Diversity of and Selection Acting on Cylindrospermopsin cyrB Gene Adenylation Domain Sequences in Florida

Abstract

Aphanizomenon ovalisporum is the only confirmed cylindrospermopsin producer identified in the United States to date. On the other hand, Cylindrospermopsis raciborskii is a prominent feature of many lakes in Florida and other regions of the United States. To see the variation in cylindrospermopsin cyrB gene adenylation domain sequences and possibly discover new cylindrospermopsin producers, we collected water samples for a 3-year period from 17 different systems in Florida. Positive amplicons were cloned and sequenced, revealing that approximately 92% of sequences were A. ovalisporum-like (>99% identity). Interestingly, 6% of sequences were very similar (>99% identity) to cyrB sequences of C. raciborskii from Australia and of Aphanizomenon sp. from Germany. Neutrality tests suggest that A. ovalisporum-like cyrB adenylation domain sequences are under purifying selection, with abundant low-frequency polymorphisms within the population. On the other hand, when compared between species by codon-based methods, amino acids of CyrB also seem to be under purifying selection, in accordance with the one proposed amino acid thought to be activated by the CyrB adenylation domain.

Cyanobacteria are oxygenic photosynthetic organisms capable of reaching very high biomass levels in freshwater, brackish, and marine environments (24, 25). Although important in global nitrogen and carbon fixation (3), they are a concern for animal and human health due to toxic compounds that some cyanobacterial species can produce, especially in freshwater ecosystems (5). Cylindrospermopsin (CYN) is a toxic alkaloid containing a tricyclic carbon skeleton with an attached hydroxymethyl uracil (23). CYN has been shown to be hepatotoxic, cytotoxic, genotoxic, and potentially carcinogenic (7, 8, 31, 43). It is biosynthesized via the action of 11 genes encoding enzymes, including an amidinotransferase, mixed nonribosomal peptide synthetase (NRPS)/polyketide synthase (PKS), other PKSs, and tailoring enzymes (20). It is proposed that CYN biosynthesis starts with the transfer of a guanidino group from arginine onto glycine to form guanidinoacetate by the action of CyrA, an amidinotransferase. This guanidinoacetate is then used as a substrate by the adenylation domain of CyrB, a mixed NRPS/PKS, for subsequent polyketide extensions (14). Two other naturally occurring variants of CYN are 7-epicylindrospermopsin, with the C-7 hydroxyl group in a different orientation (2), and 7-deoxycylindrospermopsin, lacking the C-7 hydroxyl group (22).

CYN production was first identified for Cylindrospermopsis raciborskii in Australia (23), followed by Umezakia natans (11), Aphanizomenon ovalisporum (1), Raphidiopsis curvata (16), Anabaena lapponica (37), Aphanizomenon flos-aquae (30), Lyngbya wollei (34), and Oscillatoria PCC 6506 (18), from different parts of the world. The state of Florida has been a major focus of attention associated with the CYN threat in the United States due to the high frequency and broad distribution of C. raciborskii blooms (4, 46). Chapman and Schelske (4) suggested that C. raciborskii might have entered Florida's lakes after the 1960s, based on unpublished phytoplankton data. However, Prescott and Andrews (29) reported the occurrence of Anabaenopsis (Cylindrospermopsis) raciborskii in Wooster Lake in Kansas during the 1950s, which suggests that C. raciborskii may not be a recent invader in the United States. Recent studies indicate that Cylindrospermopsis is widely distributed throughout the United States, including reports from Lake Pontchartrain in Louisiana (21), Lakes Muskegon and Mona in Michigan (12), and Sandusky Bay of Lake Erie, OH (6); however, the CYN production capabilities of these strains of Cylindrospermopsis remain unclear. A recent study of eight Cylindrospermopsis raciborskii isolates from Florida did not reveal any with the ability to produce CYN. However, the same study isolated a strain of Aphanizomenon ovalisporum capable of CYN production (46).

The objectives of this study were as follows: (i) to sequence cyrB gene adenylation domains from various lakes and rivers in Florida to define the diversity of CYN genes and reveal other potential CYN producers and (ii) to investigate the type of natural selection acting on cyrB adenylation domains within A. ovalisporum populations in Florida and between species.

