Proton-coupled electron transfer modulates the metal release of blood serum iron transferrin

Abstract

Serum transferrin (sTf) is a key iron-transport protein in vertebrates, exhibiting an extraordinary affinity for Fe(III). Typically, only ∼30% of sTf is saturated with Fe(III), leaving a significant fraction of its binding sites available for other metal ions, including heavy metals and radionuclides. While iron release under endosomal pH is well-understood to proceed via protonation mechanisms, the release pathways at physiological pH remain less clear and are subject to multiple competing mechanisms. To address this, we employed extensive multi-scale modelling—combining molecular dynamics, metadynamics, and electronic structure calculations—to probe Fe(III) release under physiological conditions. Our investigations focused on three key pathways: direct protonation, one-electron reduction, and proton-coupled electron transfer (PCET). Calculated reduction potentials of approximately 1.3 V for both synthetic and protein models indicate that direct reduction is thermodynamically unfavourable, consistent with experimental observations.

---

Introduction

Among all transition metals, iron plays a vital role in a wide range of biological processes. Humans require approximately 8–12 mg day⁻¹ of dietary iron to support essential functions, including cell growth, oxygen storage and transport, and DNA replication.^1,2

Transferrin is an 80 kDa glycoprotein with an exceptionally high affinity for Fe(III), primarily functioning in iron transport to cells.^3–6 It exists in three main subgroups: lactoferrin, found in milk; ovotransferrin, found in avian egg whites; and serum transferrin (sTf), present in blood plasma.

The sTf protein comprises a single polypeptide chain of 679 amino acid residues, organised into two homologous lobes—N-lobe and C-lobe—each subdivided into two domains (NI, NII and CI, CII, respectively). Each lobe contains a cleft that serves as a high-affinity binding site for Fe(III).

The Fe(III) binding site is formed by two tyrosines (Tyr95 and Tyr188), one histidine (His249), and one aspartate (Asp63). In addition, a synergistic carbonate anion (CO₃²⁻) completes the octahedral coordination geometry via bidentate binding.⁷ All coordinating ligands are anionic hard Lewis bases, ideally suited to bind the hard Lewis acid Fe(III) with extraordinary affinity (∼10²⁰ M⁻¹) at physiological pH.^8–10

Fe(III) exists in the high-spin sextet state (S = 5/2), with five unpaired electrons occupying the 3d orbitals.¹¹ The role of alternative synergistic anions—such as citrate and malonate—in modulating Fe(III)-sTf binding has also been systematically explored.^12–14

In addition to the first coordination shell, several secondary shell amino acid residues play a critical role in stabilising the active site architecture of sTf. Notably, Arg124 and Thr120 anchor the carbonate anion, while Lys206 and Lys298, located near Tyr188, contribute to maintaining the structural integrity of the binding site. A crystallographic water molecule is observed to form strong hydrogen bonds with His249 and Asp93. Both Arg124 and salt bridge lysine residues are proposed to participate in a protonation-mediated pathway facilitating Fe(III) release, particularly at endosomal pH.

X-ray crystallography reveals three distinct conformational states of sTf.^15,16 The open conformation, seen in apotransferrin (PDB: 2HAV),¹⁵ contrasts with the partially open structure bound to Bi(II) (PDB: 4H0 W)¹⁶ and the closed conformation found in the Fe(III)-bound state (PDB: 1A8E).⁷ Under physiological conditions, only about 30% of the sTf binding sites are saturated with Fe(III), leaving the remaining apotransferrin sites available for other metal ions, especially hard Lewis acids.

sTf is known to bind a variety of metal ions with differing binding affinities, including Al(III), Ti(IV), Cu(II), Gd(III), uranyl, and Pu(IV).^17–23 Many of these ions, particularly lanthanides and actinides, often adopt coordination numbers greater than six. In such cases, solvent molecules supplement the coordination environment to satisfy the metal's preferred geometry.

Using EXAFS data in conjunction with molecular dynamics simulations, additional coordinating water molecules have been observed in complexes with Cm(III), Th(IV), and Pu(IV), yielding hepta- and octa-coordinated geometries proposed by us and others.^24–26

The cellular uptake of Fe(III) is mediated by the sTf cycle, illustrated in Scheme 1.^27,28 The process begins with apotransferrin binding Fe(III) to form holo-transferrin, which then interacts with the transferrin receptor (TfR) on the cell surface. This interaction leads to the formation of a TfR–sTf complex, comprising two sTf molecules bound to the receptor. Subsequently, the complex undergoes endocytosis via clathrin-coated pits, which invaginate and pinch off into the cytoplasm, forming endosomes. Within the endosome, the acidic pH (∼5.6)—maintained by ATP-dependent proton pumps—facilitates the release of Fe(III) from transferrin, thereby regenerating apotransferrin for another cycle.

Scheme 1 sTF Cycle.

The mechanisms underlying Fe(III) release have been extensively studied using computational methods.^29–33 The dominant release pathways are initiated by protonation of coordinating residues—either Tyr188, via the Lys206–Lys298 salt bridge, or the carbonate anion, via Arg124. These protonation events destabilize the iron-binding site, enabling metal release under acidic conditions.

