Vitamin B1 sample collection and analysis

Dissolved (<0.22 μm) and particulate (0.22–90 μm) B1 and vitamer concentrations were measured (full details in Supplementary Methods 4). All equipment was cleaned with methanol, and samples were processed in the dark. Three to five replicate 500-ml samples were filtered (<0.17 mbar) onto 47-mm nylon filters (0.22 μm; GVS North America) and frozen immediately (−80°C) for particulate vitamin analysis. Filtrates were stored in amber HDPE bottles at −20°C until processing.

B1 and vitamers were extracted from particulates on filters with use of bead beating and drying [46]. Dissolved B1 and vitamers were quantified via solid-phase extraction C18-cartridge capture [17, 27]. B1/vitamers were quantified using a Dionex Ultimate-3000 LC system coupled to the electrospray ionization source of a TSQ Quantiva triple-stage quadrupole mass spectrometer (ThermoFisher Scientific). Transition list and details are in Supplementary Table S1 and Supplementary Methods 4. Concentrations of metabolites were determined by calibration curves run in the matrix of a quality control sample, consisting of equal volumes of each sample. B1 peak area was normalized to ¹³C-B1. Dissolved B1 concentrations were corrected with sample-specific recovery of the internal ¹³C-B1 standard. Recovery of the internal standard ¹³C-B1 in the dissolved samples was 24±7%. Estimates of dissolved B1 percent recovery vary (e.g. 55±29 [38]), likely due to sample processing, detection method, internal standard preparation, recovery calculation and calibration curves [5, 17, 27, 38, 47]. Concentrations of dissolved vitamers were corrected for recovery with values obtained from Paerl et al. [5].

Limits of quantitation (LOQ) and limits of detection (LOD) were calculated as 10× and 3× the variation in the intersample blanks for all compounds (Supplementary Table S1). For dissolved samples, the LOD was calculated as the response variation in the quality control sample rather than the inter-sample blanks (Supplementary Table S1). Chromatograms for samples with compound concentrations (B1, HMP, FAMP) between LOD and LOQ were visually inspected and analyzed batch per batch [48]. Each metabolite measurement is reported as the mean of two technical replicates and only if the metabolite was above the LOD in both HPLC–MS injections.

B1, thiamin monophosphate (TMP) and vitamers HMP (4-amino-5-hydroxymethyl-2-methylpyrimidine), FAMP (N-formyl-4-amino-5-aminomethyl-2-methylpyrimidine), AmMP (4-amino-5-aminomethyl-2-methylpyrimidine), HET (4-methyl-5-thiazoleethanol), and cHET (5-(2-hydroxyethyl)-4-methyl-1,3-thiazole-2-carboxylic acid) were measured. Compound abbreviations are provided in Table 1A. Traces of AmMP, and cHET and TMP, were detected but mostly below LOQ and excluded from further analysis (Supplementary Table S1).

One bioassay with B1-auxotrophic Vibrio anguillarum PF430–3 ΔthiE was carried out to determine bioavailable dissolved B1 [11]. Filtered water (<0.2 μm) from 4 June 2020 was diluted 1:10 with Aquil medium without vitamin and metal solution [49] and supplemented with macro nutrients (glucose 100 μM, ammonium 40 μM, phosphate 2.5 μM). Four replicates were run in parallel for initial samples (no B1 addition), an internal standard curve (5, 10, 25, 50, 75 pM) and negative control tubes (medium with no B1 addition), incubated in the dark at 16°C on a shaker. PF430–3 abundances were determined by flow cytometry as described above for RF bacterioplankton.

Nucleic acid sampling, extraction, and sequencing

For DNA and RNA extraction, 1 L of sample was gently filtered via peristaltic pump onto a 0.22-μm Sterivex filter (PES, Millipore). Filters for RNA extraction were preserved with RNAlater (Sigma-Aldrich). Filters were sealed and stored at −80°C until extraction. The filters were crushed by flash freezing and grinding. DNA was extracted with the DNeasy Blood and Tissue kit (Qiagen, Hilden, Germany) with additional lysozyme (Sigma-Aldrich) and proteinase K (Qiagen) treatments, and quantified (PicoGreen, Invitrogen). RNA was extracted with the RNeasy RNA Mini extraction kit (Qiagen), and the RNase-free DNase Set (Qiagen) was used to remove DNA. Additional DNase treatments (Turbo DNA-free kit, Invitrogen) were carried out followed by a column clean up with the RNA Clean and Concentrator-5 kit (Zymo Research, Freiburg, Germany). DNA removal was confirmed by 16S rRNA gene PCR [50].

