Efficient Lewis acid catalysis of an abiological reaction in a de novo protein scaffold

New enzyme catalysts are usually engineered by repurposing the active sites of natural proteins. Here we show that design and directed evolution can be used to transform a non-natural, functionally naive zinc-binding protein into a highly active catalyst for an abiological hetero-Diels–Alder reaction. The artificial metalloenzyme achieves >104 turnovers per active site, exerts absolute control over reaction pathway and product stereochemistry, and displays a catalytic proficiency (1/KTS = 2.9 × 1010 M−1) that exceeds all previously characterized Diels–Alderases. These properties capitalize on effective Lewis acid catalysis, a chemical strategy for accelerating Diels–Alder reactions common in the laboratory but so far unknown in nature. Extension of this approach to other metal ions and other de novo scaffolds may propel the design field in exciting new directions. A de novo designed zinc-binding protein has been converted into a highly active, stereoselective catalyst for a hetero-Diels–Alder reaction. Design and directed evolution were used to effectively harness Lewis acid catalysis and create an enzyme more proficient than other reported Diels–Alderases.

T he Diels-Alder reaction has found broad application in natural product synthesis, providing rapid access to complex molecular structures through the formation of two new σ bonds and up to four contiguous stereocentres 1 . Despite its synthetic utility, surprisingly few enzymes have been found to catalyse Diels-Alder reactions in nature [2][3][4][5] . Most of these were discovered only recently and promote intramolecular cycloadditions in polyketide biosynthesis. Designer catalysts for even more challenging bimolecular reactions that simultaneously control the reaction pathway, endo/ exo ratios and the absolute configuration of the reaction products would therefore be valuable additions to the biocatalytic toolkit. Both catalytic antibodies [6][7][8] and computationally designed enzymes have been produced for several Diels-Alder reactions 9,10 , often with useful stereoselectivities, but their efficiencies are generally modest, even after extensive laboratory evolution, reflecting their reliance on simple hydrogen bonding and hydrophobic binding for transition-state stabilization.
Lewis acid catalysis by metal ions, a well-established strategy for accelerating Diels-Alder reactions in the laboratory 11,12 , could potentially augment the capabilities of these designer catalysts. In previous attempts to combine the advantages of transition-metal and enzymatic catalysis for Diels-Alder reactions, ligand-chelated copper(ii) ions were incorporated into the binding pockets of several natural protein scaffolds [13][14][15][16] . Although these artificial metalloproteins promote the enantioselective addition of cyclopentadiene to azachalcones (35-98% ee), the reactions are slow, with reaction times on the order of several days.
In contrast with natural proteins that have been optimized by natural evolution for a specific purpose, de novo protein scaffolds that possess promiscuous binding pockets for metal ions and substrates might be more amenable to functional diversification. Consistent with this hypothesis, we recently transformed a computationally designed zinc-binding peptide into an efficient, enantiospecific metalloesterase 17 . In this Article, we show that MID1sc, a 97-amino-acid-long single-chain version of the same metal interface design 17,18 , is an equally effective template for generating a highly active and stereoselective enzyme for an abiological hetero-Diels-Alder reaction.

Results and discussion
The target reaction. As our target transformation, we chose the Lewis-acid-catalysed reaction of azachalcone 1 with 3-vinylindole 2. These compounds are common substrates for Diels-Alder reactions, related to natural units present in a variety of natural products. However, when paired, they can proceed via competing Diels-Alder and hetero-Diels-Alder pathways (Fig. 1). The development of a selective biocatalyst that promotes only one of the many possible reaction pathways thus poses a chemically interesting challenge.
Density functional theory (DFT) calculations suggest that in the absence of Lewis acid, the endo transition state (TS1-endo, 21.4 kcal mol −1 ) is ambimodal 19 , leading to both sets of products (Extended Data Fig. 1a). The inclusion of Zn(ii)(HO − )(H 2 O) 2 in the calculation shifts the cycloaddition from a slow, concerted reaction to a fast, stepwise reaction with a free energy barrier of 10 kcal mol −1 ; the second step favours the endo hetero-Diels-Alder product (Extended Data Fig. 1b).
