Comparative genomic insights into culturable symbiotic cyanobacteria from the water fern Azolla.

Species of the floating, freshwater fern Azolla form a well-characterized symbiotic association with the non-culturable cyanobacterium Nostoc azollae, which fixes nitrogen for the plant. However, several cyanobacterial strains have over the years been isolated and cultured from Azolla from all over the world. The genomes of 10 of these strains were sequenced and compared with each other, with other symbiotic cyanobacterial strains, and with similar strains that were not isolated from a symbiotic association. The 10 strains fell into three distinct groups: six strains were nearly identical to the non-symbiotic strain, Nostoc (Anabaena) variabilis ATCC 29413; three were similar to the symbiotic strain, Nostoc punctiforme, and one, Nostoc sp. 2RC, was most similar to non-symbiotic strains of Nostoc linckia. However, Nostoc sp. 2RC was unusual because it has three sets of nitrogenase genes; it has complete gene clusters for two distinct Mo-nitrogenases and an alternative V-nitrogenase. Genes for Mo-nitrogenase, sugar transport, chemotaxis and pili characterized all the symbiotic strains. Several of the strains infected the liverwort Blasia, including N. variabilis ATCC 29413, which did not originate from Azolla but rather from a sewage pond. However, only Nostoc sp. 2RC, which produced highly motile hormogonia, was capable of high-frequency infection of Blasia. Thus, some of these strains, which grow readily in the laboratory, may be useful in establishing novel symbiotic associations with other plants.

Data Summary

Genome assemblies of the newly sequenced strains are available at NCBI under the following GenBank assembly accession numbers: GCA_014222145.1 (Trichormus variabilis 9RC), GCA_014222135.1 (Trichormus variabilis ARAD), GCA_014222245.1 (Trichormus variabilis FSR), GCA_014222155.1 (Trichormus variabilis N2B), GCA_014222225.1 (Trichormus variabilis PNB), GCA_014222125.1 (Trichormus variabilis V5), GCA_014222165.1 ( sp. 2RC), GCA_014222255.1 ( sp. UCD120), GCA_014222285.1 ( sp. UCD121) and GCA_014222275.1 ( sp. UCD122). The authors confirm that all comparative data, supporting data, code and protocols have been provided within the article or through supplementary data files.

A variety of plants form beneficial symbiotic associations with cyanobacteria, allowing the plants to thrive in soils that lack key nutrients, such as fixed nitrogen. Genome sequences were determined for a group of symbiotic cyanobacteria that were isolated from the water fern Azolla, a plant that can provide nitrogenous fertilizer to rice plants, particularly in developing countries. Genes and pathways that confer key characteristics important in symbiosis include those for nitrogen fixation, motility and the ability to use sugars for growth. Some of these strains, which grow readily in the laboratory, can infect plants other than Azolla and, thus, may be useful for extending these ecologically important symbiotic associations to agriculturally important plants.

Introduction

Species of the floating, freshwater fern Azolla are found throughout the world in Asia, Africa, and North and South America in temperate, tropical and subtropical habitats. Through a well-characterized symbiotic association with the cyanobacterium Nostoc azollae, which fixes nitrogen in the leaves, the Azolla plant can serve as a highly effective green manure for growing rice in Vietnam and China [1]. The cyanobiont, which is found exclusively in the periphery of an extracellular cavity on the dorsal side of the fern leaves, is transmitted during sexual reproduction vertically to the next generation via cyanobacterial filaments that are associated with the megasporocarp of the fern, without de novo infection [2-4]. Within the cavity, the cyanobiont differentiates a high percentage of heterocysts, which fix nitrogen using photosystem I and glycolysis for energy while relying on plant photosynthesis for fixed carbon.

A free-living, heterocyst-forming cyanobacterium, originally called Anabaena azollae, was isolated from Azolla and grown in culture as early as 1979 [5], but by 1986 and 1987 there was evidence, based on RFLPs, that the strain freshly harvested from Azolla was distinct from the free-living strains [6, 7]. Genetic differences in nif gene restriction sites were confirmed and it was also demonstrated that the major cyanobiont lacked the 11 kb nifD excision element that was present in the cultured strain [8]. Characterization of the freshly isolated cyanobiont cells, including sequencing of 16S rRNA genes, demonstrated that there were typically only minor differences in strains from different plants; however, other cyanobacteria have also been detected [9-13]. More recently, nearly 60 bacterial isolates, comprising nine genera outside the cyanobacteria, have also been cultured from Azolla [14].

The sequencing of the genome of a true, non-culturable cyanobiont, Nostoc azollae 0708, confirmed that it is different from the culturable forms, with reductive evolutionary degradation of the genome, such that the strain cannot grow outside of the plant [15]. Comparison of the phylogenies of six species of Azolla with those of their cyanobionts shows clear evidence of a pattern of co-speciation between the plant and its cyanobacterial partner [16, 17]. While Nostoc azollae 0708 is non-culturable [15, 18], the role of the culturable strains found in the symbiotic association remains unknown.

Culturable cyanobacteria form various types of symbiotic associations with plants, including intracellular associations (e.g. with Gunnera) and endophytic associations, within specialized cavities (e.g. Azolla, Blasia and Anthoceros) [19-23]. A culturable strain from Azolla, first named Anabaena azolla, was isolated in 1979 from Azolla caroliniana. The strain was difficult to isolate and propagate from Azolla and required disruption of the plant tissue to free the cyanobacteria from the fern leaf cavities, suggesting that it was not a surface contaminant [5]. The cultured strain was very similar to Anabaena flos-aquae ATCC 22664, a strain that showed cross-reactivity to antibodies made against Newton’s culturable strain of Anabaena azollae. [5]. Zimmerman analysed 10 strains cultured from Azolla from multiple laboratories using morphology, enzymes and lectins to determine their similarity. Five of the strains appeared to be nearly indistinguishable from the free-living strain, () variabilis ATCC 29413, while the other five were all quite divergent from the N. variabilis ATCC 29413-like strains [24]. Meeks isolated strains N1 (UCD120), A1 (UCD121) and A2 (UCD122) from an extract of cyanobacterial cells from Azolla caroliniana that was used to infect the hornwort, Anthoceros. The cyanobacteria that subsequently grew symbiotically in Anthoceros were excised from the plant tissue, and cultured on cyanobacterial medium [8].

Worldwide, several cyanobacterial strains have been isolated from Azolla, cultured in the laboratory and stored. We obtained and grew 10 strains from the Zimmerman and Meeks collections for comparisons of their genomes, to determine the similarities and differences among these isolates [8, 24]. We were particularly interested in the reported similarity of several of these strains, isolated from Azolla, to N. variabilis ATCC 29143 [24], a strain that was first isolated as Anabaena flos-aquae A-37 [25] from a sewage oxidation pond in Mississippi, with no known relationship to Azolla. Its name was later changed to [26]; however, based on data presented here, we have called it Nostoc variabilis ATCC 29413. The strain was characterized by several laboratories but the early work by Wolk’s laboratory on this strain led to its becoming a model strain for cyanobacterial physiology, nitrogen fixation and heterocyst formation [27-31].

In this study we were primarily interested in answering three questions: (1) How similar are the 10 strains, isolated from Azolla in different locations worldwide, to each other and to well-characterized model strains? (2) Could they infect a plant? (3) What genes do they share that might shed light on the characteristics that define symbiotically competent cyanobacteria? We compared the genomes of the 10 strains to model strains N. variabilis ATCC 29413 and ATCC 29133, and to other cyanobacteria isolated from symbiotic associations with moss and lichens [21, 32]. , isolated from an association with the cycad Macrozamia [33], forms associations with several plants and is a member of a clade that includes other cyanobacteria found in plant associations [34-36]. The phylogenetic relationships among the strains and among some of the genes thought to be associated with symbiosis are presented.

