A designed photoenzyme for enantioselective [2+2] cycloadditions

The ability to program new modes of catalysis into proteins would allow the development of enzyme families with functions beyond those found in nature. To this end, genetic code expansion methodology holds particular promise, as it allows the site-selective introduction of new functional elements into proteins as noncanonical amino acid side chains1–4. Here we exploit an expanded genetic code to develop a photoenzyme that operates by means of triplet energy transfer (EnT) catalysis, a versatile mode of reactivity in organic synthesis that is not accessible to biocatalysis at present5–12. Installation of a genetically encoded photosensitizer into the beta-propeller scaffold of DA_20_00 (ref. 13) converts a de novo Diels–Alderase into a photoenzyme for [2+2] cycloadditions (EnT1.0). Subsequent development and implementation of a platform for photoenzyme evolution afforded an efficient and enantioselective enzyme (EnT1.3, up to 99% enantiomeric excess (e.e.)) that can promote intramolecular and bimolecular cycloadditions, including transformations that have proved challenging to achieve selectively with small-molecule catalysts. EnT1.3 performs >300 turnovers and, in contrast to small-molecule photocatalysts, can operate effectively under aerobic conditions and at ambient temperatures. An X-ray crystal structure of an EnT1.3-product complex shows how multiple functional components work in synergy to promote efficient and selective photocatalysis. This study opens up a wealth of new excited-state chemistry in protein active sites and establishes the framework for developing a new generation of enantioselective photocatalysts. A genetically encoded triplet photosensitizer is used to develop an efficient photoenzyme that can promote enantioselective intramolecular and bimolecular [2+2] cycloadditions by means of triplet energy transfer.

Triplet EnT photocatalysis promotes a broad range of valuable chemical transformations, including cycloadditions, electrocyclic reactions, deracemizations, migrations and rearrangements, many of which are not accessible from the ground state [5][6][7][8][9][10][11][12] . Photons offer a convenient and tunable source of energy to selectively access reactive excited-state intermediates under mild reaction conditions. Perhaps the most prominent class of photochemical reactions are [2+2] cycloadditions, which construct cyclobutanes, oxetanes and azetidines [14][15][16] . In contrast to the analogous Diels-Alder [4+2] cycloadditions, [2+2] cycloadditions are thermally forbidden, owing to incompatible ground-state orbital symmetries (Fig. 1a). The key stages of EnT photocatalysis are shown in Fig. 1b. Light energy is used to promote a triplet sensitizer from the ground state (S 0 ) to a singlet excited state (S 1 ), which then undergoes intersystem crossing to a triplet state (T 1 ). Because relaxation from T 1 to S 0 is spin-forbidden, triplet intermediates are relatively long-lived compared with singlet excited states. Commonly used triplet photosensitizers include conjugated aromatic ketones such as benzophenone, xanthone and thioxanthone, which show high quantum efficiencies for population of the triplet state. The next step involves EnT from the triplet photosensitizer to the substrate in an overall spin-allowed process that returns the photosensitizer to the S 0 level and simultaneously promotes the substrate from S 0 to the reactive T 1 state, which can then undergo a variety of intramolecular and intermolecular processes 5,6 .
Enantioselective versions of photochemical reactions have been developed through dual-catalytic strategies that combine an achiral photosensitizer with a photochemically inactive chiral catalyst [17][18][19] (Fig. 1c) or by using chiral photosensitizers 15,20 (Fig. 1d). Although powerful, these approaches do not offer a general solution, and many desirable photochemical processes are not amenable to enantioselective catalysis. Furthermore, these reactions are oxygen-sensitive, often require high catalyst loadings and are restricted to a narrow range of substrates, with even small structural changes leading to substantial reductions in activity or selectivity 21 . In principle, enzyme active sites could offer more versatile chiral environments for mediating enantioselective photochemistry, in which several functional elements can be accurately positioned within a single pocket. Through directed evolution, these active sites could be sculpted to maximize productive interactions between the protein, substrate and photosensitizer Article to deliver efficient and selective photocatalysts (Fig. 1e). However, although a handful of natural photoenzymes and engineered biocatalysts have been reported [22][23][24][25][26][27][28][29][30][31][32][33][34] , enzymes that mediate enantioselective EnT processes are unknown at present.

