Seawater was sampled at five time points (Supplementary Figure 1) in the Spring 2018 at the long-term ecological research site Helgoland Roads (54°11′03′′N, 7°54′00′′E). Samples were taken at about 1 m below the sea surface. The water was stored at 4°C in 10-L containers and processed within 2 h of sampling. Particles that had already settled during that time were resuspended by gentle inversion of the containers before the seawater was used for the experiments. The bacterial diversity of unfractionated seawater samples was compared with that of fractions obtained by sequential filtration, centrifugation, and sedimentation in Imhoff cones (Figure 1).

In this study, all filtrations were performed with a constant vacuum pressure of −200 mbar. Filters of all pore sizes had a diameter of 47 mm, an effective filter diameter of 39 mm, and an effective filter area of 1,195 mm² and were composed of a hydrophilic polycarbonate membrane (Millipore, Darmstadt, Germany). All 0.2-μm pore-sized filters had the filter code GTTP. If during filtration a decrease in flow rate was detected, indicating a lower efficient pore size because of pore clogging, the filtration was repeated with a smaller sample volume.

For community analysis by 16S rRNA amplicon sequencing, 0.5 L of unfractionated seawater was filtered directly onto 0.2-μm pore-sized filters, frozen in liquid nitrogen, and stored at −80°C.

For cell staining by catalyzed reporter deposition fluorescence in situ hybridization (CARD-FISH) experiments and for determination of total cell counts, 50 ml of unfractionated seawater was fixed with 1% formaldehyde for 1 h at room temperature (RT) and filtered onto a 0.2-μm pore-sized filter. A cellulose acetate filter (0.45-μm pore size, Millipore, Darmstadt, Germany) supported all 0.2-μm filters used for cell counting to allow an even spread of the fixed cells. Filters for CARD-FISH were stored at −20°C.

In addition to the unfractionated seawater samples, plankton net samples were taken using 20- and 80-μm pore-sized nets. For DNA extraction and subsequent amplicon sequencing, 100 ml of these concentrated plankton net samples was filtered directly onto 3-μm filters.

Sequential filtration was used to fractionate seawater into large (>10 μm) and small (10–3 μm) particle fractions, as well as a free-living fraction (3–0.2 μm) (Figure 1). For community analysis by 16S rRNA amplicon sequencing, 1 L of seawater was first filtered through a 10-μm pore-sized filter (filter code of all 10-μm filters: TCTP), then the whole filtrate was applied to a 3-μm pore-sized filter (filter code of all 3-μm filters: TSTP), and finally, 0.5 L of the 3-μm filtrate was filtered through a 0.2-μm filter. For CARD-FISH, 200 ml of fixed seawater (2% formaldehyde, 2 h, RT) was applied to a 10-μm filter, then 200 ml of the 10-μm filtrate was filtrated through a 3-μm filter, and finally 50 ml of the 3-μm filtrate was used for the 0.2-μm filtration.

Forced settlement of particles was performed with 0.5 L of seawater in 50-ml tubes in a Rotina 35R centrifuge (Hettich, Tuttlingen, Germany), applying 4,890 × g for 10 min (Figure 1). Pellets in 1 ml of bottom water were collected using a sterile glass pipette and combined as the particle fraction of the centrifugation approach. The remaining 49 ml supernatant was collected and combined as the free-living fraction of the centrifugation approach. The particle fraction was resuspended in 5 ml of artificial seawater and filtered onto a 0.2-μm filter for later DNA extraction. For CARD-FISH experiments, 1 ml of sample was fixed (2% formaldehyde, 2 h, RT), diluted 10-fold with artificial seawater, and then filtered onto a 0.2-μm pore-sized filter. The free-living fraction was filtered directly through a 0.2-μm filter for amplicon sequencing, or a 10-ml portion was fixed as aforementioned and then filtered onto a 0.2-μm filter for CARD-FISH experiments.

Imhoff cones were applied in this study to fractionate microorganisms in seawater by natural gravitation (Figure 1). In a preliminary study in Spring 2017, on Julian Days 88, 94, and 135, 1 L of seawater was sampled and directly filtered through a 3-μm filter and then onto a 0.2-μm pore-sized filter. In sedimentation cones, 1 L of seawater was allowed to settle for 24 h. A bottom fraction of 1 to 5 ml was withdrawn using the bottom stopcock (Figure 1). The bottom fraction was filtered onto 3-μm pore-sized filters. DNA was extracted as described in Zhou et al. (1996). The 16S rRNA gene amplicon sequencing and analysis were performed as described below.

