Community Dynamics of Free-Living and Particle-Attached Bacteria over Sequential Blooms of Heterotrophic Dinoflagellate Noctiluca scintillans and Mixotrophic Ciliate Mesodinium rubrum

ABSTRACT

During a series of blooms of Noctiluca scintillans and Mesodinium rubrum, we applied high-throughput sequencing of the 16S rRNA gene to investigate the population dynamics of free-living (FL) and particle-attached (PA) bacteria in an attempt to evaluate the influence of protozoan bloom-induced disturbances on the structuring of these two communities. Our findings revealed that the FL and PA bacterial community compositions (BCCs) displayed distinct profiles during sequential blooms, and the PA flora responded more dynamically to these pulse perturbations. The dominant bacterial groups (e.g., Flavobacteriaceae, Rhodobacteraceae, Vibrionaceae, and SAR11 subclade I) in these two communities displayed different levels of connectivity with the bloom-causative species and environmental factors. In addition, more FL bacterial groups were associated with M. rubrum, while more PA bacterial groups were related to N. scintillans. Potential endocytic bacteria of N. scintillans, particularly Vibrionaceae and Rickettsiaceae, opportunistically thrived at the peak of the bloom, suggesting that they could be important players influencing the dynamics and biogeochemical cycling of the blooms. Overall, disparities in the substrate preferences and thermal niches of various bacterial taxa as well as the short duration of the blooms (1 to 3 days) contributed to the diverse responses of the FL and PA bacterial communities to these protozoan blooms. Our research provides insight into the responses of FL and PA bacterial communities to blooms caused by protozoa like N. scintillans and M. rubrum and highlights the ecological significance of certain keystone bacterial groups during this kind of cosmopolitan protozoan bloom.

IMPORTANCE Shifts in the bacterioplankton community composition during phytoplankton blooms have been studied extensively; however, investigations on protozoan blooms are rare. This study first evaluated the impact of perturbations caused by sequential protozoan blooms of the heterotrophic dinoflagellate Noctiluca scintillans and the mixotrophic ciliate Mesodinium rubrum on the structuring of these two bacterial communities. Our findings shed light on the responses of these two bacterial communities to such cosmopolitan protozoan blooms and highlight the possible ecological significance of certain keystone bacterial groups during these blooms. This research prepares the way for more focused studies that will help in understanding the roles that bacteria play during protozoan blooms and their impact on environmental health.

INTRODUCTION

Algal blooms occur in many estuarine and coastal marine ecosystems and cause dramatic changes in water quality, biogeochemical cycling, and food web dynamics (1–3). These changes can have a significant impact on the diversity, composition, and functional traits of the bacterial community (4–6).

Increasing evidence shows that particle-attached (PA) bacteria closely associated with algae and free-living (FL) bacteria in ambient water exhibit distinct responses to algal bloom events, with the former being more sensitive (7). This can also be reversed in some cases (8). Most studies on this topic have focused on photosynthetic algal species (9–11), with only a few on protozoan species, even though their massive blooms have been frequently reported (5, 12, 13). For instance, the heterotrophic dinoflagellate Noctiluca scintillans and the mixotrophic ciliate Mesodinium rubrum (also known as Myrionecta rubra) usually form massive blooms in coastal waters worldwide (13, 14). It has been reported that they interact closely with bacteria and define the survival characteristics of each other in the environment through specific interactions, e.g., mutualism, competition, cross-feeding, and predation (15–19). Previous studies, however, have rarely examined the responses of FL and PA bacteria separately to these protozoan blooms (12, 19, 20).

N. scintillans has two feeding modes, particle feeding practiced by single cells and mucoid feeding where cells connected by mucus threads scavenge the water (21). Therefore, N. scintillans can affect the abundance of bacteria by feeding on them directly (22, 23) as well as by removing them from the water column along with other potential and actual food organisms trapped in their mucoid webs (22, 24–26). Heterotrophic bacteria entrapped in the mucoid feeding web, on the other hand, decompose organic detritus and dissolved organic matter (DOM), allowing them to thrive. Moreover, N. scintillans cells produce a substantial amount of DOM, particularly during the decay phase of the bloom, due to their secretions and lysis of senescent cells. This would fill the pool of DOM and organic detritus, actively attracting and fostering opportunistic groups (22, 27). In parallel, nutrient recycling by bacteria would increase in/on and surrounding the N. scintillans mucoid web and red tide patches, refueling the second bloom (27). This would promote the regrowth of phytoplankton, some of which are the preferred food source of N. scintillans, thus prolonging the N. scintillans bloom (22, 27). Furthermore, some associated bacteria, such as Rickettsiaceae, may provide essential growth factors for N. scintillans (18). Conversely, some algicidal bacteria have also been reported to kill N. scintillans (28–30). Therefore, the relationships between N. scintillans and bacteria are complex and may change as the bloom progresses.