Another question was whether these strains, after years of storage and growth as axenic laboratory cultures, could infect a plant. Since there is no report of successful infection of Azolla with cyanobacteria, we determined whether the cyanobacteria could infect a more tractable model plant, the liverwort Blasia pusilla [20, 23, 37–39]. The endophytic infection of Blasia by motile hormogonia occurs via pores in the extracellular dome-shaped structures known as auricles on the surface of the Blasia plant thallus. After infection, cyanobacterial filaments with a very high frequency of heterocysts grow within the cavities, the pores close and the auricles produce mucilage as well as infiltrating plant structures thought to facilitate nutrient transfer between the cyanobacteria and the plant [20, 40]. Through nitrogen fixation in heterocysts, cyanobacteria provide fixed nitrogen to the plant [41-43] in exchange for sugar from the plant [39, 43, 44]. For the newly sequenced strains and the model strains N. variabilis ATCC 29413 and ATCC 29133, we compared genes associated with symbiosis, including those for nitrogen fixation, sugar transport, motile filaments called hormogonia used for infection of plants [45], and chemotaxis. The genes of interest are described in more detail in the Results.

Methods

Genome sequencing and assembly

Genomic DNA was extracted from N. variabilis variants ARAD, 9RC, FSR, PNB, N2B and V5, UCD strains (UCD120, UCD121 and UCD122), and 2RC, grown in an eight-fold dilution of Allen and Arnon (AA/8) medium [46] containing 5 mM NH4Cl and 10 mM N-[Tris(hydroxymethyl)methyl]-2-aminoethanesulfonic acid (TES), by vortexing cells with glass beads in the presence of phenol [47, 48]. Genomic DNA was treated with RNaseA, further purified with two phenol/chloroform/isoamyl alcohol extractions followed by a chloroform/isoamyl alcohol extraction before ethanol precipitation. DNA was additionally purified and concentrated using a Bio101 Gene Clean II kit. The concentration and purity of the genomic DNA was determined using a NanoDrop (Thermo Scientific) and 0.5 ng of genomic DNA from each cyanobacterial strain was fragmented and tagged with adapters using the protocol provided by the Nextera XT DNA Library Prep Kit (Illumina). Tagged DNA was amplified and index sequences were added using low-cycle Nextera Seq PCR (index Primers i5 and i7, annealing temperature at 55 °C, and 12 cycles). PCR products were purified using AMPure XP (Beckman Coulter) beads and normalized using a bead-based method provided in the Nextera XT DNA Library Prep Kit (Illumina). The library normalization process dilutes the genomic libraries to the same concentration before pooling, thus allowing all libraries to have consistent read depth. Normalized genomic libraries were sequenced on an Illumina Miseq using an Illumina MiSeq Reagent Kit v2 with 2×150 cycles. The Illumina Miseq paired-end reads were assembled within Pathosystems Resource Integration Center (PATRIC, www.patricbrc.org) [49] using Unicycler [50] and annotated using the RAST tool kit, RASTtk [51]. Information on the genomic sequences is provided in Table S1.

Phylogenetic trees and ANI calculations

The genomic phylogenetic tree was created in PATRIC using the Codon Trees pipeline, using a concatenated alignment of 500 randomly chosen amino acid and nucleotide sequences from PATRIC’s global Protein Families (PGFams) [52]. A default setting of 0 was used for both Max Allowed Deletions and Duplications. Alignments were created using muscle [53] for proteins and the Codon align function of BioPython [54] for nucleotides. Genomic maximum-likelihood trees were produced from the concatenated alignments by the program RAxML (Randomized Axelerated Maximum Likelihood) [55] with 100 rounds of rapid bootstrapping. BioNJ distance trees for nifH, vnfR and frtA phylogeny trees were constructed using the program SeaView [56] with 100-replicate bootstrap values. The genes used for these trees are provided in Table S2. The average nucleotide identity (ANI) values for two cyanobacterial genomes were determined using ChunLab’s online ANI calculator (www.ezbiocloud.net/tools/ani) [57]. This calculator uses the OrthoANIu algorithm, which uses USEARCH. OrthoANIu breaks up two genomes into 1020 bp random fragments and identifies pairs of fragments with reciprocal best hits. It then calculates the ANI values for all reciprocal best hits using USEARCH, which serves a similar role as blastn but uses an algorithm that trades sensitivity for speed.

Genome and gene comparisons

The proteome comparisons, based on the deduced proteins for each genome, between a reference strain and closely related cyanobacterial strains (as determined by phylogeny, Fig. 1) were done using the Proteome Comparison tool in PATRIC [49], which uses blastp to calculate protein identity. The Sequence-Based Comparison tool marks each gene by colour as either unique, a unidirectional best hit or a bidirectional best hit compared to a reference genome, and the colour indicates the per cent identity between deduced proteins. The blastp results for all the genomes are displayed as a colour-based circular map, making it easier to identify different or deleted regions. The closest relative of 2RC was determined using Similar Genome Finder in PATRIC [49], using Mash/MinHash [58]. Mash reduces a genome to smaller subsets of sequences to more rapidly determine mutation distances, while MinHash uses the mutational distances and P values to rapidly compare a genome to a massive collection of sequences.

Fig. 1.

Phylogenetic distance tree of genomes for the symbiotic cyanobacterial strains isolated from plants (in blue) and similar comparator strains that were not isolated from a plant (in black). Numbers at the branch points indicate bootstrap values. The value shown to the right of each strain indicates the number of genes that match a set of 74 genes identified as ‘symbiosis genes’ [34]. The criteria for a gene match are described in the Methods. Accession numbers for the ‘symbiosis’ genes of representative strains are provided in Table S3. *UCD120 has a partial copy of the gene that it is missing compared to UCD121 and UCD122. Accession numbers for these strains are provided in Table S1. Scale represents number of substitutions per site.

Gene comparisons were made using the blastn or blastp algorithms [59] to determine nucleotide or amino acid identity, respectively. We used blastp to determine whether the 74 genes identified as ‘symbiosis’ genes in Supplement S2 of Warshan et al. [34] were present in the genomes analysed here. Because the percentage identity for some of the genes that had been compared among the strains shown in supplement S2 of Warshan et al. [34] was only around 50 % (e.g. Moss5 and Moss6 vs. ), we used a minimum cutoff score of at least 49 % amino acid identity and an E value of at least −100 to identify homologues matching the 74 genes identified in supplement S2 of Warshan et al. [34]. These criteria for a match gave results identical to those presented in supplement S2 for the strains listed there. Accession numbers for the ‘symbiosis’ genes of representative strains studied here are provided in Table S3.

Symbiosis

In the Blasia symbiosis, cyanobacteria infect and multiply in auricles, which are dome-shaped structures on the ventral surface of the thallus. Blasia plants were maintained on agar-solidified BCD medium (1 mM MgSO4, 1.84 mM KH2PO4, 10 mM KNO3, 1 mM CaCl2, 45 µM FeSO4۰7H2O, trace minerals, 0.7 % agar) [60]. All cyanobacteria strains used to inoculate Blasia were grown in Bgll0 liquid media [61]. Blasia was transferred to BCD medium lacking nitrate, a few drops of liquid cyanobacterial culture were placed atop the Blasia, and plates were incubated at 25 °C under 60 µE m−2 s−1 light for several weeks. Blasia was checked weekly for infection using a light microscope to screen several thalli (containing 20–30 auricles) to identify infected auricles. For most of the strains we found no infected auricles after screening several thalli (20–30 auricles). For poorly infectious strains, we typically found one or two infected auricles (out of 20–30 auricles). For highly infectious strains, typically about 75 % of the auricles were infected. As can be seen in the light micrographs (Fig. 9), when infection was successful, Blasia auricles were tightly packed with cyanobacterial filaments. In contrast, when infection was not successful, cyanobacteria were loosely associated with the plant cells, but the auricles were empty.

