal Orbital Hamilton Population (ICOHP) up to the Fermi level quantifies the strength of the chemical bond between atomic species. Because the ICOHP value indicates the energy associated with the interaction between atoms and orbitals, we use it to characterize the relative extents of metal-ligand back-donation as a measure of the π-acceptor character of oxyanions in our transition-metal system.

2.3. As(V) removal prediction

Using relevant equilibrium relationships, we predicted the percentage removal of As(V) on Fe³⁺-CS as a function of pH. The Gibbs free energies of binding were calculated as follows:

$$\Delta G_{\mathit{\text{bind}}}= \Delta E_{\mathit{\text{bind}}}-T\Delta S - 2.3RT\left(pH-pKa\right)$$ 2

where $\Delta E_{\mathit{\text{bind}}}$ is the electronic binding energy calculated from eq. 1 for various As(V) and P(V) molecules. $R$ is the gas constant, $T$ is the temperature, taken to be 298K, and the third term accounts for the pH-dependent free energy for deprotonated oxyanions. We assumed that an oxyanion loses half of its aqueous phase entropy upon binding; hence $\Delta S$ was taken as half the negative of the free ion entropy in solution $S_{aq}$ (from refs.^{66, 67}). We performed a sensitivity analysis to ascertain the consequence of assuming half solution-entropy loss on selective removal of As(V) and P(V) by Fe³⁺-CS. As shown in Fig. S9, we find that the selective oxyanion removal is only slightly influenced by the amount of solution phase entropy retained upon binding, except for the extreme and unrealistic case of no entropy loss. A lower onset pH for selective As(V) removal is predicted when more entropy is retained upon binding: 52% As(V) removal at pH 6.8, 58% at pH < 6, and 87% at pH < 6, for 100%, 50%, and no entropy loss, respectively. Moreover, we obtain the closest match with the experimentally measured competitive As(V) removal³⁹ amounts with 50% entropy loss. We note that while this estimation shifts the quantitative answers, it does not alter the overall characteristics of our findings. We only computed $\Delta G_{\mathit{\text{bind}}}$ for the di-protonated and mono-protonated As(V) and P(V) species ($H_{2}X{O}_4^{-}$ and $HX{O}_4^{2-})$ since those are present in environmental conditions. Equilibrium binding constants were calculated using

$$K=\exp \left(\frac{-\Delta G_{\mathit{\text{bind}}}}{RT}\right)$$ 3

where K is the equilibrium constant for a given oxyanion species. Equilibrium concentrations of the two oxyanion protonation states were calculated using the pKa values for As(V) and P(V) (6.9 and 7.2, respectively), and the relationship:

$$pKa=pH+\log \left(\frac{[H_{2}X{O}_4^{-}]}{\left[HX{O}_4^{2-}\right]}\right)$$ 4

To examine the competition between the As(V) and P(V) species, the fractional occupation of the binding sites for a given As(V) species with the number of protons $y$, was estimated using equation 5 as in⁶⁸.

$${\theta }_{As,y}= \frac{K_{As}\left[H_{y}As{O}_4^{y-3}\right]}{1+K_{As,y}\left[H_{y}As{O}_4^{y-3}\right]+ K_{P,y}\left[H_{y}P{O}_4^{y-3}\right]}$$ 5

3. Results and discussion

To understand the differences in adsorptive selectivity of CS and Fe³⁺-CS, we used DFT to calculate the binding energies of As(V), As(III), and P(V), the closest structural analog competitor to As(V). We first discuss the binding to CS and Fe³⁺-CS of the semi-protonated states of As(V) and P(V) ($H_{2}As{O}_4^{-}$ and $H_{2}P{O}_4^{-}$) and the fully protonated state of As(III) ($H_{3}As{O}_3$) because these are the most present species at neutral pH (~7). Subsequently, we report on the effects of oxyanion protonation on binding.