The mixotrophic ciliate M. rubrum can undertake photosynthesis, ingest cryptophyte organelles for a source of new organelles or nutrition (31–33), and feed on bacteria as a supplement (34, 35). M. rubrum can assimilate ammonium, nitrate, amino acids, dissolved organic carbon (DOC), and nitrogen and can excrete DOM and regenerate nutrients like other phagotrophic protists (36–38). Previous studies have shown that bacterioplankton abundances peaked after M. rubrum blooms, which was linked to the production of DOM by M. rubrum (20, 38). Romalde et al. (12) observed distinct successions of bacterial profiles and ecological functions during the formation and decay of three M. rubrum blooms, reflecting differences in the quality and quantity of the organics released by M. rubrum during the blooms. This may be attributed to its phyto- and zooplanktonic characteristics (12).

A series of blooms caused by N. scintillans and M. rubrum erupted in Port Shelter, a semienclosed bay in the eastern part of Hong Kong consisting of subtropical coastal water in the northwestern Pacific Ocean, in November and December 2014. As categorized previously by Zhang et al., based on the scale of the blooms, these bloom events can be divided into three stages: (i) 24 to 26 November 2014, with a large mixed bloom caused by N. scintillans and M. rubrum; (ii) 30 November to 1 December 2014, with a small bloom of N. scintillans followed by a small crest of M. rubrum; and (iii) 15 December 2014, with a massive N. scintillans bloom and a slight enrichment of M. rubrum (13). As mentioned above, the interactions between bacteria and protozoa during their blooms are complicated and poorly understood due to the dichotomy in the roles that protozoa play in the microbial loop (grazer and DOM provider). Therefore, in this study, we applied high-throughput sequencing of the 16S rRNA gene, together with environmental data collected over the trajectory of the sequential protozoan blooms, to (i) examine the population dynamics of FL and PA bacterial assemblages, (ii) compare the profiles of the two bacterial communities during different bloom stages, and (iii) identify the environmental factors that help shape the bacterial communities. This is the first detailed study of the changes in FL and PA bacterial community compositions (BCCs) during a complete successional trajectory of sequential protozoan blooms in coastal water. Our study also provides new insights into the ecological significance of keystone bacterial members for dynamics and biogeochemical cycling during these protozoan blooms, preparing the way for more in-depth investigations into the functions of bacteria during protozoan blooms and their impacts on environmental health.

RESULTS

Bacterial abundance and α-diversity during serial blooms of N. scintillans and M. rubrum.

The cyanobacterium Synechococcus reached its highest abundance at the start of sampling (7.60 × 10⁴ cells mL⁻¹) (Fig. 1A), which then dropped to 2,000 cells mL⁻¹ on 25 November and remained at a low level after that. The numbers of heterotrophic bacteria fluctuated from ~7.5 × 10⁵ to 15 × 10⁵ cells mL⁻¹, without a clear pattern, and peaked on 25 November and 11 December (~1.5 × 10⁶ cells mL⁻¹) when massive N. scintillans blooms occurred.

FIG 1.

Abundances (cells per milliliter) of Synechococcus (A) and heterotrophic bacteria (HB) (B) in the surface water at station PM7 from 12 November to 29 December 2014. The pink areas indicate the abundance of Mesodinium rubrum, and the blue areas indicate the abundance of Noctiluca scintillans. (Modified from reference 13.)

In total, 99,755 bacterial operational taxonomic units (OTUs) with 1,142,400 sequences were obtained in this study. In all samples, the PA bacterial community generally showed a higher Chao index (Fig. 2A) than the FL bacterial community, with the exception of three bloom peaks (Fig. 2). The Shannon diversity index followed the same trend, except at the second bloom stage when a relatively high Shannon index value of the PA bacterial community was observed. Spearman’s correlation analysis revealed no significant relationship between the abundance of Synechococcus or heterotrophic bacteria and the two bloom-causative species (see Table S1 in the supplemental material). Also, neither the Chao richness nor the Shannon diversity of the FL bacterial community was correlated with the two bloom-causative species (Table S1). In contrast, the richness and diversity of PA bacteria were significantly negatively correlated with the abundance of N. scintillans (Table S1). This result indicated that the bloom of N. scintillans has a more substantial impact on the diversity of the PA bacteria than on the diversity of the FL bacteria.

FIG 2.

Chao richness (A) and Shannon diversity (B) of free-living (FL) and particle-attached (PA) bacterial communities calculated at the OTU level during sequential blooms of Mesodinium rubrum and/or Noctiluca scintillans in Hong Kong coastal waters. The pink areas indicate the bloom periods.

FL and PA BCCs during sequential blooms of N. scintillans and M. rubrum.

