Main

Volatile organic sulfur compounds (VOSCs) are ubiquitous in the global sulfur cycle and serve as sources of sulfur for the metabolism of many bacteria1. Dimethylsulfide (DMS), the most abundant VOSC on Earth, is produced by bacteria on the order of 10 teragrams of sulfur per year2. Other VOSCs such as methylthio-propanol and methylthio-ethanol (MT-EtOH) are produced by microbial fermentation or recycling of the amino acid methionine3,4. The methylthio-alkane reductase (MAR) system in anaerobic and phototrophic terrestrial bacteria such as Rhodospirillum rubrum and Rhodopseudomonas palustris was discovered to be involved in unprecedented reductive C–S bond cleavage of VOSCs to acquire sulfur. For example, cleavage of DMS and MT-EtOH resulted in methanethiol (CH3–SH) for needed methionine biosynthesis, along with the release of methane and ethylene, respectively, as by-products5. This process required nitrogen fixation-like (NFL) genes marBHDK that had remained of unknown function for over a decade since their initial bioinformatic identification in bacterial genomes3,5. Furthermore, MAR genes were under transcriptional control of a sulfate-responsive regulator, SalR, that activated gene expression when sulfate became limiting5. These findings revealed that anaerobic bacteria with MAR turn to VOSCs for sulfur acquisition when sulfate is scarce. However, the structural and functional basis for reductive C–S bond cleavage by an NFL system remained unknown.

The nitrogenase superfamily of enzymes characterized so far is composed of distinct reductases involved in nitrogen, carbon and energy metabolism. Nitrogenase (N2ase) performs essential biological nitrogen fixation by reducing N2 to ammonia. It also reduces protons, acetylene, CO and CO2, among other small double- and triple-bond-containing compounds6,7,8 (Fig. 1). Alternately, the NFL dark-operative protochlorophyllide oxidoreductase (DPOR) and chlorophyllide oxidoreductase (COR) reduces porphyrin ring C=C double bonds for bacteriochlorophyll biosynthesis9,10, and Ni2+-sirohydrochlorin a,c-diamide reductase (CfbCD) performs similar reductions for methanogenesis cofactor F430 biosynthesis11,12 (Fig. 1a). A hallmark of the three isoforms of N2ase, which are Mo (NifHDK), V (VnfHDGK) and Fe (AnfHDGK) N2ase, is the presence of two complex metallocofactors, the [8Fe-7S] P-cluster and the active site M-cluster. For Mo and Fe N2ase, the M-cluster is a [7Fe-9S-C-X-homocitrate] metallocofactor with X = Mo or Fe13,14,15 (Fig. 1b). In V N2ase, the M-cluster is [7Fe-8S-CO3-C-V-homocitrate] with a carbonate ligand replacing sulfur S3A. M-cluster biosynthesis requires NifB, a radical S-adenosyl-L-methionine enzyme that joins two 4Fe-4S clusters and inserts a central carbide to form the initial [8Fe-9S-C] L-cluster (NifB-co)13,16. Further assembly of the FeMo-co and FeV-co requires chaperones NifEN and VnfEN, respectively, whereas FeFe-co assembly occurs in situ13. By contrast, regarding the NFL systems for bacteriochlorophyll and F430 biosynthesis, the active site for tetrapyrrole ring reduction is composed of a 4Fe-4S cluster and a hydrophobic substrate binding cavity9,17,18 (Fig. 1). This comparison suggested that each member of the nitrogenase superfamily has distinct structural features, metallocofactors, substrate specificities and catalytic mechanisms, including the MAR system for reductively cleaving VOSCs.

Fig. 1: Nitrogenase and NFL systems.
figure 1

a,b, Subunit composition (a) and metallocofactor composition (b) of the three isoforms of N2ase and the NFL systems, which are MAR for reductive C–S bond cleavage of VOSCs5, DPOR and COR for bacteriochlorophyll synthesis9,10 and CfbCD for methanogenesis F430 cofactor synthesis11. For V and Fe N2ase, an additional G-subunit (δ) is involved in M-cluster association and interactions between components26,48,70. The mar1 and mar2 clusters are P-like and FeFe-co-like metallocofactors whose precise identity is not fully resolved. c, Reactions catalysed by N2ase. d, Reactions catalysed by MAR.

Source data

Full size image

Here, we present the native isolation and purification of MAR, its catalytic and spectroscopic characterization and an atomic-resolution structure of the oxidized enzyme. Together, they reveal a nitrogenase-like system that utilizes complex metallocofactors matured by MarB or NifB akin to N2ase for reductive catalysis, and a small binding protein called MarS that regulates activity during unfavourable conditions.

Results

MAR is a two-component system for S acquisition from VOSCs

Initially, we endeavoured to natively purify the MAR system from Rhodospirillum rubrum, given that any requisite genes beyond marBHDK were unknown. We grew the organism anaerobically under MAR-inducing conditions (MT-EtOH as the sulfur source)5 and developed MAR activity assays for VOSC reduction by whole cells and subcellular fractions. MAR enzyme catalysis required ATP, an electron donor, and anoxygenic conditions, analogous to N2ase and other NFL systems7,10,11,19 (Table 1 and Extended Data Fig. 1a–c). MAR activity was observed exclusively in the soluble fraction, which was further purified by anion exchange chromatography to 100-fold enrichment in activity. Denaturing gel electrophoresis revealed the presence of multiple protein species (Extended Data Fig. 1b), preventing unambiguous identification of the MAR system and revealing that <0.3 % of total cellular protein was composed of MAR. Thus, MAR abundance appears to be ~10-fold lower than when N2ase is induced for N2 fixation (1–10% of total cellular protein)19.

Table 1 MAR purification and activity
Full size table

To overcome this limitation, we constructed a homologous expression system for marBHDK in R. rubrum such that synthesized MarD possessed an N-terminal six-histidine tag (MarDHis6) and MarH possessed a C-terminal tag (MarHHis6), as previously done for N2ase20,21 (Fig. 2c). MAR genes were expressed under anaerobic conditions in the R. rubrum marBHDK/nflDK deletion strain (ΔmarBHDK ΔnflDK). Addition of the histidine-tags did not alter MAR activity in vivo (Extended Data Fig. 1d–f). Through the constructed expression system, total MAR activity increased 10-fold in R. rubrum, which again was recovered in the soluble fraction (Table 1). Nickel-affinity purification and anion-exchange polishing further enriched the MAR system by >300-fold to ~99% purity, which corresponded to a MAR abundance of 1% of the total cellular protein (Table 1 and Fig. 2d). The recovered MAR system was composed of a 62 kDa MarH dimer and a 228 kDa MarDK tetramer as resolved by size-exclusion chromatography (Fig. 2e,f, fractions I and II). In addition, an apparent 550 kDa higher molecular weight species was observed to contain R. rubrum protein A3441 in complex with MarDK as quantified by proteomics and matrix-assisted laser desorption–time of flight (MALDI-TOF) mass spectrometry (Fig. 2d–f, fraction III; Extended Data Fig. 2). This is consistent with two MarDK tetramers (434 kDa) complexed by two A3441 dimers (88 kDa), and this complex is involved in MAR activity regulation in response to sulfate and light availability (vida infra). The gene for this protein (Rru_A3441) is located at base 3,963,327 on the R. rubrum chromosome (NCBI: NC_007643.1) near other genes of unknown function, whereas the marBHDK genes are located at base 951,392. We designate the A3441 protein as MarS and corresponding gene marS for the MAR switch-off regulator.

