Genomic insights into the biosynthesis and physiology of the cyanobacterial neurotoxin 3-N -methyl-2,3-diaminopropanoic acid (BMAA)

Cyanobacteria are an ancient clade of photosynthetic prokaryotes, present in many habitats throughout the


Introduction
3-N-methyl-2,3-diaminopropanoic acid (syn: α-amino-β-methylaminopropionic acid, MeDAP; β-N-methylaminoalanine, BMAA) is a neurotoxin that was first isolated from seed of Cycas micronesica K.D.Hill (Cycadaceae) (Vega and Bell, 1967;Nunn, 2017).BMAA is a non-encoded amino acid (i.e.not coded for by a codon in the genetic code).It has since been detected in a range of organisms, including some species of cyanobacteria (Cox et al., 2003(Cox et al., , 2005;;Banack et al., 2007;Downing et al., 2011;Downing and Downing, 2016).Interest in BMAA was generated by the possibility that it may be a contributory causative agent of the chronic neurological complex, amyotrophic lateral sclerosis/parkinsonism/dementia (ALS-PDC) of Guam (Nunn, 2017), but this remains a controversial matter (Chernoff et al., 2017;Dunlop et al., 2021).Ecological and physiological functions for BMAA have not been determined and metabolic pathways for the biosynthesis of BMAA in cyanobacteria are currently unknown.Among other benefits, knowledge of such pathways would allow targeting of specific biosynthesis genes by polymerase chain reaction (PCR)-based methods, which could be used as an early-warning system for the potential presence of cyanotoxins in water resources and other environments (Humbert, 2017;Kurmayer et al., 2017).
BMAA occurs free and as two bound forms (Vega and Bell, 1967;Polsky et al., 1972;Dossaji and Bell, 1973;Murch et al., 2004;Faassen et al., 2016) but the chemical complexes in which BMAA is incorporated (corresponding to the soluble bound form and TCA-precipitated bound form) have not been elucidated.All three fractions of BMAA may or may not be present in organisms shown to produce BMAA (Rosén et al., 2016) and it cannot be dismissed that free and bound BMAA moieties may be derived from different metabolic precursors (Nunn and Codd, 2017).
The extent of BMAA biosynthesis throughout the Cyanoprokaryota has not been systematically determined due, to some extent, to analytical limitations (Faassen, 2014;Mantas et al., 2021).It is widely accepted that analytical methods using tandem mass spectrometry (MS/MS) methods are the most suitable for the identification and quantification of BMAA (Cohen, 2012;Faassen et al., 2012;Faassen, 2014).However, variation still exists regarding sample processing and separation methods (Faassen, 2014) and inter-laboratory comparisons are scarce (Faassen et al., 2016).Despite such problems, the ability of axenic, single strain cultures of some cyanobacteria to biosynthesise BMAA has been unequivocally confirmed (Eriksson et al., 2009;Downing et al., 2011;Downing and Downing, 2016).
In this study, we used bioinformatics tools to investigate hypotheses concerning BMAA biosynthesis in cyanobacteria through an assessment of the presence or absence of enzymes in six known potential metabolic pathways, across 130 cyanobacterial genomes.We show that most enzymes involved in pathways leading to the putative precursor of BMAA (2,3-diaminopropanoic acid, 2,3-DAP) in other species, were not found in cyanobacteria.Genes coding for SbnA and SbnB, by whose concerted action the biosynthesis of 2,3-DAP is known to occur in Staphylococcus aureus Rosenbach 1884 (Staphylococcaceae), were found, though limited to a subset of cyanobacterial species.We highlight the potential physiological role of 2,3-DAP in siderophore formation in some cyanobacterial species and show that the pam gene cluster, responsible for directing the biosynthesis of peptide-bound BMAA in Paenibacillus larvae (White 1906) Ash et al. 1994 (Paenibacillaceae), was not detected in 130 cyanobacterial species, nor was it found in 93 genomes of Paenibacillus Ash et al. 1994 (other than P. larvae) or two diatom species.We also show that the presence, in some cyanobacterial species, of genes putatively encoding the enzymes 2,3-diaminopropionate ammonia-lyase (DAPAL,EC 4.3.1.15)and reactive intermediate deaminase A (RidA, EC 3.5.99.10) could explain the failure to detect 2,3-DAP in analytical studies.The biosynthesis of 2,3-DAP in cyanobacteria appears to be either restricted to a small subset of cyanobacterial species, or there may be multiple, additional, routes for the biosynthesis of this amino acid.
