Regulation of Three Nitrogenase Gene Clusters in the Cyanobacterium Anabaena variabilis ATCC 29413

The filamentous cyanobacterium Anabaena variabilis ATCC 29413 fixes nitrogen under aerobic conditions in specialized cells called heterocysts that form in response to an environmental deficiency in combined nitrogen. Nitrogen fixation is mediated by the enzyme nitrogenase, which is very sensitive to oxygen. Heterocysts are microxic cells that allow nitrogenase to function in a filament comprised primarily of vegetative cells that produce oxygen by photosynthesis. A. variabilis is unique among well-characterized cyanobacteria in that it has three nitrogenase gene clusters that encode different nitrogenases, which function under different environmental conditions. The nif1 genes encode a Mo-nitrogenase that functions only in heterocysts, even in filaments grown anaerobically. The nif2 genes encode a different Mo-nitrogenase that functions in vegetative cells, but only in filaments grown under anoxic conditions. An alternative V-nitrogenase is encoded by vnf genes that are expressed only in heterocysts in an environment that is deficient in Mo. Thus, these three nitrogenases are expressed differentially in response to environmental conditions. The entire nif1 gene cluster, comprising at least 15 genes, is primarily under the control of the promoter for the first gene, nifB1. Transcriptional control of many of the downstream nif1 genes occurs by a combination of weak promoters within the coding regions of some downstream genes and by RNA processing, which is associated with increased transcript stability. The vnf genes show a similar pattern of transcriptional and post-transcriptional control of expression suggesting that the complex pattern of regulation of the nif1 cluster is conserved in other cyanobacterial nitrogenase gene clusters.


Introduction
Filamentous heterocyst-forming cyanobacteria fix atmospheric nitrogen to ammonium under oxic growth conditions. Nitrogen fixation occurs in specialized cells called heterocysts that differentiate at regular intervals in a filament in response to an environment that is deficient in fixed nitrogen [1][2][3][4]. Heterocysts, which comprise 5%-10% of the cells in a filament, have a glycolipid layer that may restrict oxygen diffusion into the cell, lack oxygen-evolving photosystem II activity, and have increased respiration, all of which serve to protect nitrogenase from oxygen [5][6][7][8][9]. Anabaena variabilis is unusual among the heterocyst-forming cyanobacteria in that it has three nitrogenases, which are expressed in cultures grown in different environmental conditions (reviewed in [10]). No other well-characterized cyanobacterial strain has three nitrogenases; in fact, no other characterized strain has even two. The primary nitrogenase that is expressed in cultures growing in an oxic environment that is deficient in fixed nitrogen but has adequate molybdate is the heterocyst-specific Mo-nitrogenase encoded by the nif1 genes [11,12]. In an oxic environment that is low in fixed nitrogen and molybdate, but with vanadate, A. variabilis synthesizes an alternative, heterocyst-specific V-nitrogenase, encoded by the vnf genes [11,13]. The third nitrogenase, a Mo-nitrogenase, encoded by the nif2 genes is made in vegetative cells only under anoxic growth conditions in an environment that is low in fixed nitrogen with molybdate [14][15][16]. Synthesis of all three nitrogenases is repressed in cells grown with a source of fixed nitrogen.
The alternative V-nitrogenase comprises two VnfD (α-subunit), two VnfK (β-subunit) and four δ-subunits, VnfG, forming a heterooctomeric enzyme with two FeV-cofactors [41][42][43]. Like NifH, VnfH, with a [Fe4-S4] cofactor transfers electrons to dinitrogenase. The V-nitrogenase shows different efficiency in substrate interactions than the Mo-nitrogenase; it is relatively inefficient in reducing dinitrogen and thus produces more hydrogen than the Mo-nitrogenase and, unlike the Mo-nitrogenase, it can reduce ethylene to ethane [44]. Because it is an inefficient nitrogenase and produces hydrogen, the V-nitrogenase of A. variabilis has been used to produce hydrogen in an outdoor bioreactor [45].
In the Proteobacteria, the nif genes are structured into multiple operons, including nifHDK, encoding the structural proteins of nitrogenase, nifBQ, producing the proteins required for FeMo-cofactor assembly, nifUVSM, whose products are needed for Fe-S cluster formation, and nifENX, encoding scaffolding proteins for the assembly of the nitrogenase complex [46,47]. In the Proteobacteria, nif genes are under the control of the NtrBC nitrogen regulatory system, which controls synthesis of the regulatory proteins, NifA and NifL [48]. Activation of nif genes in the absence of oxygen and fixed nitrogen requires NifA, as well as the alternative σ 54 RNA polymerase [48]. Similarly, the vnf genes of Azotobacter vinelandii are controlled by the activator VnfA [49,50].