Results

Overall genome comparisons

The newly sequenced strains that were isolated from Azolla (the first 10 strains shown in Table 1) fall into three distinct clusters. A phylogenetic analysis of strains 9RC, ARAD, FSR, N2B, PNB and V5, isolated from Azolla indicated that they are very closely related to the model strain N. variabilis ATCC 29413 (Fig. 1). The ANI values for these strains compared to N. variabilis were greater than 99.9 % (Table S4). The ANI value for the non-symbiotic strain isolated in India, YBS01, is nearly identical to the others in this group. Phylogenetically, this group is also very closely related to the previously described symbiotic strains Moss5 and Moss6 [21] (Fig. 1) with ANI values of 98 % relative to members of the N. variabilis group (Table S4). A comparison of the circular proteome maps (based on the deduced proteins for each genome), shown as concentric rings with colour-coded amino acid identity values (Fig. 2a), confirms that N. variabilis ATCC 29413 is more similar to the strains ARAD, PNB, FSR, N2B, V5 and 9RC than to Moss5 and Moss6 (which are virtually identical to each other). In contrast, Moss5 and Moss6 show lower identity values throughout the genome, with several regions of much less than 90 % identity. All of these strains, except N. variabilis and YBS01, were isolated from a symbiotic association either with Azolla (ARAD, PNB, FSR, N2B, V5 and 9RC) or with feathermoss (Moss5 and Moss6) [21], and have geographically diverse origins (Table 1). The well-characterized non-symbiotic strain sp. PCC 7120 has an ANI value of 92 % compared to the N. variabilis-like strains.

Table 1.

Origins of cultured / strains

Strain designation	Plant source	Source of strain	Location	Citation
Strains grown and characterized in this study
2RC	Azolla pinnata	R. Caudales	Rutgers University, USA	[24]
9RC	Azolla caroliniana	R. Caudales	Rutgers University, USA	[24]
A1 (UCD121)	A. caroliniana	J. Meeks	UC Davis, CA, USA	[8, 24]
A2 (UCD122)	A. caroliniana	J. Meeks	UC Davis, CA, USA	[8]
ARAD	Azolla filiculoides	E. Tel-Or	Hebrew Univ., Israel	[24, 118]
FSR	A. pinnata	R. Fisher	VA Commonwealth Univ., USA	[24]
N1 (UCD120)	A. caroliniana	J. Meeks	UC Davis, CA, USA	[8]
N2B	A. caroliniana	J. Newton	USDA, Peoria, IL, USA	[5, 24]
PNB	A. pinnata	S.Nierzwicki-Bauer	Rensselaer Poly. Inst., USA	[24]
V5	A. pinnata	I. N. Gogotov	IMPB RAS, Pushchino, Russia	[63]
Nostoc (Anabaena) variabilis ATCC 29413	None	C. P. Wolk	Michigan State Univ., USA	[62]
Nostoc (Anabaena) ATCC 29413 strain FD	None	C. P. Wolk	Michigan State Univ., USA	[62, 119]
Comparator strains for this study
Nostoc punctiforme ATCC 29133	Gymnosperm cycad Macrozamia sp.		Australia	[120]
Nostoc Moss2	Moss		Sweden	[21]
Nostoc Moss3	Moss		Sweden	[21]
Nostoc Moss4	Moss		Sweden	[21]
Nostoc Moss5	Moss		Sweden	[21]
Nostoc Moss6	Moss		Sweden	[21]
Nostoc linckia z1	None		Israel	[121, 122]
Nostoc linckia NIES-25 (IAM M-251)	None		Japan	[123]
Nostoc (Anabaena) YBS01	None		Meghalaya, India	Accession number CP034058

Fig. 2.

Proteome amino acid identity maps based on the deduced proteins for each genome for the symbiotic cyanobacterial strains and similar comparator strains that were not isolated from a plant. (a) Strains similar to N. variabilis ATCC 29413. (b) Strains similar to . (c) Strains similar to N. linckia z1. The red box in (a) indicates the absence of an 11 kb excision element in strain FSR.

() variabilis ATCC 29413

() ATCC 29413 strain FD

ATCC 29133

() YBS01

UCD120, UCD121 and UCD122, isolated from Azolla, are very closely related to the model strain and Moss2, their closest known relatives (Fig. 1). The three UCD strains share ANI values greater than 99 % compared to each other and 93 % compared to (Table 1). A comparison of the proteomes (Fig. 2b) shows that the three UCD strains are more similar to each other than they are to and somewhat less similar to the moss symbiont, Moss2. The phylogenetic tree based on the genomes confirms these relationships (Fig. 1).

In contrast, strains related to 2RC form a third cluster, distinct from N. punctiforme. Nostoc 2RC is most closely related phylogenetically to the symbiotic strain, Moss3, and N. linckia z1, with ANI values of 96.6 and 96.8 %, respectively. It is less closely related to , UCD120, UCD121 and UCD122 with ANI values of about 82 % (Table S4). The proteome map (Fig. 2c) indicates that N. linckia z1, which did not come from a plant (Table 1), is more similar to symbiotic strains 2RC and Moss3 than to the non-symbiotic strain N. linckia NIES-25 or the symbiotic strain Moss4. N. linckia z1 has ANI values of 98 % for Moss3 and 89 % for N. linckia NIES-25 and Moss4 strains (Table S4). Among the three clusters of strains, only the branch containing , Moss2 and the three UCD strains comprises known symbiotic strains, while the other two branches include both known symbiotic and non-symbiotic strains. Notable are the nearly identical strains that comprise the N. variabilis-like group, of which only two strains were isolated as free-living cyanobacteria (Table 1). Similarly, Moss3, which came from a plant, shares 98 % ANI (Table S4) with N. linckia z1, which did not.

N. variabilis ATCC 29413 has four circular plasmids, A, B, C and D, as well as E, a 37 kb linear element. When the genome of N. variabilis was originally sequenced by JGI, a laboratory variant strain of N. variabilis called FD [62] was inadvertently the source of the DNA. When N. variabilis was sequenced in another laboratory [63] they found that N. variabilis had plasmid D, which was missing in the FD strain that was sequenced by JGI. We have confirmed the lack of plasmid D in strain FD. Most of the strains that are nearly identical to N. variabilis have genes homologous to those in plasmids A, B, C and D and the linear element E; however, FSR lacks some genes homologous to those in part of plasmid A and both FSR and PNB lack genes homologous to those in the linear element (Fig. 2a). The free-living strain YBS01, like FSR and PNB, also lacks genes homologous to those in the linear element. In contrast, Moss5 and Moss6 only have genes homologous to those in plasmids A and C, and even those show regions of low similarity compared to N. variabilis ATCC 29413 (Fig. 2a). UCD120, UCD121, UCD122 and Moss2 have genes with some similarity to genes in the five plasmids present in ; however, the gene similarities are weak, indicating that these four strains probably lack the plasmids found in (Fig. 2b).

