Glycolipid composition of the heterocyst envelope of Anabaena sp. PCC 7120 is crucial for diazotrophic growth and relies on the UDP‐galactose 4‐epimerase HgdA

Abstract The nitrogenase complex in the heterocysts of the filamentous freshwater cyanobacterium Anabaenasp. PCC 7120 fixes atmospheric nitrogen to allow diazotrophic growth. The heterocyst cell envelope protects the nitrogenase from oxygen and consists of a polysaccharide and a glycolipid layer that are formed by a complex process involving the recruitment of different proteins. Here, we studied the function of the putative nucleoside‐diphosphate‐sugar epimerase HgdA, which along with HgdB and HgdC is essential for deposition of the glycolipid layer and growth without a combined nitrogen source. Using site‐directed mutagenesis and single homologous recombination approach, we performed a thoroughly functional characterization of HgdA and confirmed that the glycolipid layer of the hgdAmutant heterocyst is aberrant as shown by transmission electron microscopy and chemical analysis. The hgdA gene was expressed during late stages of the heterocyst differentiation. GFP‐tagged HgdA protein localized inside the heterocysts. The purified HgdA protein had UDP‐galactose 4‐epimerase activity in vitro. This enzyme could be responsible for synthesis of heterocyst‐specific glycolipid precursors, which could be transported over the cell wall by the ABC transporter components HgdB/HgdC.

The synthesis of HGLs and deposition of the hgl layer probably constitute a multistep pathway involving products of different genes (Awai, Lechno-Yossef, & Wolk, 2009;, and many questions remain open. It is known that a type I secretion system (T1SS)-like transporter is involved in the efflux of HGLs from the inside of the developing heterocysts to form the hgl layer (Fiedler, Arnold, Hannus, & Maldener, 1998;Maldener, Fiedler, Ernst, Fernández-Piñas, & Wolk, 1994;Staron, Forchhammer, & Maldener, 2011. This transporter is composed of the TolC homolog outer membrane protein HgdD, the periplasmic membrane fusion protein DevB, and the inner membrane ABC transporter DevCA. The DevBCA-HgdD efflux pump is essential for the hgl layer formation and heterocyst function (Fiedler et al., 1998;Staron et al., 2011).
Several homologs of the devBCA gene cluster in the genome of Anabaena sp. have been identified Staron, 2012). Some are important for diazotrophic growth and heterocyst maturation Shvarev, Nishi, Wörmer, & Maldener, 2018;Staron & Maldener, 2012). The cluster all5347/all5346/all5345 (hgdB/hgdC/hgdA) is of particular interest because the ATPase-coding gene devA is replaced by the hgdA gene coding for a putative epimerase. This gene cluster is essential for proper hgl layer deposition and growth of Anabaena sp. without combined nitrogen source , but the functions of the protein HgdA (All5345) was unknown.
For the nitrogen stepdown experiments, cells were washed three times in BG11 0 medium and cultivated afterward in BG11 0 .

