Abstract
Many dinoflagellate species make toxins in a myriad of different molecular configurations but the underlying chemistry in all cases is presumably via modular synthases, primarily polyketide synthases. In many organisms modular synthases occur as discrete synthetic genes or domains within a gene that act in coordination thus forming a module that produces a particular fragment of a natural product. The modules usually occur in tandem as gene clusters with a syntenic arrangement that is often predictive of the resultant structure. Dinoflagellate genomes however are notoriously complex with individual genes present in many tandem repeats and very few synthetic modules occurring as gene clusters, unlike what has been seen in bacteria and fungi. However, modular synthesis in all organisms requires a free thiol group that acts as a carrier for sequential synthesis called a thiolation domain. We scanned 47 dinoflagellate transcriptomes for 23 modular synthase domain models and compared their abundance among 10 orders of dinoflagellates as well as their co-occurrence with thiolation domains. The total count of domain types was quite large with over thirty-thousand identified, 29 000 of which were in the core dinoflagellates. Although there were no specific trends in domain abundance associated with types of toxins, there were readily observable lineage specific differences. The Gymnodiniales, makers of long polyketide toxins such as brevetoxin and karlotoxin had a high relative abundance of thiolation domains as well as multiple thiolation domains within a single transcript. Orders such as the Gonyaulacales, makers of small polyketides such as spirolides, had fewer thiolation domains but a relative increase in the number of acyl transferases. Unique to the core dinoflagellates, however, were thiolation domains occurring alongside tetratricopeptide repeats that facilitate protein-protein interactions, especially hexa and hepta-repeats, that may explain the scaffolding required for synthetic complexes capable of making large toxins. Clustering analysis for each type of domain was also used to discern possible origins of duplication for the multitude of single domain transcripts. Single domain transcripts frequently clustered with synonymous domains from multi-domain transcripts such as the BurA and ZmaK like genes as well as the multi-ketosynthase genes, sometimes with a large degree of apparent gene duplication, while fatty acid synthesis genes formed distinct clusters. Surprisingly the acyl-transferases and ketoreductases involved in fatty acid synthesis (FabD and FabG, respectively) were found in very large clusters indicating an unprecedented degree of gene duplication for these genes. These results demonstrate a complex evolutionary history of core dinoflagellate modular synthases with domain specific duplications throughout the lineage as well as clues to how large protein complexes can be assembled to synthesize the largest natural products known.
Introduction
Dinoflagellates are unicellular aquatic eukaryotes with an interesting and complicated evolutionary history.1,2 Generally speaking, they can be divided into 2 main groupings with the heterotrophic, often parasitic syndiniales at the base of the dinoflagellate lineage and the often mixotrophic “core dinoflagellates” extending out into the distal branches. 2 All of the core dinoflagellates have a chloroplast or evidence of a lost chloroplast with multiple symbiotic events occurring throughout the lineages.3,4 Although many core dinoflagellates are mixotrophic,5,6 the majority of dinoflagellate “algae” that form harmful algal blooms are photosynthetic, Noctiluca scintillans (Macartney) being the exception. Toxic dinoflagellates are exclusively photosynthetic and there is evidence that toxin synthesis may initiate in the chloroplast,7,8 indicating a potential relationship between photosynthesis and natural product synthesis in dinoflagellates. Amphidinium carterae (Hulbert) is a basal, photosynthetic dinoflagellate that makes the toxin amphidinol as well as many derivatives termed amphidinolides,9,10 indicating that the acquisition of a plastid and toxicity are early events in the evolution of the core dinoflagellates. Many dinoflagellate toxins pose human health concerns by a variety of mechanisms 11 as well as ecological and trophic impacts.12,13
The toxins themselves are almost universally polyketides, that is, they are formed from sequentially added acetate subunits that are modified prior to the addition of the next acetate subunit. 14 The workhouse enzymatic domain in the synthesis of polyketides is the ketosynthase (KS) domain, a condensation domain that incorporates malonyl-CoA into an existing acyl chain as acetate with the release of CO2 driving the reaction.15,16 Analogous to this reaction are non-ribosomal peptide synthases that also perform a condensation type reaction with substrate specificity provided by an adenylation domain that binds a carboxylic acid, often an amino acid, and passes it to the condensation domain for incorporation.17,18 The enzymes that incorporate each building block work along with modifying domains to form synthetic modules that can create very complex biomolecules and are responsible for many known naturally occurring compounds including antibiotics.19-21 Labeling studies have shown that dinoflagellate toxins exclusively incorporate acetate from malonyl-CoA,10,22-25 unlike bacteria that often incorporate similar subunits such as propionate or butyrate, 26 and the toxins often start with (or are occasionally extended by) small amino acids like glycine or other carboxylic acids like glycolate.24,27 There is also evidence for alkylation by methionine or acetate as well as side-by-side “alpha” carbons from acetate explained by the deletion of carbon by a theorized Favorskii rearrangement removing the beta carbon from one acetate. 28 Toxins range in complexity and size from the 31 carbon Gymnodimine 29 to the 164 carbon maitotoxin that has 98 stereocenters. 30
In spite of their complexity, synthesis of the backbone of dinoflagellate toxins utilizes the same core machinery as lipid synthesis. Lipids as a secondary metabolite are differentiated from natural products in that they are fully saturated, highly regulated, and usually synthesized and modified in the chloroplast, mitochondrion, and cytosol.31-33 Thus, there is frequently a segregation of genes, phylogenetically and physically, involved in lipid synthesis from those involved in secondary metabolite synthesis, including in dinoflagellates. 34 In terms of acetate incorporation, all dinoflagellate toxins and lipids rely on the aforementioned ketosynthase domains along with several biologically universal modification domains: ketoreductases (KRs), dehydratases (DHs), and enoyl reductases (ERs) to form a sequentially reduced backbone structure and acyl transferases (ATs) and thioesterases (TEs) to move and terminate growing acyl chains. These enzymatic domains interact with the substrate and each other via a reaction center created by transferring the phosphopantetheinyl arm of coenzyme A onto a carrier protein.35,36 One key difference between lipid and other secondary metabolite synthesis is that lipid synthesis is iterative, utilizing a single carrier protein called the acyl carrier protein while natural products are made with multiple modules with a homologous carrier domain called a thiolation domain. Whether the particular chemistry of each module is a PKS, an NRPS, or a hybrid system; a thiolation domain acts as the reaction center for all of these modular synthases. Likewise, a thiolation domain would be the reaction center for each module involved in toxin synthesis in dinoflagellates. This is useful when dealing with dinoflagellates since the type I multi-domain polyketide synthases and non-ribosomal peptide synthases found in fungi 37 and usually associated with eukaryotes are relatively uncommon in dinoflagellate transcriptomes with most transcripts containing one or rarely a few domains that would have to be combined into a multi-enzyme synthetic complex, 38 similar to the type II polyketide synthases and non-ribosomal peptide synthases usually found in prokaryotes.17,39 This is not surprising since dinoflagellates often encode genes as tandem repeats of gene copies rather than gene clusters of common metabolic function, 40 but this also makes phylogenetic reconstruction difficult even for single domains due to a high copy number of very similar sequences.
