Introduction

Cylindrospermopsin (CYLN) is an alkaloid that is produced by various cyanobacteria, including Cylindrospermopsis raciborskii, Aphanizomenon ovalisporum, Anabaena bergii, Umezakia natans, and Raphidiopsis curvata (Banker et al. 1997; Harada et al. 1994; Li et al. 2001; Schembri et al. 2001). It interferes with multiple metabolic pathways, causing hepatotoxicity (Hawkins et al. 1985), general cytotoxicity (Runnegar et al. 1994), and neurotoxicity (Kiss et al. 2002), while it may also induce carcinogenesis (Humpage et al. 2000). Human poisoning by CYLN has been reported from Australia after the ingestion of contaminated drinking water and is colloquially known as the “Palm Island Mystery Disease” (Bourke et al. 1983), “Barcoo Disease,” or “Belyando Fever” (Hayman 1992). Although incidents of CYLN poisoning have been limited to Australia so far (Saker et al. 1999b), the producer organisms are recent and highly invasive neophytes in temperate and tropical regions all over the world (Neilan et al. 2003). CYLN thus represents a public health concern on a global scale.

The molecular structure, biosynthetic pathway, and putative biosynthesis enzymes of CYLN have been recently described (Burgoyne et al. 2000; Ohtani et al. 1992; Schembri et al. 2001; Shalev-Alon et al. 2002). As depicted in Fig. 1a, the molecule consists of a sulfated, tricyclic backbone that contains a central guanidino moiety and is linked to hydroxymethyluracil (Ohtani et al. 1992). There are two naturally occurring stereoisomers, CYLN and 7-epi-CYLN, depending on the position of the hydroxy group on C-7, while the hydroxy moiety is missing in 7-deoxy-CYLN. CYLN is assembled by successive condensations of five intact acetates to a guanidinoacetate starter unit, followed by tailoring reactions, such as C-methylation, ketoreductions, sulfation, cyclizations, and possibly a carbamoylation to complete the biosynthesis. The guanidinoacetate is the product of a transamidination of glycine from an as yet unidentified guanidino donor (Burgoyne et al. 2000).Three putative biosynthetic genes, aoaA, aoaB, and aoaC, encoding an amidinotransferase, a hybrid nonribosomal peptide synthetase/polyketide synthase, and a polyketide synthase, respectively, have been identified in A. ovalisporum ILC-146 (Fig. 1b) (Schembri et al. 2001; Shalev-Alon et al. 2002). It is thought that AoaA catalyzes the synthesis of guanidinoacetate, which is recruited by AoaB for the successive polyketide extension by AoaC and further polyketide synthase modules (Fig. 1a).

Figure 1
figure 1

(A) Biochemical reactions that three putative cylindrospermopsin biosynthesis gene products AoaA, AoaB, and AoaC are thought to catalyze and (B) arrangement of aoaA, aoaB, (partial gene) and aoaC (partial gene) in A. ovalisporum. The light and dark gray shadings indicate nonribosomal peptide synthetase and polyketide synthase modules, respectively. The brackets with primer names indicate the regions that were amplified and sequenced for phylogenetic analysis. A, aminoacyl adenylation; ACP: acyl carrier protein; AT, acyl transferase; KR: ketoreductase; KS, β-ketosynthase.

Full size image

To date, no functional or evolutionary studies have been carried out on AoaA, AoaB, and AoaC. In the present study we investigated the molecular mechanisms of precursor recruitment for CYLN biosynthesis by AoaA and AoaB using three-dimensional protein modeling. In addition, we determined and compared the molecular phylogenies of CYLN producing cyanobacteria from Australia, Japan, and Israel to those of their putative CYLN biosynthetic genes and discuss the evolution of toxigenicity in these microorganisms.

Materials and Methods

Cyanobacterial Strains and Culture Conditions

Cyanobacterial strains used in the present study (Table 1) were grown in Jaworski medium (Thompson et al. 1988) in static batch culture at 26°C under continuous illumination (10 μmol m−2 s−1).

Table 1 Cyanobacteria strains used in the present study
Full size table

DNA Extraction

DNA techniques were carried out according to standard procedures (Sambrook et al. 1989). Total genomic DNA was extracted from cyanobacterial cells by lysozyme/SDS/proteinase K lysis following phenol–chloroform extraction as described previously (Neilan 1995). DNA in the supernatant was precipitated with 0.6 volumes isopropanol, washed with 70% ethanol, dissolved in TE buffer (10:1), and stored at –20°C.