| DNA manipulations
To construct an insertion mutant of hgdA by homologous recombination, an internal fragment of the gene was amplified by PCR (see Table A2 for primers) with 1 µl of the wild-type Anabaena sp. culture as a template and cloned into the XhoI-restricted suicide vector pRL277 (Table A1) using Gibson assembly (Gibson et al., 2009) ( Figure A2a). The resulting plasmid pIM695 was transferred into wild-type Anabaena sp. cells by triparental mating, followed by selection on streptomycin-and spectinomycin-containing BG11 agar plates. In the antibiotic-resistant Anabaena sp. colonies, where a single recombination event between the hgdA gene in the genome and its internal fragment in the pIM695 vector had occurred, the hgdA gene was disrupted by the pRL277 vector ( Figure 1A, A2a). Full segregation of one selected mutant (SR695) colony was confirmed by PCR ( Figure A2b) with a small piece of the mutant colony as template.
To localize HgdA in Anabaena sp. filaments, a plasmid with a translational fusion of the HgdA C-terminus with the superfolder GFP (sfGFP) (Pédelacq, Cabantous, Tran, Terwilliger, & Waldo, 2006) was constructed following the method described in .
The 3′-end of hgdA and the entire sfGFP were amplified by PCR and cloned into the XhoI-restricted suicide vector pRL277 using Gibson assembly. The resulting plasmid pIM717 was transferred into wildtype Anabaena sp. cells using triparental mating, followed by positive colony selection on streptomycin-and spectinomycin-containing BG11 agar plates. Anabaena sp. colonies contained the hgdA gene fused with sfGFP (strain SR717). The fusion was confirmed by PCR.
For complementation of the SR695 mutant, the hgdA gene under control of the glnA promoter (Valladares, Muro-Pastor, Herrero, & Flores, 2004) was cloned into the EcoRI-restricted selfreplicating plasmid pIM612, which bears a neomycin-resistance cassette (Bornikoel, 2018), using Gibson assembly. The resulting plasmid pIM774 was transferred into mutant SR695 cells, and positive colonies were selected on BG11 agar plates containing neomycin, streptomycin, and spectinomycin. The presence of the undisrupted hgdA gene in the complemented mutant colonies was confirmed by PCR ( Figure A2b).
For overexpression of the hgdA gene in E. coli, hgdA, followed by sequences encoding a Strep-tag and His-tag at the 3′-terminus was cloned into plasmid pET15b (Novagen, Merck) digested with NcoI, with help of Gibson assembly to yield plasmid pIM753.

| RNA isolation and RT-PCR
RNA was isolated at different time points after nitrogen stepdown using UPzol reagent (Biotechrabbit, Henningsdorf) according to the manufacturer's instructions from wild-type Anabaena sp. cells grown in bottles as described above. The purity and concentration of the extracted RNA were estimated by electrophoresis and GelQuantNET software (biochemlabsolutions.com). Reverse transcription (RT) reactions were performed using the Applied Biosystems RT-reaction kit. The primers used for all PCR reactions are listed in Table A2.

| Microscopy
For light and fluorescence microscopy, wild-type and mutant Anabaena sp. cells were placed onto agarose-covered glass slides and observed under a Leica DM 2500 microscope connected to a Leica DFC420C camera or a Leica DM5500 B microscope connected to a Leica monochrome DFC360 FX camera.
Fluorescence of GFP and BODIPY was recorded using a BP470 40-nm excitation filter and a BP525 50-nm emission filter. Cyanobacterial autofluorescence was captured using a 50-nm BP535 excitation filter and a 75-nm BP610 emission filter. Images were exposed for 80-150 ms in the fluorescence channels. Images of sfGFP and BODIPY fluorescence were taken as Z-stacks with 0.4µm intervals. Z-stacks were subsequently used to do 3D deconvolution using the integrated function of the Leica ASF software (Leica Microsystems). Images of fluorescence were recolored by the Leica ASF software based on the filters used.
For electron microscopy, cells were fixed and postfixed with glutaraldehyde and potassium permanganate, respectively (Fiedler et al., 1998). Ultrathin sections were stained with uranyl acetate and lead citrate and examined with a Philips Tecnai 10 electron microscope at 80 kHz.
Briefly, 1 ml of Anabaena sp. cell suspension was centrifuged at 4,000 × g for 10 min, washed with PBS buffer, and resuspended in 200 µl PBS. BODIPY (1 μl of 50 ng/ml in DMSO) was added.
The cell suspension was incubated in the dark for 30 min at room temperature and examined by light and fluorescence microscopy.
Fluorescence or phase-contrast images were captured with a Leica DM 5500B microscope connected to a Leica monochrome DFC360 FX camera.
Cells were stained with alcian blue following the protocol described in (McKinney, 1953). Cell suspensions were mixed with 1.5% Alcian blue in water (at a ratio of 20:1) and incubated at room temperature for 5 min.
For triphenyl tetrazolium chloride (TTC) staining, cell suspensions were mixed with TTC solution (0.01% TTC, w/v, in the final mixture) and incubated in the dark for 10 min at room temperature (Fay & Kulasooriya, 1972). Filaments stained with TTC or alcian blue were examined using a Leica DM 2500 microscope connected to a Leica DFC420C camera.
In brief, wild-type and mutant cells of equal chlorophyll a concentration [measured according to (Mackinney, 1941)] were pelleted and resuspended in methanol-chloroform (1:1) and pelleted again to remove cell debris. The solvents of the supernatant were evaporated under air in a fume hood.

