Streptomyces thermoautotrophicus does not fix nitrogen

Streptomyces thermoautotrophicus UBT1 has been described as a moderately thermophilic chemolithoautotroph with a novel nitrogenase enzyme that is oxygen-insensitive. We have cultured the UBT1 strain, and have isolated two new strains (H1 and P1-2) of very similar phenotypic and genetic characters. These strains show minimal growth on ammonium-free media, and fail to incorporate isotopically labeled N2 gas into biomass in multiple independent assays. The sdn genes previously published as the putative nitrogenase of S. thermoautotrophicus have little similarity to anything found in draft genome sequences, published here, for strains H1 and UBT1, but share >99% nucleotide identity with genes from Hydrogenibacillus schlegelii, a draft genome for which is also presented here. H. schlegelii similarly lacks nitrogenase genes and is a non-diazotroph. We propose reclassification of the species containing strains UBT1, H1, and P1-2 as a non-Streptomycete, non-diazotrophic, facultative chemolithoautotroph and conclude that the existence of the previously proposed oxygen-tolerant nitrogenase is extremely unlikely.

N 2 :H 2 :CO 2 :O 2 . The strains grow more slowly on a 9:8:2:1 mix of CO:N 2 :O 2 :CO 2 , reaching maximal growth within four days of incubation at 60 °C. Limited germination could be observed on some carbon sources in the absence of H 2 or CO, but the only carbon source found to sustain growth reproducibly was pyruvate. Growth in liquid was poor, either in suspension in pyruvate or as a pellicle in the presence of H 2 or CO. Serial passage on mineral media solidified with gellan gum and in the presence of H 2 or CO was robust, as was serial passage on plates containing pyruvate and NH 4 (DSM260 media), suggesting that the initial assessment of this species as an obligate chemolithoautotroph was premature.
The strains did not grow above 80 °C or below 42 °C. H1 and UBT1 form a non-fragmenting, branched substrate mycelium and, after at least 24 hours of growth, produce aerial hyphae that septate into chains of Electron microscopy shows smooth, branching substrate mycelia, and hairy/rugose decorations on mature spores, which are arranged in straight, unbranched chains ((C), magnification is 6,810x). Spore formation is robust on media containing NH 4 Cl ((D), magnification is 3,600x), but when grown on media lacking NH 4 Cl the ratio of substrate to aerial mycelia is higher, and the hyphae assume varied morphologies ((E), magnification is 3,290x). In cross section, unknown vacuoles and other subcellular features appear, and may be more prominent in mature spores (lower center, identifiable by rugose exterior) than in other hyphae ((F), magnification is 20,000x).
Scientific RepoRts | 6:20086 | DOI: 10.1038/srep20086 grey-pigmented spores, the most mature form of which bear hairy to rugose surface decorations 23 . Substrate hyphae, aerial hyphae, and free spores are all highly hydrophobic, and remain buoyant atop aqueous solutions.
S. thermoautotrophicus strain H1 was isolated from soil obtained from a charcoal pile in Hasselfelde, Germany, by adding soil particles to liquid media under a hydrogen atmosphere, and transferring the resulting buoyant mycelia after 9 days of culture. The original liquid phase enrichment of S. thermoautotrophicus from soil retained spores that formed non-mycelial colonies, which grew well on complex media. 16S rRNA analysis of these colonies indicated that they were primarily related to thermophilic Firmicutes, including Bacillus, Brevibacillus, and Geobacillus. Interestingly, liquid phase growth of this initial consortium resulted in extensive pellicle growth, which was lost upon subsequent purification of S. thermoautotrophicus after serial passage and selection of single colonies on plates. Growth of the selected strain on plates remained robust, however.
S. thermoautotrophicus strain P1-2 was isolated from soil situated near an active coal seam fire in Centralia, PA, USA. Its nutritional requirements were identical to strain H1. PCR with universal RNA primers (27F and 1492R 24 ) produced a single amplicon, Sanger sequencing of the amplicon indicated 99.3% identity with one of the 16S rRNA loci present in both genomes from the H1 and UBT1 strains ( Supplementary Fig. 1).