In nitrogen-fixing cyanobacteria there are no homologues of NtrBC, NifA or NifL and there is no homologue of VnfA in A. variabilis. The global nitrogen regulatory protein, NtcA, is required for nitrogen fixation in heterocyst-forming cyanobacteria; however, it is also required for heterocyst formation so its role in activation of nitrogen fixation genes is not yet known [4,51,52]. While no sigma factor specifically associated with nitrogen regulation, like the σ 54 factor in Proteobacteria, has been identified in cyanobacteria, the sigma factor encoded by sigE is important, but not essential, for expression of the nif genes in Anabaena sp. PCC 7120 [53].
In A. variabilis, and in most nitrogen-fixing cyanobacteria whose genomes have been sequenced, the nif gene clusters comprise, in the same order, nifB, fdxN, nifS, nifU, nifH, nifD, nifK, nifE, nifN, nifX, and nifW. All nif clusters also have hesAB and fdxH as well as several conserved unidentified ORFs [54]. In A. variabilis and in Anabaena sp. PCC 7120, the nifD gene is interrupted by an 11-kb element that is removed from the chromosome of heterocysts by an excisase, XisA, late in heterocyst differentiation [55][56][57]. Transcription of the nif genes was first reported over 30 years ago [58,59]; however, little progress has been made in identifying key regulatory mechanisms. It has been assumed, based on Northern blot results and the assumed similarity to Proteobacteria, that the large nif cluster in cyanobacteria comprises several distinct operons: nifB-fdxN-nifS-nifU [58,60], nifHDK [59,61,62], as well as hesAB [63], fdxH [64] and, by default, nifENXW. In addition to the large conserved cluster, nifP is located just upstream of the nifVZT operon [39]. In A. variabilis, nifP is located about 11 kb downstream from the 3' end of the large nif1 cluster. The best evidence for nif promoters in Anabaena are for those genes in which the apparent transcription start sites have been mapped. These include nifB [58,60,[65][66][67], nifH [58,68], hesA [63,67], and fdxH [64]. Recent work from our lab that is described in more detail here has shown that there is a strong promoter driving nifB1 and a separate promoter for hesA1 in A. variabilis, but there is no promoter for nifH1 or fdxH1 [65,69]. While most of the expression of the large cluster of nif1 genes in A. variabilis is driven by the nifB1 promoter, there are additional weak promoters, including one in the nifU1 gene and in the nifE1 gene, that supplement transcription from nifB1 [65,69]. While neither nifH1 nor fdxH1 has a promoter, the "transcription start sites" that were mapped upstream of these two genes are actually processed 5' transcript ends, not 5' primary transcription start sites [65,69].

Organization and Evolution of nif/vnf Gene Clusters in A. variabilis
The organization of the three nitrogenase gene clusters of A. variabilis is shown in Figure 1 [11,16,70]. The nearest relative of the nif1 cluster of A. variabilis is the sole nif cluster in Anabaena sp. PCC 7120; however, the nif cluster in Anabaena sp. strain PCC 7120 has a 55-kb excision element in fdxN that is not present in A. variabilis [71,72]. The nif1 cluster is also very similar to the nif clusters in other heterocyst-forming cyanobacteria. In Anabaena spp., the nitrogenase encoded by these nif1-type genes is expressed only in heterocysts, even under anoxic growth conditions [16,73]. The similarly organized nif2 cluster in A. variabilis is most similar in overall gene organization and gene similarity to the sole nif cluster of Chroococcidiopsis thermalis PCC 7203, a strain that belongs to a group of unicellular non-heterocystous cyanobacteria that grow in extreme environments and fix nitrogen only under anoxic conditions [74]. The similarity of the nif2 gene cluster to the nif cluster in Chroococcidiopsis thermalis PCC 7203 is interesting, since the Chroococcidiopsis group is the closest relative of the heterocyst-forming cyanobacteria, based on 16S rRNA phylogeny [75]. In particular, the unusual fusion of the nifE and nifN genes into a single gene in the nif2 cluster and in the nif cluster of C. thermalis suggests that these genes have a common ancestor. Another major difference between the nif1 and nif2 clusters in A. variabilis is the presence of an excision element only in the nifD1 gene. That excision element is present in most, but not all, of the nif clusters in the genomes of sequenced heterocyst-forming cyanobacteria. Although the size of the element and the genes present in these excision elements varies among