PCR Amplification and Sequencing

Fragments of the putative CYLN biosynthesis genes, aoaA, aoaB, and aoaC (Fig. 1), and 16S rRNA gene were amplified by the polymerase chain reaction (PCR) using the primer pairs listed in Table 2. PCR was performed in 20-μl reaction volumes containing 1× Taq polymerase buffer, 2.5 mM MgCl2, 0.2 mM deoxynucleotide triphosphates, 10 pmol forward and reverse primers, between 10 and 100 ng genomic DNA, and 0.2 units Taq polymerase (Fischer Biotech, Perth, Australia). Thermal cycling was performed in a GeneAmp PCR System 2400 Thermocycler (Perkin Elmer Corp., Norwalk, CT, USA). Cycling began with a denaturing step at 94°C for 3 min followed by 30 cycles of denaturing at 94°C for 10 s, primer annealing at 50°C or 55°C for 20 s, and a DNA strand extension at 72°C for 1 min. Amplification was completed by a final extension step at 72°C for 7 min. DNA was separated by agarose gel electrophoresis in TAE buffer (40 mM Tris–acetate, 1 mM EDTA, pH 7.8) and visualized by UV translumination after staining in ethidium bromide (0.5 μg/ml).

Table 2 PCR primers used in the amplification and sequencing of 16S rDNA, putative amidinotransferase (AMT), peptide synthetase A-domain (NRPS), and polyketide synthetase β-ketosynthase domain (PKS) genes (the length of PCR products and the 5′-binding sites are given in parentheses)
Full size table

Automated DNA sequencing was performed using the PRISM Big Dye cycle sequencing system and a model 373 sequencer (Applied Biosystems Inc., Foster City, CA, USA). Sequence data were analyzed using ABI Prism-Autoassembler software, and percentage similarity and identity to other translated sequences determined using BLAST in conjunction with the National Center for Biotechnology Information (NIH, MD). Accession numbers are presented in the legends to Figs. 6a to 6d.

Phylogenetic Analysis

DNA sequences were aligned using the multiple alignment tool from Clustal X (Thompson et al. 1997). Sequence alignments were manually confirmed. Phylogenetic trees were reconstructed from a pairwise distance matrix (Jukes and Cantor 1969) using neighbor-joining (Saitou and Nei 1987). All alignments were bootstrapped with 1000 resampling events. Alignments were also applied to parsimony and maximum likelihood programs of the Phylip package version 3.6 (Felsenstein 1989). Phylogenetic trees and sequence alignments were reproduced using the software NJPlot (Perrière and Gouy 1996) and TeXshade (Beitz 2000), respectively.

Three-Dimensional Protein Modeling

A suitable template for the three-dimensional modeling of AaoA was determined by a PSI-BLAST search of the nonredundant PDB database using BLOSUM62 with 10 iterations (http://www.bioinformatics.ljcrf.edu/pdb_blast/). A structural alignment was obtained by sequence-structure comparison (FUGUE) of AoaA against the HOMESTRAD database (Shi et al. 2001). The protein was modeled using the interactive three-dimensional graphics programs, O (Jones et al. 1991) and LSQMAN (Kleywegt 1996). The AoaA model was prepared, using the human arginine:glycine amidinotransferase (GATM) as a template structure (PDB accession no. 1JDW), by manually replacing each residue of GATM with the corresponding amino acid of AoaA, as determined by the structural alignment described above. Rotamer libraries were used to replace side chains. Equivalent rotamers were used when replacing a side chain unless it resulted in a steric clash. The final model, consisting of residues 4 to 372, was verified using Verify3D (Luthy et al. 1992). Substrate arginine was placed into the catalytic site of the AoaA model by using the atom coordinates of substrate arginine in GATM (PDB accession no. JDW4). Graphic representations of protein models were prepared using the software Pymol (Molecular Graphics System, deLano Scientific LLC).