Genome sequences indicate a novel genus. Streptomyces thermoautotrophicus strain UBT1 was grown and sequenced independently at two different facilities (Rosario, Argentina, and Aachen, Germany), and S. thermoautotrophicus strain H1 at a third (Boston, MA, USA; Fig. 2). Optical mapping data (OpGen, Gaithersburg, MD) were collected for strain H1, and indicated the presence of a single, circular chromosome of 5.25 Mbp. A combination of short reads (Illumina), scaffolding reads (Pacbio), and optical mapping consolidated genome assembly of strain H1 into a single scaffold 4.90 Mbp in length (of which 4.5 Mbp is sequenced data, and 0.4 Mbp is scaffolding Ns), containing an estimated 85% of the genome, with 5 additional contigs > 1 kbp in length that could not be aligned to the map (and that could collectively account for another 1.3% of the genome), and one 48 kbp contig that is likely a plasmid, based on the presence of multiple genes encoding phage-and plasmid-related putative functions. The genome sequences of strain UBT1 are more fragmentary, but have similar lengths and gene content, including the possession of a 55 kbp plasmid (which shares no significant similarity with that of the putative H1 plasmid). The genome characteristics are compared to those of other Actinomycetes in Supplementary Table 1. Comparison of the H1 and UBT1 (Aachen) genomes by the Genome Blast Distance Phylogeny (GBDP) approach, as implemented with default settings and bootstrapping on a web server (http:// ggdc.dsmz.de; date of access 21/06/2015) 25 yielded a distance estimate of 0.0067, comparable to a DNA-DNA hybridization (DDH) value of 94.9%, predicting a 97.18% probability that DDH > 70%, and thus that the strains belong to the same species.
The circularity and relatively modest size of the H1 genome are uncommon among Streptomycetes 26 , and accordingly we undertook analysis of the nearest relatives of these strains. The UBT1 (Aachen) genome assembly contains three small subunit ribosomal RNA genes, while the more fragmentary Rosario genome assembly contains two. The H1 strain genome assembly similarly has two. Each genome contains two different 16S gene  20 scaffolds from the SPAdes assembly were aligned to the optical map, and gaps in the chromosome were filled in with N's manually. Six nodes did not align; the largest of which is a predicted plasmid. The two outer rings represent genes on the two strands of the chromosome. The inner ring shows the GC content, and prominent dips represent gaps in the assembled sequence; with length and position inferred from the optical map. sequences that share only 90% identity between them. Between the H1 and UBT1 strains, however, each 16S sequence in one strain matches the corresponding sequence from the other strain with 99% identity. The presence of significantly different ribosomal RNA operons within a single bacterial genome has been observed before, especially in thermophiles 27 . A phylogenetic tree comparing the 16S sequences shows that one of the 16S genes clusters most closely with Thermobispora bispora and Acidothermus cellulolyticus (Supplementary Figure 2a). The identities with either species are 90%. The other 16S sequence in either genome has various nearest neighbors among other actinomycetes, but shares less than 93% identity over 93% of its length with any of them, as determined by BLAST.
In the absence of conclusive evidence as to which of these divergent rRNA sequences is ancestral to the organism, and which may be the product of a more recent horizontal gene transfer or duplication event, we adopted the approach of concatenating 14 well-conserved proteins (13 ribosomal proteins and one phosphpatidate cytidylyltransferase) within the genome sequences of strains H1 and UBT1, and performing alignment and neighbor-joining analyses with similar sequences from all Actinobacteria with fully sequenced genomes that were available at the time of analysis 28 . The resulting tree, while possessing some nodes with low confidence based on the supporting bootstrap values, is largely in agreement with recent results that have followed the same approach 29 (Supplementary Figure 2b). From this approach, it appears that H1 and UBT1 are most closely related to the clade that includes the genera Acidothermus, Streptosporangium, Thermobifida, Thermobispora, and Thermomonospora. By this analysis, these strains do not belong in the genus Streptomyces, and instead are in the class Actinobacteria, subclass Actinobacteridae, order Actinomycetales. The suborder, family, and genus are all undetermined, although nearby families include Frankineae (Acidothermus), Streptosporangineae (Thermomonospora, Streptosporangium, Nocardiopsis, Thermobifida), and Pseudonocardineae (Thermobispora).
Genomes lack nitrogenases; possess multiple CODHs. No nif genes are present in any of the three draft genomes, other than a single gene annotated as "Nitrogen-fixing NifU, C-terminal:Rieske [2Fe-2S] region". This single gene is located between a predicted uptake hydrogenase and factors for the maturation of that enzyme, which is homologous to similar clusters in other organisms. Proteins with C-terminal homology to NifU have predicted functions in metal cofactor synthesis, including in non-diazotrophs, and are not diagnostic for the presence of a functional nitrogenase 30 . The genome also contains no sequences with significant homology to anf or vnf genes for the alternative Fe-or V-nitrogenases.