strains, all of them have a conserved excisase gene that removes the element during heterocyst differentiation, thereby restoring a complete nifD gene to produce the β-subunit of nitrogenase [55,72]. In contrast, the organization of the vnf genes that encode an alternative V-nitrogenase [11] is different from the two nif clusters, in part because synthesis and assembly of the V-nitrogenase depends on the products of some of the genes that make the Mo-nitrogenase, notably NifB, NifS, NifU and possibly several of the small proteins such as NifW, HesA, HesB and ferredoxins [76]. The vnf genes comprise vnfDG, a fusion of vnfD and vnfG, as well as genes vnfK, vnfE and vnfN. Unlike the vnf gene cluster in the Proteobacteria, there is no vnfH near the vnfDGKEN cluster, and in A. variabilis, vnfH is over 20 kb downstream from vnfN, with vupABC between vnfH and the other structural genes (Figure 1). Complete cyanobacterial genome sequences [77] have revealed strains that have genes very similar to the vnf genes of A. variabilis [70]. Fremyella diplosiphon UTEX 481 and Fischerella muscicola PCC 7414 have orthologs of vnfDG, vnfK, vnfE and vnfN as well as the vanadate transport genes, while Fischerella sp. PCC 9339 has orthologs of vnfDG, vnfK, vnfE and vnfN but is missing most of the vanadate transport genes. In contrast, Chlorogleopsis sp. PCC 7702 has orthologs for the vanadate transport genes, and has most of the structural genes for the V-nitrogenase; however, the fused vnfDG gene is missing the vnfD portion that encodes the α-subunit of the enzyme, which is essential for dinitrogenase activity. The presence of V-nitrogenase activity has not been confirmed in any of these strains.

Cell-Type Specific Expression of the Three Nitrogenases in A. variabilis
In cyanobacteria, the best evidence for cell-type specific gene expression comes from imaging of cells expressing reporter genes such a gfp, luxAB or lacZ fused to cyanobacterial promoters. In our research, we have often used promoter:lacZ fusions because it is easy to assay β-galactosidase in the same cultures that are used for imaging and because there is no concern that the microoxic conditions in a mature heterocyst may affect the reporter protein, which might affect levels of GFP [78]. It was first shown in Anabaena sp. PCC 7120, using a Lux reporter, that the nitrogenase genes were expressed only in heterocysts even under anoxic conditions [73]. Similarly, in A. variabilis expression of nifD1:lacZ is confined to heterocysts, whether the cells are grown under oxic or anoxic conditions ( Figure 2), indicating that some aspect of heterocyst development, possibly a heterocyst-specific activator, is required for nif1 gene expression [16].
In contrast, the nif2 genes of A. variabilis are expressed only in cells grown under anoxic conditions and expression is evident within 4-6 h after nitrogen deprivation ( Figure 3, panels A, C) [14,16]. The nif2 genes are poorly expressed in the heterocysts that form under anoxic growth conditions, and the β-galactosidase activity seen in the heterocysts ( Figure 3B) may reflect enzyme that was made in the vegetative cell prior to differentiation, rather than de novo synthesis in heterocysts. In support of this hypothesis, we observe that expression of the nif2 genes is restricted to vegetative cells in filaments that are first grown under oxic conditions, to allow heterocysts to form, and then switched to anoxic conditions ( Figure 3, panels C and D) [15].
Expression of the vnf genes, like the nif1 genes, is restricted to heterocysts in cells grown under oxic or anoxic conditions ( Figure 4) [13], suggesting that, like the nif1 cluster, expression of the vnf genes depends on a signal that is induced during heterocyst development; however, the vnf genes are not expressed unless the cells are starved for molybdate ( Figures 5 and 6) [79]. The fact that the V-nitrogenase requires NifB1 and possibly other gene products in the nif1 cluster [76] is also consistent with the heterocyst-specific expression of the vnf genes. Cells grown in the absence of fixed N with oxygen; fluorescence from cleavage of fluorescein-β-D-galactopyranoside photographed without red cut-off filter (A); or with red cut-off filter (B); Cells grown in the absence of fixed N without oxygen; fluorescence from cleavage of fluorescein-β-D-galactopyranoside photographed without red cut-off filter (C); or with red cut-off filter (D). Arrows indicate heterocysts identified by bright field microscopy. Bar = 10 μM. Reproduced from [16] with permission.

