Exploiting genetic diversity and gene synthesis to identify superior nitrogenase NifH protein variants to engineer N2-fixation in plants

Engineering nitrogen fixation in eukaryotes requires high expression of functional nitrogenase structural proteins, a goal that has not yet been achieved. Here we build a knowledge-based library containing 32 nitrogenase nifH sequences from prokaryotes of diverse ecological niches and metabolic features and combine with rapid screening in tobacco to identify superior NifH variants for plant mitochondria expression. Three NifH variants outperform in tobacco mitochondria and are further tested in yeast. Hydrogenobacter thermophilus (Aquificae) NifH is isolated in large quantities from yeast mitochondria and fulfills NifH protein requirements for efficient N2 fixation, including electron transfer for substrate reduction, P-cluster maturation, and FeMo-co biosynthesis. H. thermophilus NifH expressed in tobacco leaves shows lower nitrogenase activity than that from yeast. However, transfer of [Fe4S4] clusters from NifU to NifH in vitro increases 10-fold the activity of the tobacco-isolated NifH, revealing that plant mitochondria [Fe-S] cluster availability constitutes a bottleneck to engineer plant nitrogenases. Jiang et al. show that nitrogenase NifH component of Hydrogenobacter thermophilus fulfils the quantity and functional requirements to engineer efficient N2 fixation in mitochondria. This study contributes towards future efforts engineering diazotrophs in eukaryotic cells.

N itrogen (N) fertilizers used to increase crop productivity in intensive agriculture practices pollute groundwater and release greenhouse gasses 1 . On the other hand, subsistence agriculture practices including poor N fertilization produce low and inconsistent yields causing malnutrition and poverty 2,3 . There is large interest in engineering cereal crop varieties capable of acquiring their own N 4 . One approach to this outcome relies on functional expression of a nitrogenase enzyme by the cereal plant 5 . Nitrogenases are prokaryotic, O 2 -sensitive, two-component metalloproteins that convert inert N 2 into biologically useful NH 3 [6][7][8] . The most efficient and widespread variant, the molybdenum nitrogenase, is composed of an Fe protein (nifH-encoded) and a MoFe protein (encoded by nifD and nifK). The Fe protein (NifH) donates electrons to the MoFe protein (NifDK) that in turn reduces N 2 . Nascent NifH and NifDK polypeptides need to acquire proper quaternary structure and to receive metal clusters, one [Fe 4 S 4 ] cluster per NifH homodimer and two pairs of P-cluster and FeMo-co per NifDK heterotetramer, for functionality. We have recently reviewed the mechanisms and genetic requirements to assemble these cofactors and to mature NifH and NifDK into active Mo nitrogenase 9 . The large number of nitrogen fixation (nif) genes involved, and the sensitivity of most of the protein products towards O 2 , makes nitrogenase engineering a daunting task with issues that need to be solved stepwise.
To date, functional NifH, NifU, and NifB have been purified from mitochondria of aerobically cultured Saccharomyces cerevisiae cells 10,11 , while active NifU and NifH were isolated from chloroplasts of Nicotiana benthamiana at the end of the dark period 12 . Also, the reported low stability of the NifD protein 13 has now been improved in two recent studies that identified key residues in the NifD sequence as susceptible to cleavage upon mitochondria import 14,15 . Notwithstanding these achievements, detailed analysis of yeast mitochondria-targeted Azotobacter vinelandii NifH has been hampered by low protein solubility resulting in suboptimal yields 10 . Accumulation of mostly insoluble NifH was also reported when Klebsiella oxytoca NifH was targeted to the tobacco mitochondria 16 . The difficulty of expressing high levels of soluble and functional NifH in yeast and tobacco poses a major problem for eukaryotic nitrogenase engineering as it is the most abundant Nif protein during N 2 fixation 17 . The problem is exacerbated because, in addition to serving NifDK with electrons for substrate reduction, NifH is required to mature P-clusters onto NifDK and for the final steps of FeMo-co biosynthesis in complex with NifEN 9 . For these reasons it is essential to identify a NifH variant that is highly soluble and stable when expressed at very high levels in a plant cell, and that can perform all three NifHdependent activities. One approach to achieve this outcome would be protein engineering of well-studied NifH from model diazotrophs (e.g., A. vinelandii or K. oxytoca) aimed to introduce sequences that improve stability in the mitochondria 18 . Protein engineering has been extensively employed to obtain glyphosate resistance 19 , another important trait for crops. Alternatively, mining of phylogenetically diverse nifH sources can be undertaken in order to find natural NifH proteins with superior properties, a strategy that was successful for NifB 11 and for increasing carotenoid levels in "Golden Rice" 20 .
Here, 32 distinct nifH genes were screened for expression level and solubility in mitochondria of N. benthamiana. The nifM, nifU, and nifS genes were co-expressed because their protein products are involved in NifH folding and in the biosynthesis and delivery of its [Fe 4 S 4 ] cluster 9 . The Hydrogenobacter thermophilus NifH was identified as vastly superior to the A. vinelandii NifH in terms of expression levels, solubility, and functionality both in tobacco and yeast mitochondria. Mitochondria-targeted H. thermophilus NifH satisfied all functional and spectroscopic requirements of a nitrogenase Fe protein when purified from yeast. The screening also pinpointed the plant mitochondria [Fe-S] cluster assembly as a bottleneck for further engineering.