The 16S and 18S rRNA genes were amplified using 515F-Y/926R primers [50] and KAPA HiFi HotStart ReadyMix (Roche; Supplementary Table S2). For each sample, triplicate PCR reactions were pooled and cleaned (Gene Clean kit, MP Biomedicals). Samples were indexed, purified (Agencourt XP Beads, Beckman Coulter), quantified, pooled in equimolar ratio, and sequenced with MiSeq 2 × 300 bp, v3 chemistry (Illumina) at the GeoGenetics Sequencing Core of the University of Copenhagen.

Metagenomic sequencing was carried out on a NovaSeq 6000 S4 (2 × 150 bp, Illumina) at the National Genomics Infrastructure in Stockholm, Sweden. The library was prepared using 15 ng DNA with the SMARTer Thruplex kit (Takara). Metatranscriptomic sequencing was conducted using NovaSeq 6000 S4 (2 × 150 bp, Illumina) at the National Genomics Infrastructure in Uppsala, Sweden. The library was prepared using 200 ng total RNA per sample in the Stranded Total RNA-Prep Ligation with Ribo-Zero Plus Kit (Illumina). Sequencing summary statistics are in Supplementary Table S3.

Vitamin B1 and vitamers concentrations and correlations

Concentrations of dissolved and particulate vitamin B1, HET, HMP, and FAMP were quantified (Fig. 2). The combined particulate pool of B1 and vitamers peaked on 23 June (33.9 pM), was lowest in March (5.6), and was positively correlated with bacterial production (τ = 0.52, Supplementary Fig. S3). The combined dissolved pool (sum of the mean concentrations of B1, HET, HMP, and FAMP) of the four measured compounds was also lowest in March (117 pM) but highest in October with 210 pM, after the particulate maximum.

Open in a separate window

Figure 2

Concentrations of particulate (A) and dissolved (B) B1 and vitamers (HET, HMP, FAMP) across seasons; black and gray dashed lines indicate LOD and quantification, respectively; biological replicates are shown for each time point. Bars indicate the mean of the biological replicates; one measurement of dissolved FAMP from 6 October was removed as an outlier; for compound abbreviations, see Table 1A; particulate values normalized to particulate organic carbon are provided in Supplementary Fig. S5; dissolved measurements are corrected for percent recovery.

Total B1 (dissolved and particulate) was relatively stable throughout the year with an average concentration of 92 pM (standard deviation ±12 pM). Particulate B1 ranged from 1.2 pM in March to 5.5 pM in the beginning of July (Fig. 2A). Dissolved vitamin B1 ranged from 72 to 117 pM with concentrations above 100 pM in early summer (May/June), and lower concentrations toward fall (Fig. 2B). Bioavailable dissolved B1 for 4 June was 126±10.5 pM determined by a bioassay (Supplementary Fig. S4 and Table S8). In comparison, B1 concentration for this day based on LC–MS was 117±68.8 pM.

Particulate HET concentrations were about 10× lower than B1 (0.01–0.47 pM), and dissolved HET concentrations ranged from 3.5 to 10.0 pM (Fig. 2B). Particulate HMP (3.8–28.9 pM) peaked in early summer with 28.9 pM (23 June, Fig. 2A) and was lowest in November (3.8 pM, Fig. 2A). Maximum dissolved HMP was on 23 June (27.3 pM), and concentrations were elevated from June to August, later in summer than particulate HMP (Fig. 2B). Overall, dissolved HMP ranged from 9.9 to 27.3 pM and was positively correlated to dissolved B1 (τ = 0.64). Dissolved HMP and HET were negatively correlated (τ = −0.45, Supplementary Fig. S3) and peaked in different seasons (Fig. 2B).