| Nitrogenase activity
Nitrogenase activity was measured using the acetylene reduction method for cyanobacteria (Bornikoel, Staiger, Madlung, Forchhammer, & Maldener, 2018). Briefly, cultures were incubated in the presence of acetylene for several hours in flasks closed with gas-tight caps. Anoxic conditions were generated before incubation with acetylene by adding 3-(3,4-dichlorophenyl)-1,1-dimethylurea (DCMU, 10 µmol/L, in methanol); the sealed flasks were then filled with argon and incubated for 1 hr. For oxic conditions, this step was omitted. After incubation with acetylene, 1 ml of the gaseous phase was taken from each flask, and the amount of ethylene produced was measured by gas chromatography.
To obtain the soluble (cytoplasmic) heterocyst fraction, the heterocyst pellet was resuspended in 5 mmol/L HEPES buffer (pH 8.0) containing 1 mmol/L phenylmethylsulfonyl fluoride (PMSF). The suspension was strongly sonicated (5 × 3 min, 50% duty cycle, 5 output control). The cells were then passed through a French pressure cell (SLM instruments, Inc) at 1,100 Psi 4-5 times. The suspension was centrifuged at 3,000 × g for 30 min at 4°C to separate undisrupted heterocysts. Then, the supernatant was centrifuged at 15,000 × g for 1 hr at 4°C. The supernatant of this last centrifugation step contained the heterocyst cytoplasmic fraction; the pellet consisted of insoluble debris and membranes.
Samples were analyzed by western blotting with polyclonal antibodies raised against the peptide synthesized from the C-terminus of HgdA (NH 2 -CQTKNWLQNTDIQKLVK-COOH). Peptides were synthesized and antibodies were produced by Pineda Antibody-service (Berlin). Rabbit polyclonal antibodies raised against the PII protein of Synechococcus sp. (Forchhammer & De Marsac, 1994) were used as an internal control. After incubation with antibodies against HgdA, washing in PBS buffer containing 0.05% Tween 20 (Carl Roth) and subsequent incubation with PII antibodies, the membrane was washed again and incubated with secondary peroxidase-coupled anti-rabbit IgG antibodies (Sigma A6154). For detection, a Lumi-Light western blotting substrate (Roche) and a Gel Logic 1500 imager (Kodak) were used.

| Overexpression of hgdA and purification of HgdA
The hgdA gene was overexpressed in E. coli Lemo21 (DE3) cells carrying the pIM753 plasmid. Cells were cultivated in 5-L Erlenmeyer flasks containing 1.5 L LB medium at 37°C with continuous shaking at 120 rpm until they reached an OD 600 of 0.6. Gene expression was induced by adding isopropyl β-d-1-thiogalactopyranoside (Carl Roth) at a final concentration of 0.1 mmol/L and incubation of the flasks at 25°C overnight with shaking. After induction, cells were pelleted at 7,000 × g for 15 min at 4°C, and the pellet was resuspended in lysis buffer (20 mmol/L Tris, 200 mmol/L NaCl, 0.5% Triton X-100, pH 7.5) containing 1 mmol/L PMSF and 1 mg/ml lysozyme and incubated at room temperature for 1-2 hr. Then, the solutions were sonicated with a Branson sonifier (3 × 3 min, 50% duty cycle, 5 output control) and centrifuged at 17,000 × g for 30 min at 4°C. The supernatant, which contained extracted soluble proteins, was used for purification of HgdA by affinity chromatography using a Strep-column (IBA-Lifesciences) and Tris buffer (20 mmol/L Tris, 200 mmol/L NaCl, pH 7.5) for equilibration of the column and washing steps; the same buffer containing 2.5 mmol/L desthiobiotin was used for elution. The eluted fractions were pooled and concentrated, and the purity of the HgdA protein was checked by SDS-PAGE.
HgdA was more highly purified and its oligomeric state was estimated by size-exclusion chromatography using an ÄKTA chromatography system and a Superdex 75 10/300 column in Tris buffer (see above). To calculate the molecular masses of the proteins in the eluted peaks, a mixture of standard proteins (Gel Filtration LMW Calibration Kit, GE Life Sciences) was run through the column. The fractions corresponding to different peaks of HgdA purification were pooled, concentrated, and analyzed by SDS-PAGE.
The concentration of pure HgdA protein was determined by the Bradford method using Roti-Quant solution (Carl Roth).
Afterward, the entire sample was used for SDS-PAGE analysis.