The workflow of this study is described in Fig. 1b. The gene sequences encoding the 32 NifH variants were cloned into plant vectors for Agrobacterium tumefaciens infiltration-mediated NifH expression in N. benthamiana leaves (Supplementary Table 1, see Methods section for details). The nifH sequences were codonoptimized for S. cerevisiae because codon-usage is similar to tobacco 21 and the workflow included downstream expression of tobacco-selected NifH variants in yeast for biochemical characterization. The genes were under control of the strong and constitutive E35S promoter. Amino-terminal COX4-TS extensions were added to NifH proteins. COX4 is the 29 amino acid transit peptide of the S. cerevisiae mitochondria protein cytochrome c oxidase subunit IV (MLSLRQSIRFFKPATRTLCSSRYLLQQKP), whereas TS denotes the 28 amino acid Twin-Strep-Tag peptide (WSHPQFEKGGGSGGGSGGSAWSHPQFEK) 22 . COX4 targeted NifH proteins to the mitochondria matrix and TS was used to enable variant-independent immunoblot detection of NifH and to facilitate its purification. Importantly, the TS-tag has been shown to not significantly affect NifH functionality 12 . COX4-TS-NifH variants are hereafter denoted as NbNifH Xx where Nb stands for the host N. benthamiana, Xx denotes variants collectively, and other superscripts indicate the species from which NifH sequence was obtained. Vectors with NbNifH Xx constructs additionally contained a transcriptional unit for expression of the green fluorescent protein (GFP) that was used as indicator of successful leaf infiltration (Supplementary Table 1).
An auxiliary vector was constructed to co-express A. vinelandii nifM, nifU, and nifS and target their protein products to mitochondria via N-terminal SU9 extensions. Similar to COX4, the mitochondrial presequence of subunit 9 of the Neurospora crassa F 0 -ATPase 23 (SU9) has been shown to deliver Nif proteins to N. benthamiana mitochondria 24 . NifU and NifS assemble [Fe-S] clusters destined for Nif proteins in A. vinelandii 25 . While not essential for expression of functional NifH Av in S. cerevisiae mitochondria 10 they were required to generate high amounts of active NifB in yeast 11 . As we aimed to identify NifH variants accumulating at higher levels than NifH Av , NifU Av and NifS Av were included in this study. In A. vinelandii and other wellstudied diazotrophs NifM is involved in NifH folding or dimerization prior [Fe 4 S 4 ] cluster acquisition 9,26 . Despite nifM not being present in organisms of some selected nifH variants (Supplementary Data 1), this gene was always included in infiltration experiments for consistency.
Identification of NifH proteins suitable for expression in N. benthamiana. N. benthamiana leaves were co-infiltrated with a 1:1:1 mixture of three distinct A. tumefaciens cultures for expression of, respectively, one NbNifH Xx variant plus GFP, the auxiliary proteins NbNifM Av , NbNifU Av , and NbNifS Av , and the RNA silencing suppressor p19 to enhance the nif transgene expression (Fig. 1b) 27 . Protein extracts were prepared from the N. benthamiana leaves three days after infiltration and analyzed for accumulation of soluble NbNifH Xx using antibodies recognizing the TS-tag. Only two NifH variants were consistently detected among experiments (Fig. 1c, Supplementary Fig. 1a), namely those originating from Methanocaldococcus infernus (NbNifH Mi ) and Hydrogenobacter thermophilus (NbNifH Ht ). A third NifH variant from Methanothermobacter marburgensis (NbNifH Mm ) was detected at low levels at one occasion. In contrast, analysis of total extracts prepared from the infiltrated tobacco leaves showed that, although accumulation levels of the NbNifH Xx proteins varied significantly, 25 of the 32 variants could be detected ( Supplementary Fig. 1b). Only NbNifH expression of variants from Bradyrhizobium japonicum, Rhizobium leguminosarum bv. trifolii, Herbaspirillum seropedicae, Gloeothece sp. KO68DGA, Rhodopseudomonas palustris, Methanothermobacter thermautotrophicus, and Frankia sp. (strain FaC1) could not be demonstrated. Sequence alignments and 3D-modeling of NifH Mi , NifH Ht , and NifH Mm are shown in Supplementary Fig. 2. The 3D-models did not reveal any specific feature that would explain their superior accumulation as soluble protein in tobacco mitochondria, but all three proteins originate from thermophilic organisms (Supplementary Data 1) which could possibly explain their stability and solubility.
Activity of NifH variants isolated from mitochondria of aerobically cultured S. cerevisiae. N. benthamiana screening-identified variants and NifH Av were expressed in S. cerevisiae and purified by Strep-tag affinity chromatography (STAC) to evaluate functionality when targeted to mitochondria. For this, genes encoding COX4-TS-NifH constructs were transferred to expression vectors together with su9-nifM Av , su9-nifU Av , and su9-nifS Av under the control of galactose-inducible GAL1 or GAL10 promoters (Supplementary Table 2, Supplementary Fig. 3a). These COX4-TS-NifH variants expressed in aerobic S. cerevisiae cultures are hereafter denoted ScNifH Mm , ScNifH Mi , ScNifH Ht , and ScNifH Av (ScNifH Xx collectively).
