Introduction

Low bioavailable concentrations of nitrogen can limit primary productivity in many oceanic euphotic zones (e.g. [1]). In the warm, oligotrophic open ocean, these low nitrogen concentrations select for nitrogen-fixing organisms that can efficiently convert atmospheric N2 to bioavailable forms [2]. While our understanding of nitrogen-fixing organisms in the oceans is evolving to include non-autotrophic diazotrophs and other unexpected physiologies (e.g. [3,4,5,6]), the filamentous, colony-forming cyanobacterium Trichodesmium is still considered a critical oceanic nitrogen fixer [3, 7].

Mariners have known filamentous Trichodesmium spp. as ‘sea-saw dust’ for hundreds of years because of the massive surface blooms they can form resembling small, water-suspended wood shavings [7]. Trichodesmium filaments can aggregate in natural communities, forming 1–4 mm colonies of essentially two morphologies (i.e., radial tufts or spherical puffs; [8]) that are visible to the naked eye and thus aided these early observations. Fitting with this long-term recognition, scientists defined six morphologically described species in the late 1800s [9]. Still, oceanographers did not recognize their central role in N2 fixation until the 1960s (e.g. [2, 3]). Researchers now know that Trichodesmium has a wide distribution in the tropics and subtropics [3, 10] and, even though some appear to have lost N2 fixation capabilities [4], the genus is still an essential source of bioavailable N to the oligotrophic oceans [7, 11]. Thus, while Trichodesmium species names have existed for >100 years, experiments to understand their evolution, genomic potential, and ecological impact are still active research areas.

Members of the Trichodesmium genus are closely related with large, repeat rich genomes [12, 13]. Yet, enrichment strains and field samples can show surprising morphological and physiological character variability (i.e., pigmentation, cell size, trichome shape, growth rate, N2 fixation rate, or colony structure) and abundance differences (e.g. [8, 14,15,16]). For example, marker gene phylogenetics shows four clades of Trichodesmium [8, 14, 17], with the best bootstrap support defining the Trichodesmium thiebautii and Trichodesmium erythraeum-enriched clades I and III, respectively [8]. Recent metagenomic work from the Red Sea has shown that some single gene metrics for Trichodesmium (i.e., hetR gene) have led to misidentification of clades due to paralogous copies in certain genomes [13], but this work also supports the importance of the two major clades of Trichodesmium. Additionally, morphological and molecular fieldwork shows that members of these same two clades are commonly observed, although T. thiebautii-containing clade I is typically more abundant throughout the water column (e.g. [4, 10, 18, 19]). Thus, while there are six classically defined Trichodesmium species, T. thiebautii clade I typically dominates field populations. Despite this recognition, the internal and external factors causing the numerical dominance of T. thiebautii are poorly defined.

Herein we used metapangenomics and metaproteomics of enrichment cultures and hand-picked colonies spanning the Atlantic Ocean to define genomic features predicted to impact Trichodesmium population dynamics. Our efforts showed that predicted mobile genetic element immunity (i.e., against phage and mobile plasmids; MGE) is a defining feature of T. thiebautii, as all clade members maintain and express a conserved Type III-B CRISPR-Cas system [20].

Materials and methods

Trichodesmium colony collection

Trichodesmium colonies were collected with a hand-towed line (~150 ft) 130-µm Sea Gear plankton net on February 8th thru March 11th, 2018, during the R/V Atlantis TriCoLim cruise (AT39-05) that transected from Cape Verde to Puerto Rico (Fig. 1). Colonies were rapidly removed from the cod end and picked into tuft and puff morphologies with sterile plastic disposable Pasteur pipettes into 50 ml of sterilized seawater. These morphology-segregated samples were sequentially washed two times with sterile, local seawater, gently filtered down onto 5-µm polycarbonate membranes, and rapidly frozen in liquid N2.

Fig. 1: Map of the 2018 trans-Atlantic TriCoLim Cruise.
figure 1

Color of the station location indicates hand-picked Trichodesmium colony morphology, specifically puff (blue), tuft (black), combined (blue and black), not hand-picked (red), no metagenomic data (gray) and metaproteomic data (asterisk).

