Bacterial strain and growth conditions

A. variabilis ATCC 29413 was grown in an eightfold dilution of the medium of Allen and Arnon[36, 37] (AA/8). Phototrophic and mixotrophic cultures were grown under continuous illumination by Philips cool white fluorescent lamps, 60–70 μmol photons m^-2 s^-1. Mixotrophic cultures were supplemented with 5 mM fructose. Heterotrophic cultures were grown in the dark in the presence of 5 mM fructose. Four hundred-ml phototrophic, mixotrophic, and heterotrophic cultures in 2.8-l Fernbach flasks were inoculated from 50-ml precultures grown in the same conditions. Cultures were inoculated at a concentration of 0.05 μg chlorophyll a ml^-1, and grown on a shaker at 30°C and 140 rpm. Actively growing filaments were harvested after seven days for phototrophic and heterotrophic cultures, and after four days for mixotrophic cultures. Dissolved oxygen was monitored in representative 400-ml cultures using an optical sensor system (Fluorometrix, Stow, MA) with a paper-thin, autoclavable luminescent oxygen sensor taped on the interior bottom surface of the flask, as described in the manufacturer’s instructions.

Separation of cell type-specific contents for RNA extraction

Cultures (400 ml) were sedimented at 500 × g for 5 min at 4°C, resuspended in ~15 ml RNAlater solution (Ambion, Austin, TX), and stored at -80°C. Once thawed, suspended filaments were sedimented (500 × g, 5 min, 4°C), resuspended in 50 ml of N₂-sparged HP buffer (30 mM Hepes/30 mM Pipes/1.0 mM MgCl₂, pH 7.2), and washed three times with N₂-sparged HP buffer containing 10 mM disodium ethylenediaminetetraacetic acid (HP/EDTA). Twenty percent of the suspension was used to extract RNA from whole filaments. The rest was used to isolate and extract heterocysts, by a modification of a published method[2], and to prepare vegetative cell-specific extracts. That method reported a final ratio of ca. 0.01 vegetative cells per heterocyst. The washed filaments were resuspended in 40 ml of HP/EDTA containing 1 mg ml^-1 lysozyme and were shaken at 30°C for 5 min. The lysozyme-treated suspension was sedimented (500 × g, 5 min, 4°C), and the resulting pellet was resuspended in 10 ml of HP buffer in a test tube. The tube was immersed in an ultrasonic cleaning bath (Model 8845–4, Cole-Palmer, Chicago, IL) and was subjected to cavitation for 3 min to destroy a fraction of the vegetative cells. Heterocysts and remaining vegetative cells were sedimented (500 × g, 5 min, 4°C), and the clear supernatant fluid (vegetative cell lysate) was saved on ice for extraction of vegetative cell-specific RNA. The sedimented cells were washed twice with HP/EDTA buffer. The washed cells were resuspended in 1 ml of HP/EDTA containing 0.2 mg ml^-1 lysozyme, shaken at 30°C for 25 min, sedimented (1,000 × g, 5 min, 4°C), and the pellet was resuspended in 1 ml of HP buffer. This suspension was immersed in a 12°C sonic bath for 15 min to destroy remaining vegetative cells, and again sedimented (1,000 × g, 5 min, 4°C). The supernatant solution was discarded, and the heterocyst-containing pellet was washed three times with HP buffer. Images of the resuspended pellets confirmed a high ratio of heterocysts to fragments of heterocyst envelopes and what may be ruptured remains of vegetative cell or heterocyst protoplasts (not shown).

RNA extraction

RNA was extracted from whole filaments, isolated heterocysts, and vegetative cell extracts with the RiboPure-Bacteria kit (Ambion) as described[38]. Extracted RNA was purified with an RNeasy Mini kit (Qiagen, Valencia, CA) and eluted in 30 μl of water. RNA preparations were stored at -80°C until use. All RNA extractions were performed on three biological replicates. RNA samples were quantified using a NanoDrop ND-1000 spectrophotometer (NanoDrop Technologies, Wilmington, DE).