MATERIALS AND METHODS

Collection of environmental samples and genomic DNA isolation.

Lake and river samples were collected with a vertical integrating pole sampler. Subsamples of water were preserved with Lugol's solution for microscopic identification of cyanobacteria. Locations, coordinates, and collection dates of sampling are shown in Table S1 in the supplemental material. Separation of cells and community genomic DNA isolation from environmental samples were performed as described by Yilmaz et al. (47). Community genomic DNAs were also purified with Wizard DNA cleanup system columns (Promega, Madison, WI) to further reduce inhibitory effects.

Amplification, cloning, and sequencing.

Partial amplification of the cylindrospermopsin cyrB gene adenylation domain was performed with the M13-M14 primer pair described by Schembri et al. (33). Amplification conditions were as described by Yilmaz et al. (46). PCR bands were purified from agarose gels (1.5% [wt/vol]), using QIAquick gel extraction kits (Qiagen Inc., Valencia, CA). cyrB fragments were cloned with the pGEM-T vector system II (Promega) in accordance with the manufacturer's instructions. Plasmids were purified with Purelink quick plasmid miniprep kits (Invitrogen, Carlsbad, CA). They were sequenced using the universal M13 forward and reverse primers on both strands at the University of Florida's Interdisciplinary Center for Biotechnology Research core sequencing facility (36).

Sequence alignment and phylogenetic analyses.

Sequences obtained were manually edited. Sequence alignment was performed with the amino acid sequences, which were back translated to nucleotide sequences with Clustal X implemented in the Mega software program, version 4.1 (42). The alignment was manually checked and corrected and contained 40 taxa with 552 nucleotide positions and no gaps. Phylogenetic tree constructions were done using the neighbor-joining (NJ) method with the maximum-composite-likelihood (MCL) method of nucleotide substitution, implemented in MEGA, version 4.1 (42). Maximum-likelihood (ML) trees were constructed in the PhyML 3.0.1 software program (10) as described by Yilmaz et al. (45). Maximum-parsimony (MP) trees were computed in MEGA, version 4.1 (42) using the close-neighbor-interchange (CNI) search method with search level 3. The initial tree was set by random addition with 10 replications. All tree constructions were performed with 1,000 bootstrap replications (except for the MP tree, which was performed with 100 replications).

Genetic diversity and selection analysis.

Codon-aligned multiple sequence alignment of cyrB adenylation domain sequences was analyzed to determine descriptive statistics for the whole data set and for the clusters identified in phylogenetic analyses using the DnaSP software program, version 5.0 (17). Two frequency-based neutrality tests, Tajima's D test (41) and Fu and Li's D* test (9), were performed for the A. ovalisporum-like sequences. Neutrality can be rejected if these D statistics are significantly different from zero (13, 41). A polymorphism/divergence neutrality test, the McDonald-Kreitman test (19), was performed for the A. ovalisporum-like sequences, with other species or phylogenetic clusters used as outgroups. This test is based on the idea that under neutrality, the ratio of nonsynonymous to synonymous polymorphisms within a species should be equal to the ratio of nonsynonymous to synonymous fixed differences between species. The significance of deviation from neutrality was tested with a two-tailed Fisher exact test. All three neutrality tests were performed with DnaSP, version 5.0 (17).

To determine the type of natural selection acting on cyrB adenylation domain codons between species, reduced multiple sequence alignment of cyrB sequences were submitted to the Selecton Server, version 2.2 (39). This server allows hypothesis testing, comparing a null evolutionary model that does not allow positive selection with one that does. Except for the mechanistic empirical combination (MEC) model, all models calculate ω, the ratio between nonsynonymous (K_a) and synonymous (K_s) substitution rates, with an empirical Bayesian method at each codon site. Under the MEC model, ω is not directly equivalent to Ka/K_s but is used to calculate these ratios. Codon sites with K_a/K_s values significantly smaller than 1 indicate purifying selection, while K_a/K_s values significantly larger than 1 indicate positive selection. (39). In the output file of Selecton analysis, the reliability of the ω values is given with a confidence interval defined by the 5th and 95th percentiles of the posterior distribution. The Selecton server further projects reliability estimates for each codon on a selected protein sequence chosen from the alignment using color scales. Both the M8 and MEC models allow all types of selection to be detected (39). Sequences used for selection analyses are marked with an asterisk in Fig. 1.

