Abstract
Short time-scale observations are valuable for understanding microbial ecological processes. We assessed dynamics in relative abundance and potential activities by sequencing the small sub-unit ribosomal RNA gene (rDNA) and rRNA molecules (rRNA), of Bacteria, Archaea and Eukaryotes once to twice-daily between March and May in the surface ocean off Catalina Island, California. Typically Ostreococcus, Braarudosphaera, Teleaulax, and Synechococcus dominated phytoplankton sequences while SAR11, Sulfitobacter and Fluvicola dominated non-phytoplankton prokaryotes. We observed short-lived increases of diatoms, mostly Pseudo-nitzschiaand Chaetoceros, with quickly-responding prokaryotes including Flavobacteria (Polaribacter, Formosa), Roseovarius, and Euryarchaea (MGII), which were the exact sequence variants we also observed as temporally most-abundant in another diatom bloom at a nearby location, 3 years prior. We observed positive correlations representing known interactions among abundant taxa in chloroplastic rRNA sequences, demonstrating the ecological relevance of such interactions and their influence on the environment: 1) The kleptochlorplastidic ciliate Myrionecta 18S and Teleaulax chloroplasts (16S) were correlated (Spearman r =0.83) yet uncorrelated to Teleaulax nuclear 18S, nor any other taxon and 2) the photosynthetic prymnesiophyte Braarudosphaera bigelowii and 2 strains of diazotrophic cyanobacterium UCYN-A were correlated and each was correlated to multiple other taxa, including Braarudosphaera to a Verrucomicrobium and a Dictyophyte phytoplankter (all r > 0.8). We also report strong correlations (r > 0.7) between ciliates and bacteria and phytoplankton, possibly representing mutually beneficial interactions. These data reiterate the utility of high-frequency time-series to show rapid microbial reactions to stimuli, and provide new information about in-situ dynamics of previously recognized and hypothesized interactions.
Introduction
Natural marine microbial communities, consisting of both prokaryotes and eukaryotes, are diverse and dynamic. The interactions between microbial species and their environment and between microbial species dictate how energy and nutrients flow through the ocean (Fuhrman et al. 2015). Microbial communities are known to be seasonally variable (Gilbert et al. 2012; Fuhrman et al. 2006; Cram et al. 2014) and can show rapid responses to environmental variation, such as stratification and pulses of nutrients (Teeling et al. 2012; Needham & Fuhrman 2016). Time-series studies with sampling at various temporal-scales contribute to our understanding of different processes in the sea, such as seasonal variation or climate change. Daily or diel-scale high-resolution time-series are particularly useful for observing ecological responses to short-term perturbations, such as phytoplankton blooms and interactions of organisms, because whole microbial community turnover time is on the scale of a few days (Fuhrman & Azam 1982; Cram et al. 2014). During phytoplankton blooms, microbial communities can vary in pronounced, succession-like ways with dominant taxa shifting quickly (Teeling et al. 2016, 2012), even on time scales of one to several days (Needham and Fuhrman 2016; Needham et al. 2017).
Ecological interactions between microorganisms are of great importance in the ocean (Worden et al. 2015). Such interactions can be general, such as lineages of bacteria that repeatedly respond to increases in phytoplankton biomass and the organic matter produced by blooms (Buchan et al. 2014). However, many interactions appear to be species specific, including direct microbe-microbe interactions and can be observed at short temporal scales (Fuhrman et al. 2015). Such interactions include grazing, cross-feeding, mutualism, parasitism, symbiosis, or kleptochloroplasty (i.e., where a heterotrophic protist captures chloroplasts from another species and the chloroplast continues to function inside the grazer)(Mitra et al. 2016). Many of these interactions occur beween organisms of different domains or trophic states; e.g.,, between bacteria and eukaryotes, or between phototrophs and heterotrophs. Studying all of these organisms together allows a more complete view of components in the “microbial loop” (Azam et al. 1983).
The dynamics and ecology of microbial organisms via time-series is often assessed via sequencing of the small subunit ribosomal DNA gene (rDNA) of cellular organisms, which is conserved across all three domains of life. With current sequencing outputs from the Illumina MiSeq and HiSeq platform (paired end 2×250 or 2×300 high quality reads), it is possible to confidently discriminate taxa by as little as a single base difference in this conserved gene, which can resolve taxa at the “strain” or species level (Eren et al. 2014; Callahan et al. 2016; Tikhonov et al. 2015). We have recently shown that a single rRNA gene primer set can simulataneously assess bacteria, archaea, and eukaryotic phytoplankton via their chloroplasts (with the exception of dinoflagellates due to aberrant chloroplast sequnces) via 16S (Needham & Fuhrman 2016; Parada et al. 2016). Using the same rRNA gene primer set, the full eukaryotic community can be assessed via 18S (Needham & Fuhrman 2016; Parada et al. 2016).
A complementary approach to sequencing the rDNA is reverse transcribing and sequencing of the small sub-unit of the rRNA molecule itself (rRNA) which provides the same identity information as DNA, but the number of sequences is considered a proxy for the cumulative number of ribosomes from that taxon. This approach may reveal information about the biomasses and potential activities of taxa across the full community (Campbell et al. 2011; Hunt et al. 2013; Blazewicz et al. 2013; Lankiewicz et al. 2016). The rRNA and rDNA approaches each have benefits and uncertainties. For the rDNA, while the gene copy number varies between taxa, it is consistent within individuals of a given taxon and across time. The large majority of free living planktonic marine prokaryotes have 1-2 copies per cell (Brown & Fuhrman 2005). For chloroplasts, copy number is usually between 1-2 per chloroplast, and the number of chloroplasts per cell can vary from 1 to hundreds (depending largely on cell size; Needham and Fuhrman 2016). However, for small phytoplankton most commonly found at our location, the variation is typically low (2-4 chloroplasts for common taxa) (Needham and Fuhrman 2016). The 18S of eukaryotes, on the other hand, has a larger range in copy number, from 2 to 50,000 (de Vargas et al. 2015). Thus, comparing relative abundances for these taxa via 18S is tenuous, but the copy number relates very roughly to cellular biomass, when compared over many orders of magnitude on a log-log plot (de Vargas et al. 2015). rRNA in contrast may reflect variation of “potential activity” between and within taxa over time. However, the number of ribosomes per cell does not consistently reflect growth rate across taxa, since the relationship is very noisy and irregular between taxa, and it is not anything like a linear measure of growth rate (Blazewicz et al. 2013; Lankiewicz et al. 2016). Previous work has assessed the ratio between the rRNA and rDNA of individual taxa. This work results in an “index” that aims to examine the relative activities across taxa and describe patterns across all taxa. Such an analysis, with all its inherent complexities and complicating factors, is outside the scope of this paper.