The preliminary study indicated the possibility of a bottle effect on the community composition. Hence, in 2018, different sedimentation times (0.5, 1, 3, 6, 16, and 24 h) were tested to find optimal conditions for the spring bloom samples. Large particles that settled within 1 h were rare, whereas particles visible to the naked eye had settled after 3 h. Samples collected after 24 h of sedimentation were analyzed to explore the effects of long settlement times. A 3-h sedimentation was chosen as standard throughout the experiments.

Bottom fractions containing settled particles and associated bacteria in 1- to 5-ml volumes were combined from several cones. Of this mixed sample, 5 ml was filtered onto a 0.2-μm filter for DNA extraction. One milliliter was fixed (2% formaldehyde, 2 h, RT), diluted 10-fold in artificial seawater, and filtered onto a 0.2-μm filter for cell staining. For the free-living fraction, 0.5 L of the remaining water of one cone (top fraction) was filtered onto a 0.2-μm filter, which was later used for DNA extraction. Ten milliliter was fixed as aforementioned and then filtered onto a 0.2-μm filter for cell staining.

A further separation of the bottom fraction from the Imhoff cones was performed by centrifugation. For this, the bottom fraction of one sedimentation cone was suspended in 50 ml of artificial seawater and then centrifuged (4,890 × g for 10 min). The obtained pellet was expected to contain particle-attached bacteria, and the supernatant was expected to contain free-living bacteria, including motile chemotactic bacteria. The particle fraction of this separation consisted of all particles in 5 ml bottom water and was filtered directly onto a 0.2-μm filter for DNA extraction, or 1 ml was fixed (as aforementioned) and diluted 10-fold in artificial seawater prior to filtration onto a 0.2-μm filter. The supernatant was directly filtered for DNA extraction (0.2-μm filter), and 10 ml was fixed as aforementioned and then filtered (0.2-μm filter) for cell staining.

DNA of all 2018 samples was extracted from filters using the Power Water Kit according to the manufacturer’s protocol (Qiagen, Hilden, Germany). DNA concentrations and fragment lengths were analyzed in a capillary electrophoresis (Fragment Analyzer, Advanced Analytical, Santa Clara, United States) using the DNF-488 High Sensitivity Genomic DNA Analysis Kit (Advanced Analytical). One nanogram of DNA served as template for a 25 μl PCR with 2.5 U Taq DNA polymerase (Merck, Darmstadt, Germany), 20 mM of Tris pH 8.4, 50 mM of KCl, 0.6 mM of MgCl₂, 90 μM of bovine serum albumin, 0.0625 mM of each dNTP, 0.4 μM of barcoded forward primer Bakt_341F (5′-TATACGCCTACGGGNGGCWGCAG-3′) and 0.4 μM of reverse primer Bakt_805R (5′-GACTACHVGGGTATCTAATCC-3′). The primer pair targets the hypervariable V3 and V4 regions of the bacterial 16S rRNA gene (Herlemann et al., 2011). The amplification program involved initial denaturation for 3 min at 94°C, 30 cycles of denaturation for 45 s at 94°C, annealing for 1 min at 50°C, elongation for 1.5 min at 72°C, and a final elongation of 10 min at 72°C. Amplicons were purified using AMPure XP beads (Beckman Coulter, Krefeld) and were quantified with the DNF-473 Standard Sensitivity NGS Fragment Analysis Kit (1–6,000 bp, Advanced Analytical) on the Fragment Analyzer. Equimolar amplicon pools were sequenced on an Illumina HiSeq2500 in rapid mode with 2 × 250 bp paired-end run performed by the Max Planck Genome Centre Cologne, Germany¹.

