Dinitrogen Fixation Is Restricted to the Terminal Heterocysts in the Invasive Cyanobacterium Cylindrospermopsis raciborskii CS-505

The toxin producing nitrogen-fixing heterocystous freshwater cyanobacterium Cylindrospermopsis raciborskii recently radiated from its endemic tropical environment into sub-tropical and temperate regions, a radiation likely to be favored by its ability to fix dinitrogen (diazotrophy). Although most heterocystous cyanobacteria differentiate regularly spaced intercalary heterocysts along their trichomes when combined nitrogen sources are depleted, C. raciborskii differentiates only two terminal heterocysts (one at each trichome end) that can reach >100 vegetative cells each. Here we investigated whether these terminal heterocysts are the exclusive sites for dinitrogen fixation in C. raciborskii. The highest nitrogenase activity and NifH biosynthesis (western-blot) were restricted to the light phase of a 12/12 light/dark cycle. Separation of heterocysts and vegetative cells (sonication and two-phase aqueous polymer partitioning) demonstrated that the terminal heterocysts are the sole sites for nifH expression (RT-PCR) and NifH biosynthesis. The latter finding was verified by the exclusive localization of nitrogenase in the terminal heterocysts of intact trichomes (immunogold-transmission electron microscopy and in situ immunofluorescence-light microscopy). These results suggest that the terminal heterocysts provide the combined nitrogen required by the often long trichomes (>100 vegetative cells). Our data also suggests that the terminal-heterocyst phenotype in C. raciborskii may be explained by the lack of a patL ortholog. These data help identify mechanisms by which C. raciborskii and other terminal heterocyst-forming cyanobacteria successfully inhabit environments depleted in combined nitrogen.


Introduction
Cyanobacteria are photosynthetic organisms of great ecological importance due to their carbon fixation (photosynthesis) which is combined in several genera with N 2 fixation (diazotrophy). Because the enzyme responsible for N 2 fixation (nitrogenase) is irreversibly inactivated by O 2 , most diazotrophic filamentous cyanobacteria can only fix N 2 under micro-oxic conditions (e.g. by fixing N 2 in the dark or reducing their photosynthetic rates when fixing N 2 ) [1]. Some filamentous cyanobacteria have solved this dilemma by differentiating a proportion of their vegetative cells into heterocysts, which are non-photosynthetic N 2 -fixing cells [2]. When combined nitrogen becomes depleted most heterocystous cyanobacteria differentiate 5-10% of their vegetative cells into heterocysts, which are individually spread at regular intervals along the trichomes (intercalary heterocysts) [3]. This regularity means that each heterocyst provides the nitrogen required by the ,8-12 neighboring, non-N 2 -fixing vegetative cells [4]. Numerous studies detail the differentiation process of intercalary heterocysts [2] and the N 2 fixation strategies operative in these cyanobacteria [4,5,6]. These studies have primarily focused on the model organism Anabaena sp. strain PCC 7120 (hereafter Anabaena PCC7120) and Anabaena variabilis strain ATCC 29413 (hereafter A. variabilis ATCC29413). However, some cyanobacteria differentiate heterocysts exclusively at the end of their trichomes (terminal heterocysts), such as cyanobacteria of the genera Cylindrospermum and Cylindrospermopsis. Few studies have addressed the physiology or mechanisms related to this terminal heterocyst phenotype [7,8,9]. Although their trichomes may be composed of more than 100 cells, these cyanobacteria differentiate at the most two heterocysts [7]. It has been shown that DpatA (all0521) mutant of Anabaena PCC7120 only differentiates terminal heterocysts [10]. This suggests that patA is crucial for the differentiation of intercalary heterocysts and that the differentiation of terminal heterocysts is regulated differently. However, this mutant grows poorly when lacking combined nitrogen, suggesting that the terminal heterocysts, in spite of being able to fix nitrogen, cannot satisfy the levels of nitrogen demanded by the whole trichome [11].
In contrast to Anabaena PCC7120, which has only one type of nitrogenase, the intercalary heterocystous cyanobacterium A. variabilis ATCC29413 expresses one type of nitrogenase in the heterocysts under both aerobic and anaerobic conditions (encoded by the nif1 gene cluster), while a second nitrogenase is expressed in the vegetative cells under anaerobic conditions only (encoded by the nif2 gene cluster) [6]. Likewise, numerous filamentous nonheterocystous cyanobacteria are able to express the nif genes in the vegetative cells (e.g. under darkness; [1]). Hence, we hypothesized that the terminal heterocysts of Cylindrospermopsis may rely on both a nif1-like nitrogenase (in the heterocysts) and the co-support of a nif2-like nitrogenase expressed during the dark phase of the cycle (,low oxygen) in their vegetative cells. However, the genome sequence of Cylindrospermopsis raciborskii CS-505 (hereafter C. raciborskii), the first sequenced cyanobacterium with terminal heterocysts (Fig. 1A), revealed the presence of a single nif1-like gene cluster [12]. Thus, the presence of an additional nif-cluster was excluded, which suggests that other mechanisms must be operative. Interestingly, C. raciborskii shows a highly successful ecological competence and has recently been observed to radiate from its endemic subtropical environment into limnic water bodies in temperate latitudes worldwide [13,14]. The ecological success of this cyanobacterium may be attributed not only to its facultative diazotrophic capacity [15], but also to its capacity to differentiate spore like cells (akinetes) under adverse conditions [16,17] and the existence of multiple toxin producing phenotypes within the Cylindrospermopsis genus. Some strains produce the hepatotoxin cylindrospermopsin [16], and others produce the neurotoxin saxitoxin [17], while some strains lack toxin production entirely. The toxin production, besides representing threats to human health and livestock, may render additional competitive advantages in nature [18].
Here we have analyzed the strategy employed by C. raciborskii to survive combined nitrogen depletion while differentiating only terminal heterocysts. Our analysis focused on determining nitrogenase expression in time and space along the trichomes of C. raciborskii, and discovering the genomic basis determining the terminal heterocyst phenotype (i.e. preventing the formation of intercalary heterocysts in this cyanobacterium).

