X-ray free-electron laser studies reveal correlated motion during isopenicillin N synthase catalysis

Complementary XFEL methods reveal O2 reaction intermediates and global protein dynamics during Isopenicillin N synthase catalysis.


INTRODUCTION
Following pioneering studies demonstrating l--(-aminoadipoyl)l-cysteinyl-d-valine (ACV) is the precursor of all natural penicillins, isopenicillin N synthase (IPNS) was shown to catalyze formation of both penicillin -lactam and thiazolidine rings, in an iron-and dioxygen-dependent reaction without synthetic precedent (1)(2)(3). IPNS is a member of the Fe(II) and 2-oxoglutarate (2OG) oxygenase structural superfamily (4); such enzymes are widespread in nature and have important roles, including in collagen biosynthesis, lipid metabolism, nucleic acid repair, and signaling (5,6). 2OG oxygenases play key roles in the human hypoxic response via hydroxylation of the hypoxia inducible factors in a manner regulated by dioxygen availability (7,8). Understanding how dioxygen interacts with the 2OG oxygenase superfamily is of interest from catalytic and physiological perspectives.
2OG oxygenases typically catalyze hydroxylation reactions that are coupled to 2OG oxidation giving succinate and CO 2 (4,5). IPNS catalyzes a unique four-electron oxidation involving two challenging C─H bond cleavages during conversion of the inactive peptide ACV to the conformationally strained antibiotic isopenicillin N (IPN) ( Fig. 1 and fig. S1). Substrate analog (2,9), kinetic (10), spectroscopic (11,12), and modeling (10,11,(13)(14)(15)(16)(17) studies imply binding of dioxygen to the IPNS:Fe(II):ACV complex that yields an Fe-linked superoxide, which abstracts a C-3 hydrogen from the ACV cysteine giving a thioaldehyde, which undergoes 4-exo-tricyclization to produce a -lactam linked via S to an Fe(IV)═O species. Thiazolidine formation occurs via valine hydrogen abstraction followed by reductive elimination giving the bicyclic IPN ring system (fig. S1) (4,6,10). There is a lack of knowledge of how the IPNS protein mediates the conformational changes required for penicillin formation, and structural knowledge of the intermediate formed by reaction of IPNS:Fe(II):ACV with dioxygen has been unavailable.
Redox labile intermediates are not amenable to synchrotron analysis because they react via x-ray-induced photoelectric effects (18). X-ray free-electron lasers (XFELs) provide intense femtosecondlong x-ray pulses enabling diffraction data from thousands of micron-size crystals, limiting radiation-induced chemistry (serial femtosecond crystallography, SFX) (19). To enable time-resolved (tr) SFX studies (20)(21)(22) on IPNS catalysis, we exposed anaerobic IPNS:Fe(II):ACV microcrystals to dioxygen using acoustic droplet ejection (ADE) tape drive methods (23) with simultaneous monitoring of the Fe oxidation state by x-ray emission spectroscopy (XES) (24,25). We subsequently used 19 F NMR (nuclear magnetic resonance) to probe the solution relevance of the dynamics revealed by tr-SFX.
our IPN product complexes derived from microcrystals, the Fe refined to convergence with low occupancy (40 to 60%). However, solution studies using NMR and non-denaturing mass spectrometry (MS) provide no evidence for IPN-promoted Fe displacement (fig .  S2). Thus, the apparent loss of Fe on formation of IPN in crystallo likely reflects tight IPN binding due to lattice packing interactions, which promote release of Fe rather than IPN.
We investigated the reaction of IPNS:Fe(II):ACV with O 2 by exposing microdroplets of the anaerobic microcrystal slurry to a 100% O 2 atmosphere, regulating the reaction time by changing the ADE tape drive speed (23). On the basis of the high ratios of indexed crystal lattices to images collected, multiple datasets (for 400-, 500-, 800-, 1600-, and 3000-ms O 2 exposure times) were collected from the same microcrystal batch (table S1) enabling robust dataset comparison.
Diffusion of O 2 into microcrystals slightly alters the unit cell parameters, suggesting changes that alter fold and/or lattice packing. No clear structural differences at 400, 500, or 800 ms were apparent ( fig. S3); however, analysis of the ACV B-factors at 500 ms and longer times reveals substantial increases (22 Å 2 /anaerobic to 36 Å 2 /1600 ms; Fig. 2 and fig. S3). The ACV valine B-factors increased first (500 ms), followed by those of the entire ACV (≥ 800 ms), including the Aad side chain. The latter was unexpected, because all ACV/IPN atoms are fully ordered in both, the anaerobic ACV substrate and IPN product complexes ( Fig. 1 and figs. S2 and S3). These observations suggest that O 2 binding induces motions beyond those directly involved in covalent reaction at the active site. We observed that addition of glycerol (to minimize dehydration) reduces the reaction rate, consistent with O 2 diffusion being slowed by increased viscosity in the presence of glycerol ( fig. S4) (29).