Experimentally, 1 and 2 reacted to give cyclohexenes 3 and 3,4-dihydro-2H-pyrans 4 in a 1:3 ratio in dimethyl sulfoxide containing zinc triflate. The product ratio shifted to 1:19 in aqueous buffer containing zinc sulfate (Extended Data Fig. 2a-c). 1 H-NMR analyses of the crude reaction mixtures showed that the endo stereoisomer of the hetero-Diels-Alder adduct 4 is favoured over the exo stereoisomer by a factor of 4.6:1 in dimethyl sulfoxide and 18:1 in buffer.
Design and evolution of a novel enzyme catalyst. The starting MID1sc protein did not detectably catalyse any reaction between 1 and 2. However, Rosetta design 20 using DFT-optimized transition-state geometries predicted that two mutations, E32L and K68W, located on opposite ends of the binding pocket harbouring Efficient Lewis acid catalysis of an abiological reaction in a de novo protein scaffold Sophie Basler 1,4 , Sabine Studer 1,4 , Yike Zou 2 , Takahiro Mori 1 , Yusuke Ota 1 , Anna Camus 1 , H. Adrian Bunzel 1 , Roger C. Helgeson 2 , K. N. Houk 2 , Gonzalo Jiménez-Osés 2,3 ✉ and Donald Hilvert 1 ✉ New enzyme catalysts are usually engineered by repurposing the active sites of natural proteins. Here we show that design and directed evolution can be used to transform a non-natural, functionally naive zinc-binding protein into a highly active catalyst for an abiological hetero-Diels-Alder reaction. The artificial metalloenzyme achieves >10 4 turnovers per active site, exerts absolute control over reaction pathway and product stereochemistry, and displays a catalytic proficiency (1/K TS = 2.9 × 10 10 M −1 ) that exceeds all previously characterized Diels-Alderases. These properties capitalize on effective Lewis acid catalysis, a chemical strategy for accelerating Diels-Alder reactions common in the laboratory but so far unknown in nature. Extension of this approach to other metal ions and other de novo scaffolds may propel the design field in exciting new directions.
the zinc ion, would improve substrate binding and transition-state stabilization while retaining protein stability. These substitutions replace charged side chains that might hinder substrate binding with neutral groups; in addition, Trp68 was expected to shield the relatively open active site from bulk solvent. In fact, introducing these substitutions into MID1sc afforded the enzyme DA0, which exhibited low but detectable activity (approximately twofold over the background level at 5 µM enzyme) and a product profile similar to that observed in the absence of protein (Extended Data Fig. 2c,d).
To increase its catalytic efficacy, DA0 was optimized by laboratory evolution. Eight residues lining the putative substrate-binding site were subjected to cassette mutagenesis, and the resulting variants were screened spectroscopically for consumption of the azachalcone in the presence of 2 and excess zinc. On the basis of the recovered sequences, five of these residues were re-randomized using restricted amino acid alphabets and shuffled (Extended Data Fig. 3a; Supplementary Tables 2 and 3). Screening of the combinatorial library yielded the enzyme DA1, which contained the mutations H35C and I64G and provided detectable activity in cell lysates. This double mutant was subsequently subjected to two consecutive cycles of error-prone polymerase chain reaction, focused mutagenesis of hot-spot residues, and combinatorial shuffling of beneficial mutations to afford the highly active variant DA7 (  Tables 1-3).
DA7 contains a total of 12 mutations, distributed equally over the N-and C-terminal helix-turn-helix fragments of the starting scaffold (Fig. 2b). They include the replacement of His35, a non-coordinating histidine near the metal binding site, by a cysteine, and substitution of His39, one of the original zinc ligands, with a valine. These mutations imply a major change in Zn(ii) coordination. Aside from these substitutions, only the two amino acids introduced by computation, Leu32 and Trp68, and the first-round I64G mutation affect residues lining the putative binding pocket. All the other changes are more distant and presumably enhance activity by stabilizing the scaffold and/or subtly tuning active-site conformations during the steps involved in catalysis.
