Prokaryotic and eukaryotic microbiomes associated with blooms of the ichthyotoxic dinoflagellate Cochlodinium (Margalefidinium) polykrikoides in New York, USA, estuaries

While harmful algal blooms caused by the ichthyotoxic dinoflagellate, Cochlodinium (Margalefidinium) polykrikoides, are allelopathic and may have unique associations with bacteria, a comprehensive assessment of the planktonic communities associated with these blooms has been lacking. Here, we used high-throughput amplicon sequencing to assess size fractionated (0.2 and 5 μm) bacterial (16S) and phytoplankton assemblages (18S) associated with blooms of C. polykrikoides during recurrent blooms in NY, USA. Over a three-year period, samples were collected inside (‘patch’) and outside (‘non-patch’) dense accumulations of C. polykrikoides to assess the microbiome associated with these blooms. Eukaryotic plankton communities of blooms had significantly lower diversity than non-bloom samples, and non-bloom samples hosted 30 eukaryotic operational taxonomic units (OTUs) not found within blooms, suggesting they may have been allelopathically excluded from blooms. Differential abundance analyses revealed that C. polykrikoides blooms were significantly enriched in dinoflagellates (p<0.001) and the experimental enrichment of C. polykrikoides led to a significant increase in the relative abundance of eight genera of dinoflagellates but a significant decline in other eukaryotic plankton. Amoebophrya co-dominated both within- and near- C. polykrikoides blooms and was more abundant in bloom patches. The core bacterial microbiome of the >0.2μm fraction of blooms was dominated by an uncultured bacterium from the SAR11 clade, while the >5μm size fraction was co-dominated by an uncultured bacterium from Rhodobacteraceae and Coraliomargarita. Two bacterial lineages within the >0.2μm fraction, as well as the Gammaproteobacterium, Halioglobus, from the >5μm fraction were unique to the microbiome of blooms, while there were 154 bacterial OTUs only found in non-bloom waters. Collectively, these findings reveal the unique composition and potential function of eukaryotic and prokaryotic communities associated with C. polykrikoides blooms.

Introduction natural plankton community to assess allelopathic effects on planktonic assemblages. This study revealed the presence of a parasitic dinoflagellate, Amoebophrya spp., as well as a number of novel bacteria associated with C. polykrikoides blooms.  [10]. Shinnecock Bay is a shallow, well-mixed system that exchanges with the Atlantic Ocean through the Shinnecock Inlet. Great Peconic Bay is a shallow, well-mixed system located between Long Island's north and south forks and exchanges with the Atlantic Ocean. All samples were collected between 10:00am and 12:00pm. Sample collection did not involve any protected or endangered species and did not require any specific permits as samples were collected in regions open to the public. At each sampling point, samples were first collected within the visually dense, darkly-colored bloom patches (>1,000 C. polykrikoides cells mL -1 ) and then in regions adjacent to patches where the water was not discolored (<100 C. polykrikoides cells mL -1 ), designated 'patch' and 'non-patch' samples, respectively. For molecular analysis, patch and non-patch water was filtered onto 0.2μm and 5μm polycarbonate filters and immediately frozen at -80˚C.

DNA extraction, illumina sequencing and analysis
To extract nucleic acids, 1 mL of cetyltrimethyl ammonium bromide (CTAB) buffer with fresh beta-mercaptonethanol was added to the 0.2μm and 5μm polycarbonate filters, vortexed, heated to 50˚C for 20 minutes, and frozen at -80˚C until processing. Genomic DNA extraction was performed using the CTAB method [43]. Following extraction, double-stranded DNA was quantified on a Qubit1 fluorometer using a dsDNA BR Assay kit. Alongside samples, 0.2μm and 5μm filter controls were extracted and DNA quantified (dsDNA = 0 ng/μl) as above to ensure that sample sequences were not a result from contamination during sample processing. Samples were normalized to an equal quantity of DNA for sequencing and were sent to Molecular Research Labs (Shallowater, Texas, USA) for amplicon sequencing. The 16S rRNA gene V4 variable region (~300bp) was amplified using bacterial primers A519F: 5´CAG CMG CCG CGG TAA and 802R: 5´TAC NVG GGT ATC TAA TCC [44]. The V7/V8 region of the 18S rRNA gene (~450bp) was amplified using primers 1183F: 5´AAT TTG ACT CAA CAC GGG and 1631aR: 5´TAC AAA GGG CAG GGA CG [45]. For each sample, an identifying barcode was placed on the forward primer and a 30 cycle PCR using the HotStarTaq Plus Master Mix Kit (Qiagen, USA) was performed. The following PCR conditions were used: 94˚C for 3 minutes, followed by 28 cycles of 94˚C for 30 seconds, 53˚C for 40 seconds and 72˚C for 1 minute, and a final elongation step at 72˚C for 5 minutes. Successful amplification was determined by visualizing PCR products using a 2% agarose gel. The amplification of filter controls did not result in any products and therefore were not sequenced. Samples (31 for 18S and 24 for 16S, S1 and S2 Tables, respectively) were pooled together for each respective primer region in equal proportions based on their molecular