Structures of the sulfite detoxifying F420-dependent enzyme from Methanococcales

Methanogenic archaea are main actors in the carbon cycle but are sensitive to reactive sulfite. Some methanogens use a sulfite detoxification system that combines an F420H2-oxidase with a sulfite reductase, both of which are proposed precursors of modern enzymes. Here, we present snapshots of this coupled system, named coenzyme F420-dependent sulfite reductase (Group I Fsr), obtained from two marine methanogens. Fsr organizes as a homotetramer, harboring an intertwined six-[4Fe–4S] cluster relay characterized by spectroscopy. The wire, spanning 5.4 nm, electronically connects the flavin to the siroheme center. Despite a structural architecture similar to dissimilatory sulfite reductases, Fsr shows a siroheme coordination and a reaction mechanism identical to assimilatory sulfite reductases. Accordingly, the reaction of Fsr is unidirectional, reducing sulfite or nitrite with F420H2. Our results provide structural insights into this unique fusion, in which a primitive sulfite reductase turns a poison into an elementary block of life.

Methanogenic archaea are main actors in the carbon cycle but are sensitive to reactive sulfite. Some methanogens use a sulfite detoxification system that combines an F 420 H 2 -oxidase with a sulfite reductase, both of which are proposed precursors of modern enzymes. Here, we present snapshots of this coupled system, named coenzyme F 420 -dependent sulfite reductase (Group I Fsr), obtained from two marine methanogens. Fsr organizes as a homotetramer, harboring an intertwined six-[4Fe-4S] cluster relay characterized by spectroscopy. The wire, spanning 5.4 nm, electronically connects the flavin to the siroheme center. Despite a structural architecture similar to dissimilatory sulfite reductases, Fsr shows a siroheme coordination and a reaction mechanism identical to assimilatory sulfite reductases. Accordingly, the reaction of Fsr is unidirectional, reducing sulfite or nitrite with F 420 H 2 . Our results provide structural insights into this unique fusion, in which a primitive sulfite reductase turns a poison into an elementary block of life.
When cold seawater permeates through sediments or enters hydrothermal vent walls, a partial oxidation of sulfide (HS − , S 2− ) results in the formation of (bi)sulfite (HSO 3 − ), SO 3 2− , a highly reactive intermediate of the sulfur cycle 1 . Methanogenic archaea are extremely sensitive to this strong nucleophile, which results in the collapse of methanogenesis, their central energy metabolism 2 . Despite its toxic effects, many hydrogenotrophic methanogens thrive in environments where they are exposed to fluctuating SO 3 2− concentrations, especially methanogens living in proximity to hydrothermal vents or in geothermally heated sea sediments 3-6 .
When exposed to SO 3 2− , the hyperthermophile Methanocaldococcus jannaschii 3 expresses high amounts of the Group I coenzyme F 420 -dependent sulfite reductase (referred to as MjFsr), which confers not only protection, but also the ability to grow on SO 3 2− as sole sulfur source (for example, in the absence of S 2− ) 5,7 . Because of this trait, the fsr gene has been used as a genetic marker 7,8 .
Fsr is composed of an N-terminal half belonging to the F 420 -reducing hydrogenase β-subunit family (FrhB; Supplementary   Fig. 1) and a C-terminal half made of a single sulfite/nitrite reductase repeat 5,9 (S/NiRR, from here on referred to as sulfite reductase domain). All known sulfite reductases reduce SO 3 2− using a magnetically coupled siroheme-cysteine-[4Fe-4S] center 10 . This metallocofactor is also used by nitrite reductases to reduce nitrite (NO 2 The F 420 H 2 -oxidase domain flanks a sulfite reductase core A single-wavelength anomalous dispersion experiment was performed to solve the MjFsr crystal structure. MtFsr was solved by molecular replacement, using MjFsr as a template. The crystal structures of both Fsr superpose well (Extended Data Fig. 3a) and were refined to 2.30 Å for MjFsr and 1.55 Å for MtFsr ( Fig. 1 and Extended Data Table 1). Since MjFsr has pseudo-merohedral twinning and a lower resolution compared to MtFsr, the latter was used for the in-depth structural and biochemical analysis.
As shown in Fig. 1, Fsr is organized as follows: the N-terminal ferredoxin domain (MtFsr residues 1-57 containing two [4Fe-4S] clusters) is linked to the F 420 H 2 -oxidase domain (MtFsr residues 58-336, harboring the flavin and one [4Fe-4S] cluster), which is connected to the C-terminal sulfite reductase domain (MtFsr residues 339-484, 546-618) that binds the siroheme-[4Fe-4S] and has an inserted ferredoxin domain (MtFsr residues 485-545, containing two [4Fe-4S] clusters). The tetrameric structure of the protein is established by a dimer of two homodimers over a large contact area through the two additional reduction step and transfer the sulfur species intermediate to the sulfur-carrier protein DsrC used for energy conservation (Extended Data Fig. 1) 15 . In the absence of DsrC, DsrAB releases some S 2− , as well as the reaction intermediates trithionate and thiosulfate [15][16][17] .
Structural and evolutionary studies suggest that aSirs and dSirs originated from a common progenitor 12,14 , a primitive Sir that contained a catalytic siroheme-[4Fe-4S] and was operating by itself. The gene encoding this ancestral enzyme was duplicated, and in the dSir case, the duplicated version evolved into DsrB, while DsrA was retained for structural function. In the case of aSir, the original and duplicated genes fused and only one active siroheme-[4Fe-4S] was retained. On the basis of sequence and phylogenetic analyses, it has been suggested that fsr evolved before the duplication event and therefore represents a primordial sulfite reductase 5,18,19 . Alternatively, fsr could have arisen through lateral gene transfer followed by gene fusion events.
Besides its evolutionary importance, the electron-donor module of Fsr, the F 420 H 2 -oxidase, is directly fused to its sulfite reductase domain. This fusion allows the enzyme to perform the entire six-electron reduction of SO 3 2− on its own via an unknown electronic relay, using electrons from reduced F 420 . The coenzyme F 420 is a deazaflavin derivative present at high cytoplasmic concentrations in methanogens 5,[20][21][22] and can be reduced by the F 420 -reducing hydrogenase (FrhABG; Supplementary  Fig. 1). Due to the difference in the redox potentials of the F 420 /F 420 H 2 (∆E 0 ′ = −350 mV) and HSO 3 − /HS − (∆E 0 ′ = −116 mV) couples, the overall   Supplementary Fig. 3), which is similarly bound in Fsr and FrhB ( Supplementary Fig. 4). No electron density could be found despite cocrystallization with F 420 H 2 (see Methods). Nevertheless, the reduced F 420 -binding site is presumably located in a positively charged cleft that would complement the charges of the acidic gamma-carboxy groups ( Supplementary Fig. 3c) 25,26 .