The superoxide-dependent nitrogenase was previously identified to involve two molybdenum hydroxylases, St1 and St3, which were homologs of CODH. St1 was reported to be the putative dinitrogen reductase, and St3 a true CO dehydrogenase responsible for generation of the superoxide anion radical. Both were reported to possess the same heterotrimeric structure of typical CODH enzymes or other molybdenum hydroxylases 22 . The UBT1 genome contains four operons encoding predicted aerobic CODH, and the H1 genome contains three ( Table 2). Two of the UBT1 operons contain large subunit (coxL) genes encoding the AYXCSFR motif characteristic of form I aerobic CODH 31 , and are located near coxDEF or coxG accessory genes necessary for the maturation of the functional CODH complex. The other two loci contain large subunits with alternative motifs that suggest categorization with form II CODH, or other molybdenum hydroxylases 32 . The medium subunits of all four loci have sequences consistent with the ability to bind flavin adenine dinucleotide (FAD) cofactor, despite the previous finding that St1 (encoded by sdnM) does not contain FAD 22 LI90_1082  TH66_09435  TR74_17685  --100  44   CoxS  LI90_1083  TH66_09430  TR74_17680  --100  61   CoxL  LI90_1084  TH66_09425  TR74_17675  --100  62   CoxL  LI90_1777  TH66_03290  TR74_06615  --100  39   CoxM  LI90_1775  TH66_03330  TR74_06625  --99  50   CoxS  LI90_1776  TH66_03295  TR74_06620  --100  35   CoxM  LI90_1945  TH66_19340  TR74_24565  --100  48   CoxS  LI90_1944  TH66_19345  TR74_24560  --100  30   CoxL  LI90_1946  TH66_19350  TR74_24570  --100  35   CoxM  -TH66_01980  TR74_16700  --100  48 CoxS  (Fig. 3), suggesting that the UBT1 strain sequenced here is the same strain as reported on in that period. The identity between the N-terminal protein sequences published for the St1 proteins with any of the UBT1 or H1 CODH sequences, however, is low (maximum identity between any H1 or UBT1 CODH protein and St1M, S, and L respectively is 53%, 50%, and 35%) 21 . Nucleotide sequences coding for the St1 subunits, based on sequencing clones from a library of DNA that hybridized with degenerate oligonucleotides derived from the N-terminal St1 sequences, have previously been determined 33 . These sequences have also recently been uploaded to the GenBank database (GI: 589823865, 589823867, and 589823869, for sdnMSL, respectively). The highest identity of any of the proteins encoded by these genes to any of the H1 or UBT1 sequences at the amino acid level is 50%, 63%, and 62%, for the proteins encoded by sdnM, S, and L, respectively (Table 2). These proteins in our S. thermoautotrophicus genomes are all annotated as CoxM, S, and L proteins.
The St2 protein was previously identified as a manganese-dependent superoxide dismutase (Mn-SOD), encoded by the sdnO gene. Similarly to St1, a nucleotide sequence for the entire sdnO gene had been recorded 33 , and has recently been uploaded to GenBank (GI:588295007). The UBT1 and H1 genomes each contain one predicted Fe/Mn-SOD; identity with the St2 protein sequence encoded by the reported sdnO gene sequence is 38% at the amino acid level ( Table 2).

Superoxide-Dependent Nitrogenase genes match Hydrogenibacillus schlegelii. The proteins
encoded by sdnMSL and sdnO have higher identities, however, with proteins from other strains which have partial sequences within the GenBank database. A BLAST search revealed that the closest match for sdnL was to a partial coxL gene sequence (GI:38679249) in Hydrogenibacillus schlegelii (formerly Bacillus schlegelii), for which no full genome sequence was available. In order to investigate the relationship between the sdn sequences and H. schlegelii, we undertook sequencing the genome of that organism as well, and report here a draft sequence for the genome of H. schlegelii strain DSM2000. This strain is a known chemolithoautotroph, with similar optimum temperature for growth as S. thermoautotrophicus and similar capacity to grow on CO or H 2 . Accordingly, the draft genome contains genes encoding ribulose bisphosphate carboxylase/oxygenase (RuBisCo), phosphoribulokinase (PRK), CODH and uptake hydrogenase. H. schlegelii has not been described as a diazotroph. Its genome lacks nif/anf/vnf nitrogenase genes, and we found that the strain failed to grow on media lacking NH 4 Cl, with no visible turbidity in cultures incubated for up to one month, either when grown chemolithotrophically on H 2 /CO 2 , or when grown heterotrophically on pyruvate.