weight and DNA concentrations. Pooled samples were then purified using calibrated Ampure XP beads and subsequently used to prepare a DNA library by following Illumina TruSeq DNA library. Paired-end (2x300) sequencing was performed on an Illumina MiSeq following the manufacturer's guidelines. Sequence data was processed using the Quantitative Insights Into Microbial Ecology v1.9.1 pipeline (QIIME, http:// qiime.org [46]). Raw sequences were depleted of barcodes, paired-end reads joined, depleted of primers, demultiplexed, and quality filtered using the default parameters in QIIME. The resulting quality filtered sequences were then clustered into operational taxonomic units (OTUs) at 97% similarity with UCLUST [47] using the open reference clustering protocol and SILVA release v119 (http://www.arb-silva.de/) as the reference set. The representative sequence set was aligned using PyNAST [48] and taxonomically classified using UCLUST [47]. For 18S specifically, all non-algal OTUs were not considered in order to focus specifically on algal assemblages. Further, the classes Dinophyceae and Syndiniophyceae were considered as a single group representing total dinoflagellates and from here on are referred to as Dinophyceae. Since species specificity with the QIIME pipeline was typically not possible, representative sequences for the most abundant OTUs were extracted and species specificity (percent identity) was determined using BLAST. Similarly, for 16S, since our focus was on prokaryotes, all chloroplast and mitochondria (S2 Table) related sequences were removed from OTU tables and not further considered in analyses. In addition, OTUs identified as Prochlorococcus were reassigned to Synechococcus II as another distinct group assigned as Synechococcus was already identified within our dataset [49]. NCBI BLAST results supported this reassignment as 'Prochlorococcus' OTU consensus sequences were 100% identical to Synechococcus sp. MV0605E (accession ID KU867943.1). Furthermore, phylogenetic analysis of sequences from the informatically identified Prochlorococcus and Synechococcus suggests that they are an intermixed group (data not shown). Finally, the ecology of these two organisms further supports this reassignment as Prochlorococcus is absent from eutrophic coastal regions including estuaries whereas Synechococcus inhabits a broader niche including eutrophic estuaries [50].
For the V7/V8 region of the 18S rDNA, the 30 samples generated 3,802,417 paired end reads with an amplicon size of~450bp. After quality filtering and joining reads a total of 3,345,085 reads clustered at 97% identity into 41,828 OTUs. Overall, algae represented 24 to 97% of total reads (prior to removal of non-algal OTUs) with an average of 78% for all 30 samples (S1 Table). For the 519F/802R region of 16S rDNA, the 24 samples generated 3,951,820 paired end reads with an amplicon size of~300bp. After quality filtering and joining reads a total of 3,617,344 reads clustered at 97% identity into 16,905 OTUs (S2 Table). The 870,170 and 4,570 reads assigned as chloroplasts and mitochondria, respectively, were removed from the dataset and not considered in analyses (S2 Table).

Culture and culturing conditions
A locally isolated culture of C. polykrikoides (CP1, isolated from the Peconic Estuary, NY; [10]) was used for this study. CP1 was cultivated in sterile GSe medium [51] with a salinity of 32 PSU, made with autoclaved and 0.2 μm-filtered aged coastal Atlantic Ocean water (40.79698N, 72.46068W), at 21˚C in an incubator with a 12:12 h light:dark cycle, illuminated by a bank of fluorescent lights that provided a light intensity of~100 μmol quanta m -2 s -1 to cultures. Antibiotics (stock solution, Thermo Scientific HyClone Penicillin (10,000U mL -1 ) Streptomycin (10,000μg mL -1 ) in 0.85% NaCl) were added to the CP1 culture at a final concentration of 1% by volume to discourage microbial contamination.

Allelopathy experiment
An experiment was performed to assess the natural plankton community's response to the addition of environmentally relevant densities of C. polykrikoides. While the allelopathic effects of C. polykrikoides on natural phytoplankton communities and cultures have been assessed microscopically [23], the effects on smaller phytoplankton and bacterial communities have never been examined. During October 2014, triplicate 330mL bottles were half-filled with unamended water from Old Fort Pond, NY. A control was established whereby the other half of the bottle was filled with GSe media and, for the treatment, the other half of the bottle was filled with C. polykrikoides (CP1) culture resulting in final cell densities of 2,200 cells mL -1 . To ensure that the effects seen by the addition of C. polykrikoides were due to allelochemicals and not nutrients, saturating concentrations of N (88μM), P (3.6μM), and Si (88μM) were added to all bottles. All experimental bottles were incubated at ambient light and temperature for 48 h in Shinnecock Bay at the Stony Brook Southampton Marine Science Center [52]. At the end of the incubation, contents of each experimental bottle were filtered onto 0.2μm polycarbonate filters that were then preserved as above and sequenced, in triplicate, as individual biological replicates.