| Epimerase activity assay
To test the epimerase activity of HgdA, an established colorimetric glucose oxidase-horseradish peroxidase (GOD-POD) coupled assay was used (Beerens, Soetaert, & Desmet, 2013;Moreno, Rodicio, & Herrero, 1981;Pardeshi, Rao, & Balaji, 2017). In brief, 1 mmol/L UDP-Gal dissolved in 20 mmol/L Tris buffer containing 200 mmol/L NaCl, pH 7.5 was incubated with different amounts of purified HgdA in a total reaction volume of 22 μl at 37°C for 1 or 2 hr. Reactions were stopped and proteins were acid-hydrolyzed with 3.5 μl of 0.4 N HCl at 100°C for 6 min. The mixture was neutralized with 3.5 μl of 0.4 N NaOH. Aliquots (7.5 μl) were taken from each reaction mixture and applied to a 96-well plate. The GOD-POD assay was used to detect released glucose according to the manufacturer's instructions (Sigma-Aldrich). The reaction was stopped and the color was developed by adding 100 μl of 6 N HCl per well. Afterward, the absorbance at 540 nm was read by a TECAN Spark 10M plate reader.

| The hgdA gene product is homologous to NDPsugar epimerases
The hgdA gene is the third gene in the previously described cluster involved in heterocyst formation, all5347/all5346/all5345 (hgdB/ hgdC/hgdA) [ Figure 1a; ]. It encodes a protein of 333 amino acids with a predicted molecular mass of 36.7 kDa. According to the results of a search using the NCBI BLAST tool, HgdA is a putative nucleoside-diphosphatesugar epimerase belonging to the NAD-dependent epimerase/dehydratase family of the short-chain dehydrogenases/reductases (SDR) superfamily. An additional in silico search for homologs of HgdA using the PaperBLAST tool, which searches for homologs of a given protein in published articles (Price & Arkin, 2017), revealed SDRs, including several epimerases, similar to HgdA ( Figure   A1). However, according to the phylogenetic tree built by the web service Phylogeny.fr (Dereeper et al., 2008;, based on multiple sequence alignments of selected HgdA homologs found by PaperBLAST, HgdA is more closely related to epimerases ( Figure 1b).

| HgdA protein localizes specifically to heterocysts
To study the function of HgdA, we used semi-quantitative RT-PCR to mature heterocysts. GFP fluorescence was equally distributed within the heterocyst and sometimes formed small foci (Figure 2b).
This observation is in line with the prediction that HgdA is an epimerase, which is a soluble enzyme.
In western blot analysis, a protein cross-reacting with the HgdAspecific antibody was only visible in the sample obtained from the soluble heterocyst fraction, but not in the vegetative cell fraction

| The hgdA gene is essential for diazotrophic growth
To investigate the function of the hgdA gene in more detail, we created a mutant of this gene in Anabaena sp. by inserting an antibiotic resistance gene via homologous recombination (Figure 1a, A2a). The mutant, SR695, was completely segregated ( Figure A2b). In medium with a combined nitrogen source, mutant SR695 did not differ from the wild-type in cell and filament morphology or in growth.
However, the mutant was not able to grow diazotrophically, even though the filaments formed heterocysts after nitrogen stepdown ( Figure 3a,b,d). The mutant SR695 was complemented by introducing the self-replicating vector pIM612 (Bornikoel, 2018) carrying the full-length hgdA sequence under control of the P glnA promoter (Valladares et al., 2004). The complemented mutant (SR695c) clearly grew better than the mutant SR695 at 7 days after nitrogen stepdown ( Figure 3a).