RNA quality and cell-specificity controls

The separate purifications of total RNAs from vegetative cells and from heterocysts from the same culture took close to 5 h. Because of this unavoidable time constraint, our experiments may provide reliable information only for RNAs that are stabilized by Ambion RNAlater and, perhaps, abundant. The quality of the extracted RNA was tested on an RNA 6000 Nano LabChip (Agilent Technologies, Santa Clara, CA) using a 2100 Bioanalyzer (Agilent Technologies). Reverse transcription followed by quantitative real time-PCR (RT-qPCR) was used to test the cell specificity of RNA extractions. The rbcL gene (Ava_3907) was used as a vegetative cell-specific gene and nifK (Ava_3930) was used as a heterocyst-specific gene[2, 39]. The RNAse P RNA gene (rnpB), constitutively expressed in A. variabilis, was used as an internal control for data normalization[40]. In addition, PCR reactions were performed using RNA and cDNA as templates and rnpB _F and rnpB _R as primers to control for possible contamination of our purified RNA samples with genomic DNA. The gene-specific primers (Additional file1) were designed using Primer Express 3.0. First-strand cDNA was prepared by reverse transcription using Superscript II reverse transcriptase (Invitrogen, Carlsbad, CA) and a combination of random primers (Invitrogen). 1.5 μl of reverse transcription reaction mixture was used for each RT-qPCR reaction. Each reaction mixture contained 2 μM of each gene-specific primer and 7.5 μl of Power SYBR green PCR master mix (Applied Biosystems, Foster City, CA). RT-qPCR was performed with the three biological replicates on an ABI 7900HT Fast Real-Time PCR System (Applied Biosystems). Relative fold changes in transcript levels were calculated using a standard curve for relative quantification (pools of 1 pg to 250 pg of cDNA were used).

Microarray experiments

cDNA was synthesized from the twenty-seven RNA samples (three culture conditions, and triplicate RNA extractions from each of whole filaments, vegetative cells, and heterocysts) by the University of Wisconsin-Madison Gene Expression Center. DNA end-labeling, hybridization, scanning, and data normalization were performed by NimbleGen (Reykjavík, Iceland), which provided the final data file. Cy3-labeled cDNAs were hybridized to NimbleGen expression array chips (Product no. A4385-00-01) that represent 5,657 ORFs in the A. variabilis genome (GenBank accession no. CP000117) excluding a 49-ORF incision element (GenBank accession no. NC_014000). Each ORF was represented by seventeen 60-mer oligonucleotides. Each oligonucleotide was present four times on the array. The twenty-seven microarray data files were normalized against each other using quantile normalization[41]. Expression array data were analyzed using ArrayStar 3.0 (DNASTAR, Madison, WI). Microarray data have been deposited in the National Center for Biotechnology Information Gene Expression Omnibus database (http://www.ncbi.nlm.nih.gov/geo/, accession number GSE46076).

In this paper, upregulation of a gene in a given cell type means upregulation in comparison to the other cell type in the same condition(s). A gene will be said to be transcribed at background, just above background, very low, low, moderate, high, and very high levels in a particular condition when its normalized transcript level is in the range of ≤150, 151–200, 201–600, 601–2,000, 2,001-6,000, 6,001-20,000, or 20,001-60,000 signal intensity units (SIU after normalization) in that condition, respectively. A distinction between “background” and “just above background” is somewhat arbitrary: some genes in one of these categories may belong in the other.

Statistical data analyses

Cell extracts and enzyme assays

For enzyme assays of A. variabilis grown in phototrophic conditions, cultures were harvested by centrifugation when chlorophyll concentration reached 8 μg/ml, and stored at -80°C. To prepare crude extracts from whole filaments, cells from 200-ml cultures were resuspended in 10 ml lysis buffer (50 mM Tris–HCl, pH 8.4, containing 1 mM phenylmethylsulfonyl fluoride and one protease inhibitor cocktail tablet [complete mini, EDTA-free, Roche Diagnostics, Indianapolis, IN]). Cells were lysed by two passages through a French press maintained at 4°C (4,000 to 5,000 psi). After centrifugation of the whole filament lysate (2,000 × g, 15 min, 4°C), the supernatant solution was dialyzed twice against 20 mM Tris–HCl (pH 7.2), with a total dialysis time of 24 h (SpectraPor dialysis tubing, 12,000-14,000 Da cut-off, Spectrum Laboratories, Rancho Dominguez, CA). Dialysis was required to remove phosphates from the lysate. The dialyzed filament lysate was used in enzyme assays. To prepare crude extracts of enriched heterocyst fractions, heterocysts were purified as described for RNA purification. Purified heterocysts were resuspended in 1.5 ml lysis buffer and lysed using zirconia beads in a Mini-BeadBeater (Biospec Products, Bartlesville, OK) on high speed setting (1 min, 4°C). After centrifugation (1,600 × g, 10 min, 4°C), the supernatant solution―representing the soluble extract―was concentrated by ultrafiltration, and used for protein and enzyme activity assays. Protein concentrations were determined using the Bio-Rad protein assay kit (Bio-Rad, Richmond, CA), with bovine serum albumin as the standard.

