Desulfovibrio diazotrophicus sp. nov., a sulfate-reducing bacterium from the human gut capable of nitrogen ﬁ xation

Sulfate-reducing bacteria (SRB) are widespread in human guts, yet their expansion has been linked to colonic diseases. We report the isolation, sequencing and physiological characterization of strain QI0027 T , a novel SRB species belonging to the class Desulfovibrionia . Metagenomic sequencing of stool samples from 45 Chinese individuals, and comparison with 1690 Desulfovibrionaceae metagenome-assembled genomes recovered from humans of diverse geographic locations, revealed the presence of QI0027 T in 22 further individuals. QI0027 T encoded nitrogen ﬁ xation genes and based on the acetylene reduction assay, actively ﬁ xed nitrogen. Trans-criptomics revealed that QI0027 T overexpressed 42 genes in nitrogen-limiting conditions compared to cultures supplemented with ammonia, including genes encoding nitrogenases, a urea uptake system and the urease complex. Reanalyses of 835 public stool metatranscriptomes showed that nitrogenase genes from Desulfovibrio bacteria were expressed in six samples suggesting that nitrogen ﬁ xation might be active in the gut environment. Although frequently thought of as a nutrient-rich environment, nitrogen ﬁ xation can occur in the human gut. Animals are often nitrogen limited and have evolved diverse strat-egies to capture biologically active nitrogen, ranging from amino acid transporters to stable associations with bene ﬁ cial microbes that


Summary
Sulfate-reducing bacteria (SRB) are widespread in human guts, yet their expansion has been linked to colonic diseases. We report the isolation, sequencing and physiological characterization of strain QI0027 T , a novel SRB species belonging to the class Desulfovibrionia. Metagenomic sequencing of stool samples from 45 Chinese individuals, and comparison with 1690 Desulfovibrionaceae metagenomeassembled genomes recovered from humans of diverse geographic locations, revealed the presence of QI0027 T in 22 further individuals. QI0027 T encoded nitrogen fixation genes and based on the acetylene reduction assay, actively fixed nitrogen. Transcriptomics revealed that QI0027 T overexpressed 42 genes in nitrogen-limiting conditions compared to cultures supplemented with ammonia, including genes encoding nitrogenases, a urea uptake system and the urease complex. Reanalyses of 835 public stool metatranscriptomes showed that nitrogenase genes from Desulfovibrio bacteria were expressed in six samples suggesting that nitrogen fixation might be active in the gut environment. Although frequently thought of as a nutrient-rich environment, nitrogen fixation can occur in the human gut. Animals are often nitrogen limited and have evolved diverse strategies to capture biologically active nitrogen, ranging from amino acid transporters to stable associations with beneficial microbes that provide fixed nitrogen. QI0027 T is the first Desulfovibrio human isolate for which nitrogen fixation has been demonstrated, suggesting that some sulfate-reducing bacteria could also play a role in the availability of nitrogen in the gut.

Introduction
Sulfate-reducing bacteria (SRB) are present in the mouth and the gut of~50% of the human population (Nava et al., 2012;Pasolli et al., 2019). SRB thrive in the gut, releasing hydrogen sulphide (H 2 S) as a product of sulfate reduction. H 2 S is a potent genotoxin and has been linked to chronic colonic disorders and inflammation of the large intestine (Coutinho et al., 2017). Likewise, the presence of some Desulfovibrio species has been implicated in chronic periodontitis, cell death and inflammatory bowel diseases such as ulcerative colitis and Crohn's disease (Loubinoux et al., 2003). The detrimental role of SRB is, however, not firmly established. H 2 S can also act as a signalling molecule or energy source for mitochondria (Gong et al., 2011;Carbonero et al., 2012). Moreover, by using hydrogen, SRB help in the efficient energy acquisition and complete oxidation of substrates produced by fermentative bacteria (Fischbach and Sonnenburg, 2011). SRB could, therefore, have a dual role within the gut microbiome. Relatively few strains from healthy individuals have been characterized. Cultivated SRB isolated from humans include Desulfovibrio piger, D. fairfieldensis, D. desulfuricans and D. legallii (Loubinoux et al., 2003;Vasoo et al., 2014), with D. piger described as the most abundant in samples obtained from Western countries (Carbonero and Gaskins, 2015). The physiology and genomic potential of strains isolated from non-Western countries remain largely unexplored.
Biological nitrogen fixation is the process by which gaseous dinitrogen (N 2 ) is reduced to biologically available ammonia (NH 3 ) by diazotrophic microbes (Seefeldt et al., 2009). Diazotrophy was reported in some Desulfovibrio species from free-living communities and the termite gut (Postgate and Kent, 1985;Kuhnigk et al., 1996;Fröhlich et al., 1999). All animals require access to a source of fixed nitrogen (Fuller and Reeds, 1998). Most of the readily available nitrogen in humans comes from food, but nitrogen fixation can occur in the microbiome of some human populations (Igai et al., 2016). In the human microbiome, the potential to fix nitrogen has been so far linked to Klebsiella and Clostridia (Bergersen and Hipsley, 1970;Igai et al., 2016). The human host can regulate the nitrogen available to the microbiome through the diet and intestinal secretions (Reese et al., 2018), so bacteria that can fix nitrogen could potentially have a competitive advantage in the gut environment.
We report here the isolation and characterization of a novel species of the genus Desulfovibrio, strain QI0027 T , isolated from a stool sample provided by a Chinese male donor. Surprisingly this strain encoded the genes for nitrogen fixation. The goals of this study were (i) to characterize strain QI0027 T , (ii) to identify if diazotrophy is commonly distributed among Desulfovibrionaceae associated with humans and (iii) to examine if the genes required for nitrogen fixation are expressed and functional. We used metagenomic sequencing of 45 human donors, as well as analysed comprehensive collections of genomes recovered from metagenomes to examine the distribution of QI0027 T . We used physiological tests to examine the nitrogenase activity. Finally, we used transcriptomics to determine the genes that are expressed by QI0027 T in response to the absence of a source of fixed nitrogen and reanalysed 835 stool metatranscriptomes to search for the expression of nitrogenases belonging to Desulfovibrio.