A [4Fe-4S] cluster relay connects both active sites
The distance between the isoalloxazine ring from the FAD to the closest siroheme-[4Fe-4S] is approximately 40 Å. Electrons delivered by reduced F 420 must therefore travel through an electron-transfer relay of metallocofactors. The first part of this relay, located in the N-terminal ferredoxin and F 420 H 2 -oxidase domains, shares high structural homologies with FrhBG. Indeed, FrhG and the N-terminal ferredoxin domain of Fsr are located at the same position of the F 420 -oxidoreductase domain (Fig. 2b,c), resulting in a similar electron relay. This homology suggests a common origin that may have evolved by fusion (for Fsr) or by duplication and fusion (for FrhG).
As illustrated in Fig. 3, the overall electronic path consists of five [4Fe-4S] clusters connected by short edge-to-edge distances (<11.5 Å). Dimerization is critical because half of the relay is provided by the second protomer. An intraelectron transfer between both Fsr dimers is unlikely due to the long distance between the nearest clusters (that is, 18.9 and 19.5 Å).
The electrons on the isoalloxazine ring can be transferred directly to the [4Fe-4S] cluster 1, which is located in the F 420 H 2 -oxidase domain. Sequence analyses indicated four [4Fe-4S] clusters and the one coupled to the siroheme 5 . But both Fsr structures revealed an additional cluster ([4Fe-4S] cluster 1), which has a noncanonical binding sequence (PCX 40 CX 54 CX 2 C). Strikingly, the four predicted clusters have completely different binding residues compared to primary structural analysis (Extended Data Fig. 5). Each [4Fe-4S] cluster has a divergent protein environment: cluster 1 is surrounded by basic residues; clusters 2 and 5 have a hydrophobic shell; clusters 4 and 6 are in a more polar environment; and cluster 3 has a glutamate ligand. These differences may reflect the need to establish a 'redox potential ladder' to allow a smooth one-way transfer of electrons. To investigate the electron-transfer path, electrochemical experiments followed by EPR spectroscopy were performed.

Redox properties of the metallocofactors
EPR spectroscopy at 10 K (Extended Data Fig. 6a-d) revealed that in as-isolated MtFsr high-spin (S = 5/2) and low-spin (S = 1/2) signals typical for the siroheme in sulfite reductases 27,28 were absent, neglecting the sharp axial S = 5/2 EPR signal around g = 6, which, quantified by double integration of its simulation spectrum (g = 6.22, 5.92 and 1.98), is at most 3% of MtFsr. Apparently, on purification under strictly anaerobic conditions, the siroheme remains in its ferrous state. After methylene blue oxidation or on dye-mediated redox titration with E m,7.5 = −104 mV (all potentials refer to potentials versus the H 2 /H + normal hydrogen electrode) an intense rhombic S = 5/2 EPR signal with g = 6.7 and 5.1 appeared (Fig. 4a,b). The spectrum could be simulated with three components: Article https://doi.org/10.1038/s41589-022-01232-y a main species with g = 6.70 and 5.10 (78%), a less abundant species (19%) with g = 6.80 and 5.08, but narrower linewidth, and the sharp axial g = 6 species already seen in as-isolated MtFsr. For both rhombic components g = 1.95 was taken as the third g value, as the experimental spectrum contained a weak [3Fe-4S] 1+ signal from limited [4Fe-4S] 2+ breakdown upon oxidation. In sulfite reductase and other hemoproteins multiple high-spin species are common 29 . Addition of SO 3 2− to methylene blue-oxidized MtFsr led to disappearance of the siroheme ferric high-spin signals and formation of a weak low-spin EPR signal, of which only the highest g value (2.8) was detectable, as in other sulfite reductases 30 .
In an enzyme approaching the complexity of the complex I, it is not feasible to determine all individual redox potentials of its five regular [4Fe-4S] 1+/2+ cubanes and the siroheme-bridged cubane. First, on the basis of distances in Fsr, extensive magnetic coupling 31 between neighboring cubanes is anticipated, blurring individual EPR features. Second, the coupling between the ferrous siroheme and its cysteine-bridged reduced cubane leads to complex mixtures of sharp g = 1.94, broader g = 2.29 and very anisotropic S = 3/2 mimicking signals 32 . Third, we had to avoid sodium dithionite inherently containing SO 3 2− and therefore used sodium borohydride-reduced F 420 , while following the solution potential with mediators. One [4Fe-4S] 1+/2+ cubane with simulated g values of 2.064, 1.927 and 1.85 was reduced at a relatively high potential and is also detected in as-isolated Fsr (Fig. 4a,c). From the amplitude of the second derivative of the experimental EPR spectrum at g = 2.064, E m,7.5 = −275 mV was estimated from fitting to the Nernst equation with n = 1 (Fig. 4c). The signal 'disappeared' on further reduction with E m,7.5 = −350 mV in a manner indicating cooperativity (n = 2). As super-reduction to [4Fe-4S] 0 is unlikely (E m = −790 mV (ref. 33)), we interpret this phenomenon as reduction of two neighboring clusters of the g = 2.064 cluster. This cluster thus is number 2, 3 or 4′ (the siroheme cubane typically has a very low potential 27 ). In the absence of sufficiently differing EPR features below −350 mV we double integrated the EPR spectra. On the basis of iron content divided by 24 (siroheme does not release Fe ions in acid) we quantified 4.5 ± 0.5 spin/subunit at the lowest attainable potential (−526 mV), which most likely corresponds to the five regular clusters. A fit for the spin integral as a function of the redox potential included the experimental E m,7.5 = −275 mV and E m,7.5 = −350 mV for both neighboring clusters. Avoiding overfitting, we could satisfactorily reproduce the data for five redox transitions with three midpoint potentials: one at E m,7.5 = −275 mV (experimental), one at a low potential to represent the lowest potential region (E m,7.5 = −435 mV) and three times E m,7.5 = −350 mV for the other three clusters (which includes the two clusters leading to broadening of the g = 2.064 signal).
In the low-field region, a species with unusual g values was detected (simulated g values 5.05, 3.05 and 1.96) at very low potential (Fig. 4d). It was accompanied in some samples by an isotropic g = 4.3 signal. But, since the integrated intensity was maximally 5% of the g = 5.05 species and non-Nernstian behavior was seen, it was not considered physiologically relevant. It has previously been shown that such a g = 5.05 species is not from a S = 3/2 system but from transitions of the siroheme-Fe 2+ exchange coupled to [4Fe-4S] 1+ ( J/D ≈ −0.2 and E/D ≈ 0.11, in which J, D and E are the effective Heisenberg exchange coupling parameter and the spin Hamiltonian zero-field splitting parameters of the spin quintet, respectively; Extended Data Fig. 6c) 32 . In full agreement with findings on the Escherichia coli assimilatory reductase 27 a very low potential (E m,7.5 = −445 mV) was estimated.