FIG. 1.

Neighbor-joining tree of partial environmental cyrB adenylation domain sequences. Bootstrap values above 50% are shown next to the nodes (MP/ML/NJ). Sixteen sequences used in Selecton and HyPhy analyses are shown with an asterisk. Branch lengths are proportional to the number of substitutions per site (see the scale bar).

Similarly, reduced codon-aligned cyrB sequences were also analyzed with the HyPhy software package implemented at the Datamonkey server (http://www.datamonkey.org/) to infer sites under positive or negative selection (28). Sequences were analyzed by the single-likelihood ancestor counting (SLAC), fixed-effects likelihood (FEL), and random-effects likelihood (REL) methods with default significance level settings to test selection at each codon site. All three are codon-based maximum-likelihood methods and can infer both positively and negatively selected sites. SLAC is the most conservative of all three methods, followed by FEL and REL (28).

Residues important for amino acid activation in the CyrB adenylation domain were determined by submitting the translated Aphanizomenon ovalisporum FAS-AP1 cyrB sequence to the Bioinformatics Toolbox (http://www-ab.informatik.uni-tuebingen.de/software/NRPSpredictor).

Nucleotide sequence accession numbers.

New cyrB sequences obtained in this work were deposited in GenBank under accession numbers HQ712082 to HQ712115.

RESULTS AND DISCUSSION

Genetic diversity and phylogenetic analyses of cyrB sequences.

To define the variation in CYN genes in Florida and discover new CYN producers, we collected water samples from 17 different lakes and rivers in Florida over a 3-year period. Forty-three samples were tested, 10 of which tested positive for the cyrB gene. Among the partial cyrB adenylation domain sequences (552 bp) obtained from environmental samples, approximately 48% were exact matches for that of A. ovalisporum, while 44% of the sequences were very similar (>99% identity) to the sequence for A. ovalisporum cyrB (herein, “A. ovalisporum-like” indicates A. ovalisporum and similar sequences). Therefore, we propose that approximately 92% of the sequences probably belong to A. ovalisporum and the variation observed in sequences is related to population diversity (Table 1). All A. ovalisporum-like sequences originated at lake and river sites in the St. Johns River system, which flows north from central Florida to the Atlantic Ocean near the city of Jacksonville, spanning a 500-km reach. The presence of A. ovalisporum-like sequences only in the St. Johns River samples suggests environmental conditions favoring the growth of this taxon. Indeed, microscopic examination confirmed the presence of A. ovalisporum in samples from the St. Johns River (M. Cichra and E. J. Phlips, unpublished data). A. ovalisporum has been reported to occur in systems with elevated conductivity (35). Salt water from the Atlantic Ocean and a salt spring that flows into Lake George (located in the middle of the river reach) affect the conductivity of the lower St. Johns River (26, 27).

TABLE 1.

Genetic diversity estimates for cyrB adenylation domain and neutrality tests^a

Sample	No. of sequences	No. of haplotypes	No. of segregating sites	π	θw	D value	D* value
A. ovalisporum-like	31	31	41	0.0058	0.0186	−2.53***	−3.97**
Cylindrospermopsis-like	5	5	6	0.0044	0.0052
Cylindrospermopsis	2	2	1	0.0018	0.0018
Aphanizomenon sp.	2	2	1	0.0018	0.0018
All	40	40	144	0.0249	0.0613

^aNumber of sequences analyzed, number of haplotypes and number of segregating (polymorphic) sites are shown. π, average nucleotide diversity per site between two sequences; θw, an estimator of genetic diversity per site based on segregating sites; D and D*, Tajima's and Fu and Li's neutrality tests, respectively; ***, P < 0.001; **, P < 0.02.