Full size image

DNA isolation and sequencing

High-quality DNA was isolated from ~50 frozen colonies per station via Qiagen DNeasy Powersoil Kit (Germantown, MD) using the manufacturer’s protocol with the following exceptions. DNA quality and quantity was determined via NanoDrop UV-Vis spectrophotometer and Qubit Fluorometer, respectively (ThermoFisher; Waltham, MA) and twelve TriCoLim samples were 150 PE Illumina sequenced by Novogene (Sacramento, CA) to a final depth of 25 Gbps. DNA was isolated from frozen T. thiebautii H94 samples using the same protocol as above and was sequenced via 250PE Illumina MiSeq at the USC Epigenome Center (1.8 Gbps total) because the original assembly in [12] was poor quality. Raw reads are available at NCBI’s SRA under the BioProject PRJNA828267.

Isolate and field MAG assembly

Both the field samples and T. thiebautii H94 reads were run through similar assembly pipelines, but H94 was assembled on KBase (https://www.kbase.us), while the former were on a Linux server. The quality of reads was checked with FastQC v0.11.2 [21] and trimmed to enhance stats using Trimmomatic v0.33 [22]. MAGs were assembled de novo using metaSPAdes v3.12.0 [23] for H94 and MEGAHIT v1.2.6 [24] for the TriCoLim samples. These programs were chosen because each yielded the best assembly stats. Binning of contigs was performed via MaxBin2 v2.2.4, quality was checked with CheckM v1.1.3 [25], and phylogenetic placement of the MAGs was determined with GTDB-tk v1.3.0 [26]. Field MAGs were dereplicated using fastANI [27] with a cutoff of 98.5% ID. Dereplicated bins >50% CheckM complete were hand refined in Anvi’o v7 [28, 29] until the contamination level was below 5%. T. erythraeum strain 2175 was downloaded from NCBI and hand-refined in Anvi’o [28, 29] to remove contaminating contigs using the TriCoLim reads, and its final genome stats were determined with CheckM [25].

Phylogenomics

Higher quality Trichodesmium MAGs (>50% complete; Supplementary Table 1) and nearest relative genomes downloaded from the NCBI Assembly page were run through the program GToTree v1.6.12 to define a initial guide tree based on 251 cyanobacterial core protein Hidden Markov Models [30, 31]. Alignment and partition files from GToTree were piped to IQtree v2.1.4-beta in ModelFinder optimality mode (models LG + F + R10) with 1000 ultrafast bootstraps to generate the phylogenomic tree [32]. The tree was visualized and edited in the interactive Tree of Life (ITol; [33]).

Metapangenomics

The metapangenomic pipeline in Anvi’o was used to characterize shared and distinct blast-defined gene clusters (GCs; mcl-inflation 10) in the MAGs and determine if these GCs were represented in the TriCoLim reads [28, 29]. Briefly, this pipeline creates a contig database for all MAGs that was then annotated with prodigal [34], COGs [35], PFAMS [36], KOFAM [37] and KEGG [38]. Reads were recruited to contigs in the Anvi’o database with TriCoLim read samples using bowtie2 v 2.4.1 [39], matching Sequence Alignment Map (SAMs) were converted to Binary Alignment Maps (BAMs) with SAMtools v1.11 [40], and BAMs were profiled across the TriCoLim read sets using Anvi’o to determine environmental auxiliary and environmental core genes (EAG and ECG, respectively). The reads were square root normalized to compress the results and allow one to visually see the presence of T. erythraeum in the recruitment heat map. COG categories per 100 kb was determined by exporting the annotation from Anvi’o, determining the COG categories per MAG, summing those results per clade, and then analyzing and graphing in R v 4.0.3 (2020-10-10) with R Studio v1.4.1103 [41]. Differences between COG category counts per clades were tested for statistical significance using ANOVA in the R package rstatix [42]. The same general pipeline was used to determine single gene copy per clade and toxin:antitoxin GCs per clade. The Anvi’o summary data was converted into a matrix via the scripts in [43] and used with the R package micropan [44] to generate Heaps’ law alpha value and genome fluidity estimates.

CRISPR-Cas analyses

We scanned all Trichodesmium assemblies and those from their nearest relatives with CRISPRCasTyper [45]. This tool aids in the sometimes difficult task of identifying and typing CRISPR arrays and disparate cas loci based on the currently defined 44 subtypes and variants [20]. Because many of the MAGs were fragmented, CRISPR-Cas system portions on other contigs are shown with double slashes if (1) the pieces were found on the edges of their located contigs, and (2) the associated cas genes are still predicted to be part of the subtype III-B, I-D, or III-D systems defined in [20]. Additional annotations for accessory genes (purple) and hypothetical genes (gray) were determined by CRISPRCasFinder, and BLAST [46,47,48].

