Computational Design of Metalloproteins and Catalytic Centers

Introduction

Metalloproteins constitute a significant fraction of all proteins across the domains of life and perform essential functions including electron transfer, oxygen transport, hydrolysis, and redox catalysis [1]. The incorporation of metal ions into protein scaffolds imparts unique chemical reactivity that organic cofactors alone cannot achieve [2]. Computational design of metalloproteins aims to engineer novel or enhanced catalytic function by rationally placing metal ions within designed or repurposed protein scaffolds [3, 4]. This endeavor requires a deep understanding of metal coordination geometry, the identity and spatial arrangement of coordinating amino acids, and the quantum mechanical nature of metal–ligand interactions [2, 5]. Recent advances in machine learning, molecular dynamics, and quantum mechanics/molecular mechanics (QM/MM) methods have accelerated the creation of artificial metalloenzymes with tailored activities [6, 7, 8]. This review provides a technical overview of the principles and computational workflows used to design metal-binding sites and catalytic centers in proteins, emphasizing structural validation and parameterization strategies.

Metal Coordination Geometry and Coordinating Amino Acids

Metal ions in proteins adopt characteristic coordination geometries determined by the metal's electronic configuration, oxidation state, and the ligand field exerted by side chains, backbone carbonyls, and water molecules [1]. Common geometries include tetrahedral (Zn²⁺, Fe²⁺ in rubredoxins), square planar (Cu²⁺, Ni²⁺), octahedral (Mg²⁺, Fe²⁺/Fe³⁺ in heme), and trigonal bipyramidal (some Cu²⁺ sites) [2]. The most frequent metal‑coordinating residues are histidine (imidazole nitrogens), cysteine (thiolate sulfur), aspartate and glutamate (carboxylate oxygens), methionine (thioether sulfur), and tyrosine (phenolate oxygen) [5]. The hard‑soft acid‑base (HSAB) principle governs the preference: hard acids (e.g., Mg²⁺, Ca²⁺) bind hard ligands (oxygen from Asp, Glu, water), while soft acids (Cu⁺, Zn²⁺) favor soft ligands (Cys, His) [1].

The spatial arrangement of coordinating ligands defines the metal‑binding pocket and determines spectral and redox properties [6]. For example, the histidine brace motif, consisting of two histidines and an N‑terminal amine, is employed by copper‑dependent lytic polysaccharide monooxygenases and has been successfully engineered into a de novo designed scaffold [4]. Design efforts must consider not only the primary coordination sphere but also the secondary sphere that provides hydrogen bonds, electrostatic stabilization, and steric constraints [3, 9]. The orientation of imidazole rings relative to the metal center influences the redox potential and ligand affinity [2]. A summary of common metal coordination geometries and their preferred amino acid ligands is provided in Table 1.

Table 1. Common metal coordination geometries and typical coordinating residues.

Accurate assignment of metal ions in crystal structures is critical for computational design. Elemental spectroscopy methods such as X‑ray fluorescence can validate metal identity and occupancy, correcting misassignments in the Protein Data Bank [10]. The adoption of a multimodal framework that integrates protein surface properties and sequence information has improved residue‑level metal‑binding site recognition [11].

Computational Approaches for Metal‑Binding Site Design

Machine Learning for Site Prediction and Redox Tuning

Machine learning has emerged as a powerful tool for predicting metal‑binding residues from sequence and structure [5]. A protein surface‑aware multimodal framework that combines graph convolutional networks and sequence embeddings achieved high accuracy for residue‑level classification of binding sites across nine metal types [11]. Machine learning models have also been applied to predict redox potentials of iron‑sulfur clusters, a key parameter for designing electron‑transfer pathways [6]. These models use features such as solvent accessibility, hydrogen‑bonding patterns, and electrostatic potential to estimate reduction potentials with errors below experimental uncertainty [6].

For the design of entirely new metalloenzymes, deep learning architectures like RFdiffusion have been employed to generate protein backbones that host metal‑binding motifs [12]. Evolutionary analysis combined with diffusion‑based methods enabled the design of aggregation‑resistant frataxin variants that maintain iron‑sulfur cluster assembly [12]. These generative approaches expand the conformational space beyond naturally occurring folds.

Molecular Mechanics Parameterization of Metal–Ligand Interactions

A major challenge in computational metalloprotein design is the accurate parameterization of metal–ligand bonds within classical molecular mechanics (MM) force fields [2]. Metal coordination is inherently quantum mechanical, involving partial covalency, charge transfer, and polarization [13]. Two principal strategies exist for handling metals in MM: the nonbonded approach, in which the metal is treated as a point charge with van der Waals parameters, and the bonded approach, where explicit harmonic bonds, angles, and dihedrals are defined between the metal and donor atoms [2].