Catalytic properties of the evolved metalloenzyme. The kinetics of the reaction were determined by measuring the dependence of the initial velocity (v 0 ) on the concentration of 1 at several fixed concentrations of 2 in the presence of excess Zn(ii), which is absolutely required for activity ( Fig. 2c; Extended Data Fig. 4e). The steady-state parameters were obtained by globally fitting the data to a sequential binding mechanism. The catalytic rate constant k cat was 10 s −1 and the Michaelis constant (K M ) for 1 and 2 was 82 and 160 µM, respectively. Steady-state parameters were similarly obtained for DA0, DA1 and DA4, a fourth-round variant. Over the course of evolution, k cat increased more than 600-fold, while the  Table 4). As a result, the catalytic efficiency of the reaction, given by k cat /(K M,1 K M,2 ), improved by more than five orders of magnitude (Fig. 2d). These steady-state parameters compare favourably with those reported for a recently discovered Diels-Alderase, isolated from white mulberry, that produces an isoprenylated flavonoid natural product called chalcomoracin via an intermolecular [4 + 2] cycloaddition. 21 The second-order rate constant for the reaction of 1 and 2 in aqueous buffer containing zinc sulfate (k buffer = 0.027 M −1 s −1 ) provides a benchmark for the chemical prowess of DA7 (Extended Data Fig. 4a). Catalytic proficiency 22 , defined as the ratio [k cat / (K M,1 K M,2 )]/k buffer , measures the enzyme's ability to lower the activation barrier of the reaction and represents a lower limit on its affinity for the rate-limiting transition state. The value calculated for DA7, 2.9 × 10 10 M −1 , indicates that the evolved metalloenzyme is more proficient than all kinetically characterized Diels-Alderases (Fig. 2e), including other designer enzymes that promote bimolecular Diels-Alder reactions (Supplementary Table 5) and natural enzymes like SpnF 23,24 , AbyU 25 and IccD 26 , which perform unimolecular [4 + 2]-cycloadditions in the biosynthesis of insecticidal, antimicrobial and antifungal natural products (Supplementary Table 6). This extraordinary proficiency is due to both high effective molarity (EM = k cat /k buffer ) and the relatively tight binding reflected in the low K M,1 and K M,2 values.
In addition to being remarkably active, DA7 is highly stereoselective, generating exclusively the endo hetero-Diels-Alder product (4R,6R)-3,4-dihydro-2H-pyran with >99% enantioselectivity (Fig. 2f). This was not the case for variants from the early rounds of evolution. DA0, for example, preferentially forms the endo (4S,6S)-3,4-dihydro-2H-pyran isomer, but only with 36% enantioselectivity, and the product additionally contains 5% of the competing exo stereoisomers (Extended Data Fig. 2d). Although the transition state and the product of the DA7-catalysed reaction are structurally homologous, the enzyme catalyses >10 4 turnovers per active site (Extended Data Fig. 4f). As a result, preparative-scale reactions in aqueous buffer at room temperature afforded quantitative conversion of reactants to the optically pure product, providing a mild and practical alternative to standard chemical synthesis.
Structural and computational characterization. To understand the molecular changes that enabled the emergence of this proficient metalloenzyme, we crystallized DA7 and determined its structure to 1.5 Å resolution ( Fig. 3a; Supplementary Table 7). Like the zinc-binding peptide MID1sc from which it originated, the evolved enzyme is a helical bundle consisting of two helix-turn-helix motifs connected by a flexible linker. As anticipated, though, the H35C and H39V mutations altered the coordination sphere of the catalytic Zn(ii) ion (His39, His61, His65 → Cys35, His61, His65) and induced a major conformational change that reduced the crossover angle of the two helix-turn-helix fragments by >30° (Extended Data Fig. 5a-d). This structural rearrangement created a more enclosed hydrophobic pocket in the core of the bundle to accommodate both diene and dienophile. Because a thiolate is a better Zn(ii) ligand than a neutral histidine, the H35C mutation early in the evolution likely facilitated the emergence of this innovation.
