Catalytic mechanism for Renilla-type luciferases

The widely used coelenterazine-powered Renilla luciferase was discovered over 40 years ago, but the oxidative mechanism by which it generates blue photons remains unclear. Here we decipher Renilla-type catalysis through crystallographic, spectroscopic and computational experiments. Structures of ancestral and extant luciferases complexed with the substrate-like analogue azacoelenterazine or a reaction product were obtained, providing molecular snapshots of coelenterazine-to-coelenteramide oxidation. Bound coelenterazine adopts a Y-shaped conformation, enabling the deprotonated imidazopyrazinone component to attack O2 via a radical charge-transfer mechanism. A high emission intensity is secured by an aspartate from a conserved proton-relay system, which protonates the excited coelenteramide product. Another aspartate on the rim of the catalytic pocket fine-tunes the electronic state of coelenteramide and promotes the formation of the blue light-emitting phenolate anion. The results obtained also reveal structural features distinguishing flash-type from glow-type bioluminescence, providing insights that will guide the engineering of next-generation luciferase‒luciferin pairs for ultrasensitive optical bioassays. Renilla luciferase is a popular bioluminescent enzyme, but the molecular details of its mechanism of action on luciferins such as coelenterazine remained elusive. Now, protein crystal structures and biochemical analyses provide an atomistic description of its catalytic mechanism.

The widely used coelenterazine-powered Renilla luciferase was discovered over 40 years ago, but the oxidative mechanism by which it generates blue photons remains unclear. Here we decipher Renilla-type catalysis through crystallographic, spectroscopic and computational experiments. Structures of ancestral and extant luciferases complexed with the substrate-like analogue azacoelenterazine or a reaction product were obtained, providing molecular snapshots of coelenterazine-to-coelenteramide oxidation. Bound coelenterazine adopts a Y-shaped conformation, enabling the deprotonated imidazopyrazinone component to attack O 2 via a radical charge-transfer mechanism. A high emission intensity is secured by an aspartate from a conserved proton-relay system, which protonates the excited coelenteramide product. Another aspartate on the rim of the catalytic pocket fine-tunes the electronic state of coelenteramide and promotes the formation of the blue light-emitting phenolate anion. The results obtained also reveal structural features distinguishing flash-type from glow-type bioluminescence, providing insights that will guide the engineering of next-generation luciferase-luciferin pairs for ultrasensitive optical bioassays.
Bioluminescence is a fascinating phenomenon involving the emission of light by a living creature. There is enormous interest in harnessing bioluminescent systems to design ultrasensitive optical bioassays and enable a circular bio-economy [1][2][3][4] . Bioluminescent organisms generate light via the oxidation of a substrate (a luciferin), which is catalysed by a class of enzymes called luciferases 5 .
One of the most popular bioluminescent reporters is a luciferase isolated from the sea pansy Renilla reniformis, a soft coral that displays bioluminescence upon mechanical stimulus 6 . Renilla luciferase, henceforth referred to as RLuc, is a 36 kDa protein that is active as a monomer 7,8 . RLuc displays remarkable sequence and structural similarity to a family of haloalkane dehalogenases (HLDs), indicating a common evolutionary history [9][10][11] . Unlike HLDs, which catalyse hydrolytic cleavage of carbon-halogen bonds in halogenated hydrocarbons (EC 3.8.1.5) 12 , the RLuc luciferase is a cofactor-independent monooxygenase (EC 1.13.12.5) 7,8 that catalyses the conversion of coelenterazine (CTZ) to coelenteramide (CEI). The blue light emission of Renilla luciferase has fascinated scientists for decades. However, despite intensive efforts 11,[13][14][15] , a detailed understanding of its catalytic mechanism at the molecular level remains elusive.
Here we present an atomistic description of the catalytic mechanism of Renilla-type luciferases that was inferred from cocrystal structures of AncFT and RLuc8 luciferases complexed with either a synthesized non-oxidizable CTZ analogue (azacoelenterazine, azaCTZ) or a catalytic product. These complexes reveal key structural features of the catalytic machinery responsible for CTZ conversion. Moreover, we delineate the biophysical factors responsible for flash-type and glow-type bioluminescence, and show how the electronic state of the CEI ion is tuned to favour blue light emission. The proposed mechanism is supported by data from spectroscopic, mutagenesis and computational experiments. Collectively, our findings provide mechanistic understanding of Renilla-type catalysis and will facilitate the engineering of next-generation luciferase-luciferin bioluminescent pairs.
Having synthesized azaCTZ, we studied its in vitro effects on the bioluminescent reactions of AncFT and RLuc8 by performing steady-state kinetics experiments. These experiments confirmed that azaCTZ is a non-oxidizable CTZ analogue that mimics the binding of CTZ. A detailed description of the kinetic analysis is provided in Supplementary Notes 1 and 2 while the analysis of luminescence kinetics validity and relationship to CTZ oxidation is summarized in Supplementary Figs. 4 and 5. Numerical integration of the luminescence progress curves obtained for the two enzymes in the presence of mixtures of CTZ and azaCTZ provided precise estimates of their steady-state kinetic parameters that agreed with previously reported values (Fig. 1c,d) 13 . Global fittings based on kinetic data recorded in the absence and presence of different concentrations of azaCTZ were systematically analysed to clarify the mechanisms by which the inhibitor interacted with RLuc8 and AncFT. In the case of AncFT, a model in which the inhibitor competes with the substrate for a single binding site gave the best agreement with the kinetic data ( Fig. 1c and Supplementary Figs. [6][7][8]. Additionally, the dissociation constant for the enzyme-inhibitor complex (K I = 0.092 ± 0.002 μM) calculated using the competitive model is similar to the Michaelis constant of AncFT, further supporting the assumption that azaCTZ binds to the active site of this enzyme in a very similar manner to the native substrate. AzaCTZ is thus a pure competitive inhibitor of bioluminescent CTZ degradation catalysed by AncFT ( Supplementary Fig. 2).
A more complex mixed-type inhibition model gave the best fit to the kinetic data acquired for the reaction of RLuc8 ( Fig. 1d and Supplementary Figs. 8-10). The full time-course kinetics of the substrate-to-product conversion provided additional estimates of the equilibrium dissociation constants for the enzyme-product complex, making it possible to explain the more complex inhibitory mechanism in this case. The proposed model includes competitive binding of azaCTZ to the active site of RLuc8 (in keeping with the model selected for AncFT) but also suggests that the inhibitor can bind to an enzyme-product complex. The fact that RLuc8 can simultaneously bind to the substrate and the inhibitor but does so with affinities CTZ, or 2-(p-hydroxybenzyl)-6-(p-hydroxyphenyl)-8-benzylimidazo(1,2-a)pyrazine-3-(7H)-one, is the most common marine luciferin. It features an imidazo(1,2-a)pyrazine ring system that emits a photon after undergoing an O 2 -mediated oxidation whose other products are CEI and CO 2 (refs. [16][17][18][19]. CTZ can spontaneously (non-enzymatically) emit a photon; this process is known as chemiluminescence and is favoured in aprotic solvents such as dimethyl sulfoxide (DMSO) 20,21 . As shown in Fig. 1a, mechanisms for CTZ oxidation have been suggested by McCapra 16 and Goto 17 , and more recently updated by other authors 22,23 . It was originally assumed that the reaction starts via deprotonation of the N7 nitrogen to form a CTZ anion that reacts directly with O 2 at the C2 carbon to form a CTZ peroxide ion. However, no known photoproteins or luciferases have an amino acid in the vicinity of the N7 nitrogen of the bound CTZ that could potentially mediate this initial deprotonation. Griffiths and coworkers therefore proposed that CTZ may bind to a protein as its O10H tautomer, or as an already deprotonated species (Fig. 1a) 23 . Following the oxygen addition, the resulting nucleophilic peroxide moiety performs an intramolecular attack on the C3 carbonyl to form a cyclic dioxetanone intermediate. Subsequent ring decomposition via decarboxylation leads to loss of CO 2 and the formation of an aminopyrazine product (the CEI ion) in a singlet excited state. While this is the light-emitting species in chemiluminescence, bioluminescent systems probably protonate the N1amide ion to form a light-emitting neutral CEI species. Moreover, some photoproteins and also possibly some luciferases remove another proton from the 6-(p-hydroxyphenyl) substituent to form a light-emitting 6-(p-hydroxyphenolate) ion. Finally, it has been demonstrated that some bioluminescent systems do not produce CEI as the major product of CTZ oxidation but instead convert CTZ into coelenteramine (CNM) and (p-hydroxyphenyl)acetic acid 22 . It is not currently clear whether CTZ oxidation proceeds via a single electron transfer process involving radical intermediates, as seen in the oxidation of firefly luciferin by the luciferase of Photinus pyralis 24 .
A key barrier to a deeper understanding of Renilla-type catalysis is the limited availability of structural data on substrate-and product-bound enzyme complexes. Crystallization of native RLuc is challenging, so mutagenesis experiments were performed to obtain mutants more amenable to crystallization 25,26 . As a result, a stabilized eight-point mutant designated RLuc8 (ref. 26 ) was crystallized by Loening and coworkers 14 . A partial density for CEI bound outside the active site was observed in one of their structures, although this structure is unlikely to be biologically relevant. In addition, we recently cocrystallized multiple RLuc8 mutants in the presence of excess CTZ. These efforts yielded a CEI-bound RLuc8 structure, providing the biologically relevant insights into a postcatalytic complex 13 . Unfortunately, the lack of a stable non-oxidizable CTZ analogue and the intrinsic flexibility of extant RLuc mutants seriously complicate the acquisition of high-resolution structures of catalytically favoured Michaelis enzyme-substrate complexes.
High protein stability is known to facilitate crystallization [27][28][29] , and it has been demonstrated that the structures of ancestral enzymes can be used as molecular scaffolds to unravel the structures and catalytic mechanisms of extant enzymes that are difficult to crystallize [30][31][32][33][34][35] . We previously used ancestral sequence reconstruction to obtain a stable ancestral enzyme Anc HLD-RLuc that was reconstructed from the catalytically distinct but structurally related HLDs and Renilla luciferase 11 . This ancestor enzyme turned out to have dual functions, with dehalogenase activity comparable to that of contemporary HLDs as well as promiscuous luciferase activity substantially lower than that of the stabilized RLuc8. We recently subjected Anc HLD-RLuc to insertiondeletion backbone mutagenesis and thereby discovered key structural elements responsible for acquisition of luciferase activity. In particular, transplanting a loop-helix fragment from the extant RLuc luciferase into the Anc HLD-RLuc ancestor yielded a fragment-transplanted AncFT enzyme with 7,000-fold higher catalytic efficiency and 100-fold longer Article https://doi.org/10.1038/s41929-022-00895-z   The reaction is initiated by the dissociation of a proton from the O10 oxygen to produce anionic O10-CTZ, whose C2 carbon then attacks O 2 to form the 2-peroxoCTZ ion. The distal oxygen of 2-peroxoCTZ performs an intramolecular nucleophilic attack on the C3-carbon to form a cyclic dioxetanone intermediate whose subsequent decarboxylative decomposition releases CO 2 and forms an anionic excited-state aminopyrazine product: the CEI ion. This unstable anionic species then decomposes with the emission of light, resulting in chemiluminescence. Bioluminescent systems protonate the CEI anion at the amide position to form the corresponding neutral species. Light can be emitted from this neutral species, but more often another proton is removed to form the 6-(p-hydroxyphenolate) ion, which emits blue light with an emission maximum at roughly 480 nm. Some bioluminescent systems do not produce CEI as the major product of CTZ oxidation but instead convert CTZ into CNM Article https://doi.org/10.1038/s41929-022-00895-z that are more than an order of magnitude lower than those of AncFT (K m = 1.66 μM and K I1 = 16.1 μM for RLuc8, as compared to K m = 0.042 μM and K I1 = 0.092 μM for AncFT) can be explained by the greater conformational flexibility of RLuc8. These kinetic data agree with our anaerobic equilibrium binding experiments, which also showed that the affinity of AncFT for CTZ and CEI greatly exceeded that of RLuc8 ( Supplementary Fig. 11). Collectively, these results indicate that azaCTZ could be a valuable tool for probing precatalytic enzyme-substrate complexes through X-ray crystallography.