We used clustal in the program Geneious Prime (Biomatters, San Diego, CA) to align Cas10 protein sequences from all genomes, and RaxML v8 to generate the maximum likelihood phylogenetic tree with 100 bootstraps [49].

Virome assembly

We screened for putative phage genome, prophage or plasmid fragments in TriCoLim and enrichment culture assemblies using the virsorter2, DRAM-V, and checkV pipelines for viruses and metaplasmidSPAdes for plasmids [50,51,52,53]. The contig sequences supplied from these efforts were used to generate custom blast databases [48] that were subsequently BLAST screened with Trichodesmium spacers defined above and grouped with FastANI [27].

Proteome analysis of Trichodesmium enriched field samples

The raw proteome spectra were collected from Trichodesmium colony samples published in a prior work [54] and newly analyzed for this study. Specifically, raw spectra were searched with the Sequest algorithm implemented in Proteome Discoverer 2.2 using a custom-built Trichodesmium genomic database. To avoid inflation of the sequence database and later misinterpretation of phylogenetic signals, only one version of any identical/redundant protein sequences were included in the database, with the possible phylogenetic attributions for the redundant proteins noted in downstream phylogenetic analyses. Sequest mass tolerances were set to ±10ppm (parent) and ±0.6 Dalton (fragment). Fixed Cysteine modification of +57.022, and variable N-terminal acetylation of +42 and methionine modification of +16 were included. Protein identifications were made with Peptide Prophet in Scaffold (Proteome Software) at the 80% peptide threshold minimum, resulting in an estimated peptide false discovery rate (FDR) of 1.5% and an estimated protein FDR of 0.0%. Relative protein abundances are reported as normalized total exclusive spectral counts, so only spectra corresponding to a specific peptide for a given protein were considered. This avoids the problem of overlapping peptides in the phylogenetic analysis. The values are normalized to total spectral counts identified in each sample. The peptides identified for the CRISPR-Cas proteins were further checked using the Metatryp 2.0 tool (www.metatryp.whoi.edu) [55] to ensure phylogenetic specificity of the signals.

Transcriptome read recruiting

Trichodesmium colony metatranscriptomes were downloaded from NCBI SRA (projects PRJNA381915 and PRJNA374879) and mapped against all cas genes from H94 and cas10 from MAG T. thiebautii Indian Ocean that is a representative of cluster 3. Read quality was checked with FastQC v0.11.2 [21], trimmed with Trimmomatic v0.33 [22], recruited to cas genes with Bowtie v 2.4.1, converted to BAMs with Samtools v1.11, and profiled in Anvi’o v7.0. Average read depth values were normalized with the constitutive Trichodesmium gene rotA, as in [56]. The data were visualized in R v 4.0.3 (2020-10-10) with R Studio v1.4.1103.

Results and discussion

All N2-fixingTrichodesmium genomes are “low” protein-encoding

Past work with a handful of Trichodesmium isolates shows that their genomes are low protein-encoding (i.e., ~63%) and enriched in selfish DNA elements;[12] however, these observations have never been studied systematically across the genus. To address this, we assembled 15 new Trichodesmium MAGs from our hand-picked colonies obtained on the 2018 Atlantic Ocean spanning TriCoLim cruise (Fig. 1) and compared them to previously published isolate genomes [12] and three MAGs from a Tara Oceans analysis [4]. Two of our previously published USC Trichodesmium Culture Collection (USCTCC) genomes, T. thiebautii H94 and T. erythraeum 2175, were significantly improved by MiSeq resequencing the former and Anvi’o refining both using reads from TriCoLim. Supplementary Table 1 lists the final refined CheckM [25] statistics for all MAGs and genomes. Similar to a prior three isolate comparison [12], Supplementary Table 2 shows that higher-quality Trichodesmium MAGs have low GC% and much lower coding (~64%) than the bacterial average of ~90% [57], and are large, with an average length of ~6.5 ± 0.9 MB, and relatively gene sparse, only encoding for an average of ~5396 ± 784 proteins.

There are four N2-fixing clades of Trichodesmium

To improve Trichodesmium cladistics, we performed phylogenomics using 251 conserved core genes from Trichodesmium genomes and MAGs >50% complete. The resulting tree in Fig. 2A shows that the N2-fixing members of the genus are divided into four major clades and suggests that T. thiebautii assembled from the Atlantic (Clade A) are phylogenomically different from those from other basins (Clade B). However, our read recruiting and metaproteomics (see below) indicate that genomes with high identity to T. thiebautii B are also present and active in the Atlantic Ocean, but they did not assemble with high quality from TriCoLim samples (e.g., Supplementary Table 1; St11_bin2_1_1 and St14_bin2_1 are >98% ANI with T. thiebautii B H94 and MAG_Trichodesmium_thiebautii_Indian). Lastly, while Delmont has shown that a distinct branch of Trichodesmium are non-N2-fixers [4], BLAST searches of the TriCoLim MAGs with T. erythraeum IMS101 nif genes confirmed that all of these MAGs encode for diazotrophy (i.e., when missing in a MAG annotation, nif genes fragments were consistently found at the end of contigs).