Phosphoserine phosphatase activity was measured at 30°C as described[44], using 13–270 μg protein in each assay. The phosphate released was quantified using the malachite green method[45] on a DU-650 spectrophotometer (Beckman, Fullerton, CA).

Concentration of dissolved oxygen in cultures

To avoid potential contaminations, particularly in cultures grown with fructose, cultures were shaken under ambient air, but not bubbled. Dissolved O₂ was monitored during growth to confirm that cultures were fully aerobic (data not shown). Between inoculation and harvest, the dissolved O₂ in phototrophic and mixotrophic cultures increased from 6.1 mg l^-1 just after inoculation to 7.5 mg l^-1 O₂ at harvest time (7.5 mg l^-1 is the O₂ saturation value at 30°C). The dissolved O₂ in heterotrophic cultures varied between 6.1 mg l^-1 and 6.3 mg l^-1 during the entire growth period.

Quality and cell-specificity of RNA extractions

Only heterocyst RNAs from phototrophic cultures showed evidence of degradation, with most of the degraded RNA species over 200 nt long (Additional file2: Figure S1). Because reverse transcription of bacterial RNA used random primers, and because each gene on the microarray was represented by seventeen probes, microarray experiments were nonetheless likely to capture most of the abundant RNAs. RNA extractions from heterocysts of phototrophic cultures, repeated for nine biological replicates, yielded similar degradation results. The samples that looked the least degraded were used for microarray experiments. Heterocyst RNAs from phototrophic cultures show, otherwise, trends in transcript levels very similar to those observed with heterocyst RNAs from mixotrophic and heterotrophic cultures (see Overall microarray assessment section), suggesting that RNA degradation in extracts from phototrophic cultures is not a major limitation in our experiments. PCR reactions using RNA samples as templates never showed a PCR band and always showed a PCR band with cDNA controls (data not shown), indicating that our RNA preparations were devoid of contamination by genomic DNA.

The cell specificity of our RNA preparations was tested by RT-qPCR. We chose nifK and rbcL as cell specificity marker genes because it is well established that under oxic conditions nifK is expressed only in heterocysts and rbcL is expressed mostly, perhaps only, in vegetative cells[4, 14]. Ct (threshold cycle) values for rnpB did not vary by more than 5% between heterocysts and vegetative cells in all three growth conditions (not shown), validating our choice of rnpB as a constitutively expressed gene that can be used to normalize the transcript levels of other genes across experiments. The relative rbcL signals obtained from heterocyst RNA were only 7.6% and 6.9% of those obtained from vegetative cell RNA in phototrophic and mixotrophic cultures, respectively (Figure 1). In contrast, the relative nifK signals obtained from vegetative cell RNA were only 11.8% and 10.1% of those obtained from heterocyst RNA in phototrophic and mixotrophic cultures, respectively. A conservative interpretation of these results is that heterocyst RNA preparations were over 9% and over 93% cell-specific for phototrophic and mixotrophic conditions, respectively. Vegetative cell RNA preparations were over 88% and 89% cell-specific for phototrophic and mixotrophic conditions, respectively. With heterotrophic cultures, the cell specificity of heterocyst RNA and vegetative cell RNA preparations never appeared to be above 83%, even though RNA extractions were repeated eight times, each time making the first lysis step gentler and the last lysis step harsher to better separate RNA from the two cell types.