Photoenzyme design and evolution
To develop a selective photoenzyme, we chose the computationally designed Diels-Alderase DA_20_00 as a host scaffold 13 . The DA_20_00 active site contains designed hydrogen-bonding residues (Tyr121 and Gln195) intended to promote the Diels-Alder reaction, which may prove useful in supporting catalysis of [2+2] cycloadditions. These residues are embedded within a large hydrophobic pocket suitable for accommodating a bulky aromatic photosensitizer (Fig. 2a). To unlock photocatalytic activity, an engineered Methanococcus jannaschii tyrosyl-tRNA synthetase/tyrosyl-tRNA (MjTyrRS/MjtRNA Tyr ) pair was used to incorporate 4-benzoylphenylalanine (BpA) at several positions around the DA_20_00 active site pocket 35 (Extended Data Fig. 1). Intramolecular [2+2] cycloaddition of quinolone 1 was selected as the target transformation 20 (Fig. 2b). Benzophenone (BP) is a suitable photosensitizer for this reaction and gives rise to racemic straight and crossed chain products (1a and 1b) as a 1.4:1 mixture of regioisomers (Fig. 2c). Placement of BpA at position 173 within the hydrophobic pocket of DA_20_00 provided the active photoenzyme EnT1.0, which is a more effective catalyst than BP and shows modest amounts of regioselectivity (2:1, 1a:1b) and enantioselectivity (46% e.e. for the major product (-)-1a). EnT1.0 activity is strictly dependent on light and the presence of the BpA173 photosensitizer (Fig. 2c).
To improve EnT1.0 activity, we initially mutated eight residues that lie in proximity to BpA173 to alanine (Extended Data Fig. 2). A M90A mutation gave rise to a substantial threefold increase in conversion to 1a, along with a modest improvement in enantioselectivity to 60% e.e. (EnT1.1; Fig. 3 and Extended Data Fig. 2). This improved variant facilitated detection of enzyme activity in clarified cell lysate and enabled the development of a directed-evolution workflow suitable for high-throughput engineering of triplet EnT photoenzymes. This evolutionary workflow relies on uniform irradiation of enzyme variants arrayed in microtitre plates, which was achieved using a commercial light-emitting diode (LED) array (see Methods). The observed coefficient of variance across an assay plate using purified EnT1.0 was shown to be less than 5% ( Supplementary Fig. 1), a value that is in line with established high-throughput screening methods.
The evolutionary strategy consisted of two rounds of saturation mutagenesis targeting residues in the active site and second coordination sphere, which were individually randomized using NNK degenerate codons. Individual variants arrayed in 96-well microtitre plates were irradiated for 30 mins at 365 nm using an LED array in the presence of substrate 1 and reactions analysed by high-throughput ultra-performance liquid chromatography (UPLC; Supplementary Figs. 2 and 3). The most active (about 1%) clones, identified on the basis of conversion to 1a, were then evaluated as purified proteins for improved activity and enantioselectivity ( Supplementary Fig. 4). Beneficial mutations identified during each round were subsequently combined by DNA shuffling. Following evaluation of approximately 3,500 library members, an EnT1.3 variant emerged with improved activity, stability and selectivity. EnT1.3 contains five mutations (EnT1.0 M90A, Q149D, P196R, K225E and A229S; Extended Data Fig. 3) and achieves a substantial tenfold improvement in reaction conversion of 1 to 1a compared with EnT1.0 following 30 min of irradiation (Fig. 3c,d). The enhanced performance of EnT1.3 arises from a combination of a fourfold increase in initial rate (Extended Data Fig. 4 c,d, Enantioselective EnT photocatalysis can be achieved using achiral photosensitizers in combination with chiral cocatalysts 18,19 or using chiral photosensitizers 20 . e, The approach to photoenzyme development described in this manuscript, involving selective installation of a genetically encoded photosensitizer into a protein scaffold and subsequent optimization of the modified protein by directed evolution. reduced susceptibility to photo-deactivation, which probably arises in EnT1.0 from an intramolecular photo-crosslinking process involving the benzophenone side chain (Extended Data Fig. 5). EnT1.3 also offers high enantiocontrol, generating (-)-1a with >99% e.e. Notably, EnT1.3 is more active at 4 °C than at room temperature (Extended Data Fig. 6), which may reflect an increased lifetime of the photosensitizer triplet state at lower temperatures 36 . To gain insight into the kinetic parameters of EnT1.3, we determined initial reaction velocities across a range of substrate concentrations ( Supplementary Fig. 7a). These measurements show that EnT1.3 has a high affinity for 1 (K M < 40 μM), which precludes accurate determination of K M owing to the detection limits of high-performance liquid chromatography (HPLC) analysis. A k cat of 1.3 ± 0.04 min −1 was determined under the assay conditions used throughout enzyme engineering, although this value is linearly dependent on light intensity and so can be increased at higher powers ( Supplementary Fig. 7b).
In contrast to small-molecule photocatalysts, which are highly sensitive to oxygen owing to triplet quenching, EnT1.3 is tolerant of aerobic buffers (Fig. 3d) and can achieve >300 turnovers under these conditions (Extended Data Fig. 7). Presumably the EnT1.3 active site excludes oxygen to minimize triplet quenching and favour productive EnT to substrate. In line with these observations, laser pulse experiments show that the triplet lifetime of the enzyme-bound BpA chromophore is very similar in aerobic and anaerobic buffers, which contrasts with the oxygen sensitivity of triplet benzophenone in solution (Supplementary Fig. 8). With only 1.5 mol% EnT1.3, near complete conversion of 1 to optically pure (-)-1a can be achieved within 2 h (Supplementary  Table 3), underscoring the efficiency of our engineered photoenzyme. Compared with small chiral photocatalysts, EnT1.3 achieves higher conversion, regioselectivity and enantioselectivity using lower catalyst loadings 20 . The enzyme also operates efficiently at ambient temperatures compared with the cryogenic temperatures needed with small chiral catalysts. To demonstrate practical utility, we performed a preparative-scale biotransformation to afford enantiopure (-)-1a, along with its minor regioisomer 1b, as the sole products in essentially quantitative (95%) yield ( Supplementary Fig. 9).