Fig. 2: Phylogenomic tree of Trichodesmium and nearest relatives and COG enrichment per 100Kb of Trichodesmium clades.
figure 2

In (A), TriColim bins are named by station, USCTCC strains are in bold, and names beginning in “MAG” are from (4). Trichodesmium MAGs >80% complete are color coded by clade (i.e.), T. thiebautii (Blue), T. erythraeum A (Black), T. erythraeum B (Red), while MAGs < 80% are shown in gray. Other Trichodesmium MAGs in Supplementary Table 1 were excluded from the tree due to low completion values. B shows the normalized and summed quantity of COG categories for each MAG per clade. Significantly different categories determined by ANOVA of the mean are denoted above the bracket (p < 0.05 = *, <0.01 = **, <0.001 = ***, <0.0001 = ****).

Full size image

Since there are no published isolate genomes for T. thiebautii clade A and T. erythraeum clade B, reconciling past species designations and predicting their physiological and morphological characters was not directly possible. However, we attempted to place these MAGs in context with previously isolated strains [8, 58] by comparing 16S-23S internal transcribed spacer (ITS) gene sequences via blast. The isolate ITS hits ranged from 86–100% ID to our MAGs and old strain/clade designations from Hynes et al (8) generally tracked well with our cladistics. However, the high identity hits (>99.5%) allowed us to make three conclusions. (1) T. erythraeum strains (6-1, 6-2, 6-5), that formed a weak subgroup in a prior analysis [8], are members of T. erythraeum clade B, and thus we predict this group has phycoerythrobilin-rich red cells ~6.5–9.5 µm in diameter that can form colonies or loose aggregates. (2) Trichodesmium contortum with larger diameter cells (~20–30 µm) and bright red coloration [8] is likely a member of T. erythraeum clade A. (3) The ITS does not contain enough information to determine if previous Trichodesmium hildebrandtii, Trichodesmium tenue, and Trichodesmium spiralis isolates are in either T. thiebautii clade A or B. Because of this taxonomic uncertainty, hereafter we forgo using other Trichodesmium species names and simply use the broad clade designations (i.e., T. erythraeum A & B and T. thiebautii A & B).

Trichodesmium genomes have many paralogous genes dominated by predicted mobile genetic elements

To understand broad-level genome evolution in the genus, we explored copy number enriched gene families in Trichodesmium genomes and MAGs. Our results show many paralogous GCs shared by all Trichodesmium genomes, with some found in very high copy numbers per clade. Interestingly, each clade’s top ten duplicated GCs are similar but not 100% identical in sequence or copy number (Supplementary Table 3). Thus, our data corroborate a prior finding that Trichodesmium genomes are repeat-rich [12] and show that these duplications are commonplace in situ. Annotation of these paralogous GCs shows that they are enriched (~78%) in “selfish DNA elements” like transposases, retrons, or group II introns. Since there is evidence that “selfish DNA elements” can be involved in bacterial genome re-arrangement or niche adaptation [59, 60], these results suggest that transposition or other related duplication generating processes may be important evolutionary mechanisms in Trichodesmium.

T. thiebautii MAGs are enriched in specific clusters of orthologous genes (COG) compared to T. erythraeum

To begin to understand the selective pressures driving speciation in the genus, we next characterized the genomic potential of Trichodesmium in a phylogenomic context. At first glance, the average gene number per genome is greater in T. thiebautii than T. erythraeum (Supplementary Tables 2). As shown in Fig. 2B, many of the COG categories per 100 kb in each clade are statistically indistinguishable by ANOVA. However, there are five enriched categories in T. thiebautii A, T. thiebautii B, or both. These include Lipid Metabolism (I), Intracellular trafficking (U), Defense Mechanisms (V), Mobilome (X), and genes not categorized by COG. While most of these COG enrichments were dominated by transposases, repeat-filled genes, or hypothetical genes, the Lipid Metabolism and Defense genes had many informative annotations. Closer inspection of the Lipid Metabolism genes showed that T. thiebautii A has increased acyl-carrier proteins, many of which appear to be involved in polyketide synthases or annotated with multiple functions. These findings suggest increased secondary metabolite production in this clade. However, despite the interest in understanding how Trichodesmium acquires Fe (e.g. [16, 61],), none of these clusters are predicted to encode siderophores. Lastly, the T. thiebautii Defense category was enriched in putative toxin-antitoxin proteins [62] antiphage systems [63], and CRISPR-Cas genes [20], suggesting phage interactions are a selective pressure in the two T. thiebautii clades.