AzaCTZ-and CEI-bound AncFT structures
To obtain structural insights into the RLuc catalytic mechanism, we attempted cocrystallization of AncFT with excesses of azaCTZ or its native ligand CTZ. It was expected that the latter ligand would be converted by the enzyme into the catalytic product CEI. After screening cocrystallization conditions and optimizing hits, well-diffracting crystals were obtained. The protein structures were then solved by molecular replacement and the initial models obtained were further refined through several cycles of manual building and automatic refinement, yielding structural models with low deviations from ideal geometry (Supplementary Table 3). These AncFT structures show a canonical αβα-sandwich architecture and closely resemble the previously reported structure of apo AncFT 13 , with root-mean-square deviation (r.m.s.d.) values on Cα atoms ranging from 0.1 to 0.3 Å (Supplementary Fig. 12).
Inspections of the electron density maps unambiguously revealed azaCTZ or CEI molecules bound in the active-site pocket of the corresponding cocrystal structures ( Fig. 2 and Supplementary Fig. 13). Both azaCTZ and CEI adopt a Y-shaped conformation when bound to the enzyme pocket. The R 1 2-(p-hydroxybenzyl) substituent is deeply buried in the active-site cavity, where it is anchored in the slot p2 tunnel via multiple polar and nonpolar interactions. In addition, its terminal hydroxyl moiety forms a hydrogen bond with the backbone carbonyl of S143 (roughly 2.8 Å). The triazolopyrazine core of azaCTZ is positioned in close proximity to the conserved catalytic machinery, which consists of a catalytic pentad that was first characterized in the structurally related HLDs 12,37 . The triazolopyrazine core of azaCTZ forms direct contacts with the carboxyl side chains of D118 (roughly 3.3 Å) and W119 (roughly 3.3 Å). The former residue is a nucleophilic aspartate, while the latter tryptophan functions as a halide-stabilizing residue in the HLD reaction 12,37 . The remaining two substituents, namely the R 2 6-(p-hydroxyphenyl) and R 3 8-benzyl, occupy the main p1 tunnel and form multiple interactions, most of which are hydrophobic or aromatic in nature. The R 3 8-benzyl group forms π-π stacking interactions with F260 (roughly 3.7 Å) and F259 (roughly 4.9 Å), while the hydroxyl group of the R 2 6-(p-hydroxyphenyl) substituent forms a hydrogen bond with the carboxyl side chain of D160, an aspartate residue located at the rim of the catalytic cavity (helix α4).
A similar binding mode is observed in the CEI-bound AncFT complex structure (Fig. 2b). There are no great structural differences between the protein backbones of the azaCTZ-and CEI-bound AncFT structures (r.m.s.d. roughly 0.2 Å). In both cases, active-site pocket residues wrap tightly around the bound ligand, preventing both its random motion and free access of solvent molecules. Numerous protein residues including F178, L183, S187, F284, H283, E142, V144, W154, I221 and P218 are involved in this first shell surrounding the bound ligand. Together, these cocrystal structures provide snapshots and molecular details of enzyme-substrate and enzyme-product complexes in catalytically favoured states.