While ScNifH Mm , ScNifH Mi , and ScNifH Ht were purified to near homogeneity (Fig. 2a), SDS-PAGE analysis of ScNifH Av showed additional slower migrating co-eluting proteins. Mass spectrometry confirmed that these were contaminants (Fig. 2a). ScNifH Av solubility was low and much protein was lost to the pellet fraction when preparing the soluble cell-free extract (CFE) explaining its poor purification yield (about 11 mg per kg of S. cerevisiae cells) (Supplementary Fig. 3b-e, Supplementary Table 3). The yield of ScNifH Mm was also relatively low, in line with the inferior result in the N. benthamiana screening. In contrast, the yields of ScNifH Mi and ScNifH Ht were ca. 20 times higher. Iron (Fe) quantification of purified samples was variable but indicated that ScNifH Ht was isolated largely as holo-protein containing one [Fe 4 S 4 ] cluster per dimer (Supplementary Table 3). Consistently, immunoblot analysis showed that ScNifH Ht , ScNifM Av , ScNifU Av , and ScNifS Av had been efficiently targeted to the mitochondria ( Supplementary Fig. 4).
Activities of purified ScNifH Xx variants were determined in vitro using the acetylene reduction assay (ARA) and compared  Fig. 5), indicating that this NifH variant is not compatible with NifDK Av . In contrast, ScNifH Mi was activated to some extent by NifU Av , and further by direct reconstitution, indicating that the A. vinelandii NifUS machinery is not optimal for NifH Mi ( Supplementary Fig. 5). However, activities were very low compared to the as-isolated ScNifH Ht protein (Fig. 2b). This could be explained by NifH Ht harboring more of the conserved amino acid residues known to be important for the interaction with NifDK Av ( Supplementary  Fig. 2). Importantly, soluble accumulation of ScNifH Ht in mitochondria was 20-fold higher than ScNifH Av (Supplementary Table 3), which translates into at least 10-fold higher in vivo activity and fulfills NifH quantity requirements for nitrogenase engineering. Thus, ScNifH Ht was further characterized.
ScNifH Ht exhibits NifH-characteristic spectroscopic signals and is functional in vivo. Purified ScNifH Ht protein presented ultraviolet-visible (UV-vis) absorption spectra typical of O 2sensitive [Fe-S] cluster-containing proteins (Fig. 3a). Aminoterminal sequencing revealed that amino acid residues EQKP remained after COX4 processing (Fig. 3b), where conversion of glutamine (Q) to glutamic acid (E) could be due to deamination performed by the mitochondrial matrix N-terminal amidase NTA1 29 . Electron paramagnetic resonance (EPR) confirmed that ScNifH Ht protein contained an [Fe 4 S 4 ] cluster with similar signal intensity and g-values as NifH Av (Fig. 3c), suggestive of successful maturation into functional Fe protein.
The NifH variant chosen to engineer N 2 -fixing plants must perform P-cluster maturation and FeMo-co biosynthesis in addition to serve as electron donor for substrate reduction. We therefore tested whether H. thermophilus NifH could revert the Nif − phenotype of A. vinelandii DJ77 (ΔnifH strain) 30 . For this, ts-nifH Ht was introduced by transformation into DJ77 and the resulting strain UW481 was tested for diazotrophic growth and in vivo acetylene reduction activity. UW481 showed diazotrophic growth both in solid and liquid media ( Supplementary Fig. 6a, b), and immunoblot analysis demonstrated sustained AvNifH Ht expression and acetylene reducing activity indicative of active nitrogenase ( Supplementary Fig. 6c, d). These data strongly indicate that NifH Ht can replace the functions of native A. vinelandii NifH to some extent, which requires productive interactions with at least apo-NifDK Av , NifDK Av , and NifEN Av proteins.
ScNifH Ht is active in substrate reduction, P-cluster formation and FeMo-co synthesis. Each individual NifH-dependent activity was then analyzed in vitro using pure ScNifH Ht preparations (Fig. 3d). P-cluster maturation was determined by supplementing CFE of A. vinelandii DJ77 (ΔnifH) with ScNifH Ht . The DJ77 extract is devoid of FeMo-co and contains inactive apo-NifDK Av with immature P-clusters. The P-cluster maturation assay using DJ77 CFE relies on positive outcomes of three distinct activities performed in two sequential reactions (Fig. 3d). In the first reaction (Step I + II) pure NifH and FeMo-co are added to DJ77 CFE resulting in NifH-dependent reductive coupling of the two [Fe 4 S 4 ] P-cluster precursors to form mature P-clusters (Step I), followed by FeMo-co insertion into P-cluster containing apo-NifDK Av to generate active NifDK Av (Step II) (Fig. 3d). Tetrathiomolybdate is then added to prevent further FeMo-co insertion, separating the maturation (Step I + II) and activity (Step III) reactions. Activation of DJ77 apo-NifDK Av by ScNifH Ht demonstrated its P-cluster maturation activity (Fig. 3e).