One-quarter of all Trichodesmium MAGs have shared gene clusters

To explore differences between Trichodesmium clades, we examined specific gene cluster presence/absence, annotated function, and detection in our TriCoLim reads. This effort allowed us to determine if the functionalities enriched or depleted in each clade in Fig. 2B were caused by distinct, new gene clusters, paralogous duplications, or deletions. Based on BLAST clustering, there are 1454 single and paralogous GCs in the conservative Trichodesmium core found in all genomes (Fig. 3A). Thus, approximately 1/4 of each genome is conserved core gene content. The total pangenome count was 10,054 genes. Pangenome modeling with the R package micropan [44] obtained Heap’s power law alpha estimates of ~1 for all Trichodesmium MAGs together and slightly >1 for T. thiebautii and T. erythraeum MAGs individually, indicating that these pangenomes are either “completely” sampled with this dataset (i.e., closed) or slowly growing logarithmically [64].

Fig. 3: N2-fixing Trichodesmium Metapangenomic visualization.
figure 3

A shows blast-defined conserved gene clusters (GCs) in a MAG as filled colored rings (blue for all T. thiebautii, red for T. erythraeum B, and black for T. erythraeum A). Lighter fill colors indicate that those GCs are missing from that MAG. Singleton GCs (i.e., appearing in only one MAG) are mostly shown between 9 and 11 o’clock in the diagram. The innermost rings of the diagram indicate number of contributing genomes to a GC, single copy genes (SCG), number of genes in a GC, and max number of paralogs. Continuing outwards, if the GC has annotation it is marked in green, while if it does not it is white. The outermost two rings show whether a GC is environmentally core (green) or auxiliary (red; i.e., the redder the color, the less commonly the GC was observed in TriCoLim reads) and GC homogeneity (i.e., high homogeneity = all green fill). Clear groupings of clade specific auxiliary gene clusters (AGC) are labeled on the edge of the diagram. B shows BLAST ANI clustering at the top and the ANI heat map in red from 96 to 100. The black heat map shows square root normalized read recruiting to each MAG from TriCoLim (Blacker bars = higher relative read recruiting). C shows statistical analysis of singleton GCs per clade with ANOVA statistical support shown above the brackets (p < 0.05 = *, 0.01 = **).

Full size image

Others have argued that genome fluidity (ϕ), a metric of genome dissimilarity, is a better method for estimating the likelihood of identifying new genes as more genomes in a taxonomic group are sequenced [65, 66]. We determined the MAG genome fluidity values for all Trichodesmium (ϕ = 0.303 ± 0.10) and the major clades of T. thiebautii (ϕ = 0.24 ± 0.04), and T. erythraeum (ϕ = 0.18 ± 0.03). Strict interpretation of these data suggests a 30% chance of identifying new genes as more Trichodesmium genomes are sequenced—again fitting with a growing/open pangenome. Additionally, our data predicts that the likelihood of discovering new genes is higher as more T. thiebautii are sequenced compared to T. erythraeum. While it is important to note that these ϕ values will likely improve with increased numbers of genomes in each clade [65], the data are consistent with the T. thiebautii pangenome being more ‘open’ than T. erythraeum with the former likely experiencing increased horizontal gene transfer (HGT).

T. thiebautii dominated the read recruiting regardless of colony type, and intra-clade average nucleotide identity (ANI) of the MAGs was very high (Fig. 3B; black and red heat maps, respectively). Thus, in situ quantification of each MAG was not possible because of likely random read recruiting among high ANI genomes [67]. However, since this issue would likely only underestimate the abundances of each clade, we report that T. thiebautii MAGs were recruiting at least 1–2 orders of magnitude more reads than T. erythraeum from TriCoLim colonies.