Extensive modeling of nonribosomal peptide synthetase adenylation domains (A-domains) has been carried out (Challis et al. 2000; Stachelhaus et al. 1999). Eight topologically conserved amino acid residues determine the substrate specificity and were identified for the A-domain of AoaB as described elsewhere (Challis et al. 2000; Stachelhaus et al. 1999). The substrate determining residues were compared to an assigned database of residues, where the substrate for the corresponding A-domain has been determined experimentally, and an unassigned database, where the substrate is putative or unknown (Challis et al. 2000).

Results and Discussion

Functional Modeling of AoaA

Burgoyne et al. (2000) have demonstrated that guanidino acetate is the starter unit in cylindrospermopsin biosynthesis, however, they failed to demonstrate how guanidino acetate is produced. While the authors demonstrated that an intact molecule of glycine is incorporated into guanidino acetate, no incorporation of label was achieved in feeding experiments with ubiquitously 13C- and 15N-labeled arginine. Guanidino acetate is found in a wide variety of organisms and is the product of amidinotransferases, a monophyletic enzyme family that is highly conserved in both prokaryotic and eukaryotic organisms (Bedekar et al. 1998). While the natural guanidino acceptor varies in amidinotransferases from different organisms, the preferred guanidino donor is arginine in all cases investigated (Bedekar et al. 1998; Lee et al. 2002; Srivenugopal and Adiga 1980). Considering the high level of conservation in the structure and function of amidinotransferases, it is surprising that arginine should not be the guanidino donor in the biosynthesis of guanidino acetate in C. raciborskii. On the other hand, the failure to achieve incorporation of label from arginine into cylindrospermopsin in the study by Burgoyne et al. (2000) is not conclusive evidence that arginine is not the natural substrate in cylindrospermopsin biosynthesis. Reasons for this experimental failure can be manifold. For example, not all cyanobacteria provide basic amino acid transporter systems (Montesinos et al. 1997), and evidence for the uptake of arginine by C. raciborskii was not presented.

AoaA provided the highest amino acid sequence similarity (PSI-BLAST of PDB: 40% identities/56% similarities) to the human arginine:glycine amidinotransferase, GATM (PDB accession no. 1JDW). A FUGUE search revealed high structural similarity among AoaA, GATM, and the Streptomyces griseus arginine:inosamine phosphate amidinotransferase, StrB1 (PDB accession no. 1BWD) (FUGUE Z-score = 42.29). As shown in the structural alignment in Figure 2, residues forming hydrogen bonds or salt bridges with the substrate arginine (D170, M302, R305, R322, S354, S355) and those forming the catalytic triad (D254, H303, C407) were completely conserved in GATM, AoaA, and StrB1, with the exception that M302 was substituted by serine in AoaA and deleted in StrB1. In contrast, substrate glycine binding residues were only conserved in GATM and AoaA, and not in StrB1, which utilizes inosamine phosphate as the natural guanidine acceptor (Fritsche et al. 1998).

Figure 2
figure 2

Structural alignment of amidinotransferases. AoaA, A. ovalisporum amidinotransferase; AT, human amidinotransferase; StrB1, S. griseus amidinotransferase. (□), nonconserved; (), similar; (), conserved; (■), all match. Bracket with arrow indicates β-sheet; bracket indicates α-helix. H9, helix H9; flp, 300-flap. Letters above alignment indicate substrate binding residues of AT; letters below alignment indicate substrate binding residues of StrB1. G, glycine binding residue; I, inosamine phosphate binding residue; R, arginine binding residue. Amino acids in italics are catalytic residues.

Full size image

Two secondary structures, helix H9 and 300-flap, are involved in conformational changes of GATM during catalysis (Fritsche et al. 1999). Upon entering the catalytic channel, substrate arginine sterically clashes with N300, displacing the 300-flap toward helix H9, which in turn is displaced to avoid the sterical collision of P299, H236, and N300 with R235. The regions corresponding to H9 and the 300-flap were structurally conserved in AoaA, but their amino acid sequences differed from GATM. In particular, P299 has been replaced by isoleucine, and N300 by phenylalanine.