If a transcript is more abundant in filaments than in vegetative cells, and yet this transcript is only modestly more abundant―or even less abundant―in heterocysts than in filaments, the heterocyst level of that transcript is likely under-represented in our experiments (examples, including nitrogenase [nif1] transcripts, are presented below). When transcript levels in whole filaments are consistent with transcript levels in vegetative cells and heterocysts, and in particular when specific genes are transcribed at high levels across cell types and growth conditions, and in the absence of contradictory information, we consider those genes―or whole pathways―active in heterocysts. On the other hand, transcript levels only slightly above background level in heterocysts will not be considered as evidence that genes or intact pathways are active in heterocysts, even though they may be. We are trying to be conservative in our interpretations in this first effort to use microarray data to identify active pathways in vegetative cells and heterocysts of N₂-fixing filaments, especially because the importance of major enzymatic pathways (including nitrogen fixation, the processing of sucrose by invertase, the oxidative pentose phosphate cycle, and cytochrome oxidase activity) might otherwise be misinterpreted.

Overall microarray assessment

The experimental metrics report provided by NimbleGen (not shown) gives summary statistics that can be used to help identify potential problems during hybridization. All metrics for the twenty seven microarray experiments were within the manufacturer’s suggested ranges.

The normalized microarray data are shown in Additional file3. The coefficients of determination (R² values) between the twenty seven experiments were calculated to quantify experimental variability between biological replicates (Additional file4). Reproducibility was high for biological replicates of the same experiment, as indicated by R² values ranging between 0.857 and 0.998. The R² values between microarrays using heterocyst RNAs isolated from different culture conditions were also high, between 0.827 and 0.976. These results also include the experiments with the partially degraded heterocyst RNAs extracted from phototrophic cultures, suggesting that partial degradation of the RNA has only a minor effect on overall hybridization results. The R² values between microarrays using vegetative cell RNAs and whole filament RNAs isolated from the same culture types also were high, between 0.876 and 0.994, reflecting the fact that filaments comprise mostly vegetative cells. In contrast, microarray results varied more when comparing vegetative cell RNAs extracted from different types of cultures (R² values between 0.542 and 0.817) or when comparing heterocyst and vegetative cell RNAs from the same cultures (R² values between 0.400 and 0.871). These results make sense based on the respective metabolic functions of vegetative cells and heterocysts (see explanation below).

In all growth conditions and for each cell type, signal intensities were not normally distributed (Figure 2, left panels). A high number of genes with low intensity signals (log₂ [intensity] below 7.0) is found across all experiments, independent of cell type and culture condition, and may correspond to genes whose RNA is disproportionately labile. The proportion of genes with low signal intensity in the heterocysts of phototrophic cultures is not higher than it is in vegetative cells or whole filaments in the same culture conditions (Figure 2, top left panel). This observation suggests that the poorer quality of the RNA extracted from the heterocysts of phototrophic cultures did not substantially bias the results.

Microarray experiments with RNA from whole filaments were used to validate the results of the experiments performed with cell-specific RNA. In N₂-fixing A. variabilis filaments, transcript levels of any gene, i, should be consistent with the equation, F_i = aV_i + bHt_i. Assuming that heterocysts and vegetative cells contain similar amounts of RNA and assuming that RNA is extracted with the same yield from whole filaments, vegetative cells, and heterocysts, a + b should equal 1, with the a-value ranging between 0.9 and 1 and the b-value ranging between 0 and 0.1. Linear modeling was applied to reduced data sets (Additional file5), where genes that showed average transcript levels below 128 across experiments and genes with high variability between biological replicates were removed (see Additional file6 for details). The values of a and b were determined for the three growth conditions (Additional file6). With the exceptions that the a-value was above 1 in phototrophic and heterotrophic conditions and the b-value was below 0 in heterotrophic conditions, the calculated values for a and b were generally in the ranges expected from the frequency of heterocysts in filaments (i.e., a ~ 0.92 and b ~ 0.08). Although we do not know whether heterocysts and vegetative cells have the same amounts of mRNA, equal amounts of cDNA were used in all hybridization experiments, possibly biasing the values of a and b during linear modeling. Our results remain consistent with the idea that for most genes the transcript level of a gene in heterocysts contributes little to the transcript level of this gene in whole filaments. Thus for most genes, transcript levels in whole filaments closely approximate transcript levels in vegetative cells.