Here we apply rRNA and rDNA sequencing to study the full cellular microbial community -- bacteria, archaea, and eukaryotes -- from seawater samples collected from the photic zone once to twice per day over about 1.5 months via an Environmental Sample Processor (ESP), which also provided continuous physical and chemical measurements. During the course of the sampling, a short-lived bloom of phytoplankton occurred, allowing us examine the dynamics and activities, before, during, and after the bloom. Additionally, we found that the members of two well-known symbioses were commonly found in high abundance during our time-series: 1.) the ciliate Myrionecta and the chloroplasts of cryptophyte Teleaulax (Johnson et al. 2016) and 2.) the diazotrophic cyanobacterium UCYN-A and haptophyte alga Braarudosphaera bigelowii (Zehr et al. 2016). This allowed us to assess the in-situ abundances and physiological dynamics of these relationships, which provides insight into the nature of these associations. We also explore other strong co-occurrence patterns between phytoplankton, potential eukaryotic grazers, and prokaryotes to examine potential new interactions.
Methods
Sampling
An Environmental Sample Processor (ESP) (Scholin et al. 2009), which autonomously draws seawater samples and filters them sequentially while also recording depth, temperature, conductivity, and chlorophyll fluorescence was deployed about 1 km offshore of Santa Catalina Island, California, USA in about 200 m of water (33 28.990 N, 118 30.470 W) from 13 March to 1 May, 2014. The ESP was tethered to the bottom and thus sampled via Eulerian sampling. Due to tides and tidal currents, the depth systematically varied slightly over the course of a day. Over the first 5 days, the depth of sampling was between 5 and 10 m. The ESP malfunctioned on day 6. After the instrument was restored two weeks later, samples were collected between 7 m and 15 m for the remainder of the time series (Figure 1). One L water samples for molecular analysis were drawn once (at 10 AM, 18 March to 23 March) or twice per day after the interruption of sampling (10 AM, 10 PM, 9 April to 1 May). The samples were pre-screened with 300 μm mesh and then sequentially collected on a 1 μm AE filter (Pall Gelman) and a 0.22 μm Durapore filter (Millipore) with an HA backing filter. All filters were stored in RNAlater at ambient seawater temperature until ESP retrieval (1 May), upon which the filters were stored at −80 C until processing. The ESP recorded depth, temperature, conductivity, and chlorophyll fluorescence measurements every 5 minutes.
Satellite imagery
Level-2 remote-sensing reflectances (Rrs) from Aqua MODIS (MODerate Resolution Imaging Spectrometer) were used to produce daily maps of surface chlorophyll-a concentrations over the 01 March 2014 - 30 April 2014 time period. About one third of all the images were discarded because of cloud coverage. The surface chlorophyll-a concentrations were derived by applying a local empirical algorithm to the Rrs(488)/Rrs(547) remote-sensing reflectance ratio (Trinh et al. 2017). This local empirical algorithm was parameterized specifically for MODIS using in situ measurements made in the coastal waters off the Los Angeles, CA area (i.e., our region of study).
DNA/RNA extraction
Each AE and Durapore filter was aseptically cut in half, one half for DNA extraction, the other half for RNA extraction. DNA was extracted and purified from the Durapore filters using a hot SDS extraction protocol (Fuhrman et al. 1988) and from the AE filters using a NaCl/cetyl trimethylammonium bromide (CTAB) bead-beating extraction (Countway et al. 2005). Each of these methods was modified to include lysozyme and proteinase K lysis steps (30 minutes at 37 C and 30 minutes at 50 C, respectively). Supernatant from either method was subjected to phenol/chloroform/isoamyl alcohol purification, precipitated overnight in ammonium acetate and ethanol, then resuspended in TE buffer. Each half-filter underwent two sequential lysis steps, and the extracted DNA was combined. RNA was extracted from the Durapore and AE filter via RNeasy kit (Qiagen), as per manufacturer’s instructions, including an on-column DNAse step. For the AE filters, a second DNA-removal step was performed on 10 ng of RNA with Invitrogen DNAse I, Amplification Grade (Cat. Number: 18068015).
Reverse transcription and PCR
RNA was reverse transcribed to cDNA using SuperScript III from Invitrogen using random hexamers, with 0.1 ng for the Durapore size fraction and all of the RNA from the Invitrogen DNAse treated AE RNA (input 10 ng). cDNA was cleaned up with 2x Ampure beads. Cleaned cDNA was then amplified for 30 cycles via PCR with 5’ Hot Start master mix with 515F primer (5’-GTGYCAGCMGCCGCGGTAA-3’) and 926R primer (5’ - CCGYCAATTYMTTTRAGTTT-3’), which amplifies bacterial, archaeal, and chloroplast 16S, as well as eukaryotic 18S (Parada et al. 2016). We confirmed that RNA extracts were devoid of significant DNA by performing no-RT PCR reactions and observing an absence of amplification in an agarose gel. DNA from Durapore (0.5 ng) and AE (0.05 ng) was amplified 30 and 35 cycles, respectively. For the AE DNA PCR reactions, 5 extra PCR cycles and 10-fold reduced DNA template were necessary because of an inhibitory affect of the RNALater on the extracted DNA. After PCR, products were cleaned and concentrated with Ampure beads and pooled. All samples were sequenced in one MiSeq 2×300 run at University of California, Davis.