Fig. 2: Purification and subunit composition of MAR.
figure 2

a, Plasmid-based expression system for marH gene with poly-histidine sequence (red star) for synthesis of MarH with C-terminal His-tag and pulldown of native MarH. b, Abundance of overproduced MarH (using pPUF-MarHHis6) and natively produced MarH in the wild-type (WT) and marBHDK deletion strain. c, Plasmid-based expression system for MAR genes with poly-histidine sequences (red stars) for synthesis of MarB N-terminal His-tag, MarH with C-terminal His-tag, MarD with N-terminal His-tag and MarK. d, SDS–PAGE of MAR system purification after cell lysis, affinity purification and ion-exchange purification. e,f, Analytical size-exclusion chromatography of isolated MAR system (e) and corresponding SDS–PAGE analysis of subunit composition (f). Roman numerals in e correspond to the lanes in f. Maroon spheres indicate the MarS protein, which interacts with MarDK to form higher molecular weight species and regulate MAR activity. Purification experiments were repeated n = 2 times with independent proteins samples with the same results. Prot. Std., protein standard.

Source data

Full size image

MarH possesses a conserved arginine residue, Arg100, which is also found in all N2ases and other NFL systems except COR, and is potentially involved in posttranslational regulation of activity5. In R. rubrum, Mo N2ase NifH Arg101 is the site of ADP-ribosylation, which prevents NifH binding to NifDK and switches off N2ase activity during periods of darkness and ammonia influx (R. rubrum numbering; Arg100 in A. vinelandii)22. ADP-ribosylation is performed by dinitrogenase reductase ADP-ribosyltransferase (DRAT) and removed by dinitrogenase reductase activating glycohydrolase (DRAG), and their activity is regulated by P-II family regulators. Mass spectrometry analysis of MarHDK proteins showed no posttranslational modifications, and any shifts observed in MarH electrophoretic mobility were due to oxidation or reduction of the peptide (Extended Data Fig. 2a–c). Therefore, the MAR system is not regulated posttranslationally by DRAT and DRAG. This is consistent with the separate roles of MAR and N2ase in sulfur and nitrogen metabolism, respectively.

For reductive C–S bond cleavage of VOSCs by isolated MAR components, MarH alone and MarDK alone exhibited no activity (Fig. 2f, fractions I and II; Fig. 3c). When combined together under the same reaction conditions as before, VOSC cleavage was observed at specific rates of 1–3 nmol substrate per nanomole of protein per minute (Fig. 3c,d). Surprisingly, this is 10-fold lower in specific activity compared with N2ase for N2 reduction23. Clearly, in vivo MAR abundance and activity is sufficient for meeting cell sulfur demands, so we quantified MarH abundance through expression of his-tagged MarH as bait for recovery of native MarH (Fig. 2a,b). The overproduced MarHHis6 versus native MarH were recovered at abundances of 1% and 0.1% total cell protein, respectively (Fig. 2b and Supplementary Fig. 6; n = 2). This confirmed our initial indications that when the MAR system is induced by limiting sulfate conditions, its abundance is 10-fold lower compared with when N2ase is induced by limiting ammonia conditions. For R. rubrum, the cellular elemental ratio is C1H1.7O0.4N0.2S0.002, which means that 100-fold less sulfur is needed for cell growth than nitrogen24. Therefore, the 10-fold lower specific activity of purified MAR and 10-fold lower abundance of MAR versus N2ase is fully sufficient to fulfil the 100-fold lower sulfur demands, indicating that the observed purified activities are indeed reflective of in vivo activity (Fig. 3c,d). Altogether, this conclusively demonstrates that MarH and MarDK form an oxygen sensitive two-component system that functions in reductive VOSC cleavage in an ATP- and electron-dependent manner (Fig. 1).

Fig. 3: MAR activity and metallocofactor characterization.
figure 3

a, UV–vis absorbance spectra of purified MarH and MarDK (Fig. 2f, fractions I and II) under dithionite-reduced conditions. b, ICP-MS metal analysis of MarDK compared with Mo, V and Fe N2ase23,36,37. The hash symbol indicates the limit of detection in mol mol−1 protein: Fe <0.1; V <0.04; Mo <0.004. n.d., not determined. c, Functional requirements and activity of purified MarH and MarDK fraction without bound MarS protein (Fig. 2f, fractions I and II) for each known VOSC substrate. MarH = 0.8 µM dimer and MarDK = 0.4 µM tetramer. d, Specific activity of MAR with each VOSC substrate obtained by weighted linear regression and standard error calculation from data in c, and specific activity of Fe N2ase and Mo N2ase for N2 from Harris et al., 2018 (*)23. e, CW X-band EPR spectra (ν = 9.38 GHz) of 435 µM MarH dimer (‘MarH’) and 135 µM MarH dimer + 70 µM MarDK tetramer (‘MarHDK’) without ATP (dashed) and with ATP (solid). Subtraction of resting-state MarHDK from MarHDK under turnover (shown as x3 for clarity) with simulated signals results in observed ‘E1(H)-like’ (blue) and ‘E1(H)*-like’ (red) species overlaid. The asterisk indicates unidentified residual signal. f, MAR activity for MT-EtOH cleavage to ethylene increases, then is inhibited as a function of increasing CO concentrations. Enzyme molar ratios correspond to MarH dimer to MarDK tetramer with MarDK = 0.4 µM. Averages and standard deviation error bars in b–d and f are for n = 3 independent experiments. In c–f, all reactions contained 4 mM ATP, 6 mM dithionite, 1 mM substrate and ATP regeneration system unless otherwise indicated.