3.1. As(V) binding to neat chitosan

We calculate that neat CS binds As(V) exothermically by −0.2 eV, as shown in Fig. 2a. The optimized binding configuration, shown in Fig. 2b, reveals that As(V) binds in a bidentate configuration via weak H-bonding to a CS amine site and a neighboring OH site. When As(V) is unbound and far from the binding site, there is minimal overlap between the electronic states of the As(V) O-atom and the CS H-atom, Fig. 2c. The As(V) O-atom states populate the higher energy regions of the system near the Fermi, while the H-atom states populate lower energy regions. Subsequently, when As(V) is within proximity of the binding site, there is a slight redistribution and overlap of electronic states between As(V) O atom and CS H-atoms, indicative of weak interaction, shown in Fig. 2d. The weak interaction results in a mean H⋯O bond distance (1.76 Å), which is slightly longer than the typical H⋯O hydrogen bond (~1.7Å⁶⁹) and nearly double O-H covalent bond length (~0.98 Å), indicating that neat CS binds As(V) via weak H-bonding. There is no significant charge transfer between As(V) and CS (only −0.08$e^{-}$) and we calculate an ICOHP value of −0.9 eV/bond for the H⋯O bond, consistent with the negligible overlap of states and weak binding energy. Overall, we predict that neat CS is a poor binding agent for As(V), as is seen experimentally²⁴, due to the absence of significant chemical bonding, which relegates CS-As(V) interactions to weak H-bonds.

Figure 2.

(a) Oxyanion binding energies on neat CS, (b) Optimized structure for CS-As(V), DOS plots for (c) before and (d) after As(V) binding to neat CS. H, O, N, C, and As are represented by white, red, light blue, dark brown, and green spheres, respectively.

CS binds P(V) in a nearly identical manner to As(V) with a binding energy of −0.2 eV (Fig. 2a). P(V) also binds in a bidentate configuration with neighboring amine and hydroxyl sites with a mean CS-P(V) bond distance (1.81 Å) only 0.05 Å longer than that of As(V), in agreement with the equal binding energies for both oxyanions, as shown in Fig. S3. In addition to the CS-P(V) bond distance, which shows that CS also binds P(V) via H-bonding, the PDOS reveals a similar electronic structure for the CS-P(V) interaction like As(V), indicative of its weak covalent character, see Fig. S3. We also calculate nearly equal charge transfer −0.14$e^{-}$ and mean ICOHP value, −0.8 eV/bond, respectively, for P(V) and As(V), in agreement with the similar binding energies, confirming that both oxyanions bind with equal strength and mechanism to CS. Fig. 3 shows the COHP plots for binding both oxyanions to neat CS. We attribute the similarity in binding behaviors between P(V) and As(V) to their nearly identical structure and charge on O atoms, which equates to highly similar electrostatic interactions with CS. Our results agree with the experimental work by Yamani and coworkers³¹, who showed low removal efficiencies for As(V) with background P(V) using neat chitosan.

Figure 3.

COHP plots for the H⋯O H-bond for (a) As(V), (b) P(V), and (c) As(III) bound to neat CS. Inset are ICOHP values that characterize bond strengths.

CS binding of As(III) is weaker than the other oxyanions with a binding energy of only −0.1 eV. The weak As(III) binding is characterized by a longer H⋯O bond distance of 2.19 Å compared to the shorter bonds for As(V) and P(V), 1.76 Å and 1.81 Å, respectively. Further, the lack of charge transfer and less negative ICOHP value (−0.2 eV/bond) indicates no chemical bond was formed. CS binds As(III) less preferentially than As(V) and P(V) because of its weaker electrostatic attraction to the charge neutral As(III) (H₃AsO₃) compared to the negatively charged As(V) and P(V) ($H_{2}As{O}_4^{-}$ and $H_{2}P{O}_4^{-}$). Overall, we predict relative binding preferences of P(V) = As(V) > As(III) with neat chitosan.

3.2. As(V) binding to Fe³⁺-chitosan

We investigated As and P(V) binding to Fe³⁺-CS. The crosslinking of CS with Fe³⁺ introduces high-energy Fe 3d-states into the Fe³⁺-CS electronic structure near the Fermi level, shown in Fig. 4a. The geometry of the chain is unchanged, with only a minor reorganization of two internal H-bonds of the CS tetramer in the vicinity of the crosslink. Additionally, two new H-bonds were formed between the tetramer and the dimer.

Figure 4.