A prototypical sulfite reductase
The C-terminal domain of Fsr represents the simplest sulfite reductase crystallized so far. While Fsr shares the common fold of sulfite reductases (Extended Data Fig. 7a and Supplementary Fig. 6) 9,13,14 , it lacks the large N-and C-terminal extensions found in aSirs and dSirs, which presumably serve to strengthen dimerization and to interact with partners 34 (Fig. 5a-c). Without these extensions, Fsr is much more compact-possibly a thermophilic trait. Each Fsr protomer contains one functional siroheme center. In comparison, dSirs harbor one functional and one structural siroheme center in each DsrAB heterodimer, while aSirs have lost one siroheme-[4Fe-4S] site (Extended Data Fig. 7b-d).
Although Fsr is phylogenetically more distant from aSirs than from dSirs, it superposes well with the first and second halves of aSirs The inserted ferredoxin domain in Fsr is at the same position as the ferredoxin domain in DsrA or DsrB (Extended Data Fig. 7a,c,d and Supplementary Figs. 9 and 10). There is a remarkable three-dimensional conservation of the electron connectors between Fsr, DsrA, DsrB and even the aSir from Zea mays, where the external [2Fe-2S] ferredoxin sits on the core of the sulfite reductase 35 (Fig. 5a-c). Such a conserved position suggests a common origin, but could also be due to the restricted access of the [4Fe-4S]-siroheme and the selection pressure towards an optimized distance for electron transfer.

Fsr has traits of assimilatory sulfite reductases
While the sirohemes of DsrAB are partially surface exposed to interact with DsrC (Extended Data Fig. 1) 13 , the Fsr sirohemes are buried but still The binding of the siroheme in MjFsr and MtFsr is highly conserved. It is mainly anchored by positively charged residues from one protomer, while the dimeric partner binds the adjacent [4Fe-4S] cluster establishing the siroheme-[4Fe-4S] center, as reported for other sulfite reductases 14 . On the basis of the observed electron density, we tentatively modeled a SO 3 2− bound to the siroheme iron (2.3 Å; Extended Data Fig. 8b) in MjFsr. In MtFsr, the axial ligand is a single atomic species at all sites of the asymmetric unit, which is in proximity but not covalently bound to the iron (2.9 Å; Extended Data Fig. 8c). The anion HS − was modeled in the electron density based on the pH 5.5 in the crystallization solution. This species could be the result of cocrystallizing Fsr with reduced F 420 , which might have forced the complete reduction of bound SO 3 2− . In MjFsr, four positively charged residues (Arg 355, Arg 423, Lys 460 and Lys 462), which are perfectly conserved across sulfite reductases (Fig. 5d,e and Supplementary Figs. 7 and 9), bind the SO 3 2− and two water molecules. In MtFsr, the modeled HS − is bound by Arg 423, Lys 460 and Lys 462, and one water molecule is stabilized by Arg 355 (Fig. 5f). Group II Fsr found in the genome of anaerobic methanotrophic archaea 6 (except for 'Candidatus Methanoperedens nitroreducens') and Methanosarcinales, should have a larger binding pocket and two arginines of Group I Fsr are replaced by a lysine and glycine. This suggests that the functionally uncharacterized Group II Fsr has a different substrate specificity 6,17 . Interestingly, the second isoform found in M. thermolithotrophicus harbors one arginine but exchanged the other one for a threonine (Thr 438; Supplementary Fig. 2), indicating an alternative physiological function.
The active site of Fsr shows the same traits as an assimilatory sulfite reductase: an arginine at position 388, and the coordination of the siroheme-coupled [4Fe-4S] cluster by the canonical motif (CX 5 CX n CX 3 C; Fig. 5d,f). In comparison, DsrAs contain a conserved threonine where aSirs have arginine (αThr 136 in Desulfovibrio vulgaris and αThr 133 in Archaeoglobus fulgidus) and the catalytically active [4Fe-4S] cluster coupled to the siroheme of DsrBs is coordinated by the canonical motif CX n CCX 3 C (Fig. 5e). Fsr must therefore follow the same catalytic path as aSirs; the six-electron reduction of SO 3 2− to S 2− should be unidirectional, without the formation or consumption of intermediates (for example, thiosulfate or trithionate). MtFsr did not accept thiosulfate as an electron acceptor, which is in agreement with the findings for MjFsr 5 . We also monitored F 420 -reduction by MtFsr with S 2− as substrate (up to 10 mM) and observed no reaction. The addition of 10 mM S 2− to 1.4 mM of Na 2 SO 3 also had no effect on the F 420 H 2 oxidation rate. Taken together, these results support that Fsr indeed acts like an aSir.
On the basis of its equal V max but six-fold lower K m value (Table 1), MtFsr prefers NO 2 − over SO 3 2− , a property that may expand its role from sulfite detoxification to ammonium production, as  ) in vitro with a relative activity of 20.7 ± 7.5% compared to SO 3 2− (see Methods). These promiscuous activities could expand the physiological range of the enzyme, but also its biotechnological applications.

Discussion
Some methanogens show a remarkable tolerance to SO 3 2− , one of the sulfur-reactive species that can cause oxidative damage to the methanogenic machinery. Besides the possibility that those methanogens can keep low intracellular SO 3 2− concentrations through pumping mechanisms, the cytoplasmic Group I Fsr is used as a first line of defense to convert toxic SO 3 2− into HS − , which can then be used for sulfur assimilation. The efficient SO 3 2− detoxification strategy of Methanococcales relies on the enormous amount of expressed Fsr, which constitutes 5-10% of the cellular protein (Extended Data Fig. 2b,c and Methods), but also on the use of abundant F 420 H 2 , which can be rapidly regenerated via H 2 oxidation by Frh 22 .
Fsr discloses a 'cofactor swapping' between two subunits forming a homodimer in a head-to-tail configuration, which dimerizes with a second homodimer, creating a butterfly-shaped tetramer. As a result, the centrally located sulfite reductase domains are surrounded by F 420 H 2 -oxidase domains. These shuttle electrons via three [4Fe-4S] cluster from one subunit to the other two [4Fe-4S] cluster and the siroheme-cysteine-[4Fe-4S] cofactor of the other subunit within the functional dimer. In contrast to the bidirectional hydrogenase Frh, which maintains an isopotential of E′ 0 ≈ −400 mV (ref. 25), the different metalloclusters of Fsr must establish a downhill redox potential from the FAD to the siroheme-[4Fe-4S]. Our electrochemical and spectroscopic studies indicate that the electrons carried by F 420 H 2 are immediately transferred to the siroheme-[4Fe-4S] (Fig. 4a,b and Extended Data Fig. 6a). The metallocofactors should ensure efficient electron transfer rather than serving as a transient storage, and a cascade of redox potential from −380 mV (F 420 /F 420 H 2 redox potential under certain physiological conditions 22    Article https://doi.org/10.1038/s41589-022-01232-y perform a two-electron reduction to allow the transfer of the sulfur intermediate to DsrC. In contrast, aSirs and Fsr perform a three times two-electron reduction to release HS − . A positively charged environment around the active site attracts SO 3 2− and an organized water network has been proposed to provide fast proton transfer via the Grotthuss mechanism, allowing successive SO 3 2− reduction (Extended Data Fig. 8a) 16,37 . Despite a strikingly similar position of the residues involved in substrate binding, aSirs/Fsr and dSirs react differently. With the possibility of genetically modifying M. maripaludis or M. jannaschii, it would be worthwhile to exchange the residues that confer aSir traits at the active site (Arg 388, Cys 428) with dSir ones and observe the effects on the phenotype 7,8 .
Throughout evolution, sulfite reductases have been kept to detoxify SO 3 2− as well as to conserve energy by dissimilatory SO 3 2− reduction or oxidation of H 2 S 38 . Based on sequence and structural similarity with enzymes from different superfamilies, it has been proposed that modern sulfite reductases originated from a primordial Sir/Nir that functioned as a self-complementary homodimer 18 . A snapshot of this progenitor can be derived from the Fsr structure, as the organization of its sulfite reductase domain is highly simplified (Extended Data Fig. 9). The evolution of Fsr is still a matter of debate but it needs to be thoroughly studied, as its discovery has reinforced the question of whether sulfate respiration or methanogenesis was the primeval means of energy conservation during the evolution of early Archaea 39,40 . Both metabolisms, related to each other, possibly coexisted or even coexist still 6,18,41 . Methanogens might have lost the genes required for complete sulfate dissimilation over time, but kept the sulfite reductase to adapt to environments where SO 3 2− fluctuations do occur. However, M. thermolithotrophicus appears to use a complete sulfate-reduction pathway, as it is able to grow on sulfate as its sole sulfur source 4 . This assimilation pathway requires SO 3 2− as an intermediate, and Fsr is expected to orchestrate its reduction. Although further studies need to investigate whether this methanogen can also express other enzymes of the sulfate-reduction pathway, the structural elucidation of Fsr provides the first snapshot of a sulfate reduction-associated enzyme in a methanogen.