In addition to pH effects, the transferrin receptor itself plays a significant role in modulating the reduction potential of iron, thus further influencing iron release and recycling.^34,35

A critical aspect of the sTf cycle is its lack of specificity for iron, allowing the transport of other potentially harmful metal ions, including radionuclides, into cells, which can negatively impact metabolism.^36–49 To better understand this phenomenon, a significant body of research has explored the conformational changes that occur during the binding and release of various metals, including Fe(III) and other heavy metal ions, to serum transferrin sTf.^29–34

Existing literature by Steinlein et al.,⁵⁰ Bali et al.,⁵¹ and Dhungana et al.³⁵ highlights that the acidic conditions within endosomes are crucial for triggering the release of metal ions from sTf. Their research specifically demonstrates that lower pH facilitates metal dissociation through electrostatic repulsion. Furthermore, the binding of sTf to its receptor also plays a role in modifying the reduction potential, which aids in metal ion release inside the cell. Additionally, Luch and Mason⁵² provided valuable insights into the transferrin-mediated iron delivery system and its implications for understanding the metal release mechanism in sTf.^53,54

Preventing the cellular accumulation of toxic heavy metals and radionuclides necessitates understanding their decorporation. Unlike endocytosis, the process of removing these harmful ions at physiological pH is poorly understood. This study utilises multi-scale modelling simulations and electronic structure calculations to investigate potential metal release pathways at physiological pH. The unbinding of Fe(III) was modelled, and this mechanism may be transferable to other metal ions.

Three distinct Fe-release pathways from sTf at physiological pH were examined, as shown in Fig. 1.

Fig. 1 Possible iron release pathways in serum transferrin. Color captions: Fe(III) (green), Fe(II) (orange), N (blue), O (red) and C (grey), and hydrogen (white).

1. Protonation pathway: protonating the amino acid ligand increases electrostatic repulsion between Fe(III) and the active site, thereby promoting metal release. This process mirrors the endocytosis pathway.

2. Reduction pathway: reducing the Fe(III) ion to Fe(II) weakens metal–ligand coordination, significantly impacting binding. The effectiveness of this pathway depends on the reduction potential of sTf under physiological conditions.

3. Proton-Coupled Electron Transfer (PCET) pathway: in this mechanism, protonation of the amino acid ligand drives the reduction of Fe(III).

Two well-characterised synthetic model (SM) complexes have been studied to mimic the geometric and electronic structures of serum transferrin (sTf). The first is iron(III) tris(catecholate) and the second is iron(III) bis(N-methylimidazole)(2-benzimidazol-2-ylmethyl)phenolato(2-oxo-3-methylbenzoato).^55–57 These complexes effectively replicate the octahedral coordination geometry and high-spin sextet (S = 5/2) state observed in the native sTf-Fe(III) system.

Both models also exhibit reduction potentials comparable to those of Fe(III) in sTf, making them valuable analogues for mechanistic investigations. Notably, studies suggest that proton-coupled electron transfer (PCET) pathways—where protonation of coordinating amino acid ligands facilitates Fe(III) reduction—play a key role in enabling iron release. These synthetic models thus provide important insights into the PCET-mediated iron release mechanism in the biological system.

In this study, we have systematically investigated all three Fe(III) release pathways under physiological pH conditions using a combination of electronic structure calculations and multi-scale modelling simulations. Initially, the reduction potentials of both synthetic mimics and sTf protein models were computed through density functional theory (DFT)-based electronic structure methods.

Subsequently, the protein cluster models were analysed in greater detail using multi-scale modelling to capture the conformational dynamics and assess the feasibility of iron release under various conditions. These investigations employed combined metadynamics and molecular dynamics (MetaD/MD) simulations on the full sTf protein, allowing us to explore the free energy landscape and identify favourable metal release pathways.

Experimental section

Choice of cluster models

We constructed two SM and one protein-based model (PM) from their respective X-ray crystal structures, denoted as SM1, SM2, and PM, as illustrated in Fig. 2.^7,55–57 SM1, iron(III) tris(catecholate), is a tri-anionic complex, where the three counter-cations were omitted in our calculations for simplicity.

Fig. 2 Chosen synthetic models (a) SM1 and (b) SM2 and protein model (c) PM.

In the SM1_CI model, we treated SM1 as a neutral species by explicitly incorporating three piperidinium counter-cations, consistent with those observed in the X-ray crystal structure.⁵⁷

SM2 is iron(III) bis(N-methylimidazole)(2-benzimidazol-2-ylmethyl)phenolato(2-oxo-3-methylbenzoato), a neutral complex that has been proposed as a structural and electronic model for serum transferrin (sTf).⁵⁶

In the PM, the active site, along with key secondary shell interactions, was extracted from the X-ray crystal structure of serum transferrin (PDB: 1A8E).⁷ This model has been employed in our earlier studies to investigate the electronic structure of Fe(III)⁴⁴ and its binding to various actinide ions.^24–26 In our benchmarking efforts, we compared this protein model with simplified cluster models and found that inclusion of secondary coordination shell residues is essential for accurately representing the electronic structure of Fe-bound transferrin.²⁴

The central metal ion in PM is Fe(III), coordinated by the following amino acid residues: Asp63⁻, Tyr95⁻, Tyr188⁻, His249⁻, and a bidentate carbonate ion (CO₃²⁻). These primary ligands confer an overall charge of −3 to the core binding site. The protonation states of these residues were chosen based on our previous studies on Fe(III) and actinide (III/IV) binding to sTf, as well as on literature reports.^22,32 Notably, the anionic form of His249 at physiological pH is supported by DFT calculations from Rinaldo and Field,³² and Mujika et al.²²

We also included key secondary coordination shell residues—Arg124⁺, Thr120, Lys206, Lys296⁺, and a solvent water molecule—to capture the hydrogen-bonding network around the active site. With these additions, the overall net charge of the PM system is −1, and the model consists of 120 atoms.

Computational methods

Geometry optimization was initially performed using both the gas phase and with a continuum solvation model with a dielectric constant of 5, approximating the protein environment. The resulting metal–ligand bond lengths closely matched those from experiment, justifying subsequent gas-phase optimisations. However, solvation effects were included in the final energetic evaluations. This hybrid strategy has been successfully applied by our group to various metal–protein systems, showing good correlation with experimental results.^58–61 Importantly, due to the extensive hydrogen bonding among the residues, no artificial constraints were necessary during geometry optimisations.

