Engineering electron tunneling rates
Natural photosynthetic reaction centers achieve photochemical charge separation by anchoring a light-activated pigment between two electron transfer chains. One chain provides an electron acceptor for the excited pigment, and the other an electron donor that restores the pigment ground state while creating a charge-separated donor-acceptor pair (Fig. 1a). While electron accepting or donating chains have been developed in modified natural proteins18,19,20,21,22,23,24,25,26 and in a de novo protein with a crosslinked ruthenium trisbipyridine derivative15, our design condenses the extended electron transfer chains of natural reaction centers to a core elementary donor-pigment-acceptor triad, symbolized as DPA27,28.
Upon light activation forming an excited pigment singlet state (D1P*A), the pigment may undergo radiative decay to the ground state (DPA) or relax to the triplet excited state (D3P*A) via intersystem crossing. This triplet state may in turn decay to the ground state by phosphorescence (k3, Fig. 1b). During the excited state lifetime, electron transfer to a suitably placed acceptor forms DP+A− with rate k1. Charge-separated states can short-circuit through charge recombination (k4, forming DPA) or undergo a further charge separation (k2, forming D+PA−). In the absence of additional donors, acceptors or catalytic redox reactions the charge-separated state eventually decays by direct electron transfer (k5) or indirect electron transfer through reverse reactions of k−2 and k−1 (Fig. 1b).
To engineer the appropriate cofactor spacing, we estimate rates of downhill intraprotein electron tunneling using a well-tested empirical expression, Eq. (1) (see Methods)1,3,29. This tunneling expression not only allows us to understand the engineering of natural electron transfer proteins1,2,30 but also to consider the relative performance of artificial DPA triad designs in advance of construction (Fig. 1c, d).
Light-active redox protein design
To begin reverse-engineering photosynthetic reaction centers, we use first principles of protein folding to construct a single-chain framework for positioning DPA cofactors that consists of an extended four α-helix bundle coiled-coil. Such constructs are called maquettes9,31,32,33,34 and they perform the same role as the scale maquettes made by architects to evaluate their designs prior to constructing the final product. Maquette amino acid sequences are not based on any specific natural protein. Instead, they employ charge complementarity and a binary pattern of polar and nonpolar residues to enforce a desired helical threading9,32. Further details of RC maquette design will be described elsewhere; see Supplementary Table 1 for amino acid sequences.
The RC maquette uses histidine (His) residues in the hydrophobic core to coordinate the central metals of tetrapyrroles for site-specific cofactor binding. A single His ligates a light-activatable Zn porphyrin in the pigment site10,11,12,35, while two His residues on opposing helices bis-His ligate an Fe porphyrin in the acceptor site10,14,31,36,37. The RC maquette can bind different electron acceptors including heme B or Fe(III) 2,4-diacetyl deuteroporphyrin IX (DADPIX) tightly with KD ≤ 10 nM and with Em values of −0.19 and +0.01 V vs SHE, respectively. Amphiphilic Zn tetrapyrroles show the strongest binding to our de novo proteins12. In the results reported here, we use Zn 5-phenyl, 15-(p-carboxyphenyl) porphyrin (ZnP, KD ≤ 10 nM, Em 0.91 V) as our pigment, P. See Supplementary Fig. 1 for spectro-electrochemical determination of Em values of Fe and Zn porphyrins.
Distances between cofactors are the primary determinants of electron transfer rates in the RC maquette. Inter-cofactor distances are subtly influenced by core packing, the helical register, and the orientations of porphyrin rings, but larger distance changes are more easily accomplished by moving cofactor-ligating residues along the helices in increments of a full helical turn (~5.2 Å). Previous multi-cofactor de novo protein designs10,13,14,38 have edge-to-edge distances between porphyrins that are too short or too long for high-yield charge separation in DPA triads as shown in the theoretical contour plot of Fig. 1c (calculated from Eq. 1 with the kinetic scheme of Fig. 1b). Instead, the RC maquette needs a ~13–18 Å P-to-A distance (corresponding to four or five helical turns between the P- and A-site His residues) in order to achieve stable charge separation for various porphyrin acceptors with a wide range of redox properties. Details of the electron tunneling yield calculations are provided in Methods.