Analyses of the genome sequences for several strains isolated from lichens and feathermoss may provide new information on genes that are specifically associated with symbiosis [21, 34, 35]. These authors took a bioinformatics approach to identify genes and gene families that are associated specifically with strains isolated from lichens and feathermoss. In the case of feathermoss, 74 deduced proteins were identified in all the cyanobacterial plant isolates but not in a related non-symbiotic strain [34]. We determined how many of these ‘symbiotic’ genes were present in the genomes of the 10 strains that we sequenced here. Only sp. 2RC had matches for all 74; however, the N. variabilis-like strains isolated from Azolla had 73 (Fig. 1). The UCD strains had 71 genes (the missing gene in UCD120 is probably a sequencing artefact; the strain has a partial, identical 34 aa sequence on a small contig). In the data provided for the feathermoss symbionts, only one non-symbiotic strain, sp. CALU996, was provided as a negative control [34]. Consistent with the lack of matches for sp. CALU996, we found no matches for one of its closest relatives, CCY9414; however, another strain of CENA596 had 18/74 genes. Surprisingly, another close relative of sp. CALU996, sp. 7107, had matches for 51/74 genes. We found that several other strains that have no known symbiotic association also had many of these genes; the Nostoc linckia strains (73–74), PCC 7122 (46/74), sp. PCC 7524 (45/74) and sp. PCC 7120 (55/74). The relatively large number of these ‘symbiosis’ genes in sp. PCC 7120 may reflect its similarity to N. variabilis ATCC 29413, a strain that we show here is symbiotic. As more symbiotic strains are sequenced, gene profiling techniques will probably be refined and improved, providing new information on the genes and metabolic pathways that are needed for infection and symbiosis.

nif genes and excision elements

Most cyanobacterial nitrogenase gene clusters are highly conserved both in gene sequence and in the organization of the nif genes in a single large cluster under the control of the primary nifB promoter [64-66]. One difference is glbN, cyanoglobin, a gene of unknown function located just upstream of nifH in some strains of , including [67, 68]. The glbN gene was present in UCD120, UCD121 and UCD122, and in 2RC, but was absent in all of the strains that are virtually identical to N. variabilis ATCC 29413.

There is considerable diversity among nitrogen-fixing strains in the excision element that interrupts the nifD gene in many cyanobacterial nif gene clusters (Table 2) [69]. This nifD element, which is excised during heterocyst development, leaving an intact nifD gene, was first identified in PCC 7120 [47, 70–72]. The nifD excision element of N. variabilis ATCC 29413 is very similar to that in PCC 7120 in the middle region of the excision element but, except for xisA, shows no similarity at the end regions of the element (Fig. 3). Among the cyanobacterial strains isolated from Azolla, strains V5, ARAD, N2B, 9RC and PNB have nifD excision elements identical to the nifD element in N. variabilis. In contrast, the genome of strain FSR, which is otherwise virtually identical in sequence to N. variabilis, lacks the nifD excision element completely, which is also shown in the proteome comparison (red box in Fig. 2a). The free-living strain YBS01 also has an excision element virtually identical to those in the N. variabilis ATCC 29413 group (data not shown). The nifD excision elements in the symbiotic strains, Moss5 and Moss6, are similar to each other with respect to gene composition but differ from N. variabilis at the end closest to nifH. The nifD excision element in other strains was more variable in size and composition than in the N. variabilis-like group (Fig. 3). The nifD excision element in UCD121 was most like the element in , with about half of the genes in the nifD element shared between them (Fig. 3). All the genes present in the nifD excision element of UCD121 were also present in UCD120 and UCD122 (data not shown); however, they were not all on a single contig. The similarity of the nifD elements in the strains characterized here correlates with the phylogenetic relationship of the strains. Consistent with the phylogenetic tree (Fig. 1), the nifD element of is most similar to that of Moss2, while the nifD elements of the two Nostoc linckia strains and Moss3 are most similar to each other (data not shown). 2RC is unusual because it has a very large 58 kb nifD excision element; thus, it has many genes that are not similar to those in any of the other nifD elements. Its apparent difference compared to the other nifD elements is due primarily to the presence of an 18.4 kb insertion that spans the middle third of the nifD element in 2RC (Fig. 3). Outside of this insertion, which includes an integrase gene, 2RC and N. linckia z1 share about 75 % of the genes in the nifD element. However, N. linckia z1, and Moss3 and Moss4 have genomic regions outside the nif gene clusters with multiple genes that are similar in gene composition and organization to those in the 18.4 kb insertion in 2RC, including the integrase gene. This implies that this 18.4 kb insertion element is mobile. Among all the nifD excision elements analysed here, the only genes that are shared by all the strains are the excisase gene, xisA, and a gene encoding a small hypothetical protein (Ava_3922, Npun_F0406) shown in purple in Fig. 3. The lack of genes of known function (except for the excisase) in the various excision elements and the absence of the element in FSR indicates that it does not have an essential function in nitrogen fixation because all the strains are able to grow in a medium lacking fixed nitrogen.

Table 2.

Excision elements in cyanobacterial strains

	nifD element	fdxN element	hupL element
Nostoc variabilis ATCC 29413	11 074 bp	No	No
Nostoc PCC 7120	11 289 bp	59 428 bp	9 435 bp
Nostoc punctiforme	23 723 bp	No	No
Nostoc 2RC	58 153 bp	No	5 872 bp
Nostoc YBS01	11 074 bp	No	No
Nostoc Moss5	11 879 bp	38 259 bp	No
Nostoc Moss6	11 879 bp	38 259 bp	No
Nostoc 9RC	11 074 bp	No	No
Nostoc N2B	11 074 bp	No	No
Nostoc ARAD	11 074 bp	No	No
Nostoc FSR	No	No	No
Nostoc V5	11 074 bp	No	No
Nostoc PNB	11 074 bp	No	No
Nostoc UCD120	Yes*	No	No
Nostoc UCD12	24 028 bp	No	No
Nostoc UCD122	Yes*	No	No
Nostoc linckia NIES-25	31 130 bp	No	13 051 bp
Nostoc linckia z1	32 500 bp	No	No
Nostoc Moss2	25 314 bp	No	No
Nostoc Moss3	32 989 bp	No	No
Nostoc Moss4	20 012 bp	No	No

*Located on multiple contigs.

Fig. 3.

Maps of excision elements interrupting the nifD gene (in red) in the symbiotic cyanobacterial strains isolated from plants and similar comparator strains that were not isolated from a plant. The excision element in 2RC has an 18.4 kb insertion in a gene that is a homologue of a gene in N. linckia, indicated by the red arrow.

PCC 7120

YBS01

Moss5

Moss6

UCD120

UCD12

UCD122

Moss2

Moss3

Moss4

Among the cyanobacterial strains characterized here, none have the large excision element found in PCC 7120 in the nif cluster gene, fdxN [73]; however, Moss5 and Moss6 have a smaller element (Table 2) with a similar excisase gene in the same location. PCC 7120 also has an excision element that interrupts the heterocyst-specific uptake hydrogenase, large-subunit gene, hupL [74]. Like the other excision elements, the hupL element is excised during heterocyst development by an excisase, XisC. This hupL element is absent in all the strains shown in Fig. 1 except 2RC and N. linckia NIES-25 (Table 2), where it is found at the same location as in PCC 7120 (Fig. 4). The hupL excision elements all have a similar excisase gene, xisC, and also share one or two genes for hypothetical proteins. It appears that while excision elements have no vital function, they are potentially useful taxonomic markers.

Fig. 4.

Maps of excision elements interrupting the hupL gene (in red) in the symbiotic cyanobacterial strain 2RC, isolated from Azolla, and comparator strains that were not isolated from a plant.