Six percent of the obtained sequences from the environmental samples were very similar (>99% identity) to cyrB sequences of both C. raciborskii from Australia (551/552 base pair identity) and Aphanizomenon sp. from Germany (548/552 base pair identity). Due to higher similarity, we designated these sequences as C. raciborskii-like. The sequences originated from two lakes in west-central Florida, Lake Valrico and Little Lake Wilson. Samples from these lakes contained C. raciborskii, Anabaena circinalis, Microcystis aeruginosa, Planktothrix sp., and Aphanizomenon sp. The CYN producer A. ovalisporum was not observed in either of the lakes. Without further isolation and testing of the species observed in these samples, it is not possible to pinpoint the sources of the observed cyrB sequences. Nevertheless, our results demonstrate that variation exists within Florida for cyrB adenylation domain sequences, some of which might be from undiscovered CYN-producing species. To date, none of the 21 strains of C. raciborskii isolated in our laboratory have demonstrated the ability to produce CYN (M. Yilmaz and E. J. Phlips, unpublished data). In addition, no amplification products of cyrB were observed in samples from the Harris Chain of Lakes (Lakes Griffin, Dora, and Harris), although C. raciborskii was present in all samples. The integrity of every DNA preparation was checked by amplifying a portion of the phycocyanin intergenic spacer region; therefore, PCR inhibition from DNA of these samples was ruled out (data not shown). The specific sources of the 6% of sequences similar to that of C. raciborskii remain a target for continued investigations.

In addition to these environmental cyrB sequences, we also downloaded other cyrB sequences from the National Center for Biotechnology Information (NCBI) database (http://www.ncbi.nlm.nih.gov/) for C. raciborskii, Aphanizomenon sp., and Oscillatoria PCC 6506. Forty unique sequences (full data set) were used in the alignment and subsequent phylogenetic analyses after removal of duplicates. All three phylogenetic tree construction methods yielded similar trees. The NJ tree with bootstrap supports from all three methods is shown in Fig. 1. All A. ovalisporum-like cyrB sequences clustered together with that of A. ovalisporum FAS-AP1, with high bootstrap support. Two Aphanizomenon sp. cyrB sequences clustered with C. raciborskii-like sequences with high bootstrap support; however, they formed a subcluster within this group with moderate bootstrap support. C. raciborskii-like sequences formed another subcluster with high bootstrap support. A Lake George sample collected on 2 September 2008 (LG090208m) was a sister to Aphanizomenon sp. and C. raciborskii-like sequences. The Oscillatoria PCC 6506 cyrB sequence was used as the outgroup in phylogenetic trees since it was the most distant sequence relative to all other sequences (82 to 83% identity to other cyrB sequences).

Average nucleotide diversity (π × 100) within each group was very low, from 0.4% in the Cylindrospermopsis-like group to 0.58% in the A. ovalisporum-like group. Analysis of all sequences yielded an average nucleotide diversity of 2.49% (Table 1).

Selection acting on cyrB adenylation domains.

The cyrB adenylation domain sequences obtained from lake and river systems in Florida were used to examine the evolutionary forces acting on these genes on relatively short time scales, particularly within the A. ovalisporum population. To determine whether polymorphisms observed within the A. ovalisporum-like cluster deviate from neutrality, we used two tests, Tajima's D test and Fu and Li's D* test. A. ovalisporum-like sequences yielded significant negative D values for both statistics. This result suggests that there is an excess of low-frequency variants within the A. ovalisporum-like group, indicating the action of purifying selection on cyrB adenylation domains in Florida. On the other hand, Tajima's D value for nonsynonymous sites was −2.53222 (P < 0.001) and −2.03673 (P < 0.05) for synonymous sites, suggesting that some of these mutations might be weakly deleterious (13). Thirty-one sequences were analyzed within the A. ovalisporum-like group. Fifteen of these sequences had a single nucleotide change from the A. ovalisporum FAS-AP1 sequence. It is possible that some of these single nucleotide polymorphisms represent errors introduced by Taq polymerase during PCR amplification. For this reason we performed the same analysis after removing these sequences. Tajima's D and Fu and Li's D* statistics were significantly negative when performed on the remaining 16 A. ovalisporum-like sequences (−2.096 [P < 0.05] and −2.40 [P < 0.05], respectively). For all comparisons within the MacDonald-Kreitman test, ratios of nonsynonymous to synonymous changes within A. ovalisporum-like sequences were significantly different from those fixed between species (Table 2). There were more replacement changes within species than between species, suggesting the presence of slightly deleterious replacement mutations under purifying selection. Similar results were obtained when sequences with a single nucleotide change from that of A. ovalisporum FAS-AP1 were removed from the data set before analysis.

