Microbiomes of the bloom-forming alga, Phaeocystis globosa, are stable, consistently recruited communities with symbiotic and opportunistic modes

Phaeocystis is a cosmopolitan, bloom-forming phytoplankton genus that contributes significantly to global carbon and sulfur cycles. During blooms, Phaeocystis species produce large carbon-rich colonies, thus creating a unique interface for bacterial interactions. While bacteria are known to interact with phytoplankton—e.g., they promote growth by producing phytohormones and vitamins—such interactions have not been shown for Phaeocystis. Therefore, we investigated the composition and function of P. globosa microbiomes. Specifically, we tested whether microbiome compositions are consistent across individual colonies from four P. globosa strains, whether similar microbiomes are re-recruited after antibiotic treatment, and how microbiomes affect P. globosa growth under limiting conditions. Results illuminated a core colonial P. globosa microbiome—including bacteria from the orders Alteromonadales, Burkholderiales, and Rhizobiales—that was re-recruited after microbiome disruption. Consistent microbiome composition and recruitment is indicative that P. globosa microbiomes are stable-state systems undergoing deterministic community assembly and suggests there are specific, beneficial interactions between Phaeocystis and bacteria. Growth experiments with axenic and nonaxenic cultures demonstrated that microbiomes allowed continued growth when B-vitamins were withheld, but that microbiomes accelerated culture collapse when nitrogen was withheld. In sum, this study reveals interactions between Phaeocystis colonies and microbiome bacteria that could influence large-scale phytoplankton bloom dynamics and biogeochemical cycles.