| The aberrant cell envelope of heterocysts of mutant SR695 cannot provide microoxic conditions for nitrogenase activity
We investigated whether heterocysts of the mutant SR695 provide microoxic conditions necessary for nitrogenase activity. We incubated mutant and wild-type cultures with triphenyl tetrazolium chloride (TTC) (Fay & Kulasooriya, 1972) and observed the dark crystals of reduced TTC only in wild-type heterocysts (Figure 3b). Lack of dark TTC crystals in mutant heterocysts indicated that their inner environment was oxic.
Mutants with defects in heterocyst envelope layers only have nitrogenase activity when incubated under anoxic conditions (Ernst et al., 1992). We assayed nitrogenase activity under oxic and anoxic conditions based on the measurement of acetylene reduction . The mutant SR695 had nitrogenase activity only under anoxic conditions, whereas the wild-type had nitrogenase activity under both oxic and anoxic conditions ( Figure 3c).
Hence, the mutant heterocysts did not provide the microoxic conditions required for nitrogenase activity.
We analyzed the heterocyst envelope in more detail. We were able to detect both cell layers These observations are comparable with those of the hgdA mutant FQ1647 . We did not find any structural differences in the hep layer or in the ultrastructure of vegetative cells between wild-type and mutant SR695.
We analyzed the glycolipid composition of mutant and wild-type hgl layers using TLC of methanol extracts of both strains after nitrogen stepdown. We analyzed the content at two temperatures (20 and 28°C) because the ratio of the major HGL forms can vary at different temperatures (Bauersachs, Stal, Grego, & Schwark, 2014;Wörmer, Cires, Velazquez, Quesada, & Hinrichs, 2012). Both wildtype and mutant extracts contained the major HGLs, HGL 26 keto-ol and HGL 26 diol (Perez, Wörmer, Sass, & Maldener, 2018) at both temperatures ( Figure 4). However, the diol:keto-ol ratio of wild-type and mutant SR695 differed. At 28°C, the diol:keto-ol ratio of wild-type heterocysts was higher than that of the mutant (Figure 4). At 20°C, the wild-type contained more of the keto-ol form than at 28°C.
Nevertheless, the wild-type was still different from the mutant, which did not show a temperature dependent ratio change ( Figure 4).
As previously reported, a mutant in the upstream gene hgdB shows a similar phenotype at 28°C . We also found that at 20°C, differences in growth between wild-type and mutant SR695 were not as prominent as at 28°C (Figure 4, lower panels).

| The protein HgdA is soluble and forms dimers in vitro
For the biochemical characterization of the HgdA protein, we overexpressed the gene in E. coli and purified the protein by affinity chromatography, followed by size-exclusion chromatography. The major peak of HgdA in the size-exclusion chromatography elution profile corresponded to the dimeric form; additional peak shoulders, probably representing monomeric and other oligomeric forms of HgdA, were also present ( Figure 5a).
On SDS-polyacrylamide gels, the band of purified HgdA consisted of the monomeric form. However, when purified HgdA was incubated with the amino-reactive cross-linker BS 3 , which forms covalent bonds between interacting proteins, also dimeric and other oligomeric forms were detected (Figure 5b).
We modeled the structure of HgdA using the Swiss model online

| HgdA fulfills a UDP-galactose 4-epimerase function in vitro
Based on the sequence similarity of HgdA to UDP-galactose 4epimerase, we tested whether HgdA converts UDP-galactose to UDP-glucose (Moreno et al., 1981). The enzyme catalyzed the conversion at a rate of approximately 30-40 nmol min −1 nmol HgdA −1 depending on the protein concentration. UDP-glucose production by HgdA increased when higher concentrations of the enzyme were used; the activity was considerably lower when tested at 99°C (Table 1, Figure A3).
UDP-galactose 4-epimerases use NAD as a cofactor, which is constantly bound in the conserved cofactor-binding glycine-rich site in the Rossmann fold (Allard et al., 2001;Beerens et al., 2015;Bellamacina, 1996;Rossmann, Moras, & Olsen, 1974). However, we were unable to extract or detect NAD from the enzyme using standard protocols (Creuzenet, Belanger, Wakarchuk, & Lam, 2000). In place of the conserved NAD-binding motif GXXGXXG, the HgdA sequence has a GIDEFIG motif, with the second glycine replaced by glutamate ( Figure A1). An NCBI BLAST search showed that such a motif is also found in HgdA homologs in several other cyanobacteria.