Genome sequencing and phylogenomics
As part of a study to isolate and characterize SRB from the intestinal tract of healthy Chinese donors, we isolated strain QI0027 T . Using short-read and long-read sequencing, we sequenced and assembled a circular chromosome of 2.9 Mb genome with a 62.1% GC content. The closest relative of QI0027 T was D. legallii (Genome statistics of QI0027 T and D. legallii are shown in Table 1). These bacteria grouped within the Desulfovibrio genus based on a well-supported phylogenomic tree reconstructed using 30 marker genes (Shivani et al., 2017) (Fig. 1). Even though the 16S rRNA gene of QI0027 T shared a 99% identity with the partial 16S rRNA sequence of D. legallii, the average nucleotide identity (ANI) of their whole genomes was 86.49%, the digital DNA-DNA hybridization (dDDH) was 31.4% (29%-33.9%  1. Distribution of the nitrogenase gene cluster among Desulfovibrionaceae bacteria. A maximum-likelihood phylogenomic tree was reconstructed using the amino acid sequences encoded by 30 single-copy marker genes. The tree was rooted with Clostridium termitidis CT1112. Arrows represent the presence of genes encoding the dinitrogenase reductase (nifH), the molybdenum-iron (Mo-Fe) protein (dinitrogenase nifD and nifK) and the iron-molybdenum cofactor biosynthesis (nifN-B and nifE). The arrow representing nifH is colour coded depending on the nifH subclade where the gene clusters. The country of origin for each MAG is shown with the ISO3 code. *Genes recovered using targeted assembly for QI0027 T MAGs. + Not reclassified by Waite et al. (2020).

Physiology
Similar to other Desulfovibrionaceae bacteria, QI0027 T is a Gram-negative bacterium with a curved rod shape (Fig. 2). Cells were 1.4-5 μm long. QI0027 T grew at a temperature between 15 and 46 C, with an optimum growth temperature of 37 C. QI0027 T could grow with 0-0.51 M NaCl, with optimum growth at 0.068-0.171 M NaCl. This salinity concentration range is comparable with human physiological saline (0.9% wt./vol., 0.15 M) (Oxley et al., 2010). Colonies on Postgate C plates were black due to the accumulation of iron sulphide precipitates ( Supplementary Fig. 1a). Based on disc diffusion assays with antibiotics, QI0027 T was resistant to streptomycin (Table 1).
Unlike the clearly motile D. legallii, QI0027 T was mostly non-motile but showed occasionally multiple lophotrichous flagella based on a spatial assay on soft agar and scanning and transmission electron microscopy (SEM and TEM) (see Fig. 2 and Supplementary  Fig. 1b). SEM and negative staining combined with TEM showed that D. legallii showed a single polar flagellum with both EM techniques (Fig. 2). The flagella of QI0027 T were instead longer and thinner than those compared to D. legallii, which might explain why we could not observe them with SEM, as the sample processing is harsher on the cells compared to the negative staining (Fig. 2). QI0027 T encoded the genes for flagella synthesis, including the hook-basal body complex (fliE), the basal-body rod (flgB, flgC), the cytoplasmic components fliI and the membrane-bound components (fliO, fliP, fliQ, flhA, flhB, fliR) of the flagellar protein export apparatus (reviewed by Anderson et al., 2010). QI0027 T also synthesized the negative regulator of flagellin synthesis flgM. We thus hypothesize that QI0027 T regulates the expression of flagella synthesis, which might be a key factor during the colonization and sensing of areas rich in nutrients in the gastrointestinal tract (Soutourina and Bertin, 2003).

Distribution of QI0027-related species in human microbiome metagenomes
Metagenomic reads with >95% similarity to QI0027 T were detected in two further individuals based on sequencing of 45 stool metagenomes from Chinese donors using read mapping (see Supplementary Dataset 3 and Supplementary results and discussion -Identification of QI0027-related species in Chinese metagenomes). We successfully recovered a high-quality draft genome of QI0027 T species (ANI similarity of 99.49%) from metagenomic reads using a targeted assembly approach of Desulfovibrionaceae with the combined 45 stool metagenomic reads from Chinese donors. This targeted assembly approach was used to evaluate if the genome of this species could be recovered in high-complexity stool metagenomes. We were not able to recover any D. desulfuricans MAG, even though this was the species that could be isolated more often with our isolation efforts in Chinese stool samples (13 out of 14 Desulfovibrionaceae isolates).
We further identified 20 genomes from the same species as QI0027 T (ANI > 98%) using a large unified collection of metagenome-assembled genomes (MAGs) obtained from extensive diversity of lifestyles, ages and countries (Almeida et al., 2020). Samples included in the analyses were obtained from 170 samples from Africa, 3590 from Asia, 4586 from Europe, 3298 from North America, 262 from Oceania and 183 from South America (Almeida et al., 2020). The genome collection included 1690 Desulfovibrionaceae genomes classified as D. piger, D. desulfuricans, D. fairfieldensis, Desulfovibrio sp., Bilophila sp., B. wadsworthia and Mailhella sp. Genomes from the same species as QI0027 T were assembled from stool metagenomes obtained from 14 donors residing in China, three in the UK, one in Sweden, one in Denmark and one in the USA (Metagenome studies from Qin et al., 2014;Bäckhed et al., 2015;Xie et al., 2016;He et al., 2017;Li et al., 2017;Wen et al., 2017). Since genomes of organisms that are more abundant are easier to recover from metagenomes (Albertsen et al., 2013), the distribution of QI0027 T among humans from diverse continents suggests that this species might be more prevalent in Asian individuals.