In vitro FeMo-co synthesis (Fig. 3d, Step II) 9 was determined by combining purified preparations of ScNifH Ht , apo-NifDK Av containing P-clusters but devoid of FeMo-co 31    NifB-co bound to the carrier protein NifX Av33 or NifB protein supplemented with Fe and S 34 . As for the P-cluster maturation assay, tetrathiomolybdate was added before the ARA (Fig. 3d, Step III). Figure 3f shows that ScNifH Ht supported FeMo-co synthesis in vitro. Importantly, ScNifH Ht and ScNifB Mt (Methanothermobacter thermautotrophicus NifB isolated from S. cerevisiae) 11 acted together in the NifB-dependent in vitro FeMo-co synthesis assay in which NifB-co was concomitantly synthesized by ScNifB Mt rather than added in purified form. This result proved compatibility of two essential proteins for N 2 fixation, ScNifH Ht and ScNifB Mt , when produced in yeast mitochondria. It also showed interspecies compatibility with NifDK Av and NifEN Av , altogether constituting the conserved biochemical core of nitrogenase.
ScNifH Ht activity in substrate reduction was demonstrated by the ARA and by reduction of N 2 into NH 3 . ARA titration was carried out with a fixed quantity of NifDK Av and increasing amounts of ScNifH Ht . Maximum NifDK Av activity was achieved at molar ScNifH Ht to NifDK Av ratios larger than 40 (Fig. 3g), similar to reactions with the natural counterpart NifH Av 35 . This result suggests that the maximum activity that can be achieved combining ScNifH Ht with NifDK Av is 1000 units (i.e., half of the activity with NifH Av ). In addition, ScNifH Ht supported N 2 reduction into NH 3 by NifDK Av . Importantly, the ratio of NH 3 to ethylene produced by NifDK Av was similar independently of using NifH Av or ScNifH Ht (Fig. 3h).
As-isolated NbNifH Ht was inactive but could be activated by [Fe 4 S 4 ] cluster reconstitution. NbNifH Ht was purified from A. tumefaciens-infiltrated leaves of N. benthamiana. Plants were grown under long-day conditions (16 h light/8 h dark) and leaves were processed at the end of the dark period. Genes encoding Step I P-cluster maturation Step II FeMo-co biosynthesis N  Step I+II: Step III:  Ammonia production (nmol NH3 · min Fig. 3 Characterization of ScNifH Ht . a UV-vis absorption spectra of as-isolated and air-exposed ScNifH Ht . b Processing site (black arrow) of COX4 mitochondria targeting signal (blue) as determined by N-terminal sequencing of ScNifH Ht . The conversion of Q to E (marked by *) could be due to deamination processes. c EPR signal of as-isolated ScNifH Ht (190 μM) compared to NifH Av (71.2 μM). Experimental data (black lines) and simulations (blue lines) of each protein are shown together with g values. d Schematic representation of NifH-dependent activities tested in e-h. Figure adapted from Burén et al. 9 . Copyright 2020 ACS under CC BY 4.0 http://creativecommons.org/licenses/by/4.0. e In vitro P-cluster maturation of apo-NifDK Av present in CFE of A. vinelandii DJ77 (ΔnifH) after addition of ScNifH Ht or NifH Av as indicated (±). Tetrathiomolybdate was added (following step I + II) to inhibit further apo-NifDK Av activation during the ARA (step III). Data represent mean values (n = 2 technical replicates). f NifB-co and NifB-dependent in vitro FeMo-co synthesis using ScNifH Ht or NifH Av . Tetrathiomolybdate was added (following step II) to inhibit further apo-NifDK Av activation during the ARA (step III). A 20:1 molar ratio of ScNifH Ht to NifDK Av was used in the ARA (step III). Data represent mean values (n = 2 technical replicates (NifB-co + NifH Av ), n = 3 technical replicates (NifB-co + ScNifH Ht ), n = 4 technical replicates (NifB Av + NifH Av ), n = 5 technical replicates (ScNifB Mt + ScNifH Ht ), n = 3 technical replicates (ScNifB Mt ), n = 4 technical replicates (ScNifH Ht )). Blue and red dots correspond to independent experiments. g Titration of NifDK Av activity with ScNifH Ht . Positive control reactions performed with NifH Av and NifDK Av at 40:1 molar ratio gave 1692 ± 4 units (nmol ethylene formed per min and mg of NifDK Av ). Reactions lacking NifH (negative control) gave 2.5 ± 0.8 units. Data represent mean values (n = 2 technical replicates). h ARA (red dots, left y-axis) and N 2 -reduction assay (blue squares, right y-axis) using ScNifH Ht and NifDK Av (step III). NifH Av was used as control. Data represent mean values (n = 4 technical replicates).
NbNifH Ht , NbNifM Av , NbNifU Av , and NbNifS Av (together with p19 and GFP) were piled up in a single plant-expression vector for co-expression (Methods section and Supplementary Table 1). Purified NbNifH Ht did not exhibit brown color of [Fe-S] clusters and was inactive in the ARA when combined with NifDK Av (Fig. 4a). Therefore, we reconstituted NbNifH Ht [Fe 4 S 4 ] cluster in vitro either by mixing with Fe, L-cysteine, DTT, and EcNifS Av (direct reconstitution), or by incubating with [Fe 4 S 4 ] clusterloaded EcNifU Av (NifU-mediated reconstitution). Both methods activated the NbNifH Ht as determined by the ARA (Fig. 4a) 36 . Immunoblots detected two differently migrating NbNifU Av species in purifications from tobacco leaves (Fig. 4c). Amino-terminal sequencing showed that both species were cleaved either one or seven amino acids into the TS-tag (Fig. 4d).