A total of 19.8 million 16S rRNA reads were paired with BBmerge v37.82 applying default settings and allowing no mismatch in the overlapping region (Bushnell et al., 2017). A total of 12.9 million merged reads were then de-multiplexed and quality trimmed using MOTHUR (Schloss et al., 2009). The command “trim.seqs” was used with the setting of a minlength = 300, maxambig = 0, maxhomop = 8, allfiles = T, and checkorient = T. The final high-quality merged sequences (8.9 million) were classified with the SILVAngs pipeline (Quast et al., 2013) using SILVA release version 132. We used a minimum alignment length of 150, a minimum quality score of 30 (minimum length of a sequence/reads), and a similarity threshold of 0.98 for the creation of operational taxonomic units (OTUs). Each OTU was classified according to its most related genus. The relative read abundance of sequences affiliating to chloroplasts, mitochondria, Eukarya, or Archaea were determined for each sample. After their proportional quantification, they were removed to focus the dissimilarity analysis on bacterial reads (Supplementary Figure 2). Read abundance was normalized for each sample using the “decostand()” function (method = total) (Oksanen et al., 2020). A minimum of 26,231 and maximum of 312,329 reads per sample were analyzed after processing. Raw reads of all samples were deposited at the European Nucleotide Archive (ENA) and are accessible through the accession number PRJEB41742.

An enrichment factor was calculated by dividing the relative read abundance of an OTU in a fraction by the relative read abundance of the OTU in unfractionated seawater at the respective time point. For example, the relative read abundance of SAR 11 clade Ia varied in unfractionated seawater samples across all five time points (min–max: 2.3–18.9%) but was always higher in the filter-fractionated free-living samples (min–max: 21.7–39.3%) and lower in the large filter-fractionated particle fractions (min–max: 0.37–1.6%). This led to an average enrichment factor of 4.15 for the filter-fractionated free-living and an enrichment factor of 0.10 for the large particle fraction, indicating enrichment (EF > 1) and depletion (EF < 1), respectively.

CARD-FISH filters were processed following the protocol of Pernthaler et al. (2004) with four modifications. (i) The agarose (0.1% w/v, LE agarose, Biozym, Hessisch Oldendorf, Germany) embedding of the filters was done on parafilm instead of glass to avoid particle loss: a drop of agarose was applied to the parafilm. The filter was carefully dipped into the drop from both sides before being placed with the filtered side facing down on the parafilm. (ii) After the lysozyme treatment, filters were incubated with 60 U of achromopeptidase (Sigma, Taufkirchen, Germany) in 10 mM of NaCl and 10 mM of Tris HCl pH 8.0 for 30 min at 37°C to permeabilize the cell walls. Then filters were rinsed with MilliQ water. (iii) Endogenous peroxidases were inactivated with 0.15% hydrogen peroxide in methanol for 20 min at RT. (iv) Finally, filters were embedded in a 4:1 ratio of Citifluor to VectaShield plus 4′,6-diamidino-2-phenylindole (DAPI; 0.5 ng L^–1) as described in Nikrad et al. (2012).

Cells were manually counted with a Zeiss Axio Imager D2 microscope (Zeiss, Oberkochen, Germany) equipped with a Zeiss AxioCam MRm camera, utilizing the software Axio Vision (v 4.8). Detection of cells was based on fluorescence signals of DAPI using the microscope filter DAPI HC (AHF, Tübingen, Germany) and of Alexa 488 using the filter EGFP ET (AHF). The latter was used in CARD-FISH for the probes EUBI-III and Non-EUB (control) (Amann et al., 1990). Per sample, a minimum of 2,000 cells or the total cell abundance in 100 grids (size of one grid: 15,625 μm²) were counted. The local density of cells on the filters was analyzed to determine the spatial proximity to a particle and other cells. We differentiated seven types of cell distribution: (1) single cells with equal spreading across the grid; (2) cells in direct contact with a transparent exopolymer particle (TEP); (3) an algae; (4) diatom shells; (5) cell consortia with a larger local cell density compared with the surrounding background cell density, but without stained contact points; (6) cells in filaments of at least three cells; and (7) cells in aggregates with a central distinctive connection among them (Supplementary Figure 3). The total cell count per milliliter of seawater was calculated with the following equation:

Particles with cells were analyzed for their abundance and approximate size. Based on their appearance, they were categorized as transparent polymeric particles, algae, or diatom debris.

Statistical analyses, data transformations, and graphing were performed in R studio (R Core Team, 2014). Non-metric multidimensional scaling (NMDS) analysis based on Bray–Curtis dissimilarity indices was performed with the “vegdist()” function (“vegan” package; Oksanen et al., 2020). Stress was determined with the “metaMDS()” function, and a Shepard diagram was created to determine the fit between observed dissimilarity against the ordination distance (non-metric and linear). Statistical evaluation of the communities between the methods was performed with a permutational multivariate analysis of variance (PERMANOVA) using distance matrices [method = bray, “vegdist()” function, “vegan” package; Oksanen et al., 2020]. This analysis was followed by a pairwise comparison of the method effect [“pairwise.perm.manova()” function, method = euclidian].