The malleability of the RLuc8 active-site pocket
Structural characterization of RLuc luciferase and its mutants (for example, RLuc8) has proved to be challenging and previous studies provided little understanding of its catalytic mechanism 13,14 . We therefore performed a new round of cocrystallizations with RLuc8 mutants in the presence of molar excesses of either azaCTZ or CTZ to obtain molecular insights into its catalysis. The cocrystallization experiments using the RLuc8-D162A and RLuc8-D120A mutants yielded well-diffracting crystals, and the corresponding structures could be solved by molecular replacement. Crystallographic and refinement statistics for the resulting complexes are presented in Supplementary Table 4. As for the AncFT proteins, interpretation of the electron density maps of the RLuc8-D162A complexes unambiguously identified azaCTZ or CEI. We also observed that CNM, another product of CTZ oxidation 22 , was bound to RLuc8-D120A ( Fig. 3c and Supplementary Fig. 14). In every instance, these ligands were located inside the catalytic pocket.
A common feature of all the newly determined RLuc8 complex structures is a voluminous active-site cavity (Fig. 3). This is most pronounced in the RLuc8-D162A/azaCTZ complex, where we found as many as three azaCTZ molecules bound in the same active-site pocket. One azaCTZ molecule is deeply buried in the cavity with an orientation resembling that seen in the AncFT/azaCTZ complex (Fig. 2a), while the remaining two azaCTZ molecules are rotated through roughly 180° and occupying the remaining volume of the main p1 tunnel (Fig. 3a). A similarly open catalytic pocket is seen in the RLuc8-D162A/ CEI and RLuc8-D120A/CNM complexes in which only a single ligand (CEI or CNM, respectively) is bound (Fig. 3b,c). These ligand-bound structures show that RLuc8 explores a very large conformational space and provide a structural basis for the accommodation of two or more ligands in its catalytic pocket, in keeping with the results of our kinetic experiments (Fig. 1d).
To demonstrate this extreme malleability, we generated a gallery of RLuc8 crystallographic snapshots illustrating active-site cavity opening states ranging from minimally to maximally open (Fig. 4a). The main structural elements responsible for this malleability are the α4 helix, the L9 loop and the L16 loop. Conformational sampling of these structural elements dramatically changes the catalytic pocket's volume and shape. Bulky aromatic residues in the L9 loop (W153 and W156) and the L16 loop (F261 and F262) appear to play particularly important roles in determining the openness of the active site. Although this conformational flexibility may be important in some steps of the catalytic cycle, it greatly complicated our attempts to obtain a crystallographic structure of a biologically relevant Michaelis complex.

A basis for distinguishing flash-type and glow-type bioluminescence
We previously showed that although RLuc8 displays high catalytic turnover (k cat of roughly 4.7 s −1 ), it has a relatively low affinity for its substrate (K m of roughly 1.5 μM) and shows significant product inhibition (K p of roughly 1.2 μM). This may explain why the bioluminescent signal decays rapidly after a strong initial flash 13 . On the other hand, compared to RLuc8, AncFT displayed ˃20-fold higher affinity towards the substrate (K m of roughly 0.064 μM), exhibited distinctly weaker product inhibition (K p of roughly 0.5 μM) and generated a markedly stable glow-type bioluminescent signal 13 . The cocrystal structures of the two enzymes reported here are consistent with these findings. The RLuc8 catalytic pocket is voluminous and malleable (Figs. 3 and 4a), which allows simultaneous accommodation of two or more ligand molecules and explains its tendency towards product inhibition and flash-type bioluminescence. Conversely, the AncFT catalytic pocket is less malleable and can only accommodate a single ligand molecule (Fig. 2), which explains its higher affinity for CTZ, lack of product inhibition and potentially highly stable glow-type bioluminescence. The structural and kinetic data for RLuc8 highlight the importance of our engineered AncFT protein, which is intrinsically less flexible than RLuc8 and is thus a valuable surrogate for studying key chemical steps of the luciferase reaction.

Catalytic analogy between AncFT and RLuc8
Next, we aimed to demonstrate that AncFT and RLuc8 are functionally analogous in terms of their catalytic chemistry. We therefore performed AncFT RLuc8 AncFT RLuc8 77 79 158 160 conserved catalytic pentad are shown as light blue sticks, the remaining protein residues are shown as light blue lines, azaCTZ is shown as yellow sticks and the CEI is shown as green sticks. c, Structure-based sequence alignment of AncFT and RLuc8. Secondary structure elements found in AncFT and RLuc8 are shown above and below the alignment, respectively. Catalytic pentad residues are labelled with black dots and the aspartate at the rim of the catalytic pocket is labelled with a red dot.
Article https://doi.org/10.1038/s41929-022-00895-z pairwise protein backbone comparisons, revealing that the greatest similarity existed between AncFT and the apo-form of RLuc8 with a minimally open active-site cavity. Superposition of these two structures revealed that their backbones have very similar geometries (Fig. 4b). The positioning of the side chains of the putative catalytic pentad residues is identical, allowing analogous chemistry. Moreover, the D160 aspartate residue at the rim of the cap domain in AncFT, which forms a hydrogen bond with the R 2 6-(p-hydroxyphenyl) substituent of the substrate, overlaps well with the D162 residue in RLuc8. Structural comparisons thus showed that AncFT and RLuc8 have very similar active-site chemistries.
To experimentally validate the functional roles of the residues highlighted in our cocrystal structures, we performed structure-based mutagenesis of AncFT and RLuc8. All of the generated mutants were recombinantly prepared as soluble proteins. As shown in Fig. 4c, mutating the conserved catalytic pentad residues in AncFT severely affected its luciferase activity. The most compromising effects were observed for the AncFT-D118A/W119F/H283F mutants, which retained only roughly 0.3, 2 and 0.1% of the activity of AncFT (100%). A less severe inactivating effect was observed for AncFT-N51A (roughly 16%). An aspartate-to-alanine mutation in the cap α4 helix (AncFT-D160A) had only a minimal effect on AncFT catalysis (roughly 91%).
Extensive mutagenesis experiments with RLuc8 have been reported previously 14,15,26,38,39 . We repeated some of these experiments in this work and also performed some new ones. Crucially, we found that the catalytic pentad residues are also critical for RLuc8 activity (Fig. 4d). Mutations in the active-site loops L9 and L16 showed that residues with bulky aromatic side chains (W153, W156 and F262) are catalytically important; this is consistent with the structural data showing that these residues make aromatic π-π contacts with CEI. Accordingly, a b c  Article https://doi.org/10.1038/s41929-022-00895-z the double-point mutation RLuc8-W156A/F262A caused a >16,600-fold reduction in luciferase activity (Fig. 4d), demonstrating the importance of these bulky aromatic residues. In contrast, mutagenesis of polar and charged residues in the L9 loop showed that they are not critical for the luciferase reaction (Fig. 4d). Collectively, our mutagenesis experiments showed that the residues of the conserved catalytic pentad in AncFT and RLuc8 are functionally essential, confirming the catalytic analogy between these two enzymes. This conclusion is also justified by molecular dynamics simulations of key protein-ligand complexes of both AncFT and RLuc8, as discussed below.