Positive electron density appeared at the Fe coordination site trans to Asp 216 in the 1600-and 3000-ms SFX datasets ( Fig. 3 and  fig. S5). The 1600-ms dataset electron density was initially fit and refined with a coordinated water, but difference features remained and this model did not adequately reflect the scattering. The best refined fit to the 1600-ms dataset includes an O 2 bound "end-on" to the Fe (figs. S5 and S6), consistent with modeling ( fig. S7) (11,16).
In refinements for the 1600-and 3000-ms datasets, we omitted those ACV atoms with very weak or lacking 2mF o -DF c density, which did not generate mF o -DF c difference features when removed. These atoms are present but are disordered in intermediates rendering them invisible to x-ray diffraction (XRD), despite the use of 35-fs x-ray pulses. By contrast, before reaction initiation and following its completion, all nonhydrogen ACV or IPN atoms are visible. Thus, as O 2 -initiated reaction progresses, the ACV becomes dynamic, with the extent of disorder varying with the substrate region.
The results imply that O 2 binding causes the ACV Val isopropyl group to rotate away from the Fe to avoid a steric clash; similar rotation is observed in synchrotron cryo-structures for the IPNS:Fe:ACV:NO complex, wherein NO is an O 2 surrogate ( Fig. 3 and fig. S7). By contrast with the IPNS:Fe:ACV:NO complex, in the 1600-ms dataset, electron density for the ACV Val is weak and the Val isopropyl methyls were deleted in the refinement models [ACV conformation (conf.) A, 20% occupancy (− O 2 ) and conf. B, 80% occupancy (+O 2 ); Fig. 3 and fig. S5].
Similar considerations apply to the 3000-ms dataset, where we fit and refined the ACV with the entire Val isopropyl deleted. In this dataset, we also observed decreased electron density, consistent with partial disorder, for both the valine carboxylate and Aad side chain (conf. C; Fig. 3 and fig. S5). In the 3000-ms dataset, an additional conformation (conf. D) was proposed in which the ACV cysteinyl methylene rotates such that the pro-S C Cys, ─H bond is close to the distal O of the Fe-O 2 (2.4 Å compared to 3.1 Å in conf. C). This produces a complex aligned for stereospecific cysteinyl 3─H bond cleavage (10) leading to a thioaldehyde, which is preorganized for stereoelectronically favored -lactam formation ( Fig. 3 and figs. S5 and S6). Thus, the tr-SFX results reveal that, at this stage of the reaction, the ACV Cys/Val atoms involved in thioaldehyde and -lactam formation appear to be more ordered than those Aad and Val atoms that are not directly involved in these steps of the reaction cycle.
The tape drive setup enables simultaneous collection of tr-SFX and tr-XES data (23), allowing assignment of the Fe oxidation states (30). XES analysis of the anaerobic IPNS:Fe:ACV microcrystal complexes supports the Fe(II) oxidation state ( Fig. 4A and 4B, and fig. S6). In the O 2 -bound datasets (1600 and 3000 ms and 500 ms without glycerol; figs. S4 and S6), changes are observed in the XES and difference spectra (Fig. 4A), especially in the full width at half maximum (FWHM) (25) of the K 1 peak (Fig. 4B). These changes are similar in magnitude to those on going from Fe(II) to Fe(III) in Mn/Fe containing R2a ribonucleotide reductase (23), in the diiron center of methane monooxygenase (31), and in small-molecule Fe complexes (32), supporting formation of an IPNS:Fe(III):ACV:O 2 − intermediate consistent with solution studies (10,11,17) and our tr-SFX data.