In phototrophic and mixotrophic conditions, few genes in the reduced data set behaved as outliers, with transcript level data that did not closely conform to the equation F_i = aV_i + bHt_i. (Outliers are not discussed for heterotrophic conditions because the value of b was not reliable: see Additional file6). Deviation from the linear equation suggests that RNA is degraded in one type of cell or the other. The most conspicuous outliers (i.e., the points farthest from the plane defined by F = aV + bH_t) were identified in each growth condition by calculating weighted residuals as a proportion of each gene’s transcript level using equation 1 (Additional file6). Two sets of outlier genes in phototrophic conditions warrant mention. The nif1 genes, nifB, S, U, H, D, K, E, N, X, and W (Ava_3912, Ava_3914-3917, Ava_3930, Ava_3932-3934, and Ava_3937, respectively) were the 3^rd to 12^th outliers for which F_i > > aV_i + bHt_i. The transcript levels of nif1 genes and of related maturation genes should be strongly upregulated in heterocysts compared to vegetative cells[2, 39, 46], and the signal intensities for these genes should be ca. 10-fold lower in whole filaments than in heterocysts. Instead―especially in phototrophic conditions―signal intensities for nif1 genes were nearly always higher in whole filaments than in heterocysts, implying that the signal intensities in heterocysts were at least 10-fold lower than expected. This observation suggests that the nif1 transcripts are specifically targeted for rapid degradation in heterocysts upon separation of the heterocysts from vegetative cells under aerobic conditions. Transcripts of nif1 genes may represent a large fraction of the degraded transcripts seen in heterocysts of phototrophic cultures (Additional file2: Figure S1). These results might be related to the degradation of nifHDK transcripts observed in PCC 7120[47]. Because certain nif1 transcripts accumulated to up to 44% of the most abundant transcript in heterocysts in these conditions (consistent with the very large amount of protein attributable to Nif in non-denaturing gels of A. variabilis heterocysts[2]), the seemingly artificially low transcript levels for nif1 genes likely caused a factitious increase of transcript levels for all other genes in the heterocysts of phototrophic cultures. Therefore, moderate upregulation (below 5-fold) of genes other than nif1 in the heterocysts of phototrophic cultures may not be meaningful. Second, five PS II genes (Ava_4121, Ava_0593, Ava_1597, Ava_3553, and Ava_2460, four of them psbA genes) are the top two and the top 13^th to 15^th outliers. These genes have signal intensities in filaments that are 1.6- to 38-fold lower than in vegetative cells. This trend in transcript levels of psbA genes is reminiscent of what happens in cyanobacteria subjected to oxidative damage (see Targeted analysis-Photosystems).

General analysis of microarray results

The only other use made of the reduced data sets (Additional files5 and6) was to highlight the differences in transcript levels between vegetative cells and heterocysts in the different growth conditions using volcano plots (Additional file2: Figure S2). P values for those plots were calculated using two-tailed t-tests with unequal variances. Two hundred eighty, 144, and 545 genes were significantly upregulated (over 2-fold difference with p < 0.01) in vegetative cells in phototrophic, mixotrophic, and heterotrophic cultures, respectively. Of these genes, 22.5% to 24.3% had unknown products. Five hundred sixty five, 505, and 301 genes were significantly upregulated (over 2-fold difference with p < 0.01) in heterocysts in phototrophic, mixotrophic, and heterotrophic cultures, respectively. Of these, 36.8% (in phototrophic conditions) to 46.2% (in heterotrophic conditions) were genes with unknown products. Of the genes with unknown products that were upregulated in one type of cell versus the other, 77%, 86%, and 51% were upregulated in the heterocysts in phototrophic, mixotrophic, and heterotrophic conditions, respectively. In summary, although transcript levels in vegetative cells and heterocysts are highly correlated (Additional file6), many genes were significantly upregulated in one cell type versus the other in each growth condition.

PCA was used to determine how gene transcript patterns relate to cell type and culture conditions. In the three culture conditions, principal components for the whole filament were close to those for vegetative cells, but not to those for heterocysts (Figure 3), agreeing with the fact that vegetative cells typically represent 90% to 95% of total cells in the filaments. Principal components for vegetative cells varied significantly between growth conditions. These results agree with the fact that vegetative cells are responsible for uptake of carbon and energy, and for the generation of reductant, and with the fact that carbon, energy, and reductant are the parameters that vary between growth conditions. In contrast, principal components for heterocysts varied little between growth conditions. Heterocysts are consistently responsible for nitrogen fixation. The lack of change of principal components for heterocysts in heterotrophic conditions suggests that access to light is not among the top determinants of transcript levels in heterocysts. The fact that heterocyst-specific PCA results (Figure 3) and volcano plots from phototrophic cultures (Additional file2: Figure S2) are not clearly distinguishable from those of mixotrophic and heterotrophic cultures helps to validate our decision to use seemingly partially degraded heterocyst RNAs from phototrophic cultures for our microarray studies.