Sequence Analysis
All commands run during data analysis and figure generation are available via Figshare (10.6084/m9.figshare.5373916). Sequences are available via EMBL study accession number PRJEB22356. Demultiplexed samples were trimmed via a sliding window of 5, trimming where the window dropped below q20 via Trimmomatic. Sequences less than 200bp were removed. For 16S analysis, forward and reverse reads were then merged with a minimum overlap of 20, minimum merged length of 200 and maximum differences (in overlap region) of 3 using USEARCH (Edgar 2013). 18S forward and reverse reads did not overlap so this merging step retains only 16S. A separate analysis is necessary for the 18S (see below). Primers were removed from the sequences with cutadapt (Martin 2011). Chimeras were detected de novo and reference based searching with QIIME identify_chimeric seqs.py and with the SILVA gold database (Pruesse et al. 2007) as the reference (Caporaso et al. 2010). Merged 16S reads were then padded to make them all the same length with o-pad-with-gaps via the Oligotyping pipeline (Eren et al. 2013). Then the sequences were “decomposed” with Minimum Entropy Decomposition (MED) default settings (Eren et al. 2014). MED decomposes the sequences into types that are distinguished by as little as a single base, based on an assessment of the underlying sequence variability and positions of high variability. We recently confirmed such an approach to be suitable for our assays via custom made marine mock communities (Needham et al. 2017). We refer to these highly resolved sequences as Amplicon Sequence Variants (ASVs).
Sequencing classification was performed on representative sequences from the 16S ASVs via the SILVA (Pruesse et al. 2007), Greengenes (McDonald et al. 2012), and PhytoREF (Decelle et al. 2015) (for classification of Chloroplast 16S) using QIIME assign_taxnomy.py with UCLUST. Additionally we classified against the NCBI database of cultured organisms (see Needham and Fuhrman 2016) with BLASTn (Altschul et al. 1990). All classifications and representative sequences are available via Figshare (Public Project Link: https://goo.gl/nM1cwe). Often classifications vary slightly between database sources, with the NCBI matches providing information “closest cultured relative” which may not be updated in the more curated databases since they may lack the most up-to-date sequences available (e.g., see UCYN-A below). For the 16S non-phytoplankton sequences, we generally display the SILVA and Phytoref classification of the 16S ASVs of non-phytoplankton and non-phytoplankton, respectively.
Some classifications were noted to be conflicting or lacked satisfactory resolution for a given database. The reasons vary but were most often due to sequences missing from databases or misleading annotations of sequences within a database. Therefore in the following cases, we performed further manual curation of our classification: 1.) Prasinophyte sequences were all manually curated because an abundant prasinophyte sequence was initially annotated as Ostreococcus and Bathycoccus via searches against the NCBI and Phytoref databases, respectively. By manual inspection, we found that the ASV that had the discrepancy perfectly matched an Ostreococcus genome sequence, and made the change throughout the manuscript (Supplementary Figure 1). 2.). We found that UCYN-A sequences are generically classified by the SILVA database as Cyanobacterial Subsection I, Family I (i.e., same groups as to Prochlorococcus and Synechoccocus). To resolve the UCYN-A sequences to their respective sub-groups we downloaded the sequences of UCYN-A1 (Zehr et al. 2008) (gi|284809060) and UCYN-A2 (Bombar et al. 2014)(gi|671395793) from NCBI and compared them to our representative ASV sequence. UCYN-A1 and UCYN-A2 are 3 bp different in the V4 to V5 amplified region we used (i.e., 99.2% similar) and each of our 2 UCYN-A ASVs matched one of each of the representative genomes at 100% (Supplementary Figure 2), thus we label them accordingly.
We split the 16S data into two datasets, the “phytoplankton” and “non-phytoplankton.” “Phytoplankton” included those sequences determined by the Greengenes taxonomy to be of chloroplastic orgin and cyanobacteria. “Non-phytoplankton” included the remaining bacteria and archaeal 16S sequences. After this step, samples that did not have greater than 100 reads in a given dataset were removed from further analysis. The low value of 100 increased sensitivity for to samples with a low number of reads but, the average was 16,576 ± 9,302 SD for non-phytoplankton and 13,426 ± 11,478 SD for phytoplankton.
The 18S amplicon sequencing products are too long to overlap given the MiSeq 2×300 forward and reverse sequencing that we used. The following steps were taken to process these data. 1.) The data were quality filtered the same as for the 16S analysis via Trimmomatic. 2.) The resulting quality-filtered forward and reverse reads were trimmed to 290 and 250, respectively. Reads that were shorter than those thresholds were discarded. The sequences were trimmed to the different lengths due to the difference in read qualities between the forward and reverse reads (forward is higher quality). Given these trimmed sequence lengths, the 16S reads will overlap but the 18S reads will not. 3.) We collected the 18S reads, by running all the reads through PEAR merging software, using default settings, and retained the unassembled reads. 4.) The forward and reverse reads of the unassembled reads were then joined with a degenerate base, “N”, between two reads. This approach is suitable for the k-mer based classifier we used (Jeraldo et al. 2014). Due to relatively low numbers of 18S sequences 646 ±403 SD reads per sample), we did not perform MED, but clustered the sequences into OTUs at 99% sequence similarity via QIIME pick_otus.py, using the UCLUST option. 18S OTU representative sequences were classified with assign_taxnomy.py via RDP against the PR2 database and SILVA database, and against the NCBI database as previously described using BLASTn. We generally use the PR2 classifications. All classifications and representative sequences are available via Figshare (https://goo.gl/nM1cwe). For the 18S data, we generally report the data as proportions of non-metazoan 18S sequences, except where specified.
Phylogenetic trees were generated for the most abundant unique sequence from the ASVs (16S) and OTUs (18S) with MUSCLE default settings with a maximum of 100 iterations (Edgar 2004). Phylogenies were reconstructed using PhyML default settings (Guindon et al. 2010). Notably, in the 18S tree, Myrionecta is divergent from the rest of the ciliates due to a very aberrant 18S sequence which has been previously reported (Johnson et al. 2004).