(a) DOS plot before As(V) binding to Fe³⁺-CS, (b) Oxyanion binding energies on Fe³⁺-CS, (c) DOS plot for As(V) bound to Fe³⁺-CS, (d) Schematic diagram of the orbital interactions involved in the Fe-O bond between Fe³⁺-CS and As(V). The diagram illustrates the back-donation of electron density from Fe 3d orbitals to As(V) π* antibonding orbitals, (e) Optimized structure for As(V) bound to Fe³⁺-CS. H, O, N, C, Fe, and As are represented by white, red, light blue, dark brown, light brown, and green spheres, respectively.

As(V) binds to Fe³⁺-CS in a profoundly different manner than the neat CS case. As(V) binds in a bidentate configuration to the Fe³⁺ center and an OH group and does so nearly ten times more strongly (−1.9 eV) than neat CS (see Fig. 4b). The increased binding strength is due to increased coulombic attraction between the negatively charged As(V) and the positively charged Fe³⁺ and to additional chemical bonding. Two bonds form: a covalent bond between an O atom of As(V) and Fe³⁺ and an H-bond between As(V) and a nearby OH group on CS. A similar mixture of inner-sphere metal-oxyanion and CS-oxyanion H-bonds has been predicted for As(V) binding to Fe³⁺-CS^{39, 70}. Further, we note that the inner sphere complex is via a covalent bond, unlike previously reported³⁹. The covalent bond forms due to the availability of similarly high energy O electronic states from As(V) and Fe³⁺ states near the Fermi level. Before binding, both O states from As(V) and Fe states form the near-Fermi region, as shown in Fig. 4a. These electronic states redistribute upon As(V) binding (see Fig. 4c), allowing mixing and the formation of a set of bonding and antibonding molecular orbitals. The electron density donated by the O atom forms the σ-bond with the Fe.

Meanwhile, Fe³⁺ back-donates electron density, partially filling the As-O π* anti-bond and forming additional bonding and antibonding states. The back-donation of electrons also lengthens the As-O bond of the O bound to Fe³⁺ by 0.05Å, consistent with the partial filling of their antibonding orbitals⁷¹. The interaction between the O 2p and Fe 3d electrons yields a strong coordinated Fe-O bond between As(V) and Fe³⁺ (ICOHP = −4.9 eV/bond) accompanied by lowered overall electronic energies that stabilize these states^{72, 73}. The ICOHP value for the 3d-π* orbital interaction from back-donation is −0.8 eV/bond. The orbital interactions in the Fe-O bond are shown schematically in Fig. 4d. COHP analysis corroborates the filling of these π* antibonding states in agreement with the partial charge plots for the participant bands near the Fermi level, as shown in Fig. 5a. The strong interaction causes the Fe-O bond length to be 1.89 Å, as shown in Fig. 4e. This Fe-O bond length is only 0.06Å shorter than the length deduced from the second shell fit of the Fe³⁺-As(V) complex (1.95Å) obtained via EXAFS analysis³⁹. In addition, there is excellent agreement between our DFT calculations and EXAFS for the As-O bond of As(V) when bound to Fe³⁺-CS, 1.72Å and 1.70–1.71ų⁹, respectively. Relevant bond lengths in the Fe³⁺-As(V) complex are summarized in Table 1.

Figure 5.

COHP plots for the Fe-O bond for (a) As(V), (b) P(V), and (c) As(III) bound to Fe³⁺-CS. Inset are ICOHP values that characterize bond strengths and partial charge plots for the highlighted filled bonding and antibonding regions for As(V).

Table 1.

Bond lengths and binding energies for As(V), As(III), and P(V) bound to Fe³⁺-CS

Oxyanion	pH	Species	ΔEbind (eV)	As/P-Fe distance (Å)	As/P-O bond length (Å)	Fe-O bond length (Å)	Ref
As(V)	-	H2AsO4−	−1.9	3.23	1.72	1.89	This work
	6	H2AsO4−	-	3.29	1.70–1.71	1.95	[39]
As(III)	-	H3AsO3	−1	3.44	1.92	2	This work
	6	H3AsO3	-	2.91	1.79–1.8	1.61	[39]
P(V)	-	H2PO4−	−1.8	3.14	1.55	1.89	This work