The nonmetric multidimensional scaling (NMDS) plot showed a clear separation between FL bacterial communities and PA bacterial communities, indicating that bacterial lifestyle has a significant impact on the BCCs (Fig. 3). This result was further confirmed by analysis of similarity (ANOSIM) and permutational multivariate analysis of variance (PERMANOVA) (R = 0.688 and P < 0.01 by ANOSIM; R² = 0.358 and P < 0.01 by PERMANOVA) (Table 1). PA bacterial samples were further divided into four distinct clusters: samples collected on 25 November (the first peak of the mixed bloom), 15 December (the third peak of the huge N. scintillans bloom), 24 and 29 December (postbloom), and other samples (nonbloom samples). These results were consistent with the ANOSIM and PERMANOVA results, which revealed that bloom disturbance had a greater impact on the PA bacterial community (R = 0.622 and P < 0.01 by ANOSIM; R² = 0.148 and P < 0.01 by PERMANOVA) than on the FL community (R = 0.373 and P < 0.01 by ANOSIM; R² = 0.0635 and P < 0.05 by PERMANOVA).

FIG 3.

Ordination plot of a nonmetric multidimensional scaling (NMDS) analysis based on the relative abundances of OTUs of free-living (FL) and particle-attached (PA) bacteria during sequential blooms of Mesodinium rubrum and/or Noctiluca scintillans.

TABLE 1.

Pairwise comparisons of bacterial communities based on ANOSIM and PERMANOVA^a

Comparison	R value by ANOSIM	R2 value by PERMANOVA
Size fraction (FL versus PA)	0.688**	0.358**
Bloom vs nonbloom
FL bacteria	0.373**	0.0635*
PA bacteria	0.622**	0.148**

^a“Bloom” indicates the samples collected at the peaks of the first (25 and 26 November) and third (15 December) large blooms. Note that the values are the R and R² values by ANOSIM and PERMANOVA, respectively. *, P < 0.05; **, P < 0.01. FL, free living; PA, particle attached; ANOSIM, analysis of similarity; PERMANOVA, permutational multivariate analysis of variance.

Proteobacteria (average relative abundance of 59.9%) were the most abundant bacteria in the FL bacterial community across all samples (Fig. 4A). Within this phylum, Rhodobacteraceae and SAR11 subclade I dominated the FL bacterial community, with no clear pattern of variation (Fig. 4C). The phylum Bacteroidota (average relative abundance of 21.2%) was the second abundant phylum, and this phylum was dominated by Flavobacteriaceae, of which the relative abundance decreased when blooms occurred. In addition, Actinobacteriota, represented by Actinomarinaceae, also decreased during bloom periods. Margulisbacteria and Patescibacteria increased during the first bloom, while Thermoplasmatota, represented by marine group II, became abundant immediately after the blooms. Cyanobacteria comprised 11.8% to 16.6% of the FL bacterial communities at the start of sampling but became a minor part (<1.8%) of the FL communities afterward, consistent with the observations by flow cytometry (Fig. 1A). During the first mixed bloom, the relative abundance of unclassified bacteria increased by 3- to 21-fold compared to those before and after the bloom (Fig. 4A and B), implying that some novel species may be associated with the bloom.

FIG 4.

Community compositions of free-living (A and C) and particle-attached (B and D) bacteria at the phylum and family levels during sequential blooms of Mesodinium rubrum and/or Noctiluca scintillans in Hong Kong coastal waters. The minor group includes bacterial families with proportions of less than 1% of the total reads. Dotted boxes indicate the bloom periods.

Proteobacteria and Bacteroidota were also the dominant phyla in the PA bacterial community (Fig. 4B). Their compositions, however, were more diverse than those of FL bacteria (Fig. 4A). Vibrionaceae (Proteobacteria) and Flavobacteriaceae (Bacteroidota) were the most prominent families (Fig. 4D). They exhibited opposite responses to the bloom of N. scintillans. On 25 November and 15th December, Vibrionaceae were exclusively present, accounting for 72% and 56.6% of the total population (increased 100- and 10-fold over the prebloom samples), respectively, whereas the relative abundance of Flavobacteriaceae decreased from 21.0% on 24 November to 2.0% (10-fold) on 25 November and from 11.1% on 11 December to 0.8% (14-fold) on 15 December. During these two bloom periods, minor fractions of phyla such as Cyanobacteria and Planctomycetota also decreased.