Computational modelling of enzyme-ligand complexes
We performed triplicate 10 ns long molecular dynamics simulations of the putative mechanistic ground states to determine whether the Article https://doi.org/10.1038/s41929-022-00895-z reaction mechanism inferred for AncFT can be extrapolated to RLuc8. To this end, we selected the most energetically favourable conformations for each putative chemical step and compared their binding geometries in the two enzymes. Because we had previously established that N53 and W121 (N51 and W119 in AncFT) stabilize O 2 and the leaving CO 2 , we focused on the positions of the other residues involved in the catalytic mechanism (Fig. 4b).
The residues D120 (AncFT-D118), E144 (AncFT-E142) and H285 (AncFT-H283) facilitate and stabilize proton transfer from the enzyme to the luciferin. In the E.CTZ complexes, the histidine residues of these triads form stabilizing hydrogen bonds with the aspartate and glutamate residues: the H285 to D120 and H285 to E144 distances in RLuc8 are 2.4 and 2.7 Å, respectively, while the corresponding distances in AncFT (H283 to D118 and H283 to E144) are 2.5 and 2.2 Å, respectively. At this point, the distance from N1 of the substrate to the aspartate is below 3.5 Å. In the E.2-peroxy-CTZ complexes, dioxygen is already bound to the substrate, the H-bond distances are <3 Å, and the distance from the substrate N1 nitrogen to the catalytic aspartate is 3.3 Å in RLuc and 3.1 Å in AncFT. These distances are optimal for proton transfer from the enzyme's histidine to the substrate, via the aspartate.
In the E.dioxetanone complexes, the proton has been transferred to the substrate and the distance from the substrate to the aspartate is slightly increased. However, the stabilizing hydrogen bond between the histidine and the glutamate remains; these two residues show little fluctuation during the putative chemical steps. Longer distances between the product, the aspartate and the histidine are seen in the E.CEI complexes. As shown in Supplementary Fig. 15, the distances between these residues and from the catalytic residues to the substrate are higher on average in the RLuc8 complexes. However, the orientation of the ligands and catalytic residues in both enzymes are very similar and are stable in all steps, suggesting that these enzymes use analogous catalytic mechanisms. Furthermore, the simulations confirmed that the conformational dynamics of the RLuc8 is significantly pronounced compared to the AncFT (Supplementary Note 3). Especially L9 and L16 in RLuc8 show the high flexibility necessary for proper accommodation of the bulky substrate, in accordance with the crystallographic structures of the ligand-bound enzyme complexes.

CTZ enters the enzymatic pocket with a deprotonated core
We also determined the absorbance peak maxima of CTZ, CTZ analogues and CEI at various pH values ( Supplementary Fig. 18), allowing us to determine the pK a values and monitor the protonation of individual ionizable groups in both CTZ and the CEI (Supplementary Fig. 19 and Supplementary Table 6). As shown in Fig. 5a, comparison of the protonation states of CTZ and CEI with the pH profile of the rate of chemiluminescent CTZ autooxidation revealed that no CTZ conversion occurred when the substrate was fully protonated. At pH values above the pK a of the CEI amide group, the conversion proceeded but luminescence was significantly attenuated (Supplementary Fig. 20). We thus showed that the O10-deprotonated form of CTZ (or its N7-deprotonated tautomer) and the amide-protonated CEI are critical for efficient oxidation and bright luminescence, which is consistent with the crystallographic results.
Our cocrystal structures revealed that no protein residue could potentially sequester a hydrogen atom from the N7-CTZ form. The deprotonation of the O10-CTZ form could theoretically be mediated by the halide-stabilizing tryptophan W121 (AncFT-W119), but this scenario does not seem to be favourable. Indeed, our experiments showed that there was no detectable increase in the deprotonation of this group on mixing RLuc8 with CTZ under anaerobic conditions (Fig. 5b). This makes sense because the pK a of this group is 7.55, which is close the physiological pH. Consequently, around half of the CTZ molecules in the bulk solvent will exist in the deprotonated form, so enzymatic deprotonation is not required. We conclude that CTZ enters the luciferase active site with its imidazopyrazinone core deprotonated.
In contrast, the capping 6-(p-hydroxyphenyl) group of CTZ was deprotonated by RLuc8 after binding (Fig. 5b), indicating that the α4 helix aspartate D162 (AncFT-D160) could mediate this deprotonation. This assumption was validated by the finding that the RLuc8-D162A mutant cannot deprotonate the 6-(p-hydroxyphenyl) group (Fig. 5b). Our data thus confirmed that the nucleophilic D162 residue (AncFT-D160) plays a key role in fine-tuning the emission wavelength by generating a negatively charged emitter, that is, the 6-p-hydroxyphenolate CEI ion.

Halide ions compete with O 2 for the enzyme binding site
Molecular oxygen (O 2 ), which is a cosubstrate of the luciferase reaction, was previously shown to be bound between two halide-stabilizing residues, N51 and W119, in the reconstructed ancestral enzyme Anc HLD-RLuc 9 . This halide binding site is typically occupied by a halide ion during the dehalogenation reaction in HLDs 12,37 . To validate this finding, we measured the luciferase activity of RLuc8 in the presence of various halide anions (Fig. 5c). The tested halides had substantial inhibitory effects, suggesting that they compete with O 2 for the halide-stabilizing residues. Bromide and iodide ions exhibited stronger inhibitory effects than chloride ions (Fig. 5c).

CTZ oxidation proceeds via a superoxide radical
Bui and Steiner have suggested that proper mutual positioning of the substrate and molecular oxygen in the cofactor-independent monooxygenase active site allows O 2 to attack the deprotonated CTZ via a radical mechanism 40 , similarly to what was shown in cofactor-independent dioxygenation of N-heteroaromatic compounds at the α/β-hydrolase fold 41 . We therefore used electron paramagnetic resonance (EPR) spectroscopy combined with spin trapping to capture a putative superoxide radical intermediate. The EPR-silent compound 5,5-dimethyl-1-p yrroline-N-oxide (DMPO) was used as a spin trap agent to form the EPR-active adducts DMPO-O 2− and/or DMPO-OOH by trapping the superoxide species generated in the CTZ luminescence reaction. The EPR spectrum obtained after subtracting the background from a control experiment ( Supplementary Fig. 21) is shown as a blue line in Fig. 5d together with a simulated spectrum (green line) obtained by combining four components (C1-C4) representing different radical species in the sample. The spectra are shown in normalized form in Fig. 5d, but they did not contribute equally to the final spectrum; their individual contributions are shown in Supplementary Table 7 together with the hyperfine coupling constants and the g-factor for each component. Component C1 has the typical parameters of a superoxide radical [42][43][44][45][46] , confirming the generation of superoxide in the reaction. Component C2 can also be attributed to a superoxide radical, despite a disagreement of a H γ for this species 47 . Component C3 can be attributed to a hydroxyl radical, which is expected to be present due to the well-known dismutation of DMPO-OOH into DMPO-OH 46,48 . Component C4 has typical features of a carbon-based radical, and it is probably a CTZ C2 carbanion radical species. Using all four components (C1 to C4) provides not only reasonable statistics (R 2 = 0.8569), but also perfectly explains the shape of the experimental spectrum, especially the peak in the centre, which was not the case when one of the species was omitted ( Supplementary Fig. 22). To our knowledge, this is the experimental evidence of superoxide radical generation during DMSO-activated CTZ chemiluminescence (Fig. 5d). These findings support the proposal that CTZ conversion proceeds via a charge-transfer radical mechanism.