3.3. As(III) and P(V) binding to Fe³⁺-chitosan

P(V) also binds nine times more favorably to Fe³⁺-CS, −1.8 eV, than neat CS. Like As(V), P(V) binds to Fe³⁺ in a bidentate configuration via an Fe-O covalent bond and an H-bond. The Fe-O bond length between P(V) and Fe³⁺-CS, 1.89 Å, is also within the covalent Fe-O bond length, as shown in Fig. S4. Prior to binding, P(V) has a similar electronic structure as As(V) with high energy O and Fe states near the Fermi level available for mixing and Fe³⁺-P(V) complex formation (see Fig. S4). Upon P(V) binding, there is a redistribution and overlap of the P(V) O states and Fe states to form a set of bonding and antibonding molecular orbitals and, thus, a covalent bond. The interactions stabilize electronic energies, causing a strong coordinate Fe-O bond between P(V) and Fe³⁺ (ICOHP = −4.7 eV/bond) consistent with the calculated bond length (1.89 Å). Like As(V), filled π* antibonding states below the Fermi level are seen in the COHP plot (Fig. 5b) for the Fe-O bond between P(V) and Fe³⁺ which again highlight that the bond is strengthened by Fe³⁺ back-donation (ICOHP for 3d-π* orbital interaction = −0.7 eV/bond), partially filling the As-O π* antibonding states.

Fe³⁺-CS shows a slight energetic preference (0.1 eV) for As(V) over P(V), which we attribute to a difference in their π-acceptor strengths (back-donation extents) and molecular size that cause stronger Fe-O and H-bonds for As(V) than P(V). We quantify back-donation extents and thus π-acceptor strengths via the ICOHP values for the 3d-2p orbital interaction in the Fe-O bond of the Fe³⁺-oxyanion complex, which has a more negative value (stronger) for As(V) (−0.8 eV/bond) than P(V) (−0.7 eV/bond). In addition, the H-bond is slightly shorter and stronger (1.63Å; ICOHP=−1.32 eV/bond) for As(V) than P(V) (1.67Å; ICOHP=−0.57 eV/bond). Furthermore, Bader charge analysis reveals less mean electron densities on the As(V) O-atoms than on P(V) O-atoms (−0.27$e^{-}$), suggesting that the less polarized oxyanion is a stronger π-acceptor in this metal-ligand interaction.

As(III) also binds more strongly to Fe³⁺-CS, −1 eV, than neat CS but less strongly than the other two anions. As(III) binds in a monodentate configuration, unlike As(V) and P(V), and binding is less preferred by 0.9 and 0.8 eV, respectively. This weaker binding causes a longer Fe-O bond (2Å) for As(III) (see Fig. S4) than for As(V) and P(V), 1.89Å and 1.89Å, respectively. The Fe-O bond length for the Fe³⁺-As(III) complex is 0.39Å longer than the length reported for the second shell fit of the Fe³⁺-As(III) complex (1.61Å) obtained via EXAFS analysis³⁹. The discrepancy in bond lengths may be due to the ~0.2Å variance in interatomic distances during EXAFS fitting noted by the authors. However, there is better agreement between our DFT calculations and EXAFS for the As-O bond of As(III) when bound to Fe³⁺-CS, 1.92Å and 1.79–1.8ų⁹, respectively.

As(III) exhibits a distinct electronic structure from As(V) and P(V), both before and after binding to Fe³⁺-CS. In contrast to the charged pair, the Fe³⁺-CS-As(III) system is such that the near-Fermi region is populated mainly by Fe states with little contribution from As(III) O states. Most As(III) O states are significantly lower in energy than the Fe states and thus are further removed from the Fermi level, as shown in Fig. S4. The distribution of the As(III) O states indicates higher energetic stability and hence, less reactivity of the O atoms than those of As(V) and P(V). The energetic mismatch between Fe³⁺ and As(III) states results in less mixing of the metal-ligand states during complex formation compared to As(V) and P(V), ultimately resulting in weaker binding. This is in agreement with the less negative ICOHP value (weaker bond) for the Fe-O bond for As(III), −3.5 eV/bond, than for As(V) and P(V), −4.9 eV/bond and −4.7 eV/bond, respectively.