The contour plot of Fig. 1d indicates that the electron donor must be placed closer to P to assure sufficiently rapid electron tunneling rates that can compete with short-circuiting DP+A− charge recombination, k4 (see Fig. 1b). D-P distances up to ~10 Å and Em values less than 0.75 V will give the best D+PA− yield. To reproduce PSII-like photo-oxidation activity, we chose a Tyr-P distance of 5 Å for Tyr oxidation, and we assembled a metal cluster at a distance of 10 Å from P. The bi-nuclear metal cluster uses a two-His/four-Glu motif seen in many natural metal clusters such as bacterioferritin21. The basis of our sequence selection in the metal cluster donor region of the RC maquette comes from an earlier de novo protein maquette named Due Ferri 3 (DF3)39,40,41. The sequence of DF3 was modified to integrate it into the electron donor region of the RC maquette in several ways. First, redox-inactive amino acids replaced Trp and Tyr residues that were not intended to be electron donors. Next, the loop from DF2t42 replaced a His-containing loop from DF3 to avoid nonspecific His interactions with cofactors. A flexible linker was added to connect the two halves of the dimer. Finally, four helical Gly residues (one per helix) were inserted where the electron donor region meets the pigment binding site in order to increase local backbone flexibility and facilitate the transition from one domain to the next. The DeGrado and Lombardi groups independently developed an analogous strategy to fuse a DF protein to a synthetic Zn porphyrin-binding maquette11. The operation of the Tyr and metal ion donors in the RC maquette is compared with an unnatural maleimide-functionalized ferrocene donor (Fc) anchored to a Cys residue that replaces a Tyr residue at two different D-P distances (either Y168C for Fc168 or G164C/Y168L for Fc164).
Crystal structures of the RC maquette confirm the intended coiled-coil design with pigments and redox centers assembled at the designed positions (Fig. 2). To the best of our knowledge, this represents the first crystal structure of a designed protein containing two unique porphyrins.
The RC maquette has many structural features in common with natural proteins, despite the lack of sequence similarity and the absence of evolutionary natural selection. Hydrophobic core packing, rotameric states of heme-ligating histidines, and second shell histidine-threonine hydrogen bonds are similar to the cytochrome b subunit of cytochromes bc1 and b6f (Fig. 2a)43. Even the orientation of the maquette heme relative to the superhelical bundle axis is conspicuously similar, suggesting that the particular conformation of heme in the natural proteins is not precisely controlled for any functional purpose, but rather follows directly from structural constraints demanded by this protein fold. The light-active pigment binding site differs from the cytochrome b heme site by providing only a single His for ligation.
In the electron donor module, a single Leu for His residue exchange (L71H mutation) in the bundle core immediately provides an H-bond for the Tyr hydroxyl, mimicking the proton-coupled electron transfer geometry of donor Tyr Z in PSII44,45 (Fig. 2b). In the nearby metal binding site, Glu/His ligation of Cd(II) in the RC maquette accurately reproduces the geometry of the binuclear Mn(II) cluster shown to be photochemically active in modified bacterioferritin (Fig. 2c)20,21.
Crystal structures with Zn(II) (PDB ID: 5VJS and 5VJT) show only one metal in the di-metal site, with a similar Tyr-to-closest metal distance as in the Cd(II) structure (PDB ID: 5VJU); the second metal site in the Zn(II)-containing structures is vacant (Supplementary Fig. 2). The Zn(II)-containing structures come from an acidic crystallization condition (pH 4.4–4.5) that likely protonates a metal site Glu, suppressing binding of the second metal. At the same time, nearby deprotonated Asp residues on the exterior are free to recruit a Zn(II) ion in a tetrahedral coordination geometry.
Spectroscopic titrations show that 2.0 pentacoordinate Co(II) ions bind the RC maquette at neutral pH in solution and that 2.0 Mn(II) ions bind the RC maquette-heme-ZnP complex with high affinity (Supplementary Fig. 3). The successful construction of the electron donor site in the L71H RC maquette represents a first step in exploring the Tyr-mediated photo-oxidative assembly of a PSII-like Mn cluster for water oxidation in a designed protein46.
Light-driven electron transfer dynamics
Transient absorption spectroscopy in the visible spectral region reveals charge-separated states and electron transfer rates in the RC maquette. Singular value decomposition (SVD) of the transient spectra (Fig. 3, Supplementary Figs. 4–11, and Supplementary Table 2) resolves a time and wavelength dependence that is fit to an elementary kinetic model of first order reactions between the excited, charge separated and ground states to generate rate constants (Fig. 3c) and spectra of the intermediate states (Fig. 3d). Details of data analysis are given in Methods.