Metal Transport and Its Effect on Nitrogenase Gene Expression
In many bacteria, including Escherichia coli, Rhodobacter capsulatus and Azotobacter vinelandii, high-affinity molybdate transport is mediated by an ABC-type transport system encoded by modABC genes [82][83][84][85]. ModA is the periplasmic component that binds molybdate, ModB is the transmembrane component of the permease, while ModC provides the energy from the cytoplasmic side of the membrane. Transcription of the modABC operon of E. coli is negatively regulated by dimers of ModE that are bound with four molecules of molybdate [86]. A high-affinity molybdate transport system in A. variabilis, with a Km for transport of molybdate of about 0.3 nM, is encoded by genes modA and fused genes modBC [79,87] that are located 2.7 Mb apart in the 6.36 Mb chromosome [70]. Mutants in this transport system cannot fix nitrogen unless molybdate is supplied at high concentrations (about 1 mM) or unless vanadate is supplied to allow assembly of the V-nitrogenase [79]. Cells starved for molybdate and vanadate express the nif1 and vnf nitrogenase genes in heterocysts ( Figure 5); however, because they cannot make nitrogenase, these nitrogen-starved cells produce a very high frequency of heterocysts and overexpress the nitrogenase genes. The addition of vanadate to Mo-starved cells has no effect on the expression of the nif1 genes, while the addition of molybdate, which allows the Mo-nitrogenase to function, turns off expression of the vnf genes, but also decreases nif1 gene expression and reduces heterocyst frequency compared to the Mo-starved cells ( Figure 5). In contrast, the vnf genes are expressed only in the absence of molybdate, with or without vanadate [13]. NifH1, which is made in cells starved for molybdate, can substitute for VnfH in a vnfH mutant strain [13]. Further, in a strain with a vnfH promoter mutation that allows vnfH to be expressed in cells grown with molybdate, VnfH can substitute for NifH1 when that strain has a nifH1 mutation. Thus, the two dinitrogenase reductases for the Mo-nitrogenase and the V-nitrogenase in heterocysts are able to function in place of each other suggesting that they are not involved in determining the metal specificity of these two nitrogenases [13]. This has not been shown in vivo for any other organism; however, using the nitrogenase for A. vinelandii, it has been shown in vitro that VnfH can replace NifH for the synthesis of the FeMo-cofactor and for maturation of the Mo-nitrogenase [88].
Between the vnfDGKEN and vnfH genes in the A. variabilis genome are the vupABC genes encoding the vanadate transport system that supplies vanadate for the V-nitrogenase [89] (Figure 1). The high-affinity vanadate transport system, with a Km of about 3 nM is, to date, the only vanadate transporter that has been characterized. The vanadate transport genes, like the V-nitrogenase genes, are repressed by molybdate [89]. These genes are most similar to the tungstate transport genes of Eubacterium acidaminophilum. Similar genes are not present in the complete genomes of other bacterial strains that are known to have a V-nitrogenase, including A. vinelandii, Rhodopseudomonas palustris, and Methanosarcina barkeri, although the complete genome sequences of the cyanobacteria Fremyella diplosiphon UTEX 481, Chlorogloeopsis sp. PCC 7702, and Fischerella muscicola PCC 7414 have orthologs of the vanadate transport genes.
When A. variabilis is grown in a medium without fixed nitrogen and with less than 1.0 nM Mo and V, the cells become starved for both metals; however, slow growth continues, accompanied by low levels of nitrogen fixation [80]. This slow growth is abolished in a nifDK1 mutant lacking the heterocyst-specific Mo-nitrogenase, but slow growth continues in a mutant lacking the V-nitrogenase, suggesting that only the Mo-nitrogenase is able to support slow growth in an environment with little molybdate or vanadate. Tungstate is transported by the molybdate transporter and could, theoretically, be incorporated into a nitrogenase [87]. The addition of tungstate, vanadate, or molybdate to cells starved for these metals resulted in an increase in nitrogenase activity, as measured by acetylene reduction, after two hours and this increase required new protein synthesis, suggesting that new nitrogenase was being synthesized with all these metals [80]. While tungstate functioned about as well as vanadate in supporting acetylene reduction, the cells to which tungstate was added did not grow any better with tungstate than with no added metal and did not produce ethane (Figure 7) [80]. A mutant lacking the V-nitrogenase showed no increase in nitrogenase activity upon addition of tungstate, suggesting that the V-nitrogenase, rather than the Mo-nitrogenase, was able to incorporate tungstate (Figure 7). Tungstate was able to substitute for molybdate in repressing transcription of a Mo-transport gene, but not the vnfH gene, which was, however, repressed by Mo [80]. This suggests that the Mo-dependent regulator of the molybdate transport system, probably the product of the modE homolog located just upstream of modA [79], interacts differently with molybdate/tungstate than the Mo-dependent regulators of the vnf genes, VnfR1 and VnfR2 [81] discussed in more detail below.

Figure 7. Metal-induced increase in nitrogenase activity in cells starved for molybdate and vanadate.