The tr-SFX datasets reveal clear changes of average B-factors in two structurally separate regions: (i) the C-terminal region including 10 [which is important for productive catalysis (3,33)] and the loop connecting 10 with the protein core and (ii) 3, for which no function is ascribed and which is located on the exterior of the double stranded  helix core (DSBH) fold ( Fig. 2 and figs. S3 and S4). Two other exterior regions also refine with increased B-factors (amino acids 105 to 120, including 5) and an exterior loop (amino acids 194 to 206), although these do not exhibit changes correlating well with increasing O 2 exposure time. The 10 B-factors increase with time and remain high in the IPNS:Fe:IPN complex. Those for 3 increase with time, but in the IPNS:Fe:IPN complex ( Fig. 2 and fig. S3), they refine back to values close to those for anaerobic IPNS:Fe:ACV. Comparison of the F obs − F obs isomorphous difference maps for the 400-and 3000-ms datasets reveals clear evidence for dynamic changes in the 3/11 conformations, consistent with the calculated mF o -DF c maps ( fig. S3). B-factor analyses of the tr-SFX O 2 -exposed datasets reveals increased dynamics of 3; however, we were unable to confidently fit and refine discrete new conformations because the electron density maps indicated that these are present at low occupancy (10 to 15%). Therefore, we attempted to trap thermodynamically unstable conformations by exposing single IPNS:Fe:ACV crystals to O 2 and then plunging them into liquid nitrogen for cryo-synchrotron analysis. Analysis of ~100 IPNS:Fe:ACV crystals exposed to O 2 for times ranging from 30 to 600 s shows evidence for electron density corresponding to O 2 binding trans to Asp 216 [although photoelectric reduction likely occurs (18)] with a trend to disorder of the ACV valine, along with increased 10 and 3 B-factors. Notably, under the cryo-conditions, along with the starting conformations as observed previously (I), we unequivocally observed the formation of discrete new conformations for 3 and 11 (II) (Fig. 5 and fig. S7).
Similar changes in valine disorder, increased B-factors and the new 3/11 conformations, were observed on formation of the IPNS:Fe:ACV:NO complex in single crystals ( Fig. 5 and fig. S7). A structure of the IPNS:Fe:ACV:NO complex (PDB: 1BLZ) (26) has been reported; alternative confs. of 3 and 11 due to NO binding were not refined, possibly due to low occupancies of the flexible regions. As for the tr-SFX-observed O 2 binding and as reported in a synchrotron derived structure (26), NO binds trans to Asp 216 , but with a slightly different orientation to O 2 (Fig. 5D). These NO or O 2 -exposed structures were refined with both 3/11 conformations (I and II)-the IPNS:Fe:ACV:NO complex and a representative IPNS:Fe:ACV:O 2 complex with 1:1 occupancy of conformations I and II were selected for deposition. The difference between the cryo-condition work, where we observed the discrete new conformations for 3/11, and the tr-SFX studies, where we did not accrue unequivocal evidence for them, likely in part reflects higher mobility under the latter room temperature conditions.
The combined tr studies enable experimentally based proposals for the underlying interactions involved in dynamics during IPNS catalysis. Although motions other than those involving 3/11, including the C-terminal region in substate binding and product release, are likely involved, our studies reveal a critical role for 3/11 dynamics in -lactam ring formation. O 2 binding induces mobility of the ACV valine weakening its interaction with Ser 281 11 causing 11 to move away from ACV (Fig. 5, fig. S7, and movie S1). This increases the active site volume by ~13 Å 3 , so enabling productive movement of the CV unit of ACV to initially form a thioaldehyde and then a monocyclic -lactam (Fig. 6).
As a consequence of the movement of Ser 281 11 and other 11 residues, conformational changes occur throughout 3 on the protein surface (amino acids 47  [Tyr 224 is conserved and adjacent to Leu 223 , which is part of a hydrophobic pocket binding the ACV valine isopropyl (26,33)] and His 62 3 and Tyr 90 3 ( Fig. 5C and fig. S7). Tyr 90 is adjacent to Tyr 91 , which binds the ACV Aad amino group via a hydrogen bond  ( Fig. 5C). The Aad side chain, including its amino group, becomes disordered following O 2 exposure but is ordered in the IPN product complex. 3 thus plays a central role in conformational changes involved in the conversion of ACV to IPN via dynamic interactions with both the ends of ACV/intermediates, i.e., with the Aad amino group and the Val carboxylate.
We then used 19 F-NMR to investigate whether the conformational changes observed for 3 in crystals occur in solution, by monitoring changes of IPNS 19 F-labeled (34) on 3 (IPNS*; Fig. 7  and fig. S8). The results reveal distinct signals for the apo, Fe(II), and Fe(II):ACV complexes. Addition of NO to IPNS*:Fe:ACV, but not IPNS*:Fe or IPNS*:Cd:ACV ( Fig. 7 and fig. S8) complexes, resulted in movement of the IPNS* 19 F signal ( Fig. 7 and fig. S8), supporting the proposal that O 2 /NO binding induces 3 dynamics in presence of ACV, since NO should not bind to the Cd(II) complex.