The DSM2000 genome contains a single locus encoding a putative molybdenum CODH operon (TR75_12445-55) with > 99% identity with the sdn operon at the nucleotide level. Translations of the open reading frames (ORFs) within also demonstrated > 99% identity to the St proteins (Table 2). It was further determined that a Mn-SOD within the H. schlegelii genome (TR75_10445), bore 100% identity at the amino acid level to the St2 protein (Table 2). Extensive overlap was also found at the nucleotide level between the DSM2000 genome sequence and the region around the coding sequence for sdnO (99.62% identity, over 7.2 kbp), previously sequenced by the Meyer group.
Strains grow on media lacking NH 4 Cl, but fail to incorporate significant 15 N 2 into biomass. All S. thermoautotrophicus strains tested were capable of continuous culture on mineral media lacking added NH 4 + but peculiarities regarding their growth were apparent. Specifically, growth was far more robust in the presence of 30 mM NH 4 + . Furthermore, growth was only supported on plates solidified with 0.5-1.0% gellan gum, or on agar fortified with charcoal or bio-char. A CHN analysis on the gellan gum used returned a nitrogen content of 0.095%. Growth on mineral media solidified with Noble agar was not observed. In addition, growth on plates containing either gellan gum or agar plus bio-char was greatly enhanced by the presence of plates containing 30 mM NH 4 + within the same atmosphere, presumably due to the former cross-feeding on NH 3 volatilizing from the latter.
The high number of CODH enzymes encoded by the S. thermoautotrophicus genomes and the purported identity of the St1 nitrogenase as a modified CODH prompted us to examine other published genomes containing high numbers of such enzymes. We identified Pseudonocardia dioxanivorans CB1190 as another strain whose sequenced genome encodes multiple molybdenum hydroxylase enzymes (10 operons, two of which are putative type I CODH enzymes 34 ), and which has been reported to be a diazotroph 35 . The strain CB1190 similarly lacks nif, anf, and vnf genes, yet reportedly displayed a similar ability to grow on media lacking NH 4 + as an additive. This strain was accordingly included in subsequent experiments to test for diazotrophy using isotope labeling.
In order to determine whether the growth of our strains was independent of exogenous combined nitrogen, we undertook labeled nitrogen assays at four independent laboratories (Table 3). Plates were inoculated with spores of the strains indicated and incubated in the presence of acid-washed 15 N 2 gas. After 3-5 days, biomass was scraped from plates and subjected to isotope-ratio mass spectrometry (IRMS). At three sites with thorough 15 N 2 gas washing, no enrichment of 15 N was detected in biomass relative to plates supplemented with NH 4 Cl. At one site, a slight enrichment of 15 N in biomass was observed. We note, however, that the 15 N 2 gas used in this experiment (Aldrich catalog #364584, Lot#SZ1670V) was found in a recent report 36 to contain significant amounts of contaminating 15 NH 4 + and 15 NO 3 − /NO 2 − . The small amount of 15 N detected in the biomass of our samples, is much less than the amount that would be present in unwashed gas ( Supplementary Fig. 3); thus it is possible that not all of this contaminating fixed 15 N was captured by washing for 30 minutes in 5% H 2 SO 4 . Fixation was also not detected in P. dioxanivorans CB1190.

Discussion
We conclude that there is no evidence for, and substantial evidence against, the existence of an oxygen-tolerant nitrogease as previously described in Streptomyces thermoautotrophicus. Our major empirical lines of evidence are both genomic and phenotypic. First, using genome sequencing, we failed to find either canonical or the proposed novel nitrogenase enzymes in our S. thermoautotrophicus isolates. Rather, sequences of the proposed enzymes are present in another thermophilic, non-diazotrophic bacterium. Second, extensive multi-site fixation experiments using 15 N 2 gas failed to demonstrate incorporation into biomass. Growth experiments suggest that S. thermoautotrophicus strains are highly effective nitrogen scavengers, but not nitrogen fixers.