Functional categorization of microarray data

To determine which pathways are upregulated in the different growth conditions and in the different cell types, the 5,657 ORFs represented in the microarrays were classified in sixteen functional categories (Additional file3). Fourteen categories were based on the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway database[48], Blastp searches[49], and previous publications of gene functions. ORFs annotated only with a protein domain name were arbitrarily included in the Other functions category and those annotated as hypothetical proteins or proteins of unknown function were arbitrarily grouped in the Unknown category. The Other and Unknown categories contained 1,802 and 2,201 genes, respectively (Additional file3). Since filaments consist mostly of vegetative cells, distribution of transcript levels per functional category was highly similar in whole filaments and vegetative cells in each growth condition tested, as expected (Figure 4). Because sources of carbon and energy are the parameters that vary between growth conditions, the pathways that were upregulated in vegetative cells (and whole filaments) varied widely from one growth condition to another. In contrast, distribution of transcript levels in terms of functional category varied little in heterocysts across growth conditions, agreeing with the fact that heterocysts perform the same main metabolic function, N₂ fixation across the three growth conditions (Figure 4). These results agree with our PCA results.

The genes involved in phycobilisome assembly, photosynthesis, and CO₂ uptake/fixation were clearly upregulated in vegetative cells in phototrophic conditions. Transcript levels of these genes decreased in mixotrophic conditions, and even further in heterotrophic conditions, where all carbon and reducing power come from fructose. Genes involved in electron transfer and respiration were unexpectedly down-regulated in heterocysts across growth conditions. This observation does not support the common understanding that heterocysts actively respire[50, 51] as a way to decrease intracellular O₂ concentrations[4, 52, 53]. However, this appears to be another instance in which, at least under heterotrophic conditions and for several oxidase subunits, the transcript level in heterocysts is likely under-represented.

Targeted analysis

In this section our results will be described in terms of individual pathways, with a particular focus on pathways that we plan to study later by metabolic flux analysis (e.g., central carbon metabolism as well as nitrogen fixation and amino acid synthesis).

Amino acid biosynthesis

Whereas synthesis of Gln and Glu in N₂-fixing filaments has been the focus of many studies because they are responsible for ammonia assimilation after N₂ fixation, where and how the other amino acids are synthesized have not been looked at in much detail. Starting from the amino acid biosynthetic genes identified in A. variabilis in the KEGG database[57, 58], Blastp comparisons were used to verify all annotations and to identify which pathways are active. Not all pathways and genes could be identified with certainty, in particular enzymes involved in amination (i.e., Asn synthetase) and transamination reactions. The pathways shown in Figure 5 (extra comments in Additional file8) and Additional file7 represent the predominant amino acid biosynthetic pathways in A. variabilis based on the KEGG database, pathways that are common in the bacterial world[59, 60], known amino acid synthesis pathways in cyanobacteria, and pathways supported by earlier isotope labeling studies.

Amino acid biosynthetic genes were typically either upregulated in vegetative cells or transcribed at similar levels in the two cell types (Figure 5). Only select genes appeared upregulated in heterocysts (e.g., Ava_1668, with p ≤ 0.05) (Figure 5). A few instances were found in which multiple genes encoding isozymes showed different transcript patterns. Most amino acid biosynthetic genes are not organized in operons in A. variabilis, so one gene can be transcribed at a very low level, while all other genes in the pathway are transcribed at significant levels. Several genes showed background level transcripts across experiments, possibly due to mRNA instability, making it impossible to predict in which cell type these genes are transcribed (Figure 5). Using a signal intensity cutoff of 200 as the minimum, transcript levels in heterocysts plus the phosphoserine phosphatase activity detected in the crude extracts of heterocysts of phototrophic cultures (footnote f of Figure 5) suggest that Gln, Glu, Ser, Gly, Cys, Thr, and Pro are actively produced in heterocysts. Whether or not the other protein amino acids are actively synthesized in heterocysts is unclear based on our data, because of genes not identified or of transcript levels below 200 SIU for some genes in a given pathway (Figure 5).