Statistical Analysis
Pairwise correlations between parameters were performed using eLSA (Xia et al. 2011, 2013). Missing data were interpolated linearly (typically only a few dates per dataset, except for rRNA of 18S which had 12 (of 53) days missing), and p and q values were determined via a theoretical calculation (Xia et al. 2013). Due to the two weeks of missing data and number we did not consider time-lagged correlations for this time-series. Only correlations that had p and q < 0.005 were considered significant. Mantel tests were performed in R (R Core Team 2015) with “mantel” as part of the vegan package (Oksanen et al. 2015), on a fully overlapping datasets of 41 days. We excluded the rRNA 18S dataset from this analysis since it was the most limited in number of total number of days appropriate for analysis.
Results and Discussion
Over the initial 6 days, conditions at the sample location, 1 km off Catalina Island, CA, USA, were relatively stable, with chlorophyll concentrations of 0.5-1.5 μg/L (Figure 1). After the 6th sample, the ESP malfunctioned. During the 15-day period when it was non-operational, satellite data indicated that a modest increase in phytoplankton chlorophyll occurred throughout the Southern California region. The increase in phytoplankton biomass reached closest to the location of the ESP 4 days before the instrument was repaired and sampling continued (Figure 1, Supplementary Figure 3). When sampling resumed, the chlorophyll concentrations were still elevated (though below peak levels according to satellite data) and remained between 1-5 μg/L for the remainder of the time-series. We also noted cyclical patterns within the chlorophyll data, apparently reflecting a combination of diel phytoplankton physiological variations (Dandonneau & Neveux 1997) and effects of tidal height and probably tidal current related depth variations of the instrument (Figure 1).
Overall community dynamics
Small sub-unit sequencing of rRNA and rDNA retrieved 16S sequences of archaea, bacteria and chloroplasts, as well as 18S of eukaryotes. Almost all of the non-phytoplankton taxa (bacteria and archaea) that we observed in these near-surface are presumed to have largely heterotrophic lifestyles, though many have phototrophic capability via proteorhodopsin (Fuhrman et al. 2008) or bacteriochlorophyll (Béjà et al. 2002; Schwalbach & Fuhrman 2005), and some are chemoautotrophs (e.g., Thaumarchaea (Könneke et al. 2005)). Although we recognize that divisions between classically defined trophic levels are increasingly recognized as being “fuzzy,” including the common occurrence of various kinds of mixotrophs (Worden et al. 2015), we aimed to separate the 16S data into functional guilds of classically-defined chlorophyll-a based phytoplankton and everything else (presumably largely heterotrophs). This enables a first approximation of the influence of the classic “base of the food web” (“phytoplankton”) on the rest of the food web. Due to the difficulty of accurately predicting the primary lifestyle of many eukaryotic taxa determined by 18S, because of the unknown presence of chloroplasts in some lineages and the ability of many protists to be mixotrophic (Mitra et al. 2016), we generally analyzed them as one group. For most analyses, we excluded metazoan sequences (e.g., copepods) that appeared in the data sporadically (and unintentionally, when metazoans or their fragments passed the 300 μm prefilter), and their very high 18S copy number would alter the interpretation of a primary focus, the microbial eukaryotes.
At the broadest level, the 16S rDNA sequences tended to be from non-phytoplankton taxa, especially in the smaller size fraction, averaging 58% of the total in the large size-fraction (1-300 μm) and 85% in the smaller size-fraction (0.22 - 1 μm)(Figure 2). Phytoplankton (via 16S) made up the majority of the rest of the rDNA sequences, 39% and 15%, respectively, of the large and small size fractions. Correspondingly, 18S made up 3.6% and 0.1% in the large and small fractions, respectively. In the large fraction rDNA, relative proportion of phytoplankton and non-phytoplankton varied considerably over the time-series, with the non-photosynthetic prokaryotes being more abundant following the increase in chlorophyll concentrations (Figure 2). This suggests increased heterotrophic biomass in response to the increase in algal biomass. 18S sequences reached maximum of 20% after the region-wide increase in chlorophyll (Figure 2).
In contrast to the rDNA, the relative proportion of phytoplankton rRNA was higher than non-phytoplankton rRNA in the larger size fraction (65% and 35%, respectively) including up to 75% rRNA sequences from 19 to 22 March and for most samples between 14 to 17 April in the larger size fraction. An exception was following the small phytoplankton bloom, when the 16S rRNA sequences from non-photosynthetic taxa (like the rDNA) were up to 70% in the large size fraction. In the smaller size fraction, the proportions of non-phytoplankton and phytoplankton rRNA were of roughly equal proportion (averages of 53% and 43%, respectively), with the exception of following the phytoplankton bloom when non-phytoplankton made up >95% of the rRNA sequences in the small size fraction for several sampling dates (Figure 2). In both size fractions, 18S rRNA always constituted less than 10% of the total reads, and were almost always negligible in the small size fraction (Figure 2).
Dynamics of individual phytoplankton taxa
Within the phytoplankton community, Synechococcus was typically the dominant taxon in both rRNA and rDNA, in both size fractions (Figure 3). In the larger size fraction, 2 different Synechococcus ASVs were the most abundant taxon in 24 and 44 of 50 days in rDNA and rRNA, respectively. In the smaller size fraction, a single Synechococcus taxon was dominant in all 47 rDNA samples, and in 52 of 53 of the rRNA samples, with Prochlorococcus exceeding it on a single date in rRNA.