Finally, we point out some other inconsistencies in the original characterization of the proposed oxygen-tolerant nitrogenase system. Apart from the superoxide dependent nitrogenase, no known aerobic reduction of nitrogen to ammonia is known. An unprecedented mechanism in the scheme is the use of superoxide as an  electron donor for a biologically productive reaction. No other productive biological use of superoxide is known, although many cells produce superoxide as a toxin. Usually superoxide is a toxic byproduct of cellular metabolism. With a standard potential of − 330 mV 37 it is comparable to the NADH or NADPH potential of − 320 mV, but much more toxic due to its reactivity. The Fe protein of the MoFe nitrogenase has an apparent midpoint potential of − 470 mV 38 , and the physiological electron donor is ferredoxin or flavodoxin. Therefore it seems unlikely that superoxide is sufficiently reducing to drive the nitrogenase reaction. In addition, none of the putative superoxide dependent nitrogenase system proteins have a plausible ATPase domain identified, even though ATP hydrolysis was required for the activity. In the original work, labeled nitrogen was not used during the purification of the nitrogenase, instead an ammonia production assay 21 was used, based on the reaction of ammonia with sodium phenate and hypochlorite 39 . This assay is very sensitive, but is known to have a high background 40 , which may have contributed to a misidentification of nitrogenase activity. There is absolutely no genomic evidence that S. thermoautotrophicus is diazotrophic. No MoFe, VFe, or Fe-only nitrogenase genes are present in the draft genome sequences presented here. The UBT1 strain was previously found to lack nif genes by dot blotting 19 , and its putative oxygen-tolerant nitrogenase was characterized by protein lysate fractionation and assays for ammonia production in vitro 21 . The identities of the three protein complexes of the proposed oxygen-tolerant nitrogenase were previously found to be two variants of molybdenum hydroxylases related to CODH, and an SOD-like protein 21,22 . However, N-terminal sequences of the proposed heterotrimeric St1 nitrogenase, and of the homodimeric St2 nitrogenase reductase, as well as the sequences of their respective genes sdnMSL and sdnO, do not match any sequences present in the draft genomes of strain UBT1 or the novel isolate H1 21,22 .
The genes for the proposed oxygen-tolerant nitrogenase are in fact present in the genome of another thermophilic non-diazotrophic bacterium. The sequences for sdnMSL and sdnO match with identities between 99.1% and 100% at the nucleotide level the coxMSL and sod gene sequences from Hydrogenibacillus schlegelii DSM2000, an isolate closely related to H. schlegelii DSM9132, which was studied by the Meyer group in the same period as S. thermoautotrophicus [41][42][43] . S. thermoautotrophicus and H. schlegelii overlap closely in their nutritional requirements and optimal growth temperature, and contamination of cultures of S. thermoautotrophicus is therefore a possibility. If the Meyer group did contaminate their liquid culture of S. thermoautotrophicus with H. schlegelii, these are the sequence data we would observe. We conclude that the St1 and St2 proteins are in fact the aerobic CODH and superoxide dismutase proteins from a strain of Hydrogenibacillus schlegelii.
In addition to the absence of the putative oxygen-tolerant nitrogenase genes, our extensive phenotypic characterization does not support the claim the S. thermoautotrophicus is diazotrophic. The strain of S. thermoautotrophicus we have received from D. Gadkari represent the closest known link to that which was claimed to be diazotrophic in 1992. Independent samples of that strain were cultured by different personnel at two independent sites, found to grow poorly on media lacking NH 4 Cl, grew reproducibly only on media containing additives that may contribute combined nitrogen, and did not incorporate 15 N 2 into biomass. We fail to see convincing evidence for diazotrophy in this strain. In addition, strain H1, isolated from a burning charcoal pile using procedures detailed for the isolation of UBT1 by the original authors 20 , and strain P1-2, isolated from soil overlying a coal seam fire, share morphology and physiology with UBT1 that are indistinguishable by any test performed thus far. They qualify as members of the same species as UBT1, according to whole-genome prediction of DNA-DNA hybridization and 16S rRNA sequence identity for strains H1 and P1-2, respectively. While there is no guarantee that independent isolates of a particular species of bacteria will share identical metabolic capabilities, it is striking that the H1 and P1-2 strains, which were not cultivated for prolonged periods in the laboratory before being subjected to 15 N 2 incorporation assays, show a similar capacity for minimal but reproducible growth on media lacking added NH 4 Cl, yet also fail to show significant incorporation of acid-washed, labeled dinitrogen into biomass.
Similar to our results in S. thermoautotrophicus, the abundance of CODH paralogs in the genome of P. dioxanivorans failed to predict functional diazotrophy in that strain, as measured by 15 N 2 assays. Subsequent dialogue with the authors responsible for the isolation and initial characterization of the P. dioxanivorans species confirmed that, contrary to early reports, the sequenced strain lacks known nitrogenase genes, and that the strain now available is not diazotrophic (L. Alvarez-Cohen, personal communication).