Microbial diversity and statistical analysis
All microbial diversity analyses were conducted using QIIME v1.9.1. Prior to executing diversity analysis scripts on field samples, mitochondrial and chloroplast related OTUs as well as all non-algal OTUs were filtered from 16S and 18S OTU tables, respectively. Field sample datasets were then processed using core_diversity_analyses.py scripts to rarefy samples and assess alpha-(Chao1, Shannon Diversity Index and Simpson Index) and beta-(weighted unifrac visualized via Principal Coordinates Analysis (PCoA)) diversity measurements. An Analysis of Similarity (ANOSIM) was conducted using weighted unifrac distances to assess the differences in community composition among variables.
In addition, the core microbiome of field samples for both 16S and 18S datasets were computed for each variable (patch >0.2μm, patch >5μm, non-patch >0.2μm, and non-patch >5μm) using compute_core_microbiome.py scripts in QIIME and visualized using Venny 2.0 [53]. Core microbiomes were defined as genera found in 100% of the samples within each variable (as above). Differential abundance analyses were also conducted, using the Phyloseq and DESeq packages [54][55][56] in R v3.2.3 (R Core Team 2013) to compare taxa abundances across experimental treatments as well as field sample variables (patch >0.2μm vs non-patch >0.2μm, and patch >5μm vs non-patch >5μm) for both 16S and 18S data. Briefly, raw 16S and 18S read counts were loaded into a phyloseq object and modeled with Deseq using a negative binomial distribution. Prior to modeling, counts were internally normalized for size factors using the median ratio method [54] and dispersions were estimated using a parametric fitting. Wald significance testing to test for significant log2 fold changes in abundance (α = 0.05) and p-values were adjusted using the Benjamini-Hochberg procedure to correct for multiple testing. Principal component analyses (PCA) were conducted on Hellinger transformed (y 0 ij ¼ p yij yiþ ) abundances of bacterial sequences [57] using the prcomp function (stats package) in R v3.3.3 to determine the groups of variables (region: patch, non-patch; size fraction: >0.2 and >5μm; year: 2011, 2012, 2013) that behaved similarly and the correlation of taxa (phylum or genus) to each group (biplot arrows).
To investigate potential functional differences among the microbial communities between regions, size fractions and years, predicted metagenomes were generated using PICRUST2 (Phylogenetic Investigation of Communities by Reconstruction of Unobserved States) QIIME2 plugin with default settings [58][59][60][61][62]. Prior to analysis the 16S OTU abundances were normalized (relative abundances) and the resulting predicted KEGG KO gene family and Metacyc pathway abundances were visualized with PCoA (QIIME2 Version 2019.1). PERMA-NOVA (QIIME2 diversity plugin) was used to test for significant differences among the predicted metagenomes between groups (region, size fraction, year) and was used to identify differentially abundant gene families and pathways among groups. Differential abundance of KEGG gene families and Metacyc pathways were assessed using STAMPS v 2.1.3 software [63] with a multiple sample ANOVA and Tukey-Kramer post-hoc analysis with Benjamini-Hochberg FDR multiple comparison correction. Genes and pathways were filtered for p<0.05 and an eta-squared effect size of >0.8.

Phytoplankton assemblages within-and outside-of C. polykrikoides patches
An Analysis of Similarity (ANOSIM) revealed significant differences in the community composition among patch and non-patch (p<0.001) samples, and among years (p<0.001) but not among size fractions. All alpha diversity metrics revealed no significant differences among size class (>0.2 or >5μm) or year (2011, 2012 and 2013) but indicated non-patch samples were significantly (p<0.01) more diverse than within patch samples (S3 Table). A principal coordinates analysis (PCoA) of weighted unifrac distances showed clustering among years ( Fig 1A) and sample type (patch, non-patch; Fig 1B) with the diversity between these variables being significantly (p<0.05) different.
Comparing size fractions within patch and non-patch communities. These differences among size fractions are further highlighted when comparing size fractions within patch and non-patch communities. Within Cochlodinium patches, differential abundance analyses demonstrated that the SAR11 clade, Rhodospirillales, and Oceanospirillales were all significantly  there were 259 lineages that were significantly differentially abundant, 103 and 156 of those lineages were enriched in the in the >0.2 and >5μm size fractions, respectively (S14-S21 Figs).