The breakdown of phycobiliproteins in heterocysts has been studied as a possible major source of amino acids for de novo protein synthesis in heterocysts[61, 62]. All phycobiliprotein-encoding genes were still transcribed at significant levels in the heterocysts of phototrophic cultures (Figure 6). nblA (Ava_3383), encoding a protein required for the breakdown of phycobiliproteins was upregulated 2.2-fold (p < 0.01) in the heterocysts of phototrophic cultures, but not in other growth conditions. The alanine dehydrogenase gene Ava_0176, required for the breakdown of phycobiliproteins in Synechococcus PCC 7942[63], was downregulated in heterocysts across growth conditions. These collective results suggest that while the breakdown of phycobiliproteins may contribute much of the amino acids needed during heterocyst differentiation, it may contribute little to protein repair and protein de novo synthesis in mature heterocysts. This conclusion is consistent with labeling experiments that showed that newly forming and mature heterocysts of A. oscillarioides incorporated significant levels of ¹³C and ¹⁵N in cultures grown with NaH¹³CO₃ and ¹⁵N₂[64].

CO₂ fixation

In cyanobacteria, ribulose-1,5-bisphosphate carboxylase oxygenase (RuBisCO) is often concentrated in subcompartments—the carboxysomes—in which CO₂ is concentrated to high levels by a carboxysome-specific carbonic anhydrase[82–84]. Transcript levels of the genes encoding the high-affinity bicarbonate ABC transporter CmpABCD reflect the varying cellular need for CO₂ in the three growth conditions (Figure 7). Most carboxysome-related transcripts were significantly more abundant in the vegetative cells than in the heterocysts of phototrophic cultures (Figure 7). The abundant transcripts detected for most carboxysomal genes in heterocysts could have been produced in vegetative cells or proheterocysts (immature heterocysts) prior to their differentiation into heterocysts or could have been synthesized in the heterocysts. CcmK1, CcmK2, and CcmM are present at significant levels in the heterocysts of PCC 7120[26]. The carbonic anhydrase encoded by Ava_2165 is similar to CcaA, the carboxysome-specific carbonic anhydrase in many cyanobacteria, but Ava_2165 was transcribed at background level across experiments. Most Calvin cycle genes were upregulated in the vegetative cells of phototrophic cultures. They were transcribed at significant levels across most experiments, with the exception of Ava_3290, encoding triosephosphate isomerase, whose transcript levels were very low or not distinguishable from background (Figure 7 and Additional file7).

Development-related genes and certain uncharacterized genes

Our analysis was not expected to illuminate the process of heterocyst differentiation, per se, for three interrelated reasons: (i) Steady state cultures were needed for prospective metabolic flux analysis. (ii) RNA extracted from vegetative cells in steady state cultures is combined with RNA extracted from proheterocysts that have not yet formed the envelope that protects against cavitation. Presumably as a result, genes involved in early, transient developmental processes such as deposition of the heterocyst envelope appear upregulated in vegetative cells (Additional file7). (iii) Correspondingly, RNA is isolated from mature or almost fully mature heterocysts only when their envelope protects them extensively against cavitation. Therefore, our analysis provides the opportunity to highlight ongoing processes characteristic of mature heterocysts, including their interactions with vegetative cells, and to gain further insight into the physiological differences between mature heterocysts and vegetative cells. To those ends, we have provided (in Additional file7) a section on known heterocyst developmental genes and (in Additional file9) sections on differentially transcribed genes. Among these genes are Ava_2748 and Ava_2050, which are likely parts of one or two carbohydrate uptake transport (CUT) 1-family ABC transporters, known to be specific for di- and oligosaccharides. These two genes represent possible members of the yet unknown sucrose transporter. Although we recognize that under phototrophic conditions, transcript levels in heterocysts are inflated because of the low nif1 transcript levels, the 204-fold and 35-fold higher transcript levels for Ava_2050 and Ava_2748, respectively, in heterocysts than in filaments still suggest that these genes are upregulated in heterocysts in these conditions. Other genes from Additional file9 that have high transcript levels, and are strongly upregulated, in heterocysts may also warrant inquiry.

Under phototrophic conditions, 77% of the genes with unknown products that were upregulated in one type of cell versus the other were upregulated in heterocysts. Despite the limitations of our data (e.g., degradation of the nif1 transcripts), these observations suggest to us that much of what is unknown about the metabolism of phototrophic, N₂-fixing cultures takes place in heterocysts.