The three-dimensional structure of AoaA was modeled using the crystal structure of GATM (1JDW) as a template. AoaA contained nine insertions (C105, C148, P149, V238, H239, Q270, E271, D295, E296), which were not modeled. The quality of the model was verified by plotting the mean 3DVerify score of both 1JDW and the AoaA model versus the residue number (Fig. 3). The mean score of all residues in the AoaA model (0.3778) was only slightly lower than that of the template (0.5051), indicating that the majority of residues were folded correctly. Two regions of the model (A42 to K52 and G226 to P238) provided residues with negative scores and were probably folded incorrectly. Whereas the first region was located on the surface of the protein and probably has only structural function, the second region included the 300-flap. Another misfold not evident from the 3DVerify score occurred near helix H9. The last four residues of the loop leading up to H9, PIHS in GATM, were replaced in AoaA by functionally different amino acids, WEWP. In the model, the two hydrophobic tryptophans were exposed to the solvent at the surface of the enzyme, while the charged glutamate was buried and the proline was twisted into an unlikely rotamer. This misfold had flow-on effects up to residue 250, which included the entire helix H9.

Figure 3
figure 3

Verify 3D mean score plotted versus residue number. Abbreviations as explained in the text. (—) Mean score of 1JDW; (—) mean score of AoaA model.

Full size image

In GATM, substrate arginine forms 11 hydrogen bonds to amino acid residues M302, H303, R322, S354, S355, and C407. All residues were 100% conserved in sequence and topology in the AoaA model, apart from M302, which was replaced by serine (S247). This replacement is a significant change, as a large hydrophobic amino acid was exchanged with a small polar amino acid, and its effect on substrate specificity can only be speculated on. The topolgies of substrate binding and catalytic residues of AoaB are shown in Figure 4. In GATM, substrate arginine forms a hydrogen bond between its α-amino group and the main chain carboxylate of M302. The residue corresponding to M302 has been deleted in StrB1, and substrate arginine forms a hydrogen bond with the main chain carboxylate of the adjacent T226 (Fritsche et al. 1998). It cannot be excluded that S247 in AoaA performs a similar function as M302 and T226 in GATM and StrB1, respectively.

Figure 4
figure 4

Topology of substrate binding and catalytic residues with arginine in the catalytic site of AoaA.

Full size image

Substrate glycine is held in the catalytic site of GATM by hydrogen bonds from its carboxylate to the side chains of H303 and via a water molecule to Y164, while the amino group is hydrogen-bonded to the side chain of S354. These residues were 100% conserved in both sequence and topology in the AoaA model.

Due to regional errors involving helix H9 and the 300-flap, it was not possible to determine whether AoaA would undergo the same substrate-induced conformational changes as GATM. This is, however, not essential to the function of the enzyme but is thought to increase the catalytic rate by aiding in the release of ornithine from the catalytic site (Fritsche et al. 1999).

The high structural similarity of the AoaA model to GATM, in addition to the high conservation of substrate binding residues, suggested that AoaA may catalyze the transamidination of glycine from arginine. Biochemical studies on AoaA will be required to verify this claim.

Upon synthesis, guanidino acetate is thought to be recruited by the hybrid NRPS/PKS, AoaB, for polyketide extension (Burgoyne et al. 2000). So far, no other NRPS has been described that utilizes guanidino acetate as a substrate. The eight substrate specificity conferring residues of the AoaB A-domain (DFHFITHD), determined as described by Challis et al. (2000), did not closely match any other A-domain from either the assigned or the unassigned databases (http://www.raynam.chm.jhu.edu/∼nrps/). The most similar A-domains activated proline, methyl-proline, and pipecolic acid (Fig. 5), all of which provide modified α-amino groups. Unfortunately, the overall sequence similarity of the AoaB A-domain to other A-domains with a known crystal structure was too low (<40%) for reliable modeling of its three-dimensional structure, and it could not be determined how the eight substrate binding residues described above may interact with the guanidino acetate substrate. One interesting feature of AoaB is that the eighth substrate binding residues, which in other A-domains may be one of the hydrophobic amino acids leucine, isoleucine, or valine, consisted of an aspartate (D362). Like other A-domains, AoaB provided a conserved lysine (K558), which in GrsA forms a hydrogen bond to the α-carboxylate of the substrate amino acid (Conti et al. 1997). AoaB also provided a conserved aspartate (D258), the side chain of which forms a second hydrogen bond to the α-amino group of the substrate in GrsA. Using the atom coordinates of the substrate and substrate binding residues of GrsA, and replacing these residues with the corresponding residues of AoaB, a simple model of the catalytic site was prepared (data not shown). In this model, guanidino acetate was held in the catalytic site by hydrogen bonds between its α-carboxylate and the K558 side chain, between its imino nitrogen and the D258 side chain, as well as between its guanidino amine and the side chain of D362. Other residues, such as T330 and H262, may also play a role in substrate binding. A crystal structure is required to verify this proposal.

