Introduction

It is well known that bacteria and phytoplankton dynamics are closely linked in coastal marine environments, with correlations frequently observed between bacterial and phytoplankton biomass [14]. Many studies have shown interactions between phytoplankton and bacterial communities through the measurement of bulk community parameters, such as production, respiration, and nutrient fluxes [e.g., 4, 5, 26, 36]. In contrast, little is known about how these communities interact with each other at the species composition level. Interest in harmful algal blooms has revealed that marine bacteria are capable of stimulating or inhibiting phytoplankton growth [13, 49], killing phytoplankton [24, 29, 33, 49], or altering phytoplankton physiology (i.e., production of algal toxins) [19]. Several researchers have reported evidence for species-specific interactions between bacteria and phytoplankton, which has led to the conclusion that bacteria can play a major role in controlling phytoplankton dynamics [16, 18, 24, 49, 50]. For example, Fukami et al. [15] found that natural bacterial communities collected during a Gymnodinium nagasakiense bloom inhibited Skeletonema costatum, but stimulated G. nagasakiense. Later, Fukami et al. [17] isolated Flavobacterium sp. str. 5 N-3, which was found to have algicidal properties against G. nagasakiense but to have no effect on Chattonella antiqua, Heterosigma akashiwo, or S. costatum. These findings have important implications, especially if they hold true for interactions between bacteria and members of the general phytoplankton community that form the base of carbon cycling in coastal marine environments.

There is also some evidence that differences in the quality of organic matter produced by different types of phytoplankton cause shifts in the species composition of bacterial communities utilizing this organic matter [48]. Phytoplankton are known to release up to 25% of the total organic carbon fixed by photosynthesis into the surrounding “phycosphere,” or region immediately surrounding and influenced by algal cells [9]. This dissolved organic matter (DOM) is then rapidly consumed and remineralized by the bacterial community [4, 5, 26, 36]. Thus, changes in phytoplankton community composition may influence the composition of bacterial communities that function as part of the microbial loop.

Although phytoplankton can be identified using conventional microscopy and morphology-based identification, until recently it was not possible to reliably assess the composition of bacterial communities. The development of molecular techniques has dramatically improved bacterial community analyses and has made it possible to assess how bacterial and phytoplankton communities interact at a species composition level. In the current study, we used denaturing gradient gel electrophoresis (DGGE) analysis of bacterial 16S rDNA genes to investigate community dynamics of free-living bacteria (FLB) and attached bacteria (AB), operationally defined here as bacteria associated with particles in the 5–100 μm diameter size range. We then used ordination analyses to assess whether FLB or AB community dynamics were linked to changes in the phytoplankton community. Our results indicate that species composition shifts in the AB and phytoplankton communities were correlated. These relationships were not evident between the phytoplankton and FLB communities, suggesting that a close physical association was necessary for links between bacterial and phytoplankton community dynamics to occur.

Methods

Study Sites and Sample Collection

Seawater samples were collected from several stations in the southwestern Bay of Fundy (Fig. 1) on 23 February 2000, 2 May 2000, 31 July 2000, and 25 September 2000 from the R/V Pandalus. This area has been well studied by the Canadian Department of Fisheries and Oceans, which has maintained an intensive program analyzing phytoplankton community diversity and abiotic factors since 1987. With the widest tidal ranges in the world, these well-mixed waters are rich in nutrients and highly productive. The Bay of Fundy also harbors high cell densities of the toxic dinoflagellate Alexandrium fundyense [25, 31, 38, 39] and may serve as a source of this harmful alga for other regions in the Gulf of Maine.

Figure 1
figure 1

Study site and sampling stations in the Bay of Fundy.

Full size image

For bacterial community analyses, near-surface seawater samples were obtained by using polypropylene buckets rinsed twice with 70% ethanol and with seawater from the sampling location. Seawater was successively filtered through 100-μm diameter pore-size filters to remove large particulates (e.g., zooplankton and marine snow, which were not analyzed), and subsequently filtered through 5-μm and 0.22-μm pore-size filters to collect attached and free-living bacteria, respectively. Filtration was conducted at sea immediately after sample collection and filters were frozen on dry ice and held at −70°C until analysis. For phytoplankton community analyses, a near-surface sample was taken by bucket and 250 mL was immediately preserved with 5 mL formalin:acetic acid [44].

Phytoplankton Community Composition

Fifty-mL fixed seawater samples were allowed to settle for 16 h in counting or sedimentation chambers. All phytoplankton larger than 5 μm were identified and enumerated (as cells L−1) using a Nikon inverted microscope. Further identification was done using either a JEOL JSM-5600 scanning electron microscope (SEM) or a Hitachi 2400 SEM [44].