QI0027 T can fix nitrogen
Genome analysis of the strain QI0027 T revealed the presence of the minimal gene set for nitrogen fixation (Dos Santos et al., 2012). These genes included the key gene nifH, encoding the dinitrogenase reductase; nifD and nifK, encoding the MoFe dinitrogenase; as well as nifE, nifN and nifB, encoding the FeMo cofactor biosynthesis machinery. The nifN and nifB genes were fused into a single gene (nifN-B) and two genes encoding PIIlike nitrogen regulatory proteins ( glnB1 and glnB2) were encoded between nifH and nifD as observed in Clostridium and other diazotrophic Desulfovibrio (Supplementary Fig. 2) (Chen et al., 2001;Seefeldt et al., 2009). To physiologically validate the nitrogen fixation ability of strain QI0027 T , we checked the expression pattern of the genes encoding the nitrogen fixation machinery under conditions of nitrogen limitation compared to nitrogen excess. As expected, we observed that all genes in the nif operon were significantly induced under nitrogen limiting conditions ( Fig. 3A and B).
We demonstrated that strain QI0027 T was able to fix nitrogen based on an acetylene reduction assay performed in similar growth conditions used for the RNA-seq transcription analysis (Fig. 4). The specific nitrogenase activity rate for strain QI0027 T was estimated as 7.6 ± 5.2 nmol ethylene Á mg protein À1 Á h À1 . As expected, no acetylene reduction was observed when the growth media was supplemented with yeast extract or with 18.54 mM NH 4 Cl (referred to as nitrogen excess). To our knowledge, this is the first Desulfovibrio isolate from the human gut for which diazotrophy has been physiologically demonstrated.
Further evaluation of our whole transcriptome analysis revealed that the strain QI0027 T overexpressed 43 genes under nitrogen-limiting conditions, while an additional 19 genes were overexpressed under nitrogen excess conditions ( Fig. 3A and Supplementary Dataset 2). Under nitrogen excess conditions, QI0027 T had a higher expression of genes related to cell cycle and cell division, RNA processing, glycolysis, respiration, as well as cobalamin, ATP, and tryptophan synthesis ( Fig. 3A and B).
Moreover, QI0027 T overexpressed the antitoxin HigA, a system that has been linked to survival response during environmental and chemical stresses, including amino acid starvation, growth and programmed cell death (Schureck et al., 2014).
Besides the nitrogen fixation genes overexpressed under nitrogen limiting conditions discussed above, we observed overexpression of the urease enzyme complex ureABCEFGD and the reversible urea uptake system urtABCDE. ureABCEFGD is used to reversibly transform urea into ammonia and CO 2 . Since QI0027 T cultures with limiting nitrogen did not have urea in the media, ammonia resulting from N 2 fixation was likely converted to urea and the excess excreted. Urea is less toxic than ammonia, more soluble, and can be used as an intermediate molecule for nitrogen excretion (Levin et al., 2009). Alternatively, some QI0027 T cells might have scavenged urea to use it as a nitrogen source. Two transporters, from the TRAP and DctA families, which can be used for the import of organic acids (Kelly and Thomas, 2001), were overexpressed under nitrogen-limiting conditions, suggesting that efficient transport of dicarboxylates is required to support or regulate nitrogen fixation (Dixon and Kahn, 2004;Huergo and Dixon, 2015;Bueno Batista and Dixon, 2019).
Interestingly, under nitrogen-limiting conditions, QI0027 T overexpressed one methyl-accepting chemotaxis protein (MCP) (p-value <0.05 and FDR <0.05) and A. Differentially expressed genes by QI0027 T cultures with limiting or excess nitrogen. p-value cut-off for FDR = 0.05; minimum fold change = 3. B. Overview of differentially expressed pathways under nitrogen-limited or excess conditions by QI0027 T cultures. Under nitrogen limiting conditions, genes related to nitrogen fixation, urea production and urea transport are overexpressed. Under nitrogen excess conditions, tryptophan synthesis, peptidoglycan biosynthesis and molybdopterin cofactor are overexpressed. p-value cut-off for false discovery rate (FDR) = 0.05; minimum fold change = threefold. MCP = methyl-accepting chemotaxis protein; Gln = glutamine; Trp = Tryptophan; Gly-3P = Glyceraldehyde 3 phosphate. Reactions are not balanced. three MCPs were overexpressed under nitrogen-excess conditions (p-value <0.05 but FDR >0.05) (Supplementary Dataset 2). MCP proteins can be used as chemical sensors (Ortega et al., 2017). The MCP overexpressed under nitrogen-limiting conditions has a ligand-binding domain of Tar and Tsr chemoreceptors, which could potentially mediate responses towards optimal oxygen, redox and oxidizable substrates (Ortega et al., 2017) that may support biological nitrogen fixation.

Prevalence of diazotrophy among Desulfovibrionaceae
Several members of the Desulfovibrionaceae family possessed the minimal nif gene set (nifH, nifD, nifK, nifE and nifN-B) for nitrogen fixation but in some cases, these genes may have been lost in several lineages ( Fig. 1) (see Supplementary results and discussion -nifH was likely acquired through horizontal gene transfer). We searched for the genes related to nitrogen fixation among all available Desulfovibrionaceae genomes using tblastn and combined the gene presence/absence with our phylogenomic reconstruction of the Desulfovibrionaceae bacteria. Among the lineages that lack the nitrogen-fixing genes, we identified genera that have been linked to disease (e.g. Bilophila and Lawsonia). Within the Desulfovibrios, neither the closed genome of D. piger nor other almost complete D. piger or D. fairfieldensis genomes encoded the key gene for nitrogen fixation nifH. These Desulfovibrio species are the most abundant and widespread in Western individuals (Carbonero and -Gaskins, 2015). Desulfovibrio desulfuricans isolates obtained from free-living environments, as well as our own 13 D. desulfuricans isolates obtained from eight Chinese donors, encoded the genes to fix nitrogen. All these D. desulfuricans strains grouped closer to D. legallii and QI0027 T .
From all Desulfovibrionaceae MAGs recovered from human metagenomes (Almeida et al., 2020), only D. desulfuricans and Desulfovibrio MAGs that belong to the same species as QI0027 T based on an ANI >98% encoded the nifHDKEN-B genes (Fig. 1). Twelve D. desulfuricans MAGs, and 20 QI0027 T MAGs encoded nif genes. Four of these 20 QI0027 T MAGs originally assembled by Almeida et al. (2020) did not include the nif genes. However, using a targeted assembly approach with the metagenomic reads and the QI0027 T reference genome, we recovered the nitrogenase genes for the four QI0027 T MAGs. Based on phylogenetic analysis, the nifH gene grouped closer to nifH encoded by QI0027 T showing that using stringent mapping was enough to recover the nifH genes from QI0027 T . In agreement with our genome searches on the closed genome of D. piger, as well as on draft genomes of isolates from D. fairfieldensis, Bilophila and Lawsonia, none of the human-derived MAGs from these Desulfovibrionaceae species encoded the nitrogenase genes (Fig. 1). From the isolates and MAGs that have been sequenced thus far from Desulfovibrios that encode the genomic repertoire for nitrogen fixation, only D. legallii, which was isolated from a shoulder joint infection, D. desulfuricans and QI0027 T species have been isolated from mammals ( Fig. 1).