Other researchers have identified bacteria capable of growth on trace environmental sources of combined nitrogen 44,45 , and previous claims of diazotrophy in various strains have subsequently been found to be premature 46,47 . A high-affinity pathway for assimilation of scarce combined nitrogen may prove a viable alternative to the maintenance of a complex pathway for nitrogen fixation when ammonia and nitrate are scarce. Efficient scavenging of combined nitrogen is certainly a more parsimonious explanation for our data than a novel nitrogenase that is orthogonal to all known systems and insensitive to oxygen.
While it remains possible that novel nitrogenases will be uncovered as the resources for querying the genomes of unculturable bacteria advance, any such enzymes must be supported by sensitive functional assays in order to verify their activities. Among such assays, measuring the incorporation of isotopically-labeled dinitrogen gas into biomass by mass spectrometry remains the most sensitive and specific 48,49 . The high sensitivity of isotope-ratio mass spectrometry, however, can also be a liability, as when the input reagent may be contaminated with fixed 15 N, a fact that may have contributed to the confusion regarding the status of UBT1 and other strains as putative diazotrophs.
We cannot rule out the possibility that the strains used in the previous publications cited possessed a novel nitrogenase, which was subsequently lost upon prolonged cultivation in the laboratory. Any such enzyme, however, would likely not be constituted of subunits with the sequences previously published, as their presence in H. schlegelii does not correspond with the ability to fix nitrogen, nor does their absence in UBT1 limit its ability to grow on media containing no added NH 4 Cl. Should a viable sample of diazotrophic UBT1 be recovered, we welcome the chance to characterize it further. However, given the evidence in hand we conclude that the existence of the proposed oxygen-tolerant nitrogenase system is extremely unlikely. Regarding the reassignment of the phylogeny of this species: the identity between the 16S rRNA sequences in the H1, UBT1, and P1-2 strains is sufficiently high to warrant their classification as members of a single species. When compared to other Actinobacteria, however, identity falls below the commonly suggested guidelines for placing these strains within a known genus 50 . More sensitive protein sequence alignments support this divergence, and suggest an early-branching member of the order Streptosporangiales. It is unclear at the current level of resolution available from full genome sequences whether these strains best fit as members of the family Nocardiopsaceae, Thermomonosporaceae, or a new family. In any case, the level of divergence from categorized species is sufficient to recommend that Streptomyces thermoautotrophicus be reassigned to a new genus, and described as a non-diazotroph on the basis of this work.
S. thermoautotrophicus strain H1 was isolated from soil taken from a charcoal pile in Hasselfelde, Germany. Soil particles were immersed in mineral media 20 and incubated in the presence of H 2 and CO 2 at 60° C. Pellicles emerged after several days, and were transferred and enriched on plates of mineral media plus 0.5% gellan gum (Sigma) and 0.075% MgCl 2 *6H 2 O in a defined atmosphere of 40% H 2 , 10% CO 2 , and 50% air, or else in 45% CO, 5% CO 2 , and 50% air 51 .
S. thermoautotrophicus strain P1-2 was isolated from soil located above an underground fire where a coal seam has been burning for several decades (Centralia, PA) 52 . The soil temperature in situ at the time of collection was 60 °C, and the collection site was immediately adjacent to steam issuing from a fissure. Isolation proceeded as for the H1 strain, except that isolation was conducted at an independent facility, liquid enrichment was in DSMZ261 mineral media 53 , and plates were solidified with 1% gellan gum and 0.15% MgCl 2 .
Pseudonocardia dioxanivorans CB1190 was grown on ATCC medium 196 35 for routine maintenance, and was grown on NFIX mineral media plus glucose and gellan gum for labeled nitrogen assays. Azotobacter vinelandii strains DJ and DJ100 (Δ nifD) 54 were grown on Burk's medium 55 with or without 30 mM NH 4 Cl for maintenance and labeled nitrogen assays. P. dioxanivorans and A. vinelandii were grown in air at 30 °C. Hydrogenibacillus schlegelii DSM2000 was grown on DSMZ Medium 260 or 261 56 , for heterotrophic or chemolithotrophic growth, respectively. To test for diazotrophy in this strain, variants of these media to omit NH 4 Cl were also prepared. H. schlegelii was grown at 55 °C in either air or an atmosphere of 45% H 2 , 10% CO 2 , and 45% air.

Labeled nitrogen assay.
In all experiments, separate canisters of 15 N 2 gas were purchased from Aldrich chemical (St. Louis, MO). The gas was acid-washed prior to introduction to the culture vessels: 15 N 2 gas was retrieved from the canister and injected into a Schott bottle filled with water and sealed with a rubber septum. Water was withdrawn upon introduction of the gas to equalize pressure and allow continued introduction of the gas until approximately half the volume of the bottle was occupied by the gas. Then concentrated H 2 SO 4 was introduced to bring the concentration of H 2 SO 4 in the water within the bottle to 5%. The gas was incubated for the indicated period of time with agitation before being extracted and introduced to the culture chamber.