We attribute the differences in binding strength to the charge neutrality of As(III) compared to the negatively charged As(V) and P(V), which causes the latter to interact more favorably electrostatically and chemically with the positively charged Fe³⁺. The stronger binding preference of P(V) over As(III) we have calculated (0.8eV) is in contrast to the slightly preferential removal of As(III) in the presence of P(V) measured by Pincus et al.³⁹. The disagreement between our results and the experiment may be due to our consideration of the TMC binding sites in isolation which does not account for possible electrostatic repulsions and thus weakened binding between adsorbed anionic species (As(V) and P(V)) at neighboring sites. Such repulsive forces would likely be significantly less for the charge-neutral As(III) and thus have less effect on its binding. Further, the large experimental error reported in the experimental work could indicate large inhomogeneities associated with As(III) binding, which makes chemically accurate predictions difficult. Nonetheless, our model does indicate that Fe³⁺ significantly suppresses the preference for P(V), in this case by 0.2eV, relative to neat CS. Overall, the presence of P(V) is expected to impede As(III) adsorption on Fe³⁺-CS more severely than As(V), in agreement with the previous findings of Pincus and coworkers³⁹.

3.4. Effects of protonation on oxyanion binding to neat chitosan

We examined the role of pH and oxyanion charge state on binding affinity and selectivity by calculating binding energies for As(V), As(III), and P(V) binding to neat CS as a function of oxyanion protonation extent, i.e., fully protonated ($H_{3}X{O}_y$), di-protonated ($H_{2}X{O}_y^{-}$), mono-protonated ($HX{O}_y^{2-}$) and deprotonated ($X{O}_y^{3-}$) where X represents As or P, while $y$ represents the number of O atoms (4 for As(V) and P(V), and 3 for As(III)).

Oxyanion deprotonation destabilizes the O electronic states (increase in energy), as highlighted by the upward shifts of those states relative to the Fermi level with decreasing protonation extent. This is shown in the PDOS plots of Fig. 6 for As(V). The closer the oxyanion electronic states are to the Fermi level, the more basic (reactive) the oxyanion⁷⁴. As expected, we calculate that As(V) binds more strongly with decreasing protonation extent, 0.2 eV, −0.2 eV, −0.3 eV, and −0.8 eV for H₃AsO₄, H₂AsO₄⁻, HAsO₄²⁻, and AsO₄³⁻, respectively, as shown in Fig. 7a, consistent with the decreased stability of the O states and increased electrostatic interactions with decreasing protonation extent. The fully protonated As(V), having the most stabilized O states, does not bind favorably with neat CS (0.2 eV), while the most destabilized deprotonated As(V) binds most strongly, −0.8 eV. The higher bond strengths with decreasing protonation extent cause shorter H-bonds, from 2.31 Å for the fully protonated state to 1.58 Å for the deprotonated state. The optimized structures for all As(V) species bound to neat CS are shown in Fig. S5.

Figure 6.

DOS plots for the different As(V) species in the Fe³⁺-CS system (a) H₃AsO₄, (b) H₂AsO₄⁻, (c) HAsO₄²⁻, and (d) AsO₄³⁻.

Figure 7.

(a) Binding energies for As(V), As(III) and P(V) bound to neat CS at various protonation states (b) Binding energies for As(V), As(III) and P(V) bound to Fe³⁺-CS at various protonation states.

P(V) exhibits an identical monotonic trend in binding strength with protonation extent, 0.2 eV, −0.2 eV, −0.3 eV, and −0.9 eV for H₃PO₄, H₂PO₄⁻, HPO₄²⁻, and PO₄³⁻, respectively, as shown in Fig. 7a. The optimized structures for all P(V) species bound to neat CS are shown in Fig. S6. The chemical physics governing the P(V) trend is similar to that of As(V). Notably, neat CS has no preference for As(V) over P(V) at the various protonation states, and hence, we expect both species to be bound equally at all pH conditions and P(V) to inhibit As(V) removal.

Our calculations show that As(III) exhibits a nonmonotonic trend in binding strength with decreasing protonation extent on neat CS, −0.1 eV, −0.4 eV, −0.5 eV, and −0.2 eV for H₃AsO₃, H₂AsO₃⁻, HAsO₃²⁻, and AsO₃³⁻, respectively, shown in Fig. 7a. Optimized structures for all As(III) species bound to neat CS are shown in Fig. S7. As(III) is preferred over both As(V) and P(V) in all but the deprotonated state, where it is less preferred. We attribute this nonmonotonic behavior to the difference in As(III) molecular structure between the deprotonated and other states. The deprotonated As(III) has larger O-As-O bond angles (105°) compared to the other protonation states (100°), which strains the As(III) molecule and weakens its binding. This strained state is unlike As(V) and P(V), whose bond angles are preserved in all protonation states, and thus their binding energy trends are monotonic.