DISCUSSION
Our studies provide new insights how correlated motions induced by O 2 binding enable IPNS to catalyze -lactam formation, including by transiently altering the active site volume and arranging the substrate conformation in preparation for ring formation. Sequence comparisons reveal that 3/11 and key residues involved in the associated correlated motions are conserved in IPNS ranging from prokaryotes to an animal ( fig. S9A) (35). 11 (which extends one  sheet of the core DSBH fold) and 3 (one of the two conserved exterior N-terminal helices) are present in multiple 2OG oxygenases, ranging from those of cephalosporin biosynthesis which are closely related to IPNS to those involved in DNA/RNA repair, histone demethylation, lipid metabolism, and the human hypoxic response ( fig. S9B) (5,6). Thus, although there are likely variations, and the extent of protein dynamics during catalysis is likely underestimated in crystallographic studies, the types of correlated motions observed by tr-SFX during IPNS catalysis are likely of widespread relevance. Detailed knowledge of how oxygenases work should enlighten research concerning their biological roles as sensors, on engineering them to alter the course of biosynthesis and on modulating their activity for therapeutic benefit.
The roles of dynamics in tuning catalysis of individual steps are of general interest in catalysis, particularly for enzymes that catalyze chemically challenging reactions involving substantial conformational changes (36). tr-SFX is a powerful method for uncovering these, including, as shown here, motions involved in steps after substrate binding and before product release. We hope increased knowledge of how IPNS works will promote the use of tr-SFX to study synthetically challenging reactions and inspire the discovery of new types of nonprotein catalysts making densely functionalized and conformationally strained ring systems. However, our results demonstrating that motions outside the active site are important in catalyzing conversion of a "simple" peptide to the penicillin nucleus suggest that such biomimetic catalysts may need to be macromolecular.

MATERIALS AND METHODS
Chemicals for preparation of buffers and crystallization screens were from commercial suppliers and were used without further purification. ACV was synthesized by solid phase peptide synthesis and purified using a Shimadzu high-performance liquid chromatography (HPLC) system, equipped with a SunFire semiprep column (C18, 5 m, 150-mm length, 10-mm diameter). The mass of ACV was confirmed by LC-MS; Agilent Technologies 1260 Infinity Series, equipped with a 6120 quadrupole mass spectrometer using a Merck Chromolith Performance C18 (100 mm by 4.2 mm) HPLC column. ACV purity was confirmed by NMR (Bruker AVIII HD 600 equipped with a BB-F/H N 2 Prodigy CryoProbe). Recombinant wt-IPNS (pJB703) from Aspergillus nidulans in NM554 Escherichia coli was produced by the reported procedure (37).

Cloning of pCOLD_IPNS
DNA (codon-optimized for expression in E. coli) of IPNS (GeneArt; Thermo Fisher Scientific, UK) was inserted into the pCOLD I vector (Addgene, UK) using Sal I and Not I restriction sites. This vector enabled production of IPNS with an N-terminal hexa-histidine tag (6xHis) with an N-terminal C human rhinovirus protease cleavage site as reported (38).

Mutation of pCOLD_IPNS S55C
A Ser 55 to Cys variant of codon-optimized IPNS produced with the pCOLD vector was obtained using the Q5 site-directed mutagenesis kit (New England Biolabs, USA) followed by DpnI digestion at 32°C level, green, modeled occupancy of 90% for Cd). Unmodeled electron density was observed next to Trp 213 on the protein surface. Note that exposure of IPNS:Cd:ACV crystals to NO reveals no NO binding to the metal and no evidence for 3/11 rearrangements, even after multiple NO exposures with 10 crystals, consistent with the NMR analysis (B). for 5 hours. Primers used are as follows: IPNS_S55C_fwd: cattaatgtgcagcgtctgtgccagaaaaccaaagaatt and IPNS_S55C_rev: aattctttggttttctggcacagacgctgcacattaatg. The polymerase chain reaction product was transformed into XL10 Gold ultracompetent cells (Agilent Technologies, USA) by heat shock and grown overnight on 2YT agar plates containing ampicillin (50 g ml −1 ). A single colony was picked and cultured in liquid 2YT media while shaking (150 rpm) overnight at 37°C. Plasmid DNA was isolated using the GeneJET Plasmid Miniprep kit (Thermo Fisher Scientific, USA). The presence of the desired mutation was confirmed by DNA sequencing (Eurofins, Germany).
