Atomically Accurate Design of Metalloproteins with Predefined Coordination Geometries

Abstract

We report a new computational protein design method for the construction of oligomeric protein assemblies around metal centers with predefined coordination geometries. We apply this method to design two homotrimeric assemblies, Tet4 and TP1, with tetrahedral and trigonal pyramidal tris-histidine metal coordination geometries, respectively, and demonstrate that both assemblies form the targeted metal centers with ≤0.2-Å accuracy. Although Tet4 and TP1 are constructed from the same parent protein building block, they are distinct in terms of their overall architectures, the environment surrounding the metal centers, and their metal-based reactivities, illustrating the versatility of our approach.

Graphical Abstract

Despite a limited set of bioavailable metal ions and amino acids capable of metal coordination, natural metalloproteins perform diverse functions including signaling,^1–2 electron transfer,^3–4 small molecule transport,^5–6 and catalysis.^7–10 Underlying such functional diversity is an intricate interplay between protein structure and metal coordination.^11–16 While this interplay takes place at several levels (e.g., overall protein structure/dynamics, secondary coordination sphere surrounding the metal center), the core determinant of a metalloprotein’s function is the metal center itself, the ligand composition, and the geometry of the primary coordination sphere.¹⁷ An accurate control of the metal coordination geometry by the protein structure is required not only for the selective binding of a cognate metal ion^18–23 but also for tuning its inherent reactivity,²⁴ as exemplified by many biological metal centers that are scaffolded in coordinatively unsaturated and strained geometries.^{9, 25} Inspired by natural bioinorganic systems, there has been great interest in designing proteins featuring metal centers with tailored geometries and reactivities/properties.^{13, 26–27} However, despite advances in computational protein design and the development of metal search/placement algorithms,^28–31 the sub-Å positional accuracy needed for this purpose has yet to be demonstrated.

To date, most de novo designed metalloproteins have been based on α-helical motifs.^{13, 32–34} The small sizes and highly parametrizable nature of these systems have facilitated the incorporation of metal ions with desired coordination environments and proved invaluable in exploring the minimal structural requirements in proteins for metal-based functions.^{26, 33, 35–41} Yet, the same features also restrict the scope of metal active sites and geometries that can be accommodated as well as the incorporation of functionally important structural motifs (e.g., large cavities, flexible loops). Similar challenges also apply to non-α-helical peptide motifs.^42–48 Examples of larger, designed metalloproteins have entailed either coordinatively saturated metal centers,⁴⁹ resulted in unexpected deviations from targeted coordination geometries,^50–52 or relied on recreating secondary structure elements adopted from natural metalloproteins.^53–54 We developed an alternative design approach (Metal-Templated Interface Redesign) based on the metal-directed self-assembly of protein building blocks into oligomeric architectures.^55–57 Using the structures of these assemblies as a template, the protein-protein interfaces bearing the nucleating metal centers are engineered to increase preorganization for metal binding and obtain diverse metal-based functions.^{22, 58–64} However, the structural outcome of metal-directed protein assembly is not always predictable, and the resulting interfacial metal centers are generally–but not always–coordinatively saturated, limiting access to alternative coordination geometries of interest.^{11–12, 57}

To overcome these limitations, we have developed a new computational design strategy, in which oligomeric protein assemblies are built around metal centers with predefined coordination geometries. As our initial target, we chose a coordinatively unsaturated, tris-histidine-coordinated Zn^Il center with an exchangeable ligand (Zn-His₃X) in a tetrahedral geometry, found in the active sites of enzymes such as carbonic anhydrases and matrix metalloproteinases.^65–69 As our model building block, we used an engineered, heme-free variant of cytochrome cb₅₆₂, ApoCyt, a four-helix bundle protein that is stable, tolerant to mutations, and crystallizes readily.⁷⁰ The first step in our workflow involved defining the geometric parameters for the targeted metal coordination geometry (Figure 1a). An ideal tetrahedral Zn-His₃X center is C₃ symmetric and can be described with five parameters: d₁ and Θ₁ for the Zn-His bond distance and His-Zn-His bond angles, respectively, and Θ₂, Θ₃, and Θ₄ for the rotation of the imidazole ring (Figure 1a). For a tetrahedron, Θ₁ = 109.5°, and d₁ was set at 2.0 Å.⁷¹ Using simple Zn^II(imidazole)₃(OH) models, we performed a series of density functional theory (DFT) calculations, which revealed that varying Θ₂ had the least effect on the energy of the system, whereas deviations of Θ₄ from 0° resulted in largest increases in energy. Based on these results, we used the following ranges in subsequent design stages: −40°≤ Θ₂ ≤+40°, −15°≤ Θ₃ ≤+15, and −15°≤ Θ₄ ≤0°.

Figure 1.

Workflow for the design of protein assemblies with predefined metal coordination geometries.