Cell distributions between different particle types, the colonization of cells and varying particle types, and the calculated diversity indices of each sample group were treated as comparisons between parametric groups. Significant differences among the defined parameters were determined with an analysis of variance [“aov()” function], followed by a Tukey multiple pairwise comparison [“TukeyHSD()” function] t-test. Residuals versus fit plots were used to check for homogeneity of variances, which was additionally tested by a Levene test [“leveneTest()” function, “car” package, R Version 3.5.3; Fox and Weisberg, 2019]. A normality plot of residuals was used to test for the normal distribution of the data. This was additionally tested by a Shapiro–Wilk test [“shapiro.test()” function] after extraction of the residuals [“residuals()” function]. If the data distribution did not meet the homogeneity of variance or violated a normality distribution, the one-way ANOVA was replaced by a non-parametric Kruskal–Wallis rank sum test. The rank sum test was followed by pairwise comparisons using Tukey and Kramer (Nemenyi) test with Tukey distance approximation for independent samples [posthoc.kruskal.nemenyi.test (dist = “Tukey”) function, “PMCMR” package, R Version 3.5.3; Pohlert, 2014].

Further R packages were used for data visualization and data transformation: ComplexHeatmap (Gu et al., 2016), circlize (Gu et al., 2014), picante (Kembel et al., 2010), rioja (Juggins, 2020), colorspace (Zeileis et al., 2009), and dplyr (Wickham et al., 2018).

Microscopy of the 10-, 3-, and 0.2-μm pore-sized filters confirmed the effectivity of filtration. Particles including diatoms and Phaeocystis cells were collected on the 10- and 3-μm filters, whereas the 3–0.2 μm fraction contained almost exclusively single bacterial cells (98.3–100%) (Table 1). Only 44–71% of the cells present in unfractionated seawater were recovered on 0.2-μm filters after sequential filtration, with less than 1% of all cells detected on the 10- and 3-μm filters (Table 1). The cell loss coincided with shifts in OTU relative read abundances. To quantify the abundance shifts, we derived enrichment factors that were calculated as ratio of the relative read abundance in a fraction community divided by the relative read abundance in the unfractionated seawater community (Figure 3). We also calculated the mean enrichment factor for the OTUs with, on average, the highest relative read abundance in seawater (Figure 4).

The 3–0.2 μm fractions were enriched in reads affiliated with SAR11 and depleted in several flavobacterial clades, notably Aurantivirga and Formosa (Figure 3). The fact that the latter was also depleted in the particle fractions (>10 and 10–3 μm) suggested cell loss of Flavobacteriales. The unfractionated seawater communities and derived free-living communities resembled each other closely, which was supported by pairwise statistical comparisons (Figure 2A and Supplementary Table 2), albeit with a lower diversity (Shannon–Wiener index) after filtration because of a lower evenness in relation to seawater communities (ANOVA, P < 0.001, Supplementary Figure 4).

Particle filters (>10 and 10–3 μm) were dominated by particle-attached bacteria. On average, less than 40% were single cells (Table 1). These cells were smaller than 3 μm based on DAPI staining. Most cells were in direct contact with transparent polymeric particles, algae, or diatom shells (Supplementary Figure 3). A small portion of cells appeared clustered together in cell consortia larger than 3 μm (Supplementary Table 3). Bacterial cell density on particles was highly variable. In general, transparent polymeric particles were slightly more populated [0.02 ± 0.02 cells (μm² particle) ^–1] than algae [0.01 ± 0.01 cells (μm² particle) ^–1] and diatom shells [0.01 ± 0.01 cells (μm² particle) ^–1] (Supplementary Figure 5).

The >10 and 10–3 μm fractions were significantly different from each other and from unfractionated seawater and the 3–0.2 μm fractions (ANOSIM and PERMANOVA, P < 0.001; Figure 2A and Supplementary Table 2). The diversity of the filtration-separated particle communities was higher than in unfractionated seawater and free-living communities because of greater OTU evenness (ANOVA, P < 0.001) (Supplementary Figure 4).