Source data

Full size image

MAR structure parallels the nitrogenase fold

Given the successful isolation of MarDK, we sought to determine the structural architecture of MarDK in the absence of MarH (non-turnover conditions) for comparisons with N2ase. Initial attempts in cryogenic electron microscopy (cryo-EM) imaging of the MarDK fraction with MarS were unsuccessful due to population heterogeneity. Therefore, we analysed the MarDK species without MarS by cryo-EM (Fig. 4 and Extended Data Fig. 4). While MAR proteins were purified and handled under anoxic conditions, the final cryo-EM grid blotting was performed under atmospheric conditions (20% O2) due to technical limitations (Methods). However, how oxygen affects MAR is of particular interest, as it is for nitrogenase, because both systems must be properly protected from oxygen for activity in biotechnological applications. On average, the structure was refined to 2.4 Å resolution, and there was a region of no to lower resolution (>3.5 Å) for MarD residues 1–65, 188–197 and 372–409 and MarK residues 1–26 and 88–93 near the surface of MAR, which corresponded to residues predicted to reside underneath and adjacent to where MarH or MarS binds based on AlphaFold 3 models (Extended Data Fig. 5 and 9, Supplementary Fig. 1 and Supplementary Data 1–3). It is currently unknown whether this flexible region is due to lack of MarH or MarS binding, partial denaturation at the air–water interface or protein oxidation during cryo-EM grid deposition. From the cryo-EM structure, it is evident that MarDK is arranged in an α2β2 configuration with C2 symmetry, as seen in all previously characterized N2ase, DPOR and COR systems (Fig. 4). Furthermore, MarD and MarK peptides adopt an architecture similar to the nitrogenase-fold of three subdomains, each composed of four β-sheets (domains I and II) or five β-sheets (domain III) brought into proximity by five α-helices15,25 (Extended Data Figs. 6 and 7). While the nitrogenase fold is conserved across N2ase, DPOR, COR and MAR systems, there are key differences in the resulting architecture26,27. The Mo N2ase has an extended NifK N-terminal domain that pushes up on NifD and skews the DK–DʹKʹ interface. This is absent in MarDK, VnfDGK and AnfDGK systems, giving MAR an overall architecture that parallels the alternative N2ases (Fig. 4).

Fig. 4: MAR cryo-EM structure and nitrogenase superfamily architecture.
figure 4

a, Cryo-EM electron density map contoured at 7 sigma and corresponding structural model for MarDK in the oxidized state with locations of MAR metallocofactors (mar1, mar2) indicated (Figs. 5 and 6). b–d, Structural comparisons of Fe N2ase (Anf; 8OIE)26 (b), Mo N2ase (Nif; 8CRS)71 (c) and DPOR (Chl; 2XDQ)72 (d). Relevant metallocofactors are indicated.

Source data

Full size image

MAR metallocofactors resemble N2ase P- and M-clusters

Cryo-EM analysis of MarDK revealed that six cysteine residues are associated with an electron density envelope consistent with a P-like cluster that we designate as the mar1 cluster (Fig. 5a,b). This is in stark contrast to the four coordinating cysteine residues of the 4Fe-4S clusters by DPOR/COR9,15 (Fig. 4d and Extended Data Figs. 6 and 7). In the fully reduced state (PN), the P-cluster iron atoms of N2ase are coordinated by six conserved cysteine residues: NifD Cys62, 88, 154 and NifK Cys70, 95, 153 (Fig. 5d; A. vinelandii numbering)15. AlphaFold 3 models generated for MarDK predict the same six conserved R. rubrum MarDK cysteines are positioned to coordinate a P-cluster or P-cluster-like cofactor (Fig. 5c and Supplementary Data 1). These are MarD Cys66, 91, 158, and MarK Cys32, 56, 121, whose R. rubrum numbering will be used henceforth5. However, in the cryo-EM structure, there are notable differences. MarD Cys66 is not associated with the mar1-cluster electron density, whereas in the AlphaFold 3 model it is positioned to coordinate Fe #7 of a P-cluster (Fig. 5a–c). Rather, in the cryo-EM structure, MarK Cys56 coordinates Fe #7 of the mar1 cluster as compared with coordination of Fe #4 and Fe #5 predicted in the AlphaFold 3 model and equivalent coordination of N2ase P-cluster Fe #4 and Fe #5 by homologous NifK Cys95. However, MarK sequences also possess a seventh cysteine residue, MarK Cys52, which in N2ase is a glycine. MarK Cys52 is observed to coordinate mar1-cluster Fe #4 and Fe #5 in the cryo-EM structure, as compared with the Alphafold 3 model (Fig. 5a–d). Lastly, in the cryo-EM structure, there is a visible ingress in the electron density where Fe #8 of a P-like mar1 cluster would otherwise be expected, such that MarD Cys91 only coordinates Fe #2 as compared with coordination of Fe #2 and Fe #8 in the AlphaFold 3 model and equivalent coordination of N2ase P-cluster Fe #2 and Fe #8 by homologous NifD Cys88. Altogether, the configuration of the six coordinating cysteines in the electron density indicates a P-like cluster in an oxidized state where one of the four central Fe atoms (Fe #8) is missing (Fig. 5a,b). For N2ases, depending on available coordinating residues for oxidation states beyond the catalytically relevant P2+ state, one or more of the central Fe atoms is labile to oxidative damage28,29. For example, the Mo N2ase NifK residues Ser188 or Tyr98, which coordinate the P2+ oxidized state and provide some oxygen protection28, are absent in MarK15,30 (A. vinelandii numbering; Fig. 5d). Rather, for MarK, these residues are Gly151 and Lys59, respectively. MarK Lys59 is poised to form a salt bridge with MarD Glu68, which in Mo N2ase is also a tyrosine (NifD Tyr64; A. vinelandii numbering). The mar1-cluster state in the cryo-EM structure most closely resembles the P-cluster of the previously characterized NifK Ser188Ala mutant, which exhibits reversible Fe loss under oxidizing conditions29 (Fig. 5e). Thus, the mar1-cluster environment is clearly different from the N2ase P-cluster site for accessing variable coordination across different redox states. Given that the oxidation state of the mar-1 cluster as measured in cryo-EM is currently unknown, we consider that the MarD Cys52 and Cys56 may be involved in coordinating the mar-1 cluster in other redox states throughout the catalytic cycle31.

Fig. 5: MarDK mar1 cluster and N2ase P-cluster coordination.
figure 5

a, Cryo-EM electron density contoured at 5 sigma with 2 Å buffer around the mar1 cluster and coordinating cysteine residues. b, Same electron density contour as in (a), also showing the local MarDK protein environment. The electron density for the mar1 cluster indicates an oxidized P-like cluster with missing Fe #4. c, AlphaFold 3 model of MarDK in the mar1-cluster region, showing that the six cysteines conserved with N2ase are predicted to be positioned to coordinate a metallocofactor consistent with a reduced P-cluster (PN shown) as in N2ase. d, Structure of A. vinelandii NifDK (8CRS)71 in the P-cluster region showing PN coordination by the six conserved cysteines. In N2ase NifK Gly91 replaces MarK Cy52 (orange). e, Progressive oxidation of labile P-cluster irons of A. vinelandii NifK S188A, which is analogous to the MarK G151 residue of MAR28,29. Superscript D indicates the MarD or NifD subunit and superscript K indicates the MarK or NifK subunit.

Source data

Full size image

For catalysis the N2ase M-cluster is coordinated by a conserved cysteine and histidine residue: NifD Cys275 and His442 (Figs. 4c and 6c; A. vinelandii numbering)15,25. Coordinately, previous sequence alignment suggested that the conserved MarD Cys270 and nearby His450 could serve as ligands for a MAR catalytic metallocofactor5. In the cryo-EM density map, there was no indication of a metallocofactor in this region. This is consistent with the presumed catalytic metallocofactor being oxidatively damaged during deposition for cryo-EM. Interestingly, we observed a second histidine in this region that is not found in NifD, MarH His429, which is located on the loop after helix 18 (Fig. 6a–c). MarD His429 is in closer proximity to MarD Cys270 than MarD His450, which is located on the loop before helix 19 (9.6 Å versus 10.5 Å). Indeed, when we substituted MarD Cys270 with Ala or MarD His429 with Phe or Leu, this resulted in a completely inactive enzyme, analogous to when the Mo N2ase M-cluster ligands NifD Cys275 and NifD His442 are mutated32 (Fig. 6d). Substitution of MarD His450 and the only other cysteine residue in the region, MarD Cys73, had no effect. Collectively, these results suggest that MarD His429 and MarD Cys270 coordinate a metallocofactor, which is supported by AlphaFold 3 modelling showing an L-cluster coordinated between these residues (Fig. 6c and Supplementary Data 1).