Besides Synechococcus, in the larger size fraction, a variety of eukaryotic phytoplankton (chloroplasts) were found to be periodically most abundant, including Ostreococcus (14 days), Teleaulax (6 days), Imantonia (5 days), Braarudosphaera (3 days), and Pseudo-nitzschia (1 day) (Figure 3). Several other diatom ASVs, mostly Chaetoceros sp. and Pseudo-nitzschia sp., peaked in abundance for a few days following the small phytoplankton biomass increase (potentially already decreasing by the time we resumed sampling) (Supplementary Figure 4). Pico-eukaryotic phytoplankton taxa (Micromonas, Ostreococcus) increased in numbers steadily over the second half of the time-series, and ultimately were the second and third most represented taxa in the phytoplankton rDNA on average, overall (Supplementary Figure 4). In addition, 2 ASVs of the diazotrophic, symbiotic unicellular cyanobacterium UCYN-A were cumulatively 1.1% and 5.6% on average in the large size-fraction rDNA and rRNA, respectively. UCYN-A constituted up to 25% of all rRNA phytoplankton sequences in that size fraction (more detail below) (Supplementary Figure 4). This observation of high rRNA and rDNA presence of UCYN-A a productive upwelling region is highly significant from an oceanographic standpoint since they are diazotrophs, and may be an important source of bio-available nitrogen in these surface waters even during spring and proceeding increases in phytoplankton biomass. These short-term dynamics and observations of high prevalence complements their previously documented activity at this location throughout the year where they were reported as particularly active in summer and winter (Hamersley et al. 2011).
In the smaller size fraction, besides Cyanobacteria, Ostreococcus, Micromonas, Bathycoccous, and Pelagomonas were commonly abundant (Figure 3, Supplementary Figure 4). It appears that these taxa tended to be equally split between both size fractions, with the exception of Pelagomonas, which was primarily found in the small size fraction. Generally, Cyanobacteria tended to be a higher proportion in the rRNA than rDNA, while the opposite was the case for the eukaryotic phytoplankton in the small size fraction. This suggests that cyanobacteria are likely among the more active members of the phytoplankton community.
Dynamics of individual non-phytoplankton taxa
A single SAR11 generally dominated ASV the rDNA of the non-phytoplankton prokaryotic communities in the small size fraction (most abundant on 44 of 47 samples). The larger size fraction was dominated by a more diverse set of taxa. However, in both the smaller and larger size fractions, the rRNA dominance shifted among 11 and 10 different taxa, respectively, with particularly rapid dynamics following the increase in phytoplankton biomass (Figure 3). For the large size fraction, the same taxa tended to dominate in both the rRNA and rDNA. In the rDNA, the dominance shifted between Fluviicola (24 days), Roseovarius (12), Polaribacter (3), Roseibacillus (3), Verrucomicrobia (1), and Marine Group II Archaea (1).
Previously, we a reported on a larger diatom bloom that occurred 3 years earlier at a location about 20 km away (Needham & Fuhrman 2016; Needham et al. 2017). We also had daily-resolved data for this time-series. For that study we generated 99% OTUs and then discriminated ASVs within the abundant OTUs (i.e., > 2.5% relative abundance on any given day, or 0.4% on average). Overall, 119 of the 279 bacterial and archaeal ASVs that we detected here were also reported in the previous study. 15 of the 20 ASVs that became most abundant during the present study were also among the ASVs in the previous study. Several of the ASVs became most abundant in both time-series: members of Flavobacteria (Polaribacter and Formosa), Verrucomicrobia (Roseibacillus and Puniceicoccaceae) Marine Group II Archaea, and Roseovarius, and SAR11. The rapid day-to-day variation 8 April - 12 April is similar to what we observed previously and in both time-series, and the same ASVs of Polaribacter, Roseibacillus, and Marine Group II Archaea became most abundant in response to increases in chlorophyll, while Roseovarius, Puniceicoccaceae, and SAR11 peaked during more stable conditions. However, the response here is not as pronounced as in 2011, probably because that was a larger bloom, with presumably a larger release of organic material. The consistency between years of phytoplankton bloom response, even among exact sequence variants, is similar to those reported from the North Sea (Chafee et al. 2017).
Often, particular ASVs were observed within both size fractions, but in the smaller size fraction, their temporal variation and overall relative abundances were reduced due to the sustained high relative abundance of SAR11 ASVs (cumulatively 23% and 30% in the rRNA and rDNA in 0.2-1 μΜ, respectively versus 2% and 6% in the 1-300 μm size fraction). Besides SAR11, other non-photosynthetic taxa that were higher in the smaller fraction were the gammaproteobacteria SAR92 and SAR86, and the alphaproteobacterium, OCS116 (Figure 3, Supplementary Figure 5). Notably a Vibrio ASV peaked up to 30% for one date prior to the bloom (up to 30% in rRNA and 2% in rDNA). This is surprising considering that Vibrios are typically thought to be “bloom-responders” but here were very active before the bloom, instead.
Dynamics of individual eukaryotic taxa via 18S
The whole eukaryotic community (1-300 μm) via 18S was often dominated by metazoans, such as herbivorous copepods (Paracalanus) and larvaceans (Oikopleura, which can graze particles as small as bacteria), with a copepod OTU (Paracalanus sp.) being most represented in the rDNA on 34 of 50 samples and larvacean OTU (Oikopleura dioica) being most represented on 16 of 44 dates in the rRNA (Supplementary Figure 6). Excluding metazoans, we observed 20 different ‘most abundant’ organisms via rDNA over the 51 sample points, including 21 days by Ciliates (10 days by Myrionecta), 11 days Chlorophytes (Ostreococcus (4), Bathycoccus (5), Micromonas (2)), and 9 dates Dinoflagellates (primarily Gyrodinium and Gymnodinium 4 and 2 dates, respectively) (Figure 3, Supplementary Figure 7). Similarly, ciliates were typically the most represented taxon in the rRNA (29 of 44 days), but in contrast to the rDNA, Stramenopiles were commonly most represented (14 of 44) dates, including a MAST-3 relative of Solenicola (99% match to clone FGII (Accession: HM163289) which is usually found associated with chain-forming diatoms (Padmakumar et al. 2012; Gómez et al. 2011) (6 dates), Pseudo-nitzschia (4), and MAST-1 (Massana et al. 2004), a distant 87% best match to Rhizidiomyces (Hypochytrids) (4). As suggested by the higher number of most dominant taxa, the Bray-Curtis community similarity metric showed that the eukaryotic community via 18S was much more variable than the 16S based prokaryotes and phytoplankton (Supplementary Figure 8). The reasons that the dominance patterns vary between rRNA and rDNA are probably a combination of copy number differences and levels of activity, even given that dormant cells have a baseline level of rRNA (Blazewicz et al. 2013).