Expression of nitrogen fixation genes in human stool metatranscriptomes
To identify if Desulfovibrio bacteria can express the nitrogen fixation genes in the human gut environment, we searched for nif genes transcripts in 835 stool metatranscriptomes available through the Human Microbiome Project database (Consortium, 2019; Lloyd-Price et al., 2019). Metatranscriptome reads more than 95% similar to Desulfovibrio nif genes were detected in six samples. From these six faecal samples, reads aligning to nifH from QI0027 T were present in two metatranscriptomes. In addition, reads aligning to nifH from Desulfovibrio desulfuricans subsp. desulfuricans DSM 642 were found in one sample (Supplementary Dataset 4). The low abundance of Desulfovibrio nif genes transcripts detected in these metatranscriptome datasets may reflect the abundance of diazotrophic Desulfovibrio in the gut microbiota. As the expression of nitrogen fixation genes is often controlled by oxygen, oxygen exposure during stool sample preparation may have also contributed to the low abundance of nif transcripts detected. The detection of Desulfovibrio nif transcripts in the human gut metatranscriptome might indicate that nitrogen fixation can occur by Desulfovibrio in the gut.

Discussion
We have presented evidence that QI0027 T is a novel species with nitrogen-fixing ability. This novel species was misclassified by commonly used taxonomic classifiers that rely on clade-specific marker genes from metagenomes and could only be identified using wholegenome sequencing and assembly (of MAGs or pure cultures). Our study combines cultivation, genome sequencing, physiological characterization, differential expression analysis and metagenome sequencing to overcome the bias inherent to each technique.
Based on our acetylene reduction assay and differential expression analyses, strain QI0027 T fixed nitrogen only when we did not provide a source of fixed nitrogen in the media, such as yeast extract or NH 4 Cl, which are known to inhibit the metabolically expensive process of nitrogen fixation (Hardy et al., 1968). Our results show therefore that the genes for nitrogen fixation are functional in strain QI0027 T but that their activity can be inhibited depending on fixed nitrogen availability. The rate of acetylene reduction from QI0027 T was in the range of 1.16-13.18 nmol ethylene Á mg protein À1 Á h À1 , which would be equivalent to~0.001-0.261 nmol ethylene Á g stool À1 Á h À1 (for assumptions see Materials and methods). Igai et al. (2016) showed a rate of acetylene reduction in unfrozen stool samples of 0.055-1.608 nmol ethylene Á g stool À1 Á h À1 , while Bergersen and Hipsley (1970) detected 0.75 nmol ethylene Á g stool À1 Á h À1 acetylene reducing activity in one European stool sample. Our extrapolation of the rate of acetylene reduction per gram of stool shows that the nitrogen fixation ability of QI0027 T could contribute at rates similar to those observed in human stool samples attributed previously to only other taxonomic groups (Igai et al., 2016).
The rate of nitrogen fixation by QI0027 T was lower compared to the environmental isolates D. desulfuricans, Nitratidesulfovibrio vulgaris, Megalodesulfovibrio gigas, Maridesulfovibrio salexigens, Desulfocurvibacter africanus and Dissulfuribacter thermophilus (7.6 ± 5.2 vs. 42-918 nmol ethylene Á mg protein À1 Á h À1 ) (Postgate and Kent, 1985). Using this assay, three out of 15 environmental strains belonging to the Desulfovibrionaceae family were shown not to fix nitrogen. The authors hypothesized that the right conditions for nitrogen fixation might have yet to be found. However, this pattern can be explained now as a true absence of nitrogen fixation ability since the key gene for nitrogen fixation, nifH, has been gained and lost several times in the Desulfovibrionaceae (Fig. 1, see Supplementary results and discussion -nifH was likely acquired through horizontal gene transfer).
Only a few members of the Desulfovibrio encoded nifH. The ability to fix nitrogen is not present in the Desulfovibrionaceae that are the most abundant in the gut but rather prevails in Desulfovibrio species that tend to be in lower abundance. Desulfovibrio desulfuricans is often recovered from human stool samples (Coutinho et al., 2017). Some of these D. desulfuricans strains might also fix nitrogen, such as the case of our 13 D. desulfuricans Chinese gut isolates that encoded the nitrogenase operon. However, not all strains of D. desulfuricans encode the genes to fix nitrogen (Fig. 1). Indeed, D. desulfuricans formed two major clusters with an 80%-85% ANI between the two clusters, indicating that these species will need to be reclassified in future studies. The contribution to nitrogen fixation in the human gut by Desulfovibrionaceae is therefore likely not limited to QI0027 T .