The top 30 most abundant OTUs identified in the FL and PA communities (30/99,755 of the total OTUs, consisting of 50.54% of the total reads) were affiliated mainly with Proteobacteria and Bacteroidota (Fig. 5). Rhodobacteraceae HIMB11 and SAR11 subclade I were prevalent in the FL bacterial community throughout the survey and became more abundant during or after the blooms (Fig. 5A). According to the definition of an opportunist here, there were 6 OTUs in the FL population that were classified as opportunists (Fig. 6A). They were OTU3 (Vibrionaceae), OTU7 (Vibrionaceae), OTU38 (Cyanobiaceae), OTU39 (Burkholderiaceae), OTU42 (Flavobacteriaceae), and OTU153 (Pseudoalteromonadaceae). Except for OTU39, which had a high relative abundance at the beginning of the sampling period, all of these opportunistic OTUs increased promptly during or after the blooms. Nevertheless, the most striking OTUs for PA bacteria were OTU3 and OTU7, whose relative abundances dramatically increased at the plateau of the first and third blooms, respectively, yet appeared at very low levels during the pre- and postbloom periods (Fig. 5B). In the PA population, there were more opportunistic OTUs than in the FL population (Fig. 6B). The relative abundances of the opportunistic PA taxa OTU7 (Vibrionaceae), OTU39 (Burkholderiaceae), OTU101 (Silvanigrellaceae), OTU123 (unclassified Bacteria), OTU205 (Rickettsiaceae), and OTU254 (unclassified Bacteria) increased sharply during some of the blooms.

FIG 5.

Heat map showing the relative abundances of the top 30 most abundant OTUs of free-living (A) and particle-attached (B) bacteria during successive blooms of Mesodinium rubrum and/or Noctiluca scintillans in Hong Kong coastal waters. The data were square root transformed. Dotted boxes indicate the bloom periods. Each OTU identifier and taxonomic assignment is provided to the right of the heat map.

FIG 6.

Heat map showing the relative abundances of the opportunistic OTUs in the free-living (A) and particle-attached (B) bacterial communities during successive blooms of Mesodinium rubrum and/or Noctiluca scintillans in Hong Kong coastal waters. The data were square root transformed. Each OTU identifier and taxonomic assignment is provided to the right of the heat map.

Statistical Analysis of Metagenomic Profiles (STAMP) software was used to analyze the significant differences between the frequencies of bacteria (at the family level) that occurred in the FL and PA consortia (Fig. 7). It seems that Vibrionaceae and the minor bacterial group preferred a particle-attached lifestyle, as evidenced by their higher proportions in the PA community. In contrast, the SAR lineages (including SAR11 subclade I, SAR86, and SAR116) were more prevalent in the FL habitat than in the PA habitat. In addition, Rhodobacteraceae and Actinomarinaceae also showed higher frequencies in the FL population than in the PA population. Other bacterial families, such as Sphingomonadaceae, unclassified Gammaproteobacteria, unclassified Alphaproteobacteria, Halieaceae, and the NS9 marine group, showed comparable frequencies in the FL and PA habitats.

FIG 7.

Statistical Analysis of Metagenomic Profiles (STAMP) analysis of the relative abundances of bacterial families enriched or depleted between the particle-attached (PA) and free-living (FL) consortia.

Links between the bacterial communities, the two bloom-causative species, and environmental factors.

As shown in Fig. 8, the FL bacterial groups showed more connections to M. rubrum than to N. scintillans, whereas the PA bacterial groups had the opposite relationship. Notably, SAR11 subclade I in the FL consortium was negatively correlated with M. rubrum, whereas Vibrionaceae in the PA consortium were strongly positively correlated with N. scintillans and M. rubrum. Several bacterial families were also closely associated with chlorophyll a (Chl a), for example, Actinomarinaceae in the FL community and Vibrionaceae and unclassified Gammaproteobacteria in the PA community.

FIG 8.

Schematic diagram of the main free-living (FL) and particle-attached (PA) bacterial groups during sequential blooms of Mesodinium rubrum and/or Noctiluca scintillans and subnetworks built around the two bloom-causative species (N. scintillans and M. rubrum) (red circles), bacterial families (free-living and particle-attached bacteria) (yellow circles), and environmental factors (temperature, salinity, and nutrients) (blue circles). Network diagrams are built based on pairwise Spearman’s correlations, and only statistically significant correlations are shown (r ≥ |0.5|; P < 0.05). The gray and blue lines represent positive and negative correlations, respectively. The different widths of the lines represent the strengths of the correlations.

In addition, some bacterial families were closely related to environmental factors. In the FL community, for instance, Caulobacteraceae, Halieaceae, SAR11 subclade I, Actinomarinaceae, and Sphingomonadaceae were strongly correlated with temperature, as were Flavobacteriaceae, Halieaceae, marine group II, Caulobacteraceae, and the minor groups of bacteria in the PA community. In addition, Vibrionaceae in both the FL and PA consortia were favorably related to the inorganic nutrients NH₄⁺-N and PO₄³⁻-P, which was largely from the nutrient regeneration and cell lysis of N. scintillans (Fig. 8) (13). On the other hand, marine group II, the NS marine group, and SAR116 in the FL consortium were negatively correlated with NO₃-+NO₂⁻-N. Meanwhile, SAR11 subclade I, SAR116, and unclassified Alphaproteobacteria from the PA consortium were negatively correlated with DOC. Overall, there were more intercorrelations among bacterial groups but few correlations with nutrients, and bacteria were more closely linked to temperature than to nutrients.