Introduction
Half of the photosynthesis on Earth is performed by phytoplankton, making them critical components of the global carbon cycle [1]. Like most organisms, phytoplankton do not act in isolation but instead interact with bacteria living in close proximity-either on phytoplankton cell surfaces or in the chemically influenced area surrounding phytoplankton (i.e., the phycosphere), a region analogous to the rhizosphere in plants [2]. While the phycosphere is not as extensively studied as the rhizosphere [3], it clearly hosts specialized, specific interactions with larger implications for marine ecosystem functioning and biogeochemical cycles [4]. For example, phycosphere bacteria can stimulate the growth of specific phytoplankton by producing growth hormones [5] or supplying a limiting nutrient, such as a B-vitamin [6][7][8] or fixed-nitrogen [9]. Conversely, phycosphere bacteria can manipulate phytoplankton communities by secreting algicidal molecules (e.g., roseobacticides) that selectively kill certain phytoplankton [10] or by competing for essential nutrients [11,12]. Even small shifts in phytoplankton communities can alter how much organic carbon is recycled in the surface ocean or exported to deep water [13] since phytoplankton taxa vary in their C:N:P ratio [14], morphology, and sinking rates [15]. Therefore, phytoplankton-bacteria interactions are central to fully understanding phytoplankton dynamics and carbon cycling in the oceans. However, the stability, specificity, and currencies of exchange between phycosphere bacteria and diverse phytoplankton taxa are just beginning to be characterized.
The majority of our current knowledge regarding phytoplankton-bacteria interactions centers on interactions between diatoms and bacteria (reviewed in [16]) due to the ecological significance of diatoms on a global scale [17], as well as their experimental tractability (e.g., [5,18]). Such studies have identified common bacterial inhabitants of phycospheres-including bacteria belonging to the orders Rhodobacterales, Flavobacteriales, Alteromonadales, and Oceanospirillales [18,19]. Moreover, experiments with diatoms have demonstrated that phycosphere community assembly processes are mediated by both the phytoplankton and colonizing bacteria [20] and lead to highly reproducible bacterial community compositions [19]. However, it is unclear whether these traits extend to other phytoplankton taxa, particularly those that may be in competition with diatoms.
The Haptophyte genus, Phaeocystis, is an ideal system for extending studies of phytoplankton-bacterial interactions. Phaeocystis is a bloom-forming genus found from pole to pole. Notably, different bloom-forming species of Phaeocystis are specialized within latitudinal or thermal niches, often competing with diatoms [21]. P. globosa competes with diatoms to be the dominant spring bloom constituent in the North Sea, where it is considered a nuisance alga [22]. P. globosa is further distributed across temperate, subtropical, and tropical regions globally and has more recently started causing harmful algal blooms in Southeast Asia [23,24]. The two polar species, P. pouchetii and P. antarctica, contribute to spring blooms in the Arctic / N. Atlantic oceans and the Southern Ocean, respectively [21]. In the N. Atlantic, P. pouchetii has begun to regularly replace diatoms in the spring bloom [25] and in the Southern Ocean, there is continued interannual variability as to whether P. antarctica or diatoms dominate spring blooms [26]. While not yet systematically studied, existing evidence supports the hypothesis that bacterial interactions could influence Phaeocystis blooms: consistent bacterial communities have been found in association with discrete samples of two P. globosa strains studied independently [27,28] and P. antarctica blooms have been shown to strongly influence bacterioplankton community composition [29].
Bloom-forming Phaeocystis species share the ability to form large colonial structures (up to a centimeter in diameter) that are made up of thousands of cells embedded in a mucopolysaccharide matrix [30]. The carbon-rich colony substrate creates a unique physical and chemical interface for interaction between bacteria and Phaeocystis [31]. Bacteria are attracted to polysaccharides produced by phytoplankton [32] and the specific polysaccharides excreted by a phytoplankton cell contribute to which bacteria it encounters or is colonized by [18,33,34]. The concentrated availability of distinct polysaccharides in the Phaeocystis colony matrix, along with other chemical properties-such as the presence of amino sugars [35] and abundant dimethylsulfoniopropionate (DMSP) [36]-suggests a unique bacterial consortium would be attracted to and thrive in the colonial Phaeocystis microenvironment [18]. However, given the extensive functional redundancy among bacteria [37], it remains to be determined how specialized bacterial communities associated with Phaeocystis colonies are relative to other plankton and whether Phaeocystis microbiomes are stable-state systems.
In this study, we tested the specificity and stability of bacterial communities associated with Phaeocystis colonies by evaluating the microbiomes of individual colonies from replicate culture batches of four P. globosa strains that were isolated from globally wide-ranging locations across several decades. We further tested the reproducibility of P. globosa colony microbiome compositions by removing original microbiomes with antibiotics and evaluating compositions of recruited microbiomes after inoculation with bacterial communities in seawater. Finally, we tested two possible functional roles for P. globosa microbiomes by challenging axenic and nonaxenic cultures with nitrogen and vitamin stress. Nitrogen deprivation was chosen because nitrogen availability is a key factor believed to influence temperate and subtropical Phaeocystis blooms [25,38] and because fixed-nitrogen is one currency of exchange between phycosphere bacteria and some diatoms [39]. B-vitamin limitation was chosen because (i) the availability of B-vitamins affects phytoplankton bloom dynamics [40,41], (ii) P. globosa requires vitamins B 1 and B 12 to grow [42], and (iii) phycosphere bacteria can be sources of B-vitamins [7,8,43]. Overall, we found P. globosa colonies hosted remarkably consistent bacterial communities, both before antibiotic treatment (original microbiomes) and in the recruitment trial. These results demonstrate that, like many diatoms, Phaeocystis colonies regulate convergent bacterial community assembly towards a stable and reproducible core microbiome, albeit a different core microbiome than previously observed for other phytoplankton groups. Furthermore, the full consortia of co-cultured bacteria (in media and associated with colonies) differentially impact P. globosa growth depending on environmental conditions.

Phaeocystis globosa culture sources
We included four strains of Phaeocystis globosa in this study to investigate the consistency of Phaeocystis-associated bacterial community compositions across strains. Three strains-CCMP1528, CCMP629, and CCMP2710-were obtained from the National Center for Marine Algae and Microbiota (NCMA, Boothbay, ME, U.S.A.). We isolated the fourth strain in January 2019 from the Kuroshio current in the East China Sea (28.22 ºN, 127.66 ºE) by collecting surface water (5 m depth) and filtering it through 63 µm mesh-size nylon mesh before enriching with 10% L1-Si media [44]. Nutrient enrichment triggered a bloom of colonial P. globosa and individual P. globosa colonies were subsequently isolated, rinsed with sterile artificial seawater, and used to establish mono-algal cultures of a new P. globosa strain, referred to as "Kuroshio" herein. Together, these four strains represent wide geographic diversity and have been maintained in culture for widely varying times, from several decades to just a few years (Fig. 1). The year that each strain was originally isolated is displayed below each strain's name. "CCMP" strains were obtained from the National Center for Marine Algae and Microbiota in Boothbay, Maine, U.S.A. The Kuroshio strain was isolated for this study.