The bonded approach requires force constants and equilibrium values derived from quantum chemical calculations or high‑resolution crystal structures [13]. For example, QM/MM cluster models of [NiFe]‑hydrogenase have yielded detailed pictures of proton transfer pathways and provided reference geometries for parameterizing the dinuclear active site [13]. Molecular dynamics simulations with correctly parameterized metal sites can capture correlated motions essential for catalysis, as demonstrated by paramagnetic NMR studies of a model hydrogenase [14]. The use of QM/MM to address long‑standing catalytic questions directly informs the design of artificial variants [13].

Table 2 lists typical parameters needed for bonded‑ model metal sites in common force fields (e.g., CHARMM, AMBER).

Table 2. Representative metal‑ligand bond parameters for MM force fields.

QM/MM and Multiscale Modeling

For catalytic reactions involving bond breaking and formation, QM/MM methods are indispensable [13, 2]. The catalytic center is treated quantum mechanically (often using density functional theory, DFT), while the remainder of the protein is described with a classical force field. This approach has been used to study the mechanism of artificial copper metalloenzymes that perform C‑H bond activation using H₂O₂ and O₂ [8], as well as the proton transfer steps in [NiFe]‑hydrogenase [13]. Large QM‑cluster models that include second‑sphere residues and solvent molecules improve accuracy of energy profiles [13, 2].

Hybrid quantum/classical docking protocols have also been developed to predict binding modes of covalent and non‑covalent ligands at metalloenzyme active sites [15]. These methods use QM‑derived partial charges and allow for metal‑ligand bond formation during docking, which is essential for designing transition‑state analogs or mechanism‑based inhibitors [15].

Catalytic Active Site Design Strategies

De Novo Design of Metal‑Binding Scaffolds

De novo design creates protein backbones from scratch to bind metals with predefined coordination geometry [4, 7]. The design of four‑helix bundles to bind metalloporphyrin cofactors represents an early success [7]. More recently, a histidine‑brace copper‑binding site was engineered into a three‑helix bundle, demonstrating that the characteristic bidentate coordination of copper can be recapitulated in a minimal scaffold [4]. The resulting artificial enzyme catalyses oxidative reactions using O₂ as oxidant [4, 8].

Another de novo approach uses random heteropolymers as enzyme mimics. These polymers, composed of a subset of amino acids, can catalyse ester hydrolysis and exhibit Michaelis‑Menten kinetics without a defined tertiary fold [16]. While not strictly proteins, these systems inform on the minimal requirements for catalytic function and may be integrated into designed metalloprotein frameworks.

Redesign of Existing Scaffolds

Rational redesign of natural metalloproteins is an efficient route to new catalysis. Myoglobin, an oxygen‑binding heme protein, has been extensively engineered to perform peroxygenase, oxidase, and carbene transfer reactions by mutating residues in the distal heme pocket [3]. The strategy of “functional site scaffolding” involves grafting a preorganized catalytic motif onto a stable protein framework [3]. Similarly, artificial heme‑copper enzymes have been designed by introducing a copper‑binding site adjacent to the heme iron in myoglobin, mimicking the active site of cytochrome c oxidase [9]. These designs require careful optimization of the metal‑metal distance and electron transfer pathways [9].

Stability‑activity trade‑offs are a common obstacle; computational approaches that simultaneously optimize conformational stability and catalytic turnover have been developed, for example in lipoxygenases [17]. Structure‑guided design of ferritin nanocages has also enabled the creation of thermoresponsive channels for controlled drug release, relevant for veterinary therapeutic applications [18, 19].

Verification of designed metalloenzymes involves comparison of calculated and experimentally determined metal‑ligand distances and bond angles from X‑ray crystallography or cryo‑electron microscopy [10]. Three‑dimensional structural inspection should confirm that all coordinated residues are within expected distances (typically 1.9–2.3 Å for first‑row transition metals) and that the geometry around the metal matches the intended model [2]. Deviations beyond 0.2 Å may require reparametrization or redesign of the pocket [10].