When the RC maquette is assembled with only P and no donor or acceptor (Fig. 3b, ZnP monad and Supplementary Fig. 4a), a 2 ns 532 nm laser pulse bleaches the ground state Soret absorption of P at 424 nm, and the broad absorption from 440 to 550 nm of the excited triplet state increases, decaying to the ground state in 3–5 ms. This rate is similar to that observed in other artificial protein designs12.
When an acceptor is also bound to make a PA (ZnP-ferric heme) dyad, the same triplet signature is observed but now decaying at 60 ± 30 µs, presumably because P* reduces the acceptor. The absence of any monad-like P* decay component with a 3–5 ms lifetime indicates a near-unity quantum yield of A reduction by P*. As expected, charge recombination in the dyad is rapid, because the large driving force closely matches the reorganization energy. Consequently, the P+A− intermediate does not accumulate in the dyad, and a spectroscopic signal of reduced heme cannot be resolved (see Fig. 3b and Supplementary Fig. 4b). Triplet inhibition of charge recombination is not apparent. When the heme acceptor in the dyad is inactivated by pre-reduction with dithionite, P* reverts to monad-like behavior with a 3–5 ms decay to the ground state.
To trap a long-lived charge separated state, the RC maquette requires adding a donor to the dyad to form a DPA triad. Substitution of donor Tyr for Leu168 stabilizes heme reduction in the D+PA− charge separated state (see Y-ZnP-heme triad in Fig. 3a–d, Supplementary Figs 6–8, and Supplementary Table 3). The D+PA− charge-separated state difference spectrum identified by SVD (Fig. 3d, blue) is dominated by the signature of heme reduction (Fig. 3d, black dashed line) with minor contributions from 3ZnP* and ZnP+, providing conclusive evidence that heme functions as the electron acceptor in the Tyr-ZnP-heme B triad. The 3ZnP*/ZnP+ difference spectrum has very little shift in the Soret band47, causing this difference to disappear in the noise in the Soret region. The heme redox signal persists for 3 ms, much longer than the 50 µs 3ZnP* bleach recovery. Known extinction coefficients for heme and ZnP provide a yield estimate of 11% from the excited triplet state, which in turn permits an estimate of Tyr to P+ ET of ~30 µs (Fig. 3c and Supplementary Fig. 6a).
As in natural PSII, Tyr oxidation is a proton-coupled electron transfer (PCET) reaction45. Depending on design and environmental conditions, electron transfer may involve Tyr deprotonation before thermodynamically favorable ET (observable at pH values near the Tyr pKa of ~10)48, endergonic ET before PT, or concerted PTET. Figure 3a–d suggest tyrosinate ET at pH 9.5, with an Em value of ~0.72 V48. Engineered proton acceptors may increase the yield of electron transfer by lowering the pKa of Tyr and increasing the population of tyrosinate. While the introduction of a hydrogen bond from Tyr to a His residue appears to have little impact on the lifetime or yield of charge separation by itself, the addition of Asp31 adjacent to the His71-Tyr168 hydrogen bond raises the yield from 2 ± 1% to 11.5 ± 0.7% and shortens the lifetime of charge separation from 40 ± 20 ms to 3 ± 1 ms at pH 9.5 (Supplementary Figs. 6a, 7, and 8a and Supplementary Table 3). Asp31 was intended to make a secondary hydrogen bond to His71, analogous to the D1-Asn298/D1-His190/D1-Tyr161 (YZ) hydrogen bond network in PSII49,50. Without a crystal structure of the L31D/L71H RC maquette mutant, however, we cannot confirm that the Asp31-His71 hydrogen bond forms; it remains possible that Asp31 raises the quantum yield of electron transfer by altering the dynamics and hydration of Tyr168 rather than by acting directly as a proton acceptor for His71.
The carboxylate-rich binuclear metal binding site of the RC maquette extends the electron donor chain. Spectroscopic monitoring of Co(II) binding yields KD values of 0.5 and 15 µM for the first and second metal bound (Supplementary Fig. 3a). The Irving-Williams series51 predicts the trend of metal complex stability as Mn(II) < Fe(II) < Co(II); thus the Co(II) affinity represents an approximate upper limit for Mn(II) and Fe(II) affinities.
Mn(II) binding stabilizes the binuclear metal site, increasing the RC maquette melting temperature (monitored by circular dichroism) from 81 °C to >100 °C. Binding of two Mn(II) ions per protein is confirmed by a bathochromic shift in the porphyrin Soret band of the ZnP/heme RC maquette (Supplementary Fig. 3).