A cell pellet (25 g) was resuspended (1:4 w:v) in buffer G [50 mM tris (pH 7.5), 200 mM NaCl, and 5 mM imidazole] supplemented with deoxyribonuclease I (10 g ml −1 ), phenylmethylsulfonyl fluoride (10 g ml −1 ), and lysozyme (0.2 mg ml −1 ) by stirring at 4°C and for 30 min. Cells were lysed by sonication (9 s on: 9 s off, 60% amplitude, 12-min total time, 4°C; Sonics Vibra-Cell), cell debris was removed by centrifugation (58,000g, 30 min, 4°C), and the supernatant was filtered (0.45 m). The filtrate was loaded onto a nickel affinity column (5 ml HisTrap HP, GE Healthcare, USA), pre-equilibrated with buffer G (20 CV), and eluted with a gradient from buffer G to buffer H [50 mM tris (pH 7.5), 200 mM NaCl, 500 mM imidazole, and 20 CV]. Fractions containing IPNS were identified by SDSpolyacrylamide gel electrophoresis (PAGE) and concentrated by centrifugation [10,000 molecular weight cut-off (MWCO), 3000g, 4°C; Merck Millipore, USA], before purification by Superdex 75 (300 ml, GE Healthcare, USA) chromatography using a column pre-equilibrated with buffer I [25 mM tris (pH 7.5), 500 mM NaCl, and 1 CV]. Fractions containing protein were identified by ultraviolet analysis and analyzed by SDS-PAGE. Fractions containing purified IPNS_S55C were concentrated using a centrifugation tube (10,000 MWCO, 3000g, 4°C; Amicon Ultra). Removal of the N-terminal hexa-histidine affinity tag was achieved by incubation with 3C protease (1:100, w/w) overnight at 4°C. Successful cleavage was confirmed by protein MS (Quattro Premier XE, Waters, USA). Untagged IPNS_S55C was subsequently purified using a nickel affinity column (5 ml HisTrap HP, GE Healthcare, USA) using buffer G (5 CV). . Note that the predicted mass difference based on the calculated mass of CH 2 C(O)CF 3 (110 Da) was higher by 18 Da than observed, suggesting that the ketone exists mainly in its hydrated form, as observed for a different protein (34,40,41), and consistent with our crystallization analyses (Fig. 7). The reaction was stopped by buffer exchange using PD-10 columns, pre-equilibrated with buffer J. The BTFA-labeled IPNS_S55C (IPNS*) variant was concentrated to a volume < 2 ml and further purified by size exclusion chromatography using a Superdex 75 column (300 ml, GE Healthcare, USA) pre-equilibrated with buffer J. Protein containing fractions were subsequently dialyzed as described above. The apo-IPNS* was concentrated to 50 mg ml −1 , aliquoted, and stored at −80°C. Trypsin digestion was performed to confirm the labeling (fig. S10).

Tryptic digestion
Tryptic digestion was performed according to a reported procedure (42). Labeled IPNS* (15 l, 1 mg ml −1 ) was treated with dithiothreitol (DTT; 2 l, 85 mM), in ammonium bicarbonate buffer (10 mM), for 40 min at 56°C. The protein was then alkylated with iodoacetamide (7 l, 55 mM) for 30 min and then treated with DTT (3 l, 85 mM) for 10 min at room temperature (rt) in the dark. Trypsin (3 l; 1:20, w/w) was added and reacted for 16 hours at 37°C. The sample was then diluted with acetonitrile (120 l; 1:100, v/v) and allowed to react for 3 hours at 37°C. Digestion was stopped by the addition of formic acid (7.5 l), and the mixture was then vacuum centrifuged (Eppendorf, Germany) to dryness. The residue was redissolved in H 2 O (18 l) supplemented with aqueous formic acid (0.1% v/v). After pre-equilibrating a ZipTip column (Merck Millipore, USA) with aqueous acetonitrile (98:2, 0.1% formic acid), the sample was aspirated into the ZipTip column. The fragments were eluted from the ZipTip column using aqueous acetonitrile (40:60, 0.1% aqueous formic acid), vacuum centrifuged to dryness and redissolved in aqueous acetonitrile (98:2, 0.1% formic acid), and analyzed by nanoLC (Thermo Elite, Thermo Fisher Scientific, USA), and data were analyzed using BSI PEAKS studio 8.5. The results from the trypsin digestion are shown in fig. S10.

Nitric oxide exposure of IPNS crystals and solutions
Experiments involving nitric oxide (NO) were carried out by two researchers using a fume hood. A two-chambered, four-valve glass apparatus ( fig. S11) was used for controlled NO exposure, minimizing O 2 and user exposure. Anaerobic protein/crystal samples, sealed in J Young valve NMR tubes (5 mm, Norell, USA), were attached to the NMR valve without opening the NMR tube ( fig. S11). The system was first purged with argon (10 min) with the gas outlet and separating valves open. The gas outlet valves were closed, under a slight overpressure of argon. Chamber 2 was sealed by closing the separating valve, and vacuum was applied by opening the chamber 2 outlet valve. To create a mild vacuum in the NMR tube, the outlet valve was closed before opening the NMR valve. After equilibration (~2 min), the NMR valve was closed and the chamber 2 outlet valve was reopened. This process was repeated twice. Chamber 1 was purged with NO [1000 parts per million (ppm) in nitrogen, (0.1% NO and 99.9% N 2 ), 10 min; BOC, UK]. After purging chamber 1, the outlet valve was closed while the system was under a positive pressure, and the balloon was partially filled with NO. To expose the sample to NO, the separating and NMR valves were opened. Samples were incubated with NO for 10 to 30 min to yield a pale pink solution (12) in the NMR tube. The NMR valve was closed and samples was removed for analysis. To remove NO from the system, all valves were opened and flushed with argon for several minutes. Note that, for safety reasons, no vacuum was applied once the system was filled with NO.