Amino acid biosynthesis

The current model for nitrogen assimilation after N₂ fixation has NH₃ incorporated into Gln by glutamine synthetase (encoded by glnA) in heterocysts[6, 17]. Gln is then transported to vegetative cells, where—together with 2-oxoglutarate—it serves as substrate for glutamate synthase (encoded by glsF) to produce two Glu[6, 7, 103]. Finally, one Glu is transported back to the heterocysts, where it becomes the substrate of glutamine synthetase for Gln synthesis. The Gln-Glu exchange between vegetative cells and heterocysts results in a net nitrogen flux to the vegetative cells. The second glutamate synthase substrate, 2-oxoglutarate, may also be produced in heterocysts by isocitrate dehydrogenase[7, 104]. In PCC 7120, glutamine synthetase activity is regulated by protein-protein interactions with the inhibitory protein IF7 (encoded by gifA), whose expression is negatively controlled by NtcA[105]. In A. variabilis, IF7 is encoded by Ava_0148, adjacent to glnA. Glutamate dehydrogenase (encoded by gdhA) does not seem to play an important role in nitrogen assimilation[103, 106], but it can give a competitive advantage in non-exponential phases of growth, possibly related to the fact that ATP is not required for its activity[107]. Other authors[108] suggest that Arg may be an alternative carrier of fixed nitrogen from heterocysts.

Prior results[6, 39] with A. cylindrica and PCC 7120 led to the expectation that under photoautotrophic conditions, glutamine synthetase would be active in both types of cells whereas glutamate synthase would be active specifically in vegetative cells. However, in A. variabilis, glutamine synthetase, glutamate synthase, and isocitrate dehydrogenase appear to be actively transcribed in both types of cells. Our results therefore suggest that Glu and Gln, as well as Ser, Gly, Cys, Thr, and Pro, are actively produced in heterocysts. For all other protein amino acids, our results are inconclusive. The phenotypes of several amino acid ABC transporter mutants of PCC 7120 are known[65–68], but molecules as large as 623-Da calcein may move back and forth passively between vegetative cells and heterocysts, perhaps through the septosome, without transiting through the periplasm[69, 70]. Until it has been proven or disproven that amino acids are transported through the septosome, the phenotypes of amino acid ABC transporters cannot be used to help reach a conclusion that a particular amino acid is or is not synthesized in heterocysts.

It is not excluded that only certain steps of amino acid synthesis take place in heterocysts, with precursors being provided by vegetative cells. Because heterocysts funnel much of their energy and redox power into N₂ fixation, some redox or energy-intensive enzymatic steps might take place only in vegetative cells. For example, the two genes in the Lys pathway that are transcribed at the lowest levels in heterocysts encode enzymes that require ATP (aspartate kinase) and NAD(P)H (dihydrodipicolinate reductase) (Figure 5). Future metabolic flux analyses with diazotrophically grown A. variabilis must test for the possible transport of Ala, Asn, Asp, Lys, Arg, Phe, Tyr, Trp, Ile, Val, Leu, His, and Met (or their precursors) between cell types.

CO₂ fixation

A. variabilis CcmM has an N-terminal carbonic anhydrase domain followed by three RbcS-like domains. In Thermosynechococcus elongatus, which lacks a ccaA homolog, CcmM is the carboxysomal carbonic anhydrase[84]. PCC 7120 and A. variabilis CcmM proteins contain all of the conserved residues thought to be required for T. elongatus CcmM carbonic anhydrase activity[84]. Because PCC 7120 lacks a ccaA homolog and because ccaA is barely transcribed in A. variabilis, CcmM―with high to very high transcript levels in vegetative cells―may be the carboxysome-specific carbonic anhydrase in PCC 7120 and A. variabilis. In Synechococcus PCC 7942, the average RuBisCO complex contains eight RbcL molecules and five RbcS molecules, and CcmM has been shown to form complexes readily with RbcL[110]. It is tempting to speculate that in vegetative cells of A. variabilis, CcmM assembles with incomplete RbcL-RbcS complexes to form a bi-functional RuBisCO-carbonic anhydrase. Such an assembly might permit efficient channeling of CO₂ from carbonic anhydrase to RuBisCO. It would also explain why the rbcS transcript was 13-fold less abundant than the rbcL transcript in the vegetative cells of phototrophic cultures (Figure 7).