2. Routes to the biosynthesis of 2,3-DAP and BMAA in other taxa and potential relevance in cyanobacteria 2.1.The biosynthesis of BMAA from 2, The simplest explanation for the biosynthesis of BMAA is through the 3-N methylation of 2,3-DAP, found free and as simple derivative forms in plant species (Gmelin, 1959;Nunn and Codd, 2017) (Fig. 1).
However, mechanistic support for this pathway (Brenner et al., 2003) is lacking in cyanobacteria.First, although the possible genes encoding cysteine synthase-like and methyltransferase enzymes occur in cyanobacteria (Aráoz et al., 2010), the existence of such enzymes with specificity for the proposed substrates remains speculative.Second, since this pathway requires ammonium, it may be difficult to reconcile with the results of Downing et al. (2011), who showed that BMAA levels in the non-nitrogen-fixing cyanobacterium Microcystis PCC 7806 (Microcystaceae) increase under conditions of nitrogen (ammonium/nitrate) starvation and decrease when ammonium is added.Finally, this pathway assumes that BMAA is biosynthesised by direct methylation of free 2,3-DAP, which has not been found in any cyanobacterial species or in cycads, which can accommodate cyanobacteria, to date.
There are at least four known biosynthetic routes to 2,3-DAP.Two are complex pathways in plants, and the others are simpler mechanisms found in Streptomyces Waksman and Henrici 1943 (Streptomycetaceae) and Staphylococcus Rosenbach 1884 (Staphylococcaceae).
Peptide-bound BMAA occurs in the bacterium Paenibacillus larvae (Müller et al., 2014).Complex antibacterial, antifungal and cytotoxic peptide paenilamicins are biosynthesised by an elaborate hybrid NRPS/polyketide synthetase (PKS) system (pam), in which 2,3-DAP is methylated within the specialised cluster to form BMAA (Garcia--Gonzalez et al., 2014;Müller et al., 2014).The pam gene cluster has been found only in P. larvae.There are more than 200 characterised Paenibacillus species occurring naturally in the environment (Nunn and Codd, 2019).Paenilamicins are similar in structure to the antibiotic peptide galantin 1 (Shoji et al., 1975;Sakai and Ohfune, 1992), also isolated from P. larvae subsp.pulvifaciens (Nakamura 1984) Heyndrickx et al. 1996 (Sakai and Ohfune, 1990).It is hypothesised that both galantins and paenilamicins are biosynthesised by similar NRPS mechanisms, comprising a transmethylase, which produce peptide-bound BMAA (Nunn and Codd, 2017).However, the biosynthetic route to this peptide has not been investigated biochemically.Peptides larger than paenilamicins and galantins and insoluble in protein precipitants may exist in cyanobacteria, which would explain the bound BMAA found in pellet fractions (Nunn and Codd, 2017).To date, galantins and paenilamicins are the only peptides known to contain this neurotoxin.
Another possibility is that BMAA biosynthesis could occur via the dehydration or desulfydration of a serine or cysteine residue, respectively, after insertion into a peptide chain by an NRPS module (Nunn and Codd, 2017).This would generate a dehydroalanine (DHA) residue, enclosed within a protein or peptide.The addition of ammonium to the double bond of the DHA residue would generate 2,3-DAP (Eiger and Greenstein, 1948), which, if N-methylated, could yield peptide-bound BMAA.If hypothetical peptides were subject to metabolic turnover, free BMAA would be released intracellularly.Dehydroamino acids are abundant in cyanobacteria (Siodłak, 2015) and are chemically reactive (Humphrey and Chamberlin, 1997).Available methods for obtaining peptide-bound DHA enzymatically have not been described, and it is currently not possible to bioinformatically test the feasibility of this pathway for the production of 2,3-DAP and/or BMAA in cyanobacteria.

Cyanobacterial species encoding enzymes in the pathway to 2,3-DAP from uracil in Albizia julibrissin
Several enzymes can catalyse the first two reactions in the pathway to 2,3-DAP from uracil (Fig. 2) and further information is needed regarding their substrate specificity in order to test the feasibility of these reactions in cyanobacteria.
Our results indicate that NADPH-dependent DUD (EC 1.3.1.2) was not found in any of the 130 cyanobacteria in our dataset.Protein sequences with some similarity to DUD are present in 125 cyanobacterial species, however they are functionally annotated to enzymes other than DUD and similarity to a sequence model of known DUD is low (Supplementary Table S1).It is currently not possible to discern if these protein sequences have the same substrate specificity and/or are able to catalyse the synthesis of 5-amino-5,6-dihydrouracil from 5-aminouracil.