As both NbNifU Av species showed the same N-termini processing, we concluded that the faster migrating polypeptide was truncated at the C-terminus.
Extended dark period combined with Fe fertilization produced active NbNifH Ht in mitochondria of N. benthamiana leaves.  (Fig. 4e) but allowed for isolation of active protein as shown below. About 6 mg of NbNifH Ht was consistently isolated per kg of N. benthamiana leaves (Fig. 4f, Supplementary Fig. 7c, Supplementary Table 5). Amino-terminal sequencing showed that mitochondria-targeted NbNifH Ht accumulated as two species (similar to NbNifU Av ), one in which two amino acid residues from the TS-tag were removed with the COX4 signal and another that was processed five amino acid residues further into the TS-tag (Fig. 4g). Functionality of NbNifH Ht isolated from leaves of Fe fertilized tobacco plants following 16 h of darkness was determined using ARA. NbNifH Ht preparations consistently showed activities but these were low compared to those of [Fe 4 S 4 ] cluster-reconstituted NbNifH Ht (Fig. 4h). This result suggested that NbNifH Ht accumulated as two species in tobacco mitochondria, where inactive protein likely lacking [Fe 4 S 4 ] cluster was more abundant than functional and [Fe 4 S 4 ] cluster-containing NbNifH Ht . Consistently, the Fe content of purified NbNifH Ht preparations were below detection limit (Supplementary Table 5). Altogether the results indicate that while soluble NbNifH Ht accumulates in good quantity in mitochondria of N. benthamiana leaves, engineering of additional protein components or biosynthetic pathways will be required to improve [Fe 4 S 4 ] cluster acquisition or stability.

Discussion
The first study reporting production of active NifH in yeast proved that mitochondria is a suitable organelle for hosting O 2sensitive Nif proteins under aerobic growth conditions 10 . Despite being a valid proof-of-concept, further developments with A. vinelandii NifH were limited by low yields as only a small portion was soluble in the mitochondrial matrix. Similar solubility issues were later reported for K. oxytoca NifH targeted to N. benthamiana mitochondria 16 and are confirmed in this study using immunoblot screening and STAC. Identifying the best possible NifH protein for eukaryotic (plant) expression was therefore of uttermost importance. NifH is the most abundant Nif protein required for N 2 fixation in A. vinelandii 17 . Besides being the Fe protein component of Mo nitrogenase, NifH is essential to the assembly of both NifDK cofactors, namely the P-cluster and the FeMo-co 9 .
NifH proteins for nitrogenase engineering in plants should: (i) be stable and soluble at high levels in the mitochondrial matrix, and (ii) be compatible with the NifDK component from a wellstudied model-diazotroph if their own NifDK components are not available in purified form. Compatibility is important when evaluating function of candidate NifH variants. In our case it meant that any selected NifH Xx should be compatible with Nif-M Av (if NifH Xx is not NifM-independent), NifUS Av for maturation and [Fe 4 S 4 ] cluster synthesis/insertion, and NifDK Av for nitrogenase activity measurements. We note that this requirement introduces a selection bias and that the screening could have overlooked NifH variants that were superior to that of H. thermophilus if combined with different NifDK.
The NifH variants tested in this study were selected from a curated dataset of hundreds of NifH sequences by favoring aerobic or plant-associated origins, to overcome the inherent O 2sensitivity of NifH, and functionality at moderate temperatures. We also hypothesized that NifH variants from archaea could function better in a eukaryotic environment as this domain of life is believed to be more closely related to the Eukaryota 37 , and because our previous work expressing archaeal NifB variants in yeast had shown them to be superior to those of bacterial origin 11 .
We expected that most NifH variants would be partly soluble in tobacco mitochondria when expressed together with the accessory proteins NifU Av , NifS Av , and NifM Av . However, only NifH from M. infernus and H. thermophilus were consistently detected in soluble tobacco extracts, in addition to M. marburgensis that was occasionally detected at lower levels. Two of these NifH proteins originated from archaea and the third from a bacterium. One possibility could be that the NifM Av protein was not expressed at sufficient levels in the tobacco mitochondria and that only these three NifH variants did not require NifM for maturation. However, low levels of NifM expression appear to be enough for NifH maturation in K. oxytoca 38,39 . A more plausible explanation can be found in the thermophilic nature of M. infernus, H. thermophilus, and M. marburgensis. It has recently Fig. 4 Characterization of the NbNifH Ht protein. a Activation of as-isolated NbNifH Ht protein with [Fe 4 S 4 ] clusters either by direct chemical synthesis or by EcNifU Av -mediated reconstitution. Activity using NifH Av (positive control) was 1773 ± 10 units (nmol ethylene formed per min and mg of NifDK Av ). Data represent mean values (n = 2 technical replicates). b Immunoblots showing the effect of Fe fertilization on NbNifU Av and NbNifS Av total protein expression. NifS antibody was probed on the same membrane after incubation with Streptactin-HRP. c NbNifU Av protein purified from tobacco plants. The lower band (marked by *) indicates a faster migrating NbNifU Av polypeptide. d Processing sites (black arrows) of the COX4 mitochondria targeting signal (blue) as determined by N-terminal sequencing of the full-length and the faster migrating NbNifU Av . e Immunoblots showing accumulation of NbNifH Ht at the end of 8 h or 16 h night (dark period). f NbNifH Ht protein purified from tobacco leaves. g Processing sites (black arrows) of COX4 signal (blue) as determined by N-terminal sequencing of NbNifH Ht . h ARA of five independent NbNifH Ht STAC-purifications (1)(2)(3)(4)(5). Measured activities using NifH Av (positive controls) and without NifH (negative controls) were, respectively, 2406 ± 53 and 4.6 ± 1.3 (purification 1), 1773 ± 10 and 3.2 ± 2.2 (purifications 2 and 3), and 1692 ± 3.9 and 2.5 ± 0.8 (purifications 4 and 5). All activities are in nmol ethylene formed per min and mg of NifDK Av . Data represent mean values (n = 2 technical replicates). Uncropped immunoblots and gels are shown in Supplementary Fig. 10.
COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-020-01536-6 ARTICLE COMMUNICATIONS BIOLOGY | (2021) 4:4 | https://doi.org/10.1038/s42003-020-01536-6 | www.nature.com/commsbio been reported that the temperature inside respiring mitochondria of cultured human cells is around 50°C 40 , even when the external medium is maintained at 38°C. Whether the same drastic effect on temperature holds true for mitochondria of a leaf cell is not known to us, but it could explain in part the outcome of our NifH screening. None of the two highest expressed NifH proteins originated from proven diazotrophs. We are not aware of any study investigating diazotrophy in the archaeon M. infernus. However, NifB Mi cured the Nif − phenotype of an A. vinelandii nifB mutant strain 41 and, as NifB has no other known function than biosynthesis of nitrogenase active-site cofactors, it is likely that M. infernus is in fact a diazotroph. On the other hand, N 2fixation has been tested but not observed in H. thermophilus TK-6 42 . Interestingly only six NifH variants in our library originated from organisms having genes with high similarity to A. vinelandii nifM. Perhaps other prolyl isomerases could substitute for NifM in these organisms. Whether NifM (and the NifUS machinery) is required for maturation of the three selected NifH proteins (especially NifH Ht ) in mitochondria will be investigated in future work.
Mitochondria-expressed ScNifH Ht was the only variant that supported relevant nitrogenase activity when combined with NifDK Av . Its activity corresponded to roughly half of that using ScNifH Av even if the ScNifH Ht to NifDK Av molar ratio was increased well above 40 normally used for ARA. Emerich and Burris showed that NifH proteins can function with NifDK from other organisms 35 , but this study only combined proteins from bacteria. An optimal growth temperature of 72°C has been reported for H. thermophilus TK-6 43 , which could explain lower ScNifH Ht activity in substrate reduction assays. However, our prediction from this study and previous work on NifB is that suboptimal working temperature of Nif proteins from thermophiles is a price worth paying when engineering nitrogenase in eukaryotes, as solubility and stability of these variants is so much improved.
One observation of this study was that the specific activity of as-isolated NbNifH Ht protein was lower than ScNifH Ht . We think this was caused by poor [Fe 4 S 4 ] cluster availabilityand hence inefficient incorporationor poor NifH [Fe 4 S 4 ] cluster stability within the leaf cell mitochondria. In this context, it is not known how Fe fertilization increased accumulation of soluble NbNifU Av . More available Fe could increase mitochondria [Fe-S] clusters biosynthesis and [Fe 2 S 2 ] cluster occupancy in NbNifU Av which, in turn, would provide stability to the protein. A compatibility issue between NbNifH Ht and NbNifU Av and NbNifS Av is unlikely since EcNifU Av could effectively activate NbNifH Ht in vitro. NbNifH Ht misfolding in mitochondria is also unlikely as it was efficiently activated by reconstitution of its [Fe 4 S 4 ] cluster. It is however likely that protection by respiratory O 2 -consumption in leaf is lower than in yeast. NbNifH Ht exposure to O 2 during leaf processing is also a possibility making this a purely technical problem. While leaves were kept in liquid nitrogen and lysis and purification were performed inside an anaerobic glove box, it is difficult to completely rule out that some O 2 trapped within the leaf was released during tissue disruption.
In conclusion, this study shows that genetic diversity can be exploited to identify, from a very large pool of sequences, the most adequate Nif protein components to engineer a eukaryotic nitrogenase. Modular cloning techniques, gene synthesis with codon optimization, and other synthetic biology tools permit building multi-protein pathways with components of very diverse origin. In this case the NifH protein from H. thermophilus was identified as soluble in mitochondria of both S. cerevisiae and N. benthamiana accumulating at much higher levels than the A. vinelandii homologue. This example is relevant not only because the identified variant performed all three NifH-essential reactions, namely P-cluster maturation, FeMo-co biosynthesis, and NifD-K Av reduction, but also because NifH Ht formed functional interspecies interactions with NifB, NifEN, and NifDK proteins, altogether representing the four proteins constituting the core of diazotrophy.