Figure 5
figure 5

Alignment of substrate binding residues of NRPS adenylation domains. (□), Nonconserved; (), similar; (), conserved; (■), all match. Access codes are given, followed by the protein name, adenylation module number, and three-letter amino acid code of the substrate amino acid. 3h4mPhe, 3-hydroxy-4-methoxyphenylalanine; mePro, methylproline; DAB, diaminobutyric acid; DhbF, 2,3-dihydroxybenzoylglycine synthetase; NosA/C, nostopeptolide synthetase; SafA, saframycin synthetase; Cda1/2, calcium-dependent antibiotic; GrsA, gramicidin synthetase; PvdD, pyoverdin synthetase; McyC, microcystin synthetase; PstB, peptide synthetase; TyrB, tyocidin synthetase; SnbD, pristinamycin/virginiamycin synthetase; Pps4, fengycin synthetase; RapP, rapamycin synthetase; FkbP, FK506 synthetase; PvsA, pyoverdin synthetase A; AoaB, cylindrospermopsin synthetase.

Full size image

Molecular Phylogeny of CYLN Producing Cyanobacteria

The molecular phylogeny of CYLN producing cyanobacteria was investigated using partial 16S rDNA gene sequences (954 bp). A phylogenetic tree of the 16S rDNA from CYLN producing strains and reference strains is given in Fig. 6a. As already reported elsewhere (Neilan et al. 2003), C. raciborskii strains group closely together and form a separate branch within the order Nostocales. In contrast to their morphological classification, a close relationship among U. natans, A. bergii, and A. ovalisporum became apparent, and positioned these strains within the same group as Anabaenopsis and Cyanospira (Fig. 6a). Neilan et al. (2003) reported that U. natans has a distant phylogenetic position to other cyanobacteria, however, on closer inspection, the 16S rDNA determined in that study may have been a PCR hybrid. In the present study, the sequence similarity among U. natans, the Australian and Israeli isolates of A. ovalisporum, Anabaenopsis, and Cyanospira was at least 98%, while that of A. bergii was slightly lower (97%). The Israeli isolate A. ovalisporum ILC-146 is not included in Figure 6a, since only a short fragment (479 bp; AF044269) was available from the database. This fragment was, however, 99% identical to the corresponding region of the A. ovalisporum APH028 16S rDNA. Other members belonging to the genera Aphanizomenon and Anabaena were located on a branch well separated from the above stains, and provided a sequence similarity of 93% or less to the above strains. Unfortunately, no data or culture was available for the CYLN-producer strain Raphidiopsis curvata, and therefore its phylogenetic relationship to other CYLN producers could not be established. Phylogenetic trees reconstructed by parsimony and maximum likelihood methods did not differ in their topology and geometry from those produced by the neighbor-joining method.

Figure 6a
figure 6

Phylogenetic affiliations of selected filamentous, heterocyst forming cyanobacteria derived from partial 16S rDNA gene sequences. The phenogram was reconstructed from a pairwise distance matrix (Jukes and Cantor 1969) using the neighbor-joining method (Saitou and Nei 1987). The scale represents the number of substitutions per 100 nucleotides. Bootstrap values (1000 resampling cycles) above 500 are shown and represent the statistical significance at each node. Cylindrospermopsin producing strains are in boldface. Accession numbers: AY897614, AF516724, AF516729 to AF516736, AF516738, AF516743 to AF516747, AF247577, AJ133154, AF092504, AF067819, AJ781131, AF160256, AY038033, AF044270, AY494701, AB039003, AY513505, AJ309733, AY038036, AF027655.