Diazotrophic Growth, nifH Levels and Nitrogenase Activity
Microscopic examinations verified that C. raciborskii trichomes differentiate exclusively terminal heterocysts when grown lacking Figure 1. The cyanobacterium C. raciborskii grown under diazotrophic conditions. (A) Light micrograph of C. raciborskii trichomes, magnified 40x. Note the terminal heterocysts at the end of the trichomes (white arrows); one heterocyst is magnified. (B) Percentage of trichomes with 10 to .100 vegetative cells are grouped according to their trichome length (no. of vegetative cells per trichome). The bars give the average numbers in 400 trichomes from 4 independent cultures. Standard deviation is given. The dashed line denotes the point where all trichomes have developed two end heterocysts. (C) The growth of C. raciborskii cultures, grown with 2 mM of NO 3 2 to an OD 750 of 0.2, washed and transferred to new medium under the denoted light regime. Each point corresponds to the mean of 4 biological replicates. Standard deviation is given. (D) Western-blot analysis of NifH expression during a 12L/12D cycle (above). The NifH western-blot membrane was stripped and re-labelled with an antibody against the 30S ribosomal protein S1, revealing protein levels loaded in each well (below). The closest molecular marker band is indicated, and black and white bars represent the light and dark phase of the cycle. (E) Nitrogenase activity was analyzed by ARA for C. raciborskii under diazotrophic growth and continuous light (MLA 0 Cont light), diazotrophic growth and 12L/12D cycle (MLA 0 12/12), and grown in MLA N under 12L/12D cycle (MLA N 12/12). Oxygen concentration was measured for MLA 0 (Fig. 1A). In diazotrophic cultures of C. raciborskii (OD 750 of 0.55-0.65), terminal heterocysts may be separated by a few or even up to .100 vegetative cells (Fig. 1B). One heterocyst always starts to differentiate at one end of the trichome and when the trichome has elongated to approx. 30 cells, another heterocyst starts to differentiate at the opposite end (Fig. 1B). In these diazotrophic cultures (OD 750 of 0.55-0.65), 81% (SD 6 2.97) of the trichomes developed two end-heterocysts, and each heterocyst contained an average of 22.7 (SD 63.91) vegetative cells. However, up to 40.1% (SD 6 10.07) of the C. raciborskii trichomes were composed of considerably more than 50 vegetative cells each (Fig. 1B). Notably, while the terminal heterocyst phenotype persisted, the number of vegetative cells per trichome increased as the cultures approached the stationary phase (data not shown). This is in stark contrast to the situation in Anabaena PCC7120, in which there are generally 7-15 vegetative cells per heterocyst [19].
Several approaches were used to determine whether the terminal heterocysts of C. raciborskii are the sole sites for N 2 fixation, and whether N 2 fixation may take place in the vegetative cells under any conditions (e.g. in darkness). First, we compared diazotrophic growth under continuous light and 12/12 light/dark (12L/12D) cycles, respectively (Fig. 1C). The cultures were grown in 2 mM NO 3 2 as an N-source (MLA medium) under continuous light or under 12L/12D cycles until they reached an OD 750 of ,0.2. Then they were washed and half of the cultures were maintained in the MLA medium, while the other half were resuspended in medium lacking combined nitrogen sources (MLA 0 ). All cultures were incubated under their corresponding light regime until they reached the stationary phase (Fig. 1C). The growth of C. raciborskii was stimulated under all conditions, particularly under continuous light, both in the presence and absence of a combined nitrogen source. However, diazotrophic growth was considerably lower under the 12L/12D regime, clearly indicating that darkness neither supports nor enhances diazotrophic growth in C. raciborskii.
To determine when nitrogenase is synthesized in diazotrophic cultures grown under 12L/12D cycles, protein levels were determined by western-blots based on an antibody raised against NifH (Fe-protein; approx. 34 kDa). As seen in Figure 1D (upper image), higher NifH levels were synthesized in the light compared to the dark phase of the cycle. Similar NifH levels were observed in three separate diel studies (data not shown). The same western blot membranes were then stripped and re-labelled with an antibody raised against the 30S riboprotein S1. As seen in Figure 1D (lower image), this loading control shows that similar protein levels were present at each time point, supporting the NifH protein variations observed.
Next, the nitrogenase activity profile of C. raciborskii was analyzed by the acetylene reduction assay (ARA; Fig. 1E) under three growth conditions: i) diazotrophic growth under continuous light (MLA 0 cont. light); ii) diazotrophic growth under a 12L/12D cycle (MLA 0 12/12); and for comparison, iii) non-diazotrophic growth conditions under 12L/12D cycle (MLA N 12/12). Additionally, the oxygen levels of the MLA 0 12/12 cultures were monitored in parallel. As expected, no nitrogenase activity was observed in the non-diazotrophic cultures. In the MLA 0 12/12 cultures the nitrogenase activity showed a distinct light dependent pattern, being close to zero in the dark phase while it peaked 3 h into the light phase and then decreased when approaching the new dark phase of the cycle. Notably, the oxygen concentration of the cultures also increased and remained high during the light period, which may explain the decrease in nitrogenase activity in the MLA 0 12/12 cultures beyond the peak at 3 hours (Fig. 1E). Under continuous light, the nitrogenase activity fluctuated but was pronounced at all time points analyzed (Fig. 1E), stressing the light dependence of the nitrogenase activity. These results demonstrate that the nitrogenase activity of C. raciborskii, as in other examined heterocystous cyanobacteria, is dependent on energy generated during the light phase of the cycle.