NMR studies
NMR experiments on IPNS* [the 19 F-BTFA-labeled IPNS_S55C mutant isoform (trifluoroacetonyl labeled); Fig. 7 and fig. S8] were recorded using a Bruker AVIII HD 600 equipped with a BB-F/H ProdigyN 2 CryoProbe, in 5-mm regular or J Young valve NMR tube (5 mm; Norell, USA), at 298 K, unless stated otherwise. 19 F NMR spectra were referenced to CF 3 CO 2 H (100 M, at  F = −76.55 ppm) and processed with 3-or 30-Hz Lorentzian line broadening using MestReNova 14.1 (MestReLabs, Spain; www.mestrelab.com) and TopSpin 3.6.1 (Bruker, Germany; www.bruker.com). Sample preparation was performed under anaerobic conditions (<2-ppm O 2 ) in an anaerobic chamber (Belle Technology, UK). Solutions [in 25 mM tris-d 11 buffer (pH 7.5)] were deoxygenated by argon purging (30 min) before placing in an anaerobic chamber. Solids (FeSO 4 and ACV) and NMR tubes were transferred into the glovebox and left to equilibrate (16 hours). Solutions of apo-IPNS* (50 mg ml −1 , 1.35 mM, and 50 l) were transferred to the anaerobic chamber immediately before use. Appropriate stock solutions of ACV (10 mM), FeSO 4 (100 mM), and CF 3 CO 2 H (10 mM) were prepared in a glovebox. The total volume of a typical sample was 450 l and contained 10% (v/v) D 2 O. After equilibration, the sample was transferred from an Eppendorf tube (1.5 ml) into a 5-mm J Young valve NMR tube (Norell, USA) for analysis.
For the addition of subsequent solutions to the sample, a J Young valve NMR tube was transferred into the glovebox and equilibrated for 5 to 10 min before opening. The appropriate solution was added, and the tube was closed and inverted to equilibrate. For exposing a sample inside a J Young valve NMR tube to nitric oxide (NO), the tube was connected to an appropriate glass apparatus ( fig. S11), and the sample was treated as described above. The J Young valve was closed before disconnecting the tube and inverted carefully several times to allow mixing of the gas in the headspace with the solution.

Preparation of salt free IPNS and nondenaturing MS
Purified IPNS was used for nondenaturing electrospray ionization MS after buffer exchange into ammonium acetate (1 M). Buffer exchange was performed using Zeba Micro Spin desalting columns (Thermo Fisher Scientific) followed by 3-hour dialysis with ammonium acetate (1 M). Samples were mixed with the different (co-) substrates dissolved in Milli-Q after different incubation times. Nondenaturing MS data were acquired using an Orbitrap extended mass range prototype (43) machine. Data were recorded in positive ion mode, from a static nanospray source, using a gold-plated capillary prepared in house. Nitrogen was used as a collision gas [the pressure of around 1 × 10 −9 mbar recorded in the Orbitrap, no HCD voltage (0 V) applied]. The capillary temperature was set to 30°C, and a spray voltage of 1.4 kV was applied. Acquired data were processed and analyzed using Thermo Xcalibur 4.13 (Thermo Fisher Scientific, UK) (43).

Optimization of microcrystals
Crystallization was conducted within an anaerobic chamber maintained at 2 ppm or less O 2 (Belle Technologies, UK) with plates, solutions, and other equipment used for crystallization deoxygenated within the chamber for at least 24 hours. The IPNS solution that was used to grow crystals was deoxygenated in the anaerobic chamber for 1 to 2 hours before use.