Online content
Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41589-022-01232-y.

Reagents used for this study
Lists of reagents and providers are provided in Supplementary Table 1. The cultivation media were transferred in a 1 l pressure-protected Duran laboratory bottle with a magnetic stirring bar. The Duran flask was closed with a butyl rubber stopper and degassed by applying 3 min of evacuation, followed by 30 seconds of ventilation with 1 × 10 5 Pa N 2 atmosphere, under constant magnetic stirring. This was repeated 15 times and at the final ventilation step an overpressure of 0.3 × 10 5 Pa N 2 was applied.

Trace element composition
A 100-fold-concentrated trace element solution was prepared by first dissolving 1.36 g nitrilotriacetic acid (7.1 mM) in 800 ml dH 2

Anaerobic growth of Methanococcales
For all studied archaea, cell growth was measured spectrophotometrically by measuring the optical density at 600 nm (OD 600 ). To control the purity of the culture, samples were taken and analyzed via light microscopy. Both methanogens were cultivated at 65 °C, unless stated otherwise, with 1 × 10 5 Pa of H 2 /CO 2 in the gas phase. M. jannaschii was cultivated in flasks and M. thermolithotrophicus was cultivated in flasks or a fermenter.

Growth of M. jannaschii
Duran bottles (10× 1 liter) were sealed with butyl rubber stoppers and the gas phase was exchanged for H 2 /CO 2 (80:20, 1 × 10 5 Pa). A 100-ml portion of anaerobic cultivation media was transferred into each bottle (ratio 1:10 of medium/gas phase), with 1 mM Na 2 SO 3 as a sole sulfur source. A portion of 5 ml of overnight culture (OD 600 of 0.9) was used as an inoculum for 100 ml media. No additional reductant was added. The cultures were placed at 65 °C, with standing for at least one hour, followed by overnight shaking at 180 rotations per minute without light. The cells were collected in exponential phase with a final OD 600 of 1.83 by immediately transferring them in an anaerobic tent (N 2 /CO 2 atmosphere at a ratio of 90:10), followed by anaerobic centrifugation for 30 min at 6,000g at 4 °C. The cell pellet was transferred in a sealed bottle gassed with 0.3 × 10 5 Pa N 2 and flash frozen in liquid N 2 to be stored at −80 °C.

Growth of M. thermolithotrophicus for Fsr crystallization
M. thermolithotrophicus was grown in a fermenter at 50 °C with 10 mM sulfate (SO 4 2− ) as sole sulfur substrate. Since SO 3 2− could be an intermediate in the SO 4 2− reduction pathway it would require the expression of Fsr. Therefore, 1.5 l of anaerobic cultivation medium with 10 mM SO 4 2− were continuously bubbled with H 2 and CO 2 (80:20, 2 × 10 4 Pa) and inoculated with 100 ml preculture (OD 600 of 4.2). Since the fermenter is an open system, we set a more alkaline pH (7.6) to prevent evaporation of produced S 2− . Here, it should predominantly be present in the form of HS − , and not H 2 S, and therefore stay for longer time in the medium. The pH was checked every two hours by using a pH indicator. The cells were grown until late exponential phase (OD 600 of 2.97) and then immediately transferred in an anaerobic tent (N 2 /CO 2 atmosphere at a ratio of 90:10). Cells were collected by anaerobic centrifugation for 30 min at 6,000g at 4 °C. A 1.5-l culture with an OD 600 of 2.97 yielded 19.25 g of cells (wet weight). The cell pellet was transferred in a sealed bottle, gassed with 0.3 × 10 5 Pa N 2 , flash frozen in liquid N 2 and stored at −80 °C.

Growth of M. thermolithotrophicus for Fsr activity assays
To perform enzymatic activity assays, M. thermolithotrophicus was directly grown on 2 mM Na 2 SO 3 . The ten 1-l Duran bottles were sealed with butyl rubber stoppers and the gas phase was exchanged for H 2 and CO 2 (80:20, 1 × 10 5 Pa). A 100 ml of anaerobic cultivation media containing 50 mM MES at pH 6.2 was transferred in each bottle (ratio of 1:10 of medium/gas phase), with 2 mM Na 2 SO 3 final as a sole sulfur source. A 5-ml portion of overnight-grown culture (OD 600 of 1.7) was used as an inoculum for 100 ml of media. No additional reductant was added. The cultures were placed at 65 °C, with standing overnight. The cells were grown until early exponential phase (OD 600 of 0.8), since we assumed that most SO 3 2− has not been converted into HS − yet and that Fsr should be highly expressed and active. The cells were immediately collected by transferring them in an anaerobic tent (N 2 /CO 2 atmosphere at a ratio of 90:10), followed by anaerobic centrifugation for 30 min at 6,000g at 4 °C. The cell pellet was transferred in a sealed bottle, gassed with 0.3 × 10 5 Pa N 2 , flash frozen in liquid N 2 and stored at −80 °C.

Sulfite growth inhibition
M. thermolithotrophicus was grown on different Na 2 SO 3 concentrations to determine the growth-inhibiting threshold. For this, 250-ml serum flasks were sealed with a butyl rubber stopper and the gas phase was exchanged for H 2 and CO 2 (80:20, 1 × 10 5 Pa). A 10-ml portion of anaerobic cultivation media with a pH set at 6.2 with 50 mM MES was transferred into each bottle. Then, different Na 2 SO 3 concentrations (2 mM, 10 mM, 20 mM, 30 mM and 40 mM final) were added in triplicate as a sole sulfur source, and 2 mM Na 2 S was used as a control. The cultures grew at 65 °C for 22 hours, with standing. The three biological replicates for each setup are represented as dots in Extended Data Fig. 2a, with the standard deviation shown as bars.