Nitrogen fixation in a nutrient-rich environment
The human gut is a nutrient-rich environment in which most bioavailable nitrogen comes from amino acids present in food. Expression of nitrogen fixation-related genes affiliated to Desulfovibrio bacteria was found on six stool samples. However, read counts were low and it is difficult to relate gene expression in stool, which is often exposed to oxygen, to in situ activity within the gut for an oxygen-sensitive process like nitrogen fixation. Active nitrogen fixation in the human gut would suggest that this process is not only a potential survival strategy when outside a mammal host. If nitrogen fixation is a high-energy demanding process (Rascio and La Rocca, 2013), and the gut is nutrient-rich, which conditions could favour the persistence of Desulfovibrio species that can fix nitrogen?
It is commonly assumed that N 2 fixation can occur only when bioavailable nitrogen is limiting (<1 μM) (Knapp, 2012). However, nitrogen fixation can occur in environments with relatively high input of fixed nitrogen such as oceanic waters with high nitrate or ammonia concentrations (30 μM NO 3 À ; 200 μM NH 4 + ) as well as the human gut (12-30 mM ammonia in stool, which can be increased with higher protein consumption) (Knapp, 2012;Scott et al., 2013;Igai et al., 2016;Shiozaki et al., 2018). Humans belonging to a population from Papua New Guinea that had low protein intake (e.g. fixed nitrogen), as well as Japanese individuals that had a higher protein intake, were shown to have a gut microbiome that was actively fixing nitrogen (Igai et al., 2016).
Nitrogen fixation was independent of the habitual nitrogen intake and was confirmed based on the acetylen reduction assay or 15 N incorporation (Bergersen and Hipsley, 1970;Igai et al., 2016). In these human populations, nitrogen fixation by the microbiome corresponded to only 0.01% of the standard nitrogen requirement for humans. Nitrogen fixation was attributed to Klebsiella and Clostridia bacteria based on nifH sequences recovered from cloning and metagenomic analysis, including Clostridia encoded nifH-Cluster 3 sequences recovered from Chinese faeces (Igai et al., 2016). However, the microbial contribution to the nitrogen intake from humans could be population specific depending on the dietary habits, especially when low protein diets are consumed. Based on our acetylen reduction assay, 18.54 mM ammonium inhibited nitrogen fixation, which is within the range of ammonium detected in human stool (12-30 mM) (Scott et al., 2013). Ammonium is, however, rapidly absorbed and metabolized to urea by the liver. Thus, although the gut may have on average an excess of nitrogen because of the host diet, there may be microniches where bioavailable nitrogen is locally limiting or a specific nitrogen-limited diet may result in sub-inhibitory levels of ammonia. More than 90% of nitrogen is absorbed in the small intestine and studies using germfree animals have shown that most gut microbes increase the protein requirement of the host (Fuller and Reeds, 1998;Reese et al., 2018). In the large intestine, the bacterial load increases which could create microniches where fixed-nitrogen becomes limiting (Dai et al., 2011). The host might secrete nitrogen in the large intestine to control the microbial load and select for preferred microbial taxa (Reese et al., 2018). Therefore, having nitrogen fixation ability would still be a selective advantage. SRB must compete for resources and space to thrive in the gut. Hydrogen, the most used energy source by SRB, is also used by acetogens and methanogens. Niche partitioning could be a mechanism to cope with competitors, as has been observed for even members of the same species (Sayavedra et al., 2015;Ailloud et al., 2019;Ansorge et al., 2019). Fixed nitrogen could then be used for the synthesis of amino acids, which in turn could be used as a common exchange currency with other members of the gut microbiome or the host (Lesser et al., 2007;Mohamed et al., 2008).
An alternative explanation for the presence of diazotrophs in environments with nitrogen excess is that nitrogen fixation is used to maintain an ideal intracellular redox state, analogous to bacteria which can grow heterotrophically but still retain energetically expensive CO 2 fixation pathways like the Calvin-Benson cycle (Tichi and Tabita, 2000;McKinlay and Harwood, 2010). This has been interpreted as a means to regenerate electron carriers like NAD + under anaerobic conditions, where there can be an excess of reducing equivalents (Bombar et al., 2016). Nitrogen fixation is also energetically expensive and consumes reducing equivalents. So far, the use of N fixation as an electron sink has only been proposed for phototrophs, but perhaps nitrogen fixation could fulfil a similar function in anaerobic niches of the gut, with the additional benefit of producing bioavailable nitrogen. Further work will be required to disentangle which factors trigger nitrogen fixation by Desulfovibrio in the human gut and to understand the contribution of bioavailable nitrogen through nitrogen fixation, compared to other nitrogen assimilation pathways such as the import of free amino acids, ammonium, urea, nitrate and DNA.
The type strain, QI0027 T (=DSMZ 109475 T = NCIMB 15228 T = NCTC 14354 T ), was isolated from a faecal sample donated by a 25-year-old male healthy human in Wuxi, China. The G + C content of the type strain is 62.1%.

Isolation
Strain QI0027 T was isolated from a human stool sample in Wuxi, China in June 2016. The 25-year-old male donor was healthy. The donor was provided with a stool collection kit, which included an anaerobic bag, an anaerobic pack (Mitsubishi Gas Chemical, Japan) and ice. The sample was placed inside an anaerobic cabinet within 30 min of collection. A 1 g stool sample was homogenized and used for dilution (ranging from 1:10 to 1:1 Â 10 7 ) using anaerobic 1Â PBS. Dilutions were plated onto anaerobic Postgate C agar plates, which contained per litre of distilled water: 6 g sodium lactate; 4.5 g Na 2 SO 4 , 1 g NH 4 Cl, 1 g yeast extract, 0.5 g KH 2 PO 4 , 0.3 g sodium citrate, 0.06 g MgSO 4 .7H 2 O, 0.004 g FeSO 4 Á7H 2 O, 4 ml resazurin (0.02% wt./vol.), 0.04 g CaCl 2 Á2H 2 O, 0.5 g L-Cysteine hydrochloride and 15 g agar. pH was adjusted to 7.5 ± 0.1 using 5 M HCl and the media was sterilized by autoclaving. Plates were incubated at 37 C for 2 weeks. Single black colonies were selected and subcultured two more times until a pure culture of QI0027 T was obtained. The purity of the isolate was confirmed by colony morphology, light microscopy, and by checking that all cells in the culture hybridized to the deltaproteobacteria-specific fluorescence in situ hybridization probe Delta495a (Lucker et al., 2007) as described by Amann and Fuchs (2008). Additionally, 16S rRNA sequencing from colonies that showed slightly different pigmentation or size was done to confirm the purity of QI0027 T . The 16S rRNA gene was amplified with the AMP_F (5 0 -GAG AGT TTG ATY CTG GCT CAG) and AMP_R (5 0 -AAG GAG GTG ATC CAR CCG CA) set of primers according to the method from Baker et al. (2003). Purified amplicons were sequenced using the Mix2Seq Kit Overnight service (Eurofins, Germany). To confirm the identity of QI0027 T , the 16S rRNA sequence was compared against the Silva SSU r138 database (Quast et al., 2013). For long-term storage, strain QI0027 T was stored at À80 C in Protect Select tubes for anaerobes with cryobeads (TS/73-AN80, Technical Service Consultants, UK).
DNA extraction, whole-genome sequencing and assembly DNA was extracted using the GenElute Bacterial Genomic DNA kit (NA2100, Sigma-Aldrich, UK) following the manufacturer's instructions. For whole-genome sequencing, DNA was sent to Earlham Institute, Norwich, UK, where a DNA library was constructed using a modified version of Illumina's Nextera protocol known as Low Input Transposase Enabled libraries (Beier and Gross, 2006). 947 409 paired-end 150 bp reads were generated using the Illumina HiSeq 4000 platform. Reads were quality trimmed with bbduk (v.38.06) (sourceforge. net/projects/bbmap) using a minimum quality of 3. For long-read sequencing, a DNA library was prepared using the native barcoding genomic DNA expansion 1-12 kit (with EXP-NBD104, EXP-NBD114 and SQK-LSK109, Nanopore, UK) according to the manufacturer's instructions and sequenced using a MiniION platform (Quadram Institute Bioscience, Norwich, UK). To obtain a highquality closed genome, we used the hybrid assembler Unicycler (v.0.4.8) and polished the genome with Pilon (v.1.23 (Walker et al., 2014;Wick et al., 2017). The hybrid assembly resulted in a single chromosome. We confirmed that the 5 0 and 3 0 extremes of the scaffold were connected, and hence the genome was circular, by visualization of the assembly graph with Bandage (Wick et al., 2015). Genome statistics and quality checks for QI0027 T and D. legallii KHC7 were done with Patric (see Table 1) (Parrello et al., 2019). Genome sequence-based species delimitation metrics were determined using fastANI (Jain et al., 2018) and GGDC (v.2.1) (Meier-Kolthoff et al., 2013;Meier-Kolthoff et al., 2014). We report the dDDH value estimated with the recommended formula 'identities/HSP length'. The genome was annotated with PATRIC RASTtk (Brettin et al., 2015).