DHP belongs to the cyclic amidohydrolase family of enzymes, which also include allantoinase, dihydroorotase, hydantoinase, and imidase (Holm and Sander, 1997).Although these metalloenzymes possess similar active sites and may use analogous catalysis mechanisms (Huang, 2015), they have different substrate specificities and relatively low amino acid sequence identity (Hsu et al., 2010;Peng and Huang, 2014).Site-directed mutagenesis studies on a DHP from Pseudomonas aeruginosa (Schroeter 1872) Migula 1900 (Pseudomonadaceae) have shown that histidine residues at positions 59, 61, 183, and 239 (H59, H61, H183, H239) and the aspartate residue at position 316 (D316) are essential for the assembly of the binuclear metal centre of the active site, whereas the tyrosine residue at position 155 (Y155), the serine residue at position 289 (S289), and the asparagine residue at position (N337) are necessary for substrate-binding.These residues are conserved in all DHPs (Huang, 2015).
(Supplementary Table S2), except for the conserved serine residue that is substituted by threonine in five out of the six sequences functionally annotated as DHP.
34 cyanobacterial species appear to have a putative gene for βUP (Table 1).All amino acid sequences from these species were functionally annotated as Zn-dependent hydrolases (Supplementary Table S3).
Six cyanobacterial species appear to have putative genes for both DHP and βUP.These are the same species, listed above, that have putative genes for DHP (Fig. 6).In the apparent absence of a gene for NADP-dependent DUD, this thus appears to be an unlikely route to BMAA in cyanobacterial species.Nevertheless, given the presence of many cyanobacterial protein sequences similar to DUD, biochemical studies on the feasibility of this pathway for BMAA biosynthesis in these six cyanobacterial species are merited.

Cyanobacterial species encoding enzymes in the pathway to 2,3-DAP from β-(isoxazolin-5-on-2-yl) in Lathyrus sativus seedlings
Cysteine synthase (EC 2.5.1.47)appears to be ubiquitous in cyanobacteria with 127 of the 130 species examined having two or more copies of a putative gene coding for the enzyme.In Hydrocoleum sp.CS-953 (Microcoleaceae), genetic evidence for the presence of this enzyme is lacking; this may be due to errors in sequencing, assembly, genome annotation and/or functional annotation.O-acetyl-S-serine-dependent enzymes, such as cysteine synthase, have conserved alanine, phenylalanine and glycine residues at positions 132 (A132), 152 (F152), and 185 (G185), respectively (Kobylarz et al., 2016).These residues are conserved in most cyanobacterial cysteine synthases (Supplementary Table S4).
Although S-2,3-diaminopropionate N-oxalyltransferase (EC 2.3.1.58)is included in enzyme databases such as BRENDA (Chang et al., 2021), no corresponding amino acid or nucleotide sequence is available.In the protein sequence database at NCBI (https://www.ncbi.nlm.nih.gov) four sequences are associated with EC 2.3.1.58,which are all functionally annotated as cysteine synthase A. S-2,3-diaminopropionate N-oxalyltransferase appears to have been described once in Lathyrus sativus (Malathi et al., 1970), but since then has not been purified or sequenced.Whether the enzyme is specific to L. sativus, or was mistakenly characterised, is undetermined.Nevertheless, until further data become available on the enzyme's substrate specificity, catalytic site and taxonomic distribution, no conclusion can be advanced regarding the use of this pathway (Fig. 3) for BMAA formation in cyanobacteria.
Site-directed mutagenesis studies by Kobylarz et al. (2016) showed that, in addition to R132, Y152 and S185, substrate specificity also required the positively-charged residues lysine at position 100 (K100), and arginine at position 224 (R224), thought to be responsible for S-glutamate binding.A four amino acid insertion: glycine at position 126 and 127 (G126, G127), tyrosine at position 128 (Y128), and leucine at position 129 (L129) was also shown to be moderately conserved in SbnA and in homologs that biosynthesise 2,3-DAP (Kobylarz et al., 2016).The substrate specificity of SbnB is still uncertain (Beasley et al., 2011).Nevertheless, it was shown to differ from ornithine cyclodeaminases (including VioK) and alanine dehydrogenases as its active site has been expanded to accommodate a larger substrate than ornithine or alanine (Kobylarz et al., 2014).This is evident by the presence of an arginine at position 94 (R94), forming a salt bridge to the ACEGA terminal carboxylate (Kobylarz et al., 2014).
No species appear to have orphan copies of sbnA or sbnB (i.e.encoding SbnA but not SbnB, or vice-versa).