Fig. 6: MarDK mar2-cluster region and MAR metallocofactor maturation by MarB and NifB.
figure 6

a,b, Cryo-EM of oxidized MarDK (a) and AlphaFold 3 model of MarDK (b) in the proposed mar2-cluster region. c, The structure of NifDK (8CRS)71 showing coordination of FeMo-co. The loop before helix 18 (black) contains MarD His429 required for MAR catalytic activity, which is absent in NifD. For NifD, the loop before helix 20 (orange) contains the M-cluster coordinating His442 along with left-handed helix hL. NifD and MarD helix h9 shown in green. d, In vivo MAR cleavage of VOSCs for site-specific amino acid substitutions of MarD. The hash symbol indicates a value below detection limit of 0.05 μmol. gDCW, grams dry cell weight. e, In vivo MAR cleavage of VOSCs when marBHDK, nflDK, nifB and nifEN genes are present on the chromosome, deleted or restored in trans from a plasmid. NifB or MarB is required for activity but not NifEN. MAR genes on the chromosome are regulated by SalR and expression is repressed by 1 mM sulfate5. When mar and nif genes are on the plasmid, they are constitutively expressed. OD, optical density. The asterisk indicates strains could not grow on DMS or MT-EtOH as the sole S source. All other strains grew to a final optical density of 2.0 in 48 h. Averages and standard deviation error bars in d and e are for n = 3 independent experiments. Superscript D indicates the MarD or NifD subunit and superscript K indicates the MarK or NifK subunit.

Source data

Full size image

Coordination of an active site metallocofactor in the region of MarD His423 and Cys270 is further supported by mutational analysis of MarD α-helix 9. In Mo N2ase, NifD α-helix 9 contains Gln191 and His195 for coordinating the substrate and M-cluster labile sulfur S2B above the M-cluster active site7,15,33 (Fig. 6c; A. vinelandii numbering). In MarD, these are replaced by hydrophobic Trp195 and Phe199 residues, respectively, with adjacent His194 and Asp200 (Fig. 6a,b; R. rubrum numbering)5. MarD Trp195 was required for catalysis (Fig. 6d), providing initial evidence that this residue may be involved in structuring the protein environment around the active site metallocofactor for cleavage of hydrophobic VOSCs.

Consistent with MarDK possessing multiple complex metallocofactors, it exhibits a broad multifeatured ultraviolet–visible (UV–vis) absorption spectrum near 400 nm as seen for N2ase34,35 (Fig. 3a). Inductively coupled plasma mass spectrometry (ICP-MS) metal analysis of MarDK reveals a metal content of ~32 Fe per MarDK tetramer and negligible vanadium or molybdenum (Fig. 3b). The iron content is similar to the 30–32 Fe found in N2ase from the P- and M-cluster (Fig. 1a), and parallels Fe N2ase, which contains only Fe23,36,37. This is in stark contrast to DPOR and COR, which contain only 8 Fe per tetramer due to one 4Fe-4S cluster at each heterodimer interface9. To further probe the MAR metallocofactors, we performed electron paramagnetic resonance (EPR) spectroscopy on MarH and the MarHDK complex under turnover and non-turnover conditions. The continuous-wave (CW) X-band EPR spectrum of isolated MarH in the absence of ATP exhibits signals consistent with a spin state (S) mixture of S = ½ species (g-tensor values: g1, g2, g3 = [2.03, 1.93, 1.85]; giso = 1.94) and S = 3/2 species (E/D = 0.21; D < 0; where E/D is the rhombicity of the zero field splitting tensor and D is the zero field splitting tensor) in a 1:5 ratio. In the presence of ATP, the EPR lineshapes of the S = ½ and S = 3/2 signals change (Fig. 3e and Extended Data Fig. 3b–d), similar to observations in the EPR spectra of the [4Fe-4S]+ cluster of NifH, VnfH and AnfH38,39,40. This observed behaviour suggests that, analogous to the iron proteins of nitrogenase40, MarH undergoes a conformational change upon binding of ATP. The EPR spectrum of MarHDK in the absence of ATP exhibits an attenuated S = 3/2 signal relative to MarH alone and an intense, near-axial signal around g = 2, characteristic of a [4Fe-4S]+ cluster, which we attribute primarily to MarH36,37,38 (Figs. 1a and 3e and Extended Data Fig. 3c). In the presence of ATP, several new features are observed spanning from g = 2.02 to 1.77. Subtraction of the resting-state spectrum reveals two major rhombic species with g1, g2, g3 = [1.98, 1.91, 1.78] (giso = 1.89) and g1, g2, g3 = [2.02, 1.97, 1.84] (giso = 1.94) in an approximately 1:1 ratio (Fig. 3e and Extended Data Fig. 3a). These features closely resemble those of the Fe N2ase E1(H) species (g1, g2, g3 = [1.965, 1.928, 1.779]; giso = 1.89) and E1(H)* species (g1, g2, g3 = [2.009, 1.950, 1.860]; giso = 1.94), which are catalytic intermediates observed under analogous turnover conditions. These signals are not expected to arise from canonical FeS clusters36,39. Spin quantitation of the EPR signal of MarHDK with and without ATP indicates the signals derive from active holo-enzyme (Extended Data Fig. 3d). Consistent with the elemental analysis that shows the presence of only iron, there are no EPR signals indicative of a FeMo-co or FeV-co at low fields in samples of MarHDK15,29,37,41,42 (Extended Data Fig. 3b,c). Taken together, these results are consistent with a second metallocofactor at the active site of MarDK that resembles FeFe-co bound to MarD Cys270 and His429, which we designate the mar2 cluster.

Lastly, similarity between mar2 and FeFe-co is further supported by considering the CO inhibition profile for MT-EtOH reduction to ethylene by MAR. As δCO was increased from 0 to 0.4 atm, MAR activity for MT-EtOH reduction to ethylene concomitantly increased above baseline. However, at δCO > 0.4 atm, the observed activity decreased until ultimately rendering the enzyme inactive (Fig. 3f and Extended Data Fig. 2f). This behaviour resembles that observed for acetylene reduction by the V N2ase. While the Mo N2ases show a strict inhibition by CO for a 2e− reduction of acetylene to ethylene at δCO ≤ 0.001 atm (ref. 43), the 2e− reduction of acetylene to ethylene by V N2ase displays a similar initial activity enhancement in the presence of CO under low electron flux conditions, followed by a decrease when δCO > 0.1 atm. The presence of CO also enhances the 4e− reduction of acetylene to ethane, up to 0.1 atm. This biphasic behaviour is attributed to the two separate binding sites for CO on the M-cluster43,44. The similarity in CO enhancement and inhibition profiles between MarDK and VnfDK provides initial evidence that MarDK interacts with CO during VOSC reduction in a similar manner as nitrogenases.