Patch vs non-patch within the >0.2μm fraction. Differential abundance analyses of major microbial orders and genera (i.e. taxonomic orders and genera with the highest relative abundances in Figs 5 and 6) revealed there were no significant differences between the patch and non-patch samples within the >0.2μm fraction (S3C and S4C Figs). Comparing patch and non-patch samples within the >0.2μm fraction for all genera, however, revealed that 2 lineages (Thermoplasmatales Marine Group II uncultured and Thermoplasmatales Marine Group II other) were considered significantly enriched in non-patch samples, while uncultured Acidimicrobiaceae was significantly enriched in patches (p<0.01 for all; S5A Fig).
Patch vs non-patch within the >5μm fraction. Among the >5μm fraction, differential abundance analyses demonstrated that the order Rhodobacterales was significantly (p<0.001) enriched in patch samples while the orders Rhodospirillales and Flavobacteriales were significantly (p<0.01) enriched in non-patch samples ( Comparing the microbiomes among each sample type, there were 217 shared OTUs, and 4, 1, 154 and 13 unique OTUs for patch >0.2μm, patch >5μm, non-patch >0.2μm, and nonpatch >5μm, respectively ( Fig 3C). The core bacterial microbiome was dominated by an uncultured bacterium (designated as '1' in Fig 6) from the SAR11 clade for patch and nonpatch samples in the >0.2μm fraction, and overall looked very similar (Fig 3D). For the >5μm size fraction, the patch microbiome was co-dominated by an uncultured bacterium from Rhodobacteraceae and Coraliomargarita (Verrucomicrobia), while the non-patch samples were co-dominated by Winogradskyella (Flavobacteriaceae), the cyanobacteria Synechococcus (type II) and Coraliomargarita (Fig 3D). Among these microbiomes, there were four and 154 unique OTUs for the patch and non-patch samples in the >0.2μm fraction, respectively, originating from two (an unassigned bacterium, and an uncultured bacterium from Rickettsiales) and 66 unique lineages, respectively. There were one and 13 unique OTUs for the patch and nonpatch samples in the >5μm fraction, respectively, originating from one (the Gammaproteobacteria, Halioglobus) and seven unique lineages, all of which, however, had relative abundances of <1%.

The natural plankton community's response to the addition of C. polykrikoides
The addition of a culture of C. polykrikoides resulted in marked changes in the relative abundances of the eukaryotic plankton (Fig 7). Differential abundance analyses between the control and treatment revealed that eight dinoflagellates were significantly enriched with the addition of C. polykrikoides including Amoebophrya (p<0.001), Gyrodinium spirale (p<0.001), Polykrikos (p<0.001), Karlodinium (p<0.001), Prorocentrum (p<0.001), Symbiodinium (p<0.001), Gonyaulax (p<0.05), and Pelagodinium (p<0.001; S22 Fig). While the relative abundances of Chaetoceros and Apedinella also increased in the treatment (Fig 7), the addition of C. polykrikoides caused 'other algae' to decrease by nearly 50% compared to the control (Fig 7) and caused the significant (p<0.05) decline of a number of cryptomonads, chrysophytes and other non-dinoflagellate genera (S22 Fig). Due to the mode in which allelopathic agents are secreted by C. polykrikoides, the culture was not filtered prior to its addition to the appropriate experimental bottles and, therefore, 16s data was not evaluated.

Discussion
Biotic interactions are likely to have a myriad of direct and indirect influences on the occurrence of harmful algal blooms. This study identified the microbial consortium associated with C. polykrikoides blooms and revealed that C. polykrikoides blooms were enriched in dinoflagellates and depleted in diatoms (>5μm fraction) among eukaryotes, and enriched in Rhodobacteraceae and depleted in Flavobacteriaceae, Rhodospirillaceae and the SAR11 clade among prokaryotes. Amoebophrya was highly abundant both within and near blooms while lineages of Geminigera sp., Cylindrotheca closterium, Leptocylindrus spp., an uncultured eukaryote from MAST-3J, and Pirsonia sp., were unique members of the core eukaryotic microbiomes found inside blooms. In contrast, lineages of Gonyaulax sp., Rhaphoneis sp. (Bacillariophyceae), two cryptophytes (Katablepharis sp., Leucocryptos sp.), two Prymnesiales (uncultured eukaryote and OLI16029), two uncultured eukaryotes (Mamiellophyceae and Chrysophyceae), and an uncultured eukaryote from the Syndiniales were only found outside of blooms. Among microbes, two lineages (an unassigned bacterium, and an uncultured bacterium from Rickettsiales) from the >0.2μm fraction, and one (the Gammaproteobacteria, Halioglobus) from the >5μm fraction were unique to the core microbiome of blooms. Collectively, these observations provide new insight into the composition and potential function of the eukaryotic and prokaryotic communities associated with and inhibited by C. polykrikoides bloom.