Full size image

Filamentous heterocyst-forming cyanobacteria comprise a monophyletic group, but the generic and specific assignments of its members are based on classical morphological definitions, and often do not reflect their true phylogeny (Iteman et al. 2002). Apart from Cylindrospermopsis (Neilan et al. 2003), the phylogeny of CYLN producers has not been previously examined by molecular methods. As evident from the present study, the CYLN-producer strains A. ovalisporum and Anabaena bergii may have been taxonomically misclassified. Iteman et al. (2002) divided filamentous heterocyst-forming cyanobacteria into six subgroups. Two of these groups, group IV, containing Anabaenopsis/Cyanospira, and group V (Nodularia), provide a high tolerance to elevated levels of salinity, pH, and temperature (Iteman et al. 2002). This trait is also shared by the Australian isolate Aphanizomenon ovalisporum APH028A, a bloom of which was attributed to the high salinity and alkalinity of the water (Shaw et al. 1999). No physiological data were available for A. ovalisporum ILC-146, U. natans, and A. bergii 283A. Based on shared physiological features and the high similarity in rDNA and ITS sequences, Iteman et al. (2002) suggested that Cyanospira and Anabaenopsis should be assigned to a single genus. The close relationship of U. natans, A. ovalisporum, and Anabaena bergii 283A to Anabaenopsis/Cyanospira suggests that they may have to be included in such a revised genus.

Phylogeny of CYLN Biosynthesis Genes

Twenty-one cyanobacterial strains, nine of which produced CYLN (Table 1), were screened by PCR for the presence of putative CYLN biosynthesis genes, using the primer pairs listed in Table 2. Positive PCR products were sequenced using the same primers. As depicted in Figure 1, primer set CYLATF/R amplified almost the entire aoaA gene (1105 of 1179 bp), while CPSF/R amplified a 478-bp fragment of the aoaB A-domain, which included the eight substrate specificity conferring residues. Finally, the primer set A205PKF/R was used to amplify part (514 bp) of the β-ketosynthase domain of aoaC.

All three putative CYLN biosynthesis genes, aoaA, aoaB, and aoaC, were exclusively present in CYLN producers, with one exception. The Hungarian strain, C. raciborskii Hung1, a nonproducer of CYLN, possessed an aoaB homologue but no homologues of aoaA or aoaC. Corresponding homologues of aoaA, aoaB, and aoaC in the different strains and species examined in the present study shared at least 95% nucleotide sequence identity. Multiple sequence alignment divided the corresponding aoaA, aoaB, and aoaC homologues into two distinct groups (Figs. 6b to 6d), one containing sequences exclusively from Cylindrospermopsis and the other containing sequences from Aphanizomenon, Umezakia, and Anabaena, with one exception. The aoaB homologue of C. raciborskii Hung1 grouped with those of Aphanizomenon/Umezakia/Anabaena (Fig. 6b). No DNA of A. bergii 183A was available at the time of the present study, however, a fragment of a NRPS sequence (AF170843) outside the region investigated in the present study showed greater similarity (99%) to that of Aphanizomenon than to Cylindrospermopsis (96%).

Figure 6b
figure 7

Phylogenetic tree of amidinotransferase gene sequences from CYLN producing cyanobacteria. The phenogram was reconstructed as described in the legend to Fig. 6a. Cylindrospermopsin producing strains are in boldface. Accesssion numbers: AY897605 to AY897613, AF395828, and AAM48710.

Full size image
Figure 6c
figure 8

Phylogenetic tree of nonribosomal peptide synthetase gene sequences from CYLN producing cyanobacteria. The phenogram was reconstructed as described in the legend to Fig. 6a. Cylindrospermopsin producing strains are in boldface. Accession numbers: AY89790 to AY89797, AAF15891, AAM33468, AAF76934, AF204805.

Full size image
Figure 6d
figure 9

Phylogenetic tree of polyketide synthase genes from CYLN producing cyanobacteria. The phenogram was reconstructed as described in the legend to Fig. 6a. Cylindrospermopsin producing strains are in boldface. Accession numbers: AY897598 to AY897604, AAF76933, AAG38956, AAM33470, CAD19091, AAG38958.