EnT1.3 is a versatile [2+2] cyclase
We next explored the photocatalytic activity of EnT1.3 towards [2+2] cycloadditions of allyloxy-quinolones, alken-1-yl-quinolones and allyloxy(methyl)-quinolones to generate optically enriched products (2a-13a; Fig. 4). High conversions and selectivities were achieved in most cases. With substrates 3, 8 and 12, introduction of a gem-dimethyl moiety led to a reduction in enantioselectivity, probably owing to a high degree of shape complementarity between the EnT1.3 active site and substrate 1 (vide infra). Consistent with this interpretation, the less highly evolved EnT1.2 variant gave improved activity and selectivity with 3 and 12, whereas EnT1.1 A229S Y37L was found to be a superior variant for conversion of substrate 8. The ability of EnT1.3 to generate 2a and 7a with high levels of stereocontrol is particularly noteworthy. Analogous reactions with small chiral photosensitizers proceed with poor selectivity, as cyclizations to form six-membered-ring analogues are relatively slow and competing dissociation of the excited substrate from the photosensitizer erodes e.e. 5,15,21 . These examples show that our enzymes can mediate selective transformations even when downstream excited-state chemistry is normally slow. Small-molecule chiral photosensitizers also rely on complementary dual hydrogen-bonding contacts between the substrate and catalyst to achieve enantioselective photochemistry 20 . To demonstrate that protein catalysts are not constrained in the same manner, we next investigated cycloaddition of an N-methyl derivative 13. EnT1.3 affords optically enriched 13a, along with its regioisomer 13b, with modest selectivity (2:1 regioisomeric ratio (r.r.), 36% e.e. for 13a), which could be further enhanced by first introducing a rational Y121F mutation (4:1 r.r., 78% e.e. for 13a), followed by a S271C substitution (8:1 r.r., 98% e.e. for 13a) identified through subsequent directed evolution. This further engineering demonstrates how protein active sites can be readily adapted to augment catalytic function and shows that our photoenzymes can mediate selective transformations of substrate classes that are beyond the reach of existing small-molecule systems. Finally, to expand synthetic utility, we explored the application of EnT1.3 towards bimolecular [2+2] cycloadditions of 2-quinolone. Despite being optimized for an intramolecular  reaction, EnT1.3 shows remarkable selectivity in bimolecular processes using methyl vinyl ketone or ethyl vinyl ketone as co-substrates, achieving 97% e.e. in both cases (Fig. 4). Taken together, these studies highlight EnT1.3 and its variants as a versatile and powerful platform for the enantioselective construction of cyclobutane rings.