We performed geometry optimisations using four different density functionals: BP86-D3BJ, B3LYP-D3BJ, PBE-D3BJ, and M06, each in conjunction with the def2-TZVP basis set (see Table S1 for details). Both gas-phase and solution-phase optimizations were carried out. The latter employed the COSMO continuum solvation model with a dielectric constant of 5, using the BP86/def2-TZVP level of theory to mimic the protein environment.

For the native PM with Fe(III) at the active site, we benchmarked the optimised geometries against the X-ray crystal structure (PDB: 1A8E). Across all four functionals, the Fe–Tyr188 bond length was predicted to lie in the range of 2.05 to 2.16 Å, which is longer than the reported X-ray value. In contrast, the experimental Fe–Tyr188 and Fe–Tyr95 bond lengths are 1.79 Å and 1.97 Å, respectively. The unusually short Fe–Tyr188 bond length observed in the crystal structure may arise from crystallographic inaccuracies, which are not uncommon in protein structures with resolutions around 1.60 Å.

Based on these comparisons, the BP86-D3BJ/def2-TZVP level of theory was deemed most appropriate for accurately capturing the metal–ligand geometry. Furthermore, the same computational setup provided excellent agreement with experimental data for the geometries and reduction potentials of the two synthetic models, lending additional confidence to the reliability of our chosen methodology (vide infra).

Choice of multi-scale models

We selected the N-lobe of sTf from the crystal structure (PDB ID: 1A8E)⁷ as the starting configuration for all calculations. The protonation states of amino acid residues within the iron-binding site (cf.Fig. 2) were assigned based on their predicted pK_a values at serum pH, as determined using established computational protocols.⁶²

To investigate the Fe(III) release mechanisms, we constructed four distinct models of sTf:

1. Fe(III)_sTf: the native Fe(III)-bound form of transferrin, serving as the baseline model.

2. Fe(III)_LH_sTf: a protonated variant, where the carbonate ligand is protonated, based on insights from DFT-based energetics.

3. Fe(II)_sTf: a reduced form of sTf in which the central metal is Fe(II).

4. Fe(II)_LH_sTf: a PCET model, in which the metal is reduced to Fe(II) and either the carbonate or Tyr188 is protonated.

The computational methodologies, including equilibrium and non-equilibrium molecular dynamics (MD) simulations, are provided in the SI.

Results and discussion

Geometries of Fe(III) and Fe(III) sTf models

The optimised structures of Fe(III) in the SM1, SM2, and PM models (Fig. 3) correspond to the experimentally observed high-spin sextet (S = 5/2) state. In the case of SM1, the X-ray crystal structure (CCDC 1110250)⁵⁷ reveals the presence of three piperidinium counter-ions. In the absence of these counter-ions, the optimised Fe–O bond lengths in SM1 are symmetric, averaging 2.087 Å. However, in the SM1_CI model, which includes the three piperidinium cations, the Fe–O bonds become asymmetric, as depicted in Fig. 3a and b.

Fig. 3 Optimized structures of Fe(III) species in synthetic and protein models. (a) SM1, (b) SM1_CI, (c) SM2 and (d) PM. The experimental values are in parentheses.

This asymmetry arises due to specific hydrogen-bonding interactions: one proton is transferred from a piperidinium ion to a catecholate ligand, while the other two counter-ions form strong hydrogen bonds (bond lengths <1.6 Å) with the remaining catecholate ligands. These micro-geometric changes induce a distortion in the Fe–O bond distances, in excellent agreement with the X-ray structure, validating the chemical accuracy of our model.⁵⁷

For SM2, the optimised Fe–X bond distances also show excellent agreement with experimental data (Fig. 3c).⁵⁶ Notably, the computed Fe–O bonds are on average 0.2 Å shorter than the Fe–N bonds, which reflects the difference in local hardness between oxygen and nitrogen donors. The oxygen ligands, being harder Lewis bases, interact more strongly with Fe(III), whereas the imidazole groups form weaker Fe–N interactions.

Taken together, the close match between the computed geometries and experimental structures for both synthetic models (SM1 and SM2) supports the validity of our computational protocol, lending confidence to its application to the PM.

The DFT-optimised structure of the protein model (PM) is shown in Fig. 3d. In addition to the directly coordinating amino acid residues (Asp63⁻, Tyr95⁻, Tyr188⁻, His249⁻, and CO₃²⁻), we incorporated key secondary shell residues, including Thr120, Arg124, Lys296, and Lys206, to capture essential hydrogen-bonding interactions. The calculated Fe(III)–ligand bond lengths are in excellent agreement with the X-ray crystal structure.⁷

Notably, the inorganic carbonate coordinates to Fe(III) in an asymmetric bidentate fashion, with Fe–O distances of 2.170 Å and 2.079 Å. This asymmetry arises from strong hydrogen bonding between Arg124/Thr120 and the carbonate, stabilising the structure and contributing to the bond length disparity.

The effect of one-electron reduction of Fe(III) to high-spin Fe(II) (quintet state) is shown in Fig. 4. In all models, this reduction results in significant elongation of Fe–ligand bond lengths, consistent with a metal-centered reduction process. For example, in SM1, the Fe(II)–ligand bond length increases to 2.208 Å, an elongation of ∼0.12 Å compared to the Fe(III) form.

Fig. 4 Optimized structures of Fe(II) species in synthetic and protein models, (a) SM1, (b) SM1_CI, (c) SM2 and (d) PM.

In the SM1_CI model, the reduction is further coupled to a second proton transfer from the piperidinium counter-ion to a bound catecholate oxygen, as depicted in Fig. 4a and b. This proton-coupled electron transfer (PCET) is essential, as the di-anionic Fe(II) species without protonation is unstable and unable to retain the added electron. These results indicate that proton assistance is crucial for the reduction process in SM1, further supporting the PCET mechanism.