Evaluation of P. globosa colony and media microbiomes
Cultures of each P. globosa strain were synchronized by inoculating 35 ml of L1-Si media in triplicate 50 ml plastic culture flasks with 1 ml of maintenance culture, incubating for five days, and transferring 1 ml to a newly prepared culture flask. Cultures were incubated in a TOMY digital biology CLE-series plant growth chamber with cool-white fluorescent lighting (16 h light / 8 h dark diurnal light cycle, ~ 50 µmol m -2 s -1 light) maintained at 22ºC. After five such transfers, individual colonies and free-living cells were harvested for DNA extraction. Three colonies from each culture replicate were rinsed three times with sterile artificial sweater (ASW) before careful transfer (with <10 µl of ASW) to an Axygen Maxymum Recovery PCR tube with 30 µl 10% mass:volume Chelex 100 resin beads in PCR-grade water. The colony and Chelex mixtures were vortexed for 10 s, incubated at 96ºC for 20 minutes, vortexed again for 10 s, and placed on ice [45]. Free-living bacterial cells in Phaeocystis media were collected by gravity-filtering each culture replicate through 100-, 70-, and 40-µm mesh-size cell strainers to remove colonies and then filtering under gentle vacuum pressure through 3.0-µm (to remove free-living Phaeocystis cells) and 0.2-µm pore-size polytetrafluoroethylene (PTFE) filters (Millipore) to collect bacterial biomass. DNA was extracted from the 0.2-µm pore-size filters by following the manufacturer's protocol for the DNeasy PowerSoil kit (Qiagen), including the optional heating step for enhanced cell lysis. The v3-v4 region of the 16S rRNA gene was amplified from extracted DNA from both extraction methods following the Illumina guide for "16S Metagenomic Sequencing Library Preparation" (https://support.illumina.com/content/dam/illumina-support/documents/documentation/chemistry _documentation/16s/16s-metagenomic-library-prep-guide-15044223-b.pdf). The 16S rRNA gene amplicon sequencing libraries were sequenced with 300x300-bp v3 paired-end chemistry on the Illumina MiSeq platform.

Evaluation of recruited bacterial communities following antibiotic disruption of Phaeocystis microbiomes
To determine whether Phaeocystis strains reproducibly recruit specific bacteria to their microbiomes, the original microbiomes of each strain were disrupted and then each strain was exposed to bacterial assemblages in seawater before microbiome composition was assessed a second time. Original microbiomes were disrupted through antibiotic treatment: 1 ml of actively growing cultures were transferred to 50 ml culture flasks with 35 ml of L1-Si media and 1 ml of an antibiotic cocktail with 12 000 units/ml Penicillin G, 300 units/ml Polymyxin B, 50 µg/ml Chloramphenicol, 60 µg/ml Paromomycin and 6.25 mg/ml Cephalexin. After 48 h, 1 ml of each antibiotic-treated culture was transferred to L1-Si media without antibiotics and allowed to recover from antibiotic treatment for five five-day culture cycles. Microbiome disruption was checked by incubating aliquots of each antibiotic-treated culture with Hoechst 33342 DNA-binding dye and imaging each aliquot with confocal laser scanning fluorescence microscopy (Zeiss LSM 780). Additionally, aliquots of antibiotic-treated cultures were transferred to Difco Marine Broth 2216 (BD Biosciences) at the end of each ensuing culture cycle and monitored for bacterial growth by light microscopy (Olympus CKX53). No bacteria were observed in antibiotic-treated cultures with fluorescent microscopy, nor was bacterial growth detected with light microscopy in marine broth incubations. Henceforth, antibiotic-treated cultures were considered axenic.
Inoculating seawater for microbiome recruitment trials was collected outside the coral lagoon of Okinawa Island (Japan) by bucket cast and filtered through 63 µm-mesh-size nylon mesh before transport back to the lab. Inoculating seawater was further filtered through 3 µmand 0.45 µm-pore-size PTFE filters to remove eukaryotic cells. Two 50 ml replicates of the inoculating seawater were filtered through 0.2 µm pore-size filters to assess the initial composition of bacterial communities inoculating Phaeocystis cultures. L1-Si media for recruitment trials was prepared with 50% inoculating seawater and 50% sterile artificial seawater. Culture replicates for microbiome recruitment trials were initiated with 1 ml of actively-growing axenic Phaeocystis cultures in triplicate flasks. The bacterial community was allowed to equilibrate over five consecutive five-day culture cycles (transferring every fifth day) before Phaeocystis colonies (five from each replicate) and free-living media-associated bacteria were harvested for DNA extraction and amplicon sequencing following the methods described above.