NifH Levels and nifH Expression in Heterocysts and Vegetative Cells
To examine to what extent the vegetative cells are able to express and synthesize nitrogenase, vegetative cells and heterocysts of a C. raciborskii culture grown under continuous light were analyzed separately. For this purpose, a two-phase aqueous polymer partition of the two cell types was performed. When analyzed by light microscopy (LM), the two cell fractions showed an enrichment of ,95% per cell type ( Table 1). The two cell fractions were then subjected to both RT-PCR and western-blot analyses. In order to validate the results from the two cell fractions, both heterocyst and vegetative cell markers were included. The transcripts of devA and hglD [20,21] were used as markers for heterocysts. As seen in Figure 2A, these transcripts were detected only in the heterocyst enriched fractions. The transcripts of rbcS, encoding the small subunit of the RuBisCO, were used as a marker for vegetative cells [5]. Unexpectedly, both cell fractions showed presence of rbcS transcripts (Fig. 2B). This was verified by western-blots on whole cell protein extracts using an antibody against RbcL, encoding the large subunit of the RuBisCO (Fig. 2C). Further examination of the heterocyst fractions by epifluorescence-LM revealed the presence of immature heterocysts as verified by their retention of high levels of chlorophyll a (Chla) and phycoerythrine autofluorescence (data not shown). Thus, the RuBisCO detected in the heterocyst enriched fractions ( Fig. 2B and 2C) was most likely due to the presence of immature heterocysts and not to vegetative cell contaminations. To confirm this hypothesis, transcripts of the cell division-related ftsZ gene (filamenting temperature-sensitive mutant Z) and expression levels of the FtsZ protein were also examined. FtsZ levels are typically low or non-detectable in terminally differentiated non-dividing heterocysts but high in dividing vegetative cells [22], where the protein (molecular mass of 50 kDa) assembles into a ring at the site of the growing new septum. However, using this marker, it became clear that both ftsZ transcripts and the FtsZ protein were only detected in the vegetative cell fractions ( Fig. 2B and 2D), which further shows the dominance of heterocysts in the heterocyst enriched fraction.
Next, the expression of nifH and the levels of NifH in both cell fractions were examined. The expression of nifH was detected exclusively in the heterocysts of C. raciborskii (Fig. 2E), and westernblots revealed high levels of NifH in the heterocyst fractions compared to the vegetative cell fractions, which lacked this protein ( Fig. 2F; upper image). These results were reproduced in three independent experiments (data not shown). When the same membranes were stripped and re-labelled for loading control assessment with the 30S riboprotein S1 antibody ( Fig. 2F; lower image), it was evident that protein levels from the heterocyst and vegetative cell fractions were similar.