A similar conformational change was observed when we evolved MIDsc into an efficient metalloesterase. In that case, though, rearrangement occurred midway along the evolutionary trajectory and did not require substitution of His35 with a cysteine. Although the final, evolved esterase has a similar overall topology and Zn(ii) coordination geometry to DA7, its second helix is longer and, because of a Pro mutation, kinked. Like the starting scaffold, the esterase has no detectable Diels-Alderase activity.
Although we were unable to obtain cocrystals of the protein with either substrates or product, docking calculations and molecular  (2) can react via competing Diels-Alder and hetero-Diels-Alder pathways to afford cyclohexenes 3 and/ or dihydropyrans 4, respectively. In addition to the reaction pathway, an optimal catalyst should control the endo/exo ratio and the absolute configuration of the reaction products.
dynamics (MD) simulations provided mechanistic insights into DA7 catalysis ( Fig. 3b-d). As in the solution reaction, the protein-bound zinc ion activates the azachalcone by chelation, facilitating attack of the 3-vinylindole to generate a zwitterionic intermediate, which rapidly cyclizes to the hetero-Diels-Alder product. Consistent with the experimentally observed enantioselectivity, only the transition state leading to the endo (4R,6R)-3,4-dihydro-2H-pyran product fits in the binding pocket with strong coordination to Zn(ii) and without steric clashes. A restrained MD simulation of the transition-state complex maintains all these interactions. By contrast, the enantiomeric transition state cannot fit into the binding site and still chelate Zn(ii). Restrained simulation of the latter reduces steric clashes at the expense of Zn(ii) coordination (Extended Data Fig. 6).
In the transition state leading to the observed product, the N-terminal helix-turn-helix fragment positions the azachalcone via numerous van der Waals contacts and a salt bridge between the guanidinium group of Arg28 and the carboxylate of the diene (Fig. 3b). The dienophile binds against the C-terminal helix-turnhelix fragment, placing its indole ring in a small hydrophobic pocket created by the I64G mutation and its vinyl moiety against the si face of 1 (Fig. 3c; Extended Data Fig. 6). The zwitterionic intermediate and flanking transition states are additionally stabilized by a hydrogen bond between the indole NH and the amide side chain of Gln80 ( Fig. 3d; Extended Data Fig. 7). This interaction is supported by a small hydrogen-bonding network consisting of Gln31 and Tyr84, a residue that appeared late in evolution (Fig. 3d). The efficiency of DA7 can thus be ascribed to a combination of Lewis acid catalysis, enthalpic stabilization of the transition state by the close-fitting binding pocket, and a strategically placed hydrogen-bonding interaction.
The hypothesis that the carboxylate substituent of 1 forms a salt bridge with Arg28 is supported by the fivefold lower specific activity observed for the unsubstituted azachalcone 5, which cannot make this interaction (Fig. 4). MD simulations of DA7 further suggested that slight displacement of Met87 at the base of the pyridine binding pocket would enable the binding of larger heterodienes. In accord with this prediction, the enzyme accepts isoquinoline derivative 6 as efficiently as the original azachalcone (Fig. 4). By contrast, chalcone 7, which can only bind Zn(ii) in a monodentate fashion, is a 100-fold poorer substrate than the isosteric azachalcone 1 (Fig. 4). Replacement of the indole NH in the dienophile with an oxygen is even more deleterious. 3-Vinylbenzofuran (8), an isosteric analogue of 2, is completely inactive (Fig. 4), highlighting the importance of the hydrogen bond between the dienophile and Gln80. The 23-fold reduction in efficiency observed when Gln80 was replaced with alanine reinforces this conclusion (Extended Data Fig. 4e).