DISCUSSION

Impacts of serial blooms of N. scintillans and M. rubrum on FL and PA BCCs.

Overall, bacterial lifestyle and bloom disturbance were important factors in the construction of BCCs during the progression of sequential blooms of N. scintillans and M. rubrum (Fig. 8 and Table 1). Nonetheless, PA bacteria fluctuated more dynamically than did FL bacteria and were more susceptible to bloom perturbation, as indicated by their stronger association with the bloom-causative species (e.g., N. scintillans) (see Table S1 in the supplemental material) and dramatic changes in diversity, richness, and community composition when a bloom occurred (Fig. 2 and 4). This was consistent with previous research on phytoplankton blooms (7, 39–42). A possible explanation for the greater influence of bloom perturbation on PA bacteria is that the durations of blooms in the current study were short (just 1 to 3 days), and their demise was ascribed mainly to the wind rather than spontaneous cell death (13). Moreover, attached to particles, bacteria were exposed to higher concentrations of organic polymers (43, 44) and were less susceptible to protozoan predation (45). It has been well recognized that the availability of substantial new particulate organic matter (POM) and DOM during a bloom can provide an array of ecological niches, being a niche-based filter that selectively promotes the growth of particular groups (4, 42, 46). Thus, bacteria with abilities for particle association, chemotaxis, and enzymatic degradation are expected to predominate during the bloom, reducing the richness and diversity of PA bacterial communities (4, 46). Likewise, we found more distinct compositions, lower richness and diversity, as well as more opportunists in the PA community than in the FL community during bloom progression (Fig. 6). Furthermore, ecological relatedness was closer and more complex in the PA consortium than in the FL consortium (Fig. 8). The development of opportunistic PA bacterial populations also differed from that of FL flora (Fig. 6).

Moreover, FL bacteria were shown to be more closely related to M. rubrum, while PA bacteria were more directly connected to N. scintillans (Fig. 8). This is probably owing to the physiological differences between these two causative species. M. rubrum, for example, swims in jumps of short durations (47), making it difficult for bacteria to attach to it. Additionally, M. rubrum is in an ambiguous trophic position in the microbial food web; it is known to excrete DOM, and it is also involved in the removal of DOM via the indirect stimulation of heterotrophic bacterial activity and perhaps direct heterotrophy, competing with bacteria (20, 36, 38). On the other hand, N. scintillans can provide a large amount of DOM for bacterioplankton either directly by the exudation (especially mucus formation) and excretion of DOM or indirectly through cell lysis (20, 27, 41, 48). The red tide-forming dinoflagellate N. scintillans studied here is purely heterotrophic, and research has shown that its cell and mucus harbor substantial amounts of bacteria (18, 49). This may be why the bloom of N. scintillans had a more significant influence on the composition and profile of the PA BCC (this study; 19, 27). However, additional research is required to understand if these associations were event specific or of general significance.

Keystone bacterial groups and their habitat preferences.

In microbial consortia, a few bacterial groups may play crucial roles as keystone species, and they are essential for bloom progression and biogeochemical cycling.

Flavobacteriaceae, well known as colonizers and degraders of alga-derived POM and polymeric DOM (4, 41, 46), were the most prevalent bacterial group in both the FL and PA consortia (this study; 4, 50). It has been demonstrated that some Flavobacteria phylotypes have evolved a PA lifestyle, while others exhibit an FL lifestyle, and their responses to bloom-induced disturbance are often distinct (51). This is also evident in our study. For example, the relative abundance of Flavobacteriaceae decreased more dynamically in the PA consortium than in the FL consortium during blooms and was enriched again immediately after the bloom. In addition, the rare taxon OTU42 in the FL consortium developed as an opportunist at the second bloom stage, whereas the opportunists OTU112 and -187 in the PA consortium increased sharply following the blooms. This indicates that Flavobacteria taxa had a high degree of metabolic niche specialization on various organic compounds released by the bloom-causative species, i.e., N. scintillans and M. rubrum, in the present study. Further study based on the metagenome approach may need to be conducted to unveil genome trait differences of FL and PA Flavobacteria taxa, which may help us understand the molecular mechanisms underlying the niche separation of Flavobacteria taxa.