Growth of M. thermolithotrophicus for titrations and EPR spectroscopy
Due to the high demand of MtFsr for titration and EPR spectroscopy experiments, M. thermolithotrophicus was grown in one 10-l fermenter with SO 4 2− as a sole sulfur substrate and in another 10-l fermenter with SO 3 2− as a sole sulfur source, to boost MtFsr natural expression. The fermenter containing SO 4 2− was performed as described above with an inoculum of 350 ml (OD 600 of 3.2). A 7.4-l culture with an OD 600 of 4.8 yielded 74 g of cells (wet weight). In the SO 3 2− fermenter, M. thermolithotrophicus was grown at 50 °C in 7 l anaerobic cultivation medium with a pH of 6.2 supplemented with 5 mM SO 3 2− as a sole sulfur substrate, continuously bubbled with H 2 and CO 2 (80:20, 2 × 10 4 Pa). A 600-ml Article https://doi.org/10.1038/s41589-022-01232-y preculture (OD 600 of 2.34) was used as inoculum. The cells were grown until an OD 600 of 2.48 and then immediately transferred in an anaerobic tent (N 2 /CO 2 atmosphere at a ratio of 90:10). Cells were collected by anaerobic centrifugation for 30 min at 6,000g at 4 °C and a final yield of 51 g of cells (wet weight) was obtained. The cell pellets were transferred in a sealed bottle, gassed with 0.3 × 10 5 Pa N 2 , flash frozen in liquid N 2 and stored at −80 °C.

Genome sequencing of M. thermolithotrophicus
M. thermolithotrophicus was anaerobically grown in the above-described medium and 2 mM Na 2 S was used as a sulfur source. A total culture volume of 20 ml was used. Cells were aerobically collected by centrifugation (30 min, 6,000g at 4 °C). DNA was extracted and purified based on ref. 43. Quality control, library preparation and sequencing (PacBio Sequel II) were performed in the Max Planck-Genome-Centre (Cologne).

Purification of Fsr
All steps were performed under the strict exclusion of oxygen and daylight. Protein purifications were carried out in a Coy tent with an N 2 and H 2 atmosphere (97:3) at 20 °C under yellow light. For both Fsr, three to five chromatography steps were used with some variations. Fsr purification was further followed via activity assays and on the basis of absorbance peaks at wavelengths of 280, 420 and 595 nm. Each elution profile was systematically controlled by SDS-PAGE to select the purest fractions.

Purification of MjFsr
M. jannaschii cells (13.5 g wet weight) were thawed under warm water and transferred in an anaerobic tent (N 2 /CO 2 atmosphere at a ratio of 90:10). Cells were diluted by three volumes of lysis buffer (50 mM Tricine/NaOH pH 8.0, 2 mM dithiothreitol (DTT)) and disrupted by sonication: 7 cycles at 62% intensity with 30 pulses followed by 1 min break (probe MS76, SONOPULS Bandelin). Cell debris was removed anaerobically via centrifugation (21,000g, one hour, room temperature). The protein concentration (measured by Bradford) of the supernatant was estimated to 4.68 mg ml −1 . The supernatant was transferred to a Coy tent (N 2 /H 2 atmosphere of 97:3) under yellow light at 20 °C. The sample was diluted with two volumes of lysis buffer and passed through a 0.2-µm filter (Sartorius). The filtered sample was loaded on a 10-ml Q Sepharose high-performance column (GE Healthcare), which was previously equilibrated with 5 column volumes (CV) of lysis buffer. The column was then washed with 2 CV of lysis buffer. MjFsr was eluted by a gradient of NaCl (from 0.1 to 0.

Purification of MtFsr for crystallization
Cells (19.25 g wet weight) derived from a fermenter were thawed under warm water and transferred to an anaerobic tent containing an atmosphere of N 2 /CO 2 (90:10). Cells were lysed by osmotic shock through the addition of 60 ml lysis buffer (50 mM Tricine/NaOH pH 8.0, 2 mM DTT). Cell lysate was homogenized by sonication: 3 cycles at 70% intensity with 30 pulses followed by 1 min break (probe MS76, SONOPULS Bandelin) and cell debris was removed anaerobically via centrifugation (21, The column was then washed with 2 CV of HIC buffer. A gradient of (NH 4 ) 2 SO 4 ranging from 2 to 1 M was performed for 30 min at a flow rate of 0.8 ml min −1 with a fractionation size of 1 ml. MtFsr eluted between 1.38 and 1.23 M (NH 4 ) 2 SO 4 and the respective fractions were pooled. The buffer was exchanged for the storage buffer (25 mM Tris-HCl pH 7.6, containing 10% v/v glycerol and 2 mM DTT) by using a 30-kDa-cutoff filter (6 ml, Merck Millipore) and MtFsr was concentrated to 11.06 mg ml −1 in a volume of 120 µl. The protein concentration was estimated by the Bradford method. The sample was immediately crystallized.

SO 3
2− -grown cells (8 g wet weight) were thawed under warm water and transferred to an anaerobic tent containing an atmosphere of N 2 / CO 2 (90:10). Cells were lysed by osmotic shock through the addition of 60 ml lysis buffer (50 mM Tricine/NaOH pH 8.0, 2 mM DTT). Cell lysate was homogenized by sonication: 9 cycles at 75% intensity with 30 pulses followed by 1 min break (probe KE76, SONOPULS Bandelin) and cell debris was removed anaerobically via centrifugation (21,000g, one hour at 4 °C). The supernatant was transferred to a Coy tent (N 2 / H 2 atmosphere of 97:3) under yellow light at 20 °C and was diluted with 90 ml lysis buffer, filtered through a 0.2-µm filter. The filtered sample was applied to a 10-ml DEAE fast-flow column (GE Healthcare), which was previously equilibrated with lysis buffer. The column was then washed with 2 CV of lysis buffer. A gradient of 0.1 to 0.6 M NaCl was applied for 120 min at a flow rate of 2. HAP buffer (20 mM K 2 HPO 4 /HCl pH 7.0 and 2 mM DTT). The filtered sample was applied to a 10-ml hydroxyapatite type 1 (Bio-Scale Mini CHT cartridges, BioRad) equilibrated with HAP buffer. The column was washed with 2 CV of HAP buffer and a gradient of 0.02 to 0.5 M K 2 HPO 4 in 60 min at a flow rate of 2 ml min −1 was performed. Fraction sizes of 1.5-ml were collected. MtFsr eluted between 0.25 and 0.42 M K 2 HPO 4 and the respective fractions were pooled. The pool was diluted with 3 volumes of HIC buffer (25 mM Tris-HCl pH 7.6, 2 M (NH 4 ) 2 SO 4 and 2 mM DTT). The filtered sample was applied to a Source15Phe 4.6/100 PE column (GE Healthcare) previously equilibrated with the HIC buffer. The column was then washed with 2 CV of 25 mM Tris-HCl pH 7.6, 1.6 M (NH 4 ) 2 SO 4 and 2 mM DTT buffer. MtFsr was eluted in a gradient of 1.6 to 0.8 M of (NH 4 ) 2 SO 4 in 25 min at a flow rate of 0.8 ml min −1 and a fractionation size of 1 ml. MtFsr eluted between 1.43 and 1.28 M (NH 4 ) 2 SO 4 and the respective fractions were pooled. The buffer was exchanged for the storage buffer (25 mM Tris-HCl pH 7.6, containing 10% v/v glycerol and 2 mM DTT) by using a 30-kDa-cutoff filter (6 ml, Merck Millipore) and MtFsr was concentrated to 900 µl. The concentrated sample was passed onto a Superdex 200 Increase 10/300 GL (GE Healthcare), equilibrated in storage buffer. MtFsr eluted at a flow rate 0.4 ml min −1 in a sharp Gaussian peak at an elution volume of 10.01 ml (Extended Data Fig. 2g). To determine the apparent molecular weight of MtFsr, standard proteins (conalbumin, aldolase and ferritin, purchased from GE Healthcare) were passed at the same flow rate and in the same buffer. The fractions of interest containing MtFsr were concentrated with a 30-kDa-cutoff centrifugal concentrator to 1 ml and the protein was directly used for enzymatic activity assays. The concentration of purified MtFsr, estimated by the Bradford method, was 3.41 mg ml −1 .