Conclusions and prospects
Evolution has provided complete control over the reaction pathway and product stereochemistry of an abiological hetero-Diels-Alder reaction. Since we only screened for azachalcone disappearance, both properties emerged spontaneously as the activity increased. An important challenge for the future will therefore be to see whether or not equally effective catalysts can be engineered for the other possible isomers of 3 and 4. Judging from the product profile observed for DA0, it should be possible to enrich variants capable ]/k buffer , where K TS is the apparent transition state binding affinity, k cat is the catalytic rate constant and k buffer is the rate constant for the reaction in the absence of enzyme). DA7 and its precursors are compared with catalytic antibodies (black triangles) and variants of the computationally designed DA_20_00 Diels-Alderase (grey circles). The 1/K TS values observed for the natural enzymes IccD, AbyU and SpnF, which catalyse intramolecular Diels-Alder reactions, are shown as dashed lines. f, Chiral HPLC analysis of the hetero-Diels-Alder reaction products obtained in the DA7 (orange) and background (grey) reactions. Inset: X-ray structure of the endo (4R,6R)-3,4-dihydro-2H-pyran isomer produced by DA7.
of stabilizing the enantiomeric endo hetero-Diels-Alder transition state by monitoring changes in the product distribution directly by HPLC. Although reversing the inherent reactivity of this system to produce disfavoured exo products or normal-electron-demand Diels-Alder adducts will likely prove more demanding, computational methods 27 should help to design suitable libraries that favour such reaction channels.
In the design of any new enzyme, the choice of an appropriate starting scaffold is a fundamental consideration. Most previous attempts at creating such catalysts have focused on recycling pre-existing proteins from nature [27][28][29] . Our results show that evolutionarily naive metalloproteins like MID1sc are attractive alternatives. Despite its relatively small size (97 amino acids) and simple structure, this de novo helical bundle was able to successfully harness Lewis acid catalysis to produce the valuable dihydropyran moiety with high efficiency and selectivity. Other Lewis-acid-catalysed transformations 30 and other de novo scaffolds 31 may prove similarly amenable to this approach, providing a new avenue for designing and evolving biocatalysts of value for a variety of non-natural functions.

Methods
Materials, methods and data characterization for all biological, chemical and computational experiments are described in detail in the Supplementary Information.  Fig. 2 | Chemo-and stereoselectivity. a, HPLC chromatogram (at 350 nm) and crystal structure (Supplementary Table 11) of the racemic exo Diels-Alder product 3 (only one enantiomer is shown for clarity). b, HPLC chromatogram of the racemic hetero-Diels-Alder products 4 and crystal structure of the enzymatically synthesised (4 R,6 R)-endo hetero-Diels-Alder product. The hydrogen atoms on the central six-membered ring are shown for the X-ray crystal structures in a and b. c, HPLC analysis of the Diels-Alder (3.36 min) and hetero-Diels-Alder (3.06 min) product profile under different conditions. d, Chiral HPLC chromatograms of the hetero-Diels-Alder reaction products from reactions without protein (grey) and reactions catalysed by DA0 (pink) and DA7 (orange). Retention times vary due to different column temperatures (40 °C vs room temperature). Fig. 4 | Kinetic characterisation of buffer-catalysed reaction, intermediate DA variants, and DA7. a, Non-enzymatic background reaction between Diels-Alder substrates 1 and 2 in buffer. The rates were determined in 20 mM MOPS, pH 8, 3.5% DMSO, 10 µM Zn(II), 1 mg/mL BSA at 25 °C. Error bars denote s.d. b-d, Michaelis-Menten plots for the reaction between 1 and 2 catalysed by the parental scaffold DA0 and evolutionary intermediates DA1 and DA4. An analogous plot for DA7 is shown in Fig. 2c in the main text. For DA0 (b), error bars indicate the standard deviation between three biological replicates. e, Relative activities of DA7 and the variants DA7 W16S , DA7 C35H and DA7 Q80A , which were respectively prepared to aid crystallisation and probe mechanism. Mutation of the metal binding site (DA7 C35A/H61A/H65A ) or removal of zinc with EDTA leads to complete loss of activity.