Only one of the dominant OTUs in seawater (Figure 2B and Supplementary Table 4) had a higher relative read abundance in the filtered particle fractions: Persicirhabdus (Verrucomicrobia). The other dominant seawater OTUs were depleted.

Operational taxonomic units with a high relative read abundance in seawater are, for arithmetical reasons, unlikely to have high enrichment factors. For OTUs with a relative read abundance of at least 0.1% in seawater, only few OTUs had enrichment factors >6 in a particle fraction. These were expected to spend most of their lifetime on particles. Examples of these OTUs in both particle fractions were in decreasing order of the read abundance: clade BD1-7, Sulfitobacter, Algibacter, Rhodococcus, Colwellia, Psychromonas, Winogradskyella, and Maribacter (Figure 3). A size selectivity was noticed for several OTUs: uncultured Nitrincolaceae (Oceanospirillaceae), Kiloniella, and Phycisphaeraceae were present at higher relative read abundances in the small particle communities (10–3 μm), whereas the large particle communities (>10 μm) included higher relative abundances of Sneathiella, Loktanella, and Maribacter. The taxa with a lower relative read abundance in seawater (<0.1%) and with high enrichment (enrichment factors of >25) in both particle communities were Tropicibacter, Lentisphaera, Roseobacter, and uncultured Actinomarinales. Highly enriched on particle filters but less rare in seawater were uncultured Cyclobacterium, clade KI89A, Paraglaciecola, and Phycisphaeraceae.

We extended the size-selective separation by community analysis to 20- and 80-μm pore-sized plankton net catches that were concentrated on 3-μm filters. These fractions were similar to the corresponding >10-μm communities (Figure 2). Interestingly, Gramella was 125 times more enriched in plankton net samples than in unfractionated seawater and >10-μm samples (Figure 3 and Supplementary Table 4). Also highly enriched were Aliikangiella, Costertonia, Gillisia, and Flavimarina.

Centrifugation of seawater at 4,890 × g for 10 min resulted in visible pellets. Although the relative read abundance of 16S rRNA genes originating from chloroplasts increased in two out of three samples by 50% (Supplementary Figure 2), the bacterial communities of supernatant and pellet showed no significant differences (pairwise comparison, P > 0.05, Figure 5A and Supplementary Table 5). Diversity, species richness, and evenness were not significantly different between the fractions (Supplementary Figure 4). The particle fraction had more cells than the particle fractions of other methods (Table 1). Dominant OTUs had similar relative read abundances in seawater and in the pellet and supernatant communities (Figure 5B). Among the OTUs enriched in pellet fractions were Profundimonas and Colwellia, which had the highest read abundance of 1.9% at time point 3 (onset of the diatom bloom). OTUs with a relative read abundance of ≥ 0.1% and an enrichment factor of above 18 for the pellet fraction were Ca. Megaira and Martelella (both Alphaproteobacteria), Aliikangiella, Oleispira, Profundimonas, and Pseudoalteromonas (all Gammaproteobacteria), Flavobacterium, flavobacterial clade NS10, and Lentisphaera (Figure 3).

Enrichment of particles in sedimentation cones was first explored at Helgoland Roads in Spring 2017 using a particle settlement time of 24 h. Particle communities were collected onto 3-μm pore-sized filters and were compared with seawater communities directly sampled as >3 and 3–0.2 μm fractions by sequential filtration. Statistical analyses revealed significant differences between the communities, with the largest distances between the 24-h particle samples and both types of 0-h samples (ANOVA and PERMANOVA, P < 0.001, Supplementary Tables 6, 7 and Supplementary Figure 6). Dominant free-living microorganisms (3–0.2 μm fraction) were affiliated to SAR11 clades, SAR86, SAR116, OM43, Planktomarina, Ca. Actinomarina, flavobacterial marine groups NS4 and NS5, and uncultured Cryomorphaceae, a typical free-living community for a pre-algal bloom in March at Helgoland Roads (Chafee et al., 2018) (Supplementary Table 8). At the earlier sampling days (time points 1 and 2, 2017), few particles were collected. The seawater >3-μm population showed increased relative read abundances of Fluviicola and the gammaproteobacterial genera Colwellia, Halomonas, Pseudoalteromonas, Psychrobacter, Shewanella, Vibrio, and Woeseia. Most of these gammaproteobacterial OTUs dominated the sedimentation cone bottom fraction (>3 μm), together with increased numbers of Alphaproteobacteria (Erythrobacter, Loktanella, Sulfitobacter, and Roseovarius) and the Flavobacteriaceae Gramella and Olleya. Many of these taxa are known to be part of the phycosphere but are also part of the “bottle effect,” a rapid shift in community composition after sampling in bottles (Pernthaler and Amann, 2005).