Purification of MtFsr for titrations and EPR spectroscopy
For the titrations and EPR spectroscopic measurements two separate purifications were carried out starting either with 34 g cells (wet weight) derived from a SO 3 2− -grown fermenter, or with 49.5 g cells (wet weight) derived from a SO 4 2− -grown fermenter. Cells were thawed under warm water and transferred to an anaerobic tent containing an atmosphere of N 2 /CO 2 (90:10). Cells were lysed by osmotic shock through the addition of 180 ml and 240 ml lysis buffer (50 mM Tricine/NaOH pH 8.0, 2 mM DTT), respectively. The cell lysates were homo genized by sonication: 4 cycles at 72% intensity with 60 pulses followed by 1.30 minute break (probe MS76, SONOPULS Bandelin) and the cell debris was removed anaerobically via centrifugation (21,000g, 1 h at 10 °C). The supernatant was transferred in a Coy tent (N 2 /H 2 atmosphere of 97:3), with yellow light at 20 °C.
The purification steps were carried out as described in 'Purification of MtFsr for crystallization'. In the final purification step the buffer was exchanged by dilution and concentration in storage buffer (25 mM Tris-HCl pH 7.6, containing 10% v/v glycerol and 2 mM DTT) by using 30-kDa-cutoff filter (6 ml, Merck Millipore). MtFsr derived from the SO 3 2− -grown fermenter was concentrated to 18 mg ml −1 in a volume of 4.54 ml, and for the SO 4 2− -grown fermenter MtFsr was concentrated to 20 mg ml −1 in a volume of 1.24 ml. The protein concentrations were estimated by the Bradford method.

Mass spectrometry identification
Purified MtFsr (1 µg) was digested with trypsin and analyzed by mass spectrometry (ThermoFisher Q Exactive HF coupled to an Easy-nLC 1200) as described in ref. 44.

Protein crystallization
The purified enzymes were kept in 25 mM Tris-HCl pH 7.6, 10% v/v glycerol and 2 mM DTT. Fresh, unfrozen samples were immediately used for crystallization. Crystals were obtained anaerobically (N 2 /H 2 , 97:3) by initial screening at 20 °C using the sitting-drop method on 96-well MRC two-drop crystallization plates in polystyrene (SWISSCI) containing 90 µl of crystallization solution in the reservoir.

Crystallization of MtFsr
MtFsr at a concentration of 11 mg ml −1 was cocrystallized with FAD (0.5 mM final concentration) and F 420 H 2 (15.5 µM final concentration). The protein sample (0.6 µl) was mixed with 0.6 µl reservoir solution. Thick, square-shaped, brown crystals appeared after a few days. The reservoir solution contained 200 mM lithium sulfate, 100 mM Bis-Tris, pH 5.5 and 25% w/v polyethylene glycol 3350.

X-ray crystallography and structural analysis
Crystal handling was done inside the Coy tent under anaerobic atmosphere (N 2 /H 2 , 97:3). MjFsr crystals were directly plunged in liquid nitrogen, whereas MtFsr crystals were soaked in their crystallization solution supplemented with 20% v/v ethylene glycol as a cryo-protectant before being frozen in liquid nitrogen. Crystals were tested and collected at 100 K at the Synchrotron Source Optimisée de Lumière d'Énergie Intermédiaire du LURE (SOLEIL), PROXIMA-1 beamline; the Swiss Light Source, X06DA-PXIII; and at PETRA III, P11.

MjFsr
After an X-ray fluorescence spectrum on the Fe K-edge, datasets were collected at 1.74013 Å to perform the single-wavelength anomalous dispersion experiment. Native datasets were collected at a wavelength of 0.97857 Å on the same crystal. Data were processed and scaled with autoPROC 45 . The resolution limits in each cell direction were as follows: a = 2.43 Å, b = 2.62 Å and c = 2.19 Å. Phasing (obtained maximum CFOM for the substructure determination was 69), density modification and automatic building were performed with CRANK-2 (ref. 46). The asymmetric unit of MjFsr contains two half homotetramers. The model was then manually built with Coot and further refined with PHE-NIX 47,48 . X-ray crystallographic data were twinned, and the refinement was performed by applying the following twin law -k, -h, -l. During the refinement translational-liberation screw was applied.

MtFsr
Data were processed and scaled with autoPROC. The resolution limits in each cell direction were as follows: a = 1.69 Å, b = 1.55 Å and c = 1.81 Å. The structure was solved by molecular replacement with phaser from PHENIX, using MjFsr as a template 48 . The asymmetric unit of MtFsr contains four homotetramers. This crystalline form presents a notable translational noncrystallographic symmetry (14%). The model was then manually rebuilt with Coot and further refined with PHENIX. During the refinement, noncrystallographic symmetry and translational-liberation screw were applied. In the last refinement cycles, hydrogens were added in riding positions. Hydrogens were omitted from the final deposited model. In one of the chains (chain N), the lid region 204-253 has two different conformations, and both were tentatively modeled.
All models were validated through the MolProbity server (http:// molprobity.biochem.duke.edu) 49 . B-factors, MolProbity scores and rotamer outliers in Extended Data Table 1 were calculated based on the available PDB structures with PHENIX. The other values in Extended Data Table 1 were derived from the original first PDB reports. Data collection and refinement statistics, as well as PDB identification codes for the deposited models and structure factors, are listed in Extended Data Table 1  MtFrh was required to reduce F 420 and was purified from the same batch of cells as MtFsr used for crystallization. The activity of MtFrh after each purification step was followed by the reduction of methyl viologen in the N 2 /H 2 tent (97:3). The assay was performed in 120 µl of 0.5 M KH 2 PO 4 /NaOH pH 7.6 containing 1.7 mM of oxidized methyl viologen. The addition of 2 µl from the fractions containing Frh led to a blue coloration.
MtFrh was in the same pool as MtFsr used for crystallization, for the DEAE and the Q Sepharose columns. The Q Sepharose column performed the separation of the two target proteins. MtFrh eluted between 0.48 and 0.49 M NaCl from the Q Sepharose column. The filtered sample was applied to a 10-ml hydroxyapatite type 1 (Bio-Scale Mini CHT cartridges, BioRad) equilibrated with HAP buffer (20 mM Tris-HCl pH 7.6, containing 10% v/v glycerol and 2 mM DTT) by using a 30-kDa-cutoff filter (6 ml, Merck Millipore) and MtFrh was concentrated to 4.97 mg ml −1 in 100 µl. The purified sample was aliquoted and anaerobically flash frozen in liquid N 2 and stored at −80 °C. MtFrh lost its activity after more than one cycle of thawing-freezing.