MAR maturation utilizes MarB or N2ase NifB

A striking feature of MAR gene clusters in bacteria is the presence of a highly conserved nifB homologue we designated marB5. MarB sequences contain the same conserved domains that NifB utilizes for synthesis of the [8Fe-9S-C] NifB-co (L-cluster) precursor of the M-cluster5 (Fig. 1b). Using a genetic approach, we characterized the role of MarB in MAR maturation. First, we verified that no Nif, DPOR or COR components of R. rubrum could replace MarH or MarDK, given previous reports of cross-interaction between nitrogenase superfamily components45 (Extended Data Fig. 8). This allowed us to test the requirement of MarB for MarHDK activity, which resulted in two key observations. First, in the absence of MarB, MAR activity was still present, and second, in vivo cleavage of DMS to methane decreased while MT-EtOH cleavage to ethylene increased (Fig. 6e, strains 1–5). Given the high (48%) sequence similarity between MarB and NifB5, we questioned whether NifB could function in lieu of MarB. Indeed, when NifB and MarB were both absent, no MAR activity was possible. Then, when NifB was restored in trans from a plasmid, the same differential production of methane versus ethylene was observed as when NifB was the sole maturase present on the chromosome (Fig. 6e, strains 5–9). This demonstrates that MarB and NifB individually can mature MarDK into a functional system and further establishes that M-like clusters function in systems beyond N2ase. Moreover, as with the Fe N2ase, activity does not require chaperone NifEN, indicating MarB and NifB directly interact with MarDK for metallocofactor assembly13 (Fig. 6e, strains 10–12).

MarS represses MAR activity upon sulfate influx or darkness

Based on sequence identity, MarS belongs to the phenolic acid decarboxylase regulator (PadR) winged helix family of proteins and shares 32% identity with the canonical PadR regulator from Bacillus subtilis46. The AlphaFold predicted structure (UniProt: Q2RNRO) contains a C-terminal dimerization domain, a winged-helix domain and a low-complexity N-terminal region that is 30% of the total peptide length (Extended Data Fig. 2d,e). MarS is not a sequence homologue of the V or Fe N2ase G-subunit, CowN protein for CO protection, or the Shethna protein II (FeSII) for O2 protection47,48,49,50,51. Therefore, we took an in vivo approach to determine MarS function by measuring growth and MAR activity in the wild-type (marS+) and MarS deletion strain (marS−). In the presence of oxygen, MarS did not confer any oxygen protection for growth using VOSCs (Extended Data Fig. 10c). Also, MarS was not required for CO protection, as no growth inhibition was observed in the presence of 0–0.1 atm CO. This is consistent with the observation that purified MAR is inhibited only above δCO >0.4 atm (Fig. 3f and Extended Data Fig. 10d). Instead, MarS is observed to function in switching-off MAR activity during sulfate influx, or darkness when ATP cannot be efficiently made by photophosphorylation. This is proposed to prevent unnecessary use of ATP akin to N2ase switch-off by DRAT during ammonium influx and darkness22 (Fig. 7b and Extended Data Fig. 10b). Coordinately, we simultaneously grew and purified MAR from marS+ and marS− strains to eliminate any effects of MarS on MAR activity. MarHDK isolated from the marS− strain exhibited 2-fold more activity than MarHDK isolated with MarS under standard reaction conditions, consistent with in vivo results (Fig. 7a). Both the decrease in MAR activity and coordination of two MarDK tetramers into a larger structure by MarS is corroborated by AlphaFold 3 models of MarS, PadR or MarH binding to MarDK (Extended Data Fig. 9 and Supplementary Fig. 1). In all models of MarS and MarDK multimerization, four MarS monomers formed two dimers that bridged two adjacent MarDK tetramers (Fig. 7c and Supplementary Fig. 1). Notably, MarS helix 3 is predicted to insert adjacent to MarD helix 5 and MarK helix 6 in a similar fashion to MarH helix 4. This is analogous to but distinct from Mo N2ase regulation by DRAG and DRAT. For NifH, conserved helix 4 interacts with NifD helix 5 and NifK helix 9 (Extended Data Figs. 6 and 7) and contains NifH Arg101 that is required for catalysis (R. rubrum numbering; Arg100 in A. vinelandii)52. NifH Arg101 ADP-ribosylation by DRAT disrupts the NifH–NifDK interaction interface to inhibit N2ase activity, whereas with MarH there is no evidence for MarH Arg100 ADP-ribosylation (Extended Data Fig. 2). Altogether, this points to a mechanism where MarS regulates MAR activity by binding and multimerizing MarDK to block key MarH interactions with MarDK. Currently, what triggers MarS binding and release for MarDK multimerization is unknown, but there is precedence for multimerization of nitrogenase. In Azotobacter vinelandii, the Shethna FeSII protein, upon oxidation of the FeS cluster, inserts between NifDK and NifH to turn off Mo N2ase and protect the metallocofactors from oxygen damage. This serves as a nucleation site for the association of additional FeSII, NifDK and NifH components into larger filamentous complexes thought to provide additional sequestration of N2ase from oxygen50,51. The binding of MAR to MarS is clearly different, as no MarH is associated in these complexes. Thus, it appears that nature has evolved mechanisms to regulate MAR activity in response to sulfate and energy availability by inhibiting MarH binding in a mechanism distinct from how nitrogenase activity is regulated by DRAT or FeSII in response to ammonia or oxygen, respectively.

Fig. 7: MarS is involved in MAR activity switch-off upon sulfate influx.
figure 7

a, Activity of MarHDK isolated from a marS deletion strain (marS−, strain ΔnflDK ΔmarBHDK ΔmarS,) compared with activity of MarHDK with MarS isolated from the marS competent strain (marS+, strain ΔnflDK ΔmarBHDK). MarH = 0.8 µM dimer and MarDK = 0.4 µM tetramer. b, In vivo MAR activity change upon influx of sulfate at time t = 0 in a marS+ (wild-type) and marS− strain (ΔmarS) initially grown on MT-EtOH as the sole sulfur source for maximum initial MAR activity. Lines are exponential fits to ethylene production by strains before and after sulfate addition showing a 2-fold change in ethylene rate constant. As the cultures grow exponentially, ethylene production follows the same exponential trend. In a and b, averages and error bar standard deviations are for n = 3 independent experiments. c,d, Representative AlphaFold 3 models of MarS and MarH binding to MarDK (Supplementary Fig. 1 and Supplementary Data 1–3).