Full size image

The phylogenetic grouping of aoaA, aoaB, and aoaC homologues followed that of the 16S rRNA genes of the producer organisms, however, the phylogenetic distance of corresponding aoaA, aoaB, and aoaC homologues between the Cylindrospermopsis and the Aphanizomenon/Umezakia/Anabaena group, as well as within the latter group, was smaller than that of their 16S rRNA genes. The sequence similarity between aoaA, aoaB, and aoaC homologues within Cylindrospermopsis strains and within the Aphanizomenon/Umezakia/Anabaena group ranged from 98 to 100% at the nucleotide level, while the similarity between the two groups was still 95 to 98%. The 16S rDNA sequence similarity within the Cylindrospermopsis group was 99%, but only 95 to 97% within the Aphanizomenon/Umezakia/Anabaena group and 91 to 92% between the two groups. Considering that 16S rDNA is one of the most conserved genes (Honda et al. 1999), the discrepancy in the divergence between respective 16S rDNA genes and the putative CYLN biosynthesis genes becomes even more apparent. There is mounting evidence that horizontal gene transfer is an important factor in shaping bacterial genomes (e.g., Palenik et al. 2003; Sawada et al. 2002). The present data suggest that a gene transfer event may have taken place between the Cylindrospermopsis and Aphanizomenon/Umezakia/Anabaena groups, as well as among the Aphanizomenon/Umezakia/Anabaena group. The aoaB homologue in the Hungarian isolate C. raciborskii Hung1 could represent either a remnant or an otherwise ancestral intermediate of a functional CYLN gene cluster and an evolutionary link between the Aphanizomenon/Umezakia/Anabaena and Cylindrospermopsis groups.

There is indirect evidence that CYLN originated in Australia and spread to other countries in the world. CYLN poisoning has been reported in Australia since times of early European settlement, then referred to as “Belyando Fever” or “Barcoo Disease” (Hayman 1992), and the largest number of reported CYLN producing species occurs in Australia. Up until early last century, C. raciborskii was limited to tropical regions, with a primary evolutionary center in the deep lakes of tropical Africa and a secondary center in Northern Australia, from where it is believed to have recently spread to other parts of the world, including Europe (Padisak 1997). The invasiveness of C. raciborskii has been described (Neilan et al. 2003), while A. ovalisporum may share similar physiological features (Shaw et al. 1999). The close phylogenetic relatedness between Israeli and Australian A. ovalisporum isolates indicates that they may have originated from a common source and are capable of being dispersed over vast distances.

Conclusion

The AoaA model suggested that the putative function of AoaA is the production of guanidino acetate from glycine with arginine as the guanidino donor. As opposed to Strb1, secondary structures involved in substrate induced conformational changes in GATM were present in AoaA, however, their amino acid sequence was poorly conserved, and the topology of these structures could not be modeled accurately. No prediction was made about whether AoaA would undergo substrate induced conformational changes or whether it would be static as in the case of Strb1.

Previously established models were used to attempt the prediction of the substrate of the AoaB A-domain. Secondary structural alignment revealed the presence of a universally conserved lysine (K558) and aspartate residue (D258) in AoaB, which in other A-domains forms hydrogen bonds to the α-carboxylate and α-amino group of the amino acid substrate, respectively. In addition, AoaB provided a second aspartate (D362) residue, which in other A-domains may be one of the hydrophobic amino acids, leucine, isoleucine, or valine. Assuming that the catalytic residues of AoaB provide the same topology as in GrsA guanidino acetate may be held in the catalytic site by hydrogen bonds between its α-carboxylate and K558, between its imino-nitrogen of the guanidino group and D258, as well as between its positively charged guanidino amine and D362.

CYLN producing strains have been assigned to several genera of filamentous heterocyst-forming cyanobacteria based on morphological characteristics. The molecular phylogeny of CYLN producing strains determined in the present study clearly showed that the morphological classification is not precise. The CYLN producing strains examined belonged to two phylotypes. One consisted of strains belonging to the monophyletic and monospecific C. raciborskii and the other, containing Aphanizomenon/Umezakia/Anabaena, should be combined into a revised genus together with Anabaenopsis/Cyanospira.

There was no evident correlation between phylogeny and the capacity for toxin production. The phylogeny of CYLN biosynthesis genes followed, in general, that of the producer strains, however, the divergence between Cylindrospermopsis and Aphanizomenon/Umezakia/Anabaena phylotypes, as well as between strains within the latter phylotype, was found to be greater than that between their aoa homologues. This may be the result of horizontal gene transfer of aoa homologues.

CYLN producing cyanobacteria have the capacity for global distribution and represent a public health concern worldwide. The presence of an amidinotransferase in a cyanotoxin gene cluster is unique to CYLN producing cyanobacteria and thus represents an ideal target as a gene probe for water quality monitoring and, also, for the metabolic engineering of complex biosyntheses.