In addition to the nif1 gene cluster that makes the heterocyst-specific nitrogenase, N. variabilis has a large cluster of genes that encode a second Mo-nitrogenase that functions in vegetative cells under anoxic conditions [66, 75, 76]. All the N. variabilis-like strains have nif2 genes (Fig. 5) with over 99 % amino acid identity to the nif2 gene cluster in N. variabilis ATCC 29413. 2RC is the only strain not in the N. variabilis group that has the nif2 cluster, which is identical in gene structure and organization to the nif2 cluster in N. variabilis ATCC 29413 (Fig. 5); however, its nif2 genes show only about 90 % amino acid identity to the N. variabilis homologues. Downstream from the nif2 cluster in N. variabilis is the gene for the nifB2 transcriptional activator cnfR2 [66]; however, between the nif2 genes and cnfR2 in 2RC, there are coxBAC2-type genes, encoding a cytochrome oxidase that is important for nitrogen fixation [77]. These coxBAC2 genes are absent in the nif2 region in strains in the N. variabilis group. In 2RC, the coxBAC2 genes near the nif2 cluster (Fig. 5) share 60–85 % amino acid identity with the other coxBAC2 genes near cnfR1. The coxBAC2 genes close to cnfR1 in 2RC share 77–87 % amino acid identity with the single set of coxBAC2 genes located near cnfR1 in N. variabilis while the coxBAC2 genes close to nif2 in 2RC share 63–85% amino acid identity with the single set of coxBAC2 genes in N. variabilis ATCC 29413. While it seems likely that in 2RC the coxBAC2 genes near the nif2 genes function to support Nif2, another cytochrome oxidase must function for Nif2 in N. variabilis, which has five sets of cox genes, but only one cox2-like set.

Fig. 5.

Maps of nif2-cnfR2 and the cnfR1-cox2 gene regions in the symbiotic cyanobacterial strain 2RC and the comparator strain not isolated from a plant, N. variabilis.

In addition to the second Mo-nitrogenase, N. variabilis has an alternative V-nitrogenase that is made only in the absence of Mo and functions in heterocysts [78, 79]. While the V-nitrogenase is not common in cyanobacteria, the genes for the V-nitrogenase have been found in cyanobacteria in symbiotic association with lichens in boreal and arctic ecosystems where they are expressed in lichens growing in soils deficient in Mo [80, 81]. All the N. variabilis-like strains, as well as 2RC, have vnf genes. Except for 2RC, the vnf genes described here are at least 97 % identical (Table 3). In contrast, the 2RC vnf genes show only about 75–80% nucleotide identity to the vnf genes of N. variabilis, consistent with the fact that 2RC is not closely related to N. variabilis. The two N. linckia strains, which are most similar to 2RC as well as to Moss2, Moss3 and Moss4, lack both the vnf and nif2 genes.

Table 3.

V-nitrogenase and V-transport genes (vupABC) in symbiotic and non-symbiotic strains

	Percentage nucleotide identity based on shared regions of similarity with ATCC 29413
	vnfR1	vnfR2	vnfR3	vnfH	vnfH2	vnfDG	vnfK	vupA	vupB	vupC
Nostoc variabilis ATCC 29413	100	100	−	100	−	100	100	100	100	100
Nostoc sp. YBS01	100	100	−	100	−	100	100	100	100	100
Nostoc 9RC	100	100	−	100	−	100	100	100	100	100
Nostoc N2B	100	100	−	100	−	100	100	100	100	100
Nostoc ARAD	100	100	−	100	−	100	100	100	100	100
Nostoc FSR	100	100	−	100	−	100	100	100	100	100
Nostoc V5	100	100	−	100	−	100	100	100	100	100
Nostoc PNB	100	100	−	100	−	100	100	100	100	100
Nostoc Moss5	97	100	−	99	−	99	99	94	96	95
Nostoc Moss6	97	100	−	99	−	99	99	94	96	95
Nostoc punctiforme	−	75	−	82	−	−	−	−	−	−
Nostoc UCD120	−	82†	−	82	−	−	−	−	−	−
Nostoc UCD121	−	82†	−	82	−	−	−	−	−	−
Nostoc UCD122	−	82†	−	82	−	−	−	−	−	−
Nostoc 2RC	70	73	+‡	85	+	80*	76	82	84	77
Nostoc linckia z1	−	80†	+	84	−	−	−	−	−	−
Nostoc linckia NIES-25	−	75	−	85	−	−	−	−	−	−

*Although vnfDG is a fused gene, vnfD in 2RC has 80 % identity with vnfD in N. variabilis ATCC 29413 but vnfG has only 68 % identity with vnfG in N. variabilis ATCC 29413; a 330 bp region between vnfD and vnfG in 2RC has no matching region in N. variabilis ATCC 29413.

†Truncated vnfR, about 15 % of the length compared to homologues.

‡+ = Present in the strain. Accession numbers for these genes are provided in Table S2.

−

−

sp. YBS01

−

−

−

−

−

−

−

−

−

−

−

−

−

−

Moss5

−

−

Moss6

−

−

−

−

−

−

−

−

−

−

UCD120

−

−

−

−

−

−

−

−

UCD121

−

−

−

−

−

−

−

−

UCD122

−

−

−

−

−

−

−

−

+‡

+

−

+

−

−

−

−

−

−

−

−

−

−

−

−

−

−

has homologues of only two vnf genes, vnfH and vnfR2 (the regulator of vnfH [79]), and they are nearly identical to vnfH and vnfR2 in 2RC; however, unlike 2RC, lacks the genes for the V-nitrogenase alpha and beta subunits (vnfDG and vnfK), as well as the vanadate transport genes, vupABC [82], so it cannot make a V-nitrogenase (Table 3). The presence of the entire vnf gene cluster in 2RC and the similarity of vnfR2 and vnfH to homologues in suggests that an ancestor of might have had the entire vnf cluster. UCD120, UCD121 and UCD122 are most similar to in terms of having only vnfH-like and vnfR2-like genes (Table 3); however, the vnfR gene in the three UCD strains is truncated to only about 15 % of the size of the typical vnfR gene.

The true vnfH genes in the N. variabilis-like strains that have complete vnf gene clusters are distantly related to the vnfH copies present in the strains that lack major vnf genes (Fig. 6). 2RC has all the vnf structural genes, although their organization is different from the cluster in the N. variabilis-like strains. The vnfH gene of 2RC clusters with the vnfH genes of the strains that lack the rest of the vnf genes. In contrast to the N. variabilis-like strains, 2RC has a second copy of vnfH located near the V-nitrogenase structural gene vnfDG but divergently transcribed (Fig. 6). Although we have named it vnfH2 based on its location, its function as part of the V-nitrogenase in this strain will need to be confirmed experimentally. This second vnfH gene in 2RC clusters with a group of nifH5 genes of unknown function that, in these other strains, is found downstream of the cydAB genes (Fig. 6). This group of nifH genes is phylogenetically distinct from all the other nifH copies. The gene most closely related to vnfH2 in 2RC is nifH4 in an uncharacterized strain, Microchaete diplosiphon NIES-3275; however, that nifH4 is located near the cydAB genes like the nifH5 copies of N. variabilis (Fig. 6).

Fig. 6.

Phylogenetic tree of nifH/vnfH genes for the symbiotic cyanobacterial strains isolated from plants and similar comparator strains that were not isolated from a plant. The N. variabilis group represents N. variabilis ATCC 29413 and the other nearly identical strains, V5, ARAD, FSR, N2B, PNB, 9RC and YBS01, while UCD strains represent UCD120, UCD121 and UCD122. Accession numbers for nifH/vnfH genes are provided in Table S2. Scale represents the number of substitutions per site.

The vnf genes in N. variabilis ATCC 29413 are repressed by VnfR1 or VnfR2 when Mo is present. In the absence of Mo, the repressor cannot bind and the vnf genes are expressed [79]. , the two N. linckia strains and Moss3, all lacking the structural genes for the V-nitrogenase, vnfDGK, have a vnfH homologue with a single copy of vnfR upstream of vnfH (Fig. 6). This vnfR gene is homologous to the vnfR2 gene in N. variabilis ATCC 29413, which is located just upstream of the functional vnfH gene; however, there is no information on the expression or function of the vnfH homologue in . In addition to the vnfR1 and vnfR2 genes present in the N. variabilis-like group, 2RC and Microchaete diplosiphon NIES-3275 have an additional copy, vnfR3, located just upstream of the vanadate transport genes, vupABC [82] (Fig. 7). This vnfR3 gene clusters with the vnfR1 genes that are present in all the strains that have a functional V-nitrogenase. It is interesting that while there are redundant copies of vnfR, none is located close to the V-nitrogenase genes, vnfDG and vnfK, which are regulated by VnfR. Similarly, 2RC has a copy of vnfH upstream of vnfDG and vnfK and has three copies of vnfR, but none is near the vnfDGK structural genes, suggesting that the regulatory genes have always been distant from the major vnf structural genes.