The exceptions to the multitude of single domain transcripts in dinoflagellates are several multi-domain genes that have conserved domain arrangement and sequence. Two of these are the BurA and ZmaK-like genes 41 that contain both adenylation and ketosynthase domains in what appears to be a single module. There is also a multi-module gene usually containing at least 3 consecutive ketosynthase containing modules, here referred to as the triple KS.38,42 Phylogenies of dinoflagellate modular synthase domains usually form a robust set of dinoflagellate clades but with poor support placing these clades among eukaryotic outgroups, as well as no obvious reflection of relationships within core dinoflagellates.34,43,44 This not only reveals a gap in annotated sequences that can function as outgroups to dinoflagellates but also indicates that at least some of these modular synthases are likely of bacterial origin, specifically BurA and ZmaK, which have only been described in prokaryotes and seem to have been transferred in their entirety.45,46 Thus, with the exceptions of the conserved fatty acid biosynthetic genes and the corresponding acyl carrier protein, phylogenetic comparisons to model eukaryotes or prokaryotes are generally uninformative when trying to deduce the roles of dinoflagellate modular synthases in toxin production. Likewise, the traditional nomenclature of polyketide synthases that relies on single versus multi-domain and prokaryote versus eukaryote fails to describe the domains in dinoflagellates in a useful manner.
The primary aim of this study was to survey genes that may be involved in dinoflagellate natural product synthesis, specifically toxins, without prejudice from what has been described in prokaryotes or distantly related model eukaryotes. Amphidinium carterae was used as a model because it is a basal toxic dinoflagellate 2 and has the BurA and ZmaK like genes as well as the triple KS gene in their apparent entirety and single copy.41,42 The domains selected were taken from these previously annotated multi-domain dinoflagellate transcripts resulting in several unexpected discoveries such as a large number of adenylation domains seemingly without the traditional condensation domains as well as scaffolding domains associated with specific synthetic domains. This global approach was also able to describe the relative copy number of each synthetic domain revealing several atypical relationships. One example is a large number of enoyl reductases compared to dehydratases, which is very strange since enoyl reductases theoretically act downstream and should be less abundant than dehydratases. The second portion of this survey was to place the retrieved domains into theoretical functional bins based on sequence similarity using a method that is not hampered by gene duplication and horizontal gene transfer. Although many of these synthetic domains can and have been given hypothetical function using phylogenetic inference with model systems as outgroups, the results presented here demonstrate many novel sequence clusters that are difficult to resolve phylogenetically as well as some very atypical gene expansions including acyl transferases and ketoreductases involved in lipid synthesis that were largely overlooked in previous studies. The results of this study demonstrate another way in which dinoflagellates defy the paradigms established by model systems, in this case in terms of the mechanisms of natural product (toxin) synthesis, and are presented here as a framework to be used in future biochemical experiments to validate the hypothetical functions of PKS and NRPS genes in dinoflagellates.
Materials and Methods
Transcriptome preparation and analysis
A total of 61 initial transcriptomes were selected for domain searches with the majority of dinoflagellate transcriptomes taken from the CAMERA database, originally published in (http://camera.calit2.net/mmetsp/list.php, 47 and now hosted at https://www.imicrobe.us/#/projects/104 NCBI project #PRJNA231566, as assembled contigs using Trinity. In addition, data for cultures of Karenia brevis (C.C. Davis), Karlodinium veneficum (D. Ballantine), and Akashiwo sanguinea (K. Hirasaka) that were infected with the syndinean parasite of the genus Amoebophyra were collected from previous phylogenetic studies.1,48,49 For A sanguinea the transcriptome was done with and without infection and for the K. veneficum parasite there is a genome available for comparison. 50 In addition to these transcriptomes the deep sequencing transcriptomes (using Hi-Seq) for K. brevis, 38 and 2 Gambierdiscus species, 51 G. excentricus (S. Fraga), and G. polynesiensis (Chinain and M. Faust) that were assembled using CLC (595M, 118M, 884M reads, respectively) were included. Unfortunately the transcriptomes from the 2 Gambierdiscus species in the transcriptome sequence archive were incomplete with about 70 PKS genes identified in the initial study deposited separately in Genbank. These were added back into the total domain count following domain searches. Each transcriptome was translated in all 6 frames using a Perl script and Genbank translation Table 1 (standard eukaryotic) prior to analysis.
Transcriptome BUSCO scores and domain content.
§ Percentages shown are the fraction of BUSCO genes retrieved by one (Single), multiple (Duplicate), or any number (Combined) of transcripts in each transcriptome.
Benchmarking Universal Single-Copy Orthologs (BUSCO) 52 was used to determine transcriptome quality using the Eukaryota odb9 dataset with final scores (single and multiple copy orthologs) ranging from 1.7% for Perkinsus chesapeaki to 86.1% for Alexandrium tamarense (Lebour) and a median of 77.4% (Table 1). The eukaryote database was chosen over the protist database after initial tests with the protist database gave very low scores (approximately 30% maximum, data not shown). This study is intentionally specific to the “core” dinoflagellates so only closely related outgroups were used including Perkinsus marinus (Levine), Chromera velia (R.B. Moore et al), and Triceratium dubium (Brightwell), as well as the syndinean parasite of crustaceans Hematodinium sp. and the other aforementioned syndiniales with dinoflagellate hosts. No pertinent domains were found in the transcriptomes of syndinean parasites with dinoflagellate hosts except for 3 transcripts from the K. veneficum parasite and were thus excluded from further analyses giving a total of 46 transcriptomes with a BUSCO score 64% or greater that were included in the final tabulations following domain searches. The 2 K. brevis transcriptomes using different assembly methods had similar BUSCO scores and so the Trinity assembled transcriptome was selected for the final tabulations to make comparisons with other transcriptomes, the majority of which were assembled using Trinity, more informative. Oxyrrhis and all outgroup species were given their own taxonomic bin. The forty remaining ingroup transcriptomes were placed into 7 taxonomic bins at approximately the ordinal level including the Gonyaulacales (10 species), the Thoracosphaerales (Brandtodinium), the Prorocentrales (3 species), the Peridiniales (10 species), the Dinophysiales (2 species), the Noctilucales (Noctiluca), and the Gymnodiniales (8 species) with the Suessiales (5 additional species) as a subgrouping of the Gymnodiniales. The 64% cutoff was chosen as a natural observed breakpoint for transcriptomes that had a full repertoire of domains relative to other closely related species (Supplemental Figure S1). Some of the outgroup species had lower BUSCO scores (P. marinus 30%, C. velia 54.8%, T. dubium 41.5%) than the 64% cutoff. Although the scores were low, these transcriptomes were included since most of the tabulations are based on ratios and the domain searches successfully recovered transcripts with synthetic modules, for example, 183 domain hits for T. dubium and 104 domain hits for C. velia. Also, BUSCO analysis of the P. marinus genome (Genbank Bioproject PRJNA12737) yielded a completeness score of 53.3% indicating that the alveolate sequences may not be well represented in the BUSCO database and/or that parasitism has resulted in gene reduction. A lack of sequence representation in the BUSCO database is also supported by maximum BUSCO scores of approximately 80%, even for deeply sequenced transcriptomes showing that the BUSCO scores could be used as a guide but were not quantitative.