Similar to prior studies [33,39,49], high-throughput sequencing facilitated the discovery of a number of species that have yet to be described in C. polykrikoides blooms in the western hemisphere. High-throughput amplicon sequencing provided an enhanced resolution of phytoplankton community diversity, revealing parasitic species (i.e. the dinoflagellates, Amoebophrya and Duboscquella; the nanoflagellate, Pirsonia), picoplankton (i.e. Synechococcus), and species found in low abundance (i.e. Katablepharis sp., Leucocryptos sp.), many of which have yet to be identified via traditional microscopy within blooms. Consistent with prior studies utilizing microscopy, high-throughput sequencing captured most of the common, larger phytoplankton species known to be present during the summer/fall season within this region ( [8][9][10] Hattenrath-Lehmann personal observation). Prorocentrum, a dinoflagellate known to cooccur with C. polykrikoides blooms in New York [9], however, was detected at comparatively lower relative abundances than expected, perhaps due to a lack of sequences of one of the most common species found in NY, P. gracile, in the SILVA reference database [39]. For this reason, it has been recommended [33,39] that both microscopy and sequencing be utilized for comprehensive descriptions of microbial communities. The importance of-and need for-assessing the full microbiome, including both prokaryotes and eukaryotes has become increasingly obvious as it has been demonstrated that biotic interactions are better predictors of community structure than abiotic factors [64]. In the future, long-term monitoring of the base of the food chain of coastal ecosystems using a combination of microscopy and high-throughput sequencing would provide a more accurate and comprehensive assessment of the full suite of microbial consortia that are likely to improve over time as informative databases expand and become more accurately annotated [35,65].
While the parasitic dinoflagellate Amoebophrya spp. is known to form associations with many other dinoflagellate species [66][67][68], observations of Amoebophrya infections during C. polykrikoides blooms have never been reported outside of Korean waters [68][69][70][71]. Here, using high-throughput sequencing, Amoebophrya was detected in all samples analyzed, regardless of region, size fraction, or year. Moreover, Amoebophrya sequences were 95-100% identical to a Korean isolate from Namhae (NCBI accession KF791348). Amoebophrya accounted for up to 54% of the eukaryotic community and was enriched within bloom communities. While Gyrodinium spirale and Polykrikos spp. were also present and have been previously described as hosts of the parasite Amoebophrya [68], Amoebophrya was also dominant when these two potential hosts were present at very low relative abundances (<1%; Fig 2). Therefore, while not observed microscopically, we hypothesize that C. polykrikoides was the primary host of Amoebophrya. Using high-throughput sequencing, Amoebophrya was also abundant in another NY embayment, Northport Bay, in association with blooms of Alexandrium catenella [39]. Parasitic Amoebophrya infections may be more-widespread and important to HAB dynamics than previously assumed and may require the use of molecular methods to clarify their importance [71]. Since 2014, blooms of C. polykrikoides have become less intense in NY, with blooms achieving a lower maximal biomass, persisting for a shorter period of time, and being less widespread than blooms that occurred from 2006-2013 [8,10] (C. Gobler personal observation). While the role of Amoebophrya in this change is unknown, adaptation of Amoebophrya and/or other parasitic communities toward the exertion of stronger biological control could account for such a shift.
In a manner similar to several other HABs [39, [72][73][74][75], C. polykrikoides can shape planktonic community structure via the secretion of allelopathic compounds [23]. Allelochemicals produced by C. polykrikoides have the capacity to lyse and inhibit the growth of several species of dinoflagellates, diatoms, cryptophytes, haptophytes, raphidophytes, and pelagophytes in culture as well as natural plankton communities with effects being dependent on the densities of C. polykrikoides as well as target species [23]. Consistent with these known allelopathic capabilities, C. polykrikoides dominated both in-and outside-of patches, and phytoplankton community diversity was significantly lower within C. polykrikoides patches compared to non-patch samples, suggesting species vulnerable to allelochemicals were selectively eliminated. Furthermore, Chaetoceros and Gyrodinium were more abundant outside of Cochlodinium patches, a finding consistent with prior observations [9] and the density dependency of these allelopathic effects [23]. In experiments, the addition of C. polykrikoides to a natural plankton community resulted in the enrichment of Amoebophrya, Gyrodinium spirale, Polykrikos, several other dinoflagellates, Chaetoceros, and Apedinella, but the depletion of several other genera including Thalassiosira, Picomonas and Micromonas (S22 Fig). Natural community experiments conducted by Tang and Gobler [23] demonstrated that euglena populations were significantly enhanced with the addition of C. polykrikoides culture, while Gyrodinium, Scrippsiella, Skeletonema, Chaetoceros and Thalassiosira all significantly decreased. While disparities among these studies (i.e. Chaetoceros and Gyrodinium) are likely due to density, species, and strain dependent allelopathic effects [23], it is clear that 18S sequencing was capable of describing allelopathic patterns and can enhance allelopathic studies by revealing rare members of the community that would otherwise not have been captured using microscopic methods.