In the cyanobacterium Coleofasciculus chthonoplastes, putative sbnAB appear together and are co-localised within an NRPS/PKS1 cluster (Fig. 7).The product of this cluster is unknown, but the complex resembles the NPRS assembly responsible for synthesising the siderophore amonabactin P 750 in Aeromonas hydrophila subsp.hydrophila ATCC 7966 (Aeromonadaceae) (Barghouthi et al., 1989).Given that sbnAB are involved in the biosynthesis of staphyloferrin B in Staphylococcus aureus, and that the NRPS cluster containing these genes in cyanobacteria is similar to that of a siderophore-producing cluster in Aeromonas hydrophila, we hypothesise that, in some cyanobacteria, 2,3-DAP is probably involved in the production of iron-chelators.Our previous study indicated the same physiological role for 2,4-DAB, a structural isomer of BMAA (Mantas et al., 2021).Nevertheless, in the absence of a nearby methyltransferase, a feasible route to BMAA is lacking in Coleofasciculus chthonoplastes.In the cyanobacterium Moorea producens, putative sbnAB appear adjacent on the genome, but are not co-localised within a specialised metabolite cluster, and do not have an adjoining methyltransferase (Fig. 7).Nearby genes are not suggestive of any specific biosynthetic pathway, indicating multiple and different roles of 2,3-DAP in cyanobacteria.
Moorea producens and Coleofasciculus chthonoplastes are in the same clade of the cyanobacterial species phylogeny (Fig. 6).Both are marine species, prone to forming thick mats and blooms (Siegesmund et al., 2008;Engene et al., 2012).Our results suggest that, although never detected/quantified, 2,3-DAP can potentially be biosynthesised in cyanobacteria through a pathway similar to that used by Staphylococcus aureus for the synthesis of SB.This pathway to 2,3-DAP does not appear to be widespread in cyanobacteria.
Homologs of SbnA and SbnB have also been described in a novel NRPS cluster responsible for poly(S-diaminopropanoic acid) synthesis in Streptomyces albulus PD-1 (Streptomycetaceae) (NjxA and NjxB) (Xu et al., 2015), and in the gene cluster for sulfazecin in Pseudomonas acidophila Imada et al. 1980 (Burkholderiaceae) (SulG and SulH) (Li et al., 2017) (note, diaminopropionic acid is a valid synonym of diaminopropanoic acid).In Streptomyces albulus PD-1, NjxA and NjxB, homologous to cysteine synthase/serine dehydratase and OCD respectively, use S-serine and S-ornithine as substrates and lack the conserved residues indicated in SbnA and SbnB (Supplementary Table S7-S8).From the description of Xu et al. (2015), it appears that NjxA and NjxB are more structurally similar to VioK and VioB, than to SbnA and SbnB.The substrate specificities of SulG, homologous to OCD, and SulH, homologous to cysteine synthase, have not been characterised.However, they too lack the conserved residues encoded in SbnA and SbnB (Supplementary Table S7-S8).It is possible that cyanobacterial homologs of SbnA and SbnB, with distinct amino acid residues at the catalytic site, are still capable of synthesising 2,3-DAP.
In S. albulus PD-1, the genes responsible for 2,3-DAP biosynthesis (NjxA nd NjxB) were found to be adjacent on the genome, but not colocalised within poly(S-diaminopropionic acid) synthetase gene clusters (Xu et al., 2015).The possibility that putative SbnA and SbnB in the cyanobacterium Moorea producens are responsible for the biosynthesis of 2,3-DAP for use in an unidentified, specialised metabolite cluster cannot be disregarded.
In contrast to SbnA and SbnB, to date, the active sites of VioK and VioB have not been characterised.Provisionally, VioB and VioK may be distinguished from their homologs SbnA and SbnB by functional annotation in bioinformatics databases.VioK can also be differentiated from SbnB by excluding homologous sequences encompassing active site residues characteristic of SbnB (Kobylarz et al., 2014).Given that VioB uses nitrogen atoms from ammonium as the nucleophile, while cysteine synthase uses sulfur atoms from hydrogen sulfide (Thomas et al., 2003), and that VioB uses OAS as substrate whilst SbnA uses OPS, it is not expected that these enzymes (i.e.VioB/cysteine synthase and VioB/SbnA) share identical catalytic sites (i.e.VioB sequences are not expected to contain residues A132, F152 and G185, characteristic of cysteine synthases, nor residues R132, Y152, and S185, characteristic of SbnA).The gene product of vioB, S-2,3-diaminopropionate synthetase, does not have an assigned enzyme commission (EC) number and searches in protein sequence databases yield no results with the same name.Until the active sites of VioK and VioB are described, the possibility that these enzymes share the same catalytic residues as SbnB and SbnA/cysteine synthase, respectively, cannot be excluded.