Fig. 7.

Phylogenetic tree of vnfR genes for the cyanobacterial strains that have the structural V-nitrogenase genes (black) versus comparator strains that lack these genes (blue). The N. variabilis group represents N. variabilis ATCC 29413 and the other nearly identical strains, V5, ARAD, FSR, N2B, PNB, 9RC and YBS01. Accession numbers for vnfR genes are provided in Table S2. Scale represents the number of substitutions per site.

Many cyanobacteria, even those with only one nif system, have additional nifH copies (Fig. 6) that have not been studied. N. variabilis has five copies of nifH genes (nifH5 is mentioned above) of which only three, nifH1, nifH2 and vnfH, function as part of a complete nitrogenase [83]. A phylogeny of all the nifH genes in the strains described in this study shows that the nifH copies that are most closely related to each other also share similar genes surrounding them, even (in the case of nifH5) when these nearby genes have no similarity to known nitrogenase genes. The nifH1 copies of all the N. variabilis-like strains, as well 2RC and the two N. linckia strains, are most closely related to each other. The nifH1 genes of and its close relatives UCD120, UCD121 and UCD122 are closely related to each other and to Moss2 and Moss4, but less closely related to the N. variabilis group (Fig. 6).

The nifH copy most closely related to nifH1 is called nifH2 in strains that lack the nif2 system and nifH4 in strains that have the nif2 system (Fig. 6). Although the function of nifH2 (nifH4) is unknown, the gene, like nifH1, is expressed exclusively in heterocysts [84]. Like the nifH1 genes, the nifH4 genes of the N. variabilis-like strains (nifH2 in strains lacking a nif2 system) form a cluster that is related to their homologues in 2RC, the two N. linckia strains and Moss3. The nifH1 and nifH4 groups are related to each other but are distinct from the vnfH group (Fig. 6). Only vnfH from N. variabilis ATCC 29413 has been shown to function as part of the V-nitrogenase; however, its similarity in sequence and gene context to vnfH genes in other cyanobacteria suggests that either the other strains lost the rest of the vnf genes or possibly that this nifH copy gained its VnfH function later in the evolution of the vnf genes, since the nifH1 gene of N. variabilis ATCC 29413 functions well in place of vnfH [85].

Sugar transport genes

Cyanobacteria in symbiotic associations obtain their carbon primarily from the plant in the form of sugars; hence, they must be able to take up sugar from the plant [39, 44]. Both N. variabilis and , but not PCC 7120, are capable of using fructose to grow heterotrophically in the dark [27, 86]. If the fructose transport genes of N. variabilis are expressed in PCC 7120, the latter strain gains the ability to grow on fructose heterotrophically in the dark, indicating that it is the lack of a fructose transport system that limits the use of fructose in PCC 7120 [87]. has both fructose and glucose transporters [86]; however, most other filamentous cyanobacteria have only fructose transport genes. In N. variabilis ATCC 29413 and the nearly identical strains V5, ARAD, FSR, N2B, PNB, 9RC and YBS01 the frtA-frtB-frtC genes comprise a typical ABC-transporter [87]; however, , UCD120, UCD121 and UCD122, Moss2, Moss3 and Moss4, and 2RC all have two copies of frtA (Fig. 8). The frtA1 genes are most closely related to each other as are the frtA2 genes, but the single frtA gene in N. variabilis clusters with the frtA2 genes (Fig. S1). In , UCD120, UCD121 and UCD122, and Moss2, the glucose transporter, glcP, is just downstream from the fructose transport genes (Fig. 8) but is elsewhere in strains NIES-25, Moss3 and 2RC. The glucose transporter, glcP, is absent in the symbiont Moss4 and the non-symbiont, N. linckia z1. Although grows poorly using glucose as a carbon source, loss of the glucose transporter in prevented infection of the hornwort Anthoceros, but the absence of the fructose transporter (in a strain overexpressing the glucose transporter) did not affect infection [86]. All of the strains studied here have a gene, frtR (hrmR), that makes a LacI-like repressor that regulates the expression of the frt operon [87]. While all the strains isolated from plants have sugar transport genes, other non-symbiotic strains, notably N. variabilis ATCC 29413, PCC 7107 and the two N. linckia strains, have sugar transport genes that are very similar to those in the symbionts.

Fig. 8.

Maps of fructose transport (frtABC) and glucose transport (glcP) gene regions, including genes, hrm, implicated in hormogonia regulation.

The hrm genes are involved in the regulation of hormogonia, which are plant-responsive, motile cyanobacterial filaments that initiate plant infection [20, 33]. HrmR (FrtR) regulates its own transcription and that of hrmE while hrmA and hrmU are important for the regulation of the level of hormogonia production [45, 88]. The hrm genes were absent in N. variabilis and the nearly identical strains V5, ARAD, FSR, N2B, PNB, 9RC and YBS01 and were also absent in Moss5 and Moss6, but were present in , the two N. linckia strains, UCD120, UCD121 and UCD122, 2RC, and Moss 2, Moss3 and Moss4. Although the organization of the sugar transport and hrm genes is very similar among the strains that have these genes [35], in UCD120, UCD121 and UCD122 the hrmA and hrmU genes are some distance from the rest of the hrm cluster (on a different contig) (Fig. 8).

Chemotaxis

has multiple clusters of genes with similarity to chemotaxis (che) genes from other bacteria [89] but few have been well characterized. A deletion of the entire locus of cheR-like genes (NpR0244–NpR0250) has no effect on hormogonia formation or phototaxis and these genes do not respond to the addition of a hormogonium-inducing factor [90]; however, it has also been reported that a strain with a mutation of one of the genes in this cluster, NpR0248, is impaired in hormogonia formation and motility and is unable to infect the symbiotic liverwort host, Blasia pusilla [37]. A che region with a high degree of sequence similarity to this large cluster is not present in any of the N. variabilis-like strains, the UCD strains, the two strains of N. linckia nor any of the Moss strains. However, several genes in the cluster, with about 50 % amino acid identity based on shared regions of similarity to the homologues, are present in many of the strains. The best-characterized chemotaxis genes in cyanobacteria are the hmpBCDE genes in [91-93]. Mutants in hmpB, hmpC, hmpD or hmpE prevent the formation of hormogonia and, therefore, are not motile and fail to establish a symbiotic relationship with the hornwort Anthoceros punctatus [92]. This gene cluster is well conserved in filamentous cyanobacteria, including the N. variabilis-like strains, UCD120, UCD121 and UCD122, 2RC, the two strains of N. linckia, and all the Moss strains (Fig. S2). The hmpE gene is highly variable in size in these strains due to repeat regions of variable lengths, and hmpD in Moss4 is split into two genes. The nucleotide identity for these genes between pairs of strains was similar to the ANI between pairs of strains. Although chemotaxis is surely important in symbiosis, we did not identify any che-like genes that distinguished symbiotic strains from many other non-symbiotic cyanobacteria.