IPNS microcrystals were prepared as reported (27). For setting up a batch plate of microcrystals, IPNS (200 l, c = 52 mg ml −1 ), ACV (2.2 mg), FeSO 4 ·7H 2 O (1 eq., 100 mM stock in H 2 O, 2.78 l) and seed stock solution (12 l) were transferred into an anaerobic chamber (Belle Technologies, UK). Using a 96-well plate (Corning, USA), solutions for the batch setup were prepared as follows.  (28). Crystals were prepared using the hanging drop method by combining the reservoir solutions (3 l) and protein solutions (3 l). Crystals (60 to 200 m) formed after 24 to 36 hours were harvested and cryo-cooled by rapid plunging into liquid N 2 before data collection. Single IPNS crystals were cryoprotected by transferring to a solution of mother liquor [1.7 M Li 2 SO 4 and 0.1 M tris (pH 8.5)] supplemented with 20% (v/v) glycerol before being cryo-cooled in liquid N 2 . Data for the single crystals were collected at 100 K using synchrotron radiation at the Diamond Light Source beamlines I03, I04, I04-1, and I24 and processed using the Xia2 pipelines (table S1) (44).
Note that formation of IPNS:Cd:ACV crystals under aerobic conditions was only possible when the starting enzyme was completely metal ion free, possibly due to ACV turnover by a small amount of Fe(II)-bound enzyme in solution under aerobic conditions. Needle-shaped crystals were used to prepare seeds using the PTFE Seed Bead Kit as described by the manufacturer (Hampton Research, USA). These crystals were used in batch methods (as described above) to obtain more IPNS:Cd:ACV crystals for NO exposure experiments. Note that while all complexes crystallized with Fe(II) show clear turnover in crystallo and in solution (followed by NMR and mass spectroscopy) (33,45) after removal from the anaerobic chamber and exposure to O 2 from air, the Cd(II) complex crystallized under aerobic conditions and showed no turnover of ACV to IPN.

Setup and sample injection for crystallographic data collection
Room temperature diffraction data for microcrystal slurries were collected at the MFX (Macromolecular Femtosecond Crystallography) instrument of Linac Coherent Light Source (LCLS) (46,47) (proposals LU50/P143 and LS34/P110) and at the BL2 beamline at SACLA (proposal 2017B8085) (table S1). The drop-on-tape (DOT) sample delivery method (23) was used combined with ADE at LCLS to obtain anaerobic IPNS:Fe(II):ACV complex and tr O 2 -exposed structures (23). The ejected droplets (~3.5 to 4 nl; flow rate, 7 l min −1 ) are deposited onto the conveyor belt at room temperature in a helium atmosphere. To trigger the reaction in crystallo, our method exposes the microdroplets of the anaerobic crystal slurry to a 100% O 2 atmosphere as they pass through a 60-mm-long reaction chamber for a varied time regulated by the Kapton belt speed (table S1) (23). We collected tr-SFX datasets after exposing droplets to O 2 and then applying additional reaction times of 400, 500, 800, 1600, and 3000 ms. Shorter reaction times derive from faster belt speeds, which reduces the overall time for O 2 diffusion into the microdroplets and hence reaction within the microcrystals. Longer reaction times derive from a slower belt speed and commensurately longer O 2 droplet equilibration and reaction times. Detailed information about the Kapton tape speed (millimeter per second) and the O 2 incubation times (millisecond, including incubation time in the reaction chamber and additional travel time of microdroplets from the reaction chamber until they arrive in the x-ray interaction zone) are shown in table S1. The x-ray wavelength for experiments obtained under proposals LU50/P143 was kept at 1.30 Å (9.537 keV) and LS34/P110 at 1.31 Å (9.464 keV) with a data collection rate of 30 Hz, 4 mJ/ pulse, a pulse duration of ~35 fs and an x-ray beam size at the sample of ~3 m in diameter. XRD data were collected using a Rayonix MX340-HS detector for the LU50/P143 beam time and on a Rayonix MX170-HS detector for LS34/P110 with 4 × 4 binning.
The IPNS:Fe:IPN product complex SFX data were collected using the viscous extruder and a grease-matrix carrier by first exposing anaerobic slurries of the IPNS:Fe(II):ACV complex (crystal density > 5 × 10 7 ml −1 ) to atmospheric O 2 for 30 min and then mixing the slurry with grease in the ratio 10 l slurry:90 l grease at SACLA (48). Mixing of grease with slurry was performed by using two 100-l Hamilton syringes connected over an extruder-based connector. The obtained grease matrix containing randomly oriented protein microcrystals was applied to the sample reservoir (60 l of reservoir and 2 mm in diameter) of the sample delivery system and mounted in the setup for data collection. A flow rate of 1 to 1.5 l min −1 and a nozzle dimension of 100 m were used to obtain a stable flow for the protein grease matrix. Data were collected at room temperature with a collection rate of 30 Hz and an x-ray beam size at the sample of ~1.7 m in diameter. The x-ray wavelength for the experiment was kept at 1.14 Å (11 keV) with 0.34 mJ/ pulse and 10-fs pulse length. XRD data were collected on a MPCCD (MultiPort Charge-Coupled Device) octal detector.