Immunolocalization of Nitrogenase During a 12L/12D Cycle
These results suggest that nifH expression and NifH localization are restricted to terminal heterocysts for both cultures grown under a continuous light regime and those grown under a 12L/ 12D cycle. Because the cell isolation procedure takes longer to perform than the length of the nitrogenase activity peak (see Fig. 1B), intact sectioned trichomes were also tested using transmission electron microscopy coupled to gold conjugated antibodies (immunogold-TEM) targeting NifH. To determine nitrogenase localization during peak nitrogenase activity (Fig. 1E), trichomes were sampled 3 h into the light phase from cultures grown under 12L/12D cycles. Diazotrophic Anabaena PCC7120 was used to determine a comparative immuno-TEM-labelling baseline of NifH levels in heterocysts, as well as as biological background levels of the non-NifH vegetative cells. In both cyanobacteria the cross-sectioned heterocysts showed higher levels of NifH label compared to the vegetative cells (Table 2; Fig. 3). To determine whether these differences were statistically significant, NifH levels in heterocysts and vegetative cells of both cyanobacteria were subjected to two-way analysis of variance (ANOVA) testing between the same cell types, different cell types, and the two different species. The only statistically significant differences were between the NifH levels in heterocysts compared to the vegetative cells (F-value [1, 88] = 167.99; p-value #0.001) (Table S1). Notably, the heterocysts of Anabaena PCC7120 and of C. raciborskii showed the same NifH levels (i.e. statistically not different; Table S1). Although some randomly distributed gold label was detected in the vegetative cells of C. raciborskii, this most likely represents biological background due to its low level in vegetative cells compared to heterocysts (Fig. 3). The low NifH levels detected by immunogold-TEM in the vegetative cells, suggested the potential existence of nitrogen-fixing diazocyte-like cells along the trichomes of C. raciborskii as in the cyanobacterium Trichodesmium [23,24,25]. Diazocyte cells are capable of carrying out N 2 fixation in light, although they lack the protective thick cell walls characteristic of mature heterocysts [23,24,25]. Thus, we next assessed the presence of nitrogenase in situ along intact (nonsectioned) trichomes of C. raciborskii by in situ immunofluorescence-LM. C. raciborskii cultures grown under 12L/12D cycle were again sampled during peak nitrogenase activity, and key antibody markers were used to localize immunofluorescence labels inside permabilized heterocysts and vegetative cells. An antibody against the NtcA protein, a transcription factor sensing the nitrogen status of the vegetative cells in cyanobacteria [26], was used as a positive marker for heterocysts. Although NtcA is present in all the cells of the C. raciborskii trichome when induced by the depletion combined nitrogen, it is present at even higher levels in developing heterocysts [27]. The specificity of the antibodies raised against NtcA was verified by western-blot, using whole protein extracts from diazotrophic C. raciborskii cultures, revealing a single peptide of a molecular mass of 25 kDa (data not shown). Additionally, the antibodies against FtsZ and RbcL were used as positive markers for vegetative cells. Unexpectedly it was not possible to detect the NtcA fluorescence label inside heterocysts (Fig. 4A). However, immunofluorescence-LM analyses were positive for NtcA, FtsZ  4) and a fragment of the 16S rRNA (lanes 5-8). (F) NifH levels (upper) and S1 protein levels (lower). Western-blot membranes were stripped and re-labelled with the 30S ribosomal protein S1 to verify loading similarities. Protein levels and selected gene expression levels were determined in enriched heterocysts (Het/H) or vegetative cell (Veg/V) fractions. Protein extracts from non-fractionated trichomes of C. raciborskii were used as a positive control (+). The closest molecular marker band is indicated. C. raciborskii DNA, and the RNA used to generate the cDNA, were used as positive (+), and negative (2)  and RbcL in the vegetative cells of C. raciborskii (Fig. 4A, 4B and 4C). The NtcA protein showed a strong centroplasm localization signal, as expected, (Fig. 4A [27]). The FtsZ protein was also clearly organized into cell division rings in the actively dividing cells (Fig. 4B), and the RbcL immunofluorescence label was apparent in carboxysome-like subcellular structures spread over the cell interior (Fig. 4C). Finally, when NifH was assessed, no immunofluorescence-LM signal was detected in vegetative cells of C. raciborskii (Fig. 4D), strongly arguing against the expression of nif genes in this cell type. Likewise, NifH immunofluorescence labeling was not observed in the vegetative cells of Anabaena PCC7120 serving as a control (Fig. 4E). Interestingly, the NifH2 label [28] was apparent in vegetative cells of A. variabilis ATCC29413 when examined 4 hours into the dark phase of 12L/12D cycles (Fig. 4F). We therefore conclude that the low NifH levels detected by immunogold-TEM in vegetative cells of C. raciborskii corresponded to biological background label, as was also verified by the absence of NifH immunofluorescence label in the vegetative cells of C. raciborskii.