Structural basis for efficient catalysis
To gain insights into the EnT1.3 catalytic mechanism, a crystal structure of a C-terminally truncated analogue (EnT1.3ΔC 310-314 ; see Supplementary Information) complexed with product (-)-1a was solved (1.7 Å; Fig. 5 and Supplementary Table 5). This truncation has negligible effect on catalytic activity or selectivity ( Supplementary Fig. 5) and was introduced to circumvent the C-terminus of a neighbouring chain from occluding the EnT1.3 active site in crystallo. The ligand sits in a snug active-site pocket with its aromatic ring sandwiched between His287 and the benzophenone side chain (3.6 Å and 3.8 Å, respectively; Fig. 5 and Supplementary Fig. 6). This pose presumably allows for efficient triplet EnT from the excited-state photosensitizer to the parent substrate. The well-packed active site led us to consider whether a non-covalent benzophenone sensitizer could support enantioselective catalysis in the EnT1.3 scaffold. However, reactions with EnT1.3 BpA173Ala in the presence of exogenous benzophenone led to low conversion and selectivity (Supplementary Table 3), highlighting the importance of installing the triplet photosensitizer on the genetic level. The crystal structure suggests that Trp244 may play a role in controlling reaction selectivity in favour of the formation of (-)-1a by shaping the active-site cavity to prevent addition of the exocyclic alkene to the enantiotopic face of the excited quinolone. Indeed, a W244A mutation leads to a substantial reduction in activity and racemic product formation (Extended Data Fig. 9). Ligand binding is further supported by complementary hydrogen-bonding interactions with Tyr121, which serves as a hydrogen-bond acceptor to the quinolone N-H, and Gln195, which acts as a hydrogen-bond donor to the quinolone carbonyl (Fig. 5). Notably, Tyr121 and Gln195 were designed to mediate Diels-Alder catalysis in DA_20_00 through hydrogen-bonding interactions, albeit with opposite donor/acceptor relationships to those observed in EnT1.3. With substrate 1, mutation of Tyr121 to Phe leads to a threefold reduction in activity, along with a modest decrease in enantioselectivity (67% e.e.; Extended Data Fig. 9). By contrast, this mutation improves both activity and selectivity towards N-methylated substrate 13 (Fig. 4), in which the N-methyl substituent probably occupies the position vacated by removal of the Tyr121 phenolic oxygen. The hydrogen bond from Gln195 to the quinolone carbonyl is particularly short (2.6 Å) and probably serves to lower the triplet energy of the substrate 37 . Mutation of Gln195 to Ala leads to a considerable reduction in activity and selectivity, underscoring its importance to EnT1.3 catalysis. Gln195 is anchored in position by a further hydrogen bond with Arg196, which emerged during evolution.
Notably, in apo-EnT1.3, Gln195 and Arg196 lie in markedly different orientations, with Arg196 instead interacting with Asp149 and Glu225, suggesting that substrate binding may induce formation of a productive Arg196-Gln195-substrate hydrogen-bonding network (Extended Data Fig. 10). This combined structural analysis offers a first glimpse into active-site features governing efficient EnT catalysis and provides an important blueprint for the future design of photoenzymes with new and augmented functions.