Trichodesmium auxiliary gene content and genomic average nucleotide identity (ANI) recapitulate the phylogenomic signal

While the predicted Trichodesmium core N2-fixing genome makes up ~1/4 of the genes, many auxiliary GCs are also detected. As shown in Fig. 3A, some auxiliary GCs were only found in one genome (i.e., singletons), while others associate with specific clades. The environmentally accessory genes (EAGs; i.e., not found in situ) to environmentally core genes (ECGs; i.e., found in situ) ratio shown on the outer ring indicates that many, but not all of these auxiliary GC bins, are commonly detected in Atlantic Ocean Trichodesmium colonies. Coloring the rings of Fig. 3A by phylogenomic group shows that the auxiliary gene content, average nucleotide identity (ANI; Fig. 3B), and phylogenomics of core genes (Fig. 2A) give the same relationships between Trichodesmium clades. Additionally, statistical analysis of singleton genes shows an uneven distribution in the genus, with T. thiebautii genomes maintaining significantly more (Fig. 3C). These empirical data are consistent with the genome fluidity results above and suggest mechanisms that increase novel gene content, like horizontal gene transfer, are more common in T. thiebautii.

We next took the GCs in each clade-specific bin, highlighted in Fig. 3A, to characterize enriched functionalities (Supplementary Fig. 1). The largest groups of clade-specific genes are found in the primary division between T. thiebautii AB and T. erythraeum AB, where the former shares 313 GCs and the latter shares 315, respectively. Percentage normalized COG analyses of these conserved GC bins showed four things: (1) non-COG categorized GCs dominate those found in all bins (ranging from ~44 to 79%), and 18–75% of these non-categorized GCs are hypotheticals, (2) the Tery-AB bin has much more COG diversity than the similarly sized Thieb-AB bin (Supplementary Fig. 1), (3) Thieb-AB, Thieb-B, and Tery-B bins are enriched in mobilome sequences (~10% of the bin’s GCs), while in Tery-A and Tery-AB, the mobilome GCs only account for ~5% of GCs, and (4) the Thieb-AB bin has a higher percentage of Defense COGs. While our data show that specific CRISPR-related gene duplications are common in Trichodesmium MAGs (Duplicated GCs; Supplementary Table 3), the Thieb-specific bin is enriched in CRISPR-Cas immunity genes.

T. thiebautii encodes a complete Type III-B CRISPR-Cas system, while T. erythraeum does not

The expanded, Trichodesmium-nearest relative phylogenomic tree in Fig. 4A shows cas gene detection in the lineage. These data demonstrate that the non-N2 fixing Trichodesmium, T. thiebautii, Hydrocoleum and Okeania MAGs all encode CRISPRs. In contrast, none of the 10 T. erythraeum MAGs analyzed here have them. Additionally, we could not identify any cas10 gene hits in any T. erythraeum MAGs in our dataset or from NCBI [13]. Finally, the observation that the T. erythraeum IMS101 genome is closed [12] and completely missing the CRISPR-Cas system, supports its absence from T. erythraeum in general.

Fig. 4: Presence of Trichodesmium CRISPR systems in a phylogenomic context and Cas10 phylogenetics.
figure 4

Presence or absence of CRISPR-Cas genes in a Trichodesmium MAGs and their nearest relatives phylogenomic context (A) and a maximum likelihood tree of their Cas10 protein sequences (B). In (A), the color-coded, directional shapes on the right represent detected Cas genes (yellow), CRISPR arrays (black), Cas accessory genes (purple) and hypothetical genes (gray) that were annotated by as described in the Materials and Methods. Lighter color indicates lower confidence in the annotation. Double-slashes are contig break positions near the annotated CRISPR-Cas systems, indicating that some clusters are fragmented due to breaks in the assemblies. Gene lengths are not drawn to scale. In (B), the color coding corresponds to T. thiebautii (Blue) and other relatives (Black). Cas10 protein sequences are in Supplementary Table 12.

Full size image

As the cas10 gene is diagnostic for the Type III-B CRISPR predicted to be encoded by T. thiebautii and can show significant sequence variation [68], we performed phylogenetic analyses of Cas10 protein sequences to explore the origins of this system in the lineage. The Cas10 maximum likelihood phylogeny shown in Fig. 4B suggests two-to-three Type III-B systems in T. thiebautii. Additionally, this tree indicates that these systems are likely ancestral because the phylogeny of each of the three distinct sequence clusters is roughly congruent with the phylogenomic signal shown in Fig. 4A. However, careful comparison of both trees shows that all three Type III-B Cas10 protein clusters are not conserved in every T. thiebautii assembly. BLASTN searches confirmed that the missing cas10 genes were present at contig breaks in our T. thiebautii MAGs (i.e., St18_bin1, St19_bin1, and St16_bin2_tuft) corresponding to clusters 1 & 2 in Fig. 4B. The most straightforward interpretation of these data is that most T. thiebautii assemblies do not have cluster 3, and perhaps it is currently disappearing from the Trichodesmium pangenome. Fittingly, cluster 3 is undetectable from our best-assembled MAG, T. thiebautii H94 isolate genome (566 contigs). Thus, while there is variation in III-B CRISPR-Cas system copy number in T. thiebautii, its loss in T. erythraeum is a phylogenetically constrained defining difference between these major Trichodesmium clades.