Two cyanobacterial species appear to have vioK genes, coding for ornithine cyclodeaminase: Phormidium willei (Gardner, 1927) Anagnostidis and Komarek 1988 (Oscillatoriaceae) and Mastigocoleus testarum Lagerheim ex Bornet et Flahault 1887 (Hapalosiphonaceae) (Supplementary Table S9).The putative gene was not included in a specialised metabolite cluster, and no genes functionally annotated to a methyltransferase were found in its genetic neighbourhood (Fig. 8).Although vioK was found in two cyanobacteria, it is not accompanied by  (EC 2.5.1.140,1.5.1.51)(in dark blue) are co-localised within a hybrid NRPS/PKS cluster, the product of which is currently unknown.The gene cluster was adapted from antiSMASH (v.5.1.2). (B) In Moorea producens PAL-8-15-08-1 (Oscillatoriaceae), SbnAB are found adjacent to each other on the genome, but are not co-localised within a specialised metabolite cluster.SbnA and SbnB appear on the third and second frames of translation (FR), respectively.The genomic neighbourhood was adapted from Artemis Comparison Tool (ACT; Carver et al., 2005).For protein accession numbers and functional annotations, see Supplementary Table S22.Arrows indicate the orientation of transcription.(For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)M.J.Q.Mantas et al. vioB.It is unclear whether the species encoding a gene for OCD have developed new physiological roles for the protein, or if a gene coding for S-2,3-diaminopropionate synthetase was lost from these species.In the apparent absence of VioB and given the seeming restricted distribution of OCD, the production of 2,3-DAP, and hence, of BMAA in cyanobacteria is unlikely to occur via a pathway analogous to that present in Streptomyces and Clostridium.
From the cyanobacterial species that appear to have either sbnAB or vioK (Fig. 6), only Coleofasciculus chthonoplastes was sequenced from an axenic monocyanobacterial culture (Mantas et al., 2021, their Supplementary Table S6).It is believed that this species can biosynthesise 2, 3-DAP through an analogous pathway to that of Staphylococcus aureus, since contamination by other potential 2,3-DAP-producing bacteria can be excluded.Results from the remaining species would have to be verified in genomes sequenced from axenic cultures.

The search for the pam gene cluster in cyanobacterial, Paenibacillus, and diatom species
The pam gene cluster, encoding a bound form of BMAA, has been found in Paenibacillus larvae.The gene products of pamS and pamR are responsible for the biosynthesis of 2,3-DAP, the methylation of which within the metabolite cluster could form BMAA (Müller et al., 2014).Since in the similar complex peptide galantin 1, only one of the two 2, 3-DAP residues is methylated, and in the paenilamicins, both 2,3-DAP residues are methylated, it is possible that the pam gene cluster first assembles 2,3-DAP and then methylates it within the complex.
Although it was not specified by Kobylarz et al. (2014), our results show that the 2,3-DAP constituent of paenilamicin is probably derived from the concerted action of genes homologous to sbnAB, as the active sites of both SbnA and SbnB are conserved in PamR and PamS (Supplementary Table S7-S8).In the peptide antibiotic dapdiamide from Pantoea agglomerans, the genes responsible for 2,3-DAP production (ddaA and ddaB) are also homologs of sbnA/sbnB, with conservation of the enzyme's catalytic site (Supplementary Table S7-8).It appears that, when present, 2,3-DAP, and possibly BMAA, are most likely derived from the action of sbnAB rather than of vioB/vioK.
Our results confirm that the paenilamicin gene cluster occurs in 10 Paenibacillus larvae genomes (Supplementary Table S10).However, evidence for the presence of the hybrid pam gene cluster was not found in any of the 130 cyanobacterial genomes included in our dataset (Supplementary Table S11), or in an additional 93 Paenibacillus genomes other than P. larvae (Supplementary Table S12), or in the genomes of two BMAA-producing diatom species, Thalassiosira pseudonana CCMP1335 (Thalassiosiraceae) and Phaeodactylum tricornutum CCAP 1055/1 (Phaeodactylaceae) (Jiang et al., 2014;Réveillon et al., 2016) (Supplementary Tables S13-S14).