Conclusion
In summary, our study provides a powerful demonstration of how an expanded genetic code can be used to embed entirely new modes of catalysis into proteins. This technology has allowed the development of a proficient photoenzyme that operates through a triplet EnT mechanism, a versatile reaction manifold in organic synthesis that was previously inaccessible to biocatalysis. Although EnT1.3 was tailored to promote intramolecular [2+2] cycloadditions, we anticipate that our approach can be readily adapted to alternative chemistries that are enabled by triplet EnT. In this regard, it is encouraging that EnT1.3 is able to mediate bimolecular processes and can be readily adapted to operate on substrates that are beyond the scope of small chiral photocatalysts. EnT1.3 also promotes bimolecular cycloadditions of 2-quinolone with methyl vinyl ketone or ethyl vinyl ketone to produce 14 and 15, respectively. Reaction conversions are the mean of biotransformations performed in triplicate. The absolute stereochemistry of products 2a-13a were assigned by analogy to the product (-)-1a, formed by EnT1.3. The absolute stereochemistry of 14 was assigned by comparison of HPLC retention times following a literature procedure 39 . The absolute stereochemistry of 15 was assigned by analogy to the product 14. Reaction conditions for the synthesis of 1a-15 are presented in Supplementary Table 3.

Article
The catalytic performance and structural sophistication of EnT1.3 is particularly notable, given that <1,000 protein variants were evaluated across the evolutionary trajectory. Presumably, more efficient photocatalysts will be generated through a deeper exploration of protein sequence. Likewise, combinations of high-throughput experimentation and in silico design will deliver active sites with new geometries and arrangements of functional groups suitable for mediating selective photocatalysis 38 . Thus with the mechanistic framework for embedding EnT catalysis into proteins now established, we are optimistic about the prospect of developing photoenzymes for a broad array of valuable photochemical processes.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-022-05335-3.
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.

Construction of pET-29b_EnT1.0 and variants
The original DA_20_00 design 13 was subcloned using NdeI and XhoI restriction sites into a pET-29b(+) vector containing a C-terminal His 6 -tag to yield pET-29b(+)_DA_20_00. The Ala173BpA mutation was introduced by replacing the Ala173 codon with a TAG stop codon using QuikChange site-directed mutagenesis (Agilent) to yield pET-29b(+)_ EnT1.0. Point mutants of EnT1.0 and variants were constructed using the same procedure.

Mass spectrometry
Purified protein samples were desalted on 10,000 MWCO Vivaspin centrifugal concentrators (Sartorius) using 0.1% acetic acid and diluted to a final concentration of 0.4 mg ml −1 . Mass spectrometry was performed on a 1200 Series Agilent LC in conjunction with a Agilent 6510 QTOF. A 5-μl sample injection was performed, followed by a 1-min 5% acetonitrile (with 0.1% formic acid) isocratic wash. Protein was eluted over 1 min using 95% acetonitrile with 5% water. The resulting multiply charged spectrum was deconvoluted using Agilent MassHunter Software. Protein mass spectrometry results are shown in Supplementary Information, Table S7.

Library construction
Rounds 1, 2 and 3: saturation mutagenesis. Positions were individually randomized using degenerate NNK codons. DNA libraries were constructed by overlap extension polymerase chain reaction (PCR). Primers for library generation are given in Supplementary Information, Table S11. Assembled genes and pET-29b(+) vector were digested using NdeI and XhoI endonucleases, gel-purified and subsequently ligated using T4 DNA ligase in a 5:1 ratio, respectively. Ligations were transformed into E. coli 5α cells, the resulting colonies were pooled and plasmid DNA was extracted using a Miniprep Kit (QIAGEN) to yield plasmid DNA for each library. Sequencing was performed by Source BioScience (Nottingham).

Shuffling by overlap extension PCR
After rounds 1 and 2 of evolution, beneficial mutations were combined by DNA shuffling of fragments generated by overlap extension PCR. Primers were designed that encoded either the parent amino acid or the identified mutation. These primers were used to generate short fragments that were gel-purified and mixed for assembly of the full-length gene by overlap extension PCR. Final full-length genes contain all possible combinations of mutations at specified positions. Genes were cloned as described above.