Effects of microbiomes on P. globosa growth under limiting culture conditions
To assess impacts of microbiomes on Phaeocystis growth, we monitored axenic and nonaxenic Phaeocystis cultures under nutrient-replete and two limiting conditions: nitrogen limitation and B-vitamin limitation. L1-Si media was prepared without nitrate to induce nitrogen limitation and L1-Si media was prepared without the L1 vitamin solution, which contains thiamine (vit. B 1 ), biotin (vit. B 7 ), and cobalamin (vit. B 12 ), to induce B-vitamin limitation [44]. Triplicate 50 ml sterile plastic culture flasks were prepared with 35 ml of each media type (L1-Si replete, L1-Si with no nitrate, and L1-Si with no B-vitamins) in a laminar flow hood sterilized with UV light and cleaned with ethanol. One ml of actively growing axenic or nonaxenic P. globosa was transferred to each flask so there were three replicates of axenic and nonaxenic cultures for each of the four strains (total of 72 flasks per culture cycle). Growth was monitored daily by removing 1 ml from each flask, homogenizing by pipette mixing, and measuring in vivo Chlorophyll a fluorescence with a Turner Designs Trilogy Fluorometer. Transfers to fresh media were completed for five culture cycles (5-7 days each, depending on growth) for each treatment, except for nonaxenic cultures in L1-Si media without nitrate, as growth ceased after the third culture cycle. All transfers and aliquots were made inside the clean laminar flow hood. Following each culture cycle, an additional 100 µl aliquot was taken from each flask, transferred to 1 ml sterile marine broth in a 12-well cell-culture plate, and monitored for bacterial growth for 72-hours. No bacterial growth was observed from aliquots taken from axenic cultures, whereas bacterial growth was always observed after 24-hours in aliquots from nonaxenic cultures, thus confirming axenic and non-axenic cultures, respectively.

Sequence processing
Overall, amplicons from 100 samples (n = 34 original single-colony microbiomes, n = 12 original media microbiomes, n = 40 recruited single-colony microbiomes, n = 12 recruited media microbiomes, n = 2 inoculating seawater communities) were sequenced for this project, generating a total of 17 328 537 sequencing reads (median total sequences per sample = 147 351, Fig. S1). Raw sequencing data is available from the NCBI SRA under accession PRJN779092. Following quality-filtering, denoising, and chimera and singleton removal, 13 435 004 sequencing reads remained (median remaining sequences per sample = 121 035, Fig. S1). One concern when studying phytoplankton microbiomes is the proportion of sequencing resources allocated to non-target chloroplast 16S genes instead of target bacterial 16S genes.
Lower input, single-colony samples yielded a larger proportion of chloroplast sequences than media or seawater samples: single colony samples had a mean of 22% bacterial sequences compared to 85% and 99.5% bacterial sequences for media and seawater samples, respectively (Fig. S2). However, following removal of chloroplast sequences, rarefaction curves for all samples reached ASV richness saturation (Fig. S3), indicating an adequate number of bacterial sequences remained for analysis due to sufficient sequencing depth. Chloroplast sequences were excluded from ensuing analyses.