Pili have been studied in unicellular cyanobacteria where they mediate twitching motility [94] and in where a type IV pilus-like system powers the gliding motility and polysaccharide secretion of hormogonia [93]. Mutants in NpR0117 (pilT-like) and NpR2800 (pilD-like) of have very low rates of infection of Blasia while a mutant in NpF0069 (pulG, or pilA-like) has somewhat reduced levels of infection of Blasia [95], suggesting that pili are important for the motile hormogonia that lead to infection of plants. Another putative pilA gene (NpF0676), which did not yield segregated mutants [95], is localized to rings at hormogonia cell junctions [92]. All of these putative pilA, pilT and pilD genes have homologues in all the N. variabilis-like strains, YBS01, strains UCD120, UCD121 and UCD122, 2RC, Moss2–6, and the two strains of N. linckia (Figs S3–S6). The nucleotide identity between pairs of strains for these genes was similar to the ANI between pairs of strains.

Individual plant strains associate with a variety of strains, and, similarly, individual strains associate with a variety of plant strains [10, 96]. However, there is no evidence of successful reconstitution of the symbiosis of any cyanobacterial strain with the water fern Azolla [97]; therefore, we attempted to infect the more tractable liverwort, Blasia pusilla, with the newly sequenced cyanobacterial strains. Cyanobacteria infect structures known as auricles on the surface of the Blasia plant thallus. Within the auricle, the endophytic cyanobacteria grow and differentiate a high percentage of heterocysts, filling the cavity. Because the auricles on the thallus are colourless, it is easy to distinguish the dense green infected auricles from uninfected auricles and from cyanobacterial filaments that are loosely associated with the plant surface (Fig. 9a). We infected axenic Blasia plants on agar plates with axenic liquid cultures of , N. variabilis ATCC 29413 and the newly sequenced strains. Despite the near identities of the N. variabilis-like strains V5, ARAD, FSR, N2B, PNB and 9RC, only N. variabilis ATCC 29413, PNB and V5 infected Blasia, and the infection rate for all was poor with few infected auricles (determined semi-quantitatively to be about 5%). Among the strains, and 2RC infected Blasia easily with about 70–75% infected auricles (Fig. 9a). 2RC was unusual among all these strains in that it produced abundant, highly motile hormogonia, causing the strain to swarm over the entire surface of an agar plate. It seems likely that its proficiency in producing motile hormogonia aids in its ability to infect Blasia. Among the three nearly identical strains, UCD120, UCD121 and UCD122, only UCD122 infected Blasia and the frequency was low, similar to N. variabilis ATCC 29413 (Fig. 9a). Although these three strains are nearly identical genetically, their morphology in liquid culture was different. UCD122 filaments aggregated into clumps, while liquid cultures of the other two strains had a smooth and homogeneous appearance (Fig. 9b). Thus, for UCD122, aggregation and hormogonia formation were correlated with its ability to infect Blasia. N. punctiforme normally grows in clumps [37, 90, 98]; however, there is an uncharacterized smooth variant that is easier to work with but lacks hormogonia [92]. An hmp mutant that could not differentiate hormogonia grew as dispersed suspensions [92], while a pks2 mutant that produced many highly motile hormogonia showed increased aggregation [99]. The smooth variant of infects Blasia very poorly [37], similar to N. variabilis, which also grows as a homogeneous suspension.

Fig. 9.

Blasia infection and UCD strain growth. (a) Light micrographs of uninfected Blasia auricles and Blasia auricles infected by UCD122, , N. variabilis ATCC 29413 and 2RC. Black arrows indicate green infected auricles (about 120 μm in diameter), packed with cyanobacteria, while red arrows indicate nearly colourless uninfected auricles. Infected auricles were determined as described in the methods. (b) Liquid growth characteristics of UCD strains.

Taxonomy and nomenclature

‘ ATCC 29413’ is now called ‘Nostoc variabilis ATCC 29413’, a change that is supported by both the characteristics of the strain and its phylogeny. Like other symbiotic strains, N. variabilis ATCC 29413 and its very close relatives that were isolated from Azolla produce hormogonia [100], which are required for the infection of plants. Also, the symbiotic strains can transport and use sugars, notably fructose, consistent with their symbiotic lifestyle. While there are strains that are now considered to be Trichormus variabilis [101], N. variabilis ATCC 29413 is not closely related to them (see Discussion). A phylogenetic tree based on 500 shared genes as well as the whole genome ANI values (Fig. 10) indicates that N. variabilis ATCC 29413 is closely related to other strains and is more distantly related to the only two Trichormus strains that have a sequenced genome [102, 103]. In fact, the two sequenced Trichormus strains cluster together with PCC 7122 and sp. PCC 7108 (Fig. 10), not with N. variabilis ATCC 29413.

Fig. 10.

Phylogenetic distance tree based on genomes for representative cyanobacterial strains. ANI values are included for comparison between strains indicated by asterisks. Accession numbers for these strains are provided in Table S1. Scale represents the number of substitution per site.

Discussion

Genome characterization of newly sequenced strains isolated from Azolla

Although the non-culturable strain N. azollae 0708 has been identified as the primary symbiont in Azolla [15], other cyanobacteria have also been found associated with the plant [9-13]. We sequenced the genomes of 10 cyanobacteria that were isolated from Azolla strains by different laboratories in different parts of the world (Table 1). Most were nearly identical to N. variabilis, which was not itself isolated from Azolla, and were very similar to Moss5 and Moss6 strains that were isolated from feathermoss [21]. Strains ARAD from Azolla filiculoides in Israel and V5 from Azolla pinnata in Russia were nearly indistinguishable from strains 9RC and N2B isolated from Azolla caroliniana in the USA. However, there were some differences, including the absence of the 37 kb linear plasmid in strains PNB and FSR and the lack of the nifD 11 kb excision element in FSR. In contrast, UCD120, UCD121 and UCD122 were all isolated from A. caroliniana originally collected in Ohio [8, 104]. Unlike the other strains that we sequenced, the UCD strains were first collected from Azolla and then isolated after selecting for strains that were able to form an association with Anthoceros, prior to growing them as free-living cultures [8]. These strains are nearly identical genetically but are morphologically different, especially UCD122, which aggregates into clumps in liquid culture.

One strain isolated from Azolla, 2RC, was different from the other characterized strains. 2RC is brownish in colour, aggregates into clumps in liquid culture and spreads rapidly over the surface of agar plates via its very motile hormogonia. Its genome is most similar to two N. linckia strains, which have not been reported to be symbiotic, and to Moss3, a symbiont that was isolated from feathermoss [21]. However, 2RC differs significantly from these other strains because it has two additional sets of nitrogenase genes, also present in all of the N. variabilis-like strains and in Moss5 and Moss6. These are the nif2 genes, encoding an Mo-nitrogenase that functions in vegetative cells under anoxic conditions [66, 83] and the vnf genes that make the alternative V-nitrogenase [78, 79], which are absent in all the strains closely related to 2RC, except for the vnfH-like and vnfR-like genes. The presence of these two conserved vnf-like genes suggests that these strains may have once had the full cluster of vnf genes.

Characteristics of many symbiotic strains shared by the newly sequenced strains

Culturable symbiotic strains isolated from Azolla and other plants share several physiological characteristics. All form heterocysts and fix nitrogen, which is supplied to the plant [41, 43, 105, 106], a characteristic shared with N. azollae 0708, the non-culturable Azolla symbiont [15, 107, 108]. All culturable symbionts can transport sugars, typically fructose, although many of the symbiotic strains described here have a gene for glucose transport, a characteristic shared with several non-symbiotic strains including unicellular cyanobacteria [109, 110]. In contrast, orthologues of the glucose and fructose transport genes in and N. variabilis [86, 87] are absent in N. azollae 0708 [15]; however, other transporters may mediate sugar transport in N. azollae 0708 [111].