Workflow for Computational Metalloprotein Design

The computational design of a metalloenzyme can be summarized as a sequence of steps involving target selection, scaffold identification or generation, metal‑site placement, optimization, and validation. Figure 1 presents a Mermaid diagram of a typical workflow.

flowchart TB
    A[Target reaction selection], > B[Scaffold identification or de novo generation]
    B, > C[Placement of metal ion and coordinating residues]
    C, > D[Energy minimization and side‑chain packing]
    D, > E[QM/MM or DFT geometry optimization]
    E, > F[Validation of coordination distances and angles]
    F, > G{Geometry acceptable?}
    G, Yes, > H[Molecular dynamics simulation]
    G, No, > C
    H, > I[Binding free energy or catalytic barrier calculation]
    I, > J[Redox potential prediction via ML]
    J, > K{Meet target criteria?}
    K, Yes, > L[Experimental expression and characterization]
    K, No, > B
    L, > M[Structure determination and spectroscopy]
    M, > N[Compare with computational model]
    N, > O[Refine parameters or redesign if needed]

Iteration between computational and experimental steps is essential [2, 5]. Machine learning models can accelerate the initial screening of metal‑site compatibility [11, 6], while multiscale QM/MM simulations provide accurate energetics [13, 8].

Case Examples in Veterinary and Catalytic Contexts

Iron‑Sulfur Cluster Redesign

Iron‑sulfur clusters are ubiquitous electron‑transfer cofactors in bacterial, plant, and animal metabolism [1]. Machine learning models trained on a dataset of experimentally determined reduction potentials have achieved high predictive accuracy for [2Fe‑2S] and [4Fe‑4S] clusters [6]. This capability can guide the design of redox‑active variants of ferredoxins for applications in biosensors or anaerobic bioremediation [6]. In veterinary contexts, iron‑sulfur cluster biogenesis is critical for pathogens such as Mycobacterium avium subsp. paratuberculosis (Johne's disease) [6, 1].

Artificial Copper Enzymes for Oxidative Catalysis

De novo designed copper proteins based on the histidine brace motif catalyze the oxidation of organic substrates with H₂O₂ or O₂ [4, 8]. Mechanistic studies using DFT and QM/MM revealed a two‑electron oxidation pathway involving a Cu(II)‑oxyl intermediate [8]. Such artificial enzymes hold promise for inactivating bacterial toxins or degrading pesticides in agricultural settings.

Hydrogenase Mimics for Hydrogen Production

The [NiFe]‑hydrogenase active site has been a focus of computational redesign. Paramagnetic NMR studies combined with MD simulations have correlated protein dynamics with catalytic activity, revealing that substrate access is gated by a conserved histidine residue [14]. Large QM‑cluster models provided a detailed picture of proton transfer from the Ni‑Fe site to the protein surface [13]. These insights have guided the design of simplified hydrogenase mimics with improved oxygen tolerance [13].

Future Directions and Challenges

Despite progress, several challenges remain. The stability‑activity trade‑off often forces designers to accept reduced thermostability for enhanced catalytic rate; computational methods that explicitly balance these properties are under development [17]. Accurate prediction of metal redox potentials in protein environments remains difficult due to long‑range electrostatic effects and solvent reorganization [6]. Integration of evolutionary information with generative deep learning (e.g., RFdiffusion) offers new routes to metal‑binding scaffolds that are both stable and functional [12].

The use of metal‑specific spectroscopic validation (e.g., EXAFS, X‑ray fluorescence) to confirm computational predictions will become increasingly important [10]. Finally, the application of these methods to veterinary pathogens, such as designing inhibitors of metalloenzymes in Clostridium perfringens or Escherichia coli, could yield novel therapeutic strategies [20, 21].

Conclusion

Computational design of metalloproteins and catalytic centers is a multidisciplinary field that combines coordination chemistry, structural bioinformatics, and multiscale simulation. Machine learning has accelerated the prediction of metal‑binding residues and redox properties [11, 5], while QM/MM methods provide the accuracy needed to study catalytic mechanisms [13, 8]. Parameterizing metal–ligand interactions in molecular mechanics remains a technical hurdle, but bonded models derived from QM calculations offer reliable performance. Verification of designed sites through 3D structural inspection and spectroscopic validation is essential to close the design cycle. As computational power and algorithmic sophistication continue to advance, the ability to create custom metalloenzymes for veterinary and biotechnological applications will expand.