Next, we implemented a protein docking procedure (termed Metal-Directed Protein Docking, Figure 1b) to place three C₃-symmetry-related ApoCyt monomers to form the desired Zn-His₃X center (within the allowable Θ₂, Θ₃, Θ₄ ranges), while yielding sufficiently large intermonomer interfaces that can be redesigned to stabilize the resulting assembly. Briefly, this procedure involved: (1) placement of a His residue at a manually chosen position (38 or 66) on each monomer; (2) calculation of sidechain coordinates based on a given set of geometric parameters (d, Θ’s) and the symmetry of the metal center; (3) energetic evaluation of the resulting assemblies based on solvent accessible surface areas (SASA) of the monomers and a Rosetta⁷² centroid score function to identify backbone clashes; (4) repetition of steps 1–3 to sample combinations of His positions, geometric parameters, and torsion angles to yield a library of trimeric ApoCyt structures with a Zn-His₃X center. From this library, we selected several structures for multiple iterations of interface redesign by Rosetta (Figure 1c),⁷² ultimately yielding seven designs encompassing five distinct docking geometries (Figure S1) that were then evaluated using AlphaFold2⁷³ for structure prediction (Figure 1d). Of these seven designs, five had significant disagreement (αC-RMSD >10 Å) between the computed model and the AlphaFold2 prediction, whereas two had good agreement (αC-RMSD <2.5 Å, Figure S2). Upon bacterial expression and purification of the two promising designs, one was found to be predominantly monomeric in solution, whereas the second, Tet4, formed a metal-independent trimer as desired, with a dissociation constant (K_d) of 2.7 nM for Zn^II (Figures 2a, S3). In contrast, four of the five designs that showed large deviations from AlphaFold2 predictions either failed to express in bacterial cultures or did not assemble into a trimer (Table S1, Figure S4), validating the in silico screening step. The remaining design formed a trimer but did not crystallize and possessed >10-fold weaker affinity for Zn^II than Tet4 (Figure S3).

Figure 2.

(a) Analytical ultracentrifugation profiles of apo-Tet4 (blue) and Zn-Tet4 (magenta). (b) Thermal denaturation of apo-Tet4 (blue), Zn-Tet4 (magenta) and ApoCyt (grey). (c) Superposition of experimental (magenta) and designed (grey) structures of Zn-Tet4 (PDB: 8SJG), and (d) close-up views of engineered interfacial residues. (e) Views of the Zn-³⁸His₃ center (Zn – grey sphere, water – red sphere), along with 2F_o-F_c (grey mesh, 1.0σ) and Zn-anomalous maps (blue mesh, 5.0σ). (f) Superposition of Zn-Tet4 (magenta) and apo-Tet4 (cyan, PDB: 8SJF) structures.

Tet4 displayed considerably improved thermal stability over ApoCyt, retaining nearly ~60% of its native α-helical structure at 100 °C (Figure 2b). We determined the crystal structures of Tet4 in the Zn^II-bound and apo states at resolutions of 2.4 Å and 2.3 Å, respectively. The Zn-Tet4 structure was in excellent agreement with the designed model (αC-RMSD = 1.2 Å) (Figure 2c–d). Importantly, the Zn-³⁸His₃ center, which also included an axial aqua ligand, possessed a nearly ideal tetrahedral symmetry, with d_1,avg = 2.0 Å and Θ_1,avg = 104.9° (Figure 2e, Figure 3). Overall, the design accuracy of Zn-³⁸His₃ center (based on the deviation of His N_ε atoms and Zn from target positions) was 0.12 Å. Apo-Tet4 adopted a more open trimeric arrangement compared to Zn-Tet4, whereby the αC distances between the ³⁸His residues increased from 10.3 Å to 13.4 Å (Figure 2f). This metal-dependent shift was accommodated by the malleable hydrophobic interfaces between the monomers and indicated that the desired Zn-³⁸His₃ tetrahedral geometry was obtained despite the lack of rigid preorganization of the assembly (Figure S5).

Figure 3.

Geometric parameters for Zn-Tet4 and Zn-TP1 metal centers.

To further demonstrate the utility of our method, we next targeted a trigonal planar Zn-His₃ center. In contrast to tetrahedral geometries, trigonal planar Zn^II centers are rarely observed in proteins (Figure S6). In fact, in our search of the RCSB database,⁷⁴ we could not find a metalloprotein with a trigonal planar His₃-Zn motif, suggesting that this geometry may be thermodynamically less favorable. We therefore reasoned that a stringent test of our approach would be to design a preorganized ApoCyt assembly that would enforce a trigonal planar Zn-His₃ coordination geometry, which would require an accuracy of ≤0.2 Å in the positions of His N_ε atoms to discriminate between the two geometries (assuming d₁ = 2.0 Å). Again using ApoCyt as our building block, we sampled the same set of geometric parameters as previously, except that Θ₁ was set at 120°. This search resulted in a new set of docked trimer structures, from which we chose one for interface redesign. Of the five promising design candidates with low Rosetta scores, only one candidate, TP-1, had an αC-RMSD of <2.5 Å compared to the AlphaFold2 prediction (Figure S7) and was therefore chosen for experimental characterization.