An aspartate of the conserved proton-relay system protonates the CEI ion
As reported previously 49 and verified by our data (Fig. 5a), the excited CEI product must be protonated at the amide nitrogen to avoid pronounced attenuation of its light emission. This CEI reprotonation must occur inside the enzymatic pocket and not in bulk solvent because efficient light emission requires a hydrophobic environment and restricted flexibility 50,51 . This was confirmed by our observations showing that Article https://doi.org/10.1038/s41929-022-00895-z the bioluminescence spectrum corresponds to the fluorescence emission spectrum of the enzyme-bound CEI molecule rather than free CEI (Supplementary Fig. 23). Our cocrystal structures suggest that the residue responsible for this reprotonation is probably aspartate 120 (AncFT-D118), which is part of a previously discussed conserved catalytic glutamate-histidine-aspartate triad. The side-chain carboxylate   Luminescence activity over CEI product formation rate ratio

Fig. 5 | Spectroscopic dissection of the mechanism of Renilla-type
bioluminescence. a, Comparison of pH-titrated protonation states of CTZ and CEI, and a pH profile of the chemiluminescent autooxidation rate of CTZ. Autooxidation is not observed at pH values below the pK a of CTZ. When the pH is above the pK a of CEI, autooxidation proceeds ( Supplementary Fig. 15) but luminescence is attenuated, highlighting the importance of deprotonated CTZ and protonated CEI for efficient luminescence. b, Changes in the protonation state of CTZ on anoxic binding to RLuc8 and RLuc8-D162A. The O10-hydroxy group is half-deprotonated in solution at physiological pH and no further enzymatic deprotonation is observed while the 6-PhOH group is fully protonated in solution and is actively deprotonated by the aspartate D162 on binding. c, Inhibition of the luciferase activity in the presence of halide (X − ) anions. The activity is efficiently inhibited by bromide and iodide ions, indicating competition with oxygen for the halide-stabilizing residues (N53 and W121). d, A spin-trapping experiment using DMPO to study CTZ autooxidation. The experimental EPR spectrum (blue line) was simulated (green line) with four different components (C1 to C4) whose hyperfine coupling constants are shown in Supplementary Table 6. The EPR signal reveals the presence of both superoxide and carbon-based radicals, suggesting that CTZ oxidation proceeds via a chargetransfer radical mechanism. e, Comparison of the bioluminescence spectra of RLuc8 and RLuc8-D120A with the fluorescence spectra of CEI at different pH values. The shift of the RLuc8-D120A luminescence maximum corresponds to the shift of the deprotonated CEI fluorescence maximum, demonstrating the role of aspartate D120 in reprotonating the CEI product. f, Luminescence to product formation rate ratios for selected RLuc8 mutants. RLuc8-D120A generates significantly less luminescence per formed CEI molecule than the parent RLuc8, indicating that emission in this mutant occurs mainly from the less-luminescent deprotonated CEI. This result confirms the role of D120 in tuning the emission wavelength and intensity by ensuring the reprotonation of CEI. The data are presented as best fit values ± s.e. calculated from nonlinear regression and error propagation rules. Article https://doi.org/10.1038/s41929-022-00895-z of this aspartate is located in close proximity (3.3 to 3.6 Å) to the amide nitrogen of the bound CEI. The proton-relay system conserved in the HLD fold thus seems to be critical in this final reprotonation step during the luciferase reaction. The bioluminescence spectrum of CEI in the RLuc8-D120A mutant was blue-shifted by roughly 50 nm, corresponding to the difference in the emission wavelengths of protonated and deprotonated CEI (Fig. 5e). This observation agrees with the results of Shimomura and Teranishi 20 , who found that the luminescence of the amide-deprotonated emitter has an emission maximum of 435-458 nm. RLuc8-D120A yielded a significantly lower luminescence yield per molecule of generated CEI (Fig. 5f). The attenuated light output ( Fig. 5f and Supplementary Fig. 20) and shifted emission spectrum (Figs. 5e and 6a,b) are additional indicators of emission from the deprotonated CEI form, suggesting that the RLuc8D120A mutant cannot effectively perform the reprotonation step and thus confirming the proton-transfer role of residue D120 (AncFT-D118). Finally, the ratio of CNM to CEI obtained by the action of RLuc8-D120A on CTZ was 17 times higher than that achieved with RLuc8 ( Supplementary Fig. 24). This is consistent with our cocrystal structure of RLuc-D120A, in which the CNM molecule is unambiguously present in the active site (Fig. 3c).

Tuning the electronic state of CEI to favour blue light emission
The Renilla luciferase is known for emitting blue light; its emission maximum is roughly 480 nm 7,8,15 . The emission of AncFT is slightly shifted towards higher energy wavelengths, with an emission peak at roughly 450 nm (Fig. 6a). It has been suggested that the CEI formed by the action of these enzymes on CTZ features an electron in an excited state, which loses energy and returns to the ground electronic state via the release of a visible photon. The wavelength of the emitted photon depends directly on the energy difference between the excited and ground states, which in turn depends directly on the environment around the CEI 15 . Although early studies suggested that the CEI exists as an amide anion when it is in the active-site pocket 7,20,52 , more recent works have postulated that the phenolate anion is the blue emitter in bioluminescence (Fig. 6a) 15,[53][54][55] .
Although previous structural studies did not identify a residue that could tune the electronic state of the CEI product 11,13-15 , our structures suggest that this role may be played by an aspartate localized at the rim of the enzymatic pocket. In the CEI-bound AncFT complex, a side-chain carboxylate of D160 (RLuc8-D162) is in close proximity (roughly 3 Å) to a hydroxyl group of the R 2 6-(p-hydroxyphenyl) substituent, suggesting potential hydrogen bonding between the two (Fig. 2 and Supplementary Fig. 13). Similar hydrogen bonding interactions are possible in RLuc8 (Fig. 3). Supporting this hypothesis, mutation of this aspartate to alanine caused significant shifts in the emission maxima of both AncFT and RLuc8 (Fig. 6b). Specifically, AncFT-D160A emits light at the ultraviolet edge, with a single-peaked maximum at roughly 390 nm. Conversely, RLuc8-D162A has a two-peak emission spectrum, with a smaller peak at roughly 400 nm and a major green-shifted peak at roughly 520 nm. This two-peak spectrum may be associated with a closed-to-open transition of its malleable enzymatic cavity. Furthermore, this shift in the bioluminescence of the RLuc8-D162A mutant was only observed during the conversion of substrates whose R 2 substituent bore a hydroxyl group amenable to deprotonation (Fig. 6b,c); it was not seen with CTZ-400a, which lacks this group (Fig. 6d). These observations strongly support the postulated role of D162 in deprotonating this specific substituent. Collectively, our results clearly identify the aspartate residue responsible for fine-tuning the electronic state of the CEI product.