DNA Extraction, PCR Amplification of 16S rDNA Fragments, and DGGE

Community DNAs were extracted from filters using FastDNA SPIN Kits, as recommended by the manufacturer (Bio 101, Vista, CA). The resulting DNA extracts were used as template for PCR with primers 338F-GC (the complement of EUB338) [2] and 907R [28]. PCR and DGGE analysis of PCR products was conducted as previously described [43], with a 55–70% denaturant gradient (where 100% is equivalent to 7 M urea and 40% formamide) [35] for 16 h at 70 volts. The concentrations of PCR products were determined by agarose gel electrophoresis and quantitative analysis of ethidium bromide–stained gels using molecular mass standards and a Gel Doc 2000 digital gel documentation system (Bio-Rad, Hercules, CA). For each sample, ~1000 ng PCR product was analyzed by DGGE. After DGGE, isolated bands were excised and pulverized with a sterile mortar and pestle, and DNA was eluted overnight in 50 μL 0.1 M Tris, pH 8.0, at 4°C. Partial 16S rDNAs were then reamplified from excised bands, as described above, and analyzed by a second DGGE in order ensure that heteroduplexes were resolved. In addition, for these second DGGEs, bands from different samples that appeared to migrate to the same or similar position in the original analysis were placed in adjacent lanes to confirm their relative positions.

DGGE profiles of the free-living and bacterial communities were analyzed using Quantity One gel documentation software (Bio-Rad) in order to determine the position and intensities of individual bands. DGGE profiles of communities were analyzed by subtracting the background fluorescence from each lane, and then band intensities were normalized to the total intensity of all bands in a given lane, to give relative band intensities.

DNA Sequence Analysis

Isolated bands from second DGGE gels were excised and 16S rDNA fragments were reamplified, except that no GC clamp was incorporated into the forward primer. These PCR products were purified using Qiaquick PCR preps (Qiagen, Valencia, CA) and were sequenced using an ABI 373A automated sequencer (Applied Biosystems, Foster City, CA) at the University of Massachusetts Department of Microbiology or a CEQ 2000 (Beckman-Coulter, Fullerton, CA) automated DNA sequencer at the UMass Lowell Biological Sciences Department. 16S rDNA sequences were checked for potential chimeras with the Ribosomal Database Project II (RDP II) Chimera Check program [30]. Sequences were then aligned with closely related 16S rRNA sequences from GenBank and the RDP II and alignments were edited manually using the program SeqPup. Unambiguously aligned base positions were used to construct phylogenetic trees with maximum likelihood and maximum parsimony methods using PAUP* (Sinauer Associates, Sunderland MA).

Ordination Analyses of Community Composition Data

Community ordination analyses were conducted using relative band intensity data from the bacterial communities and relative abundance data from the phytoplankton community. Canoco for Windows 4.0 (Microcomputer Power, Ithaca, NY) was used to conduct correspondence analysis (CA) and canonical correspondence analysis (CCA) ordinations. For CA and CCA ordinations, biplot scaling was used and was focused on intersample distances. The data were not transformed.

In order to determine whether attributes of the phytoplankton community accounted for significant variation in the species composition data of FLB and AB communities, CCA was used to analyze the bacterial communities with phytoplankton community characteristics as explanatory variables. The following variables were chosen as descriptors of the phytoplankton community and were calculated for each sample: Shannon’s diversity index (H); total number of phytoplankton cells L−1 (N); and the relative abundance of diatoms. This last term was chosen to reflect large shifts in community composition, e.g., communities dominated by diatoms or other types of phytoplankton. Shannon’s diversity index for the phytoplankton community (H P ) was calculated as follows:

$$ H_P = - \sum\limits_{i = 1}^{S_P } {(p_i )({\text{ln }}p_i )} $$

where S p is phytoplankton species richness and p i is the relative abundance of taxon i. These variables were used as explanatory variables in separate CCA ordinations of the FLB and AB communities. In order to directly compare free-living and attached bacterial communities on a single CCA ordination diagram, inputs of species composition data were such that each “sample” consisted of a given community type (FLB or AB) from a given sampling station and date. For example, FLB and AB collected on 2 May 2000 were treated as different samples. A nominal explanatory variable, which was set equal to 0 for FLB and 1 for AB communities, was included in the analysis. Thus, variability in the species composition data that was due to community type was accounted for by this nominal variable, and remaining variability due to other explanatory variables (i.e., phytoplankton community characteristics) could be analyzed separately. Using this approach, it was possible to effectively remove variation due to bacterial community type (attached vs free-living) and analyze variation in the data set that resulted from other factors.

GenBank Accession Numbers

The sequences were deposited in GenBank under accession numbers AY353551–AY353562.

Results

Development of the Seasonal Phytoplankton Bloom

Analysis of the phytoplankton community is described by Savin et al. [44] and summarized in Fig. 2. A total of 88 taxa were identified at the species or genus level. In May, the total numbers of phytoplankton in the water column were relatively low (1140–2540 cells L−1; Fig. 2). By 31 July 2000, a bloom had developed, with total cell numbers reaching ~250,000 cells L−1 at some sites (Fig. 2). The highest phytoplankton cell densities were observed at the Wolves, Deadmans Harbour, and Brandy Cove sites, where the community was dominated by diatoms. Total phytoplankton densities were much lower in Passamaquoddy Bay and Bocabec Bay, and at these sites, dinoflagellates outnumbered diatoms (Fig. 2). At all sites except for Passamaquoddy Bay, the number of phytoplankton cells declined from July to September and the community was composed almost entirely of diatoms. In contrast, total cell densities increased in September at Passamaquoddy Bay and dinoflagellates dominated the community at that site.