HMM assembly and domain searches
Amphidinium carterae (Hulbert) was used to create dinoflagellate specific hidden Markov models (HMMs) of domains from modular synthases. Although many robust models exist for model species, protists in general are poorly sampled and with almost no experimental verification, predictions based on those models are difficult. Four transcripts of multi-domain synthases from the A. carterae transcriptome were used (Figure 1). Each is readily found in other dinoflagellate taxa with the same domain arrangement. The first (comp6001_c0_seq1) is a hybrid PKS/NRPS, the BurA-like gene described in the bacterial genus Burkholderia that participates in the synthesis of burkholderic acid. 45 It has an unusual domain order containing 2 thioesterase, 2 thiolation, an adenylation (described by the NCBI conserved domain database as an acyl-CoA ligase), a ketosynthase, a ketoreductase, and an acyl-transferase domain. The second (comp26075_c0_seq1) is also a hybrid PKS/NRPS that is most similar on a sequence basis to the ZmaK gene described in Bacillus cereus to act in the synthesis of zwittermicin. 46 It contains 2 thiolation, an adenylation, an acyl-transferase, a ketosynthase, a ketoreductase, a dehydratase, an enoyl reductase, and a FSH1 serine hydrolase domain. While the BurA-like gene has the same domain content and arrangement as the genes from Burkholderia, the domain arrangement of the ZmaK-like gene is not similar to the ZmaK gene from B. cereus making predictions about substrates or function in dinoflagellates unreliable. The third multi-domain transcript is a straightforward multiple ketosynthase-containing set of overlapping transcripts (comp305_c0_seq1 and comp32615_c0_seq1) that have a total of 4 thiolation domains and 3 possible modules, each with a ketosynthase domain as well as a ketoreductase; a ketoreductase and a dehydratase; and a ketoreductase, a dehydratase, an enoyl reductase, and a thioesterase. This triple-KS transcript has a ketosynthase in the third module described as an acyl-transferase containing ketosynthase by the NCBI conserved domain database. Thus, an acyl-transferase may or may not be detected depending on the software used and database queried. A final A. carterae transcript (comp14261_c0_seq1) used to make HMMs is herein termed TeCATe due to the flanking thioesterase domains and repeating adenylation and condensation domains as well as a GCN5-associated N-acetyl transferase (GNAT) domain that transfers acetate from acetyl CoA to a substrate, 53 but conservation of this sequence in other dinoflagellate species is low. One additional sequence is comp17349_c0_seq1 that contains a thiolation domain and a tetratricopeptide repeat that is used in protein-protein interactions across all life in a variety of process and configurations. 54 This combination was first described in K. brevis 38 and was included to determine the prevalence and association of this repeat domain with other modular synthase domains. It is unclear if any of these transcripts participate in toxin synthesis but they are readily identifiable and the domain arrangement of the triple-KS, BurA and ZmaK like genes in A. carterae is conserved in other dinoflagellates indicating that the function is also likely conserved.
Domain arrangement of A. carterae transcripts used in hidden Markov model creation.
A total of 22 domains were used for HMM creation with sequence boundaries based on InterPro 55 annotations as implemented in Macvector V16.0.1. These included the adenylation, the ketosynthase, the ketoreductase, and the acyl-transferase domain as well as thioesterase domain 1 from BurA; the adenylation, dehydratase, enoyl reductase, and serine hydrolase domains from ZmaK; ketoreductase domain 2, thiolation domain 3, ketosynthase domain 3, dehydratase domain 3, enoyl reductase domain 3, and the thioesterase domain from the triple-KS; adenylation domain 1, thiolation domain 1, both condensation domains, and the GNAT domain from TeCATe; and finally the thiolation and tetratricopeptide repeat domains from comp_17349_c0_seq1 (Figure 1). These domains were chosen to provide replicative sampling of each domain across multiple sequences when possible.
The protein translation from the A. carterae sequence of each domain was used as the query sequence for a BLAST search across all possible protein translations of the A. carterae transcriptome with no cutoff to give as broad a sampling as possible. The aligned region of each BLAST hit was then compiled into a single file for each query domain in fasta format and aligned using Muscle V3.8.31. 56 These alignments were then each used to generate an A. carterae specific hidden Markov model (HMM) for each domain using hmmbuild in the HMMER V3.3 package. 57 Each HMM was then compressed with hmmpress and used by hmmsearch with an e-value cutoff of 1e − 10 across the protein translations of all 58 transcriptomes with the results given in tab-delimited format for processing. An e-value cutoff was given for the HMM search and not the BLAST search with the assumption that spurious BLAST hits would be represented in the HMM as aligned characters with very low bit scores and that the e-value cutoff in the HMM search would prevent propagation of these errors while maximizing sensitivity. A Perl script was then used to tabulate the data from the HMM search giving a count of each HMM for a given transcript (Supplemental Table S1). The tabulated results were summarized graphically in R V3.3.2 using the GGplot package. For redundant domains the HMM with the highest number of counts for a given transcript was used to maximize sensitivity, for example, if the 3 ketosynthase HMMs returned counts of 1, 2, and 1 the transcript would be counted as having 2 ketosynthase domains. This differentiates the domains that correspond to a specific HMM, from a domain type that equals the functional classification such as ketosynthase
Domain clustering
Protein sequences for each domain were retrieved from the 6-frame translation of each transcriptome using a Perl script and the output from hmmsearch giving the translation frame and position of the alignments for each HMM. Multiple hits within a transcript were denoted as the number out of the maximum number of that domain in the transcript, for example, Ketosynthase 1_3 for ketosynthase domain 1 out of a total of 3 along with the transcript and host identification. The sequence files were dereplicated prior to clustering to remove redundant protein sequences via a Perl script. The extracted protein sequences were output in fasta format and sequences for adenylation, ketosynthase, thiolation, acyl-transferase, and thioesterase domains were each clustered using CLANS. 58 This software uses an all by all BLAST search and the subsequent e-values are used as attraction values to group sequences. This is not as robust a method as phylogenetic inference for finding closest ancestors but is able to group replicated sequences with their ancestor in a 3-dimensional way and is useful when trying to compare and visualize many very similar sequences. Clusters were visually identified based on a high relative number of internal edges and the sequence names of each node within each cluster were exported to a text file. The sequence list of each cluster was then compared to the master list of domain counts for each transcript to determine the content of each cluster that could be annotated. The vast majority of sequences were single domain transcripts. However, if all of a particular domain from a multi-domain transcript was encompassed by a single cluster then that cluster was labeled for that multi-domain transcript. For example, if the ketosynthase domain from every BurA transcript was found in a single cluster then that cluster was labeled “BurA.”
The thiolation domains were a special case in that almost all of the retrieved domains formed just one cluster. In order to provide resolution, the acyl carrier protein sequences (presumably involved in lipid synthesis) from A. carterae (comp649_c0_seq2, comp2819_c0_seq1 and comp3690_c0_seq1) were used as BLAST queries against the other transcriptomes and a separate HMM search was performed. The thiolation domains from these sequences were then added back in to the clustering analysis. This was not necessary for the other domains where either the genes involved in lipid synthesis were retrieved in the initial HMM search or there was sufficient resolution of clusters to make the inclusion of fatty acid biosynthesis genes unnecessary. For the smaller datasets of acyl-transferase and thioesterase domains, verification of the clustering results were attempted by maximum likelihood based phylogenetic inference using RAxML 59 using rapid bootstrapping of 100 replicates and seed values of 11111 for both the bootstrapping and parsimony steps.