| D ISCUSS I ON
One of the main events in heterocyst maturation is the formation of the heterocyst-specific envelope. A variety of enzymes participate in this process, including those that are responsible for the synthesis and transport of the envelope components Fiedler et al., 1998;Huang et al., 2005 Staron et al., 2011). In this study, we investigated the function of the putative epimerase HgdA (Figure 1) in heterocyst formation. Our results partially confirmed previous findings , and in addition described the enzymatic activity of HgdA.
Transcripts of hgdA were found only when the heterocysts were almost completely mature (24-48 hr after nitrogen stepdown; Figure 2a). These time points were markedly later than activation of the devB gene, which encodes the membrane fusion component of the efflux pump transporting HGLs (Fiedler et al., 1998;Staron et al., 2011). However, the expression patterns of the hgdB and hgdC genes  are similar to that of hgdA, which indicates that products of the hgdBCA gene cluster are formed at the same time even if they probably do not comprise an operon since they were complemented separately  and that the proteins might work together. The localization of HgdA in the cytoplasm of mature heterocysts (Figure 2b, c) confirms its specific importance for these differentiated cells and demonstrates that HgdA is a soluble protein, as expected from the in silico analysis of its sequence.
Our mutant SR695, like the transposon-insertion mutant of this gene  showed a Fox − phenotype, that is, the inability to grow diazotrophically under oxic conditions. Since nitrogenase activity was detectable under anoxic conditions, this mutant shows a phenotype, which is specific for mutants with an impaired heterocyst envelope (Figure 3a-c).
Although the hgl layer was present in the mutant SR695 ( Figure 3d), its defect allowed oxygen to enter the heterocyst. The main difference between the mutant and wild-type was in the HGL composition ( Figure 4). Specifically, the aberrant ratio of the two major HGLs in the mutant, with an excess of HGL 26 keto-ol, seemed to be critical for hgl layer formation and heterocyst function at 28°C.
The same aberrant HGL ratio, which causes a Fox − phenotype, in an hgdB mutant has been found ; this finding along with the results of our expression studies might indicate that products of the hgdB, hgdC, and hgdA genes work cooperatively. Most UDP-galactose 4-epimerases form dimers or other oligomers (Allard et al., 2001), but they can also function in a monomeric state (Nayar  Note. Glucose oxidase-horseradish peroxidase assay of UDP-galactose 4-epimerase activity of recombinant HgdA. HgdA at the concentration of 10 µmol/L was incubated with 1 mmol/L UDP-galactose at indicated temperatures for 2 hr, and UDP-glucose production was measured. Bovine serum albumin (BSA) was used as negative control. Shown is the relative enzymatic activity in arbitrary units (AU) with the standard deviation of indicated experimental replicates.
& Bhattacharyya, 1997). Our size-exclusion chromatography, crosslinking, and modeling results indicated that the main active states of HgdA are probably dimers ( Figure 5).
Purified HgdA has typical UDP-galactose 4-epimerase activity in vitro (Table 1). Compared to other known UDP-galactose 4-epimerase activities, this activity was in the lower range, with some epimerases having activities several times less and others hundreds of times more than that of HgdA (Agarwal, Gopal, Upadhyaya, & Dixit, 2007;Chung, Ryu, & Lee, 2012;Guevara, El-Kereamy, Yaish, Mei-Bi, & Rothstein, 2014;Pardeshi et al., 2017;Shin et al., 2015). Since we were unable to detect or extract NAD from the purified active HgdA protein, we assume that the altered NAD-binding sequence, with glutamate replacing glycine, captures more tightly NAD.
The function of an epimerase in HGL synthesis has not been described so far. But based on our results, we suggest that HgdA converts UDP-galactose, which could derive from thylakoid degradation in heterocysts, to UDP-glucose. By a still unknown mechanism, the different activated sugar epimers determine the ratio between HGL diol and HGL keto-ol.
Nevertheless, other HGL diol biosynthetic pathways independent of HgdA must be present because mutant SR695 heterocysts contain a small amount of the HGL diol (Figure 4), which is not sufficient to support heterocyst function. The similarity of the phenotypes of the hgdA mutant and hgdB mutant  suggests that HgdA and HgdBC closely cooperate, but further investigation is required.
Altered HGL diol:keto-ol ratios have been described in other situations. For instance, during growth at higher temperatures, cyanobacteria produce higher amounts of HGL diols (Bauersachs et al., 2014;Wörmer et al., 2012), which might protect heterocysts from gas penetration under these conditions. When HGL keto-ols are prevalent and the amount of HGL diols is lower, the heterocyst cell envelope might lose its gas tightness at higher temperatures; at lower temperatures, when amounts of the keto-ol form increase, the envelope retains its gas tightness.
The deposition of the HGLs in the wild-type and in the hgdA or hgdB mutant ) differs ( Figure 6). The wild-type forms a normal hgl layer around the entire heterocyst using two exporter systems, namely DevBCA-HgdD (Staron et al., 2011) mostly at the polar neck regions and HgdBC-HgdD  at the lateral sides. In the hgdB mutant, the hgl layer is replaced by an amorphous layer at the lateral sides because the HgdBC transporter is lacking, and the hgl layer is thicker at the polar regions because of excess substrate for DevBCA-HgdD (HGLs that are not transported by HgdBC-HgdD but are still synthesized). In the hgdA mutant, both transporters are present, but because HGL production is deficient, the hgl layer is much thinner than in the wild-type.   has a thicker hgl layer at the polar regions but it is replaced by an amorphous unstructured layer at the lateral sides of the cell. Green line, cellular membranes and cell wall. Lower panels show the putative interplay between the different means of HGL synthesis and transport involving HgdA and transporters DevBCA-HgdD (Staron et al., 2011) or HgdBC-HgdD . Arrows from HgdA show the route of HGLs, produced with help of HgdA, to a transporter (DevBCA-HgdD or HgdBC-HgdD). Arrows from the question marks indicate routes of HGLs produced independently of HgdA (probably at earlier stages of heterocyst development) point, and a transporter composed of HgdBC and HgdD (TolC homolog) might export HgdA products directly or sequentially to the heterocyst cell envelope.