A blueprint for the reaction mechanism of Renilla-type bioluminescence
The results presented above allowed us to delineate a catalytic mechanism for the monooxygenation of CTZ by Renilla-type luciferases. Cocrystal structures of azaCTZ-and CEI-bound AncFT luciferase in which both ligands were captured in catalytically favoured conformations played a vital role in this process. By considering these two structures, we were able to model the binding modes of CTZ and the more short-lived intermediates 2-peroxy-CTZ and CTZ dioxetanone (Fig. 7a). The proposed catalytic reaction mechanism is depicted schematically   site with a deprotonated imidazopyrazinone core because the pK a of the core (7.55) is very close to the physiological pH, as demonstrated experimentally in this work. On the binding, the -OH group of the C6-(p-hydroxyphenyl) substituent is deprotonated by D160 to form the activated dianion O10-CTZ, which affects the emission maximum of emitted light. In the ternary Michaelis complex, the side chains of N51 and W119 position a cosubstrate molecule (dioxygen) such that it can be directly attacked by the C2 carbon of O10-CTZ. Their initial interaction occurs via a charge-transfer radical mechanism.
The next step involves radical pairing and termination to form the 2-peroxy-CTZ anion, which then cyclizes via a nucleophilic addition-elimination mechanism. This step yields a highly unstable and energetic dioxetanone structure with a deprotonated amide group. At this point, the amide group must be protonated by D118 to avoid significant attenuation of the luminescence signal due to the formation of the deprotonated CEI product. D118 also prevents hydrolysis of the amide bond by a water molecule, which would yield the side product CNM. After reprotonation, the unstable dioxetanone ring decomposes and the released energy excites the newly formed CEI product. As it returns to the ground state, the excited molecule releases a photon, representing the bioluminescence signal, together with the final products: ground-state CEI and carbon dioxide. The residue D118 is reprotonated via an interaction with H283.
Article https://doi.org/10.1038/s41929-022-00895-z in Fig. 7b. The mechanism begins with the entry of the deprotonated form of CTZ into the enzyme active site through the main p1 tunnel. The imidazopyrazinone core of CTZ is readily deprotonated in solution because its pK a of 7.55 is close to the physiological pH (Fig. 5a,b). On binding, the -OH group of the R 2 6-(p-hydroxyphenyl) substituent is deprotonated by aspartate 160 (RLuc8-D162) to give the dianionic O10-CTZ, which affects the emission maximum of the emitted light. We have previously shown 11 that a dioxygen binding site in RLuc8 overlaps with the halide binding site of related HLDs and that the binding of molecular oxygen at this site is compatible with the binding mode of CTZ in the enzymatic pocket. In the ternary Michaelis complex obtained after binding of both CTZ and molecular oxygen, the side chains of N51 (RLuc8-N53) and W119 (RLuc8-W121) 11 position the cosubstrate (O 2 ) such that it can be directly attacked by the C2 carbon atom of the activated dianion. The initial interaction occurs via a charge-transfer mechanism that generates radical intermediates. The next steps are radical pairing and termination to form a 2-peroxy-CTZ anion that is then cyclized via a nucleophilic addition-elimination mechanism to form a highly unstable energetic dioxetanone structure with a deprotonated amide group. At this stage, the amide group must be protonated by D118 (RLuc8-D120) to ensure that luminescence occurs from the protonated form of CEI rather than the significantly less-luminescent deprotonated CEI product. The presence of D118 also prevents unwanted hydrolysis of the amide bond by a water molecule, which yields the CNM side product. Following this reprotonation step, the unstable dioxetanone ring decomposes, and the released energy excites the newly formed CEI product. As it returns to the ground state, the excited molecule releases a photon, resulting in the bioluminescence signal (Fig. 7b). The residue H283 (RLuc8-H285) is then reprotonated through E142 (RLuc8-E144) to regenerate the proton-relay system for a next cytalytic cycle.

Discussion
The CTZ-using Renilla luciferase is one of the most popular reporter systems used in biological research and bioimaging technologies. However, despite intensive effort 7,8,11,[13][14][15]26,38,39 , the molecular details of its reaction mechanism remained unknown, largely due to a lack of structural data on its E.S and E.P complexes. In the first structure reported by Loening and coworkers 14 , CEI is bound outside the RLuc8 active-site pocket, and is involved in crystallographic contacts, which makes it difficult to assess the biological relevance of this binding mode. More recently, we captured CEI bound in the RLuc8 enzymatic pocket but the wide opening of the active-site cavity in this structure again made it impossible to infer the reaction mechanism at the molecular level 13 . In this work, we circumvented these problems by using the stabilized AncFT 13 surrogate enzyme and a new non-oxidizable CTZ analogue azaCTZ, to decipher the mechanism of bioluminescence catalysed by α/β-hydrolase fold luciferases.
Our kinetic experiments confirmed that azaCTZ acts as a non-oxidizable substrate analogue in AncFT and RLuc8, and is thus a valuable tool for studying their precatalytic Michaelis complexes by X-ray crystallography. A similar strategy using a different CTZ derivative was recently used to probe the substrate-binding site and mechanism of the Ca 2+ -regulated photoprotein aequorin 56 . We have also proposed a detailed reaction mechanism for Renilla-type bioluminescence that was inferred by considering multiple cocrystal structures of AncFT and RLuc8 luciferases and is supported by the results of mutagenesis, spectroscopic and computational experiments. We show that CTZ adopts a Y-shaped conformation in the active sites of these enzymes, which is required for proper positioning of its imidazopyrazinone core in the enzymatic pocket. Notably, the substrate-binding mode seen in our structures differs markedly from that predicted by Loening and coworkers 15 . We were unable to identify any residue that could potentially initiate the oxidation reaction by deprotonating the bound CTZ substrate at the N7 nitrogen or the O10 oxygen. However, by monitoring the dependence of CTZ autooxidation and chemiluminescence on the pH, we showed that only the O10-deprotonated form of CTZ (or its N7-deprotonated tautomer) can undergo efficient oxidation. Since the first pK a of CTZ is 7.55, which is close to the physiological pH, the O10-deprotonated form will be abundant in the bulk solvent and no enzymatic deprotonation is required, in accordance with the calculations of Griffiths and coworkers 23 . We thus conclude that CTZ enters the enzymatic pocket with its imidazopyrazinone core deprotonated at O10 and binds in an orientation that positions its C2 carbon perfectly for an attack on the bound cosubstrate dioxygen via a charge-transfer process that generates superoxide and CTZ radicals. We also present EPR spectroscopic evidence for the participation of the superoxide anion radical in CTZ oxidation. As long as we are aware, this is the most reliable indication of the radical mechanism that we can experimentally prove for the Renilla-type luciferase catalysed enzymatic reaction, which appears to be a common feature of cofactor-independent bioluminescence 24 .
After radical pairing and cyclization, an unstable dioxetanone intermediate with a deprotonated amide group is formed. Our cocrystal structures reveal that an aspartate of the conserved catalytic triad (Asp-His-Glu) protonates this intermediate, which is required to avoid significant attenuation of the luminescence signal. An aspartate-to-alanine mutation resulted in the formation of CNM rather than CEI. Our results thus demonstrate how a tiny change in enzymatic pocket can alter the CTZ oxidation pathway to favour CNM, which was recently identified as a main product of CTZ oxidation in the photoprotein pholasin 22 . Moreover, we showed that the evolutionarily conserved and functionally important catalytic pentad of HLDs 12,37 is also important in key chemical steps of the CTZ bioluminescence catalysed by α/β-hydrolase fold luciferases. Finally, we identified an additional aspartate at the rim of the catalytic pocket that is not critical for the catalysis but is responsible for fine-tuning the electronic state of the CEI product. The electronic state of the CEI phenolate ion is responsible for the generation of blue light with an emission peak at roughly 480 nm 20 , which is required for energy transfer to green fluorescent protein via bioluminescence resonance energy transfer, which is used by Renilla reniformis. Together with our previous results 11,13 , the findings presented here provide clear evidence that the evolutionary emergence of a functional luciferase resulted from the optimization of a pre-existing HLD-fold protein. Recent discoveries by other groups support this hypothesis and suggest that this may have happened multiple times via divergent evolution 9,10,57,58 .
The Renilla luciferase is a widely used reporter, although its relatively low stability, product inhibition and flash-type bioluminescence can limit its applications. We previously showed that the engineered protein AncFT binds the CTZ substrate tightly and exhibits markedly stable glow-type bioluminescence 13 . Here we report structures that provide a rationale for these distinct behaviours. While the AncFT enzymatic pocket is conformationally constrained and shaped to tightly accommodate a single substrate, the RLuc8 pocket is conformationally malleable and can accommodate up to three ligands. We speculate that this rather unusual behaviour of extant RLuc luciferases may be linked to the absence of its native interaction partners, CTZ-binding protein (CBP) 59,60 and green fluorescent protein 60,61 , which can modify its dynamics via protein-protein interactions.
In conclusion, the mechanistic insights into visible light production in the active sites of α/β-hydrolase fold luciferases presented herein will make it possible to extend the usefulness of these enzymes for science and society. In the near future, rational protein engineering and focused directed evolution will be used to generate customized luciferases for diverse bioluminescent technologies.