In the cyanobacterial species examined, the pam gene cluster does not appear to be a viable route of BMAA biosynthesis.So far, this specialised cluster appears to be restricted to Paenibacillus larvae, suggesting that paenilamicin production is likely not a widespread environmental source of the neurotoxin BMAA.Given the limited number of published diatom genomes, it cannot be concluded that the pam gene cluster is absent in all such species.However, our results indicate that, when BMAA is present in diatoms, it is probably not derived from paenilamicin-like peptides.The reduced number of specialised secondary clusters in the diatom species investigated (Supplementary Table S13-S14), and the absence of genes homologous to those implicated in 2,3-DAP synthesis suggest that, in these species, BMAA is probably either acquired from other sources (i.e.not by direct biosynthesis by the diatoms themselves), or that additional unknown routes to BMAA biosynthesis exist.
It is intriguing that a metabolically expensive and complex peptide such as paenilamicin can apparently only be found in P. larvae.It may be that the pam gene cluster represents a specialised version of a more general mechanism.However, although Paenibacillus spp.other than P. larvae do not appear to synthesise BMAA via the pam gene cluster, many species, including those isolated from human cerebrospinal fluid (Hehnly et al., 2020), appear to be able to synthesise other compounds, such as ectoines, derived from the BMAA structural isomer 2,4-DAB Fig. 8. Localisation and characterisation of the genomic neighbourhood surrounding the enzyme VioK in two cyanobacterial species.VioK is present in both Phormidium willei BDU 13079 (Oscillatoriaceae) (A) and Mastigocoleus testarum BC008 (Hapalosiphonaceae) (B), but is not co-localised within a specialised metabolite cluster.(A) VioK appears on the third frame of translation (FR), adjacent to a protein functionally annotated as FAD-dependent oxidoreductase.(B) VioK appears on the first frame of translation (FR), adjacent to a protein functionally annotated as (2Fe-2S)-binding protein.The genomic neighbourhood was adapted from Artemis Comparison Tool (ACT).For protein accession numbers and functional annotations, see Supplementary Table S23.Arrows indicate the orientation of transcription.

A potential explanation for the apparent absence of 2,3-DAP in cyanobacteria
2,3-DAP expresses severe cytotoxicity in some cell types, and many organisms have developed strategies that prevent endogenous accumulation of specific amino acids (Ernst et al., 2016).DAPAL (EC 4.3.1.15)is a prokaryotic type II PLP-dependent enzyme that catalyses the degradation of R-and S-forms of 2,3-DAP to 2-aminoacrylate and ammonium (Bisht et al., 2012) (Fig. 9).2-aminoacrylate is a three-carbon reactive enamine intermediate synthesised by several PLP-dependent enzymes (Downs and Ernst, 2015).This compound is formed in the reaction catalysed by VioK/VioB; however, there it is contained, bound to a specific position within the enzyme's catalytic site.2-aminoacrylate is susceptible to tautomerisation and spontaneous hydrolysis in solvent water, releasing ammonium and pyruvate (Datta and Bhadra, 1978).It can also react with other compounds, leading to covalent modification and the inactivation of enzymes involved in essential metabolic processes (Flynn and Downs, 2013;Downs and Ernst, 2015) (Fig. 9).
The similarities between DAPAL and other PLP-dependent enzymes, the fact that DAPAL exhibits amino acid sequence dissimilarity in closely related species (Khan et al., 2003;Kalyani et al., 2012;Deka et al., 2018), and that the structure of DAPAL has not been characterised in cyanobacteria, hinder the search and distinction of DAPAL homologs in the 130 cyanobacterial genomes analysed.However, an enzyme functionally annotated as DAPAL and exhibiting amino acid residues similar to EcDAPAL is present in the cyanobacterium Lyngbya confervoides (Fig. 6).Additionally, 108 cyanobacterial species were found to encode enzymes functionally annotated to biosynthetic threonine ammonia-lyases (Supplementary Table S15).Residue D120 is not conserved in any of the cyanobacterial homologs evaluated, including in L. confervoides, suggesting a different mode of abstraction of C1 protons from the S-DAP isoform, or loss of function.Given the observed similarities between the catalytic cleft of Ec/StDAPAL and cyanobacterial biosynthetic threonine ammonia-lyases (Supplementary Table S15), it is possible that the ability to degrade 2,3-DAP into 2-aminoacrylate and ammonium is widespread in cyanobacteria.Purification and molecular characterisation studies of cyanobacterial DAPAL are needed to clarify the prevalence and action of the enzyme in these species.