In the reduced SM2 model, we observe that the benzimidazole fragment detaches from the coordination sphere and no longer binds to the Fe(II) center, resulting in a penta-coordinated geometry, as shown in Fig. 4c. Upon reduction, the Fe–N bond lengths contract, whereas the Fe–O bond lengths elongate relative to the Fe(III) state. This structural rearrangement suggests that although SM2 closely mimics the coordination environment of sTf in the Fe(III) state, its reduced Fe(II) form becomes geometrically unstable, facilitating ligand dissociation and potential metal release—a behavior consistent with the proposed functional role of sTf.

For the sTf PM, reduction of Fe(III) to Fe(II) also results in substantial geometric changes, as illustrated in Fig. 4d. Notably, a proton transfer occurs from Arg124 to the bound carbonate, converting it into a bicarbonate species. This proton-coupled event mirrors the behaviour seen in the SM1_CI model, reinforcing the role of PCET in stabilising reduced states. The resulting bicarbonate binds to Fe(II) in an asymmetric bidentate fashion, further altering the coordination geometry. Additionally, the Fe–O and Fe–N bond lengths increase by approximately 0.05 Å due to enhanced electrostatic repulsion in the reduced state.

Reduction potentials of synthetic models and sTf protein

The reduction potentials of both synthetic and protein-based sTf models are reported to be more negative than −1.2 V versus the Ag/AgCl electrode, indicating that direct Fe(III) reduction is thermodynamically unfavourable under physiological conditions. The computed and experimental reduction potentials for all three models are summarised in Table 1. Experimentally, SM1 and SM2 exhibit reduction potentials of −1.60 V and −1.30 V, respectively, against the Ag/AgCl reference.^49,50

Table 1 Computed reduction potentials (in V) of synthetic and protein models

Model	Computed	Experiment
SM1	–2.570	–1.60 (ref. 37)
SM1_CI	–1.623	–1.60 (ref. 37)
SM2	–1.290	–1.30 (ref. 38)
PM	–1.515	∼–0.7 to −0.8 (ref. 63 and 64)
PM_HCO3^−	0.432

---

Our computed reduction potentials for SM1 and SM1_CI are −2.57 V and −1.62 V, respectively. The prediction for SM1_CI agrees closely with experiment, within 20 mV, validating the inclusion of explicit counter-ions. In contrast, the SM1 model without counter-ions underestimates the reduction potential by nearly 900 mV, highlighting the critical role of proton-coupled electron transfer (PCET) in stabilizing the reduced species. This result confirms that proton assistance, via piperidinium counter-ions, is essential for accurately modeling the reduction behavior of SM1.

The computed reduction potential of SM2 (−1.29 V) is in excellent agreement with the experimental value of −1.30 V.⁵⁶ The variation in reduction potentials between SM1 and SM2 can be rationalized by examining the energy and nature of the redox-active molecular orbital (RAMO), as shown in Fig. S1. In both systems, the high-spin Fe(III) center has five half-filled α-spin 3d orbitals, and the incoming electron occupies a β-spin orbital of predominantly t₂g symmetry.

In SM2, the RAMO is primarily the 3d_xy orbital with contributions from 3d_yz and 3d_xz, located at −3.99 eV, making electron acceptance thermodynamically feasible. In contrast, the RAMOs in SM1 and SM1_CI are positioned higher in energy at −2.80 eV and −3.60 eV, respectively, reflecting a more unfavorable reduction process (see Table 1).

For the PM, the computed reduction potential is −1.51 V vs. Ag/AgCl, which is comparable to those of SM1_CI and SM2, but is significantly underestimated relative to the experimental value of <–0.8 V.^63,64 This discrepancy likely arises from the neglect of dynamic protein environment and explicit solvent effects that could stabilize the reduced species in experiment. Nevertheless, the direct reduction pathway is highly unlikely at physiological pH. Similar to the SM1_CI model, the reduction of Fe(III) in PM is accompanied by a proton transfer from Arg124 to the bound carbonate, resulting in the formation of bicarbonate. This proton-coupled electron transfer (PCET) mechanism stabilizes the Fe(II) species and facilitates reduction.

The redox-active molecular orbital (RAMO) for the PM model is located at −3.77 eV, close in energy to those of SM1_CI (−3.60 eV) and SM1 (−2.80 eV). This suggests that the electron-accepting ability of PM is thermodynamically comparable to that of SM1_CI. In contrast, the deeper-lying RAMO of SM2 (−3.99 eV) reflects greater stabilization of the LUMO, consistent with its more favorable computed reduction potential. Thus, the RAMO energetics across the models correlate well with the observed reduction behavior, validating the use of orbital energy analysis in interpreting redox properties.

Iron release pathways of sTf

Thus far, our analysis has demonstrated that the pure electron transfer process is thermodynamically unfavorable for both the synthetic and protein models. Moreover, due to the exceptionally high association constant of Fe(III) with serum transferrin (sTf), iron removal under physiological pH (∼7.4) conditions remains highly challenging.¹⁰ The origin of this high binding affinity can be understood in terms of hard–soft acid–base (HSAB) theory.⁶⁵ Both Fe(III), a hard Lewis acid, and the coordinating amino acid residues in the sTf binding site (such as tyrosine, histidine, and aspartate), which are hard Lewis bases, form strong interactions that stabilize the metal complex.