Molecular cloning and site-directed mutagenesis
All genes were amplified by a standard PCR and cloned into pET21b expression vector between NdeI and BamHI sites. Mutagenesis was carried out in two step PCR using Phusion polymerase (NEB) according Article https://doi.org/10.1038/s41929-022-00895-z to the manufacturer's protocol. The list of primers used is available in Supplementary Table 8. After the mutagenesis reactions, the original template was removed by DpnI treatment (2 h at 37 °C), followed by final inactivation of DpnI (NEB) enzyme (20 min at 80 °C). The resulting plasmids were transformed into chemocompetent E. coli Dh5α cells, plated on Luria-Bertani agar (tryptone 20 g l −1 , yeast extract 10 g l −1 , NaCl 10 g l −1 , agar 15 g l −1 ) containing ampicillin (100 μg ml −1 ) and incubated at 37 °C overnight. Plasmids were isolated from three randomly selected colonies and sent for DNA sequencing (Eurofins Genomics).

Overproduction and purification of recombinant enzymes
Overexpression of proteins from pET (amp R ) was carried out in E. coli BL21 cells (NEB) cultivated in Luria-Bertani medium supplemented with ampicillin (100 μg ml −1 ) at 37 °C. Protein production was induced at 20 °C once the optical density (OD 600 ) reached roughly 0.6 by adding isopropyl-β-d-thiogalactoside to a final concentration of 0.5 mM. Before purification, cells were disrupted by sonication using Sonic Dismembrator Model 705 (Fisher Scientific). Lysates were clarified by centrifugation (21,036g per 1 h per 4 °C) using a Sigma 6-16 K centrifuge (SciQuip) equipped with a 12166 rotor. Supernatants containing the recombinant His-tagged proteins at their C-terminal ends were metal-affinity purified using Ni-NTA Superflow Cartridge 5 ml (Qiagen) installed on a fast protein liquid chromatography system (Bio-Rad Laboratories) and equilibrated in a purification buffer A (500 mM NaCl, 10 mM imidazole 20 mM potassium phosphate buffer pH 7.5). Proteins were eluted with imidazole gradient and monomers were separated using gel permeation chromatography on an Äkta fast protein liquid chromatography system (GE Healthcare) equipped with HiLoad 16/600 Superdex 200 pg column (GE Healthcare, Sweden) and equilibrated with a GF buffer (50 mM NaCl, 10 mM Tris-HCl pH 7.5). Purified proteins were concentrated to final concentrations using Centrifugal Filter Units Amicon R Ultra-15 Ultracel R -10K (Merck Millipore Ltd). Purity of proteins was verified on SDS-PAGE. Concentration of protein samples was measured using DeNovix R DS-11 Spectrophotometer (DeNovix Inc.).

Cocrystallization experiments
All crystallization experiments were done at 20 °C using the hanging-drop vapour-diffusion method in EasyXtal 15-well plates (Qiagen) with drops equilibrated against 500 μl of reservoir solution.
For azaCTZ-bound AncFT complex, enzyme concentrated to roughly 9 mg ml −1 was mixed with azaCTZ in 1:4 molar enzyme:ligand ratio. Crystals were obtained after mixing 2 μl of enzyme-ligand mixture with 1 μl of the precipitant solution consisting of 0.2 M MgCl 2 , 0.1 M BisTris pH 6.5 and 21% PEG 3350. For CEI-bound AncFT complex, enzyme concentrated to roughly 22 mg ml −1 was mixed with native CTZ in 1:2 molar enzyme:ligand ratio. Crystals were obtained after mixing 1.5 μl of enzyme-ligand mixture with 1 μl of the precipitant solution consisting of 0.2 M Mg acetate and 18% PEG 3350.
For azaCTZ-bound RLuc8-D162A complex, enzyme concentrated to roughly 10 mg ml −1 was mixed with azaCTZ in 1:2 molar enzyme:ligand ratio. Crystals were obtained after mixing 1 μl of enzyme-ligand mixture with 1 μl of the precipitant solution consisting of 0.1 M BisTris pH 6.5 and 25% PEG 3350. For CEI-bound RLuc8-D162A complex, enzyme concentrated to roughly 9 mg ml −1 was mixed with native CTZ in 1:3 molar enzyme:ligand ratio. Crystals were obtained after mixing 1 μl of enzyme-ligand mixture with 1 μl of the precipitant solution consisting of 0.04 M potassium phosphate monobasic and 16% PEG 8000. For CNM-bound RLuc8-D120A complex, enzyme concentrated to roughly 9 mg ml −1 was mixed with native CTZ in 1:3 molar enzyme:ligand ratio. Crystals were obtained after mixing 1 μl of enzyme-ligand mixture with 1 μl of the precipitant solution consisting of 0.2 M MgCl 2 , 0.1 M BisTris pH 6.5 and 20% PEG 3350.
All crystals were fished out, cryogenically protected in the corresponding reservoir solutions supplemented with 20% glycerol, and flash-frozen in liquid nitrogen for X-ray data collection.