Particulate FAMP ranged from 0.43 to 2.55 pM with higher concentrations in summer (Fig. 2A). Dissolved FAMP ranged from 16 to 112 pM and was also highest in summer, similarly to the pyrimidine vitamer HMP (Fig. 2B). Dissolved FAMP was positively correlated to dissolved HMP (τ = 0.52). Particulate and dissolved FAMP were positively correlated with temperature (dissolved τ = 0.55, particulate τ = 0.73). Particulate values normalized to POC, a proxy for B1 or vitamer per biomass, showed highest B1/POC and vitamer/POC ratios at the end of August (Supplementary Fig. S5) at peak temperature (23.1°C).

Bacterial and eukaryotic biomass and community composition

Ciliophora dominated eukaryotic biomass (23.6 μg C l⁻¹) followed by Dinoflagellata and flagellates (16.1 μg C l⁻¹, 15.0 μg C l⁻¹; Fig. 3A). Dinoflagellata and Cryptophyta were also detected among 18S rRNA ASVs (Supplementary Fig. S6). Highest eukaryotic biomass occurred in June/July coinciding with high POC concentrations (Fig. 1D). Diatoms, Euglenophyta, and Chlorophyta accounted for less than 10% of the eukaryotic biomass (Fig. 3A). Eukaryotic plankton biomass was only positively correlated with particulate HMP (τ = 0.58, Supplementary Fig. S3A).

Open in a separate window

Figure 3

Dynamics of eukaryotes and bacterioplankton taxa across seasons. Eukaryotic biomass based on microscopy identification and enumeration (A); relative abundances of dominant bacterial orders (top 16) based on 16S rRNA gene amplicon sequencing (B).

Bacterial biomass exceeded eukaryotic biomass—210 μg C l⁻¹ vs 57 μg C l⁻¹ on average—based on microscopy, flow cytometry, and published conversion factors (Supplementary Table S9) [44], suggesting that a large fraction of B1/POC was within bacterioplankton biomass (Supplementary Fig. S5). The bacterial phyla Bacteroidota, Proteobacteria, Cyanobacteria, and Actinobacteria dominated the prokaryotic community based on relative abundance of reads of 16S rRNA gene ASVs (Fig. 3B). Flavobacteriales (Bacteriodota) were most abundant, accounting on average for 36% of reads. The picocyanobacterial order PCC-6307 was abundant from late June, correlating with an increase in Chl a, to October contributing up to 38% of the reads in summer, and decreasing toward fall with declining water temperature (Figs. 1A, ​,3B).3B). PCC-6307 also correlated positively with particulate B1 concentrations (τ = 0.64, Fig. 2A), temperature (τ = 0.58, Fig. 1A), and Chl a (τ = 0.55, Fig. 1A).

Temporal change in genes and transcripts related to vitamin B1 physiology

Twelve single assemblies, which had corresponding metatranscriptomes, were surveyed for genes encoding proteins for B1 synthesis, transport, and salvage (Fig. 4). B1 prototrophs canonically possess B1 synthesis genes thiC, thiG, and thiE at a ratio of 1:1:1; thus, a lower ratio on community level indicates a prevalence of auxotrophic populations. Gene numbers for thiG (thiazole synthase) and thiE (thiamin monophosphate synthase) were similar (0.9:1), while thiC (pyrimidine synthase) was low relative to thiE (0.64:1). No significant correlations between de novo synthesis gene counts and B1/vitamer concentrations were found except a weak negative correlation between thiE gene counts and particulate HMP (τ = −0.47, Supplementary Table S7).

Open in a separate window

Figure 4

Temporal dynamics of genes and transcripts encoding proteins for B1 synthesis, transport, and salvage; normalized counts of B1 biosynthesis genes (A) and transcripts (C) and of gene and transcripts associated with B1 or vitamer transport or salvage (B, D); metagenomic counts are normalized by RPKM and by marker genes’ RPKM (A, B) and metatranscriptomic values to the normalized metagenomic counts and shown as relative transcripts per sample (C, D); for gene information, see Table 1B.

Most B1 synthesis transcripts were accounted for by thiE and thiC (41% and 49%). Highest transcription of thiG coincided with high dissolved HET concentrations in September (Figs. 2 and ​and4C).4C). Fewest thiC genes and transcripts (signaling prevalent auxotrophy) occurred on 4 June (Fig. 4C), and transcripts were positively correlated with relative abundance of Cyanobacteria (τ = 0.67).