A. variabilis strains FD (WT) and MB2 (vnfDG mutant) were grown in Mo-and V-free medium for at least 10 generation to deplete internal stores of these metals and then Na2MoO4, Na3VO4 or Na2WO4 (all at 100 nM) were added to these starved cells at 0 time. Acetylene reduction was measured for strain FD at 2 h, 4 h, 6 h and 24 h after metal addition (A) and for strain MB2 (vnfDG mutant) at 6 h after metal addition (B). The inset shows the ethane (% of ethylene) produced by strain FD with added Na3VO4, which were the only samples that produced any ethane. Reproduced from [80] with permission.

Transcription of Nitrogenase Genes
By analogy with the nif operons of other nitrogen-fixing bacteria, including Klebsiella and Azotobacter, it has been thought that the large cluster comprising most of the nif genes in Anabaena could be divided into several discrete operons, including nifBSU, nifHDK and nifENX. Northern blots appeared to confirm this and, in fact, putative transcription start sites were mapped for the nifB, nifH, hesA and fdxH genes in the nif clusters of Anabaena spp. [58,60,[63][64][65][66][67][68]. We mapped what appeared to be transcription start sites for vnfDG and vnfH (although they are actually processing sites) and confirmed that the apparent transcription start sites for nifB1 and nifH1 in A. variabilis were identical to those mapped in Anabaena sp. PCC 7120. We also identified additional weak promoters within the coding regions of nifU1 and nifE1 [65,69]. However, when we attempted to use the nifH1 promoter to drive expression of lacZ, using a 300-bp promoter fragment that extended at least 150 bp upstream from the putative nifH1 transcription start site, there was no reporter activity. A strain in which this same 300-bp fragment was used to drive expression of the nifH1 gene failed to grow under nitrogen-fixing conditions and had no nitrogenase activity [90]. The same problem occurred when we attempted to drive transcription of lacZ with the putative vnfDG promoter [81] or with the vnfH promoter [65]; there was no expression using these promoter regions, although they extended well upstream from the putative transcription start sites. However, the promoters of other genes gave good activity, including nifB1 [65], ava4025, the gene upstream of vnfDG [81], and vnfR2, the gene upstream of vnfH [65]. The answer to this puzzle came when the 5' ends of these transcripts were characterized and nifH1, fdxH1, vnfH, and vnfDG were found to have the 5' monophosphate end of a processed RNA, rather than the 5' triphosphate characteristic of a primary transcript (Figure 8) [65,69,81]. The difference in the 5' end structure of the mRNA can be determined by using a technique called RNA Ligase Mediated Rapid Amplification of cDNA Ends (5' RACE) with RNA that is treated, or not treated, with tobacco acid phosphatase (TAP), which converts a 5' triphosphate end to a 5' monophosphate end in preparation for ligation of the RNA adapter to the 5' end of the transcript. If the transcript is a primary transcript, the RNA adapter cannot ligate to the 5' triphosphate of the RNA unless it is treated with TAP; however, if the transcript is processed, it already has a 5' monophosphate and does not require TAP treatment. A 5' RACE product that is made equally well with RNA treated or not treated with TAP provides good evidence that the transcript is the result of processing in vivo. If very little 5' RACE product is made when the RNA is not treated with TAP, this indicates that the mRNA is a primary transcript. As shown in Figure 8, TAP treatment was not required to produce strong products by 5' RACE for nifH1, fdxH1, vnfDG, or vnfH, while TAP treatment was required for 5' RACE amplification of transcripts for nifB1, nifU1, hesA1, ava4025 and vnfR2 [65,69,81]. (B) hesA1, requiring TAP, is a primary transcript, but fdxH1 is a processed transcript; (C) ava4025, requiring TAP, is a primary transcript, but vnfDG is a processed transcript; (D) vnfR2, requiring TAP, is a primary transcript, but vnfH is a processed transcript. tss, transcription start site; tps, transcriptional processing site. Arrows indicate the PCR products that were sequenced to determine the 5' transcript ends whose location is shown on the gene maps. Gel images in panels A and D are reproduced from [65], in panel B from [69], and in panel C from [81], with permissions.