Round 3: improvement of EnT1.3 Y121F activity towards substrate 13.
A 75-μl volume of clarified lysate was transferred to 96-well polypropylene microtitre plates containing 25 μl of 1.2 mM substrate 13 in PBS buffer pH 7.4 with 20% DMSO as a cosolvent. Samples were irradiated at 365 nm in a UV curing LED oven (equipped with 365-nm and 395-nm LEDs, UV intensity 750 mW cm −2 , LED module size 100 × 100, NovaChem), with pulsing irradiation (10 s on, 10 s off) at 4 °C for a total irradiation time of 30 min at 100% intensity. Reactions were quenched with the addition of 100 μl of acetonitrile, the plates heat-sealed and incubated for a further 1 h at 30 °C, 80% humidity and 900 r.p.m. Precipitated proteins were removed by centrifugation at 2,900 × g for 10 min. A 100-μl volume of the clarified reaction mixture was transferred to 96-well polypropylene microtitre plates and heat-sealed with pierceable foil. Reactions were evaluated by UPLC analysis. Following each round, the most active variants were rescreened as purified proteins and evaluated by both UPLC and supercritical fluid chromatography (SFC) analysis. Proteins were expressed and purified as described above with the exception that starter cultures were inoculated from glycerol stocks prepared from the original library plate overnight cultures.

General procedure for analytical-scale biotransformations
Ninety-six-well microtitre round-bottom polypropylene plates were used for biotransformations, using HD Clear high-performance tape (Duck), 3-inch × 54.6-yard roll, to seal the samples. Biotransformations were performed at 4 °C using 0.4 mM substrate 1 and the relevant biocatalyst (20 μM) in PBS buffer pH 7.4 with 5% DMSO as a cosolvent. Samples were irradiated at 365 nm or 395 nm in a UV curing LED oven (Nova-Chem), 23 cm below the LED array (unless stated otherwise). Instrument settings: 100% intensity, 750 mW cm −2 , 10 s on/off pulse. Conditions for substrate scope characterization are detailed in Supplementary Information, Table S3. Reactions were evaluated by UPLC analysis.

Anaerobic biotransformations
For anaerobic biotransformations, samples of EnT1.3 and substrate were incubated in a glovebox overnight on ice to ensure complete removal of oxygen. Reactions were made up in the glovebox in glass vials (final volume of 500 μl) using 2.5 mol% (10 μM) EnT1.3 in PBS (pH 7.4) with 5% DMSO as a cosolvent. Twenty-five-microlitre samples were taken at 10, 20, 30, 60 and 90 min and quenched with two volumes of MeCN. Reactions were evaluated by UPLC analysis.

EnT1.3 temperature profile
Biotransformations were performed at 4 °C and room temperature in glass vials using 2.25 μM of EnT1.3 and 300 μM of substrate 1 in PBS buffer with 5% DMSO as a cosolvent (final volume of 500 μl). For reactions at 4 °C, all reaction components were incubated in a cold room maintained at 4 °C for 30 min before running reactions. For both temperatures, reactions were run using 100% intensity irradiation at 365 nm with 10 s on/10 s off intervals and 25-μl samples were taken at 10, 20, 30, 40, 50, 60, 90 and 120 min. Reactions were evaluated by UPLC analysis.

EnT1.3 cosolvent tolerance
To investigate cosolvent tolerance, analytical-scale biotransformations were performed using 1 (400 μM) and EnT1.3 (7.5 μM) in 100 μl PBS buffer (pH 7.4) with the stated concentration of MeCN or DMSO, for a total irradiation time of 30 min at 100% intensity at 365 nm with 10 s on/10 s off intervals at 4 °C ( Supplementary Information, Table S6). Reactions were run in a 96-well plate and evaluated by UPLC analysis.

Photodamage biotransformations
To investigate the effects of photodamage, 1-ml aliquots of 30 μM enzyme in PBS (pH 7.4) were pre-irradiated for 90 min (10 s on/off pulse at 365 nm) in glass vials at 4 °C before performing reactions. Control aliquots of enzyme were incubated in the dark at 4 °C. Light-exposed and non-light-exposed protein samples (DA_20_00, EnT1.0 and EnT1.