Correlations between taxa
Previously almost all correlation analyses between taxa have been between the abundance of organisms (DNA-or count-based), irrespective of activity. However for many types of interactions, it would be valuable to consider some indicator of activity level of the organisms as well. We aim to do so here by including rRNA in addition to the rDNA relative abundances in the co-occurrence patterns between taxa. We first examine known 2-organism positive interactions that occur among abundant taxa within our samples. This allows assessment of the nature of known associations in the environment. Additionally, it allows identification of organisms that may be previously unrecognized members of these associations, including those that may replace a known member under some circumstances while retaining similar function. Additionally, we examine the strong correlations across all taxa to identify possible interactions among and between domains, such as syntrophy, symbiosis, or grazing.
UCYN-A and Braarudosphaera
Researchers studying marine nitrogen fixation by molecular genetic analysis of (nifH) genes discovered a widely distributed and important group of nitrogen fixers that for several years went unidentified, but recently were found to be part of a symbiotic association (Zehr et al. 2016; Farnelid et al. 2016; Thompson et al. 2012). The organism, recently named Candidatus Atelocyanobacterium thalassa, but still commonly known as UCYN-A, is a marine unicellular nitrogen-fixing cyanobacterium with greatly reduced genome for a cyanobacterium, lacking the ability to generate oxygen (which inhibits nitrogen fixation) and possessing incomplete TCA cycle pathways (Tripp et al. 2010). Its metabolic deficiencies are evidently met by having a symbiotic relationship with algae (Thompson et al. 2012). At least four strains of UCYN-A have been reported (denoted UCYN-A1, A2, A3, and A4), based on phylogeny of the nifH sequence. These strains vary in their global distribution, size, symbiotic hosts, in addition to other genetic and probably physiological differences (Farnelid et al. 2016). The most well-supported UCYN-A symbiosis is a relationship between UCYN-A2 and the haptophyte alga Braarudosphaera bigelowii (Zehr et al. 2016; Thompson et al. 2012; Farnelid et al. 2016), in which fixed nitrogen is exchanged for organic substrates UCYN-A needs for growth. Other UCYN-A types are thought to be associated with different phytoplankton, including with species closely related to Braarudosphaera (Zehr et al. 2016).
We observed two 16S ASVs of UCYN-A, each an exact match to genomic sequences from UCYN-A types. One ASV was a perfect match to a genome sequence of UCYN-A1 (gi|284809060) and another with a perfect match to a genome scaffold of UCYN-A2 (gi|671395793) (Figure 4, Supplementary Figure 7). These two ASVs differed by 3 base pairs over the 375 base pairs 16S amplicon sequence we analyzed. The dynamics of the rDNA relative abundance of the UCYN-A1 and UCYN-A2 were similar over the full time-series (Spearman r= 0.64). There was a pronounced increase in both types from 18 April to 25 April when UCYN-A1 increased from about 0.5% to about 3% in relative abundance of all phytoplankton, while the increase in UCYN-A2 was less pronounced (it peaked to about 1.5% on 25 April). Both UCYN-A types also peaked in early March -- though the peaks were offset slightly (by 1 day via rDNA, 2 days via rRNA). Both were relatively low in early and late April.
Overall, the rRNA levels were similar in dynamics to the rDNA and to one another (Figure 4), though the mid-to-late April peaks were more similar in amplitude and timing in the rRNA than the rDNA when UCYN-A1 was about 2x as relatively abundant. The average rRNA relative abundance of UCYN-A1 and UCYN-A2 was 2.4% and 3.1%, respectively, which is was about 3 and 10x the relative concentration of their rDNA.
Likewise, Braarudosphaera bigelowii (1 bp different over 368 bp (99.7%) to an NCBI 16S chloroplast sequence from Braarudosphaera, Accession: AB847986.2, (Hagino et al. 2013) was high during March, low in early April, peaked during the middle of April and decreased after April 24 (Figure 4). The rRNA and rDNA of Braarudosphaera chloroplasts were correlated (0.64, p < 0.001), with an average rRNA to rDNA ratio of 0.93.
In general, Braarudosphaera and UCYN-A were highly positively correlated, and the best correlations were between the Braarudosphaera rDNA and UCYN-A1 rDNA (r= 0.85, Figure 4C), while the correlation to UCYN-A2 was not quite as strong (r = 0.76). Braarudosphaera rRNA was correlated to both UCYN-A1 and UCYN-A2 rRNA (r=0.81 and 0.83, respectively). UCYN-A1 rDNA was also significantly correlated to Braarudospheara rRNA (r = 0.63), but the other combinations of rRNA to rDNA and vice-versa between the taxa were not as significantly correlated (i.e., p < 0.005). Given that the literature reports a specific relationship between UCYN-A2 and Braarudospheara and between UCYN-A1 and a closely related taxon (Zehr et al. 2016), it may be that the 16S amplicon sequence does not discriminate between distinct but closely related haptophyte species that may be present with similar dynamics. An alternative speculation is that the Braarudosphaera strain(s) present at this site can be a host to UCYN-A1 and UCYN-A2.
We found that there were several other taxa highly correlated to Braarudosphaera, which implies the possibility of additional members of this known symbiosis, or at least strong affiliations. A Dictyochophyte alga (rRNA) had a particularly strong correlation to Braarudosphaera (r = 0.86), with a slightly weaker correlation between the rDNA of the two UCYN-A taxa (Figure 4). A Puniceicoccaceae (Verrucomicrobium) rRNA and rDNA was very strongly correlated to Braarudosphaera (all r >0.81) and UCYN-A1. Verrucomicrobia are often found to be particle associated (Crespo et al. 2013; Rath et al. 1998; Mestre et al. 2017; Needham & Fuhrman 2016), and were indeed enriched in the larger size fraction in our samples, suggesting possible physical attachment in an association. Other ASVs strongly correlated (r > 0.8) to Braarudosphaera were two photosynthetic algae, both Chrysochromulina, a genus that previous evidence suggests also may be a host to UCYN-A, particularly UCYN-A1 (Thompson et al. 2012; Zehr et al. 2016) (Figure 4).