Source data

Full size image

Discussion

The nitrogenase superfamily has evolved enzyme systems not only for nitrogen fixation (N2ase) and carbon/energy metabolism (DPOR, COR and CfbCD), but also for sulfur metabolism (MAR) based on the nitrogenase-fold domain (Fig. 4 and Extended Data Figs. 6 and 7). MarHDK is the catalytic system that produces methane from DMS and ethylene from MT-EtOH in R. rubrum and other organisms with MAR homologues5. This explains how ethylene would be made from MT-EtOH by organisms with MAR in anoxic soils3 and represents a promising alternative route for microbial production of this platform chemical to reduce reliance on fossil fuels.

The regulation of MAR activity through the binding and dissociation of the PadR family protein MarS is thought to be mediated by allosteric mechanisms. The canonical PadR regulator binds phenolic acid family molecules in an interdomain pocket between helix 6 and helix 5’ via residues Arg164, Lys127, His154 and Thr93’. Effector molecule binding allosterically induces PadR to dissociate from the DNA through an interdomain-reorganization mechanism46. While these effector molecule coordination residues are not conserved in MarS, we anticipate that specific residues in the analogous MarS helix6 and helix 5’ region, or the glycine-rich N-terminal tail (not present in PadR), will be involved in binding of effector molecules, albeit currently unknown, that are associated with the presence or absence of sulfate and light. Such binding would presumably result in MarS conformational changes that trigger association or dissociation from MarDK.

Strikingly, beyond MAR, the vast majority of group-IV NFL gene clusters that continue to be of unknown function also contain a conserved marB/nifB homologue. Coordinately, the NFL homologues of NifDK contain the same conserved cysteines used by N2ase for P-cluster coordination and conserved NifD Cys275 residue used by Mo N2ase for M-cluster coordination5. The observation that both MarB and NifB can mature the presumed MarDK mar2 cluster—and that this activity requires the conserved MarD residues Cys270 and the novel His429—strongly suggests something similar to the nitrogenase [8Fe-9S-C] L-cluster (NifB-co; Fig. 1b) is at the core of the catalytic metallocofactors throughout NFL systems of unknown function. Given the differential activities of the MAR system towards DMS versus MT-EtOH when assembled with MarB versus NifB, the bona fide mar2 cluster made by MarB is suggested to be distinct from the cofactor made by NifB. Alternatively, it may be the same cofactor but placed in a different orientation by the two maturases, resulting in an altered coordination environment that affects substrate binding and/or formation of catalytic intermediates. The observed EPR signal similarities of MarH to NifH, VnfH and AnfH and the similarities of MarHDK to the E1(H) and E1(H)* intermediates of Fe N2ase provide the first spectroscopic evidence that the mar2 cluster resembles FeFe-co41,53. Moreover, in Fe N2ase, these two species are postulated to interconvert via displacement of a bridging hydride within the M-cluster. This suggests that, like nitrogenase, the catalytic cofactor accumulates reducing equivalents in the form of hydrides and, thus, may be activated via sulfur release before reacting with substrate53. So far, the complex nitrogenase M-clusters and their L-cluster precursors have been unprecedented outside of nitrogenase enzymes. This has been attributed to specific evolution required to overcome the extreme stability of the N≡N bond7,33. The discovery of metallocofactors in MAR resembling the nitrogenase P- and M-clusters suggests that these catalysts evolved not only for nitrogen reduction, but also for sulfur reduction—and probably for broader roles in transforming essential elements into biologically accessible forms.

Methods

Chemicals and helper enzymes

All chemicals were of reagent purity or higher and were obtained from Sigma-Aldrich, except for ethylmethyl sulfide and (2-methylthio)ethanol, which were from ThermoFisher. Creatine phosphokinase was from Sigma-Aldrich. All restriction enzymes and T4 ligase were from New England Biolabs (NEB), and Phusion DNA Polymerase was from ThermoFisher.

Bacterial strains and growth conditions

All Rhodospirillum rubrum strains used in this study are derived from wild-type strain S1 (ATCC 25922)54 as follows:

Strain

Genotype

Reference

ΔnflDK

Δ0772:3; StR

5

ΔnflDK ΔmarBHDK

Δ0772:3 Δ0793:6; StR

5

ΔnflDK ΔmarBHDK ΔnifB

Δ0772:3 Δ0793:6 Δ0994; StR

This work

ΔnflDK ΔmarBHDK ΔnifEN

Δ0772:3 Δ0793:6 Δ2285:6; StR

This work

ΔmarS

Δ3441; StR, GmR

This work

ΔnflDK ΔmarBHDK ΔmarS

Δ0772:3 Δ0793:6 Δ3441; StR, GmR

This work

ΔnflDK ΔmarBHDK ΔnifB ΔmarS

Δ0772:3 Δ0793:6 Δ2285:6 Δ3441; StR, GmR

This work

Strains were grown in Ormerod’s sulfur-free malate minimal medium supplemented with 1 mM (NH4)2SO4 and/or 1 mM VOSC under anaerobic conditions with 95% N2:5% H2 gaseous headspace with appropriate antibiotics (50 µg ml−1 streptomycin, 25 µg ml−1 kanamycin and 1–4 µg ml−1 tetracycline)5,55. Ormerod’s medium was made sulfur free by replacing all sulfate-complexed metals with chloride versions. All in vivo growth and activity analyses were performed in 10 ml medium in anaerobic stoppered test tubes. All growth for protein purification and characterization was performed in 1- and 2-l Roux bottles. All strains were grown at 30 °C under 2,000 lux for growth experiments and 3,000 lux for protein production. All strain manipulations were performed in a Coy anaerobic chamber with 95% N2:5% H2 atmosphere. Cell growth for quantification of hydrocarbons produced from VOSCs by wild-type and modified MAR enzyme systems, cell growth for CO and O2 inhibition of MAR activity in vivo and cell growth for measuring sulfate influx and dark switch-off of MAR activity in vivo are fully detailed in the Supplementary Methods.

Plasmid construction for gene deletions and substitutions

Plasmids, primers and restriction enzyme sites for plasmid construction of truncated gene fragments for gene deletions listed in the ‘Bacterial strains and growth conditions’ section, construction of marBHDK and nifB complementation plasmids, and construction of plasmids with marDK mutations corresponding to site-specific amino acid substitutions are provided in Supplementary Tables 2 and 3. All constructed plasmids and chromosomal gene deletions were verified by Sanger sequencing.

Gene deletions

gene deletions were performed in the wild-type and Δ793:6Δ772:3 strains using homologous recombination methods established for R. rubrum3,5. In brief, all deletions were in-frame deletions except for deletion of the A3441 gene, which was a deletion and insertion of a gentamycin antibiotic resistance cassette. For each gene, 1,000 nucleotides (nt) upstream and the first 10–15 codons were amplified by PCR using primers to include a 5′ restriction site and a 3′ XbaI site. Similarly, the last 10–15 codons and 1,000 nt downstream were amplified by PCR using primers to include a 5′ SpeI site and a 3′ restriction site. PCR products were first digested and ligated together using the XbaI/SpeI sites to generate the truncated fragment, which was then digested and ligated into correspondingly digested pK18MobSacB. Plasmids were transferred to Escherichia coli stellar cells (TaKaRa) and then to R. rubrum by triparental mating with E. coli JM109:pRK2013 (ATCC 37159). First-recombination events were screened based on presence of sucrose sensitivity and kanamycin resistance (KmR) from integration of the pK18mobsacB plasmid. Second-recombination events were screened based on loss of sucrose sensitivity and KmR.