Genomic Background Determining Terminal Heterocyst Morphology
The genome of C. raciborskii lacks several genes related to heterocyst differentiation and heterocyst pattern regulation [12]. Still, it retains patA, patL and hetF homologs, which are known to be related to these functions [10,11,29,30]. As a first approach to determine if the C. raciborskii homologs conserve their functions in heterocyst pattern regulation, we analyzed their phylogenetic relationship with the homologs characterized in Anabaena PCC7120.
The Anabaena PCC7120 PatA sequence (hereafter N7120-PatA, 379 aa; encoded by all0521), was used to retrieve a total of 23 homologous gene sequences from other cyanobacteria (Table S2).   All homologs contained the conserved PatA N-terminal (PATAN) domain, and the CheY-like phosphoacceptor (receiver, REC) domain at the C-terminal end [31]. The resulting PatA phylogenetic tree showed that the C. raciborskii homolog (CR-PatA) is localized in the same clade as N7120-PatA and PatA sequences from intercalary heterocystous cyanobacteria (Fig. 5A, green shading). Identical topologies were obtained when analyzing the PATAN and REC domains separately (data not shown). Additionally, whole genome alignments show highly conserved synteny around the CR-PatA and the intercalary heterocystous PatA sequences (data not shown). Notably, CR-PatA (308 aa) has the shortest and least conserved sequence within its clade, sharing between 37-49% identical residues with PatA sequences from intercalary heterocystous cyanobacteria (which have 61-95% pairwise identity between themselves). Interestingly, a second PatA homolog was retrieved from all heterocystous cyanobacteria (Fig. 5A, yellow shading), except for C. raciborskii. Analysis of synonymous and non-synonymous substitutions for the heterocystous patA sequences indicated that CR-patA is evolving under purifying selection and is thus likely to be functional. However, the shorter linker region between the PATAN and REC domains of CR-PatA, which is 20 aa compared to 55-91 aa in the PatA sequences of intercalary heterocystous cyanobacteria (Fig. S1), suggests that although the exact nature of the regulation of PatA in heterocystous cyanobacteria is unknown, this truncation in CR-PatA might limit proper interaction of the two domains in C. raciborskii.
Anabaena PCC7120 PatL (hereafter N7120-PatL, 496 aa), encoded by all0521, contains a region of several pentapeptide repeats at its C-terminus. A total of 1001 homologs in cyanobacteria were retrieved using N7120-PatL as query sequence. A subset of PatL homologs, including single best-scoring homologs from non-heterocystous cyanobacteria, all sequences from heterocystous cyanobacteria (Table S2), and all five paralogs in C. raciborskii (CR-PatL parologs), were used to construct a PatL phylogenetic tree (Fig. 5B). The resulting phylogenetic tree shows that none of the five CR-PatL parologs (Fig. 5B, pink letters) are placed in the same clade as N7120-PatL or its best-scoring homologs among heterocystous cyanobacteria (Fig. 5B, green shading). Additionally, there was no synteny around the CR-PatL paralogs and N7120-PatL in its best-scoring homologs among heterocystous cyanobacteria. These results suggest that the CR-PatL paralogs are most likely not involved in heterocyst pattern formation.
To analyze HetF homologs among cyanobacteria, sequences from Anabaena PCC7120, Raphidiopsis brookii D9 and Trichodesmium erythraeum passed a comprehensive manual editing and selection procedure (see Information S1). The alignment of the HetFhomologs showed that all sequences contained a conserved His-Cys dyad, which constitutes the active site residues of cysteine proteases (including caspases) [32]. Our phylogenetic analysis showed that one of the R. brookii and C. raciborskii HetF paralogs (RB-HetF and CR-HetF respectively) clustered together with the homologs of heterocystous cyanobacteria (Fig. 5C, green shading). Notably, all HetF homologs of this group, except for RB-HetF, contain a predicted transmembrane helix at approximately residue 550 (although with a lower probability in Lyngbya sp. PCC 8106; data not shown). No such transmembrane region was found in the homologs of the other major clades (Fig. 5C). These results suggest that CR-HetF might have a conserved role in heterocyst pattern formation and that the transmembrane region of HetF might be important for heterocyst differentiation.