Methods
Design, assembly, and cloning of the nifH library. A curated dataset of diazotrophs 41 was used to collect nifH candidates and design the library. Genes encoding nifH variants were codon optimized for expression in S. cerevisiae and the sequence encoding pE35S::cox4-twinstrep was codon optimized for expression in tobacco (Supplementary Data 1). All genetic parts were optimized using the GeneOptimizer tool (ThermoFisher) and synthesized by ThermoFisher via the Engineering Nitrogen Symbiosis for Africa (ENSA) project. The nifH genes were synthesized and cloned into pMA cloning vector with BamHI and BstEII restriction sites flanking each gene. The pE35S::cox4-twinstrep sequence was flanked by HindIII and BglII restriction sites.
pGFPGUSplus (plasmid #64401, Addgene) and the pMA vector containing pE35S::cox4-twinstrep were digested with HindIII and BglII and used to generate the parental vector pN2SB41, containing a pE35S::cox4-twinstrep-gus-tNOS transcriptional unit in which gus was flanked by BamHI and BstEII restriction sites. The parental vector pN2SB41 and all pMA vectors containing nifH variants were digested with BamHI and BstEII and used to generate vectors pN2XJ81-pN2XJ112 (Supplementary Table 1).
All DNA digestions were performed using enzymes from New England Biolabs. Ligated products (T4 ligase, Promega) were introduced into E. coli DH5α chemically competent cells and selected on LB (Lysogenic broth) supplemented with appropriate antibiotics. Plasmid extraction was performed using Qiaprep Spin Miniprep kit (QIAGEN) and correct cloning was confirmed by Sanger sequencing (Macrogen).
Growth of S. cerevisiae, mitochondria isolations, and ScNifH purifications. S. cerevisiae for galactose-induced expression of ScNifH Mm , ScNifH Mi , ScNifH Ht , and ScNifH Av together with SU9-NifU Av , SU9-NifS Av , and SU9-NifM Av (XJ1Y-XJ4Y, Supplementary Table 3) were cultured in 4-l fermenters under aerobic conditions (0.625 l of air per minute and l of culture, 250 rpm stirring) and used for mitochondria isolations or NifH purifications as previously described 11 . Preparation of CFE and STAC purifications were performed at O 2 -levels below 1 ppm in anaerobic chambers (Coy systems or MBraun). Typically, cells were resuspended in lysis buffer (100 mM Tris-HCl (pH 8.6), 200 mM NaCl, 10% glycerol, 2 mM sodium dithionite (DTH), 1 mM PMSF, 1 μg/ml leupeptin, 5 μg/ml DNAse I) at a ratio of 1:2 (w/v). Total extracts (TE) were prepared by lysis of the cell suspensions under anaerobic atmosphere using an EmulsiFlex-C5 homogenizer (Avestin Inc.) operating at 20,000 psi. The TE was transferrred to centrifuge tubes equipped with sealing closures (Beckman Coulter) and centrifuged at 50,000 g for 1 h at 4°C (Avanti J-26 XP). The supernatant was filtered using filtering cups with a pore size of 0.2 μm, rendering cell-free extract (CFE) of soluble proteins that was loaded at 2.5 ml/min into a 5 ml Strep-Tactin XP column (IBA LifeSciences) attached to an ÄKTA FPLC (GE Heathcare). The column was washed using 75 ml washing buffer (100 mM Tris-HCl pH 8.0, 200 mM NaCl, 10% glycerol, 2 mM DTH). Strep-Tactin XP column-bound proteins were eluted with 15 ml washing buffer supplemented with 50 mM biotin (IBA LifeSciences). The elution fraction was concentrated, and biotin removed, by passing the protein through PD-10 desalting columns (GE Healthcare). Desalted eluate was further concentrated using centrifugal filters (Amicon, Millipore) with 30 kDa cutoff. Finally, the concentrated protein was snap-frozen in cryovials (Nalgene) and stored in liquid N 2 .
Soil Fe fertilization, preparation of anaerobic N. benthamiana leaf cell-free extracts, and purification of NbNifH Ht and NbNifU Av . N. benthamiana plants were grown under long day conditions (16 h light/8 h dark) with supporting light from 17:00 to 00:00 for 4 weeks. For Fe fertilization experiments, plants were irrigated (2l per week) with tap water supplemented with 1 g/l Sequestrene G100 (Syngenta). Leaves harvested after extended dark period (16 h) were kept in darkness from 17:00 (previous day) until sample collection (09:00 following morning).
Purifications of NbNifH Ht and NbNifU Av were performed at O 2 -levels below 1 ppm inside anaerobic chambers (Coy systema or MBraun). Typically, 200 g of leaf material was harvested and frozen in liquid N 2 . Leaf material was transferred into an anaerobic chamber in frozen condition and disrupted in equal amount (w/v) of lysis buffer (100 mM Tris-HCl pH 8.6, 200 mM NaCl, 10% glycerol, 2 mM DTH, 1 mM PMSF, 1 μg/ml leupeptin, 5 μg/ml DNAseI) using a blender (Oster Classic 4655) operating at maximum power and maintained at 4°C using a circulating water bath. TE was filtered through cheese cloth to remove larger debris. Preparation CFE by centrifugation, Strep-Tactin affinity chromatography, protein elution, concentration, and storage was identical as for yeast-expressed ScNifH proteins. The purification procedure for NbNifU Av only differed in that no DTH was present in the buffers.
Protein methods, antibodies, UV-vis absorption spectrum, and electron paramagnetic resonance. Protein concentrations were measured using the BCA protein assay (PIERCE) in combination with iodoacetamide to eliminate the interfering effect of DTH 45 . Colorimetric Fe determination was performed as reported 46 , and the N-terminal amino acid sequences were determined by Edman degradation (Proteome Factory AG).