As a result, protonation alone is unlikely to occur at physiological pH, thereby limiting any pH-triggered iron release. However, under acidic endosomal conditions (pH ≈ 5.5), protonation becomes more thermodynamically favorable and thus emerges as a plausible initiating step in the Fe(III) release mechanism. Previous multi-scale modeling studies from our group have shown that even under these acidic conditions, electrostatic interactions between Fe(III) and the coordinating amino acids remain strong, further complicating metal dissociation.^24–26

To better understand the protonation propensity, we computed the proton affinities of the four ligands directly coordinated to Fe(III), as listed in Table 2. Among them, carbonate exhibits the highest proton affinity (−305.9 kcal mol⁻¹), followed by anionic histidine (His255) at −300.0 kcal mol⁻¹, tyrosine at −295.0 kcal mol⁻¹, and aspartate at −284.7 kcal mol⁻¹. These values clearly suggest that protonation is a thermodynamically favorable process, particularly for carbonate and His255, which likely act as initial protonation sites in the endosomal environment. However, even with protonation, Fe(III) release remains a thermodynamically uphill process, in line with our previous findings.²⁵

Table 2 Computed absolute hardness (η, in eV) of Fe(III) and Fe(II) species and proton affinities (PA, in kcal mol⁻¹) of coordinating amino acid ligands in sTf

	Asp^−	CO3^2−	His^−	Tyr^−	Fe^3+	Fe^2+
PA	−284.74	−305.89	−300.03	−295.02	N/A
η	N/A				24.08 (8.70)	11.08 (4.27)

---

Additionally, the computed local hardness values for the naked Fe(III) and Fe(II) are 24.08 and 11.08, respectively, further confirming that Fe(II) is significantly softer than Fe(III). The values in parentheses are for the corresponding hydrated ions that follow a similar trend to those of the bare ions. This suggests that reduction to Fe(II) partially alleviates the tight binding, but only when coupled with protonation can it facilitate iron release from sTf.

Thus, reduction can create a mismatch between the harder ligands and the softer Fe(II) species. Reduction is only possible through a proton-coupled electron transfer (PCET). The optimized structure of the Fe(II) species with the bicarbonate anion is shown in Fig. 5. Carbonate is protonated based on the computed proton affinity, and in Fig. 4d, we note proton transfer from Arg124 to carbonate in the optimized structure of the Fe(II) species. Thus, we have optimized the bicarbonate while retaining the cationic Arg124.

Fig. 5 Optimized structure of Fe(II) PM with bicarbonate.

We find that bicarbonate is coordinating to the Fe(II) ion in a mono-dentate fashion with Fe–O bond lengths at 2.47 Å and 2.62 Å, respectively. The bicarbonate oxygen makes strong hydrogen bonds with Arg124 and Thr120, as shown in Fig. 5.

The computed reduction potential of this PCET species is 0.43 V against Ag/AgCl, which is favorable under the physiological conditions compared to −1.51 V without the proton assistance. Of all the competing pathways, the PCET is the most plausible and feasible metal release pathway at neutral pH. We will highlight below that our cluster-based DFT calculations are in line and indeed reflected in multi-scale modeling techniques of the sTf protein.

Multi-scale modelling of sTf

Equilibrium molecular dynamics (MD) simulations of 100 nanoseconds were carried out for all four models. The X-ray crystal structure served as the initial conformation for the Fe(III)-sTf system, while the remaining three systems used pre-equilibrated structures. To evaluate structural stability and conformational changes, the root-mean-square deviation (RMSD) of the protein backbone relative to the starting structure, along with the radius of gyration (R_g), was analyzed from the MD trajectories (Fig. S2).

The RMSD of Cα atoms remained within 1.7 Å across all models, with stabilization occurring within 2 ns, except for the Fe(III)_LH_sTf system, which showed slight fluctuations around 60 ns and 80 ns before stabilizing. R_g analysis revealed that in every case, the metal-bound sTf protein retained its closed conformation, with an average R_g of 19.7 ± 0.06 Å. These results align well with previously reported values for iron-bound transferrin.^24–26 Overall, the consistent trends observed in both RMSD and R_g confirm the structural convergence of the metal-bound sTf protein and support the validity of the simulation parameters and methodology used.

The presence of a carbonate ion and the closed conformation of the NI–NII subdomains are essential for effective metal ion binding to the apo-transferrin protein. To validate this, we calculated the center of mass (COM) distance between the NI and NII subdomains of sTf, as well as the COM distance between the metal ions and the coordinating residues in the first coordination shell at the binding site (Fig. S3). Across all four models, the average NI–NII distance was consistently within 26.1 ± 0.22 Å. Notably, the NI–NII separation in the Fe(III)-sTf model closely matches the experimentally determined value of 25.9 Å from the crystal structure of the iron-loaded closed form.⁷ Furthermore, the COM distance between the metal ion and the first-shell residues at the binding site remained within 2 Å in all models, confirming stable coordination.

Hydrogen bonding and synergistic anions

The presence of a carbonate anion is crucial for metal ion binding within the cleft of the protein. This carbonate ion is positioned near the Arg124 residue, located in the second coordination shell of the binding site. Proton transfer from Arg124 to the carbonate may facilitate cleft opening and promote metal release. To explore this interaction, we analyzed the MD trajectory to assess the proximity of Arg124 to the carbonate or bicarbonate ion. The center of mass (COM) separation between these components, shown in Fig. S4, indicates that Arg124 remains close to the carbonate throughout the simulations across all four models. We further examined the hydrogen bonding interactions between Arg124 and the carbonate ion. The Fe(II)_sTf model displayed a greater number of hydrogen bonds compared to Fe(III)_sTf. Notably, a strong hydrogen bond is observed between the side chain of Arg124 and the carbonate, supporting the possibility of proton transfer from arginine to the carbonate ion.

To quantify these interactions, we calculated hydrogen bond lifetime correlation functions, CHB(t), for all models (Fig. S5). The Fe(II)_LH_sTf system showed the largest area under the CHB(t) curve (3223), with the hydrogen bond persisting for up to ∼24 ns of the simulation. In contrast, the other three models exhibited significantly lower hydrogen bond activity, with areas under the curve measured at 34 (Fe(III)_sTf), 276 (Fe(II)_sTf), and 1092 (Fe(III)_LH_sTf).