Results
BUSCO scores
The scores from the BUSCO analysis ranged from 1.6% for P chesapeaki to 86.1% for A. tamarense (Table 1). There was also frequent duplication with up to 48.8% of the orthologs used for testing present in multiple copies in Kryptoperidinium foliaceum (F. Stein). Despite deep sequencing of several of the transcriptomes, the highest BUSCO score would not be considered a complete transcriptome, indicating that many of the “common” eukaryotic orthologs are not present or were not detected. Deep sequencing also did not guarantee a higher than average score with the G. polynesiensis transcriptome analysis resulting in a score of 68.3%. Several of the transcriptomes had very low scores such as the A. andersonii (33.3%) and P. minimum strain 2233 (39.2%) that correlated to a lower number of assembled contigs (1 M and 500k, respectively) compared to those with high scores (1.8 M for A. tamarense)
Domain tabulation
The “core” dinoflagellates were shown to have many more synthase modules relative to the syndinales and outgroups. Lingulodinium polyedra (F. Stein) possessed the most domains (total HMM hits) and domain types (unique functional group hits, eg, “ketosynthase”) with 2407 and 1304, respectively (Table 1). In total there were 55 818 HMM hits with sufficient scores (<1e − 10) across all transcriptomes (including those with low BUSCO scores) with a median value of 859 per transcriptome, although around 40% of these were the tetratricopeptide repeats predominantly occurring in the core dinoflagellates. When the number of modular synthase hits (excluding tetratricopeptide repeats) was reduced to functional domains by taking the maximum score across all HMMs for the same domain there were 27 424 domains in the core dinoflagellates compared to 1332 for the outgroup species, or an average of 669 and 222 per transcriptome, respectively (Table 2). When the 2 dinoflagellate outgroup species Hematodinium and Oxyrrhis are removed and the remaining outgroup alveolates are taken separately the average drops even further to 79. The largest difference was in the thioesterase domains that were thirteen times more abundant in the core dinoflagellates while often found in single copy in the outgroup species. Thiolation and ketosynthase domains were also much more abundant in the core dinoflagellates with a more than 6-fold increase indicating that core dinoflagellates possess a higher synthetic capacity than other dinoflagellates and alveolates on average.
Summary of Domain Types.
Abbreviations: FSH1, fission yeast serine hydrolase 1; GNAT, GCN5-associated N-acetyl transferase; TTPR, tetratricopeptide repeat.
Values shown represent the count of each domain type in each transcriptome or the sum when designated.
The GNAT and condensation domains from the TeCATe transcript were poorly represented in the core dinoflagellates and absent from the outgroup species (Table 2). This is likely due to the low number of BLAST results (5 for GNAT, 6 for condensation domain 1, and 9 for condensation domain 2) from the primary search in A. carterae that limited the creation of a robust HMM for GNAT and condensation domains. By contrast the adenylation domain from the TeCATe transcript resulted in 41 BLAST hits. The HMM search was still more sensitive than BLAST alone with a total of 13 transcripts detected with a GNAT domain using the HMM with an e-value cutoff of 1e − 10 versus 8 using BLAST with no cutoff among all transcriptomes (data not shown). Still, these domains may have been under-sampled especially for taxa more distantly related to A. carterae. The GNAT and condensation domains were the least represented among all other domains with adenylation and ketosynthase having the highest relative abundance across taxonomic groups (Figure 2). Thiolation, acyl-transferase, enoyl reductase, and ketoreductase domains were also usually well represented while dehydratase, thioesterase, and the FSH1 serine hydrolase were in relatively low abundance.
The percent of each domain relative to all domains detected in dinoflagellates using the dinoflagellate HMMs.
This picture changes when looking at multi-domain transcripts (transcripts with more than one domain type, not including multiple domain hits of the same domain type), where roughly a third of dehydratase domains and half of thiolation domains are found in multi-domain transcripts while adenylation, ketosynthase, ketoreductase, and enoyl reductase domains are predominantly found as single domains (Figure 3). These trends frequently held across taxonomic groupings except for thioesterases, which were found 10% to 15% of the time in multi-domain transcripts for the Gonyaulacales and Dinophysiales and a quarter to a third of the time as multi-domain transcripts in the other taxonomic groups. Multi-domain transcripts were the exception in core dinoflagellates accounting for 8.34% of all domain types (excluding tetratricopeptide repeats) with an average of 13.84% for each species and domain type combination.
The percent of each domain in a multidomain transcript relative to the total number of each domain found in dinoflagellates.
The relative abundance of modular synthase domains was similar across species with no obvious differences in species with a known toxin. The principal components plot based on domain counts and colored by toxin type was used to demonstrate this (Figure 4(A)) and showed a general clustering of all species, irrespective of toxin type except for 3 species: Gyrodinium instriatum, that does not make a known toxin and has a higher proportional number of adenylation domains (Axis 1 outlier on the far left), and Lingulodinium polyedra and Azadinium spinosum that make yessotoxin and azaspiracids, respectively, and have a proportionally higher number of ketosynthase domains (Axis 2 outliers on the bottom). There were however, lineage specific differences for specific domain types. Thiolation domains were more relatively abundant in the more basal Gymnodiniales compared to acyl transferase in a decreasing trend to the more distal Gonyaulacales (Figure 5). This is also visible in the plot of domains as percentages. The acyl-transferase domains make up a much higher percentage in the Gonyaulacales and Dinophysiales versus other taxonomic groups (Figure 2), although this is less obvious for the thiolation domains. There is also a high average number of thiolation domains in a transcript when comparing the Gymnodiniales to the other taxonomic groups (Figure 6).
Principal component plots of the overall domain counts within dinoflagellates (A) and the breakdown of those counts within clusters of highly similar sequences (B).
The ratio of thiolation to acyl-transferase domains in dinoflagellate and outgroup taxa.
The mean number of thiolation domains in all thiolation domain containing transcripts.
Although tetratricopeptide repeats were found in almost every transcriptome, the abundance was much higher in the core dinoflagellates (19 285 in core dinoflagellates vs 300 in outgroups or a 4-fold increase on average per transcriptome) and the combination of this repeat with thiolation domains was only found in the core dinoflagellates (Figure 7). The number of repeats varied within a transcript from 1 to 20 and the distribution of repeat number is approximately log normal in shape with low numbers of repeats being very frequent (Figure 8(A)). This distribution changes dramatically when looking at repeat numbers in transcripts with a thiolation domain where 6 and 7 member repeats are very frequent approximating a t-distribution (Figure 8(B)). The relative number of repeats among each taxonomic group did not vary greatly and approximated the relative total number of domains found.
Percentage of transcripts containing a thiolation domain and tetratricopeptide repeats in dinoflagellates and outgroup species.
Histogram of the number of tetratricopeptide repeats in repeat containing transcripts.