Metagenome sequencing, microbial profiling of stool samples and detecting the presence of QI0027 T in Chinese metagenomes
Forty five humans from China were provided with a stool collection kit as described above. Samples were stored at À80 C upon 48 h of collection. DNA was extracted from the stool samples using the FastDNA SPIN kit for faeces (MP Biomedicals, Shanghai, China) according to the manufacturer's instructions using approximately 50 mg of stool material per individual. DNA libraries were prepared using the NEB Next Ultra DNA Library Prep kit for Illumina (New England BioLabs). Approximately 10 Gb were sequenced with a HiSeq X Ten instrument as paired-end 150 bp reads. Library preparation and sequencing was done by Novogene (China). Metagenomic reads were quality trimmed with a minimum quality of 2 and human host reads were removed using as reference the human genome GCA_000001405 with bbduk. These clean reads were used to search for QI0027 T by mapping the reads against QI0027 T genome assembly with bbmap (v.38.43) using >95% mapping identity. We considered that QI0027 T species was present when >70% of the reference genome was covered by the reads.

Recovery of QI0027 T from Chinese metagenomes and MAGs
Clean reads of the 45 stool metagenomes were combined and mapped against all publicly available and our own Desulfovibrionaceae genomes. The resulting assembly from the combined reads was separated into genomes using an unsupervised binning method based on nucleotide composition, differential coverage and linkage data from paired-end reads (Alneberg et al., 2014). The resulting reads were then assembled with metaSPAdes (v.3.13.0) (Nurk et al., 2016) and binned with CONCOCT (v.1.1.0) (Alneberg et al., 2014). Genome completeness was estimated using CheckM (v.1.0.13) with the marker lineage Deltaproteobacteria (UID3218) (Parks et al., 2015) and identity was determined with GTDB-Tk (v.1.0.2) using the database release 89 (Parks et al., 2019). Binning allowed us to recover a total of four Desulfovibrionaceae genomes with a completeness ≥97.34% and a contamination ≤8.67%, including QI0027, Bilophila wadsworthia, D. fairfieldensis and D. piger. The relative abundance of reads mapping to the same species as QI0027 T was determined using bbmap (v.38.43) as the number of reads mapping to the genome/total number of reads.