Discussion
Despite its restricted heterocyst frequency, C. raciborskii is an ecologically successful freshwater cyanobacterium which in recent decades has rapidly spread from tropical and sub-tropical to temperate aquatic regions [13,14,33], e.g. from Greece [34], via Hungary, Germany, Austria, France and recently to Portugal [13,35,36,37,38,39]; as well as to Canada and New Zealand [40,41]. Increasing water temperatures ('climate change') and ecotype selection have been proposed to explain this rapid change in the distribution of the toxin-producing C. raciborskii. The ecological competence of C. raciborskii has also been attributed to its facultative diazotrophy, which allows it to compete efficiently in environments with high fluctuations in combined nitrogen availability [15]. Despite its novelty and the ecological success of C. raciborskii, nitrogen fixation in terminal heterocystous cyanobacteria has until now largely remained unstudied. Our investigation shows that C. raciborskii, grown under combined nitrogen depletion and 12L/12D cycles, performs NifH biosynthesis and presents its highest nitrogenase activity in the light period (Fig. 1). Additionally, our data demonstrates that C. raciborskii restricts nifH expression and NifH biosynthesis to its two terminal heterocysts, irrespective of trichome length (Figs. 2, 3 and 4). Thus, the possibility that C. raciborskii compensates for lack of interspersed heterocysts by fixing N 2 in the vegetative cells in the dark can be discarded, as verified by our localization of nitrogenase in intact trichomes (Fig. 4).
The question that follows is how C. raciborskii is able to sustain the long trichomes (often .100 vegetative cells long) with enough nitrogen with only two terminal heterocysts per trichome (Fig. 1A). When compared to Anabaena PCC7120, which has one heterocyst every 7-15 vegetative cells [19], both show similar NifH levels in their heterocysts (Table 2). Although Anabaena PCC7120 differentiates a higher frequency of heterocysts per trichome compared to C. raciborskii, both cyanobacteria showed similar maximum nitrogenase activity rates at their peak (12L/12D cycles): 8.37 (SD61.98) and 10.20 (SD61.43) nmol C 2 H 4 per mg of protein mL 21 h 21 for C. raciborskii and Anabaena PCC7120, respectively (n = 3 cultures with three generations of diazotrophic establishment and an OD 750 of 0.61 SD60.05 and 0.63 SD60.04 respectively). This finding points to a higher nitrogenase activity in the cyanobacterium with terminal heterocysts. Since our data clearly show that the sole sites for N 2 fixation are the terminal heterocysts, C. raciborskii has to solve the problem of how to mobilize the fixed N 2 from its terminal heterocysts to all other cells of its often long trichomes. One possibility is that C. raciborskii might possess a more efficient transport mechanism along its trichomes, perhaps combined with a more efficient detainment of the fixed N 2 within the trichomes.
C. raciborskii has the smallest sequenced genome described for any sequenced heterocystous cyanobacterium (3.88 Mbp [12]). It is in fact approaching or even smaller than, the genome size of some unicellular diazotrophic cyanobacteria such as species within the genera Cyanothece and Crocosphaera [42]. Indeed, whole genome phylogeny suggests that C. raciborskii belongs to a clade of cyanobacteria characterized by shrinking genomes, which however seem to have originated from an ancestor with intercalary heterocysts and an estimated genome