Release of metal ions from metadynamics (MtD) simulations

The binding strength between a metal ion and the protein matrix is influenced by multiple factors, including structural fluctuations, the ionic potential (z/r) of the metal, electrostatic forces, solvation effects, van der Waals interactions, hydrogen bonding, and water bridge (WB) interactions involving protein–water–protein residues. To assess the relative binding strength of the metal–protein complexes and determine the order of metal ion release from the binding site, we performed well-tempered metadynamics simulations.

Two collective variables (CVs) were employed: (i) the center of mass (COM) separation between the NI and NII subdomains, and (ii) the COM distance between the metal ion and the protein. A visual representation of these CVs is provided in Fig. S6. Metal ion release is promoted by protein conformational changes, particularly the opening motion reflected in increasing NI–NII separation.^24–26

We computed the potential of mean force (PMF) profiles for metal ion release along these CVs to capture the energy landscape of the unbinding process. These PMF profiles, shown in Fig. 6 for all four models, reveal a complex pathway characterized by multiple local minima (binding basins) and maxima (energy barriers), underscoring the intricacy of metal–protein interactions.

Fig. 6 Free energy contour surface evaluated from well-tempered metadynamics using collective variables as (i) protein–ion center of mass separation (x-axis) and center of mass separation between the NI and NII subdomains of sTf (y-axis) for the unbinding transition of metal ions from the binding site of the sTf protein. PMF profiles for all four models viz. Fe(III)_sTf, Fe(II)_sTf, Fe(II)_LH_sTf and Fe(III)_LH_sTf are shown. Iso-energy lines are drawn every 26.38 kJ mol⁻¹ as indicated by the color bar. Blue-colored regions show the bound state of the ions.

Among the studied systems, Fe(II)_LH_sTf and Fe(III)_LH_sTf—where bicarbonate serves as the synergistic anion—exhibited the lowest energy requirements and fastest unbinding kinetics. In contrast, Fe(II)_sTf and Fe(III)_sTf, which utilize carbonate as the di-anionic synergistic ligand, showed higher energy barriers for metal release.

Fig. 7 presents the one-dimensional potential of mean force (PMF) profiles for metal ion unbinding from the sTf binding cleft, plotted as a function of the center of mass (COM) distance between the metal ions and the coordinating protein residues. These profiles allow for a comparative assessment of the relative binding strength of the metal ions.

Fig. 7 PMF profiles of metal ion unbinding from the sTf binding cleft as a function of center of mass separation between the metal ion and protein residues. PMF profiles are shown for Fe(III)_sTf (black), Fe(II)_sTf (red), Fe(II)_LH_sTf(green) and Fe(III)_LH_sTf (blue) protein models respectively.

Correspondingly, Table 3 summarizes the relative dissociation energies of the metal–protein complexes. Across all models, the stability order of the metal–sTf complexes is observed to be: Fe(III)_sTf > Fe(II)_sTf > Fe(III)_LH_sTf > Fe(II)_LH_sTf.

Table 3 Computed approximate iron dissociation energies (units in kJ mol⁻¹) and approximate time for iron release (ns)

Model	Dissociation energy	Approximate time
Fe(III)_sTf	260	44
Fe(II)_sTf	230	40
Fe(III)_LH_sTf	124	13
Fe(III)_TyrH_sTf	163	>50
Fe(II)_TyrH_sTf	142	20
Fe(II)_LH_sTf	77	7

---

This indicates that the Fe(II)_LH_sTf model exhibits the lowest energy barrier for metal ion release, suggesting it is the most weakly bound among the four systems. The combination of proton transfer from Arg124 to carbonate and the reduction of Fe(III) to Fe(II) plays a crucial role in facilitating iron release. However, the data further reveal that reduction alone is insufficient to significantly enhance unbinding. Instead, favorable protonation of the carbonate ion by Arg124 appears to be a key step that promotes Fe(III) reduction, thereby enabling efficient iron release from the binding cleft.

Previous studies have suggested that Tyr188 may play a significant role in the endosomal pH-dependent iron release pathway. To further investigate this, we conducted additional simulations to examine the effects of protonating Tyr188 and explored the Fe-release mechanism using multi-scale modeling approaches. Our results show that the dissociation energies for iron release without protonation at the carbonate site are 163 kJ mol⁻¹ for Fe(III) and 142 kJ mol⁻¹ for Fe(II), with corresponding unbinding times exceeding 50 ns and 20 ns, respectively, as listed in Table 3. However, upon protonation at the carbonate site, the dissociation energies are significantly reduced—to 124 kJ mol⁻¹ for Fe(III) and 77 kJ mol⁻¹ for Fe(II)—with much shorter unbinding times of approximately 13 ns and 7 ns.

These findings are in line with proton affinity values derived from DFT calculations for both the carbonate ion and the Tyr188 residue, further supporting the importance of protonation in facilitating iron release under endosomal conditions.

Conclusions

sTf is a highly specialized protein with an exceptionally strong binding affinity for Fe(III), making it a tightly regulated carrier of iron in biological systems. However, approximately 70% of sTf exists in the apo (metal-free) form, creating opportunities for other metal ions—including toxic metals and radionuclides—to bind. This has significant biological and environmental implications, highlighting the urgent need for effective decorporation agents that can remove harmful metal ions before cellular uptake occurs.

In this study, we employed a combination of electronic structure calculations and multi-scale modeling to investigate the structural features and redox behavior of several synthetic models (SMs) of sTf, along with a detailed protein model. Our computational findings provide several important insights:

(i) Accurate redox predictions: the computed reduction potentials of the synthetic models align well with experimental values, with deviations within 100 mV. The largely negative potentials suggest that the reduction process is inherently difficult.