The members of the Rhodobacteraceae have also been found in both FL and PA consortia, during phytoplankton blooms (52), on marine snow (53), or during protozoan blooms, as in the current study. They can associate and interact with phytoplankton and other eukaryotes through a variety of mechanisms, including the uptake of compounds produced by algae, such as peptides, amino acids, putrescine, spermidine, and dimethylsulfoniopropionate (DMSP), as well as the genetic potential for chemotaxis, attachment, and secretion (54–56). For example, OTU2, identified as Rhodobacteraceae taxon HIMB11, was prevalent in the FL consortium throughout the survey (accounting for 40 to 77% of the Rhodobacteraceae sequences), especially during and following the blooms. HIMB11 possesses the genetic potential for bacteriochlorophyll-based aerobic anoxygenic phototrophic (AAnP) metabolism, the degradation of the alga-derived compound DMSP along with the production of the climate-relevant gas dimethylsulfide (DMS), and the oxidation of the greenhouse gas carbon monoxide (54). It has been shown that a high level of DMS in the surface waters of New Zealand was closely associated with N. scintillans blooms (57). Zhou et al. (19) also found that 173 genes involved in DMSP degradation exhibited positive signal intensities, especially at the plateau of the N. scintillans bloom, and the more active degradation of DMSP to DMS (by the dsyB gene product) during the postbloom stage where Rhodobacteraceae were predominant. Regarding these facts, we suspected that the high abundance of HIMB11 associated with the N. scintillans bloom might explain the high levels of DMS at the bloom/postbloom stage of N. scintillans (19, 57). According to our current investigation and previous research (18, 19, 58), Rhodobacteraceae usually display high dominance in the bacterial community and profit from N. scintillans blooms. Additionally, N. scintillans harbors a high proportion of endocytic Rhodobacteraceae (18, 59, 60). If HIMB11 or other Rhodobacteraceae taxa with similar functions were abundant, DMS emissions would likely increase during a massive N. scintillans bloom due to DMSP degradation by a large quantity of these Rhodobacteraceae taxa either in the ambient water or associated with N. scintillans as well as the increased lysis of DMSP-producing algal cells during the phagotrophy of N. scintillans (61). Given the increased frequency, intensity, and magnitude of N. scintillans blooms in recent years (13, 14), N. scintillans blooms may play an important role in moderating global warming. Nonetheless, this is beyond the scope of the present study, and more research is required to verify this hypothesis.

SAR11 subclade I is found mainly in the FL fraction of marine bacteria (this study; 62, 63) and is prominent when the phytoplankton biomass and primary production are low (62–64). This is in line with our observation that SAR11 subclade I was negatively correlated with the abundance of M. rubrum, most likely due to the phytoplanktonic characteristics of M. rubrum (65). Nevertheless, the responses of SAR11 subclade I to the blooms of N. scintillans were not always consistent with previous reports that SAR11 was suppressed by algal blooms (63, 66, 67). For example, they thrived at the third massive N. scintillans bloom (Fig. 4C), whereupon nutrient concentrations were extremely high, and the phytoplankton abundances were at a level similar to those under prebloom conditions (13). Alonso and Pernthaler (68), using coupled microautoradiography and fluorescent in situ hybridization methods, showed that SAR11 and Rhodobacteraceae largely dominated glucose uptake in the German Bight. Galactose and glucose were the most abundant monomers in N. scintillans and its mucilage samples (48). Therefore, the availability of a substantial favorable substrate (e.g., glucose) from N. scintillans may be one of the reasons for the divergent response of SAR11 clade I to the phytoplankton/M. rubrum bloom versus the N. scintillans bloom.

Vibrionaceae have been commonly detected as endophytic bacteria with various marine protists (18, 69, 70). Members of this group can metabolize a wide variety of polymeric substrates, e.g., the abundant biopolymer chitin that is commonly found in both crustacean and diatom shells in the marine environment (71, 72), as well as an array of plant/algal polysaccharides, including agar, alginate, fucoidan, mannan, cellulose, pectin, and laminarin (73, 74). Vibrionaceae may contribute to the development of marine mucilage (75) and can utilize gelatinous material or host exudates as the sole carbon and nitrogen sources (70, 74). It has been reported that N. scintillans cells can harbor up to 10⁶ bacteria (49), and Vibrionaceae are usually some of the dominant associated bacteria (18, 60). Therefore, endosymbiosis and colonization by Vibrionaceae in mucilage may promote their tight interaction with N. scintillans cells, indicated by the significant positive correlations between N. scintillans and Vibrionaceae in the PA fraction (Fig. 8). In addition, the increases and decreases in Vibrionaceae dramatically coincided with the dynamics of N. scintillans blooms. Distinct Vibrionaceae taxa categorized as opportunists prevailed in the PA fraction at different bloom stages, for instance, OTU3 at the first stage and OTU7 at the third stage (Fig. 5B). This was unlikely to result from the FL taxa hitchhiking on zooplankton cell surfaces or bacterial colonization of their cells and mucilage, as there was few Vibrionaceae in the surrounding water. Alternatively, this could be due to the effect of temperature on the development of different Vibrionaceae taxa (76, 77), as the temperatures between the two blooms varied (23.93°C versus 19.80°C). In addition, the bloom-causative species, particularly N. scintillans, which harbors a distinct assembly of bacteria, may have originated from different water bodies. In other words, the outburst of these two blooms was not due to the biological growth of autochthonous flora. This was further supported by our previous finding that wind was the main factor driving the formation of these two large blooms (13). The importance of Vibrionaceae in N. scintillans bloom development (the trigger or the consequence) and the ecological consequences of Vibrionaceae-N. scintillans interactions are unknown. Notably, Vibrionaceae taxa can be opportunistic pathogens for homoiothermic or poikilothermic organisms (78). Therefore, it is crucial to investigate the interactions between N. scintillans and Vibrionaceae and their ecological consequences on global and ecosystem scales.