Detection of genes related to nitrogen fixation and phylogenetic analysis
We searched for the nitrogen fixation related genes among the Desulfovibrionaceae genomes included in our phylogenomic tree, as well as the Desulfovibrionaceae MAGs from the collection by Pasolli et al. (2019). We used as query the amino acid sequences encoded by nifH, nifD, nifK, nifE and nifB from QI0027 T against the isolate genomes and MAGs with tblastn (v.2.2.31+) to be able to detect partial genes at the end of scaffolds (>40% similarity, >40% query coverage, minimum alignment length >50). To improve the genome assembly of the four MAGs that did not encode the nitrogenase cluster, we used read mapping against the QI0027 T reference genome using bbmap (v.38.43) with a similarity ≥95% against the reads from projects YSZC12003_3554, SRR3736997, H1M313811and YSZC12003_36012 (Short Read Archive). Results were integrated with the phylogenomic tree with iTol (Letunic and Bork, 2016).
For phylogenetic analysis, we annotated all genomes with Prokka (v.1.14.0) and retrieved nifH amino acid sequences. Representative nifH amino acid sequences from Clusters 1-4 were retrieved from Gaby and Buckley (2014). Sequences were aligned using Mafft v. 1.3.7 and realigned with ClustalW v2.1. The alignment was masked to remove alignment positions with >75% gaps. A maximum-likelihood tree reconstruction was obtained using FastTree (v. 2.1.5) with SH-like branching support values (Price et al., 2009).
Physiology of QI0027 T pH, salinity, temperature and motility. Physiological testing was done by adjusting Postgate C media without agarose in Hungate tubes (SciQuip, UK) and diluting cells 100-fold. To identify the pH range at which QI0027 T could grow, pH was adjusted by injecting deoxygenated 1.34 M HCl for acidic pHs, or 0.16 M NaOH with 10% NaHCO 3 (wt./vol.) for alkaline pHs before autoclaving. Final pH was measured before inoculation (range from 3.98 to 8.83) and after the growth of the cells. QI0027 T grew in media with a pH range between 4.5 and 6.5, and the pH of the media became more basic after cell growth ranging from 6.8 to 7.2. To determine the tolerance of QI0027 T to different salinity concentrations, growth at 0-0.513 M NaCl was investigated in liquid Postgate C medium as described previously (Lopez-Cortes et al., 2006). pH and salinity tolerance were investigated at 37 C for 5 days.
To test for the effect of temperature on growth, duplicate cultures of strain QI0012 were incubated in 10 ml Hungate tubes at a temperature range of 10-59 C for up to 7 days. Growth was monitored with a turbidimeter CO 8000 (Biochrom, UK). Motility of QI0027 T was tested using semisolid Postgate C agar (5 g L À1 of agar), using as reference for diffusion the motile D. legallii. The semisolid media was stabbed with disposable inoculating loops and the diffusion was observed after 3-5 days.
Antibiotic resistance. Antibiotic resistance was examined by disc diffusion assays on Postgate C agar according to the standard procedure outlined in the National Committee for Clinical Laboratory Standards guidelines (Wayne, 1999). Inside an anaerobic cabinet, a sterile swab was dipped in a bacterial cell suspension grown overnight and spread on reduced Postgate C agar media. Antibiotic discs (Oxoid, Thermo Fisher Scientific, UK) were placed on the surface using sterile forceps. Plates were incubated for 48-72 h at 37 C. The following antibiotics were tested: ampicillin (25 μg), chloramphenicol (10 μg), nalidixic acid (30 μg), erythromycin (30 μg), tetracycline (30 μg) and streptomycin (25 μg).
Cell morphology, FISH and transmission electron microscopy. Morphological properties, such as cell shape, cell size and motility were observed by phasecontrast light microscopy (Olympus BX60 microscope). For fluorescence microscopy, a Zeiss Axio Imager M2 microscope was used with a 100Â objective lens. Gram staining was done with the Gram staining kit (Sigma-Aldrich) according to the manufacturer's instructions. To investigate the presence of flagella, negative staining combined with transmission electron microscopy (TEM) was used. Briefly, a small drop of bacterial cell suspension was applied to a formvar/carbon-coated copper TEM grid (Agar Scientific, Stansted, UK) and left for 1 min. Excess liquid was removed with filter paper. A drop of 2% aqueous phosphotungstic acid was applied to the grid surface and left for 1 min. Excess stain was removed with filter paper and the grids were left to dry thoroughly before viewing in the transmission electron microscope. Grids were examined and imaged in an FEI Talos F200C TEM at 200 kV with a Gatan One-View digital camera. For scanning electron microscopy (SEM), cells were prepared by critical-point drying (CPD) onto filter membranes (Kal ab et al., 2008). Briefly, 100 μl of bacterial cell suspension was added dropwise onto an isopore membrane polycarbonate filter (HTTP01300, Millipore, UK). The filter was trimmed beforehand with a razor blade to identify the surface with inoculum. Cells were left to adhere to the surface for 5 min after which the filters were placed in 2.5% glutaraldehyde in 0.1 M PIPES buffer (pH 7.2) and fixed for 1 h. After washing with 0.1 M PIPES buffer, each sample was carefully inserted into a CPD capsule and dehydrated in a series of ethanol solutions (30, 50, 70, 80, 90, 3Â 100% including a final dry ethanol change). Samples were dried in a Leica EM CPD300 Critical Point Dryer using liquid carbon dioxide as the transition fluid. The filters were carefully mounted onto SEM stubs using sticky tabs, ensuring that the inoculated surface was facing upwards. The samples were coated with platinum in an Agar high-resolution sputtercoater apparatus. SEM was carried out using a Zeiss Supra 55 VP FEG SEM, operating at 3 kV.
Fatty acid analyses. For the determination of cellular fatty acids from the strains D. legallii (DSM 19129) and QI0027 T , cells were grown in Postgate C media at 37 C for 18 h. Cells in the exponential phase were pelleted by centrifugation at 4000g for 10 min. Fatty acid identification was carried out by the Identification Service of the DSMZ (Braunschweig, Germany). Fatty acids were converted to FAMEs by saponification, methylation and extraction according to Miller (1982) and Kuykendall et al. (1988). Samples were prepared according to the standard method given in MIDI Technical Note 101. Profiles of fatty acids were determined with an Agilent 6890N gas chromatograph and processed with the MIDI Inc Sherlock MIS software version 6.1.
Utilization of carbon substrates. To determine which carbon substrates could be utilized by QI0027 T and D. legallii, we used AN microplates from the OmniLog™ system (1007, Technopath, Ireland). We followed the manufacturer's instructions (AN Microplate™ procedure for testing carbon metabolism of anaerobic bacteria, draft of 3.11.2017 and (Takii et al., 2008)). Briefly, cells of the two species were grown on Biolog Universal Anaerobe agar (Technopath) inside an anaerobic cabinet until visible growth could be observed (2-3 days). Cells were collected with cotton swabs and used for preparing a cell suspension in AN inoculating fluid (Technopath) with turbidity between 0.4 and 0.48. These suspensions were used for the inoculation of the AN microplate in triplicates. To maintain an anaerobic atmosphere during the experiment, AN microplates were placed inside plastic bags which had an atmosphere of 90% N 2 , 5% CO 2 and 5% H 2 . The plates were incubated at 37 C for 36 h and the colour change caused by the response to the oxidation of the different substrates in the plate were measured with an Omnilog plate reader (Biolog). The resulting data were analysed with the opm package (v.1.3.63) (Vaas et al., 2013) implemented in R. Reactions were defined as positive or negative using k-means partitioning as implemented in the function do_disc.