Genes for B1 transporters (thiB, thiT) were present throughout the year with thiB genes being most abundant, and thiB and thiT transcript abundances were similar (Fig. 4B and D). Highest transcription of B1 transporters was on 4 June and coincided with the lowest thiC transcription and lowest relative abundances of prototrophic clusters (Supplementary Fig. S7). Putative transporters for pyrimidine vitamers (thiV, thiY, cytX, ykoF) were detected in DNA and RNA (Fig. 4B and D). Gene counts for thiV, a putative HMP sodium symporter [13], were comparable to thiB, but thiV transcripts were higher than other detected B1 transporters (thiB, thiT). Similar to thiT, transcription of thiY, coding for part of a HMP and/or B1 ABC transporter [74–76], was relatively high compared to its gene count.

Genes and transcripts for B1 salvage from pyrimidine (tenA) or thiazole (thiM) vitamers were present but in low numbers (Fig. 4B and D). While tenA transcripts were relatively stable over time, thiM transcripts were more prevalent from March to June.

Metagenome-assembled genome generation, clustering, and abundances

On average, 75% of the reads from 20 assemblies mapped to bins—indicating that bins represented most of the community (Supplementary Fig. S8). Of the 3982 bins obtained, 2180 were medium- or higher-quality MAGs forming 405 bacterial clusters (≥95% ANI; Supplementary Fig. S9). The dataset included 1025 MAGs with high completeness (≥90% completeness, <5% contamination) from 195 clusters. We established B1 genotypes of clusters by screening all MAGs in each cluster for B1-related genes. We investigated the putative B1 physiology of the 195 clusters containing minimum one high completeness MAG and the top 19 most abundant clusters (Fig. 5) in more detail. The top 19 clusters included the 14 most abundant clusters overall in the dataset across time points (Supplementary Fig. S10) and five additional clusters showing high relative abundance (among top two most abundant) at individual time points (e.g. Cluster 90).

Open in a separate window

Figure 5

B1-related genotypes of the 19 most abundant clusters; heatmap shows presence (dark boxes) or absence (gray boxes) of proteins across clusters; the gradient of the dark boxes indicates the percentage of MAGs containing the protein. Number of MAGs and maximum estimated completeness of the MAGs are shown on the left side; taxonomy is shown for each cluster on phylum (Cyano: Cyanobacteria, Proteo: Proteobacteria, Actino: Actinobacteriota, Bacter: Bacteroidota), order and genus levels; relative abundances of clusters marked with asterisks are provided in Supplementary Fig. S10; prototrophic clusters with synthesis genes thiE, thiG, thiC (A), pyrimidine auxotrophic clusters with genes thiE, thiG, and transporter thiV (B), auxotrophic clusters capable of B1 transport with thiB (C), and auxotrophic clusters with putative transporters (D); white X’s indicate proteins that are considered false positives based on manual evaluation.

Vitamin B1 genotypes across microbial populations

Only 3 of the 19 abundant clusters were B1 prototrophic (possessing thiC, thiG, thiE; Fig. 5A). Two of the prototrophic clusters were picocyanobacterial (Vulcanococcus), and the third was Gammaproteobacterial (Burkholderiales BACL14). Clusters deficient in only one B1 synthesis gene were of the order Pelagibacterales (lack of thiC, Fig. 5B) and possessed the pyrimidine vitamer transporter gene thiV, adjacent to a TDP riboswitch and B1 salvage gene tenA (Cluster 81), as previously described for Pelagibacterales [13, 77]. Most Actinobacteria appeared reliant on exogenous dissolved B1 and transport via thiB to meet their need for intact B1 (Fig. 5C, Supplementary Fig. S11), while others possessed putative B1 transporters, cytX, or ThiPerm. Beyond Actinobacteria, the transporter omr1 occurred in two SAR86 clusters (e.g. Cluster 4, Fig. 5D) and two Flavobacteriales clusters (e.g. Cluster 207, Supplementary Fig. S11) with a nearby TDP riboswitch.

Among the 195 clusters containing at least one high-quality MAG (≥90% completion; <5% contamination), only 22% were prototrophic (Supplementary Fig. S11). Several Bacteroidota and Actinobacteriota clusters lacked currently described synthesis and transport systems for B1 and vitamers (Fig. 5, Supplementary Fig. S11).