Furthermore, OTU205, belonging to the Rickettsiaceae, was classified as an opportunist in our study and accounted for 1.36% of the total sequences of bacteria in the PA consortium during the third bloom (Fig. 6B). Genome analysis of the Rickettsiaceae strain associated with N. scintillans has shown its potential to provide cofactors and amino acids to the host (18). According to a previous study, Rhodobacteria, Rickettsiaceae, Vibrionaceae, and ZB3 can be the predominantly associated bacteria in N. scintillans (18, 59, 60). It is common that a bloom-causative species harbors unique PA bacterial flora, and these bacteria may play an important role in the dynamics of harmful algal blooms (HABs) (7, 79, 80). Even though the mechanism and importance of bacterial symbiosis or parasitism in N. scintillans cells have not been fully elucidated, our results highlight the importance of considering endocytic microbial flora and bacterium-host interactions for understanding bloom dynamics as well as the microbial loop and biogeochemical cycling during blooms.

Overall, our findings support the idea that the metabolic niche specialization of bacteria on the bloom-causative species-derived substrates determined the profile and composition of microbes over the trajectory of the bloom episodes (19, 52). To better understand the bacterial response to a bloom-induced perturbation, there is a clear need for additional studies resolving interactions between specific substrate components supplied by bloom-causative species in/around the cell or in the ambient water and the specific members of the microbial community that metabolize them.

Other factors driving the changes in FL and PA BCCs.

Environmental factors invariably serve as constraint factors for bacterial communities as each bacterial taxon has its preferred environmental conditions. Fluctuations in abiotic environmental conditions result in the succession of not only the bloom-causative species but also the bacterial abundance and composition (19, 81). Nevertheless, in the current study, there were more intercorrelations among organisms than between bacteria and environmental factors (Fig. 8). This suggests that biotic factors may play a more important role than the physicochemical factors in driving the changes in BCCs during the blooming events.

Despite that, our study revealed that water temperature might also have a significant impact on the population dynamics of certain bacterial groups (Fig. 8). Of particular interest was the top dominant family of the PA community, Flavobacteriaceae, which was significantly correlated with temperature but not with the abundances of N. scintillans and M. rubrum or the nutrients that were contingent on N. scintillans (13). This indicated that temperature overrode the pulse disturbance of protozoan blooms in structuring this group within the PA consortium. Whereas the dominant family of the FL bacteria, SAR11 subclade I, was linked closely to temperature and M. rubrum abundance, the two were also significantly negatively correlated. Therefore, it is difficult to disentangle the effects of bloom-induced disturbances and temperature on the population dynamics of SAR11 subclade I.

It has been suggested that warming will promote the proliferation of protozoa, thereby exacerbating the top-down control of the phytoplankton biomass (82). Simultaneously, climate warming during the winter/early spring in temperate climate zones will favor bacterial degradation of organic matter by tightening phytoplankton-bacterioplankton coupling (83). Therefore, future climate warming scenarios may have a substantial impact on protozoan blooms in two aspects: (i) an increase in the frequency and intensity of these blooms and (ii) an increase in greenhouse gas emissions as a result of increased protozoan respiration and ingestion as well as bacterial degradation. To verify these hypotheses, however, more research is required to examine the effect of warming on protozoon-bacterium coupling and bacterial profiles and functions during protozoan blooms.

Conclusion.

In summary, sequential blooms of N. scintillans and M. rubrum had significant impacts on both the FL and PA BCCs. However, the PA bacterial community fluctuated more dramatically and was more impacted by such sudden pulse perturbations than the FL bacterial community. Variations in the BCCs between the FL and the PA communities and between different bloom stages were primarily attributable to differences in substrate utilization and habitat preference among various bacterial groups in these two assemblages. Different thermal niches of various bacterial taxa and the short duration of the blooms may also contribute to the different responses of the FL and PA bacterial communities. To better understand protozoon-bacterium interactions during protozoan blooms, a combination of molecular and classical techniques is required to determine the structure-function relationships in different bacterial consortia at different bloom stages.

MATERIALS AND METHODS

Sample collection.

The sampling methods have been described in detail previously by Zhang et al. (13). In brief, samples were taken from station PM7 (22°20.453′N, 114°17.703′E), a fixed sampling station situated in Port Shelter, Hong Kong, during a series of blooms of N. scintillans and/or M. rubrum. To collect the PA and FL bacteria, a 500-mL prefiltered (200-μm) water sample was serially filtered using 3-μm (PA bacteria)- and 0.2-μm (FL bacteria)-pore-size polycarbonate membranes (Millipore, Bedford, MA, USA). Data about total bacterial abundances, biological factors (size-fractionated Chl a and microplankton abundances), physical factors (temperature, salinity, and pH), as well as chemical factors (inorganic nutrient concentrations) have been reported previously by Zhang et al. (13). Measurement of DOC and the results (Fig. S1) are described in the supplemental material.