Generally speaking, CRISPR-Cas systems protect the cell from mobile genetic elements (MGEs; phage and mobile plasmids) via a sequence-based, targeted genome degradation [69, 70]. Many different CRISPR-Cas systems that vary in gene content and recognition molecule (RNA vs. DNA) have been described [20]. That said, while all CRISPR-Cas variants appear to provide memory-driven immunity against MGEs, the Type III-B subtype, predicted in numerically abundant T. thiebautii clades, requires active RNA transcription for function, can use other CRISPR arrays in addition to its own, and provides better protection against phage protospacer mutagenic evasion [71].

Mechanistically, Type III-B CRISPR-Cas systems operate in three steps (1) adaptation: recognition and incorporation of transcribed 30–50 bp protospacers (i.e., DNA or RNA sequences of invading MGEs; typically mediated by Cas1 or Cas1-reverse transcriptase (RT) fusion proteins, respectively [72, 73]) into CRISPR arrays as spacer sequence DNA “memories” of past attacks, (2) expression: spacer RNAs are expressed as precursor CRISPR RNA (crRNA), and (3) interference: sequence-specific crRNAs guides interfere with invading phage or plasmids by the action of the Cas10 protein [68, 69]. The absence of a Cas1-RT fusion protein in T. thiebautii suggests that the primary adaptation targets for this system are transcribed DNA MGEs. In contrast, an HD superfamily nuclease domain in T. thiebautii Cas10 proteins indicates that the interference step is likely cleaving both RNA and transcribed DNA [68, 72, 74]. Importantly, these DNA spacer sequences also provide ‘fingerprints’ of past MGE attacks that link phage/plasmid sequences with the CRISPR-Cas system containing host (e.g. [75,76,77]).

Predicted phage genome fragments assembled from TriCoLim 100% match T. thiebautii CRISPR spacers

We next sought to identify predicted MGEs from colony assemblies and determine if they matched T. thiebautii spacer sequences. While metaplasmidSPAdes identified several putative plasmids in enrichment and field samples (data not shown), none matched any T. thiebautii spacers. We also could not detect phage particles/genomes from the enrichment MAGs; however, the TriCoLim assemblies revealed 1000 s of putative phage genome fragments with contigs sizes ranging from 1000 s to >100kbps (data not shown).

Next, we asked whether these phage genomes sequences matched the T. thiebautii spacers. This effort identified seven 100% ID hits and 29 more with ANI > 90% covering ≥93% of the spacer (Supplementary Table 4). We conservatively picked the latter ID and coverage level because Type III-B crisper systems can function with mismatches, a feature that requires phage to delete ‘whole’ spacer-protospacer targets from their genomes to escape degradation [78, 79]. Unfortunately, these spacer-matching putative phage DNA fragments only ranged from 1755 to 5628 bps and were thus too small to identify the phage. All spacer-matching contigs were categorized as virstorter2 category 2 (i.e., likely phage DNA) and contained many predicted hypothetical viral genes, while one also is expected to encode a transposase (Supplementary Table 5). None of the fragments drew hits from known phage sequences in Genbank or IMG or any other gene. It is noteworthy that many of these putative phage genome fragments were detected multiple times from independently assembled TriCoLim stations (Supplementary Table 5; fastANI groupings), suggesting that some consistent phage particles were present across the transect. As past research shows that high phage relatedness selects for CRISPR-Cas systems [69, 80, 81], these data suggest that the T. thiebautii CRISPR-Cas system is defending against a relatively conserved phage group.