Diversity of original and recruited P. globosa microbiomes
The inoculating bacterial community for the recruitment trial was significantly more diverse than original and recruited Phaeocystis microbiomes ( Fig. 2; pairwise Wilcox test, p < 0.05). In addition, original and recruited colony microbiomes were significantly less diverse than original and recruited media microbiomes ( Fig. 2; pairwise Wilcox test, p < 0.001). In both these cases, colony microbiomes had about half the observed richness as media microbiomes.
The beta diversity between different sample types was computed by normalizing ASV sequence read counts to a centered log-mean ratio (CLR) using the CoDaSeq package and calculating Euclidean distances between community compositions in each sample with the phyloseq package. This combination of CLR transformation and Euclidean distance is known as the Aitchison distance and minimizes bias inherent to compositional data, such as amplicon data [52]. A Principal Coordinates Analysis (PCoA) performed on Aitchison distances revealed four clearly separated clusters: inoculating seawater communities, original media microbiomes, recruited media microbiomes, and colony-associated microbiomes (Fig. 3a). When sample points are colored by trial (original or recruitment), the colony-associated community cluster appears split by trial, but original and recruited colony microbiomes were not significantly different (Fig. 3a). Pairwise Permutational Multivariate Analysis of Variance (PERMANOVA) tests using the adonis2 function from the vegan package (999 permutations; [53]) showed original and recruited colony-associated communities were both significantly different from media and seawater-associated communities (p < 0.05) but not from each other, whereas recruited and original media-associated communities were both significantly different from the colony and seawater-associated communities as well as from each other (p < 0.05).
To evaluate the consistency of colony microbiome compositions by host strain, data were subset to include only single-colony samples, and the Aitchison distances and PCoA were recomputed (Fig. 3b). This PCoA showed original colony microbiomes were differentiated by strain. Specifically, original colony microbiomes from the Kuroshio strain, which was isolated most recently and was not kept at a culture facility, cluster separately from the other original colony microbiomes. However, this separation by strain collapses when comparing microbiomes recruited to colonies after antibiotic treatment (Fig. 3b). Moreover, PERMANOVA tests by strain performed separately for original and recruited colony microbiomes showed strain-level differences were only significant (p < 0.001) among original microbiomes. for each sample type is included below the corresponding box. Note that the observed richness in the inoculating seawater is plotted on a separate y-axis due to the much higher diversity in these samples. Boxes denote the 25 th to 75 th percentile ranges for each sample type; the center bar is the median; whiskers show 1.5 times the interquartile range above the 75 th percentile and below the 25 th percentile; outlier points are values outside 1.5 times the interquartile range above the 75 th percentile and below the 25 th percentile. Media-associated bacterial communities are significantly more diverse (higher richness) than colony-associated bacterial communities (pairwise Wilcox test, p < 0.001). distance between all samples with point shape representing sample type (single colony, media, or seawater) and point color representing trial (original or recruitment). Four main clusters are visible: colony-associated communities, original media-associated communities, recruited media-associated communities, and communities in seawater. Coloration reveals that the colony-associated community cluster is split by trial. (B) PCoA of Aitchison distances between single-colony samples only, with point shape representing trial (original or recruitment) and color representing P. globosa strains. Samples separate by strain among original colony-associated communities, especially in regards to the Kuroshio colonies, but this differentiation collapses among recruited colony samples. Phaeocystis strain-level differences in colony microbiomes were significant among the original colony samples (p < 0.001, PERMANOVA, 999 permutations), but not among the recruitment trial colony samples (p = 0.117, PERMANOVA, 999 permutations).

Composition of original and recruited P. globosa microbiomes
To further examine ASVs composing Phaeocystis colony and media microbiomes, data were subset to exclude ASVs only present in seawater samples. Subsetting reduced the total unique ASV count from 781 to 341. When the relative abundance of each remaining ASV was plotted for each sample, trends in original and recruited colony-and media-associated microbiomes were readily apparent (Fig. 4). Original and recruited colony microbiomes were remarkably consistent across strains, with most colony microbiomes including the same ASVs from the orders Alteromonadales, Burkholderiales, and Rhizobiales. Original colony microbiomes also consistently included Caulobacterales ASVs, whereas recruited colony microbiomes consistently included an Oceanospirillales ASV that was not in original colony or media-associated microbiomes but was in recruited media microbiomes and in inoculating seawater. In addition, the plot highlighted two ASVs in the order Oceanospirillales contributing to the Kuroshio strain's distinct original colony microbiomes (Fig. 3b). Original media-associated microbiomes were most consistent, with almost all ASVs detected being found in at least 9 out of 12 samples. Notably, recruited ASVs were not always also detected in seawater. This is likely due to recruited ASVs representing bacteria with very low abundance in the inoculating seawater, especially since this seawater was filtered to remove particles and eukaryotic cells that likely harbor the recruited bacteria. Alternatively, original microbiomes might not have been completely removed during antibiotic treatment, but this would require that remaining bacteria were sparse enough to evade detection with fluorescence microscopy and that they were unable to grow in marine broth tests. Overall, results demonstrate the Phaeocystis colonial microbiome is a consistent subset of co-cultured bacteria in media and is primarily composed of bacteria belonging to orders Alteromonadales, Burkholderiales, and Rhizobiales. with Phaeocystis globosa colonies and in culture media. Data were subset to include only ASVs that were present in at least one single-colony or media sample (i.e., not only present in seawater)-reducing the number of unique ASVs from 781 to 341-and the relative abundance of each remaining ASV in every sample is represented by bubble size. Bubbles are colored by P. globosa source strain or inoculating seawater and are grouped by trial (original or recruitment) and sample type (single colony, media, and inoculating seawater). Columns are biological replicates of each sample type and rows are individual ASVs. ASVs are grouped on the y-axis by taxonomic orders, which appear alphabetically.