Diffraction data collection and data processing
All X-ray data were collected at PXIII beamline at the Swiss Light Source (SLS) Synchrotron (Villigen) at the wavelength of 0.999 Å using a Pilatus 2M-F detector. The data were processed using XDS 62 , and Aimless 63 was used for data merging. Initial phases were solved by molecular replacement using Phaser 64 implemented in Phenix 65 with apo-form AncFT (Protein Data Bank (PDB) 6S97) 13 and RLuc8 (PDB 6YN2) 13 used as search models. The refinement was carried out in cycles of automated refinement in phenix.refine program 66 and manual model building in Coot 67 . The final models were validated using tools provided by Coot 67 and Molprobity 68 . Visualizations of structural data were created using PyMOL 69 . Structural superposition was carried out using the secondary structure matching superimpose tool in the Coot 67 . Atomic coordinates and structure factors of the enzyme-ligand complexes were deposited in the PDB (www.wwpdb.org) 70 under the codes 7QXQ, 7QXR, 7OMD, 7OMR and 7OMO.

Analysis and mapping of enzyme-ligand interactions
In general, enzyme-ligand interactions were studied in all monomers in the asymmetric units. Briefly, all residues within 4 Å of a ligand were found. Then all atoms within 4 Å of a ligand were coloured by element to distinguish between hydrophobic interactions and hydrogen bonds. A π-π interaction was recognized as an interaction between two aromatic rings in which either (1) the angle between the ring planes is less than 30° and the distance between the ring centroids is less than 4.0 Å (face-to-face), or (2) the angle between the ring planes is between 60° and 120° and the distance between the ring centroids is less than 5.0 Å (edge-to-face). Graphical visualizations were done in PyMOL 69 . To prepare a 2D representation of ligand interactions with AncFT, the images of chemical structures were prepared in Chem3D Pro 12.0 (PerkinElmer) software, exported as vector images to be further adjusted in CorelDRAW X6 (Corel Corporation) graphical software. The visual style for two-dimensional drawings was adapted from PoseView (BioSolveIT) software 71 .

Luciferase activity measurements
Luciferase activity measurements were performed as described previously 11,13 . Briefly, Luminescence activity measurement was performed with CTZ (≥95%, for biochemistry, Carl Roth GmbH + Co. KG) as a substrate and determined using FLUOstar Omega Microplate reader (BMG Labtech) at 37 °C. Sample of 25 μl of purified enzyme was placed into the microtiter well. After baseline collection for 10 s, the luminescence reaction was initiated by addition of 225 μl of 8.8 μM CTZ in the reaction buffer (100 mM potassium phosphate buffer, pH 7.5). Luminescence was recorded for 72.5 s and each sample was measured in three replicates (gain 3250). The area of obtained luminescence intensities peak given in relative luminescence units was recalculated to relative luminescence units per mg −1 s −1 .

EPR spectroscopy
The presence of the superoxide radical during the chemiluminescence reaction was detected using EPR using the spin-trapping technique 42 . The EPR-silent compound DMPO was used as the spin trap agent to form the EPR-active adducts DMPO-O 2− and/or DMPO-OOH by trapping the superoxide species generated in the reaction. The DMPO adducts are relatively stable nitroxide radicals and the trapped radicals can be identified according to the separation between the EPR absorption peaks caused by the hyperfine interaction of the electron with the nuclear spin around it. The splitting corresponds to the hyperfine coupling constants (denoted by a) that can be precisely calculated through spectral simulation. The simulations were performed using Easyspin 72 , a freely available package for EPR spectral simulations working for MATLAB 73 . The reactions were prepared in glass vials with a final volume of 100 μl containing CTZ (685 μM) and DMPO (200 mM) diluted in DMSO. Before EPR data acquisition, 40 μl of 50 mM phosphate buffered saline solution pH 7.5 Article https://doi.org/10.1038/s41929-022-00895-z were added to the reaction. As a control, we measured a solution with the same content except for CTZ that was substituted for pure DMSO. The excess of DMPO was intended to avoid secondary radical formation in the reaction. The measured solution was pipetted into a quartz cuvette (flat cell) and measured at room temperature using a Magnettech X-band EPR spectrometer. The spectra were obtained with 10 mW of microwave power and modulation of 0.9 G at 100 kHz. The presented results are the average of 36 accumulations of 120 G sweeps performed for 1 min each.

Computer modelling
The X-ray structures of the azaCTZ-bound AncFT (PDB 7QXR) and CEI-bound AncFT (PDB 7QXQ) complexes solved in this study were used to model E.CTZ, E.2-peroxy-CTZ, E.dioxetanone intermediate complexes. Then, PyMOL 69 was used to model the RLuc complex structures by superimposing the apo-form RLuc8 (PDB 2PSF; chain B) 14 and AncFT complexes, producing analogous RLuc8-ligand complexes for molecular dynamics simulations.

Energy minimization of structural models
The two apo structures (PDB 2PSF; chain B to represent RLuc8 and PDB 7QXR with the substrate deleted to represent AncFT) and the eight complex structures representing each ground state of the putative reaction (four states for AncFT and four states for RLuc) were minimized using YASARA 74 . We started by cleaning the structure and adding all the hydrogens that were missing at pH 7.5. The putative pentad of catalytic residues was visually inspected, and the D162 (D160 in AncFT) on the α4 helix was checked to confirm that their protonation states were as shown in Supplementary Fig. 19. We selected the AMBER FF14SB 75 force field on YASARA and used the option 'clean' that does the following: (1) detects missing bonds, adds them and assigns bond orders; (2) adds missing cysteine bridges between close cysteine atoms, provided that their positions allow bridge formation; (3) adds missing hydrogens to provide a starting point for the following analysis and (4) assigns force field parameters needed in the next steps. In the context of the AMBER force fields, ligands are automatically parameterized using GAFF2 (ref. 76 ) and AM1-BCC charges. After these steps, we ran structure minimization using the YASARA macro em_run.mcr to obtain the starting points for the production simulations. The minimization was considered as converging as soon as the energy improves by less than 0.05 kJ mol −1 per atom during 200 steps.

Molecular dynamics simulations
For the analysis and optimization of substrate, intermediates and product complexes, short molecular dynamics simulations were performed. The minimized structures of all complexes were used as starting points and 10 ns long production molecular dynamics simulations were performed using AMBER 14. These production molecular dynamics simulations were run using YASARA's macro, md_run.mcr. The systems were solvated in a cubical water box of explicit solvent, with density 0.997 g ml −1 , so that all atoms were at least 12 Å from the boundary of the box. Cl − and Na + ions were added to neutralize the protein's charge and get a final concentration of 0.1 M. All the simulations used periodic boundary conditions, the particle mesh Ewald method was used to treat interactions beyond a 10.5 Å cut-off, electrostatic interactions were suppressed more than four bond terms away from each other, and the smoothing and switching of van der Waals and electrostatic interactions were cut off at 8 Å. For the molecular dynamics simulations analysis, md_analyze.mcr was used for the r.m.s.d. and root-mean-square fluctuation (RMSF) analysis. The distances between atoms were measured in PyMOL 69 . Three replicas of the production simulations were performed to ensure the significance and convergence of the simulation.

Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability
Atomic coordinates and structural factors have been deposited in the PDB (www.wwpdb.org) 70 under accession codes 7QXR, 7QXQ, 7OMD, 7OMR and 7OMO. We will release the atomic coordinates and experimental data on article publication. The primary data from molecular dynamics simulations are available in Zenodo repository with the identifier https://doi.org/10.5281/zenodo.7241544. Source data are provided with this paper.