(ii) Role of proton transfer: the reduction of SM1 and the protein model (PM) is driven by proton transfer from counter-ions and the Arg124 residue, indicating that proton-coupled electron transfer (PCET) facilitates the reduction.

(iii) Impact of carbonate protonation: protonation of the carbonate ligand significantly increases the reduction potential, thereby promoting iron release from the strongly bound Fe(III)-sTf complex.

(iv) Mechanistic insight from simulations: molecular dynamics and metadynamics simulations support that the most favorable pathway for iron release involves both the reduction of Fe(III) to Fe(II) and the protonation of carbonate to bicarbonate, particularly in the Fe(II)_LH_sTf model.

These findings lay the groundwork for future studies targeting the decorporation of toxic metal ions through PCET mechanisms. Our laboratory is actively pursuing this direction using advanced computational approaches.

Author contributions

M. S. designed the project, was involved in DFT calculations and wrote the paper. L. M. was involved in multi-scale model calculations and wrote the paper. N. K. B. and S. P. M. were involved in the organization of data and assisted in the manuscript preparation.

Conflicts of interest

There are no conflicts to declare.

Data availability

The data supporting this article have been included as part of the SI.

Supplementary information: additional details on electronic structure calculations, well-tempered metadynamics, MD simulations and optimized coordinates of all species in Fe(III) and Fe(II). See DOI: https://doi.org/10.1039/d5dt01803j.

Acknowledgements

We thank BARC super-computing facilities.