T. thiebautii CRISPR-Cas systems are expressed in situ

Identifying conserved Trichodesmium spacers and putative phage genome fragments suggests that the CRISPR-Cas systems are active in the field. To verify, we screened our TriCoLim Trichodesmium colony metaproteomics dataset for in situ Cas protein expression (protein identifications are provided as Supplementary Tables 6–8). In this re-analysis, we identified 3498 proteins and 68058 peptides. After binning the detected proteins by clade, T. thiebautii proteins were >10x more abundant across the transect than either T. erythraeum clades (Fig. 5A—Red, Pink, Blue, and Cyan filled colored bars, respectively). This protein detection dataset agrees well with our metagenomic read mapping in Fig. 3B, where T. thiebautii consistently dominated the sequencing reads. Major metabolic proteins such as the photosystems and nitrogenase were detected in high levels across the transect (Fig. 5B) and originated mainly from T. thiebautii proteins. Non-COG categorized proteins were ×50 more abundant across the transect than even these core metabolic functions, consistent with this being one of the most enriched categories in the T. thiebautii assemblies. Additionally, the in situ expression of these non-categorized proteins suggests that they are required for environmental growth and highlights the importance of characterizing them further.

Fig. 5: Trichodesmium protein abundances across the TriCoLim transect.
figure 5

A Protein abundance data sorted into Trichodesmium phylogenetic groups. Proteins were normalized across each sample, then sorted into the respective phylogenetic group and summed. Error bars indicate the standard deviation of the averaged biological replicates. Quantitation is displayed as normalized spectral counts (see “Methods”). Core and T. thiebautii proteins are much more abundant than those derived from T. erythraeum. B Protein abundance data sorted by biological function. Again, proteins were normalized across each sample, then sorted by COG function and summed. Selected, highly abundant functions are shown. C Summed normalized protein data for CRISPR-Cas and toxin/antitoxin proteins across the TriCoLim transect. D Nonsignificant correlation of CRISPR-Cas protein abundances versus T. thiebautii total protein abundance (p = 0.88).

Full size image

Proteins involved in cellular defense, including toxin/antitoxin proteins (i.e., the toxin components of RelE and MazEF and the antitoxin component of ParD) and the CRISPR-Cas system, were identified in across the transect (Fig. 5C). The CRISPR-Cas proteins did not correlate with total T. thiebautii protein abundance, suggesting that the former are not constitutively expressed (i.e., as a function of biomass) and are instead under some regulatory control. The CRISPR proteins identified included Cas10 and Cas7, and their phylogenetic assignment at the peptide level corresponded to T. thiebautii species and were assigned to Cas10 clusters 1 & 2 from Fig. 4B. Specifically, we identified peptides that were identical to those in T. thiebautii MAGS H94, St18_bin1, St16_bin2, and MAG T. thiebautii Indian Ocean, indicating that these species were contributing to CRISPR-Cas protein production in situ (Supp. Table 8). These data also show that Thieb-B clade members are present and active in the Atlantic Ocean.

Research shows that CRISPR-Cas adaptation (i.e., protospacer incorporation into spacer arrays) requires Cas1 or Cas2 to respond to new MGE threats [20, 73]. Thus the absence of these proteins in our metaproteome could suggest that the T. thiebautii CRISPR-Cas system is not actively adapting to new phage and is perhaps performing alternative functions in the cell independent of viral immunity [82, 83]. Three observations argue against this supposition. (1) self-targeting spacers (i.e., matching alternative sites in the MAG) were not identified, suggesting that interference-based gene regulation is not occurring (e.g. [82,83,84]). (2) most of the spacers detected in each MAG are distinct from those in other MAGs, suggesting that “rapid” adaptation occurs in T. thiebautii [75, 85], (3) read recruiting from Pacific Ocean Trichodesmium community data collected by others [86] shows that all annotated T. thiebautii H94 cas genes are expressed (including cas1 and cas2), and they appear to have diel periodicity (Supplementary Fig, 2; gene targets and results in Supp. Tables 9–12). Thus, while we cannot exclude alternative CRISPR functions in T. thiebautii, our data strongly suggest CRISPR-Cas mediated phage immunity is commonplace in the clade.

Conclusion

We show that the conserved maintenance of a functional CRISPR-Cas system in T. thiebautii is a defining speciation difference between the major clades of Trichodesmium. This conservation combined with singleton gene enrichment suggests that immunity allows the recipients of transduced genes to survive and thereby increase their genetic diversity (as noted in other systems [87, 88]). More research is needed to determine if CRISPR-Cas immunity conservation in T. thiebautii is the cause or the result of their numerical dominance over T. erythraeum. Because we are only just beginning to address how MGE selection maintains CRISPR-Cas systems in global populations [80], our findings also highlight the importance of future Trichodesmium virome studies. For example, will T. erythraeum be a climate change winner as predicted [15, 89] or will increased phage infectivity reduce their future expansion?