The UV-vis absorption spectra were recorded after removal of the DTH from the protein samples using PD-10 desalting columns (GE Healthcare) equilibrated with the corresponding protein buffer wihtout DTH. DTH-free protein samples were then diluted in the same buffers and transferred to a Q6 spectroscopy cuvettes with sealing closures. Absorption (280 nm to 800 nm) was recorded using a UV-2600 spectrophotometer (Shimadzu).
EPR measurements were performed in a Bruker E500 spectrometer equipped with a resonator operating in the TE 102 mode at 9.47 GHz. Temperature was set and stabilized to 10 K by an Oxford temperature controller regulating a gas-flow cryostat refrigerated with helium. For measurements, a microwave power of 2.5 mW and a magnetic field modulation amplitude of 1 mT was used. Experimental conditions were carefully monitored to avoid over-modulation or saturation effects. Simulations of the EPR spectra were performed using the Matlab toolbox Easyspin 47 .
In vitro NifH activity. NifH activity was determined as described by Shah et al. with slight modifications 48 . Reactions were prepared inside anaerobic chambers. Purified NifH proteins were analyzed by ARA after addition of NifDK Av and ATPregenerating mixture (1.23 mM ATP, 18 mM phosphocreatine, 2.2 mM MgCl 2 , 3 mM DTH and 46 μg/ml of creatine phosphokinase, 22 mM Tris-HCl pH 7.5) in a final volume of 600 μl inside 9 ml serum vials under Ar atmosphere containing 500 μl of acetylene (1 atm). The ratio of NifH to NifDK in the assays was 40:1 unless otherwise indicated. The ARA were performed at 30°C in a shaking water bath for 15 min. Reactions were stopped by adding 100 μl of 8 M NaOH. Positive control reactions for acetylene reduction were carried out with NifH Av . Ethylene formed was measured in 50 μl gas phase samples using a Porapak N 80/100 column in a gas chromatograph (Shimadzu).
Reduction of N 2 to NH 3 was determined in reaction mixtures prepared as for the ARA but containing 100 mM 3-(N-morpholino)propanesulfonic acid (MOPS), pH 7.8, as buffer. Mixtures were prepared in volumes of 750 μl, from which 100 μl was removed at assay start to serve as background (t o ) for NH 3 measurements. After exchanging vial atmosphere for N 2 , mixtures were incubated at 30°C for 30 min, and reactions were stopped by addition of 100 μl 5 M EDTA. Twenty-five μl of the blank (t o ) and the reaction (t 30 ) were added in duplicates to 200 μl ophthaldialdehyde reagent solution (ThermoFisher Scientific) in 96-well microplate for fluorescence-based assays (Nunc). Fluorescence (Ex 390 nm, Em 472 nm) was measured using a Varioskan LUX plate reader (ThermoFisher Scientific). NH 3 production was determined from the increase in fluorescence (t 30 -t o ) against standards prepared with NH 4 Cl and recorded in the same plate.
In vitro P-cluster maturation. P-cluster maturation assays were performed inside anaerobic chambers. The in vitro assay combined isolated NifH to be tested (50 μg) with A.vinelandii DJ77 (ΔnifH) CFE (4.34 mg total protein) and an excess of pure FeMo-co (0.85 μM) in 500 μl ATP-regenerating mixture as described above. Reactions were incubated at 30°C for 30 min. Forty μl of 1 mM (NH 4 ) 2 MoS 4 (tetrathiomolybdate) were then added and mixtures were incubated for 10 min at room temperature to prevent further FeMo-co incorporation into NifDK Av during the ARA.
Apo-NifDK Av activation after P-cluster maturation and FeMo-co insertion was analyzed by ARA after addition of an excess of the same NifH species (100 μg) and ATP-regenerating mixture in a final reaction volume of 1 ml. ARA was carried out in 9 ml serum vials containing Ar and 500 μl of acetylene (1 atm) in the headspace for 15 min at 30°C. Positive control reactions for in vitro P-cluster maturation and ARA contained purified NifH Av . Ethylene formed was measured in 50 μl gas phase samples using a Porapak N 80/100 column in a gas chromatograph (Shimadzu).
Following in vitro synthesis of FeMo-co, 17.5 μM (NH 4 ) 2 MoS 4 was added to prevent further FeMo-co incorporation into apo-NifDK Av , and incubated for 10 min at 25°C. Activation of apo-NifDK Av was analyzed by addition of 500 μl ATPregenerating mixture and ScNifH Ht (2.0 μM final concentration) in 9 ml vials containing Ar and 500 μl acetylene. The ARA were performed at 30°C for 20 min. Positive control reactions for ARA contained NifDK Av and NifH Av . Ethylene formed was measured in 50 μl gas phase samples using a Porapak N 80/100 column in a gas chromatograph (Shimadzu).
For "direct reconstitution" activity assays, the activity of [Fe 4 S 4 ] cluster reconstituted NifH protein was determined using ARA. For "NifU-mediated reconstitution", as-isolated NifH protein was mixed with [Fe-S] cluster reconstituted EcNifU Av , and then immediately used for ARA.
Statistics and reproducibility. Distinct samples were used for in vitro activity measurements and sample sizes are indicated by n, where each distinct sample was measured at least two times. Mean of measured activities are shown. The data presented in the figure graphs are listed in Supplementary Data 2.