At Harvard Medical School: 15 N 2 gas was acid-washed for 30 minutes. Plates of NFIX mineral media possessing or lacking 1.5 g/L (28 mM) NH 4 Cl and solidified with 0.5% gellan gum were inoculated with S. thermoautotrophicus H1 or UBT1 and incubated in the presence or absence of 2.5% 15 N 2 for 3 days. Then biomass was collected, dried at 80 °C for 12 hours, and analyzed at the MBL Stable Isotope Laboratory (Woods Hole, MA) on a Europa 20-20 CF isotope ratio mass spectrometer equipped with a Europa ANCA-SL element analyzer. A. vinelandii was grown in suspension in Burk's medium possessing or lacking 1.5 g/L NH 4 Cl, with stirring, at 30 °C for 3 days.
At the RWTH Aachen: 15 N 2 gas was acid-washed for one hour. Plates of NFIX mineral media lacking 1.5 g/L (28 mM) NH 4 Cl and solidified with 1.0% gellan gum were inoculated with S. thermoautotrophicus H1 or UBT1 and incubated for five days in the presence of 10% 15 N 2 gas. Then biomass was collected, dried at 80 °C overnight, and sent for analysis to ISO-analytical, Crewe, UK.
At Universidad Nacional de Rosario: 15 N 2 gas was acid-washed for 30 minutes. Plates of NFIX mineral media possessing or lacking 1.5 g/L (28 mM) NH 4 Cl, fortified with 0.5% activated charcoal, and solidified with 1.5% agar were inoculated with S. thermoautotrophicus H1 or UBT1 and incubated at 60 °C for 5 days in desiccators in the presence or absence of 1.5% 15 N 2 . A. vinelandii DJ and DJ33 57 (nifD − nifK − )were grown in Burk's medium with (DJ33) or without (DJ) 1 g/L NH4 Cl for 3 days at 28 C in desiccators in the presence or absence of 0.75% 15  NH 4 Cl were each inoculated with UBT1, H1, and P1-2 strains. These plates were incubated for 3 days in 40% H 2 , 10% CO 2 , and 50% air atmospheres with 2% of the atmosphere replaced with acid-washed 15 N 2 . We also incubated plates of Azotobacter vinelandii (wild type) and A. vinelandii (nifD-mutant), which cannot fix nitrogen in Mo-containing media, as positive and negative controls, respectively. Controls were incubated on mineral media with 1% glucose either with (A. vinelandii nifD − ) or without (A. vinelandii WT) NH 4 Cl in the presence of an air atmosphere with ~2% of the atmosphere replaced with acid-washed 15 N 2 . After incubation, growth was scraped from plates and washed in Phosphate buffered saline to remove potential contaminating nitrogen. Two NH 4 -free plates worth of biomass were pooled into single samples at this point to compensate for poor growth of H1, UBT1, and P1-2 strains on these media. Biomass was dried at 60 °C for 24 hours, then analyzed for 15 N content on a GV instruments Isoprime Mass Spectrometer interfaced with a EuroVector Elemental Analyzer 3000 Series.
Sequencing and bioinformatic analyses. At Universidad Nacional de Rosario: genomic DNA from S. thermoautotrophicus UBT1 (Rosario), was extracted with cetyltrimethylammonium bromide 58 . DNA was fragmented by nebulization, ends repaired, adaptors ligated, and small fragments removed following GS-FLX Titanium library preparation. Libraries were sequenced with single-end FLX 454. The reads were assembled with Newbler v.2.5.3 59 , and the genome annotated by RAST.
At the RWTH Aachen: genomic DNA from S. thermoautotrophicus UBT1 (Aachen), was prepared with a NucleopSpin Tissue kit (Macherey-Nagel, Germany). DNA was fragmented using a Biorupter Pico (Diagenode, Belgium), and libraries were prepared with the TruSeq kit. Libraries were sequenced with the 250PE reagent kit on a MiSeq (Illumina). PacBio reads were generated by GATC (Germany). Illumina reads were trimmed for quality with Trimmomatic (V0.32) 60 , and Illumina and PacBio reads were assembled with SPAdes version 2.5.1 61 . The resulting assembly was annotated with RAST.