TABLE 2.

MacDonald-Kreitman test results of neutrality for cyrB adenylation domain sequences obtained from environmental samples^a

Comparison	Interspecific fixed nucleotide difference		Intraspecific nucleotide polymorphisms		P value
	Syn	Non-syn	Syn	Non-syn
A. ovalisporum-like vs Cylindrospermopsis	14 (14)	6 (6)	10 (6)	33 (25)	0.00066*** (0.00044***)
A. ovalisporum-like vs Aphanizomenon sp.	13 (13)	6 (6)	10 (6)	33 (25)	0.00134** (0.00086***)
A. ovalisporum-like vs Oscillatoria PCC6506	64 (64)	29 (29)	10 (6)	32 (24)	0.00000*** (0.00000***)

^aNumbers in parentheses show values obtained with the data set when sequences with a single nucleotide change from the A. ovalisporum FAS-AP1 sequence are removed. Syn, synonymous; Non-syn, nonsynonymous; ***, P < 0.001; **, 0.001 < P < 0.01.

Similarly, when this gene was compared between species on a longer time scale, a majority of codons seemed to be under purifying selection. In order to detect the type of natural selection acting on codons of the adenylation domains of cyrB genes of different species, we used several methods implemented at the Selecton Server and within the HyPhy package. For this purpose, the alignment (full data set) of cyrB sequences was reduced to 16 sequences (reduced data set) by removing 24 least-divergent A. ovalisporum-like sequences in order not to cause a bias in the analyses from redundant A. ovalisporum-like sequences. Recombination was not detected using the Genetic Algorithm for Recombination Detection (GARD) tool at the Datamonkey server, for either the full or reduced cyrB alignment set. For the reduced multiple sequence alignment of cyrB sequences, the M8 model, which allows positive selection, was not significantly better than the M8a model, which does not allow positive selection, based on the results of the likelihood ratio test (likelihood values were −1,364.3 and −1,362.74, respectively). However, positive selection was significant when MEC and M8a were compared based on the lower second-order Akaike information criterion (AICc) score of MEC (likelihood values were −1,354.97 and −1,362.8, and AICc scores were 2,720.1 and 2,733.7, respectively). The advantage of the MEC model over other models is that it takes into account the differences between amino acid replacement rates. Thus, radical nucleotide replacements will result in higher K_a values than moderate replacements. Projected positive or negative selection on amino acid sites for the MEC model is shown in Fig. 2. As can be seen, a majority of amino acid sites are under purifying selection. Although according to the MEC model positive selection in the protein was significant, none of the inferred sites was significantly positively selected.

FIG. 2.

A modified output of Selecton MEC analysis showing negatively or positively selected sites on the partial CyrB adenylation domain of A. ovalisporum FAS-AP1. Shades of gray represent purifying selection, with the darkest being significant. Empty boxes represent positively selected sites, neither of which was significant. Residues 8 Å around the substrate are underlined (32). Nine of ten residues involved in amino acid activation as determined by Stachelhaus (38) are circled. Codon 30, which is also identified as positively selected by the FEL method, is shown with an asterisk.

Among the methods implemented within the HyPhy package, only FEL identified a positively selected site (codon 30 in Fig. 2), while the same method identified 32 negatively selected sites (not shown). SLAC found two negatively selected sites, and REL did not infer any sites with K_a values greater than K_s values, suggesting all sites are under purifying selection.