Myrionecta and Teleaulax
Another known interaction between abundant taxa in our dataset is that of the ciliate Myrionecta rubra (=Mesodinium rubrum) with the photosynthetic cryptophyte, Teleaulax. In this interaction, Myrionecta is thought to phagocytize Teleaulax and retain functioning Telaulax chloroplast within the Myrionecta cells, becoming functionally phototrophic (Gustafson et al. 2000). Myrionecta is not capable of photosynthesis without this association, but, when possessing the chloroplasts, it can perform high rates of photosynthesis (Stoecker et al. 1991) and can form massive blooms (“red tides”)(Taylor et al. 1971). Myrionecta is capable of consuming many strains of cryptophytic algae from the Teleaulax/ Plagioselmis/ Geminigera clade (Peterson et al. 2012; Hansen et al. 2012; Park et al. 2007), but it appears most often associated with chloroplasts from Teleaulax amphioxae (Johnson et al. 2016; Herfort et al. 2011; Hansen et al. 2006). The exact nature and mechanisms of the interaction is unclear, but the Teleaulax chloroplasts can remain intact and harbored for days to weeks within the Myrionecta (Johnson et al. 2007; Herfort et al. 2011). The chloroplasts reportedly can be replicated within the Myrionecta with the assistance of captured Teleaulax nuclei (Johnson et al. 2007). It is unclear to what extent the relationship is most similar to kleptochloroplastic relationships, whereby chloroplasts are consumed and used until they lose function without nuclear assistance; karyoplastic ones, whereby chloroplasts can be maintained by consuming and retention of the nucleus of grazed Teleaulax, or an endosymbiosis where Myrionecta harbors Teleaulax chloroplasts permanently or nearly so. Alternatively, Myrionecta may be “farming” whole Teleaulax cells (Qiu et al. 2016), a conclusion met with firm skepticism (Johnson et al. 2016). Additionally, Myrionecta has been observed to harbor extensive microdiversity, with various strains co-existing within a location (Herfort et al. 2011) and variability in dominant types between locations (Herfort et al. 2011; Johnson et al. 2016); the nature of interactions between Myrionecta and Teleaulax may be variable between different strains. A further aspect is that the toxic dinoflagellate Dinophysis is reported to obtain its chloroplasts by feeding on Myrionecta, which in that case would be an intermediate source from Telaulax (Garcia-Cuetos et al. 2010; Sjöqvist & Lindholm 2011).
We found that Myrionecta and Teleaulax were among the most abundant and common taxa found in the eukaryotic community (18S) and phytoplankton communities (via 16S chloroplasts), respectively (Figure 3, Supplementary Figure 4 and 7). On average, the most abundant Teleaulax ASV (an exact match over the full 374 base pairs to Teleaulax amphioxeia, Supplementary Figure 9) made up 5.5% and 12.4% of chloroplast rDNA and rRNA, respectively and the most abundant Myrionecta OTU (an exact match to Myrionecta major strain LGC-2011, which was described from coastal Denmark, Supplementary Figure 10) made up 2.0% and 2.5% of rDNA and rRNA 18S sequences (Figure 5). The rDNA of these taxa increased in abundance between April 15 and April 20, and again between 24 April and 26 April. The Spearman correlation between the rDNA of these taxa (Myrionecta and Teleaulax_amphioxea_1 chloroplasts) was 0.86 and the relationship seemed highly exclusive, i.e., no significant correlations to anything else (Figure 5D).
A second abundant Teleaulax chloroplast sequence (3 base pairs different from the best match, Teleaulax amphioxeia) was also commonly detected with an average abundance of 2.0% rDNA and 1.6% rRNA. This Teleaulax chloroplast ASV was not significantly correlated with Myrionecta; however, it was significantly correlated with a Teleaulax 18S (nuclear) OTU (r = 0.67, p < 0.005, Figure 5D). Unlike the Myrionecta-Teleaulax association, these Teleaulax taxa were positively correlated with many other taxa, rDNA of Synechococcus, Alphaproteobacteria (OCS116 and Defluuivicoccus) NS5 genus of Bacteroidetes, Marine Group II Archaea, and a Sphingobacterium (all r > 0.7).
Thus, we saw a strong relationship between one ASV of one Teleaulax_1 chloroplast ASV, but not nuclei, and Myrionecta over the 1.5 month of study. Additionally, based on correlation between Teleaulax (nuclear) 18S and a second Teleaulax chloroplast, it appears the second strain of free-living Teleaulax present may not be associated with Myrionecta cells. We did observe Dinophysis in our samples (Supplementary Figure 4-5), without significant correlations to support a Myrionecta or Teleaulax interaction (i.e., secondary kleptoplastidy) specifically or consistently with these taxa at our location and timeframe; however such a statistical relationship may not be expected if the abundance of Dinophysis is not dependent on contemporaneous availability of Myrionecta-Teleaulax via a specific grazing dependency. In our study, we found the Teleaulax nuclei were regularly present at the sample site, but they did not correlate with either Myrionecta-Teleaulax chloroplast association.
Our observations of strong, consistent relationship over about 1.5 months between specific types of Myrionecta and chloroplasts from Teleaulax lends some support either the hypothesis that the relationship is an endosymbiosis (Herfort et al. 2011; Hansen et al. 2006) or that Myrionecta can maintain chloroplasts a long time with the periodic help of Teleaulax nuclei (Johnson et al. 2007). Because a single Teleaulax nucleus in a Myrionecta cell might support replication of many more captured chloroplasts than would be found in a single Teleaulax cell (Johnson et al. 2007), the possible need for Teleaulax nuclei might be hard to discern via correlations between 18S nuclear genes in our system (i.e., if the ratio of Teleaulax nuclei and chloroplasts is highly variable within Myrionecta), in contrast to the strong correlative relationship between the relative abundances of Teleaulax chloroplasts and Myrionecta 18S. Alternatively, other Teleaulax nuclei may be present but in lesser abundance (and 18S copies per cell), reducing the ability to regularly detect them in strong co-occurrence with the Teleaulax and Myrionecta.