HABs can also influence prokaryotic communities [36-39,76]. To date, only two studies have assessed the microbiome associated with C. polykrikoides blooms, both using molecular techniques that differed from each other and from the current study [9,32]. Consistent with what was found using clone library analysis of C. polykrikoides blooms in South Korea [32], C. polykrikoides blooms during this study were dominated by bacteria of the Alphaproteobacteria and Flavobacteria lineages. In addition, blooms of C. polykrikoides in NY were also dominated by cyanobacteria, which had higher relative abundances in the >5μm fraction (up to 22%) and were more abundant than in Korean blooms (<5.4%) [32]. Rhodobacterales had higher relative abundances in patches than outside of patches, a finding similar to Park et al.
[32] who reported that Rhodobacterales were more abundant during bloom peaks and had a significant positive correlation with C. polykrikoides cell densities. Similar to NY, during blooms in Korea Alphaproteobacteria were dominated by the SAR11 cluster, Rhodobacterales and Rhodospirillales, but unlike NY blooms where the Gammaproteobacteria, Alteromonadales and Oceanospirillales were consistently present, these bacteria were rarer in Korea [32]. Also consistent with the present study, using terminal restriction fragment length polymorphism (TRFLP) analysis of 16S rRNA genes, Koch et al. [9] found TRFs consistent with those predicted for Alphaproteobacteria were more abundant in bloom samples. Despite the use of different molecular techniques, there were numerous similarities among the present study, Koch et al. , Methylophaga (<0.5%), Marinobacter (<0.5%), and Jannaschia (<0.05%) were present among field samples at varying relative abundances, however, Ponticoccus was not present in the dataset. Among these genera, Winogradskyella sp. had the highest relative abundances and differential abundance analyses demonstrated it was more abundant in non-patch samples and significantly enriched in the >5μm fraction (S4 Fig), with a PCA further confirming a strong association between this bacteria and the >5μm fraction. Despite low relative abundances, Methylophaga and Marinobacter were found to be significantly more abundant in >0.2μm fraction of both patch and non-patch samples. As with any culture, both of the Korean C. polykrikoides isolates were cultivated under environmental conditions optimal for growth of this alga (i.e. ideal temperature, light intensity, and saturating nutrient concentrations) and these conditions will likely affect the free-living and epiphytic bacteria associated with the cultures. As such, bacteria found in cultures would be expected to differ from those in the field where environmental conditions will differ and are more dynamic. Accordingly, only a single dominant genus (Winogradskyella sp.) from Korean cultures [40] was also found at high relative abundances in NY field samples.
Phytoplankton blooms, in general [77], and HABs such as Alexandrium spp., Pseudonitzschia sp., Akashiwo sanguinea, and C. polykrikoides, are known to be dominated by members of specific heterotrophic bacterial lineages including Alphaproteobacteria, Gammaproteobacteria, and Flavobacteria [32,36,38,39, [78][79][80][81][82][83]. During this study C. polykrikoides blooms in NY were also dominated by the classes Flavobacteria and Alphaproteobacteria, while Gammaproteobacteria comprised a relatively smaller portion of microbial communities compared to prior HAB studies. Within those classes, C. polykrikoides blooms shared a series of microbial similarities with blooms of A. catenella and Dinophysis acuminata reported by Hattenrath-Lehmann and Gobler [39] including dominance of the orders Flavobacteriales, Rhodobacterales, and SAR11. While many of the bacteria from these two studies are members of similar lineages [39], the NS5 marine group (Flavobacteriales), an uncultured bacterium from Rhodobacteracea, Owenweeksia spp., Perlucidbaca spp. and Limnobacter spp were the dominant genera among Alexandrium-associated field samples while Dinophysis-associated field samples were dominated by 'unassigned bacteria' and an uncultured bacterium from Rhodobacteracea. Differential abundance analyses demonstrated that C. polykrikoides blooms were significantly enriched in an uncultured bacterium from Rhodobacteraceae, the core bacterial microbiome of the >0.2μm fraction of patches was dominated by an uncultured bacterium from the SAR11 clade, while the >5μm size fraction was co-dominated by an uncultured bacterium from Rhodobacteraceae and Coraliomargarita. Collectively, the high level of phylogenetic resolution provided by high-throughput sequencing demonstrates that each of these NY HABs was associated with distinct microbiomes.