Nitrogenase genes are common in cyanobacteria, especially the nif genes that encode the heterocyst-specific Mo-nitrogenase, and all the strains studied here had these genes. In contrast, complete sets of vnf genes are comparatively rare but are present in the genomes of about a dozen strains of the genera , , Chlorogleopsis, and /Aulosira (accessed via the JGI genomes database). The vnf genes have also been found in a strain associated with the lichen Peltigera [32, 80] and in three strains isolated from hornworts Phaeoceros carolinanus and Leiosporoceros dussii, and Blasia pusilla, some on plasmids [112]. The V-nitrogenase may confer a selective advantage to the plants in environments where Mo is limiting and they may be laterally transferred [112], perhaps even between strains in the same plant.

Several strains that have the vnf genes, including Aulosira laxa NIES-50, Nostoc carneum NIES-2107, Calothrix brevissima NIES-22 and PCC 7101, also have the nif2 gene cluster. There appear to be no strains that have the nif2 genes but not the vnf genes, suggesting that the nif2 genes, which form a single tight cluster, may have been lost from strains that once had both. The significance of the association of the vnf and nif2 gene clusters in these strains as well as in symbiotic strains closely related to N. variabilis is unknown.

Infection of Blasia by the strains isolated from Azolla

The culturable, free-living strains that infect plants probably do so through the differentiation of motile, non-growing hormogonia [106, 113]. In the context of symbiosis, hormogonia have been studied only in [36, 44], although they are also made in N. variabilis ATCC 29413 in cells subjected to starvation for fixed nitrogen [100]. Among the newly sequenced strains described here, hormogonia were abundant and highly motile only in 2RC. Although several of the strains were capable of infecting the liverwort Blasia, only , UCD122 and 2RC did so readily. These strains shared the characteristic of clumpy growth, probably a result of hormogonia formation [92, 99].

All the strains had similar chemotaxis genes and pili-related genes that are likely to be involved in infection [37, 92, 93, 114]; however, many cyanobacteria that have no known association with symbiosis also have homologes of these genes. Analysis of the genes and metabolic pathways that correlate with symbiotic cyanobacteria suggests that many of these pathways function in a coordinated manner to allow plant detection, infections and maintenance of the symbiotic state [21, 34]. However, we found that even genes identified from several cyanobacteria as associated with symbiosis in feathermoss [34] have homologues in strains not known to be symbiotic (Fig. 1, Table S3). Among the well-characterized genes analysed here, there appear to be no genes that clearly distinguish symbiotic strains from many other related, non-symbiotic strains of the genus .

N. variabilis ATCC 29413 has been reported not to infect Anthoceros [106] but we found that it infected Blasia, as do its close relatives Moss5 and Moss6 in a clade called Extra II [34]. However, Moss5 and Moss6 do not infect Gunnera [34], and there is one report that N. variabilis N2B (also known as Anabaena azollae N1) also does not infect Gunnera [96]. Although UCD122 infected Blasia (Fig. 9), as does its close relative [34], the nearly genetically identical strain UCD120 (ANI between UCD120 and UCD122 is 99.9%) did not. While we were not successful in infecting Blasia with UCD120, it has been shown to infect Anthoceros [115]. Therefore, it seems likely that in different laboratories, variations in the physiological conditions of the plants and the cyanobacteria lead to differences in success in reconstituting a symbiosis. Further, natural ecosystems with their plant-associated biota are likely to favour symbiosis in ways that cannot be reproduced in the laboratory.

Taxonomy and naming of () variabilis ATCC 29413

N. variabilis and its close relatives described here share an ANI of 98 % with feathermoss isolates Moss5 and Moss6, which form symbiotic associations with Pleurozium schreberi and Blasia pusilla [34]. The latter strains are virtually identical to each other and somewhat distant from the other feathermoss isolates, Moss2, Moss3 and Moss 4. N. variabilis, Moss5 and Moss6, like N. variabilis ATCC 29413, belong to a clade called Extra II, indicating their symbiosis is extracellular [34]. However, while infection of Blasia is extracellular but endophytic, infection of feathermoss is extracellular and epiphytic. Among characterized strains, only those in the clade called Extra/Intra are capable of intracellular symbiosis with Gunnera and extracellular symbiosis with bryophytes The Extra II clade, including Moss5 and Moss6, which diverged about 1500 Ma appears to have gained its symbiotic capacity by horizontal gene transfer with genes coming almost equally from the strains in the Extra/Intra clade (which includes and Moss 2), from N. azollae, and from the Extra I clade (which includes Moss3 and Moss4) [34].

Komárek and Anagnostidis created the genus name Trichormus and placed a strain, SAG 1403 4b (ATCC 29211; PCC 6309), in that genus [101]. In just the last few years NCBI has renamed any strain as Trichormus variabilis, including ATCC 29413. However, based on multiple lines of evidence, we have called ATCC 29413 Nostoc variabilis ATCC 29413 instead. As early as 1979, strains were defined as obligate photoautotrophs [61]; however, by then it was known that ATCC 29413 is capable of growing in the dark using fructose [27]. Bergey’s Manual states that the ‘Wolk strain’ (C.P. Wolk) of (i.e. ATCC 29413) was misidentified, commenting that strains of that are closely related to sp. PCC 7108 (a photoautotroph) do not include the strain ATCC 29413. Also, Bergey’s Manual states that ATCC 29413 differs from the true strains in the morphology of the trichomes, the akinetes and the hormogonia [116].

This distinction among Trichormus (Anabaena) variabilis strains is supported by our data that demonstrate the Nostoc-like characteristics of N. variabilis ATCC 29413 and by the phylogenetic trees provided here and in the online cyanobacterial taxonomy database (http://cyanophylogeny.scienceontheweb.net/). Earlier phylogenetic trees, based on either 16S rRNA or rpoB gene sequences, showed that T. variabilis HINDAK 2001/4 and T. variabilis GREIFSWALD/92 are most closely related to sp. PCC 7108, but are distant from sp. PCC 7120, the closest relative to N. variabilis ATCC 29413 shown in those trees [117]. Similarly, phylogenetic trees published for T. variabilis SAG 1403 4b (ATCC 29211; PCC 6309) [102] and for Trichormus sp. NMC-1 [103] show that these Trichormus strains are not closely related to N. variabilis ATCC 29413. N. azollae 0708, the non-culturable Azolla symbiont, has also recently been renamed Trichormus azollae by NCBI. Phylogenetic trees based on whole-genome data indicate that T. azollae clusters with T. variabilis SAG 1403 4b and Trichormus sp. NMC-1, along with sp. PCC 7108 and PCC 7122 [102, 103]. With more sequenced genomes the taxonomic relationships among the //Trichormus genera may become clearer.

Conclusions

Free-living strains that were isolated from Azolla from many parts of the world share many genes with other characterized symbiotic cyanobacteria cultured from a variety of plants. However, these genes are also found in closely related cyanobacteria that have no known associations with plants, such as two strains of Nostoc linckia. The accepted method for demonstrating symbiosis is the ability of a purified axenic cyanobacterial strain to infect an axenic plant. In this study, we found that some axenic strains isolated from Azolla could readily infect axenic Blasia, but others did so poorly or not at all, despite having ‘symbiosis’ genes and, in some cases, nearly identical genomes with strains that could infect. This may not be surprising for two reasons. First, mutations in genes or regulatory regions that are critical for producing motile hormogonia and for communicating with the plant would be expected to drastically reduce infection. These would be difficult to identify because most such genes are unknown or poorly characterized. Strains maintained in a sterile laboratory setting are very likely to lose characteristics that are important for infection and symbiosis. Second, normal infection and establishment of symbiosis occur in an ecological environment rich in bacteria, fungi, insects, and plants that probably play important roles in symbiotic infections. Thus, we speculate that in the natural environment other organisms aid cyanobacteria in establishing associations with plants that will never be reproduced in a laboratory setting. Future research on symbiosis should consider and incorporate the natural plant biome as a possible contributor to successful symbiosis.

Click here for additional data file.