Putative RidA appears to be omnipresent in cyanobacteria.homologs of Rid family proteins were found, 611 of which have a conserved arginine residue.Only seven cyanobacterial species appear to lack a RidA homolog with a conserved arginine at the catalytic site (Supplementary Table S16) (Fig. 6).There are several species encoding homologs with full amino acid sequence conservation at the active site.Many of these homologs are annotated as hypothetical proteins, and less frequently as endoribonuclease L-PSP (liver-perchloric acid-soluble protein).Apart from the active site, amino acid sequence identity was low between cyanobacterial homologs and those from other bacteria, yeast and mammals.This is consistent with the data from Mistiniene et al. (2003) and Parsons et al. (2003) who found amino acid sequence identity to vary between RidA homologs from different origins, with some members sharing <8% sequence similarity.The observed M.J.Q.Mantas et al. sequence diversity and variation in functional annotation corroborates the premise that RidA may execute myriad different roles (Niehaus et al., 2015).
Our results show that the failure to detect free 2,3-DAP in cyanobacteria may be explained by the presence of DAPAL and RidA.DAPAL would catalyse the synthesis of 2-aminoacrylate and ammonium from 2,3-DAP (Fig. 9).If free 2,3-DAP and the ammonia-lyase were present in cyanobacteria, RidA would be required to quench 2-aminoacrylate, preventing disruption of cellular functions (Downs and Ernst, 2015).The expression of RidA has been shown to be inversely correlated with iron availability (Irons et al., 2020).In conditions of iron limitation, cyanobacteria produce siderophores (Whitton, 2012).If cyanobacteria can synthesise 2,3-DAP-based iron-chelators, this would corroborate our hypothesis that, in some species, 2,3-DAP and hence BMAA, may be involved in siderophore formation.

Comparison between bioinformatics results and the analytical chemistry data from the literature
Following the methodology of Mantas et al. (2021), our bioinformatics findings are compared with those of 28 biochemical studies where evidence for BMAA-, or BMAA-and 2-4-DAB-producing cyanobacteria has been provided (Supplementary Table S17).A scenario where a particular species is shown to possess the necessary genes for BMAA biosynthesis, and is proven to produce the same compound via analytical studies, serves as a strong indicator that this species is a BMAA-producer.However, a species that lacks known genes required for BMAA formation, but is shown by analytical studies to produce the neurotoxin could be indicative of a false-positive result derived from an ambiguous analytical approach (especially probable if non-axenic cyanobacterial cultures and/or low-specificity analytical methods were employed), or suggests the existence of other, still undescribed routes for BMAA biosynthesis.24 of the 28 studies used variations of MS/MS for BMAA/2,4-DAB identification and quantification, and more than one analytical method was commonly used (either to test the accuracy and specificity of one analytical method compared to others, or to cross-check results).Only seven analyses for the neurotoxin itself were derived from all-axenic cultures, and some surveys involved a combination of axenic cultures, non-axenic monocyanobacterial cultures, and environmental samples.Others did not report on the culture type of the isolates studied.Our 130 cyanobacterial species genomic dataset and the compilation of BMAA production from the literature overlapped in seven species: Prochlorococcus marinus Chisholm et al. 2001 (Prochlorococcaceae), Planktothrix agardhii (Gomont 1892) Anagnostidis and Komarek 1988 (Microcoleaceae), Aphanizomenon flos-aquae (L.) Ralfs (Aphanizomenonaceae), Cylindrospermopsis raciborskii (Wołosz.)Seenayya & Subbaraju (Aphanizomenonaceae), Nodularia spumigena Mertens ex Bornet and Flahault, 1888 (Aphanizomenonaceae), Microcystis aeruginosa (Kutzing) Kutzing 1846 (Microcystaceae) and Microcoleus vaginatus Gomont ex Gomont 1892 (Microcoleaceae).None of these common bloom-forming species (Whitton, 2012) appear to have genes coding for relevant enzymes in the pathways to 2,3-DAP/BMAA described above (Figs.2-5), or specialised gene clusters analogous to those of dapdiamides, galatins, or paenilamicin peptides.Apart from the Microcystis aeruginosa used by Downing et al. (2011), the studies that have used these species for the identification and quantification of BMAA employed analytical techniques other than MS/MS, and/or used non-axenic cyanobacterial cultures (i.e.non-axenic monocyanobacterial cultures and/or environmental samples) in their analyses.It cannot be excluded that these correspond to false-positive results as a consequence of the use of low-specificity analytical techniques and non-axenic cultures.However, it is also possible that cyanobacteria synthesise BMAA through a yet-undiscovered route or routes.