Transcription of the nif1 gene cluster of A. variabilis appears to depend primarily on the promoter for the first gene in the cluster, nifB1. While there is a promoter inside nifU1, it is very weak compared to the nifB1 promoter and a strain in which nifH1 is driven only by the nifU1 promoter fixes nitrogen poorly compared to the wild-type strain [65]. There is no promoter upstream of nifK1 or nifE1, so their transcription depends on the nifB1 and nifU1 promoters. Like the nifU1 promoter, the promoter within nifE1 is weak, suggesting that it serves an auxiliary rather than a primary function for gene expression [69]. Further support for the importance of the nifB1 promoter in expression of the far downstream genes, including nifKENXW1 is the near loss of these transcripts in a mutant strain that lacks xisA, the gene that makes the excisase that removes the 11-kb element from nifD1 [69]. In this mutant, the nifB1 and nifU1 promoters cannot drive expression of the genes downstream from the 11-kb element, which is not excised, and these genes are poorly transcribed (Figure 9). Even hesA1, which has its own promoter, shows decreased expression in the xisA mutant, which suggests that the nifB1 promoter is capable, at least partially, of driving transcription of a gene that is 14 kb away. Consistent with the fact that the nif1 cluster encodes an enzyme that functions only in heterocysts, the nifB1, nifU1 and nifE1 promoters showed heterocyst-specific expression ( Figure 10) [69]. Figure 9. Transcript abundance of nif cluster genes in the wild-type strain compared to BP669, an xisA mutant. (A) The amount of transcript for genes in the nif cluster (relative to nifB1) was determined by RT-qPCR using RNA isolated from the wild-type strain, FD and from BP669, which cannot remove the 11-kb excision element in nifD1. Strains were grown with ammonium and then nif genes were induced by 24 h of starvation for fixed nitrogen; (B) Transcript levels in BP669 are shown relative to the wild-type strain, FD, in order to more clearly visualize the low levels of transcript for nifK1, nifE1, nifN1, nifX1, and nifW1 in the mutant. Reproduced from [69], with permission. The primary promoter for the structural genes for the V-nitrogenase, vnfDGKEN, is the Mo-repressible promoter for the gene upstream from this cluster, ava4025. The predicted product of this gene shows similarity to the periplasmic component of molybdate transporters, suggesting that it may have a role in sensing molybdate in the environment; however, a mutant in ava4025 has no apparent phenotype and the gene is not required for Mo-repression of vnfDGKEN [81]. Although the ava4025 promoter controls expression of ava4025 and vnfDG, levels of vnfDG transcript are about 500-fold higher than ava4025, perhaps resulting from increased stability of the vnfDG transcript, which is processed at the site that was initially identified as the transcription start site [11,81]. Expression of vnfDG, under the control of the ava4025 promoter is heterocyst specific (see Figures 5 and 6) [81]. Like vnfDGKEN, vnfH, encoding the dinitrogenase reductase component of the V-nitrogenase, is the result of the processing of a transcript that is made from the promoter of the upstream gene, vnfR1. Although we initially reported, based on Northern blots, that vanadate transport genes, vupABC, form an operon [89], it now seems possible that the promoter for vnfR1, located upstream of the vupABC cluster may control these genes as well as the gene between vnfR1 and vupABC, which may be vnfV, and that the vupABC transcripts may also result from RNA processing; however, this hypothesis awaits experimental support.
Little is known about the control of nif genes that function under anoxic conditions, including the nif2 cluster in A. variabilis; however, the conservation of the organization of the entire cluster suggests that these genes may also be under the control of a single primary promoter. There are striking similarities in the sequences of nifB1 and nifB2 in the region upstream of the nifB1 transcription start site, including conserved motifs, that suggest that the two nifB genes have some aspects of regulation in common ( Figure 11). To test this hypothesis we created a fusion between the upstream region of nifB2, up to and including the first conserved motif shown in turquoise (Figure 11), to the downstream region of nifB1 and then fused the hybrid promoter to GFP (strain JJ146). Expression of the nifB2:nifB1 hybrid promoter fused to GFP was localized specifically to heterocysts as was GFP expressed from the control nifB1 promoter (strain JJ72) (unpublished data and Figure 12). This heterocyst-specific expression of the hybrid promoter indicates that the conserved region upstream of the promoter of nifB1 may serve in regulation that senses oxygen levels, since that signal is the primary one that induces expression of nifB2, and that heterocyst specificity may be conferred by sequences closer to the promoter that are not shared with the nifB2 promoter and by heterocyst-specific protein(s) that may recognize these sequences. An understanding of the roles of the various elements in the promoter and upstream regions awaits a more detailed genetic analysis of both promoters and the identification of proteins that may control cell-specific transcription of the different nitrogenase genes. Figure 11. Similarity of the nifB1 and nifB2 regions upstream from the promoter. Alignment of the region upstream of the transcription start site of nifB1, shown in red, with a similar region upstream of nifB2. A putative extended −10 region is highlighted in magenta, a conserved TGAGTATA motif is highlighted in green, and another conserved motif is highlighted in turquoise. The T that is underlined at the end of the first motif in turquoise indicates the fusion site of a nifB2:nifB1 hybrid promoter (see Figure 12).