Other correlations between taxa
To gain an understanding for how the communities (i.e. phytoplankton, non-phytoplankton 16S, 18S, DNA or RNA, different size fractions) changed, overall, in relation to one another, we performed Mantel tests. While all the different communities were significantly correlated (p < 0.001), the strongest correlations were between phytoplankton via 16S and non-phytoplankton 16S (Supplementary Figure 11). The best correlation was between the rDNA of the phytoplankton and the large or particle-attached non-phytoplankton 16S, and the slightly less so to the free-living and small non-phytoplankton 16S (Figure 6, Supplementary Figure 11). This is a similar result as we previously reported (Needham and Fuhrman 2016) where there was a stronger correlation of eukaryotic phytoplankton via chloroplasts to attached bacteria and archaea than to free-living bacteria and archaea. The rDNA of phytoplankton 16S was more related to both the rRNA and rDNA of non-phytoplankton 16S communities, than the rRNA of phytoplankton 16S was to either. These strong affiliations of phytoplankton and attached bacteria and archaea are perhaps a result of symbiotic interactions where both phytoplankton and prokaryotes benefit from the association (Amin et al. 2012, 2015). Another hypothesis is that different phytoplankton communities generate different suspended and sinking marine aggregates that in turn harbor different bacterial communities. There may be stronger relationships between rDNA versus rRNA because physical attachment, potentially limited by cell numbers (rDNA), is relatively more important than the potential activity (rRNA) of cells.
The observation of relatively weak correlation between eukaryotes by 18S to the other communities may be because grazing may be less species-specific, as has been previously concluded based on co-occurrence patterns (Chow et al. 2014). Thus, the strength of correlation between particular grazers and bacteria may not be strong since grazers may be more likely to graze bacteria of similar size than in a species-specific manner. This could result in an increase in the abundance of the grazer in response to an increase in any of a variety of similarly sized bacteria.
For pairwise correlations, several of the phytoplankton-to-heterotrophic bacteria correlations are the same as those that we previously reported, including those between Rhodobacteraceae, Polaribacter (Flavobacteria), and SAR92 to diatoms Pseudo-nitzschia and Chaetoceros (Figure 6), suggesting that these correlations are specific and repeatable between different time-series even though they were separated by 3 years and about 20 km. The associations of these prokaryotic groups with phytoplankton, especially in diatom blooms, have been reported previously, with responses at time-scales from weeks to months (Teeling et al. 2012 & 2016, Buchan et al. 2012). These interactions involve finely-resolved taxa (ASVs) consistent with the patterns uncovered in the North Sea (Chafee et al. 2017).
We also observed a group of highly positively correlated Prochlorococcus ASV to various Flavobacteria and Verrucomicrobia, indicating the shared ecosystem preferences or interactions. These taxa all steadily increased in rRNA and rDNA from about 11 April until to the end of the time-series (Figure 6).
In addition to those types of interactions that were previously described, we also observed many strong correlations between heterotrophic or mixotrophic eukaryotic taxa and potential symbionts or prey (Figure 7). Of these, only 5 taxa had strong correlations (|Spearman r | > 0.7, p < 0.005) to bacteria or phytoplankton; of these 4 were ciliates. In addition to the relationship between Myrionecta and Teleaulax described previously, the ciliate OTUs of Strombidium were shown to have correlations to a variety of bacteria, including Flavobacteria, Gammaproteobacteria, relatively rare ASVs classified as mycoplasma, and Pseudo-nitzschia. Thus it appears that these particular ciliates may have specific interactions with these bacteria, and may be good targets for future analyses to determine the nature of these interactions. We observed no strong negative correlations (Spearman r < −0.7), in contrast to many strong positive ones >0.7 or even 0.8, suggesting that the specific interactions we observe were much more likely to represent mutually beneficial interactions, without strong antagonistic relationships. This observation of strong positive without strong negative correlations in fact applies to all the associations we observed in this study, generally.
Conclusions
Our results show a rapid, day-to-day response of particular microbial taxa to changes in phytoplankton. In this case we saw only a small increase in chlorophyll (especially compared to previous studies) and yet the patterns we have previously observed persisted. Observations of microbial dynamics via rRNA and rDNA yielded somewhat similar results, though the overall proportions of taxa could change between the rRNA and rDNA, with phytoplankton often dominating the rRNA sequences. Via co-occurrence patterns, our results provide new in-situ characterizations of previously observed interactions among the most abundant taxa. They suggested that the Myrionecta-to-Teleaulax chloroplast association appeared to occur independently of other microbial associations, while UCYN-A-to-Braarudosphaera co-occurred strongly with several taxa. Overall, the study reiterates the utility of short-term time-series for understanding environmental responses and microbe-to-microbe interactions where turnover times can be very fast.
Conflicts of Interest
The authors declare no conflict of interest
Acknowledgements
We thank the Gordon and Betty Moore Foundation for funding (Grant Number: 3779) and their support of this project. We thank the Scholin Lab at the Monterey Bay Aquarium Research Institute for deployment, data acquisition, quality control, and help with experimental design. In particular we thank Christina Preston, Roman Marin III, James Birch, Scott Jensen, Brent Roman, Bill Ussler, and Kevan Yamahara for their help with the ESP. We thank the Wrigley Institute of Environmental Studies for logical support, especially Gordon Boivin. We thank the National Science Foundation for financial support (Grant Number: 1136818).
Footnotes
Contacts David M. Needham, dmneedha{at}gmail.com
Jed A. Fuhrman, fuhrman{at}usc.edu
Conflicts of Interest: The authors declare no conflict of interest.