Gene complementation of marBHDK and nifB

All complementation studies utilized pBBR1-MCS3 based plasmid56, pMTAP-MCS3, which contains the R. rubrum constitutive MTA phosphorylase promoter and tetracycline resistance gene (TcR) exactly as previously used for marBHDK gene complementation in R. rubrum3,5. For each gene combination, the first gene was amplified by primers to contain an NdeI site at the ATG start codon, which correctly positioned the start codon relative to the MTA phosphorylase promoter ribosome binding site. All other genes utilized their associated ribosome binding site. All genes were amplified by PCR from the R. rubrum genome using primers and cloned into pMTAP-MCS3 using restriction sites listed in Supplementary Tables 2 and 3.

Mutagenesis and complementation of marBHDK genes

Plasmid pMTAP with the marBHDK genes is too large for amplification by PCR to directly incorporate nucleotide substitutions for site-directed mutagenesis. Therefore, marDK was digested from pMTAP-MarBHDK by KpnI and XbaI and cloned into the same sites of pUC19 (Invitrogen). Site-directed mutagenesis was performed by the Quick-Change Lightning kit (Agilent) using complementary primers. Modified marDK genes were then digested from pUC19 by KpnI and XbaI cloned into pMTAP-MarBHDK to replace the original marDK sequence.

Gas chromatography analysis of hydrocarbons

Quantification of methane, ethane and ethylene was performed using a Shimazdu GC-14A with Restek Rt-Alumina BOND/Na2SO4 column, 30 m, 0.53 mm inner diameter. Gaseous culture headspace after feeding or growth experiments was injected at 180 °C and separated isothermally at 35 °C. Eluted compounds were detected by flame ionization detector at 180 °C and identified on the basis of the retention time of methane, ethane and ethylene standard (Schott; Air Liquide). The total amount of each hydrocarbon present was calculated from the peak area as compared with standard concentration curves of the corresponding reference standard.

Overexpression plasmid for marBHDK for purification

Expression plasmids were based on R. rubrum complementation plasmid pMTAP3,5. The promoter of pMTAP-MCS3 was replaced with the R. rubrum PUF promoter. This promoter for the light harvesting complex genes pufLMB, which are next to the COR bchXYZ operon, is a strong promoter under anaerobic conditions57. The PUF promoter and marBHDK genes were cloned into pMTAP-MCS3 as described in the Supplementary Methods. This resulted in the following plasmids:

  • pPUF-MCS3

  • pPUF-M4 (pPUF plasmid with PUF promoter, marBH, PUF promoter, and marDK)

  • pPUF-MarBHis6-MarHHis6-MarDHis6-NifK (same as pPUF-M4 but with His6-tag sequences)

  • pPUF-MarHHis6

Purification of MAR

All MAR proteins produced in R. rubrum were resolved by sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS–PAGE) electrophoresis using 12% acrylamide gels (Bio-Rad), and all original gel images with cropped areas indicated are given in Supplementary Figs. 2–5.

All purification steps were performed in a Coy anaerobic chamber. Purification followed the standard published procedures for A. vinelandii nitrogenase with important aspects and differences detailed in the Supplementary Methods58. Purified protein samples were frozen by adding dropwise to liquid nitrogen and storing the resulting beads at −80 °C in anaerobic serum vials.

ICP-MS, MALDI-TOF and proteomic analysis

Rhodospirillum rubrum MarDK with MarS (A3441 protein) was resolved by SDS–PAGE, washed extensively in ultrapure water, stained with Coomassie-G stain in 40 % methanol/10 % acetic acid, destained with 40 % methanol/10 % acetic acid and washed extensively with water. The A3441 protein band was excised and analysed by in-gel trypsin digestions followed by extraction and liquid chromatography–tandem mass spectrometry quantitation by the Ohio State University Campus Chemical Instrument Center following standard protocols59. Protein A3441 was the only R. rubrum protein identified in high abundance; all other proteins of notable abundance were human keratin. MALDI-TOF of purified MarHDK and MarS proteins was performed on a Bruker Microflex LRF with sinapic acid matrix. All MADLI-TOF was performed on at least n = 2 independent sample preparations (Supplementary Fig. 2). ICP-MS of metal content was performed at the University of Georgia Center for Applied Isotope Studies. Purified MarDK + MarS was desalted by centrifugal concentration at 4 °C in 25 mM Tris pH 7.5, 50 mM NaCl treated with CHELEX resin to remove trace metals. Proteins and buffer control were digested by nitric acid and analysed by ICP-MS based on standard curves for Mg, Fe, V, Mo, Co, Cu and Ni. Analysis was performed on protein from n = 3 independent sample preparations.

MAR activity and CO inhibition assays

Activity assays in cell extracts, partially purified fractions and fully purified preparations were performed similarly to the standard purified N2ase activity assay. All experiments were performed anaerobically on a Schlenk line. Reactions of 500 µl volume were performed in 2-ml serum vials with 500 mg total cellular or soluble/insoluble protein fractions, or 13–25 µg MarH and 50 µg MarDK (with or without MarS protein) for a 1:1 or 2:1 MarH dimer to MarDK tetramer molar ratio. Reaction conditions contained 100 mM HEPES pH 7.5, 6 mM dithionite, 5 mM MgCl2, 1 mM substrate and ATP regeneration system (20 mM creatine phosphate and 0.1 mg ml−1 creatine phosphokinase). Reactions were initiated by addition of ATP by gas-tight syringe, incubated at 30 °C under nitrogen atmosphere for 5–60 min and quenched with 1/10th volume 100% trichloroacetic acid. For CO inhibition assay, the procedure was performed in the same manner, but CO was added to the nitrogen headspace at the indicated partial pressure and incubated for 1 min before initiating the reaction with addition of ATP. Protein concentration in cellular protein extracts and soluble/insoluble fractions were quantified by Bradford assay with BSA as a standard.