Like Tet4, TP1 formed a metal-independent trimer with high thermal stability (Figures 4a,b). TP1 also bound Zn^II with high affinity (K_d = 62 nM), albeit >20-fold more weakly than Tet4 (Figure S8), affirming that the trigonal planar geometry was energetically less favorable. The 1.6-Å resolution crystal structure of Zn-bound TP1 aligned nearly perfectly with the design model (αC-RMSD = 0.9 Å) as well as with the 1.5-Å resolution crystal structure of apo-TP1 (αC-RMSD = 0.3 Å), with a design accuracy of 0.21 Å for the Zn-⁶⁶His₃ center (Figures 4c,d,f). The particularly close agreement between the apo- and Zn-bound TP1 structures pointed to a high level of preorganization, which indeed enforced a considerably more planar arrangement of the ⁶⁶His₃-Zn center in TP1 compared to the ³⁸His₃-Zn center in Tet4 as evidenced by: (1) an increase in N_ε-N_ε distances by 0.2 Å (while maintaining d₁=2.0 Å), (2) a decrease in the “doming” angle (Θ_doming) from 23.8° to 13.0°, and (3) an increase in Θ₁ from 104.9° to 115.1° (Figures 3, 4e). The slight doming in ⁶⁶His₃-Zn was likely caused by an axial chloride ligand, yielding a distorted trigonal pyramidal geometry.

Figure 4.

(a) Analytical ultracentrifugation profiles of apo-TP1 (blue) and Zn-TP1 (magenta). (b) Thermal denaturation of apo-TP1 (blue), Zn-TP1 (magenta) and ApoCyt (grey). (c) Superposition of experimental (magenta) and designed (grey) structures of Zn-TP1 (PDB: 8SJH), and (d) close-up views of engineered interfacial residues. (e) Views of the Zn-⁶⁶His₃ center (Zn – grey sphere, chloride – green sphere), along with 2F_o-F_c (grey mesh, 1.0σ) and Zn-anomalous maps (blue mesh, 5.0σ). (f) Superposition of Zn-TP1 (magenta) and apo-TP1 (cyan, PDB: 8SJI) structures.

Although Tet4 and TP1 are both constructed from ApoCyt monomers, they use His residues which lie on different helices of ApoCyt for metal coordination, ultimately leading to different trimer arrangements. In Tet4, the monomers are tilted by ~31° from the C₃ axis to give a conical shape, whereas in TP1 this value is ~11° to give a parallel arrangement (Figures 5a,b). Consequently, TP1 possesses larger intermonomer interfaces than Tet4 (1420 Ų vs. 1230 Ų per monomer), likely accounting for its greater structural pre-organization for Zn^II binding, although differences in crystal packing interactions cannot be discounted. This arrangement of TP1 also results in a deeply buried Zn center, which is connected to the surface through a single file of water molecules within a hydrophobic tunnel formed along the C₃ axis. This observation suggests that metal-templated design of helical structures may complement existing approaches for the design of selective ion/water channels.^75–77 The conical arrangement of Tet1, in contrast, places the ³⁸His₃-Zn center in a surface-accessible position (Figure 5a, right), which we surmised could be used for a catalytic function. Inspired by recent work on carbonic anhydrase,⁷⁸ we examined if Zn-Tet4 could catalyze the abiological reduction of ketones via a putative Zn-hydride species. Indeed, in the presence of a phenylsilane hydride donor, Zn-Tet4 reduced 4-acetylpyridine with turnover number (TON) of 97±1 and an enantiomeric excess (ee) of 18% (Figure 5c, Table S5) in 6 h. The latter finding indicates that the protein environment surrounding the ³⁸His₃-Zn center imposes some stereoselectivity despite its surface-exposed nature. As anticipated, Zn-TP1 was inactive for the same reaction due to the inaccessibility of its active site.

Figure 5.

(a) Orientation of ApoCyt monomers in Zn^II-Tet4. (b) Orientation of ApoCyt monomers (left) and the central water channel in Zn-TP1. (c) The investigated hydride transfer reaction (top) and corresponding chiral-HPLC traces of relevant species. Reaction conditions were: 10 μmol substrate, 30 μmol phenylsilane, 0.01 μmol protein trimer and/or ZnCl₂, incubated for 6 h at 20 °C.

In conclusion, we have reported here a new method to design proteins with predefined metal coordination geometries with atomic accuracy, bringing us closer to controlling protein-based metal reactivities with the facility demonstrated in synthetic inorganic and organometallic chemistry. This method is straightforward to implement, and its versatility is demonstrated by the facile access to two considerably different protein structures and metal environments from the same building block. Although these proof-of-principle studies focused on symmetric metal centers, we envision that our method can be readily adapted for designing asymmetric metal active sites constructed between disparate protein structural motifs, particularly if complemented by rapidly evolving machine-learning-based tools for protein design.^{53, 79–80}