Furthermore, the deprotonated As(III) binds in a monodentate configuration at an amine site, while both As(V) and P(V) bind bidentate to an amine and neighboring hydroxyl site and thus bind more strongly. We note that we tested multiple binding configurations, and these structures are the most stable ones. Despite the more favorable energetics we have calculated for As(III) compared to As(V) and P(V) at the fully, di, and monoprotonated states, these preferences occur in protonation states that are very unlikely for As(III) given its pKa of 9.2 relative to both As(V) and P(V), 6.9 and 7.2, respectively. Hence, we expect As(V) and P(V) to bind more favorably than As(III) to neat CS in environmental conditions where As(III) is fully protonated but As(V) and P(V) are semi-protonated.

3.5. Effects of protonation on oxyanion binding to Fe³⁺-chitosan

Complementary calculations for binding the different oxyanion protonation states were performed for the Fe³⁺-CS system. We calculate that binding strength increases monotonically with decreasing protonation extent for As(V) bound to Fe³⁺-CS (Fig. 7b). The fully protonated state binds endothermically by 0.2 eV, while the deprotonated state binds by −3.2 eV. This binding energy trend correlates with the mean Fe-O bond strengths of the different protonation states, which have ICOHP values of −2.6, −4.9, −11.4, and −12 eV/bond for the fully, di, mono, and deprotonated states, respectively. The increased Fe-O bond strengths with decreasing protonation extent caused shorter Fe-O bonds, from 2.18 Å for the fully protonated state to 1.84 Å for the deprotonated state. We also note differences in binding configurations where the fully protonated As(V) binds monodentate, but the di, mono, and deprotonated states bind in bidentate configurations, resulting in stronger bonds for those states. Again, we attribute this binding energy trend to the differences in electronic stability among the oxyanion protonation states. The closer the oxyanion electronic states are to the Fermi level, the more basic (reactive) the oxyanion⁷⁴, causing the more favorable binding of the less protonated As(V) states.

P(V) behaves nearly identically to As(V). P(V) binding strength to Fe³⁺-CS increases monotonically with decreasing protonation extent. P(V) binds by 0.4 eV in the fully protonated state, while it binds exothermically by −3 eV when deprotonated. Thus, As(V) is preferred over P(V) at all protonation states. We attribute the preference for As(V) over P(V) to the differences in their π-acceptor character, which cause As(V) to form more stabilized molecular orbitals in the Fe-O bonds with Fe³⁺-CS than P(V), highlighted by the more negative 3d-2p orbital ICOHP values for As(V) than P(V), shown in Fig. 8a. These binding energetics thus suggest that Fe³⁺-CS would prefer As(V) over P(V) in a wide pH range across protonation states.

Figure 8.

(a) ICOHP for 3d-π* interaction in the Fe-O bond for As(V), As(III) and P(V) bound to Fe³⁺-CS at various protonation states; secondary y-axis shows As(V) binding energies at the various protonation states (b) Predicted % As(V) and P(V) removed as a function of pH with 1:6 As(V)/P(V) ratio for Fe³⁺-CS. Solid line is for As(V) while dashed line is for P(V).

As(III) binding energies to Fe³⁺-CS are nonmonotonic, being −1 eV, −2.1 eV, −2.6 eV, and −2.5 eV for H₃AsO₃, H₂AsO₃⁻, HAsO₃²⁻, and AsO₃³⁻, respectively. As(III) is preferred over both As(V) and P(V) in the fully and di-protonated states. Like neat CS, the nonmonotonic As(III) binding trend arises from its stretched O-As-O bond angles that induce strain and weaken binding in the mono- and deprotonated states. Just as in the case of neat CS, although As(III) has more favorable binding energetics than As(V) and P(V) at the fully and di-protonated states, protonation state differences of the ions based on their pKa means that the energetic effects at equivalent degrees of protonation are unrealistic. We, therefore, expect As(V) and P(V) to bind more favorably than As(III) to Fe³⁺-CS at environmental conditions where As(III) is fully protonated and As(V) and P(V) are di- or mono-protonated.