Domain clustering and gene duplication
The protein sequence clustering results in a 3-dimensional relationship of the domains where more similar sequences are spaced more closely together and are shown as points in the data space. If the points are close enough to pass a threshold of a calculated probability then a line is drawn denoting significant similarity and a group of points with interconnecting lines was denoted as a cluster. While these data do not tell us the inferred ancestry of the sequences like a phylogenetic tree would, the relationships are not forced into a bifurcating arrangement. This is helpful in visualizing many very similar sequences such as dinoflagellate domains where there is an abundance of gene duplication and strict orthology is difficult to ascertain. The results for each domain retrieved by HMM searches were compared based on the number of clusters, where a large number of clusters denotes a high relative degree of inferred functional diversity, and the size of the clusters that is an indication of the amount of gene duplication. Clusters were also searched for the annotated multi-domain transcripts to allow for comparisons of clusters across and within each domain. The number of sequences used for clustering varied substantially between domains with 15 865 adenylation; 10 118 ketoreductase; 9832 ketosynthase; 7854 thiolation; 7025 enoyl-reductase; 3324 dehydratase; 2492 acyl transferase; and 1085 thioesterase domains, following dereplication of sequences. There are also likely some false positives from the HMM search with 7887 of the 202 024 total sequences containing internal stop codons that may be from the translation of a spurious open reading frame that was coincidently similar to the HMM. These false positives as well as some truncated sequences appear in the clusterings as outlying spots. Also, sequence depth may artificially inflate or deflate the size of each cluster. The BUSCO scores for each transcriptome used in clustering were similar so this is not likely to be a dramatic effect. Likewise, this is unlikely to affect the number of clusters, only the size, since only a relatively large number of similar sequences would generate a cluster.
The clustering of thiolation domains represent an example of low diversity and low copy number with only 2 clusters formed when the acyl carrier protein was added into the dataset and a small trail off of the largest cluster containing one of the ZmaK thiolation domains (Figure 9). The main cluster has several subclusters, one containing BurA transcripts, a second containing the triple KS transcripts, a third that has several transcripts with adenylation and ketosynthase domains that are not 1 of the 4 annotated transcripts from Figure 1. There is some resolution of the subclusters separating the annotated transcripts from each other but they are very tightly linked with many internal edges. The other main cluster exclusively contains acyl carrier protein sequences from each of the transcriptomes. These differences can be seen when viewing an alignment of the binding sites from A. carterae for the phosphopantetheinyl transferase that activates the thiolation domain (Figure 9 insert). Most of the domains have similar positively charged residues following and negatively charged residues preceding the invariant serine that serves as the site of phosphopantetheinate attachment. For one of the ZmaK sites the negatively charged residue is instead positively charged and for the acyl carrier protein there is a methionine. The acyl carrier protein from Escherichia coli also has a methionine showing how conserved this residue is in the acyl carrier protein making this gene easy to distinguish from other modular synthases. Thus, there is a clear segregation of fat synthesis from other small molecule synthesis in the thiolation domain clusters that is irrespective of any implied gene origin such as horizontal gene transfer from bacteria in the case of BurA and ZmaK. This can also be used to validate the data to some degree as the ACP cluster contains approximately 230 sequences averaging to 4 per transcriptome. A. carterae has 3 readily identifiable acyl carrier proteins indicating that a reasonably expected number of sequences are present in the ACP cluster.
Cluster plot of the thiolation domains.
Other domains with limited clusters include the dehydratase, enoyl reductase, and thioesterase domains (Figure 10). The dehydratase domains (Figure 10(A)) form a single cluster of sequences including those from the triple-KS and ZmaK transcripts as well as an ancillary cluster that is annotated as a “Domain of Unknown Function” by the NCBI COG database and is similar to dehydratases involved in tyrosine metabolism by BLAST. The low copy number of the dehydratase domain is in contrast to the enoyl reductase clusters that are numerous (Figure 10(B)). Diversity is still low with a single cluster containing the ZmaK transcript and 2 other clusters that do not contain annotated transcripts but there are also many sequence fragments that form satellite points and do not cluster. The thioesterase domain clustering (Figure 10(C)) contains 3 low abundance clusters all containing annotated transcripts.
Protein sequence clusters of dehydratase, enoyl reductase, and thioesterase domains.
For the ketosynthase and adenylation domains the domain count was very high with several unknown clusters (Figure 11), similar to the enoyl reductase clustering but with a larger number of clusters. For the adenylation domain clustering (Figure 11(A)) the number of sequences was the largest with over 15 000 unique sequences. The adenylation domains from the ZmaK and BurA transcripts were found in separate clusters but for the ketosynthase domain (Figure 11(B)) the annotated transcripts all occupy a single cluster with poor resolution of subclusters and include the condensation domains from the TeCATe transcript. Both domains produced several clusters that do not contain annotated transcripts and the ketosynthase domains involved in fat synthesis labeled “FabB” form a distinct low abundance cluster.
Adenylation and Ketosynthase protein sequence clusters.
This pattern of large clusters of single domain transcripts that are similar to domains from the annotated transcripts and very small conserved clusters of fat synthesis genes appears to reverse for the acyl transferases and ketoreductases (Figure 12). Despite acyl transferases being one of the lowest abundance domains and ketoreductases one of the highest, both clusterings contain large clusters of sequences similar to the fat synthesis genes FabD (acyl transferase) and FabG (ketoreductase). There are small clusters of acyl transferase domains from the triple KS and BurA transcripts with very few single domain transcripts. The cluster containing FabD like transcripts is quite large with many of these transcripts containing ankyrin repeats that promote and regulate protein-protein interactions. 60 There is also a small cluster of single domain transcripts that are not similar to the annotated transcripts. Likewise the large ketoreductase clusters (Figure 12(B)) are comprised of the annotated transcripts contained in a single cluster, 3 clusters that do not contain annotated transcripts and a final very large cluster containing the FabG gene.
Acyl transferase and ketoreductase protein sequence clusters.
Principal components plots of the domain counts for each cluster (Figure 3(B)) gave similar results to the overall domain counts with no association between toxins produced or phylogenetic group to principal component positions. Principal component axis 1, which accounted for 91.2% of the variance differed mainly in the expansion of BurA-like domains with species on the left portion of the graph possessing large numbers of BurA-like domains while those on the right had very few. This did not correlate to intact BurA transcripts and the level of expansion was not always consistent, e.g. Karenia brevis was found to have 512 BurA-like adenylation domains but only 13 BurA-like thioesterase domains (Supplemental Table S2). In both cases this count was much higher than other species, but not equivalent.
Phylogenetic inference was attempted on the 2 smallest datasets, acyl transferases and thioesterases, to determine ancestry and compare the results to the clustering results. The resultant trees (Supplemental files 1 and 2) had zero or near zero bootstrap support for all major and minor branches up to the final bifurcations indicating that determination of ancestry was not possible for these datasets and methods. The highest scoring trees were able to replicate the clustering results to some degree with major clades mirroring the clusters formed indicating that there was some phylogenetic signal present. Also, very similar sequences or assembly variants were visible with high bootstrap support at the distal branches.