Figure 12.
Heterocyst-specific expression of the nifB1 promoter and a nifB2:nifB1 hybrid promoter. The wild-type nifB1 promoter was fused to the reporter gene, gfp, and a hybrid promoter of nifB2/nifB1, fused at the first nucleotide following the first conserved motif shown in turquoise in Figure 10, was also fused to gfp using fusion PCR, and the construct was integrated into the A. variabilis genome as described previously [65,81]. Strains were imaged by confocal microscopy 24 h after nitrogen step-down. GFP fluorescence was localized to heterocysts, which are green in the images. The map of the fusion shows the location of the tss in nifB1 at −328 bp from the translation start site and the approximate location of the putative tss of nifB2. The vertical green arrows on the map show the locations of the three conserved motifs that are highlighted in green in Figure 11.
The lack of discrete operons in the nif1 clusters is inconsistent with the differences in transcript levels for different nif genes, especially for the highly abundant nifH1. The nifH1 transcript is present in much greater quantity than nifU1, the gene directly upstream of nifH1, and the other structural genes for nitrogenase, nifD1 and nifK1, also show high levels of transcription compared to nifB1 ( Figure 9A). If a strong promoter is not directly driving transcription of nifHDK1, then the higher levels of these transcripts may result from their stability. A striking feature of the region near the transcriptional processing sites of nifH1, vnfH1, vnfDG, and fdxH1 is the presence of conserved stem-loop structures that may play a role in stabilizing the transcript [65,69]. By measuring RNA by RT-qPCR at various times after the addition of rifampin, which inhibits initiation of transcription by RNA polymerase, we determined the half-lives of most of the genes in the nif1 gene cluster [69]. The half-life of nifH1 is much longer than the genes upstream of the processing site and the half-lives of the transcripts downstream of nifH1 decline with increasing distance from the processing site [69] (Table 1), suggesting that transcript stability plays a major role in controlling the relative amount of transcript. A mutant strain in which the stem-loop structure located at the processing site of nifH1 is abolished shows a shorter half-life for the nifH1 transcript than the wild-type strain and nitrogen fixation is strongly inhibited in this mutant (Thiel; unpublished data).

Proteins Involved in Regulation of Nitrogenase Genes
The global regulator NtcA affects many genes that respond to nitrogen availability in the cell and its binding sites have been the subject of several studies [52,91,92]. Although NtcA is a regulator in all cyanobacteria, its significance in heterocyst-forming cyanobacteria is the key role it plays in sensing nitrogen starvation and initiating the complex process of heterocyst differentiation [3,4]. In addition to its role in activating genes that are required for the differentiation of heterocysts, NtcA activates expression of PipX, which is thought to work in concert with NtcA to allow full expression of late heterocyst-specific genes, including nifH, coxB3 and coxB2 (encoding heterocyst-specific cytochrome oxidases). A pipX mutant shows low levels of expression of these genes and is impaired in nitrogenase activity [93,94]. However, since the expression of the nif operon depends on low oxygen levels that result from high respiration that is mediated by the products of the cox2 and cox3 genes, the low levels of nif gene expression may be a secondary effect of relatively high oxygen levels in the heterocyst resulting from low levels of cox gene expression in the pipX mutant. In Anabaena sp. PCC 7120, patB was shown to be important for growth in the absence of fixed nitrogen [95] and we have found that expression of nifB1 requires PatB1 and expression of nifB2 requires PatB2 (Thiel; unpublished data). Similarly, in a non-heterocystous cyanobacterium a PatB homologue called CnfR has been identified as a key regulator of nif gene expression [96].
NtcA has been reported to bind weakly to a region upstream of nifH in Anabaena sp. PCC 7120 and a putative non-canonical NtcA binding site was identified [68,97]. However, recent ChIPSeq data for Anabaena sp. PCC 7120 showed that NtcA did not bind to any region upstream of nifH, but rather to a site within the coding region of nifH in Anabaena sp. PCC 7120 [52]. In this study NtcA was found to bind upstream of nifB, but the binding site was hundreds of nucleotides upstream of the transcription start site of nifB [52], suggesting that NtcA does not directly control expression from the nifB promoter. Using the nifUH1 intergenic region of A. variabilis as the target, we were unable to detect binding of NtcA and mutations in the putative NtcA binding site in this region had no effect on expression of nifH1 [65]; however, this is not surprising since we could find no evidence of a promoter in the nifUH1 intergenic region (see Section 5 above). There may be differences in nifHDK regulation between Anabaena sp. PCC 7120 and A. variabilis, and the putative NtcA binding site upstream of nifH in Anabaena sp. PCC 7120 is not well conserved in A. variabilis. However, because of the high degree of overall sequence homology between the two nif1 clusters in both strains, it seems unlikely that the same 5' nifH transcript end, found in both strains, results from fundamentally different processes. Further, a RNAseq mapping technique that identified transcription start sites (and excluded processed sites) for the genome of Anabaena sp. PCC 7120 found the anticipated nifB transcription start site at the published site, but failed to identify the putative nifH transcription start site, even though it is found at levels of at least 20-fold higher that nifB [67]. It is clear that NtcA is important for expression of the nif genes, but its effect is likely to be indirect, reflecting the fact that it may be essential for the expression of other genes whose products may act more directly to regulate expression of the nif genes.