For both Selecton and HyPhy analyses, the results were the same whether the full data set or fewer sequences were used. A majority of amino acids were still under purifying selection. Only the significance of purifying selection changed for some sites. Using the full data set did not yield any sites with significant positive selection for either method. The reduced data set or fewer sequences identified codon 30 as significantly positively selected only for the FEL analysis (data not shown).

Purifying selection acting on the adenylation domain of cyrB is in agreement with the only proposed amino acid thought to be activated by this domain, which is guanidinoacetate (20). Although some positively selected sites were proposed, this is similar to purifying selection observed for adenylation domains of microcystin biosynthesis genes (15, 44). When conserved residues for amino acid activation are aligned for the taxa used in the phylogenetic analyses of cyrB (Fig. 3), it is apparent that there is a high conservation for amino acid residues 8 Å around the substrate (32) and for the 10-amino-acid code defined by Stachelhaus (38). However, there are also nonsynonymous changes resulting in different amino acids for some species. It is difficult to estimate the impact of such replacement changes on the amino acid activated or CYN biosynthesis, especially without knowing if those sequences are coming from CYN producers or CYN mutants (14). One noteworthy observation is the replacement of polar aspartic acid (site 27 in Fig. 2 and site 6 in Fig. 3) with nonpolar glycine in a Crescent Lake clone (CL050906i). This aspartic acid residue (referred to as D258 in reference 14) is proposed to form a hydrogen bond to the α-amino group of guanidinoacetate, along with K558, which forms a hydrogen bond to the α-carboxylate of the substrate (14). Other important nonsynonymous changes were observed within codon 30 (codon 9 in Fig. 3). This site is conserved among all sequences, except in three sequences, including DUNNS050906g (C30R), LG090208m (C30R), and Oscillatoria PCC 6506 (C30F). This site was identified as significantly positively selected only in the FEL analysis. Since site 30 is not identified by all methods as positively selected, it does not provide definitive evidence that positive selection is acting on this site. However, it is 1 bp downstream from a conserved His, which is one of 10 amino acids identified as important for substrate selectivity in adenylation domains (38).

FIG. 3.

Alignment of CyrB adenylation domain residues 8 Å around the substrate. Residues identical to those in the consensus sequence are shown with dots. Nine of ten residues proposed to be involved in amino acid activation as determined by Stachelhaus (38) are underlined. Twenty-eight other sequences in the full data set of cyrB alignment shared identical residues with the A. ovalisporum consensus sequence. The conserved residues for the DUNNS050906g sequence were identical to those for the LG090208m sequence.

Similarity between 16S rRNA gene sequences of CYN-producing Cylindrospermopsis from Australia and CYN-producing A. ovalisporum (93% identity) is lower than the similarity between their CYN biosynthesis gene sequences (96% identity), which led Kellmann et al. (14) to propose that horizontal gene transfer (HGT) events occurred between these genera. With newly published cyrB sequences, a similar situation is seen between Aphanizomenon sp. 10E6 and C. raciborskii AWT205, with 94% identity for the 16S rRNA gene and 99% identity for the cyrB adenylation domain. The identities of other CYN biosynthesis genes between these two genera are high and are larger than 95% (40). Sequence identities higher than that of the 16S rRNA gene, along with various and high identities for different CYN genes, suggest that HGT events occurred between C. raciborskii and Aphanizomenon sp., probably involving specific regions at different times (40). A somewhat different situation is observed for the newly reported CYN producer Oscillatoria PCC 6506 (18). While its 16S rRNA gene identity to those of other CYN producers ranges between 87% and 92%, a partial cyrB adenylation domain sequence of this strain shows only 82 to 83% identity to those of others. A BLAST search of Oscillatoria PCC 6506 CYN biosynthesis genes against Aphanizomenon sp. 10E6 and C. raciborskii AWT 205 sequences yielded nucleotide identities ranging from 82% (for cyrB) to 93%, with a majority around 86 to 87% identity. Similar or smaller phylogenetic distances of cyrB sequences compared to 16S rRNA gene distances suggest evolution of cyrB genes independently within Oscillatoria. Higher identities for some genes (i.e., cyrJ and cyrK) between Oscillatoria and other CYN producers might suggest an HGT or different selective constraints for different CYN genes.