Impact of microbiomes on Phaeocystis growth when nitrate or B-vitamin deprived
Growth was monitored in axenic and nonaxenic culture replicates of each P. globosa strain in three media formulations: replete L1-Si with no modifications, L1-Si media with nitrate omitted, and L1-Si media with B-vitamin solution omitted (no added source of thiamine, biotin, or cobalamin). Growth was maintained in replete L1-Si media for all strains across the five transfers monitored (Fig. 5). In contrast, growth slowed for nonaxenic replicates of all four strains in L1-Si media without nitrate after the second transfer and ceased after the third transfer, but was unaffected in axenic replicates. Conversely, growth slowed for axenic replicates of all four strains in L1-Si media without B-vitamins after the fourth transfer, but was unaffected in nonaxenic replicates. These results suggest that when nitrogen is limiting, co-cultured bacteria compete for the limiting resource, whereas when vitamins become limiting, co-cultured bacteria mitigate limitation.

Discussion
This study shows similarly composed bacterial communities are found in co-culture with Phaeocystis globosa strains that were isolated from different ocean basins across several decades (Fig. 1, 3, 4). Moreover, original communities directly associated with individual colonies were a predictable subset of co-cultured bacteria in culture media, and bacterial communities recruited to colonies converged towards the original composition of colony microbiomes (Fig. 4). The consistency of these communities is evidence Phaeocystis globosa colonies harbor a persistent core microbiome and selective processes drive microbiome assembly. Core constituents of original and recruited colony microbiomes included bacteria belonging to orders Alteromonadales, Burkholderiales, and Rhizobiales (Fig. 4). Of these, only Alteromonadales bacteria are core components of diatom phycosphere communities [20], but Rhizobiales bacteria have been found in association with the diatom Asterionellopsis glacialis [54]. Burkholderiales and Rhizobiales are commonly found in plant rhizospheres [55,56]. Differences in Phaeocystis phycosphere community composition compared to other phytoplankton reflects the uniqueness of the Phaeocystis colony microenvironment. Furthermore, bacteria in co-culture with P. globosa mitigated nutrient stress when B-vitamins were withheld but accelerated culture collapse when nitrate was withheld (Fig. 5). This implies bacterial counterparts in the Phaeocystis phycosphere play a dual role alternating between mutualism and opportunism with their host [57].

The Phaeocystis globosa microbiome is a stable-state system shaped by selective assembly mechanisms
Bacterial communities in co-culture with long-term Phaeocystis cultures are consistently composed and stable through time. The original media microbiomes were almost identical in P. globosa CCMP1528, CCMP2710, CCMP629, and Kuroshio strains, despite each of these strains being isolated from a different ocean basin during a different decade. Likewise, the original individual colony microbiomes were strikingly similar (Fig. 3, 4). However, despite sharing many core constituents (Fig. 4), overall compositions of original Kuroshio strain colony microbiomes were significantly different from the original colony microbiomes of the other strains (Fig. 3b). Since the Kuroshio strain was isolated most recently and was not sourced from a common culture bank, its different original microbiome suggests time and place of isolation influences the composition of Phaeocystis microbiomes-as reported for other phytoplankton [58]-but time spent in a common culture facility reduces initial variability. Convergence of colonial microbiomes of all four strains, including the Kuroshio strain, after antibiotic treatment and exposure to bacteria from seawater (Fig. 3b, 4) is further evidence that Phaeocystis microbiomes are stable systems shaped by selective, deterministic assembly mechanisms: given access to the same bacteria, all four strains recruit bacterial communities that are not significantly different from each other (F = 1.26, R 2 = 0.09, p = 0.117).
Selection of P. globosa microbiome bacteria likely arises from the unique physical and chemical characteristics of colonies. First, the large size of colonies relative to individual Phaeocystis cells and other phytoplankton creates more potential habitat for bacteria and, therefore, increases the diversity that can be supported [59]. Second, the colony skin likely represents a physical barrier limiting access to the inner matrix [60], representing a first step in microbiome selection. Third, the colony matrix is made up of a unique medley of mucopolysaccharides, polysaccharides, and other organic molecules, including amino sugars [30,31,35], which would attract and support bacteria with unique enzymatic capabilities to exploit these resources [33]. Finally, additional community filtering may arise from feedback from and between microbiome bacteria, which can produce antibacterial compounds [61] or elicit phytoplankton defense responses [20]. For instance, the Phaeocystis colony matrix has high concentrations of acrylate, which has antimicrobial activity [27,62]. Together, these Phaeocystis traits lead to consistent microbiome community assembly across individual colonies and across host strains, as was seen in this study. However, the convergence of recruited microbiomes and their differences from original microbiomes also demonstrate some functional redundancy in microbiome bacteria (i.e., phylogenetically distinct bacteria can occupy the same niche) [37].