size of 5.65 Mbp [42,43]. Interestingly, R. brookii, the closest relative to C. raciborskii with a sequenced genome, has an even smaller genome (3.18 Mbp [12]) and lacks heterocysts as well as N 2 fixation activity altogether. In Anabaena PCC7120 a terminal heterocyst phenotype is apparent when deleting patA [10,11] and patL [29]. Also, the terminal heterocyst phenotype of the patA mutant is reverted by ectopic expression of hetF, indicating that HetF exerts its function downstream of PatA [30]. Our phylogenetic analysis of the PatA, PatL and HetF homologs (Fig. 5) provides additional support for the close relatedness between C. raciborskii, R. brookii and the third member of this shrinking genome clade, a symbiont of the fern Azolla, 'Nostoc azollae' 0708 [12,42,43]. Also, in addition to the loss of several genes with functions in heterocyst differentiation and pattern regulation, the terminal heterocyst phenotype of C. raciborskii correlates with the lack of a patL paralog, and not to the absence of patA or hetF. It has been suggested that genome reductions in C. raciborskii and R. brookii may also be a consequence of interactions with yet unidentified eukaryotic hosts. An alternative explanation is that this may simply represent another example of reductive genome evolution [12,44], a strategy used by the globally wide-spread and ubiquitous marine unicellular cyanobacterial genera Prochlorococcus and Synechococcus [42,45]. One of the few other examples of the terminal heterocyst phenotype of C. raciborskii are the so called primordia [46]. This is an intermediate life-stage between the small-celled motile hormogonia (functioning as either escape or plant colonization units), and fully mature vegetative trichomes of the heterocystous, often symbiotic, genus Nostoc [46]. However, both hormogonia and primordia are, as opposed to C. raciborskii, transient stages in the life span of Nostoc spp. In spite of the more severe genome streamlining in R. brookii, this closely related strain has more genes involved in ''aa transport and metabolism'' and ''coenzyme transport and metabolism'' than C. raciborskii [12]. The loss of heterocysts and N 2 fixation in R. brookii has been suggested to be due to a consistent availability of dissolved inorganic or organic nitrogen sources in its natural environment [15]. Our analysis also revealed that, although RB-HetF is closely related to CR-HetF and all homologs from heterocystous cynanobacteria, it lacked the conserved transmembrane region present in all representatives of this clade. Additionally, the fact that RB-HetF is represented by two neighboring genes (Information S1) might reflect the ongoing streamlining of R. brookii.
Taken together, we here show an exclusive localization of the nitrogenase enzyme in the heterocysts in a cyanobacterium differentiating only two terminal heterocysts. Additionally, it is shown that the genome of C. raciborskii lacks the patL homolog, on which other cyanobacteria with intercalary heterocysts depend. Hence, our study offers baseline information for future research regarding the mechanisms involved in terminal heterocysts differentiation and their assumed highly efficient transfer to the vegetative cells, or detainment of the fixed N 2 in the trichomes, or a combination of both. This research will also help identify means by which C. raciborskii and other terminal heterocyst forming cyanobacteria survive, and as in the case of C. raciborskii, even become ecologically successful.