Determination of nitrogenase activity
To determine the nitrogenase activity of QI0027 T to fix nitrogen, we used the acetylene reduction assay (Dilworth, 1966;Schöllhorn and Burris, 1967). The cells were first grown in media free of added fixed nitrogen (ammonia, NH 3 ), modified from Postgate B media (Postgate, 1970), which contained per liter of distilled water: 3.5 g sodium lactate, 2 g MgSO 4 Á7H 2 O, 0.25 g CaSO 4 , 0.5 g K 2 HPO 4 , 0.004 g FeSO 4 Á7H 2 O, 9 g NaCl, 4 ml resazurin (0.02% wt./vol.), 0.1 g ascorbic acid and 0.1 g thioglycolic acid. pH was adjusted to 7.5 ± 0.1 using 5 M HCl. The media was dispensed into Hungate tubes (6.2 ml gas phase) and sterilized by autoclaving. Media was supplemented with 1 ml L À1 of trace element solution SL-10 (DSMZ). Growth could be observed after 5-7 days. These cells were used to inoculate 12 tubes with a 1:100 dilution in 10 ml of media. As nitrogenase activity is inhibited when fixed nitrogen is available, we included as negative control duplicate tubes supplemented with either 1 g L À1 of yeast extract or 1 g L À1 NH 4 Cl (Hardy et al., 1968;Postgate, 1970;Riederer-Henderson and Wilson, 1970). Duplicate blank tubes without inoculum were also included as negative controls. All tubes were injected 10% vol./vol. of the gas phase with acetylene and incubated at 37 C. After 7 days of incubation, ethylene formation was quantified by using a Perkin Elmer Clarus 480 gas chromatograph equipped with a HayeSep® N (80-100 MESH) column. The injector and oven temperatures were kept at 100 C, while the FID detector was set at 150 C. The carrier gas flow was set at 8-10 ml min À1 . The ethylene standard was prepared from ethephon decomposition based on Zhang and Wen (2010). Nitrogenase activity is reported as nmol of ethylene produced per mg protein per hour. To determine whole-cell protein concentration, 5 ml of the culture was centrifuged at 3200g for 10 min at 4 C. The pellet was resuspended in 500 μl NaOH 0.1 M and incubated for 3 h at 30 C. Cells were then sonicated using an MSE Soniprep 150 instrument for seven cycles of 15 s sonication followed by at least 30 s rest on ice. Protein content from the lysates was finally measured with the Bradford assay (BioRad). To compare our results with the ARAs previously described in human stool by Igai et al. (2016) and Bergersen and Hipsley (1970), we assumed the following: (i) the bacterial load is 10 11 cells per gram (Sender et al., 2016), (ii) bacterial cells have 60-330 fg of protein per cell (Zubkov et al., 1999) and (iii) QI0027 T can be at an abundance from 0.02% to 0.06% from the total bacterial population (Supplementary Dataset 3).

Transcriptional response to the presence and absence of ammonia
To identify the genes that are expressed by QI0027 T when fixing nitrogen, we used whole transcriptome sequencing of cultures grown in modified Postgate B minimal media without any source of fixed nitrogen or with 1 g L À1 NH 4 CL. QI0027 T cells grown for 48 h in Postgate C media were centrifuged at 4000g for 3 min. Cells were resuspended in modified Postgate B media without NH 4 CL. These cells were used to inoculate six Hungate tubes with 10 ml Postgate B media, six tubes supplemented with 1 g L À1 NH 4 Cl, as well as duplicate control tubes supplemented with 1 g L À1 of yeast extract using a 1:100 dilution. Duplicate negative controls of Postgate B media supplemented with yeast extract without QI0027 T inoculum were also included. As expected, control tubes with yeast extract showed growth in 48 h. After 7 days, we divided the media from six tubes per condition into two to increase the input biomass for RNA extraction. Cells were centrifuged at 4000g for 10 min and immediately frozen in dry ice for 10 min. RNA was extracted using the RNeasy Plus Minikit (Qiagen, UK) according to the manufacturer's instructions with the following modifications. 350 μl RNeasy lysis buffer and 10 μl ß-mercaptoethanol were added to the cell pellets. Cells were disrupted in a FastPrep (MP Bio, UK) with 3 cycles lasting 20 s at 4 m s À1 in Lysing Matrix E tubes (MP Bio). Samples were centrifuged at 13 000g for 3 min, and the supernatant was used for further RNA extraction. The resulting RNA was used to synthesize cDNA using the SMARTer Universal Low Input RNA kit for sequencing (Takara, Korea), followed by library construction using the TruSeq Total RNA (NEB Microbe, New England BioLabs, USA) kit without rRNA depletion. 3 Gb per sample were generated as 100 bp paired-end reads using a NovaSeq instrument at Macrogen (South Korea).
For differential expression analyses, adapters and rRNA sequences were removed from transcriptome reads using bbduk (v.38.06). Clean reads were mapped to the QI0027 T reference genome with a minimum identity of 95% and all ambiguous mapping reported. The number of transcripts per gene was estimated with featureCounts (v.2.0) (Liao et al., 2014). Differentially expressed genes were detected with edgeR with TMM normalization Robinson and Oshlack, 2010). Pathway differentially expressed were examined with Pathway Tools (v.23.5). Operons were predicted with Rockhopper. Secretion and conjugation systems were predicted with Maccsyfinder (v.1.0.5) using the CONJScan and TXSS models (Abby et al., 2014;Abby et al., 2016). Functional domains of MCP proteins were predicted with CD-Search tool (Marchler-Bauer et al., 2014).

Author Contributions
L.S. and A.N. designed the study. L.S. conducted experiments and analysed data. T.L. isolated SRB, collected stool samples and extracted nucleic acids for metagenomes. M.B.B. and L.S. did acetylene reduction assays. B.K.B.S. suggested the impact of nitrogen fixation as electron sink and critically commented on the manuscript. C.B. did electron microscopy. Q.Z. and W.C. coordinated sample collection in China. L.S. wrote the manuscript with contributions from all co-authors.

Data Availability Statement
All sequencing data were submitted to the European Nucleotide Archive (http://www.ebi.ac.uk/ena/data/view/). Sequencing reads and the genome assembly of QI0027 T are available with the accession number PRJEB34368. Metagenomic reads obtained from Chinese stool samples without human reads are available with the accession number PRJEB38832. Transcriptomic reads are available under the accession number PRJEB38833.

Ethics Statement
This study was approved by the Ethics Committee in Jiangnan University, China (SYXK 2012-0002). All the faecal samples from healthy persons were for public health purposes and these were the only human materials used in the present study. Written informed consent for the use of their faecal samples was obtained from the participants or their legal guardians. All of them conducted health questionnaires before sampling and no human experiments were involved. The collection of faecal samples had no risk of predictable harm or discomfort to the participants. estimated from the TMM normalized values. Genes overexpressed with nitrogen deplete conditions are shown in blue. Genes overexpressed with nitrogen replete conditions are shown in orange. **p-value significant corrected using false discovery rate (FDR) correction. *p-value <0.05 and FDR <0.1. DE = Differentially expressed. DE = Differentially expressed. Supplementary Dataset 3. Relative abundance of QI0027 Trelated species in 45 Chinese stool metagenomes. The abundance was estimated based on read mapping against QI0027 T scaffolds ≥1000 bp. The relative abundance of the two Desulfovibrionaceae species predicted by MetaPhlan2 is also shown.