EnT1.3 light intensity rate profile
To investigate the effect of light intensity on the rate of reaction with EnT1.3, biotransformations were performed in 2-ml MS glass vials using 1 μM EnT1.3 and 400 μM 1 in PBS buffer (pH 7.4) with 5% DMSO as a cosolvent (500 μl total reaction volume). Reactions were positioned at a 3-cm distance from the LED array and irradiated as described at varying LED intensities and time points taken at 2 min, 4 min, 6 min, 8 min and 10 min for 30% and 40%, and 4 min, 8 min, 12 min, 16 min and 20 min for 10% and 20% and analysed by UPLC analysis. The reaction rate at each light intensity was calculated using the slope of the time course (as an average of triplicate measurements), which remained linear in each case.

Preparative-scale biotransformation
Substrate 1 (12 mg) was dissolved to a concentration of 400 μM in PBS buffer (pH 7.4) with 5% DMSO as a cosolvent (140 ml reaction volume) with 10 μM EnT1.3 (2.5 mol%). The solution was irradiated as described at 365 nm for a total reaction time of 2 h in a Pyrex dish (22 cm diameter, solution path length 0.37 cm). Once full conversion was reached (as monitored by UPLC), the solution was transferred to a separatory funnel and extracted with 3 × 20 ml ethyl acetate and the combined organic layers were washed with 3 × 20 ml brine, dried over MgSO 4 , filtered and concentrated in vacuo to afford optically pure 1a along with its minor regioisomer 1b (9:1 r.r., 11.4 mg, 95%), which required no further purification.
Reactions were run in a 96-well plate for a total irradiation time of 4 h at 100% intensity at 395 nm with 10 s on/10 s off intervals at 4 °C, quenched with one volume of MeCN and evaluated by UPLC and chiral HPLC analysis.

Biotransformations with exogenous benzophenone using EnT1.3 BpA173A
Biotransformations were performed using 10 μM EnT1.3 BpA173A, 10 μM benzophenone (taken from a 1 mM stock in MeCN) and 400 μM of substrate 1 in 100 μl PBS buffer (pH 7.4) with 5% DMSO as a cosolvent. Reactions were run in a 96-well plate for a total irradiation time of 60 min at 100% intensity at 365 nm with 10 s on/10 s off intervals at 4 °C, quenched with one volume of MeCN and evaluated by UPLC and chiral SFC analysis.

Chromatographic analysis
For UPLC analysis, reactions were quenched at the stated time points with the addition of one volume of acetonitrile. Samples were shaken at 900 r.p.m for 1 h and precipitated proteins were removed by centrifugation (2,900 × g for 10 min). For SFC and chiral HPLC analysis, the substrates and products were transferred into 1.5-ml microcentrifuge tubes and extracted with three volumes of ethyl acetate. Precipitated proteins were removed by centrifugation (14,000 × g for 15 min), the organic phase was separated and directly injected onto the SFC.
UPLC analysis was performed on a 1290 Infinity II LC system (Agilent) with a Kinetex 5 μm XB-C18 100 Å LC Column, 50 × 2.1 mm (Phenomenex). Peaks were assigned by comparison with chemically synthesized standards and the peak areas were integrated using Agilent's OpenLab software. The separation methods for all substrate(s)/product(s) and extinction coefficients used to calculate the conversion are reported in Supplementary Information, Table S8.
Chiral analysis was performed using either a SFC 1290 Infinity II system (Agilent) or a HPLC 1260 system (Agilent). Enantiomers of all reaction products 1a-12a were separated using a Daicel 87S82 CHIRALPAK IG-3 SFC column, 3 mm × 50 mm × 3 μm. For all adducts (1a-13a), the major stereoisomer formed in the biotransformations was assigned on the basis of an analogy to EnT1.3-derived (-)-1a. Peaks were assigned by comparison with chemically synthesized standards and peak areas were integrated using Agilent's OpenLab software. Separation methods for all substrate(s)/product(s) enantiomers used to determine e.e. are reported in Supplementary Table S9 for SFC methods and Supplementary Table S10 for chiral HPLC methods.