Materials and Methods
Strains, culture growth and trichome length determination

Nitrogenase activity and Oxygen concentration assessment
Cultures (10 ml) were transferred to a 17 ml vial, and 1 ml of air was replaced with 1 ml of acetylene. All vials were incubated in agitation for 1 h under the corresponding growth condition. After incubation, 0.5 ml of the gas phase was withdrawn and ethylene content was measured as described in Lundgren et al. [48]. Nitrogenase activity rates were normalized to chlorophyll a (Chla) or protein content and background ethylene values in the acetylene gas phase were subtracted from control vials lacking cyanobacteria. Oxygen concentration was measured with a pH/ oxygen Pocket Meter 340i (WTW) and normalized to protein content. Chla was extracted as described in [49], and proteins were extracted and quantified as described below.

Heterocyst and vegetative cell fractionation
To generate the heterocyst and vegetative cell enrichments, we developed a protocol based on the eukaryotic cell partitioning system described by Van Alstine [51] (see Information S1). The efficiency of this cell partition method was revised by counting 100 cells of each fraction by LM, of 4 independent cell partition experiments.

Nucleic acid extractions and gene expression analysis
Genomic DNA of C. raciborskii was extracted according to the CTAB method [52]. For total RNA extraction, diazotrophically grown C. raciborskii cultures were collected by vacuum filtration and transferred to lysing matrix tubes with RLT buffer (InviTek, Berlin, Germany). Cells were disrupted and RNA extraction was performed using InviTrapH Spin Cell RNA MiniKit (InviTek) according to the manufacturer's instructions. Samples were DNase treated (turbo DNA-free TM , Ambion), and RNA integrity was verified by RNA gel electrophoresis in 1% (w/v) agarose. RNA was reverse transcribed to cDNA using an ImProm-II TM system (Promega, Southampton, United Kingdom), and PCR amplifications were performed using the primers presented in Table S3. PCR amplifications consisted of 35 cycles with annealing at 60uC 630 s and elongation at 72uC 61 min. All PCR products were visualized by PAGE 8% (w/v) with TBE, and AgNO 3 stained [53].

Immunogold-TEM
Diazotrophically grown C. raciborskii and Anabaena PCC7120 were sampled 3 h into the light phase of their 12L/12D cycles, concentrated by centrifugation, and embedded in 2% (w/v) agar in phosphate buffer solution (PBS). Agar-embedded trichomes were fixed 30 min in 3% (w/v) paraformaldehyde in PBS, washed 3 times with PBS, dehydrated in an ethanol series (30% to 100% [v/v]), and finally embedded in LR-White TM (London Resin). NifH immunogold labelling was performed as described previously [54], using goat anti-rabbit IgG conjugated to 10-nm gold particles (Sigma-Aldrich) (1:200 in 10% [w/v] BSA in PBS). Samples were sectioned by ultramicrotome (Sorvall MT5000, Norwalk, CT, USA), and mounted on copper grids for TEM examinations. Control experiments were performed excluding the primary antibody, showing negligible non-specific binding of the secondary antibody to the sections. Statistical differences between the immunogold label in heterocysts and vegetative cells of C. raciborskii and Anabaena PCC7120 were analyzed by two-way analysis of variance (Statistica 7.0, Statsoft). All samples were observed in an EM 906 transmission electron microscope (Zeiss, Oberkochen, Germany), at 80 kV accelerating voltage.

Immunofluorescence labelling
Droplets of 30ml of diazotrophic cyanobacteria, grown under 12L/12D cycles, were placed on poly-l-lysine coated microscope glass slides and air dried. Samples were fixed in 70% (v/v) ethanol, at Phylogenetic analysis of C. raciborskii patAL and hetF For the multiple sequence alignments and phylogenetic analyses of PatA, PatL and HetF the corresponding protein sequences in Anabaena PCC 7120 (accessions NP_484565, NP_487345 and NP_487586, respectively) were used as queries in a BLASTP [55] search against the non-redundant protein database (nr, v2011-09-28). Homologous sequences with an E-value ,1e 210 and an amino acid (aa) identity $30% with a length $30% of the query sequence were retrieved (coverage threshold was raised to $50% for the PatA search) (Table S2). Multiple sequence alignments were generated using Muscle (v3.8.31, [56]) with default settings. The PatA and HetF alignments were manually edited (Information S1), and sites in alignments with gaps in $50% of sequences were removed prior to constructing Maximum Likelihood phylogenetic trees using FastTree v2.1.3 [57] with -slow andslownni settings. Ratios of non-synonymous and synonymous substitutions for patA sequences were calculated in HyPhy (v2.0020110824beta, [58]). Putative transmembrane helices were predicted for all protein sequences using the TMHMM Server v2.0 (http://www.cbs.dtu.dk/services/TMHMM/). Figure S1 Multiple sequence alignment of PatA homologs in cyanobacteria. Organism names and protein identifiers are shown in the left margin with the canonical heterocystous sequences shaded in green. The sequences are the same as in Figure 5A. Alignment columns are colored by conserved hydrophobic properties (30% cutoff). The PATAN, linker region and REC domains are indicated from the N-to the C-terminus. The C. raciborskii PatA sequence contains both the PATAN and REC domains but has the shortest linker region of the compared sequences.

Supporting Information
(JPG) Information S1 Supporting Materials and Methods.

(DOC)
Table S1 Two-way ANOVA of the immunogold-TEM NifH label in C. raciborskii and Anabaena PCC7120. df, denotes degrees of freedom; MS, denotes the mean square; * Significant difference. (DOC)