Core members of original and recruited Phaeocystis globosa colony microbiomes
Core members of the Phaeocystis globosa colony microbiome-bacteria with ASVs detected in almost all original and recruited colony microbiomes-belonged to orders Alteromonadales (family Alteromondacea), Burkholderiales (families Oxalobacteraceae and Comamonabacteraceae), and Rhizobiales (family Rhizobiaceae) (Fig. 4). Of these, only Alteromonadales bacteria are common inhabitants of diatom phycospheres [20]. In a BLAST search [63] against the NCBI nr/nt database (accessed: 10/4/2021, [64]), the three Alteromonadacea ASVs consistently abundant in colony microbiomes shared 99-100% nucleotide identity with sequences from Alteromonas macleodii. A. macleodii-a ubiquitous particle-associated marine copiotroph [65]-is an opportunistic, rather than symbiotic, member of diatom microbiomes [20] that directly competes with diatoms for nitrate [66]. Given how prevalent these bacteria are on marine particles and phytoplankton, it is unsurprising they are also associated with Phaeocystis. However, as with diatoms, we suspect the relationship between Phaeocystis and A. macleodii is more opportunistic than symbiotic in nature.
While bacteria in orders Burkholderiales and Rhizobiales are not commonly found living in association with other phytoplankton studied thus far, they are often associated with plant rhizospheres. The Burkholderiales ASV from the family Oxalobacteraceae that was consistently associated with P. globosa colonies shared 100% sequence identity with bacteria in the genus Herbaspirillum, which have been isolated from rhizospheres of many plants, and from soil and aquatic environments. Herbaspirillum strains have genomic potential to fix nitrogen, reduce nitrite, and produce siderophores [67]. Herbaspirillum strains also produce plant growth hormones-such as gibberellins and indole acetic acid [68]-and have been shown to promote rice, corn, and sugarcane growth [69]. The other core Burkholderiales ASV (family Comamonadaceae) shared 100% sequence identity with sequences deriving from the genus Delftia. Delftia spp. are considered Plant Growth-Promoting Rhizobacteria (PGPR), have capacity to produce siderophores [70], and can metabolize aromatic compounds, including benzene [71]. From the order Rhizobiales, one core ASV shared 100% identity with Ochrobactrum pseudogrigonensis sequences. O. pseudogrigonensis is a rhizosphere bacteria that promotes plant growth through induced stress tolerance, e.g., by producing antioxidants [56]. The other core Rhizobiales ASV shared 100% identity with Rhizobium radiobacter. Many R. radiobacter strains are plant pathogens, but nonpathogenic strains promote plant growth and can induce resistance to other pathogens [55]. While further work is needed to fully characterize how members of the core Phaeocystis microbiome influence Phaeocystis growth, our results indicate that, apart from A. macleodii, the core members of the P. globosa colony microbiome are broadly symbiotic bacteria with the potential to promote growth through phytohormones, nitrogen-cycling, or siderophore production.