Fe(II) binding is apparent from a similar Soret band shift. Bacterioferritin-like ferroxidase activity demonstrates that the diiron site is redox active (Supplementary Fig. 12). Light-induced electron transfer is clear as near-stoichiometric Fe(II) prolongs the lifetime of charge separation from 2.0 ms to 250 ms at pH 7.5 in RC maquette variant L71H (Fig. 3e, orange and Supplementary Figs. 8b and 10a). Tyr promotes electron transfer from Fe(II) to ZnP+ (Supplementary Fig. 10), as knocking out Tyr168 (Y168L) lowers the yield of Fe(II)-facilitated charge separation to an undetectable level.
Unbound Fe(II) is not a viable electron donor at near-stoichiometric concentrations, or even at 10-fold excess, as seen by the suppression of Fe(II)-facilitated charge separation by pre-loading the maquette with stoichiometric Zn(II). When a very large excess of Fe(II) is added (1.1 mM) in the absence of Zn, Cd, or Mn, the kinetic behavior is altogether different from stoichiometric Fe(II) concentrations: the measured ferrous heme lifetime is greater than 10 s. Thus, at very high concentrations, Fe(II) can act as an exogenous electron donor and diffuse away before charge recombination.
At neutral pH, Zn(II) and Cd(II) suppress Tyr oxidation. These metals may electrostatically inhibit proton transfer from Tyr to protonatable residues in the metal cluster site. A similar effect could apply to redox-active bound Fe(II); however, even if transient Tyr oxidation were energetically uphill, subsequent energetically favorable electron transfer from Fe(II) could kinetically stabilize charge separation. In comparison, Mn(II) oxidation after transient Tyr oxidation will not be as energetically favorable, so less kinetic stabilization is expected.
Mn oxidation is not simple. During photo-assembly in natural PSII46, initial Mn(II) photo-oxidation steps are coupled to a dark rearrangement that takes ~150 ms, while in modified purple bacterial reaction centers somewhat similar to our system, Mn(II) oxidation has a lifetime of 12 ms22,26. These rates are much slower than the rate of direct electron tunneling at these distances. We expect Mn(II) photo-oxidation to be slowed by site rearrangements on a timescale longer than the 2.7 ms charge recombination of the tyrosine-ZnP-heme triad (see Supplementary Figs. 8b and 9). Slower charge recombination is required for higher yields of Mn oxidation.
Like Fe(II), ferrocene (Fc) electron donors permit long charge separation lifetimes in the RC maquette (Fig. 1d), because their relatively low Em values inhibit the uphill reverse electron transfer from P to D+. Maleimide-functionalized Fc (Em 0.56 V in DMF) is coupled to a Cys residue to make Fc168 or Fc164 (sequences are given in Supplementary Table 1). Fc Em values are modulated by the protein environment; Fc-modified azurin ranges from 0.51 to 0.55 V depending on surrounding amino acids52. Figure 3e (cyan) and Supplementary Fig. 5a give a 350 ms lifetime for charge separation in the Fc168-ZnP-heme B triad based on the heme redox signal.
At the electron accepting end of the RC maquette, we confirm the prediction (Fig. 1c) that the electron transport chain remains functional over a wide range of acceptor midpoint potentials. Replacement of heme B in the Fc168-ZnP-heme B triad with the Fe porphyrin DADPIX, which is 200 mV more oxidizing, changes the lifetime of charge separation from 350 ms to 20 ms and the yield of D+PA− from 4% to 31% (see Supplementary Fig. 5).
As an acceptor, Co(III)PPIX achieves highly desirable electrochemical and photochemical proton reduction to H2 in natural proteins17. It binds tighter to the RC maquette (KD 250 nM) than to the natural 4-helix bundle cytochrome b56217. While bound CoPPIX is indeed photoreduced by the Zn porphyrin in the RC maquette under steady state illumination in the presence of ascorbate and EDTA as sacrificial electron donors, the quantum yield of long-lived charge separation using CoPPIX as the direct acceptor is low (see Supplementary Fig. 11). Work with CoPPIX in myoglobin (Em + 0.1 V) is orders of magnitude slower than heme reduction53, a sign that additional reorganization barriers, such as ligand exchange, accompany CoPPIX reduction. Just as slow Mn photo-oxidation is likely to be assisted by extending the donor redox chain, designs using CoPPIX as an effective terminal reductive catalytic center must pair it with a rapid initial electron acceptor in a reducing electron transfer chain.