Discussion
Modular synthases are abundant in the core dinoflagellates
The goal of this study was to investigate the abundance, diversity, domain arrangement, and evolution of the suite of enzymes likely to participate in the synthesis of dinoflagellate toxins by focusing on dinoflagellates and the synthetic domains found consistently within dinoflagellates independent of existing model frameworks. There is an inherent need to study these synthetic pathways since dinoflagellate toxins are the largest known natural products with high potential toxicity,11,61 are synthesized using many non-canonical and interesting chemistries,25,28 and have potential therapeutic uses;61,62 but we have very little understanding of how they are synthesized. Based on isotopic labeling studies it is clear that they are predominantly made of acetate units incorporated by polyketide synthases23,25 with the occasional amino acid or other carboxylic acid used as starter and extender units via non-ribosomal peptide synthases. 27 In both cases a condensation reaction incorporates a chemical unit into a growing molecule that is subsequently modified either during elongation or following the synthesis of a large portion of the molecule. 16 In order to facilitate discussion we have combined polyketide synthases and non-ribosomal peptide synthases into the term “modular synthases” to encompass both condensation reactions and the general similarities of their chemistry and genetics. For polyketide synthases, the condensation reaction is performed by a ketosynthase that usually incorporates acetate from malonyl CoA but can also facilitate the addition of other short carbohydrates 26 with the release of carbon dioxide. Non-ribosomal peptide synthases use adenylation domains to pass a specific substrate, often an amino acid, to the condensation domain with microcystin being a common example. 63 Adenylation domains can also be found in the same module as ketosynthases in a hybrid system as is the case for BurA and ZmaK that participate in the synthesis of burkholderic acid in Burkholderia species and zwittermicin in Bacillus species, respectively.45,46 The BurA and ZmaK synthetic pathways are also important to mention because the module has been fragmented in their respective bacterial genomes with separate modules occurring on distant regions of the chromosome while BurA additionally serves an unusual role in bridging pathways. Due to the processive nature of these modular synthases the pathway is generally encoded as syntenic modules with domains in a more or less linear fashion that can be used to predict the final product of synthesis. 64 Although many domains can come into play in a trans fashion, 39 the most common trans-acting domains are acyl transferases and thioesterases. These domains also do not need to be collinear with other synthetic modules and have been shown to be synthetically active when whole genomic sections have been cloned. 65 A cursory BLAST analysis of the published Polarella glacialis 66 genome shows that domains are commonly found in tandem repeats of the same domain with different domains found on different scaffolds with the exception of common multi-domain transcripts such as the triple KS (Supplemental Table S4).
Dinoflagellates regulate their gene expression largely post-transcriptionally67,68 making linear encoding of the modular synthase domains obsolete. Unsurprisingly, the vast majority of modular synthase domains have been fragmented and duplicated, similar to what has been shown for other gene families in dinoflagellates such as actin and translation initiation factors.40,69 In this study adenylation, ketosynthase, and ketoreductase domains were frequently observed as single domain transcripts, as has been observed previously.7,38,43 Both single and multi-domain transcripts of modular synthases occurred in high abundance in all core dinoflagellates and their distribution was not correlated with taxonomy, toxicity, or toxin type (Figure 4). This apparently ubiquitous synthetic capacity argues that secondary metabolite synthesis is a common feature of all core dinoflagellates, a theory supported by observations that polyketide synthesis genes are found in species that do not produce known polyketide toxins. 70 Similarly, the only phosphopantetheinyl transferase, the enzyme required to activate thiolation domains and initiate secondary metabolite synthesis, found in all core dinoflagellates was able to activate a NRPS based reporter system indicative of natural product rather than lipid synthesis. 71 This is in contrast to syndinean dinoflagellates and other alveolates that had a much lower abundance of synthetic domains with all domains in similar abundance (Table 2). This is likely due to serial duplication that is a hallmark of core dinoflagellate evolution 72 and has been shown to affect the evolution of the synthetic pathway for saxitoxin in particular. 73
Single domain transcripts exhibit domain specific patterns of duplication
Multidomain transcripts were observed in all core dinoflagellates. The triple KS and BurA-like transcripts are more or less intact across almost all of the core dinoflagellates and can be readily found by simple domain counting (Supplemental Table S3) or by looking for large transcripts with ketosynthase domains. The ZmaK-like and TeCATe transcripts are less robust and are often truncated or have missing domains but are still readily recognizable. It appears that the BurA-like and ZmaK transcripts were horizontally transferred from bacteria since they are largely absent in the syndiniales but are present in a number of bacterial species as part of conserved synthetic pathways. Although modular synthases were almost entirely absent from Amoebophyra species, multi-domain polyketide synthases were found in Hematodinium in this study (Table 2) as well as another separate transcriptomics study that determined them to be cytosolic in nature. 74 The sequence arrangement is very similar between the A. carterae transcriptome and Hematodinium genome polyketide synthases. Similarly in Toxoplasma and Cryptosporidium there are multi-module PKS genes similar to the dinoflagellate triple KS used as a model here that are theorized to process fatty acids.75,76 Dinoflagellates are known to make many poly-unsaturated fatty acids77,78 and these triple KS genes may be involved. Hematodinium, unlike the alveolate Chromera velia and diatom Triceratium dubium, has adenylation domains and condensation domains similar to the TeCATe transcript. Thus, it is possible that the triple KS and TeCATe transcripts were present in some form in the dinoflagellate common ancestor. They could either have been modified or lost such as in the Amoebophyra species that infect dinoflagellates and parasitize essential fatty acids from their host, or kept intact in species like Hematodinium that infects crustacean hosts not known to make the polyunsaturated fatty acids found in dinoflagellates.
The origin of the single domain transcripts is much harder to ascertain due simply to the sheer number of very similar sequences. Previous studies have focused on phylogenies of adenylation and ketosynthase domains that could be annotated using traditional nomenclature.34,43,79 This makes sense considering that these 2 domains represent the workhorse enzymes of modular synthesis and there is precedent for the gain or loss of a domain being diagnostic for toxicity. 80 Traditional nomenclature of polyketide synthases, however, is largely based on whether a gene is eukaryotic or prokaryotic and whether it is multi-domain or an assemblage of single domains, i.e. type I and type II, 16 and dinoflagellates have been shown to possess genes similar to both eukaryotic and prokaryotic models that are both single and multi-domain. 34 This nomenclature combined with distantly related model organisms such as humans and yeast is therefore not very useful when trying to unravel the mechanisms underlying dinoflagellate modular synthases. Traditional datasets and nomenclature can also be misleading when trying to annotate sequences since protists are notoriously under-sampled in public databases relative to yeast and vertebrate models as well as prokaryotes. In Kohli et al, 51 264 ketosynthase and ketoreductase single-domain transcripts as well as 24 multi-domain PKS transcripts were found in G. excentricus and G. polynesiensis transcriptomes using the BLAST2GO pipeline and HMMs based on annotations from previous studies. The present study using HMMs created from BLAST searches in A. carterae of the aforementioned domains in known multi-domain transcripts yielded 156 additional ketosynthase and ketoreductase domains in single and multi-domain transcripts for the 2 Gambierdiscus species, although more is not necessarily better, especially when confirming predictions experimentally is still out of reach. All possible genes could play a role in toxin synthesis ignoring bias from annotations in model organisms since it is clear that what is atypical chemistry for model organisms appears to be the norm for toxin synthesis in dinoflagellates. 28 This is also true of the BurA and ZmaK genes themselves that are atypical for prokaryote polyketide synthesis modules but appear to have been successfully transferred and retained in dinoflagellates.