Using our calculated binding energies, we predicted As(V) and P(V) uptake of Fe³⁺-CS as a function of pH, assuming an excess of anions compared to the number of Fe³⁺ binding sites and a 1:6 molar ratio of As(V) to P(V) concentration, following the experimental work of Pincus et al.³⁹. As(III) was neglected because of the strong preference for As(V) and P(V) over As(III) at environmental conditions as dictated by pKa differences. The predicted removal of oxyanions is presented in Fig. 8b, highlighting the As(V)/P(V) preference. This preference is delineated into three regimes: 1) pH < 6, where di-protonated states predominate; 2) pH between 6 and 8, characterized by varying protonation extents and equilibrium binding constants; and 3) pH > 8, where monoprotonated states dominate. In Regime 1, the competition between As(V) and P(V) is quantified using Equation 5, resulting in approximately 58% occupation of Fe³⁺ sites by As(V) due to its more favorable binding free energies (Fig. S8). Regime 2 is complex due to multiple ion species and varying binding free energies with protonation extent, leading to steady As(V) uptake achieving selective As(V) removal of ~61% by Fe³⁺-CS at pH 6. In Regime 3 (pH > 8), where monoprotonated states prevail, As(V) dominates with over 95% occupation, consistent with experimental data showing ~59% selective As(V) removal at pH 6 by Pincus et al.³⁹.

In summary, our findings indicate that selective separation of As(V) from P(V) is achievable with Fe³⁺-CS under a wide range of pH conditions. However, the adsorption of As(III) is hindered by P(V) due to disparities in oxyanion pKa and charge. Therefore, there is a need to explore modifications that enhance the preference for As(III) and increase overall As removal capacity. Nevertheless, the predicted selectivity of As(V) over P(V) highlights the practical potential of Fe³⁺-CS. Moreover, this selectivity is maintained even in excess of background P(V), underscoring the resilience and robustness of Fe³⁺-CS as a promising sorbent for As removal applications.

4. Conclusions

In this work, we have systematically utilized ab initio calculations to study the binding of As(V), P(V), and As(III) to neat chitosan and Fe³⁺-chitosan. Our results show that at environmental conditions (with As(V) and P(V) in their deprotonated states and As(III) fully protonated), neat chitosan has no preference for arsenic over phosphate because the oxyanions bind to chitosan via weak H-bonds. Crosslinking chitosan with Fe³⁺ introduces high-energy 3d states for chemical bonding with oxyanion 2p states of similar energy. Fe³⁺-chitosan, therefore, binds the three oxyanions up to ten times more strongly via coordinate covalent bonding using the 3d metal and 2p ligand electrons in the metal-ligand complex. We calculate that at environmental conditions, Fe³⁺-chitosan favors As(V) over P(V) because of the more favorable Fe back-donation to As(V) than to P(V), which causes stronger Fe-O-As(V) bonds. This favorability arises because As(V) is a stronger π-acceptor than P(V). Conversely, both As(V) and P(V) are preferred over As(III) due to the charge neutrality of As(III) compared to the negatively charged As(V) and P(V), which causes the charged pair to form stronger Fe-O bonds with Fe³⁺-chitosan.

We further elucidate a pH-dependent binding of arsenic by Fe³⁺-chitosan and predict the selective removal of As(V) over P(V) from aqueous solutions in a wide pH range, which increases at elevated pH. While we calculated a binding energy preference for As(III) over both As(V) and P(V) in some protonation states, we note it is of little consequence at environmental conditions due to differences in oxyanion pKa’s. These findings explain the improved performance of Fe³⁺-chitosan for the selective removal of arsenic over phosphate and present an opportunity to leverage the new understanding in the rational design of selective organometallic sorbent materials by exploiting metal-ligand interactions. Specifically, different transition metal-ligand combinations and coordination environments yielding stronger π-donors should be explored to leverage the difference in π-accepting properties of arsenate and phosphate. This concept can also be extended to selectively removing other heavy-metal oxyanions with identical interfering ions from water.