References

1. Y. Kohgo, K. Ikuta, T. Ohtake, Y. Torimoto and J. Kato, Int. J. Hematol., 2008, 88, 15.
2. K. Vijayalakshmy, M. Virmani, M. J. U. Lambe, R. Verma and V. Choudhiry, J. Entomol. Zool. Stud., 2019, 7, 442–444.
3. Z. Qian, H. Li, H. Sun and K. Ho, Pharmacol. Rev., 2002, 54, 561–587.
4. H. Z. Sun, H. Y. Li and P. J. Sadler, Chem. Rev., 1999, 99, 2817–2842.
5. S. Kovac, A. Ferrand, J. P. Esteve, A. B. Mason and G. S. Baldwin, FEBS J., 2009, 276, 4866–4874.
6. D. Xiu-Lian, W. Kui, K. Ya, Y. Lan, L. Rong-chang, C. Y. Zhong, H. K. Ping and Q. Z. Ming, J. Cell. Physiol., 2004, 201, 45–54.
7. R. T. A. MacGillivray, S. A. Moore, J. Chen, B. F. Anderson, H. Baker, Y. Luo, M. Bewley, C. A. Smith, M. E. P. Murphy, Y. Wang, A. B. Mason, R. C. Woodworth, G. D. Brayer and E. N. Baker, Biochemistry, 1998, 37, 7919–7928.
8. E. N. Baker, Adv. Inorg. Chem., 1994, 41, 389–463.
9. R. A. Pinkowitz and P. Aisen, J. Biol. Chem., 1972, 247, 7830–7834.
10. P. Aisen, A. Leibman and J. Zweier, J. Biol. Chem., 1978, 253, 1930–1937.
11. M. Azarkh, P. Gast, A. B. Mason, E. J. J. Groenen and G. Mathies, Phys. Chem. Chem. Phys., 2019, 21, 16937–16948.
12. Z. Ghanbari, M. R. Housain and M. R. Bozorgmehr, J. Mol. Graphics Modell., 2017, 78, 176–186.
13. A. D. Tinoco, M. Saxena, S. Sharma, N. Noinaj, Y. Delgado, E. P. Quiñones, S. E. Conklin, N. Zambrana, S. A. Loza-Rosas and T. B. Parks, J. Am. Chem. Soc., 2016, 138, 5659–5665.
14. W. R. Harris, Biochim. Biophys. Acta, Gen. Subj., 2012, 1820, 348–361.
15. J. Wally, P. J. Halbrooks, C. Vonrhein, M. A. Rould, S. J. Everse, A. B. Mason and S. K. Buchanan, J. Biol. Chem., 2006, 281, 24934–24944.
16. N. Yang, H. Zhang, M. Wang, Q. Hao and H. Sun, Sci. Rep., 2012, 2, 999.
17. Y. Nuevo-Ordoñez, M. Montes-Bayón, E. Blanco González and A. Sanz-Medel, Metallomics, 2011, 3, 1297–1297.
18. A. D. Tinoco and A. M. Valentine, J. Am. Chem. Soc., 2005, 127, 11218–11219.
19. P. B. O'Hara and S. H. Koenig, Biochemistry, 1986, 25, 1445–1450.
20. W. R. Harris and Y. Chen, Inorg. Chem., 1992, 31, 5001–5006.
21. C. Zurita, S. Tsushima, P. L. Solari, D. Menut, S. Dourdain, A. Jeanson, G. Creff and C. Den Auwer, Chem. – Eur. J., 2023, 29, e202300636.
22. J. I. Mujika, B. Escribano, E. Akhmatskaya, J. M. Ugalde and X. Lopez, Biochemistry, 2012, 51, 7017–7027.
23. M. G. Benavides-Garcia and K. Balasubramanian, Chem. Res. Toxicol., 2009, 22, 1613–1621.
24. L. Mishra, P. D. Sawant, M. Sundararajan and T. Bandyopadhyay, J. Phys. Chem. B, 2019, 123, 2729–2744.
25. L. Mishra, M. Sundararajan and T. Bandyopadhyay, Proteins:Struct., Funct., Bioinf., 2020, 89, 193–206.
26. L. Mishra, M. Sundararajan and T. Bandyopadhyay, JBIC, J. Biol. Inorg. Chem., 2020, 25, 213–231.
27. A. M. N. Silva, T. Moniz, B. de Castro and M. Rangel, Coord. Chem. Rev., 2021, 449, 214186–214186.
28. Y. Cheng, O. Zak, P. Aisen, S. C. Harrison and T. Walz, Cell, 2004, 116, 565–576.
29. H. Abdizadeh, A. R. Atilgan and C. Atilgan, JBIC, J. Biol. Inorg. Chem., 2015, 20, 705–718.
30. N. Kojima and G. W. Bates, J. Biol. Chem., 1981, 256, 12034–12039.
31. E. C. Theil and P. Aisen, in G. Winkelmann, D. Van and J. B. Neilands, Iron Transport in Microbes, Plants, and Animals, Wiley-VCH, 1987, p. 491.
32. D. Rinaldo and M. J. Field, Aust. J. Chem., 2004, 57, 1219.
33. B. K. Fındık, U. Cilesiz, S. K. Bali, C. Atilgan, V. Aviyente and B. Dedeoglu, Org. Biomol. Chem., 2022, 20, 8766–8774.
34. D. C. Kraiter, O. Zak, P. Aisen and A. L. Crumbliss, Inorg. Chem., 1998, 37, 964–968.
35. S. Dhungana, C. H. Taboy, O. Zak, M. Larvie, A. L. Crumbliss and P. Aisen, Biochemistry, 2003, 43, 205–209.
36. G. J.-P. Deblonde, M. Sturzbecher-Hoehne, A. B. Mason and R. J. Abergel, Metallomics, 2013, 5, 619–626.
37. A. Jeanson, M. Ferrand, H. Funke, C. Hennig, P. Moisy, P. Solari, C. Vidaud and C. Den Auwer, Chem. – Eur. J., 2009, 16, 1378–1387.
38. M. Sturzbecher-Hoehne, C. Goujon, G. J.-P. Deblonde, A. B. Mason and R. J. Abergel, J. Am. Chem. Soc., 2013, 135, 2676–2683.
39. Y. Shi, I. Harvey, D. Campopiano and P. J. Sadler, Bioinorg. Chem. Appl., 2010, 307578.
40. H. Sun, H. Li and P. J. Sadler, Chem. Rev., 1999, 99, 2817–2842.
41. E. Peter and M. Lehmann, Interaction of Thorium with Blood Serum Proteins in Vivo, Int. J. Radiat. Biol. Relat. Stud. Phys. Chem. Med., 1981, 40, 445–450.
42. N. Bauer, D. R. Fröhlich and P. J. Panak, Dalton Trans., 2014, 43, 6689–6700.
43. N. Bauer and P. J. Panak, New J. Chem., 2014, 39, 1375–1381.
44. L. Mishra, I. S. Singh, H. K. Patni and D. D. Rao, Radiat. Prot. Dosim., 2018, 181, 168–177.
45. N. Bauer, V. C. Smith, A. MacGillivray and P. J. Panak, Dalton Trans., 2014, 44, 1850–1857.
46. I. S. Singh, L. Mishra, J. R. Yadav, M. Y. Nadar, D. D. Rao and K. S. Pradeepkumar, Appl. Radiat. Isot., 2015, 104, 49–54.
47. M. P. Jensen, D. Gorman-Lewis, B. Aryal, T. Paunesku, S. Vogt, P. G. Rickert, S. Seifert, B. Lai, G. E. Woloschak and L. Soderholm, Nat. Chem. Biol., 2011, 7, 560–565.
48. F. Brulfert and J. Aupiais, Dalton Trans., 2018, 47, 9994–10001.
49. D. M. Taylor, J. Alloys Compd., 1998, 271–273, 6–10.
50. L. M. Steinlein, C. M. Ligman, S. Kessler and R. A. Ikeda, Biochemistry, 1998, 37, 13696–13703.
51. P. K. Bali, O. Zak and P. Aisen, Biochemistry, 1991, 30, 324–328.
52. A. N. Luck and A. B. Mason, Curr. Top. Membr., 2012, 3–35.
53. J.-M. E. Chahine and D. Fain, Eur. J. Biochem., 1994, 223, 581–587.
54. N. C. Giri, Role of Transferrin in Iron Metabolism, in Iron Metabolism - A Double-Edged Sword, IntechOpen, 2022,  DOI:10.5772/intechopen.100488.
55. T. B. Karpishin, M. S. Gebhard, E. I. Solomon and K. N. Raymond, J. Am. Chem. Soc., 1991, 113, 2977–2984.
56. M. R. McDevitt, A. W. Addison, E. Sinn and L. K. Thompson, Inorg. Chem., 1990, 29, 3425–3433.
57. B. F. Anderson, D. A. Buckingham, G. B. Robertson and J. Webb, Acta Crystallogr., Sect. B, 1982, 38, 1927–1931.
58. M. Sundararajan, ACS Omega, 2023, 8, 18041–18046.
59. G. Rajaraman, R. Kumar and M. Sundararajan, Chem. Commun., 2021, 57, 13750.
60. M. Sundararajan, B. Park and M.-H. Baik, Inorg. Chem., 2019, 58, 16250–16255.
61. V. Mishra, M. Sundararajan, A. Pathak, P. Sawant and T. Bandyopadhyay, Phys. Chem. Chem. Phys., 2025, 27, 3485–3495.
62. R. Anandakrishnan, B. Aguilar and A. V. Onufriev, Nucleic Acids Res., 2012, 40, W537–W541.
63. S. A. Kretchmar, Z. E. Reyes and K. N. Raymond, Biochim. Biophys. Acta, Protein Struct. Mol. Enzymol., 1988, 956, 85–94.
64. J. Parker, E. M. Lin, S. J. Brandt, A. D. Tinoco, A. M. Valentine and A. L. Crumbliss, J. Inorg. Biochem., 2010, 104, 1006–1009.
65. K. R. S. Chandrakumar and S. Pal, J. Phys. Chem. A, 2002, 106, 11775–11781.

---