References

[1] Kroneck PMH. Exploring the fascination of metal-sulfur bonds, vivid colors, and electron transfer through proteins: A tribute to Harry B. Gray. J Inorg Biochem. 2025. https://pubmed.ncbi.nlm.nih.gov/40816237/

[2] Taher M, Mazumdar S. Computational tools in rational metalloenzyme design. Methods Enzymol. 2025. https://pubmed.ncbi.nlm.nih.gov/41047208/

[3] Zha Z, Wang Y, Teng C et al. Redesigning myoglobin via functional site scaffolding for enhanced catalytic functions. Biochem Biophys Res Commun. 2026. https://pubmed.ncbi.nlm.nih.gov/41855859/

[4] La Gatta S, Leone L, Sgueglia G et al. Engineering a Functional Histidine Brace Copper-Binding Site into a De Novo-Designed Protein Scaffold. JACS Au. 2025. https://pubmed.ncbi.nlm.nih.gov/41169579/

[5] Noroozi Tiyoula F, Vafaee Sharbaf F, Rahimian K et al. Artificial intelligence in metalloprotein binding site prediction: A systematic review bridging bioinformatics and biotechnology. Int J Biol Macromol. 2025. https://pubmed.ncbi.nlm.nih.gov/40885350/

[6] Persico F, Galuzzi BG, Pellegrino M et al. Predicting Metalloprotein Redox Potentials with Machine Learning: A Focus on Iron-Sulfur Systems. J Chem Inf Model. 2025. https://pubmed.ncbi.nlm.nih.gov/41165319/

[7] Coronado KR, Zhu Y, Mann SI. De novo design of four-helix bundle proteins to bind metalloporphyrin cofactors. Methods Enzymol. 2025. https://pubmed.ncbi.nlm.nih.gov/41047202/

[8] Prakash D, Wu Y, Misra SK et al. Mechanism of Oxidative C─H Bond Activation by De Novo Designed Artificial Cu Metalloenzymes Using H₂O₂ and O₂. Chemistry. 2025. https://pubmed.ncbi.nlm.nih.gov/40801149/ *** Disclaimer: This article is for educational and informational purposes only. It is not intended to substitute for professional veterinary advice, diagnosis, treatment, or regulatory guidance. Always consult a licensed veterinarian or qualified specialist regarding animal health, disease diagnosis, and therapeutic decisions.

[9] Heidari H, Phan D, Lawson D et al. Design and preparation of artificial heme-copper enzymes. Methods Enzymol. 2025. https://pubmed.ncbi.nlm.nih.gov/41047215/

[10] Snell EH, Grime GW, Webb SM et al. Assessing Metal Ion Assignment Accuracy in Protein Data Bank Models via Elemental Spectroscopy. J Chem Inf Model. 2026. https://pubmed.ncbi.nlm.nih.gov/42290629/

[11] Shao B, Li P, Liu ZP. A protein surface-aware multimodal framework for residue-level metal-binding site recognition. Cell Rep Methods. 2026. https://pubmed.ncbi.nlm.nih.gov/42013856/

[12] Kırboğa KK, Küçüksille EU. Integration of Evolutionary Analysis With RFdiffusion for De Novo Design of Aggregation-Resistant Frataxin. Proteins. 2026. https://pubmed.ncbi.nlm.nih.gov/41563298/

[13] Suhagia TA, Cheng Q, Summers TJ et al. Addressing Long-Standing Challenges in Computational Enzymology With Large QM-Cluster Models of the [Ni, Fe]-Hydrogenase Proton Transfer. J Comput Chem. 2025. https://pubmed.ncbi.nlm.nih.gov/41063669/

[14] Teptarakulkarn PH, Treviño RE, Hansen AL et al. Correlating Protein Dynamics and Catalytic Activity of a Model Hydrogenase Using Paramagnetic and Biological Nuclear Magnetic Resonance Spectroscopy. J Am Chem Soc. 2026. https://pubmed.ncbi.nlm.nih.gov/41493147/

[15] Goullieux M, Zoete V, Röhrig UF. Hybrid quantum/classical docking of covalent and non-covalent ligands with Attracting Cavities. Sci Rep. 2025. https://pubmed.ncbi.nlm.nih.gov/41298542/

[16] Yu H, Eres M, Hilburg SL et al. Random heteropolymers as enzyme mimics. Nature. 2026. https://pubmed.ncbi.nlm.nih.gov/41476271/

[17] Chi H, Xia B, Shen J et al. Overcoming the Stability-Activity Trade-Off in Lipoxygenase by Integrated Computational-Assisted Structure-Guided Design. J Agric Food Chem. 2026. https://pubmed.ncbi.nlm.nih.gov/41800710/

[18] Su HC, Huang CW, Wang SH et al. Structure-guided rational design of ferritin nanocages unlocks thermoresponsive channels for accelerated drug encapsulation. Int J Biol Macromol. 2026. https://pubmed.ncbi.nlm.nih.gov/41638275/

[19] Zhao W, Huang F, Uddin S et al. Computer-aided drug design of SP94 peptide-functionalized human H-chain ferritin for targeted doxorubicin delivery. Int J Biol Macromol. 2025. https://pubmed.ncbi.nlm.nih.gov/41138863/