At Harvard Medical School: genomic DNA from S. thermoautotrophicus strain H1 was isolated with the DNeasy blood and tissue kit (Qiagen). Quality was verified with Bioanalyzer chips (Agilent). DNA was fragmented by adaptive focused acoustics (Covaris) to 400 bp or 3 kb, and libraries prepared with the TruSeq kit. Libraries were sequenced with 250 PE reagent kits on a MiSeq (Illumina). PacBio reads were generated with the SMRTbell Template Prep Kit. Reads were trimmed for quality with Trimmomatic 61 , and assembled with SPAdes 3.1 61 . The resulting assembly was annotated with RAST 28 , augmented with a bidirectional BLAST 62 search against all bacterial protein sequences from the UniProt database 63 to increase the number of genes with a putative function.
At Michigan State University: genomic DNA from H. schlegelii DSM2000 was obtained directly from DSMZ, and was also extracted from H. schlegelii DSM2000 grown in Na-pyruvate mineral media (DSM 260: 4.5 g/L Na 2 HPO 4 , 1.5 g/L KH 2 PO 4 , 0.01 g/L MnSO 4 *7H 2 O, 0.2 g/L MgSO 4 *7H 2 O, 0.01 g/L CaCl 2 *2H 2 O, 0.005g/L Ferric citrate, 1.0 g/L NH 4 Cl,1.50 g/L Na-pyruvate, and 3.0 mL DSM trace element solution 6) without shaking at 60 °C. Cells were pelleted, incubated for 30 minutes at 55 °C with 280 ul Tissue Cell Lysis solution (Epicentre) and with 20 ul Proteinase K (Roche). 200 ul MPC protein precipitation solution was then added, the supernatant was transferred to isopropanol, and washed with 70% ethanol. Libraries were prepared for sequencing from both sources of DNA. Genomic DNA was sheared on a Covaris S2 ultrasonicator using the manufacturer's recommended parameters for shearing to 500 bp. Sheared DNA was end repaired, A-tailed, and Illumina adapters were ligated using enzymatic kits from New England Biolabs 64 . Ligation products were amplified for 15 PCR cycles with Phusion polymerase (NEB) and size-selected on a 2% agarose gel. Finished libraries were pooled and sequenced on an Illumina MiSeq 150 bp paired sequencing run. Quality-filtered reads were assembled with the A5 pipeline 65 , and the resulting assembly was annotated with RAST.
The CoxMSL operon sequence from H. schlegelii DSM2000 was completed using the polymerase chain reaction (PCR). A DreamTaq kit (Thermo) was used to amplify for 35 cycles, and the amplicon was gel-purified and sequenced on an ABI 3730XL with BigDye chemistry. A quality-trimmed Sanger read was used to merge the two scaffolds in Phrap 64,66 .
Phylogenetic analyses were performed on 16S rRNA genes by downloading sequences from the NCBI, RefSeq Targeted Loci Project (revised Nov. 2010 at http://www.ncbi.nlm.nih.gov/genomes/static/refseqtarget.html; date of access 23/08/2015). 352 sequences representing Actinobacteria with sequenced genomes available were retrieved, as well as B. subtilis as an outgroup. These were aligned to the 16S sequences from the S. thermoautotrophicus genomes using Clustal Omega version 1.2.1 67 with default parameters and 30 iterations of sequence input order. The resulting alignment was submitted to the Gblocks server 68 , and processed with the default settings. The consolidated alignment was used to form a distance matrix and a neighbor-joining tree with 100 bootstrap replicates, which were then used to form a consensus tree; these operations were carried out with PHYLIP version 3.695 69 . The final tree was edited in MEGA6 70 : for the condensed tree, branches that were monophyletic according to their current classification in NCBI Taxonomy were collapsed up to the family level.
Phylogenetic analyses were performed on protein sequences by identifying highly conserved sequences as previously described 28 . The subset of these represented in both the H1 and UBT1 genomes, including 14 ribosomal proteins and a phosphatidate cytidylyltransferase, were identified. Subsequently all Actinobacterial sequences, as well as B. subtilis as an outgroup, of the corresponding protein families were downloaded from the PFAM server 71 . These sequences were combined with the S. thermoautotrophicus proteins, and the proteins from each family were concatenated by species to form a single extended protein sequence representing a single isolate. The resulting table contained 431 strains, including H1 and UBT1. These were aligned in Clustal Omega with 30 iterations of input order, the resulting alignment trimmed with Gblocks, the trimmed alignment used to form a distance matrix and neighbor joining tree with 100 bootstrap replicates, and the consensus tree all generated with PHYLIP as described above. The final tree was edited in Mega to collapse monophyletic branches to the family level, excepting the immediate neighbors of H1 and UBT1.