Impact of microbiomes on Phaeocystis growth under nitrogen and vitamin deprivation
Taken together, consistent and reproducible compositions of P. globosa microbiomes and the growth-promoting potential of most core microbiome members point toward beneficial interactions between Phaeocystis and microbiome bacteria. Interestingly, growth experiments involving axenic and nonaxenic P. globosa strains grown under nitrogen and vitamin deprivation revealed contrasting bacterial effects. Axenic cultures persisted longer than nonaxenic cultures when nitrate was withheld (Fig. 5), demonstrating competition between Phaeocystis and co-cultured bacteria for this important resource. This result aligns with the high relative abundance of A. macleodii in colony microbiomes (Fig. 4), as A. macleodii competes with phytoplankton for nitrate [66]. In a natural setting, these results might indicate microbiome interactions disadvantage Phaeocystis if nitrogen is limiting. However, given that A. macleodii is often also associated with diatoms [20], diatoms may be subject to similar ecological competition with heterotrophs. Indeed, the influence of nitrogen limitation on Phaeocystis and diatom bloom dynamics is not straightforward, and each group has come out on top when nitrogen was limiting in different ecosystems [23,25,72].
In contrast to when nitrate was withheld, axenic P. globosa growth was diminished compared to nonaxenic cultures when B-vitamins were withheld (Fig. 5), demonstrating co-cultured bacteria can produce essential B-vitamins (i.e., thiamine, cobalamin; [42]) and stave off growth-diminishing vitamin stress. This is significant because B-vitamins can be limiting in coastal areas where Phaeocystis blooms occur [41,73,74]. While several major heterotrophic cobalamin and thiamine synthesizing bacteria-namely Oceanospirillales [7], Rhodospirillales, and Flavobacteriales bacteria [75]-were in our dataset, they were more consistently present and abundant in media microbiomes than in colony microbiomes (Fig. 4). Nonetheless, two Oceanospirillales ASVs were abundant core components of original Kuroshio strain colonial microbiomes, and an Oceanospirillales and a Rhodospirillales ASV were both recruited at relatively high abundance to almost all colonies sampled in the recruitment trial (Fig. 4). The bacteria represented by these ASVs could indeed contribute to vitamin synthesis in Phaeocystis colony microbiomes. However, nutrient-limitation experiments only included Phaeocystis cultures retaining original microbiomes, where most recognized vitamin-producing bacteria were not tightly associated with colonies. Therefore, it is possible that other colony-associated bacteria are vitamin synthesizers or that vitamin synthesizers were more loosely associated with colonies (i.e., not attached to colony surfaces or inhabiting the colonial matrix). A wide diversity of marine bacteria are capable of vitamin synthesis [75,76], thus necessitating evaluation of genomic potential and gene expression in Phaeocystis microbiomes to ultimately identify bacterial vitamin sources. Regardless of which microbiome constituents produce or consume limiting nutrients, challenging axenic and nonaxenic P. globosa with different limiting conditions clearly demonstrated interactions with the same total microbiome can be beneficial under one set of conditions and detrimental under another.

Conclusions and Future Directions
Though interactions between phytoplankton and bacteria occur on small scales, they have large-scale repercussions for primary production and carbon cycling in aquatic ecosystems. A more complete understanding of these interactions that includes diverse phytoplankton taxa will improve our capabilities to model and predict changes to the marine environment. This study showed Phaeocystis globosa has consistent and reproducible core microbiomes co-cultured in media and associated with individual colonies. Members of the core microbiome have potential to produce growth-promoting phytohormones, fix nitrogen, synthesize vitamins, and release siderophores. Our experiments showed co-cultured bacteria provide vitamin resources and drawdown nitrate. Further work including metagenome and metatranscriptome sequencing to assess genomic potential and gene expression in Phaeocystis microbiomes is required to fully characterize mechanisms of interaction and currencies of exchange between Phaeocystis and bacteria. Additionally, future work should expand to other bloom-forming Phaeocystis species, determine how bacterial interactions impact Phaeocystis growth under other ecologically-relevant conditions-such as iron and phosphorus limitation-and investigate how increasing sea-surface temperatures associated with global climate change will impact microbiome interactions with Phaeocystis species. Figure S1. Total sequence read counts and number of remaining sequencing reads after each processing step in the DADA2 pipeline. The plot is faceted by sample type: single colony samples (n = 74), media samples (n = 24), and seawater samples (n = 2). Boxes denote the 25 th to 75 th percentile ranges for each sample type; the center bar is the median; whiskers show 1.5 times the interquartile range above the 75 th percentile and below the 25 th percentile; outlier points are values outside 1.5 times the interquartile range above the 75 th percentile and below the 25 th percentile. Raw sequence counts for each sample are available in table format through the following link: https://maggimars.github.io/phaeosphere/recruitmentStudy/kable.html. Figure S2. Relative abundance of bacterial and chloroplast 16S rRNA gene sequences in each sample. A mean of 22% of sequences or 31 095 sequences from colony samples were classified as bacteria. A mean of 85% of sequences or 93 042 sequences from media samples were classified as bacteria. The two sequencing replicates representing the natural bacterial assemblage at the start of the recruitment trial had the largest mean proportion of sequencing reads classified as bacteria: 99.5% (mean 82 219 sequences per sample). Figure S3. Rarefaction curves for single-colony and media microbiome samples. Rarefaction was performed with phyloseq after chloroplast sequences were removed from the data and results were plotted with the function `ggrare`. The colony sample inset plot shows samples with less than 10 000 sequences remaining after chloroplast sequences were removed. All samples reach species (ASV) richness saturation.