Purification of oxidized F 420
Since F 420 is highly sensitive to light, all steps were carried out under yellow light or by covering the sample with aluminum foil. About 10 g (wet weight) of M. thermolithotrophicus cells from a 1.5-l fermenter were anaerobically lysed by osmotic shock and sonication (see above). The sample was centrifuged at 45,000g for 60 min at 4 °C. The supernatant was transferred in a Coy tent containing an atmosphere of N 2 /H 2 (97:3). The sample was filtered and passed onto a 30-ml DEAE Sepharose column equilibrated with 50 mM Tricine/NaOH pH 8.0 and 2 mM DTT. F 420 was eluted by a gradient of 0 to 0.6 M NaCl. The samples containing F 420 were determined on the basis of the absorbance profile at 420 nm and eluted between 0.48 M and 0.58 M NaCl. Pooled fractions were moved outside the tent and diluted with one volume of HIC-F 420 buffer (25 mM Tris HCl pH 7.6, 2 M (NH 4 ) 2 SO 4 ). (NH 4 ) 2 SO 4 powder was directly added to the diluted sample to reach a final concentration of 3 M (NH 4 ) 2 SO 4 and was stirred for one hour at room temperature. The sample was centrifuged at 4,000g for 20 minutes at room temperature. The supernatant was filtered through a 0.2-µm filter and loaded on a 30-ml Phenyl-Sepharose high-performance column, equilibrated with HIC-F 420 buffer. F 420 was eluted by washing the column with the HIC-F 420 buffer, at a flow rate of 2 ml min −1 and 1-ml fractions were collected. The fractions containing F 420 were pooled and filtered through a 0.2-µm filter. The sample was diluted by 50 volumes of 5 mM Tris-HCl pH 8.0 and loaded overnight on a 5-ml Q Sepharose high-performance column, equilibrated in 5 mM Tris-HCl pH 8.0. The following steps were performed at 4 °C. The column containing the bound F 420 was washed with 5 CV of 20 mM (NH 4 )HCO 3 precooled at 4 °C. F 420 elution was performed by adding 1 M (NH 4 )HCO 3 and collected in a brown serum flask. (NH 4 )HCO 3 was removed by evacuation at 37 °C for 2 hours under constant stirring. (NH 4 )HCO 3 -free F 420 powder was obtained by freeze drying. The purity of the preparation was checked by measuring the ratio of Abs 247 /Abs 420 in 25 mM Tris buffer pH 8.8. A pure sample would have a ratio value of 0.85 (ref. 50). F 420 concentration was estimated by measuring the absorbance at 420 nm in 25 mM Tris buffer pH 7.5 (ε 420 = 41.4 mM −1 cm −1 ). The final concentration of oxidized F 420 used for this study was 3.15 mM and 7.53 mM.

Reduction of F 420 for enzyme assays
For enzyme activity assays and cocrystallization of MtFsr with F 420 H 2 , the oxidized F 420 needed to be reduced. Dithionite was not used since it contains 10-20% (m/m) sodium sulfite and generates further SO 3 2− as product. All steps were performed under the strict exclusion of oxygen and under yellow light. First, the aerobic gas phase of the F 420 stock was exchanged several times for N 2 . The sample was then transferred in a Coy tent with an atmosphere containing a N 2 /H 2 mixture (97:3). The reduction took place in 1.4 ml 200 mM KH 2 PO 4 , pH 7.0, 0.5 mM F 420 , and 5 µl of 5 mg ml −1 purified MtFrh was added. Outside the tent, in a brown serum flask, the gas phase was exchanged three times for H 2 and CO 2 by evacuation and gassing with 1 × 10 5 Pa H 2 and CO 2 (80:20) at room temperature. The reduction of F 420 was observed by the color shift from yellow to transparent. Frh was removed by passing the sample through a 10-kDa-cutoff filter. Since reduced F 420 is not stable and oxidizes with time, aliquoted F 420 H 2 without Frh was immediately flash frozen in liquid N 2 and stored at −80 °C.

Reduction of F 420 for redox titrations
F 420 is the physiological electron donor for Fsr and was therefore used as the reductant for the redox titrations. Oxidized F 420 was purified as described before. Since both the reduction of F 420 with Frh is not complete and F 420 H 2 is not stable over time, we reduced F 420 with sodium borohydride, as previously described 51 . The reduction of F 420 was performed in an anaerobic chamber with an N 2 /H 2 atmosphere of 97:3 at 25 °C. F 420 H 2 was generated by reducing 100 µl F 420 at 7.53 mM with a few sodium borohydride crystals in a 10 mM Tris-HCl solution at pH 7.6, followed by destruction of excess borohydride by acidification with 50 µl 1 M hydrochloric acid. After the hydrogen evolution ceased, the pH was readjusted by the addition of 50 µl 1 M Tris-HCl pH 8.0. The generated F 420 H 2 was prepared freshly for each experiment and used immediately.

Enzymatic assays
Enzymatic Fsr measurements were performed in 200 mM KH 2 PO 4 buffer pH 7.0 under strict exclusion of hydrogen and oxygen. F 420 was reduced by Frh as previously described. The oxidation of the reduced electron donor F 420 was followed spectrophotometrically at 420 nm. For F 420 H 2 , a molecular extinction coefficient of 33.82 mM −1 cm −1 at 420 nm was experimentally determined for the above-mentioned conditions.
The assays for the specific enzyme activity were performed at 65 °C in a 1-ml quartz cuvette closed with a butyl rubber stopper. The gas phase of the cuvette was exchanged several times with N 2 . To monitor the reduction of SO 3 ). We further tested whether MtFsr can function in the reverse way by providing 1.4 mM Na 2 S as an electron donor and 47.3 µM of oxidized F 420 . All experiments were performed in triplicate.
The appK m and appV max of MtFsr for SO 3 2− and NO 2 − were determined at 50 °C under an anaerobic atmosphere (100% N 2 ). The assays were performed in 96-deep-well plates and monitored spectrophotometrically (FLUOstar Omega Multi-Mode Microplate Reader). To determine the appK m and appV max of MtFsr, 0-500 µM Na 2 SO 3 or NaNO 2 and 50 µM F 420 H 2 were added to the 200 mM KH 2 PO 4 buffer pH 7.0 and the reaction was started by the addition of 3.8 ng MtFsr. All experiments were Nature Chemical Biology Article https://doi.org/10.1038/s41589-022-01232-y performed in triplicate with a standard deviation represented by the ± sign. Kinetic parameters were calculated based on the ic50.tk server by applying a Hill coefficient of 1 (http://www.ic50.tk/kmvmax.html).