DNA extraction, PCR amplification, and high-throughput sequencing.

Genomic DNA was extracted from the 3-μm (PA bacteria)- and 0.2-μm (FL bacteria)-pore-size polycarbonate membranes using the PureLink genomic DNA minikit (Invitrogen, CA, USA) according to the manufacturer’s protocol, eluted with Tris-EDTA (TE) buffer (10 mM Tris, 1 mM EDTA [pH 8.0]), and stored at −20°C for later use. For pyrosequencing, the V4 and V5 regions of the 16S rRNA gene were amplified from the extracted genomic DNA using the primers 515F (5′-GTGCCAGCMGCCGCGGTAA-3′) and 926R (5′-CCGYCAATTYMTTTRAGTTT-3′). The PCR mixture contained 25 μL 2× Premix Taq (TaKaRa Biotechnology, Dalian Co. Ltd., China), 1 μL each primer (10 μM), 3 μL (20 ng μL⁻¹) DNA template, and 20 μL sterilized double-distilled water (ddH₂O). The PCR cycles for the bacterial 16S rRNA gene started with an initial denaturation step of 5 min at 94°C for initialization; 30 cycles of 30 s of denaturation at 94°C, 30 s of annealing at 52°C, and 30 s of extension at 72°C; and a final 10-min elongation step at 72°C. PCR products were purified using the PureLink quick gel extraction kit (Invitrogen) and sequenced on an Illumina Nova6000 platform, and 250-bp paired-end reads were generated (Guangdong Magigene Biotechnology Co. Ltd., Guangzhou, China).

Data analysis.

Fastp (version 0.14.1) (https://github.com/OpenGene/fastp) was used to control the quality of raw read data by a sliding window (-W 4 -M 20). The primers were removed using cutadapt software (https://github.com/marcelm/cutadapt/) according to the primer information at the beginning and end of the sequence to obtain paired-end clean reads. The clean sequences were then analyzed using the MOTHUR microbial ecology community software program (version 1.44.3) (http://www.mothur.org/wiki/Download_mothur). Briefly, the clean reads were assembled using the make.contigs command. Chimera sequences were analyzed using the chimera.vsearch command and removed from the data sets. After the above-described processing steps, high-quality sequences were aligned and identified using the SILVA (version 138) reference database. Sequences that were identified as chloroplast, mitochondria, or unknown were removed. The remaining sequences were clustered into operational taxonomic units (OTUs) with cutoff values of 3%. Each sample was rarefied to 11,900 sequences. The richness estimator (Chao index), diversity index (Shannon index), coverage, and OTUs were calculated at 97% similarity. The top 30 most abundant OTUs were displayed with a heat map using the R package Pheatmap (https://cran.r-project.org/web/packages/pheatmap/). In this study, we defined rare, common, and dominant OTUs according to their relative abundances; that is, the relative abundance of rare OTUs was <0.001, that of common OTUs was ~0.001 to 0.01, and that of dominant OTUs was >0.01. If an OTU is the “dominant OTU” in more than 1 sample but fewer than 30% of the total samples, it is an opportunist.

All statistical analyses were performed using R (R Core Team 2013). Analysis of similarity (ANOSIM) and permutational multivariate analysis of variance (PERMANOVA) were used to test if the size fraction and bloom/nonbloom could significantly explain the variation in the bacterioplankton communities. The two analyses were performed at the family level using the vegan package (84). Redundancy analysis (RDA) was conducted using Canoco (version 5) (20). To visualize the community assembly structure, nonmetric multidimensional scaling (NMDS) ordination analysis based on the relative abundance of each OTU was conducted using Primer software (version 5; Primer-E Ltd., Plymouth, UK). Spearman correlation analysis was applied to analyze the relationships among the absolute abundances of the two bloom-causative species (log transformed), nutrient concentrations, and relative abundances of bacterial families, using the R cor function (https://www.rdocumentation.org/packages/stats/versions/3.6.2/topics/cor). Only highly significant associations (r value of greater than 0.5 or less than −0.5 and P value of <0.05) were visualized in the networks with Cytoscape 3.7.2 (85). To determine the significant differences between the two bacterial communities (PA versus FL), the Statistical Analysis of Metagenomic Profiles (STAMP) software package (version 2.1.3) was used (86), with the table of the relative abundance of each family as the input. P values were calculated using Student’s two-sided t test.

Data availability.

All of the sequences obtained in the present study have been deposited in the National Center for Biotechnology Information (NCBI) Sequence Read Archive under accession number PRJNA850103.