While the impacts of C. polykrikoides blooms on eukaryotic community composition and diversity were highly significant, the effects on bacterial communities were somewhat less intense. There were 154 16S OTUs that were unique to the >0.2μm, non-bloom samples, suggesting these 154 OTUs were inhibited by allelochemicals present within blooms. This finding is consistent with the observations of Koch et al. [9] who reported on four 16S T-RFLP fragments found only in non-bloom samples. There were, however, only five OTUs (four in the >0.2μm, one in the >5μm) unique to C. polykrikoides blooms and diversity analyses did not reveal significant differences between bloom and non-bloom samples for the 16S community. Given these bloom patches move horizontally with currents and tides, it is plausible that the shorter generation times of bacteria make these communities less vulnerable to the effects of C. polykrikoides, compared to eukaryotes that have significantly slower growth rates [84]. Furthermore, the cell wall of bacteria may make them more resistant to reactive oxygen species and similar allelochemicals released by C. polykrikoides [85]. Greater differences between bacterial communities may be more likely to manifest themselves over the course of the blooms as described in Park et al. [32] as organic matter inventories build and temperatures change [86]. There were far more significant differences between the size fractions within patch and nonpatch samples among bacterial communities (S6-S21 Figs), demonstrating the strong differences in particle associated bacteria vs. free-living bacteria.
Predictive tools (PICRUST2) used to determine functional potential of microbial communities revealed significant differences between size fractions, and size fractions within-and between-regions. For example, on multiple dates, the predicted metagenomic communities associated with bloom patches had significantly higher levels of metabolic pathways associated with the cytochrome c aerobic pathway than non-bloom samples. Beyond its role in mitochondrial respiration and the electron transport chain, cytochrome c can also act as an anti-oxidative enzyme, removing reactive oxygen species (ROS) [87,88]. Given that the toxic effects of C. polykrikoides are known to emanate from the production of ROS [3,16,85], this finding suggests that bacteria with a greater abundance of cytochrome c may be better suited to co-exist with dense blooms of C. polykrikoides where ROS-production is presumably high. Still, metagenomic predictive tools should be taken with caution given that functional potential results are generated from a limited database of available genomes.
The roles of the dominant bacteria found associated with C. polykrikoides blooms can also be surmised from the literature. For example, patch samples were dominated by bacteria such as Coraliomargarita, part of the phylum Verrucomicrobia, which preferentially grow on phytoplankton-derived high molecular organic compounds [89] that would likely be readily available in bloom formers such as C. polykrikoides that excrete extracellular polysaccharides [9,16,90]. Flavobacteria (Winogradskyella co-dominate in patches) are known to facilitate macromolecule conversions, converting high molecular compounds to low molecular weight compounds [77], a process that may concurrently support the growth of C. polykrikoides by providing smaller, labile compounds such as glutamic acid which this alga consumes readily [8]. Other microbial members that co-dominated within C. polykrikoides patches, such as bacterium from the AEGEAN-169 marine group and the SAR11 clade, are part of the phylum Alphaproteobacteria which are capable of assimilating dissolved organic matter (DOM) including extracellular polymeric substances (EPS) [91] which are produced in copious amounts by C. polykrikoides [16,90]. Beyond dominant microbes, it has been demonstrated that if a rarer microbe is responsible for a key metabolic process that other microbes do not carry out, it can have an enormous ecosystem impact [92] and the discovery of the metabolic repertoire of rare microbes will likely continue to advance in parallel with high-throughput sequencing efforts. C. polykrikoides has an obligate requirement for B-vitamins [18] and it has been demonstrated that the availability of both, B 1 and B 12 can shape bloom dynamics [9,93]. Vitamin concentrations and turnover inside bloom patches of C. polykrikoides are elevated, a fact not only attributed to the auxotrophic phytoplankton but also to the bacterial consortium residing within the patches [9]. While the relative abundance of potential B 12 producing-bacteria belonging to the order Rhodobacterales (class Alphaproteobacteria) [94,95] was significantly enriched in patch vs non-patch samples, the former has been shown to harbor 10-fold higher heterotrophic bacterial densities, likely leading to the observed higher vitamin concentrations found in Koch et al. [9]. Interestingly, Rhodobacterales was also more abundant in the >5μm fraction of patch samples, suggesting a potentially endosymbiotic association between these potential vitamin producers and auxotrophic C. polykrikoides.
Synechococcus was part of the core microbiome but was significantly enriched in the >5μm fraction of both patch and non-patch samples, despite it being a 1 μm cell. C. polykrikoides can feed on Synechococcus [20,96,97]. Field samples revealed that the >5μm fraction of patches (1-7%) had slightly lower relative abundances of Synechococcus than the same size fraction in non-patch samples (1-13%; Fig 6). This finding is consistent with Koch et al. [9] who found that Synechococcus densities were significantly lower in patches of C. polykrikoides and could indicate C. polykrikoides is capable of consuming Synechococcus and/or inhibiting this population via allelochemicals. Collectively, this suggests that these cyanobacteria may influence the nutrition of C. polykrikoides and potentially the C and N cycles of the surrounding microbiome.