The clustering analysis is in some ways more informative than phylogeny and annotation in that it gives an indication of the level of gene duplication for a domain within dinoflagellates and allows visualization for a large number of sequences without the preconception of a bifurcating style of evolution during speciation. It is also important to include as many domain types and not focus on ketosynthases alone since the loss of an acyl transferase or thioesterase can result in a truncated structure as been hypothesized based on chemical comparison of pinnatoxin and gymnodimine to spirolides. 28 The underlying hypothesis is that domains with sequence similarity may perform similar functions or be involved in similar pathways since this is often the primary constraint on evolution but neofunctionalization is also a possibility. Thus, the clusters were colored according to the presence of domains from the known multi-domain transcripts as a way of binning the clusters and begin to ascertain the functions of the many single domain transcripts. One case is the large number of single adenylation and ketosynthase domain transcripts that are similar to the annotated transcripts as well as other clusters of unknown similarity (Figure 11). Two reasonable explanations for this diversity are that the domains themselves were serially duplicated and fragmented from parent multidomain transcripts resulting in gene expansion, or that there was a functional constraint forcing domains acquired by other means as single domain transcripts to evolve convergently and form multidomain transcripts. It is possible that both are happening, for example, gene duplication for the discrete cluster of BurA-like adenylation domains and convergent evolution for the ZmaK-like adenylation domains that are linked to another cluster of adenylation domains of unknown origin. It is unclear if the ketosynthase domains from the central cluster containing multi-domain transcripts are performing similar functions and the outlying clusters are performing different functions such as chain length factors, or if ketosynthase domains are being acquired faster than convergent evolution is acting. For the thiolation domains convergent evolution is more likely considering that a single cluster encompasses domains from several different multi-domain transcripts while the acyl carrier proteins have their own cluster (Figure 9). This would make sense considering that dinoflagellates only have between 1 and 3 phosphopantetheinyl transferases that can activate these thiolation domains. 71
The acyl transferase and ketoreductase domain clusterings are especially interesting as the only case where the gene for fat synthesis is present in very large clusters with small clusters containing domains from the multi-domain transcripts (Figure 12). This was first described in the Symbiodiniaceae and described as FabD-like Trans ATs. 43 While horizontal transfer and duplication of entire fat synthesis gene clusters has been shown,81,82 the extensive gene duplication suggested by the data presented here for the acyl transferase and ketoreductase genes would be unprecedented. Convergent evolution is unlikely given that fat synthesis is usually tightly regulated and none of the other fat synthesis genes show this type of clustering. It is possible that FabD and FabG like genes were coopted for some other function following an initial duplication and gene expansion followed. This would also indicate that these domains are performing a function separate from the multi-domain transcripts given that that they almost always have intact acyl transferase or ketoreductase domains. The triple KS is a special case here since the second ketosynthase is annotated as an acyl transferase containing ketosynthase by the conserved domain database of NCBI and the acyl transferase HMM only detected a domain in some transcripts but not others. This means that a Trans-acting acyl transferase is possible for some of the triple KS modules if the ketosynthase has lost the acyl transferase functionality, but this is speculation given these data. In general the acyl transferase and ketoreductase clusters of unknown similarity were probably acquired later or gene duplication occurred in early dinoflagellates since they are in very low abundance in the basal species, for example, A. carterae only has acyl transferases that are BurA-like (8 copies) and FabD-like (10 copies) and only 2 of 31 ketoreductases are found in the unknown cluster (Supplemental Table S2).
The thioesterase and dehydratase clusterings paint a very different picture than the other abundant and diverse domains as one of the few cases where the domain count is consistently low with small clusters (Figure 10). There is still some gene expansion such as the Bur-A like thioesterases in K. brevis and G. spinifera that appear to have been duplicated along with other BurA-like domains, just to a lesser degree (Supplemental Table S2). This small number of thioesterases in most species indicates that for very large toxins the number of synthetic complexes is low or that synthesis is highly iterative since a thioesterase is usually necessary to terminate each portion of synthesis. 16 However, it is important to remember that the “low abundance” of thioesterases is a relative description since thioesterases are more than 9-fold more abundant in the core dinoflagellates than in the outgroup species (Table 2). The dehydratases, like the thioesterases, are usually encountered in multi-domain transcripts (Figure 3). Ketosynthase and ketoreductase domains on the other hand are abundant as single domain transcripts. When looking at the chemical structure of many dinoflagellate toxins the acetate units are frequently hydroxylated, indicating that the ketone has been modified by a ketoreductase but not a dehydratase. 28 These hydroxyls then frequently form epoxide bonds resulting in the “zipped up” structures of brevetoxin and yessotoxin. This makes the abundance of enoyl reductases strange since they would theoretically act after the dehydratases to further saturate the polyketide but the enoyl reductases are much more abundant than the dehydratases (Figure 10). It may be that many of the enoyl reductases have been coopted to operate on a substrate other than polyketides or that the dehydratases act as a chokepoint in synthesis and that their abundance is under tighter regulation or selection pressure. Given the number of enoyl reductase fragments (Figure 10(B)) it may also be that this gene is subject to a much higher level of gene duplication but that not all of the transcripts are being translated. Either way, the large number of enoyl reductases relative to dehydratases in dinoflagellates is in stark contrast to what is frequently described in prokaryote and fungal models where regulation of gene expression is much better understood and domain abundance directly correlates to the structure of the final product.
The phylogenetic analyses attempted on the thioesterase and acyl transferase domains had no bootstrap for all major nodes in spite of being able to produce clades with similar structure to the clustering output in the highest scoring trees (Supplemental Files). The only nodes with bootstrap support above 70% were those containing sequence variants from a single species or assembly variants from a single transcriptome. This is not surprising since gene copy number has made sequence phylogeny difficult for dinoflagellates in the past.1,2,40 Also, given the amount of horizontal gene transfer the concept of orthology become difficult to prove in general, 3 and in this case a functional approach is more useful if the goal is to extend hypothesis to biochemical characterization.
In general there was a lack of condensation domains despite a large number of adenylation domains in all the core dinoflagellates. Although the condensation domains in the TeCATe transcript used to construct the HMM are similar to canonical condensation domains it is quite possible that there are other condensation domains not associated with multi-domain transcripts. It is certainly true that condensation domains can have their own specificity in natural product synthesis forming both amide and epoxide bonds without the aid of adenylation domains. 83 Condensation domains are unlikely to play a large role in toxin synthesis in dinoflagellates given their almost ubiquitous use of acetate and general lack of amino acids although the frequent use of glycolate as a starter is conspicuous, 28 and an unknown trans-acting condensation domain may be critical in initiating toxin synthesis.