EPR spectroscopy
The midpoint potentials of the [4Fe-4S] centers and the siroheme of MtFsr were determined from EPR signal intensities and EPR integrals of the various redox states. All titrations were performed in a Coy tent (N 2 /H 2 , 97:3), at 25 °C in the dark. A volume of 3.32 or 3 ml for the reductive or oxidative titrations with F 420 H 2 or potassium ferricyanide at an initial MtFsr concentration of 4.07 or 2.7 mg ml −1 (in 100 mM Tris-HCl, pH 7.6), respectively, was stirred under anaerobic conditions. The solution potential was measured with an InLab ARGENTHAL (Mettler) microelectrode (Ag/AgCl, +207 mV versus H 2 /H + with in-built platinum counter electrode) in the presence of the respective mediator mix. MtFsr was preincubated for 30 minutes before each titration with the mediator mix and assay buffer. The amount of MtFsr available and the necessary protein concentration to obtain a satisfying signal-to-noise ratio for the EPR spectra precluded multiple titrations. Thus, values reported were from a single redox titration for the siroheme and from two redox titrations for the Fe/S signals.
The mediator mix for the reductive titration contained methylene blue, resorufin, indigo carmine, 2-hydroxy-1,4-naphthoquinone (50 µM), sodium anthraquinone-2-sulfonate, phenosafranin, safranin T, neutral red, benzyl and methyl viologen (all at a final concentration of 25 µM, except 2-hydroxy-1,4-naphthoquinone). For the oxidative titration the mediator mix contained methylene blue, resorufin, indigo carmine, 2-hydroxy-1,4-naphthoquinone (all at a final concentration of 20 µM). After adjustment of the potential by microliter additions of F 420 H 2 or potassium ferricyanide and 3 minutes equilibration, EPR samples were taken. For this, 300 µl of the mix were withdrawn, removed from the anaerobic glovebox in EPR tubes after attachment of a 5-cm piece of 3 mm × 7 mm (internal diameter × outer diameter) natural rubber tubing sealed with a 5-mm outer diameter acrylic glass stick at the other end. The samples were stored in liquid nitrogen until EPR spectra were recorded.
MtFsr as isolated was already in a partially reduced state. To obtain the completely oxidized form, 675 µl Fsr at 20 mg ml −1 was incubated for 30 minutes with 2 mM methylene blue. The sample was then passed through a Sephadex G-25M column (previously equilibrated with 100 mM Tris-HCl pH 7.6) to remove the methylene blue. This methylene blue-treated Fsr (1.28 ml) was collected at a concentration of 5.65 mg ml −1 and 300 µl was directly taken frozen for EPR spectroscopy of Fsr in its oxidized form.
Samples from the same methylene blue-treated Fsr (passed through a Sephadex G-25M column) at 5.09 mg ml −1 final concentration were incubated for 5 minutes with 10 mM Na 2 SO 3 , and then stored in liquid nitrogen.
All EPR spectra were recorded on a Bruker Elexsys E580 X band spectrometer (digitally upgraded) with a 4122HQE cavity linked to an ESR 900 Oxford Instruments helium flow cryostat. Cryocooling was performed by a Stinger (Cold Edge Technologies) closed-cycle cryostat driven by an F-70 Sumitomo helium compressor. Our local glassblower produced EPR tubes from Ilmasil PN tubing (outer diameter 4.7 mm and 0.5 mm wall thickness, Qsil). Before use, the tubes were extensively cleaned with pipe cleaners to remove inadvertent contaminants. EPR spectra were simulated with Easyspin 52 . The concentration of Fsr for the spin integration (using a 1 mM Cu 2+ -EDTA solution as standard) was obtained by dividing the Fe concentration, as determined with the ferene method 29 , by 24, since siroheme does not release Fe. Fitting to the Nernst equation was performed in Excel.

High-resolution clear-native PAGE
To visualize the expression levels of Fsr in HS − -versus SO 3 2− -grown cultures, and to estimate the oligomerization of Fsr, high-resolution clear-native-PAGE (hrCN-PAGE) was performed. 10 ml of M. thermolithotrophicus and M. jannaschii cultures, with either 2 mM Na 2 S or 2 mM Na 2 SO 3 as sulfur source, were grown for one night at 65 °C, with standing. Cells were collected by anaerobic centrifugation at 6,000g for 20 min at room temperature and the cell pellets were resuspended in 2 ml of 50 mM Tricine/NaOH pH 8.0 and 2 mM DTT. The cells were anaerobically sonicated four times at 70% intensity for 10 seconds, followed by a 30-second break (MS 73 probe, SONOPULS Bandelin). The hrCN-PAGE was run anaerobically and the protocol was adapted from ref. 53. Linear polyacrylamide gradient gels (8-15%) were prepared under aerobic conditions but then transferred into an anoxic chamber (atmosphere of N 2 /CO 2 , 90:10), where the gels were equilibrated in anaerobic cathode buffer (50 mM Tricine; 15 mM Bis-Tris, pH 7.0; 0.05% w/v sodium deoxycholate; 0.01% w/v dodecyl maltoside and 2 mM DTT) overnight. Fresh and anaerobic samples were diluted with the lysis buffer to a final concentration of 1 mg ml −1 and a volume of 12 µl per sample was loaded onto the gel, as well as 2 µl of the Native-Mark Unstained Protein Standard ladder (ThermoFisher). Glycerol (20% v/v final) was added to each sample and 0.001% w/v Ponceau S served as a marker for protein migration. The electrophoresis anode buffer contained 50 mM Bis-Tris buffer pH 7.0 and 2 mM DTT. The hrCN gels were run with a constant 40-mA current (PowerPac Basic Power Supply, BioRad). After electrophoresis, the protein bands were aerobically stained with Instant Blue (Expedeon).

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

Data availability
The crystal structures have been deposited in the Protein Data Bank under accession codes: 7NP8 for MjFsr and 7NPA for MtFsr. Raw crystallographic data have been deposited on Zenodo: https://doi. org/10.5281/zenodo.4751125. The data for this study are available within the paper and its Supplementary Information. Source data are provided with this paper.  suggesting a tight covalent binding. In MtFsr, the bridging-sulfur of the cysteine 472 is at a distance of 2.6 Å to the Fe-siroheme and the sulfur from the modelled HS − is 2.9 Å distant to the Fe-siroheme, indicating a loose binding of the HS − , which might result from a reduction event by X-ray radiation 60 . d, Siroheme superposition between aSirs (1AOP, 5H92), dSirs (3MM5, 2V4J) and Fsrs. Siroheme from aSirs and Fsr are coloured in green, structural siroheme/sirohydrochlorin from dSirs in black and dSirs functional sirohemes in blue. Superposition analysis shows that the functional sirohemes are arranged in a highly similar manner, whereas the conformation of the structural siroheme or sirohydrochlorin differ, which highlights the strong influence of the protein environment on the siroheme geometry.

March 2021
Data Policy information about availability of data All manuscripts must include a data availability statement. This statement should provide the following information, where applicable: -Accession codes, unique identifiers, or web links for publicly available datasets -A description of any restrictions on data availability -For clinical datasets or third party data, please ensure that the statement adheres to our policy