[20] Marimuthu SK, Ramakrishnan V, Thamotharan S. Targeting dual substrate pockets of colistin resistance conferring MCR-1 of Escherichia coli with natural products: insights from high throughput virtual screening and molecular dynamics simulations. J Biomol Struct Dyn. 2025. https://pubmed.ncbi.nlm.nih.gov/41020634/

[21] Zhai N, Zhou C, Cheng L et al. Binding differences of fluxapyroxad with succinate dehydrogenase across species: insights from in silico simulations. Pest Manag Sci. 2025. https://pubmed.ncbi.nlm.nih.gov/40899335/

[22] Bijelic V, Momoli F, Liebman M et al. Ferritin Reference Curves and Optimal Curves in Preadolescent Children. JAMA Netw Open. 2026. https://pubmed.ncbi.nlm.nih.gov/42138918/

[23] Shiga S, Sugiyama S, Ito S et al. The Design of Metal Ion-Induced Dimers Suggestive of 3D Domain Swapping. Chembiochem. 2026. https://pubmed.ncbi.nlm.nih.gov/42057663/

[24] Zheng X, Lin X, Tao J et al. Harnessing machine learning and multi-scale modeling to discover novel ALOX15 inhibitors from marine natural products. Mol Divers. 2026. https://pubmed.ncbi.nlm.nih.gov/42012717/

[25] Ratanachotpanich T, Chumgate A, Lengwehasathit K et al. Bridging In Silico design and experimental validation: Virtual screening and In Vitro assessment of biomimetic anticancer peptides inspired by horseshoe crab hemolymph proteins. Eur J Pharm Sci. 2026. https://pubmed.ncbi.nlm.nih.gov/41999784/

[26] Oliw EH. Structural analyses of oxygen channels of animal, soybean and manganese lipoxygenases. Arch Biochem Biophys. 2026. https://pubmed.ncbi.nlm.nih.gov/41933860/

[27] Sitkov N, Ryabko A, Ivanov S et al. Biosensor-Based Detection of Calprotectin and Lactoferrin as Neutrophil-Derived Markers of Inflammatory Bowel Diseases: From Molecular Pathophysiology to Point-of-Care Platforms. Int J Mol Sci. 2026. https://pubmed.ncbi.nlm.nih.gov/41898556/

[28] Lombardo L, Agnello F, Gitto R et al. A Multistep Computational Approach to Achieve a Complete Human 5-Lipoxygenase Structure and Provide a Pharmacophore Model for Further Drug Design. Mol Inform. 2026. https://pubmed.ncbi.nlm.nih.gov/41877544/

[29] Sun Y, Li K, Wu R et al. Scaffold Hopping Combined with 3D-QSAR for the Discovery of Succinate Dehydrogenase Inhibitors Containing N-Benzyloxyformamide. J Agric Food Chem. 2026. https://pubmed.ncbi.nlm.nih.gov/41592794/

[30] Oh DH, Kang JH, Lee OH et al. Cell swelling and upright mounting-based imaging for high-resolution visualization of intracellular trafficking across the BBB using conventional confocal microscopy. Drug Deliv. 2026. https://pubmed.ncbi.nlm.nih.gov/41496478/

[31] Zhou C, Ge J, Li Z et al. Rational Design of β-Ketonitrile Acaricides through Binding-Mode-Guided Isosteric Ring Replacement. J Agric Food Chem. 2026. https://pubmed.ncbi.nlm.nih.gov/41453360/

[32] Baharlouei Z, Karimzadeh F, Sanati A et al. Toward non-invasive diagnostics through AuNSs@Nano-MIP biosensor for sensitive lactoferrin detection in sweat. Talanta. 2026. https://pubmed.ncbi.nlm.nih.gov/41418616/

[33] Tsopka IC, Pontiki E, Sigala I et al. Design, Synthesis, Biological Evaluation, and In Silico Studies of Novel Multitarget Cinnamic Acid Hybrids. Molecules. 2025. https://pubmed.ncbi.nlm.nih.gov/41375178/

[34] Naïdji C, Katrib C, Devos D et al. Histo-radiological correlations in an early-stage model of a Parkinson's disease. Exp Neurol. 2026. https://pubmed.ncbi.nlm.nih.gov/41173226/

[35] Hao Z, Han J, Zhang Y et al. Discovery of Aromatic Amide Derivatives Bearing Oxime Ethers Targeting Succinate Dehydrogenase. J Agric Food Chem. 2025. https://pubmed.ncbi.nlm.nih.gov/40977077/