Data processing, model building, and refinement of SFX datasets
During the SFX experiments, data acquisition was tracked with the cctbx.xfel graphical user interface (49), which monitors for new data and submits processing jobs to the computing cluster. The jobs run the core program dials.stills_process to index and integrate the images while providing real-time feedback, as part of the larger cctbx. xfel and DIALS processing suite (50)(51)(52)(53)(54)(55). A first estimate of the detector position (distance and beam center) was obtained from a powder diffraction pattern of silver(I) behenate (Alfa Aesar). After initial spot finding of strong reflections, followed by indexing of strong reflections and integration using dials.stills_process, a round of metrology refinement was done (56). A second round of indexing and integration was performed with the refined detector position. Initial merging was performed using PRIME (57) followed by molecular replacement with PHASER (58). This provided a reference model using a target unit cell of a = 41.9 Å, b = 75.7 Å, c = 102 Å,  =  =  = 90° (space group P 2 1 2 1 2 1 ). After correction of the integrated intensities for absorption by the Kapton conveyor belt, final data integration and merging were performed using cxi.merge (21). Justification of the resolution cutoff for the merged data was determined on the basis of multiplicity in the highest-resolution shell (>10-fold) and on CC 1/2 (monotonic decrease) (see tables at the end of the Supplementary Materials with the merging statistics justifying the resolution cutoff) (20,23).
Structures were solved by isomorphous molecular replacement using the reported structural data file of IPNS [PDB: 1BLZ (26)] as a search model. All structures were iteratively fitted and refined using PHENIX (59) and Coot (60). Processing and refinement statistics for all anaerobically and O 2 -exposed IPNS structures are given in table S2.

XES analysis
The on-demand ADE/tape drive sample delivery setup enables the simultaneous collection of tr-XES and tr-SFX (23) from the same sample and x-ray pulse to study the Fe oxidation state. We used a wavelength-dispersive von Hamos spectrometer with four cylindrically bent (R = 250 mm) germanium (440) crystals orthogonal to the sample with the center of the crystals located at 75.41° with respect to the interaction point and each focusing the emitted Fe K 1,2 x-ray photons onto an ePix100 detector located below the sample x-ray interaction region. Calibration was performed using aqueous solutions of 10 mM Fe(III)(NO 3 ) 3 and Fe(II)Cl 2 (anaerobically prepared and measured) as references. Detector images were sorted by the sample hit rate as described (25). Briefly, for anaerobic, 400-, 500-, 800-, and 1600-ms datasets, two threshold values were set as 3 and 2 for thresholds I and II, respectively. For the 3000-ms dataset, the value for threshold II, which compares the number of photons per unit area inside and outside the region of interest, was reduced to 1. A linear function interpolated across two points outside the region of interest was subtracted from the images for background correction. For spectral analysis, the spectra were area-normalized in the 6380-to 6415-eV range. In this range, the recorded energy was initially spread out by 0.146-to 0.159-eV intervals. To create equal intervals and apply smoothing, the spectra were first linearly splined to a 0.001-eV interval and then processed by a second-order Savitzky-Golay filter with a 3003-point window size. This window size covers approximately the same range of points as a 19-point window applied to the original intervals. Difference spectra were calculated after smoothing. For calculation of FWHM, the spectral range for K 1 was selected as 6396 to 6407 eV. Error bars for the calculated FWHM were obtained by a bootstrapping procedure as described (25).
Exposure of anaerobically obtained single IPNS:Fe:ACV crystals to O 2 IPNS:Fe:ACV crystals (needle morphology) were grown anaerobically to the appropriate size [~5 × 5 × (100 to 150) m 3 ] (27); the anaerobic microcrystal slurry (10 l) was transferred into a 500-l tube. A second 500-l tube was filled with the IPNS crystallization buffer [100 l, 1.7 M Li 2 SO 4 , and 0.1 M tris (pH 8.5) supplemented with 20% glycerol (v/v)], and the solution was saturated with O 2 . The anaerobic crystal slurry (2 l) was removed from the glovebox and immediately mixed with the O 2 -saturated crystallization solution (8 l). The samples were incubated at room temperature and at varying timepoints (30 s to 10 min). Single crystals were mounted on nylon loops and then cryo-cooled by rapid plunging into liquid N 2 . Data for a total of ~100 cryo-cooled single crystals of IPNS:Fe:ACV exposed to O 2 were collected at the MX beamlines of the Diamond Light Source, UK. From the IPNS:Fe:ACV:O 2 datasets, one representative structure was selected for deposition (PDB: 6ZAP), which was refined to 1.36-Å resolution. In this complex, both confs. A and B of 3 are observed and were refined in 50:50 occupancy.