Summary and conclusions
Our bioinformatics survey has provided insight into potential pathways for the biosynthesis of 2,3-DAP/BMAA in cyanobacteria, which can lead to future experimental investigations.The bioinformatics results indicate potential 2,3-DAP biosynthesis in some cyanobacterial species via the staphyloferrin B pathway.The fact that, in Staphylococcus aureus, 2,3-DAP is synthesised via sbnAB for inclusion in siderophores, and that in one cyanobacterial species (Coleofasciculus chthonoplastes) putative sbnA and sbnB were found to be co-localised within an NRPS/ PKS1 cluster similar to amonabactin P 750 from Aeromonas hydrophila, suggests a functional association with siderophore biosynthesis.2,3-DAP also appears to be synthesised by homologs of sbnAB in the chemoheterotrophic bacteria Paenibacillus larvae (pamRS) and Paenibacillus agglomerans (ddaAB), indicating that this is a more common route to free and bound 2,3-DAP, and potentially to BMAA, in prokaryotes (compared to the homologous vioKB).
A homolog of the pam gene cluster from P. larvae was not detected in any of the 130 cyanobacterial genomes in our analysis, two diatom species or Paenibacillus other than P. larvae.Other mechanisms may exist that remain to be discovered.
2,3-DAP has not been reported in cyanobacteria.We have shown that, in some species, the presence of putative DAPAL and RidA could explain the failure to detect this compound from cyanobacterial analytical results.RidA expression is upregulated during periods of iron scarcity, which corroborates our hypothesis that, in some cyanobacterial species, 2,3-DAP may be implicated in environmental iron-scavenging.
The existence of many homologs of enzymes potentially leading to 2,3-DAP and BMAA biosynthesis, and the presence of uncharacterised enzymes, with amino acid sequences and structural characterisation not yet available, hinder the search for possible pathways to 2,3-DAP and BMAA in cyanobacteria.Further information regarding the enzymes in the described pathways and their specificity is needed to further test our hypotheses.Also, analytical studies using MS/MS techniques and axenic cyanobacterial cultures would be helpful to ensure accurate and consistent results concerning the identification, origin(s) and quantification of BMAA.Ultimately, the ability to synthesise 2,3-DAP and BMAA does not appear to be universal among cyanobacteria.This suggests that either this compound is not widespread in cyanobacteria, or that there are additional and as yet unknown pathways for the synthesis of 2,3-DAP and BMAA.

Experimental
Methods are based on those of Mantas et al. (2021), summarised here.For further details, see Supplementary Section S7.
iTOL (Letunic and Bork, 2016) was used to re-annotate a previously published phylogeny of 130 high-quality cyanobacterial genomes of various culture types (Mantas et al., 2021).
From the protein sequence database at the NCBI (https://www.ncbi.nlm.nih.gov),amino acid sequences with the same enzyme nomenclature as in the BRENDA database were selected for each enzyme in the known pathways of 2,3-DAP and/or BMAA biosynthesis (Figs.2-5) to build profile hidden Markov models (pHMMs) (Supplementary Section 7.1).The pHMMs for each enzyme were used to search proteins of the 130 cyanobacterial genomes for homologs using HMM search (HMMER package v.3.1b2,hmmer.org),with a default threshold (E-value ≤ 0.01).Where necessary to confirm absence, protein sequences for enzymes in the known pathways for 2,3-DAP and/or BMAA biosynthesis were also used as queries in BLAST searches (Altschul et al., 1997) against all cyanobacterial proteomes available at the NCBI.
AntiSMASH (bacterial version, v.5.1.2) (Blin et al., 2019) was used to search for the pam gene cluster of P. larvae (Supplementary Table 19) in the 130 cyanobacterial species in the dataset and in an additional 93 Paenibacillus spp.genomes (Supplementary Table S20).In two diatom genomes (Supplementary Table S21), sequenced from species previously shown to produce BMAA (Jiang et al., 2014;Réveillon et al., 2016), a combination of antiSMASH bacterial and plant (plantiSMASH) options were used to search for the pam gene cluster (Oliver et al., 2021).

Fig. 7 .
Fig. 7. Localisation and characterisation of the genomic neighbourhood surrounding the enzymes SbnA and SbnB in two cyanobacterial species.(A) In Coleofasciculus chthonoplastes PCC 7420 (Coleofasciculaceae), SbnAB(EC 2.5.1.140,1.5.1.51)(in dark blue) are co-localised within a hybrid NRPS/PKS cluster, the product of which is currently unknown.The gene cluster was adapted from antiSMASH (v.5.1.2). (B) In Moorea producens PAL-8-15-08-1 (Oscillatoriaceae), SbnAB are found adjacent to each other on the genome, but are not co-localised within a specialised metabolite cluster.SbnA and SbnB appear on the third and second frames of translation (FR), respectively.The genomic neighbourhood was adapted from Artemis Comparison Tool (ACT;Carver et al., 2005).For protein accession numbers and functional annotations, see Supplementary TableS22.Arrows indicate the orientation of transcription.(For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)