Laser pulse experiments
Laser photoexcitation experiments were carried out at 4 °C using an Edinburgh Instruments LP980 Transient Absorption Spectrometer. Samples contained 50 μM benzophenone, Ent1.0 or Ent1.3 in PBS buffer (pH 7.4) and O 2 was removed where necessary by incubation in an anaerobic glovebox (Belle Technology) for 4 h. Triplet formation was initiated by excitation at 355 nm (about 50 mJ), using the third harmonic of a Q-switched Nd-YAG laser (NT342B, EKSPLA) in a cuvette of 1 cm path length. Time-dependent absorbance difference spectra were recorded with an intensified charge-coupled device detector using a gate width of 100 ns and 50 averages. Kinetic transients were recorded using 50 averages at the specified wavelengths with the detection system (comprising probe light, sample, monochromator and photomultiplier tube detector) at right angles to the incident laser beam. Lifetimes were measured by fitting to a double exponential function using Origin Pro 9.1 software.
General procedure 2 (ref. 9 ): synthesis of 4-((allyloxy)methyl) quinolones 10-12. A solution of the corresponding allylic alcohol (2.73 mmol, 1.3 eq.) in 3.5 ml of dry tetrahydrofuran (THF) was added to a suspension of sodium hydride (60% in paraffin oil, 185 mg, 4.62 mmol, 2.2 eq.) in 3.5 ml of dry THF at 0 °C. The reaction mixture was stirred at 0 °C for 2 h and then heated to reflux for a further 2 h. After cooling to room temperature, the solution was added dropwise to a stirring suspension of 4-bromomethyl-2(1H)-quinolinone (500 mg, 2.10 mmol, 1 eq.) in 3 ml dry THF at 0 °C. Subsequently, the reaction mixture was stirred for 6 h at room temperature. The reaction was quenched by addition of aqueous 1M HCl (30 ml) and then extracted with ethyl acetate (3 × 100 ml). The combined organic layers were washed with brine (2 × 50 ml), dried over MgSO 4 , filtered and the solvent removed in vacuo to give the crude product.

General procedure 3 (ref. 15 ): synthesis of 4-(pent-4-en-1-yl) quinolones 6-9
2-Hydroxy-4-methylquinoline (500 mg, 3.14 mmol, 1 eq.) was suspended in 17.5 ml dry THF, cooled to 0 °C and treated dropwise with n-butyl lithium (2.5M in hexane, 2.76 ml, 6.9 mmol, 2.2 eq.) under a nitrogen atmosphere. The dark-red solution was stirred at room temperature for 2 h, cooled to −78 °C and treated with tetrabutylammonium iodide (872 mg, 2.36 mmol, 0.75 eq.) and the corresponding alkenyl bromide (6.34 mmol, 2.02 eq.) was added subsequently. The yellow solution was stirred at room temperature under a nitrogen atmosphere for a further 3 h, cooled to 0 °C and quenched by slow addition of 1M HCl (30 ml). The resulting solution was extracted with dichloromethane (3 × 100 ml). The combined organic layers were washed with aqueous NaHCO 3 (60%, 50 ml) and brine (2 × 50 ml), dried over MgSO 4 , filtered and the solvent removed in vacuo to give the crude product. 1-13(a/b). The corresponding quinolones 1-13 (100 mg) were dissolved in acetonitrile to a concentration of 10 mM and degassed by bubbling nitrogen through the solution for 30 min. The solution was irradiated at λ = 300 nm at room temperature until full conversion was achieved (as monitored by thin-layer chromatography). The solvent was removed in vacuo to give the crude product.

General procedure 4: synthesis of intramolecular [2+2] cycloaddition product standards
General procedure 5: synthesis of bimolecular [2+2] cycloaddition product standards 14 and 15. 2-Quinolone (100 mg) was dissolved in acetonitrile to a concentration of 10 mM with 50 mol% thioxanthone and degassed as described. Once dissolved and degassed, the corresponding vinyl ketone (10 eq.) was added and the solution was irradiated at λ = 395 nm at 4 °C until full conversion was achieved (as monitored by thin-layer chromatography). The solvent was removed in vacuo to give the crude product.        13