In summary, this study revealed the profound effects of C. polykrikoides blooms on prokaryotic and eukaryotic plankton communities. The relative abundance of dinoflagellates in general, and C. polykrikoides and Amoebophrya in particular, were enriched in bloom samples and the experimental enrichment of C. polykrikoides led to a significant increase in the relative abundance of eight genera of dinoflagellates but a significant decline in other eukaryotic plankton. Patch and non-patch eukaryotic community composition were significantly different and patch communities harbored significantly lower diversity than non-patch samples, with more than 30 unique OTUs found within non-patch samples, suggesting blooms allelopathically inhibited these OTUs. While differential abundance analyses demonstrated that the Rhodobacterales were significantly enriched in C. polykrikoides patches and that there were more than 150 unique OTUs in non-bloom samples, the overall impact of these HABs on bacterial community composition and diversity was less intense compared to the impacts on eukaryotic communities.
Supporting information S1 Table. QIIME outputs for the sequencing of the V7/V8 region of the 18S rRNA gene, including split libraries output (demultiplexed reads), total UCLUST assigned reads and the total number of algal assigned reads for time series (patch and non-patch), and experimental samples (initial, control and Cochlodinium addition) and size fractions (>0.2 and >5μm). % algal reads = (total number of algal assigned reads divided by the total number of UCLUST assigned reads) x 100. (PDF) S2 Table. QIIME outputs for the sequencing of the V4 variable region of the 16S rRNA gene, including split libraries output (demultiplexed reads), total UCLUST assigned reads and the total number of chloroplast and mitochondria assigned reads for time series (patch and non-patch) samples and size fractions (>0.2 and >5μm).  Lineages that are part of the >5μm fraction (+) and >0.2μm fraction (-) core microbiomes are in bold, and italicized if unique to a core microbiome. Only significant differential abundances (alpha <0.05) are shown. Data is grouped by phyla (taxonomy) and colored by order. Unknown orders are listed as next lowest known taxonomy and indicated with � . (PDF) S14 Fig. Differentially abundant 16S lineages among the Non-patch >0.2μm and Non-patch >5μm size fraction samples for A) Acidobacteria and B) Actinobacteria. Negative log2 fold changes represent lineages enriched in >0.2μm samples, while positive log2 fold changes represent lineages enriched in >5μm samples. Lineages that are part of the >5μm fraction (+) and >0.2μm fraction (-) core microbiomes are in bold, and italicized if unique to a core microbiome. Only significant differential abundances (alpha <0.05) are shown. Data is grouped by phyla (taxonomy) and colored by order. Unknown orders are listed as next lowest known taxonomy and indicated with � . (PDF) S15 Fig. Differentially abundant 16S lineages among the Non-patch >0.2μm and Nonpatch >5μm size fraction samples for Bacteriodetes. Negative log2 fold changes represent lineages enriched in >0.2μm samples, while positive log2 fold changes represent lineages enriched in >5μm samples. Lineages that are part of the >5μm fraction (+) and >0.2μm fraction (-) core microbiomes are in bold, and italicized if unique to a core microbiome. Only significant differential abundances (alpha <0.05) are shown. Data is grouped by phyla (taxonomy) and colored by order. Unknown orders are listed as next lowest known taxonomy and indicated with � . (PDF) S16 Fig. Differentially abundant 16S lineages among the Non-patch >0.2μm and Nonpatch >5μm size fraction samples for Alphaproteobacteria. Negative log2 fold changes represent lineages enriched in >0.2μm samples, while positive log2 fold changes represent lineages enriched in >5μm samples. Lineages that are part of the >5μm fraction (+) and >0.2μm fraction (-) core microbiomes are in bold, and italicized if unique to a core microbiome. Only significant differential abundances (alpha <0.05) are shown. Data is grouped by phyla (taxonomy) and colored by order. Unknown orders are listed as next lowest known taxonomy and indicated with � . (PDF) >0.2μm fraction (-) core microbiomes are in bold, and italicized if unique to a core microbiome. Only significant differential abundances (alpha <0.05) are shown. Data is grouped by phyla (taxonomy) and colored by order. Unknown orders are listed as next lowest known taxonomy and indicated with � . (PDF) S21 Fig. Differentially abundant 16S lineages among the Non-patch >0.2μm and Non-patch >5μm size fraction samples for A) SAR and B) Verrucomicrobia. Negative log2 fold changes represent lineages enriched in >0.2μm samples, while positive log2 fold changes represent lineages enriched in >5μm samples. Lineages that are part of the >5μm fraction (+) and >0.2μm fraction (-) core microbiomes are in bold, and italicized if unique to a core microbiome. Only significant differential abundances (alpha <0.05) are shown. Data is grouped by phyla (taxonomy) and colored by order. Unknown orders are listed as next lowest known taxonomy and indicated with � . (PDF)