Bacterial cornerstones that supply vitamin B1

Because B1 production per cell is not well constrained [12, 14] and auxotrophy/prototrophy in eukaryotic plankton is challenging to determine in-situ (due to challenges in reconstruction of environmental eukaryotic genomes), a significant unknown remains unanswered—who are key producers of B1 in-situ? In RF, biomass estimates of major eukaryotic plankton groups were not correlated with B1 or vitamers except particulate HMP (Supplementary Table S7)—thus, eukaryotic phytoplankton do not appear to be key net sources of B1 (as well as vitamers overall). Also, many common RF taxa were putative auxotrophs based on culture studies (Mamiellales, Chlorophyta; Euglenophyta; Cryptophyta) [12], except dinoflagellates and diatoms [8, 12]; Fig. 3A, Supplementary Fig. S6). No thiC transcripts classified as eukaryotic were detected, and unclassified thiC transcripts were consistently lower than bacterial thiC transcripts (Supplementary Fig. S14). High abundances of eukaryotic thiC transcripts have been reported from Diatom cultures [21], possibly the eukaryotic contribution is underestimated here by the methods selected for gene calling and mapping of metatranscriptomic reads to metagenomic assemblies. Nonetheless, there was no significant relationship between eukaryotic biomass and vitamin concentrations and a prevalence of putative auxotrophic phytoplankton taxa in RF.

Within prokaryotes, abundant B1 synthesizers were picocyanobacteria and BACL14 (Methylophilaceaea, Burkholderiales) and to a lesser extent Verrucomicrobiales (Fig. 6B). Together, these few bacterial populations accounted regularly for more than 75% of thiC transcripts. These groups are widespread and show little variation in B1-related genotype—strongly pointing to them as key B1 sources in coastal waters and beyond [11, 22, 88–90].

Cyanobacteria are key primary producers in marine and brackish systems [91–94] and “rhythm setters” for heterotrophs reliant on their organic carbon production [95]. Our in situ data extend their importance to include increasing B1/vitamer concentrations in dissolved and particulate pools (based on correlations, Supplementary Table S7), which has repercussions upon auxotroph survival and B1 trophic transfer. The mechanisms of vitamin provision may involve interactions [39, 96] or cell mortality [39, 97]. Nonetheless, vitamin measurements and correlations point to picocyanobacteria as key sources in RF, and this fits with the observation of increased B1 and/or vitamers [36], HMP, and pseudocobalamin [13, 98] availability in cyanobacterial cultures. When cyanobacteria are rare, heterotrophic BACL14 (Burkholderiales) and Verrucomicrobiales serve as prominent B1 sources based on thiC transcripts. These heterotrophs use different organic carbon sources and globally occur across freshwater and marine systems [99–102]. Speculatively the variety of lifestyles de novo B1 synthesizers possess helps maintain ecosystem B1 availability across seasons and potentially in microenvironments (e.g. on particles).

We thank Gregor Luetzenburg and Rikke Brønnum for assistance with field work, Dan Johan Kristensen for conducting the bioassay, and Weronika N. Kolodziejczyk for help with processing of POC and dissolved vitamin samples. Christian Darling, Stine Kærulf Andersen, Emil Guddal Larsen, and Ivar Thorstein Hansen from Danish Nature Agency assisted with water collection from Roskilde fjord and provided contextual data. The authors acknowledge support from the Swedish National Genomics Infrastructure (NGI). NGI in Stockholm performed DNA sequencing and NGI in Uppsala (the SNP&SEQ Technology Platform) performed RNA sequencing. Both facilities are hosted by Science for Life Laboratory (SciLifeLab) and funded by the Knut and Alice Wallenberg Foundation and the Swedish Research Council. The computations and data handling were enabled by resources provided by the Swedish National Infrastructure for Computing (SNIC) at UPPMAX (projects SNIC 2021-22-184, SNIC 2022-5-87, and SNIC 2022-6-196) partially funded by the Swedish Research Council through grant agreement no. 2018-05973. JS is financially supported by the Knut and Alice Wallenberg Foundation as part of the National Bioinformatics Infrastructure Sweden at SciLifeLab. We thank Markus Englund from the National Bioinformatics Infrastructure Sweden (NBIS) for help with uploading data to databases.