ACK N OWLED G EM ENTS
We thank Claudia Menzel for electron microscopy sample preparation, Thomas Härtner for help with gas chromatography, Oliver Betz for access to the TEM, Ritu Garg for help with size exclusion chromatography of standard proteins, and Karl Forchhammer for fruitful discussions and valuable advice. We also thank Karen Brune for critically reading the manuscript and improving the text linguistically. This work was supported by Deutsche Forschungsgemeinschaft (SFB766 and GRK1708).

CO N FLI C T O F I NTE R E S T S
The authors declare no conflict of interest.

AUTH O R S CO NTR I B UTI O N
DS and IM designed the experiments; DS and CNN performed experiments; DS and IM analyzed the data and wrote the manuscript; all authors read the final manuscript; IM supervised the project.

E TH I C S S TATEM ENT
None required.

DATA ACCE SS I B I LIT Y
The data will be available on request from the corresponding author. Arrows with numbers, primers used for the genotypic analysis of the mutants (see also \ Table A2). (b) Segregation of the hgdA mutant analyzed by PCR. SR695: mutant SR695; wt: wild-type; SR695c: SR695 mutant complemented with the hgdA gene. Primer numbers correspond to those depicted in A F I G U R E A 1 COBALT multiple alignment of several characterized HgdA homologs found using the online tool PaperBLAST. Red, highly conserved residues; blue, less conserved residues; gray, not conserved residues; yellow box, conserved GXXGXXG NAD(P) binding site; violet boxes, residues of the conserved S(X) 24 -Y(X) 3 K catalytic triad