In 2018, bacterial communities after 3 and 24 h of separation in cones were analyzed to quantify the influence of microbial growth. Bottom fraction communities of the two sedimentation times resembled each other closer than other communities (Figure 6). Growth of gammaproteobacterial Pseudoalteromonas, Colwellia, and Psychrobacter was detected as larger relative read abundance or enrichment factor in the 24-h sedimentation cone bottom fractions, complemented by lower relative read abundances and enrichment factors of Olleya, Polaribacter 1, Pseudofulvibacter, Flavirhabdus, uncultured Nitrincolaceae, Dokdonia, Sulfitobacter, Roseobacter, Nonlabens, and Cellulophaga. Such a strong growth-dependent signal was not observed in the supernatant fractions of the cones (Supplementary Tables 4, 8).

In Spring 2018, six settlement times (0.5, 1, 3, 6, 16, and 24 h) covering sinking speeds ranging from 0.3 m day^–1 (24-h settlement time) to 14.4 m day^–1 (0.5-h settlement time) were tested. The amount of particles settled at the bottom of the cones did not visibly increase beyond 3 h. This time corresponds to the collection of all particles with a sinking speed larger than 2.4 m day^–1 and provides sufficient time for chemosensory-directed swimming of motile bacteria to follow the particles to the bottom of the cone (reviewed in Stocker and Seymour, 2012).

Microscopy revealed that algae and diatom shells had settled after 3 h. The supernatant still contained particles composed of extracellular polymeric substances but was dominated by single cells (96–100%). Single cells were also abundant in the bottom fraction (63.3–99%, Table 1). Bacterial communities of unfractionated seawater and of the top fraction in cones were not significantly different in OTU diversity and relative read abundance (pairwise comparison, P > 0.05, Figure 7, Supplementary Table 9 and Supplementary Figure 4). Only 403–487 OTUs were detected in the bottom fractions, with half of the OTU numbers detected in unfractionated seawater and the supernatant (742–1,079) (Supplementary Figure 4). The reduced richness was balanced by a larger evenness resulting in no change in the overall diversity (Supplementary Figure 4).

The settled fractions at time points 4 and 5 were rich in SAR11 clade Ia as most abundant and slightly enriched OTU compared with seawater. High enrichment factors were observed for Psychrobacter, Erythrobacter, Rheinheimera, Halomonas, Pseudomonas, Oleispira, and Methylobacterium, all with ≥0.1% relative read abundance in seawater (Figure 3). Among the less frequent OTUs (relative read abundance in seawater <0.1%) with enrichment factors >10 were OTUs affiliating with the genera Roseobacter, Francisella, Acinetobacter, and Pirellula.

At two time points, the sedimentation cone bottom fractions collected after 3 h of settlement were resuspended in artificial seawater and separated for a second time by centrifugation. Most cells were recovered in the resulting supernatants. Overall, the number of particle-attached cells increased with the start of the phytoplankton bloom (time point 1 versus 3) (Table 1 and Supplementary Table 3). OTU communities obtained by the secondary separation were different from each other and from all other communities (Figure 7A). Three groups of taxa were identified and categorized as present in both fractions or enriched in either the supernatant or the pellet. OTUs with an enrichment and ≥0.1% relative read abundance in the supernatant included SAR11 clades Ia, Ib, I uncultured, II and III, SAR116, SUP05 group, Oleibacter, uncultured Nitrincolaceae, Lentimonas, flavobacterial marine groups NS7 and NS9, and Puniceicoccaceae marine group (Figure 3). OTUs specifically enriched in the pellet samples and with relative read abundances of ≥ 0.5% were Lutimonas, Fusobacterium, Ca. Megaira, and Pseudophaeobacter (Figure 3). Taxa enriched in the pellet and supernatant were in a decreasing order of relative read abundance Pseudoalteromonas, Olleya, Nonlabens, Erythrobacter, Sulfitobacter, Psychrobacter, Flavobacterium, Sphingorhabdus, Shewanella, Vibrio, Cellulophaga, Pseudomonas, Loktanella, Rheinheimera, and Lacinutrix (Figure 3).