EPR spectroscopy

Samples under turnover and non-turnover conditions for EPR were prepared as above for activity assays with the following alterations. MarDK+A3441 as added to 21.4 mg ml−1 (70 µM) and MarH to 8.6 mg ml−1 (135 µM) in a 175-µl reaction volume containing the same buffer and ATP regeneration system as above plus 20% glycerol as a glassing agent. This corresponded to a 1:2 ratio of MarDK tetramer to MarH dimer. Similarly, MarH alone in the same conditions was added to 28 mg ml−1 (435 µM). At time t = 0, no ATP or ATP to 4 mM concentration was added, rapidly mixed and immediately transferred to a quartz EPR tube. After 30 s incubation time at room temperature, samples were frozen in liquid-nitrogen-cooled isopentane. CW X-band (9.376 GHz) EPR spectra were collected using a Bruker EMXPlus equipped with an ColdEdge cryogen-free helium cryostat and recirculation system and an Oxford Instruments MercuryITC temperature controller at both 6 and 15 K. All presented spectra were obtained using a modulation frequency and amplitude of 100 kHz and 10 G, respectively. Power- and temperature-dependent experiments were performed by adjusting the microwave power from 0.06325 to 20 mW at the indicated temperatures. Background signals were removed by baseline subtraction using IGOR Pro 9.00 (Wavemetrics). The difference spectra were generated by subtraction of the spectra of MarHDK without ATP from the spectra of MarHDK with ATP at the indicated powers. EPR spectral simulations of all spectra were carried out using the EasySpin Matlab toolbox60. Relative populations of the S = ½ and S = 3/2 signals in addition to the intermediates formed during turnover were determined by optimizing the weights of the different species at 15 K, where neither species was saturated. To simulate the inhomogeneous line broadening exhibited by the two species, g-strain values of [0.0008, 0.02, 0.055] (‘E1(H)-like’) and [0.024, 0.007, 0.045] (‘E1(H)*-like’) were included in the simulation. Spin quantitation was performed by integrating the baseline corrected experimental spectra of MarH and MarHDK without and with ATP under non-saturating conditions at 15 K. The total number of spins was then corrected to a concentration by comparison to a sample of CuII-azurin with a known concentration of 186 µM. The spin concentration was calculated based on the protein concentration of the enzyme complex to determine the spins per complex.

Cryo-EM grid preparation

Thermo Scientific Vitrobot Mark IV was used for cryo-EM specimen freezing. UltrAuFoil R1.2/1.3 grids were made hydrophilic by glow discharging in a Pelco easiGlow with 30 mA current for 2 min. Then, 3.5 µl of MarDK protein at 1.9 mg ml−1 in 25 mM Tris pH 7.4, 100 mM NaCl, 2 mM sodium dithionite was applied to a glow-discharged grid. While all protein preparations and activity assays were performed in an anaerobic chamber, the final step of cryo-EM grid preparation in the Vitrobot under 100% humidity for 3–4 s was performed in the presence of air, due to technical limitations. After this blotting step, the grid was plunged into liquid ethane cooled by liquid nitrogen. Then, the grids were stored in liquid nitrogen until examination. The brief duration ultimately exposed the sample to oxygen such that the system could be oxidatively damaged. Thus, as seen for N2ase, the catalytic M-cluster can be lost to oxidative damage and the P-cluster can be oxidatively damaged such that one or more atoms are removed28.

Cryo-EM data collection

Data were collected on a ThermoScientific Titan Krios G3i cryo-TEM operated at 300 kV and at a nominal 105,000 magnification using the ThermoScientific EPU program with a nominal defocus range of −1.0 µm to −2.0 µm. Image stack files were filtered with a BioQuantum energy filter at 15 eV slit width and acquired with a Gatan K3 Direct Electron Detector (Gatan) in Counted Super-Resolution mode at a pixel size of 0.4125 Å. Total electron dose was 50 e− Å−2, and each stack file consisted of 40 frames.

Cryo-EM image processing

Images were processed exclusively using CryoSPARC61 as shown in Extended Data Fig. Fig 4. The dataset was subjected to patch motion correction with an output F-crop factor of 1/2 resulting in a pixel size of 0.825 Å. Patch contrast transfer function estimation and contrast transfer function cut-off with vision screening at 4.05 Å were performed (removing less than 1% of micrographs, suggesting the high quality of data). Blob Picker was then performed, and particles were extracted into 64 × 64-pixel boxes at a pixel size of 3.3 Å, followed by two rounds of two-dimensional (2D) classification to generate templates for Template Picker. Nine unique classes representing unique views of the second 2D classification were selected as templates and applied to the selected micrographs using Template Picker. Particles were then inspected and extracted into 64 × 64-pixel boxes at a pixel size of 3.3 Å, followed by 2D classification. Ab initio three-dimensional (3D) reconstruction of ten classes was performed using selected particles, followed by heterogeneous refinement. The particles of the best class were used for further ab initio 3D reconstruction with three classes, followed by heterogeneous refinement. Three-dimensional classification of ten classes with a target resolution of 6 Å was applied to the best class of the ab initio reconstruction, and five of the ten 3D classes that showed complete structures were selected. The particles of the selected 3D classes were reextracted into 400 × 400-pixel boxes at a pixel size of 0.825 Å. Ab initio 3D reconstruction with three classes and heterogenous refinement was subsequentially applied to the reextracted particles. Two classes that showed complete structures were selected, combined and processed by non-uniform refinement, yielding the final map at 2.35 Å resolution based on the Fourier shell correlation global resolution plot at a Fourier shell correlation of 0.143 generated by CryoSPARC from final electron density map in Extended Data Fig. 4 (ref. 61).

MarDK model building and structure refinement

An ab initio model of a tetrameric MarDK complex was generated with AlphaFold-multimer (v.3)62 as implemented in the AlphaFold colab notebook (v.1.5.2)63. N-terminal residues with a predicted random coil type conformation were truncated and the resultant model was placed into the cryo-EM map with phenix.dock_in_map64. Following model placement, discrepancies between the observed map and the predicted model were corrected with molecular dynamics flexible fitting using the ISOLDE plugin65 in ChimeraX66. The MarDK model was further refined with phenix.real_space_refine67 in combination with iterative real space refinement in Coot68. The final MarDK model was validated with MolProbity as implemented in the PHENIX software package64.

For cluster modelling, P-cluster restraints were downloaded from the PDB in CIF format (ligand ID: CLF; https://www.rcsb.org/ligand/CLF), and phenix.elbow was used to convert the restraints, without geometry modification, into a format suitable for phenix.real_space_refine67. To reflect the oxidation state of the P-cluster, the Fe #8 atom was removed along with its angle and bond length restraints. The S1/FE6 bond length restraint was increased by 0.1 Å. The complete oxidized P-cluster restraint file used for model refinement is included as Supplementary Data 4. In addition, custom tetrahedral bond length and angle restraints for phenix.real_space_refine were defined between the sulfur atom of MarDK Cys residues and P-cluster Fe atoms to have an ideal bond length of 2.3 Å with a sigma value of 0.1 Å. These restraints are included in Supplementary Data 5.

For AlphaFold 3 models, Chai-1 was utilized to predict the location and coordination of a 4Fe-4S cluster, P-cluster (likely mar1 cluster) and L-cluster (likely mar2 cluster)69, with SMILES strings of S12[Fe](S3[Fe]11)S([Fe]22)[Fe]3S12, S12[Fe](S([Fe]22)[Fe]3S42)S3([Fe]23)([Fe]14)([Fe]1S22)[Fe](S33)S1[Fe]23 and S12[Fe](S([Fe]2S23)[Fe]33S4)(S33([Fe]567)([Fe]12S7)[Fe]14S62)S[Fe]3(S53)S1[Fe]32 used to represent the clusters, respectively. Bolz-1 and Chai-1 could not render a true L-cluster from a SMILES string, so an ‘L-cluster’ with a central sulfide was used and then the central sulfide replaced with a carbide.

Reporting summary

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