The role of NtcA in expression of the nif2 cluster is also not clear. An ntcA mutant of A. variabilis failed to fix nitrogen using the nif2-encoded Mo-nitrogenase, indicating that NtcA has an important role in expression of this enzyme; however, there is no canonical NtcA binding site, GTAN8TAC, anywhere in the region that is likely to have the promoter [15]. Because of the diversity of NtcA binding sites and of their locations relative to the start of genes [52] as well as the global effect of NtcA regulation it may be difficult to assign a specific role for NtcA in expression of these nitrogenase genes.
There are at least two proteins that are repressors of vnfDG, VnfR1 and VnfR2. VnfR1 is encoded by ava4042, located upstream of the vanadate transport genes, vupABC (Figure 1) [89]. VnfR2 is encoded by ava4055, upstream from vnfH, which encodes the dinitrogenase reductase for the V-nitrogenase. The promoter for vnfR2 serves both vnfR2 and vnfH (Figure 8) [65]. These proteins with a conserved N-terminal helix-turn-helix motif show 73% protein identity and act as Mo-dependent repressors that independently repress transcription of ava4025-vnfDG in cells grown with molybdate [81]. Although each protein can repress expression of ava4025-vnfDG, only VnfR1 binds specifically, in vitro, to a region upstream of the ava4025 promoter. Cells lacking either vnfR1 or vnfR2 still show heterocyst-specific, Mo-repressed expression of ava4025-vnfDG. A mutant lacking both vnfR1 and vnfR2 expresses ava4025-vnfDG in the presence of Mo and expression is heterocyst specific (Figure 6), indicating that other factors activate expression of this promoter in heterocysts.

Conclusions
Although there are three nitrogenases in A. variabilis, and two of them are Mo-nitrogenases, the tight control of expression of the nif1, nif2 and vnf genes ensures that the cell makes the correct enzyme for the environment in which it is growing. This regulation includes differential expression of nitrogenases in response to cell differentiation, oxic versus anoxic growth conditions, and for environments with or without molybdate. The nif2 cluster has not been found in any other well-characterized heterocyst-forming cyanobacterium and it shares an evolutionary origin with nif genes from non-heterocystous cyanobacteria especially with the primitive cyanobacterium Chroococcidiopsis thermalis [75,98], suggesting that an ancestor of this unicellular cyanobacterium may represent the evolutionary origin of the nif2 genes. The vnf genes are not present in well-studied cyanobacteria; however, these genes have recently been found in cyanobacteria in a lichen symbiosis [99], suggesting that they may be represented in symbiotic interactions. The culturable cyanobionts from the water fern Azolla filiculoides, Anabaena spp. [58,100], are virtually indistinguishable morphologically and physiologically from A. variabilis and also have vnf genes [11] and nif2 genes [16]; however, the non-culturable Azolla symbiont, Nostoc azollae 0708, shows a degraded genome incapable of supporting independent growth and this strain lacks both vnf and nif2 genes [101].
While we understand at an environmental-response level and even at a whole-cell level how these three nitrogenases are regulated, information at the molecular level is still lacking. The fact that at least two of the three gene clusters employ RNA processing, and its associated transcript stability, as a regulatory mechanism suggests that this may be a more general mechanism of cyanobacterial gene regulation; however, that needs to be tested experimentally. While several promoters that show late heterocyst-specific gene expression have been identified, including a number that are described here, we still do not understand how those genes are activated late in heterocyst development. Proteins NtcA [4] and SigE [53] are important for expression of late heterocyst genes, but their specific function in controlling these genes is not known. Research is still needed to understand how the environmentally important process of nitrogen fixation and the synthesis of associated essential proteins, such as the uptake hydrogenase, ferredoxins, and cytochrome oxidases, are regulated.