Scaffolding domains and single domain transcripts are associated with toxin synthesis
Given their abundance, one could speculate that it is largely the single domain genes that are responsible for toxin synthesis with multi-domain genes like the triple KS responsible for the synthesis of poly-unsaturated fatty acids or portions of toxins like the acyl chains. It is also possible that these multi-domain genes or modules within them act on specific segments of toxin synthesis either individually or iteratively as has been proposed several times.38,42,80,84 If it is mostly single domain genes involved in toxin synthesis then the thiolation domains with tetratricopeptide repeats may be important in scaffolding protein domains and providing reaction centers for the large complexes necessary to synthesize toxins. 85 The fact that the fusion of a thiolation domain and a tetratricopeptide repeat is never found in conjunction with another domain and only present in the core dinoflagellates also correlates to toxin synthesis via single domain genes since none of the syndiniales or other alveolates make large polyketide toxins. Also the acyl transferase domains may be involved in their own scaffolding or reaction center bridging given the occurrence of ankyrin repeats in many of the FabD like acyl transferase containing transcripts. 60 The interplay between scaffolding by thiolation domains and reaction center bridging by trans-acting acyl transferases may be a driving force in the evolution of modular synthesis in dinoflagellates. Specifically, the number of acyl-transferases relative to thiolation domains increases as one moves from the most basal Gymnodiniales to the more distal Gonyaulacales (Figure 5). This shift is also evident in the decrease in the mean number of thiolation domains in a transcript (Figure 6). The occurrence of multiple thiolation domains in tandem within a transcript was first observed in the K. brevis transcriptome and appears to be a hallmark of the Gymnodiniales that include species that make sterolysins and brevetoxin10,22,86-88 as well as the Suessiales that can make zooxanthellatoxin and zooxanthellamide. 61 Unfortunately this is not diagnostic with many species that do not have a described toxin such as Pelagodinium beii having a higher average (1.47) than Karlodinium veneficum (1.28) that makes karlotoxin. The Gambierdiscus species had the highest average number of thiolation domains among the Gonyaulacales (1.58 and 1.68) that otherwise had low averages indicating that multiple thiolation domains may be one strategy in synthesizing long polyketides such as ciguatoxin. 89 Unfortunately, many polyketides in dinoflagellates are likely undescribed if they do not act as a toxin in humans making it difficult to correlate the synthesis of dinoflagellate polyketides to any molecular results.
Conclusion
In general there was no overarching signal relating domain count or domain expansion to toxin production as shown in the principal components plots (Figure 4). This is largely due to the abundance of modular synthase genes among all core dinoflagellates investigated. So if core dinoflagellates are all making polyketides, what is their purpose? Karlodinium veneficum is the only case where an ecological role has been identified, i.e. prey capture, 12 but it is also the only case where the toxin is found readily outside the cell. Another proposed role for a toxin is mediating redox potential by brevetoxin in the chloroplast of Karenia brevis. 90 This helps explain why complex polyketides are found in photosynthetic species as well as the apparent association between function and synthesis in the chloroplast since it is a major source of redox stress. Thus, focusing on “toxin” synthesis may not be advantageous in the long run versus understanding the modular synthases in dinoflagellates as a whole and their biological role within dinoflagellates. Just as subtle differences in the availability of a thioesterase or acyl transferase can radically alter the final structure of a polyketide, assays that identify known toxins can falsely label a species or strain as being non-toxic despite that organism making polyketides that are only subtly different than the toxin standards.
The data presented here shows long-term evolution along the entire scope of dinoflagellate history with the acquisition of tetratricopeptide repeats fused to thiolation domains in the core dinoflagellates and the increase in acyl transferase domains as a major component of the synthetic domain population, specifically the FabD-like acyl transferases. It also shows short-term evolution with rapid increases in the copy number of certain domains that was first shown in Symbiodinium species, 43 and appears to be a universal feature of dinoflagellate evolution that could also explain why many of the larger toxins are unique to certain lineages. While it seems like a natural progression to use molecular datasets from dinoflagellates to make predictions about the functionality of synthetic domains, existing datasets have been validated with species very distantly related to dinoflagellates, and protists in general, making these predictions unlikely to be realistic. For example the Beedessee et al paper from 2019 43 used up to date methods to predict the substrates of adenylation domains in dinoflagellates resulting in tryptophan, phenylalanine, and glycine. This is unlikely to be true since tryptophan and phenylalanine are not found in described dinoflagellate natural products that would utilize adenylation domains. Also, using the same method as the Beedessee et al paper to predict the substrate for the A. carterae adenylation domain of BurA, as well as from the original BurA sequence from Burkholderia, similarly results in phenylalanine, but this was shown to actually be a methionine modified to a propanal by radioisotopic labeling in the bacterium. 45 Thus, while abundant molecular data for many dinoflagellate species is certainly a boon in the study of dinoflagellate biology and toxin synthesis in particular, it must be tempered with biochemical validation to determine functionality and to start the process of unraveling synthetic pathways.
Supplemental Material
sj-pdf-1-evb-10.1177_11769343211031871 – Supplemental material for A Global Approach to Estimating the Abundance and Duplication of Polyketide Synthase Domains in Dinoflagellates
Supplemental material, sj-pdf-1-evb-10.1177_11769343211031871 for A Global Approach to Estimating the Abundance and Duplication of Polyketide Synthase Domains in Dinoflagellates by Ernest P Williams, Tsvetan R Bachvaroff and Allen R Place in Evolutionary Bioinformatics
Supplemental Material
sj-xlsx-2-evb-10.1177_11769343211031871 – Supplemental material for A Global Approach to Estimating the Abundance and Duplication of Polyketide Synthase Domains in Dinoflagellates
Supplemental material, sj-xlsx-2-evb-10.1177_11769343211031871 for A Global Approach to Estimating the Abundance and Duplication of Polyketide Synthase Domains in Dinoflagellates by Ernest P Williams, Tsvetan R Bachvaroff and Allen R Place in Evolutionary Bioinformatics
Supplemental Material
sj-xlsx-3-evb-10.1177_11769343211031871 – Supplemental material for A Global Approach to Estimating the Abundance and Duplication of Polyketide Synthase Domains in Dinoflagellates
Supplemental material, sj-xlsx-3-evb-10.1177_11769343211031871 for A Global Approach to Estimating the Abundance and Duplication of Polyketide Synthase Domains in Dinoflagellates by Ernest P Williams, Tsvetan R Bachvaroff and Allen R Place in Evolutionary Bioinformatics
Supplemental Material
sj-xlsx-4-evb-10.1177_11769343211031871 – Supplemental material for A Global Approach to Estimating the Abundance and Duplication of Polyketide Synthase Domains in Dinoflagellates
Supplemental material, sj-xlsx-4-evb-10.1177_11769343211031871 for A Global Approach to Estimating the Abundance and Duplication of Polyketide Synthase Domains in Dinoflagellates by Ernest P Williams, Tsvetan R Bachvaroff and Allen R Place in Evolutionary Bioinformatics
Supplemental Material
sj-xlsx-5-evb-10.1177_11769343211031871 – Supplemental material for A Global Approach to Estimating the Abundance and Duplication of Polyketide Synthase Domains in Dinoflagellates
Supplemental material, sj-xlsx-5-evb-10.1177_11769343211031871 for A Global Approach to Estimating the Abundance and Duplication of Polyketide Synthase Domains in Dinoflagellates by Ernest P Williams, Tsvetan R Bachvaroff and Allen R Place in Evolutionary Bioinformatics
Footnotes
Declaration of Conflicting Interests:
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding:
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was funded in part by a grant from NOAA-NOS-NCCOS-2012-2002987 to A.R.P. This paper is Contribution No. 5963 from the University of Maryland Center for Environmental Science and No. 21-004 from the Institute of Marine and Environmental Technology.
Supplemental Material
Supplemental material for this article is available online.
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