Figure 2
figure 2

Absolute abundances of all phytoplankton species (total phytoplankton), all dinoflagellate species (total dinoflagellates), all diatom species (total diatoms), and all “other” species (total other; includes Dictyocha sp., Dinobyron sp., Ebria tripartita, Euglenophyceae, and Mesodinium rubrum).

Full size image

DGGE Profiles of Attached and Free-Living Bacterial Communities

Relatively little diversity and variability were observed in the FLB community by DGGE (Fig. 3). A total of 21 DGGE bands were observed, with three to seven bands per sample, and, of those, several were present in all or most samples (Fig. 3). These results indicate that the community was dominated by relatively few phylotypes.

Figure 3
figure 3

DGGE profiles of 16S rRNA gene fragments amplified from the free-living bacterial community. Bands from which sequence data were obtained are marked.

Full size image

Although the AB community was also dominated by relatively few phylotypes, it exhibited both greater diversity and variability than the FLB community, with a total of 31 bands and four to 12 distinct bands per sample (Fig. 4). A paired t-test indicated that the number of DGGE bands in the AB community was statistically significantly greater than that of the FLB community (P = 0.006). In comparison with the free-living community, DGGE profiles also varied more among samples taken from different sites on the same date. This between-station variability was especially evident in samples taken in July and September, when the absolute abundances of phytoplankton were highest and most variable (Fig. 2).

Figure 4
figure 4

DGGE profiles of 16S rRNA gene fragments amplified from the attached bacterial community. Bands from which sequence data were obtained are marked.

Full size image

Community Ordination Analyses

CA ordinations of the FLB, AB, and phytoplankton communities are shown in Fig. 5. The first four ordination axes explained 79.8%, 72%, and 76.4% of the total variation in species data for the FLB, AB, and phytoplankton communities, respectively. The percentage of variation explained by the first two ordination axes for each community is shown in Fig. 5.

Figure 5
figure 5

Correspondence analysis (CA) ordination diagrams of (A) the phytoplankton community; (B) the attached bacterial community; and (C) the free-living bacterial community. Only samples with data available for all three community types were included in the analysis. Sample abbreviations are as follows: MW, MD, and ML are samples taken on 2 May 2000 from stations Wolves, Deadmans Harbour, and Lime Kiln, respectively; JW, JD, JL, JP, and JB are samples taken 31 July 2000 from stations Wolves, Deadmans Harbour, Lime Kiln, Passamaquoddy Bay, and Bocabec Bay; SW, SD, SL, SP, and SBr are samples taken 25 Sept 2000 from stations Wolves, Deadmans Harbour, Lime Kiln, Passamaquoddy Bay, and Brandy Cove, respectively.

Full size image

Four distinct clusters of sites are evident in the phytoplankton community CA diagram (Fig. 5A). The “May Group” includes all sites analyzed on 2 May 2000, the “July Group” includes the Wolves, Deadmans Harbor, and Lime Kiln sites on 31 July 2000, the “July Pass-Boca” group includes Passamaquoddy Bay and Bocabec sites on 31 July 2000, and the “September Group” includes all sites analyzed on 25 September 2000. The symbols used to represent each sample in the FLB and AB communities reflect these clusters observed in the phytoplankton CA diagram. The AB diagram shows clusters similar to the phytoplankton groups, while the FLB diagram does not (Fig. 5 B,C). In the AB community, with the exception of the September Lime Kiln site, the samples are separated into the May Group, the July Pass-Boca Group, the July Group, and the September Group, as defined by the phytoplankton clusters. In contrast, in the FLB community, all July and September samples cluster together and the May Group is dispersed.

CCA ordination diagrams using phytoplankton community characteristics as explanatory variables are shown in Fig. 6. In order to analyze both the free-living and attached bacterial communities in the same ordination, a “sample” was defined as a given station sampled on a given date for either the free-living or attached bacterial community. Thus, the data input for sample “May-Wolves-free-living” included all the DGGE bands from the free-living community at the Wolves station on the May sampling date, with values of 0 entered for all the attached bacteria DGGE bands. A dummy explanatory variable that was set equal to 0 for FLB samples and to 1 for AB samples was included in the analysis. Using this approach, the first ordination axis, which has the highest eigenvalue (0.71), was highly correlated with the dummy explanatory variable that differentiated FLB and AB (correlation coefficient of −0.984) and not correlated with the phytoplankton community characteristics that comprised the remaining explanatory variables (correlation coefficients with axis 1 ≤ |0.052|; Fig. 6A). This result is evident from Fig. 6A, in which sample scores for axes 1 and 2 of the FLB and AB communities were plotted and the two community types are separated clearly along ordination axis 1.

Figure 6
figure 6

CCA of attached bacteria (AB) and free-living bacteria (FLB) communities using phytoplankton community characteristics (diversity [H], total cell number [N], and % diatoms) as explanatory variables, as well as a “dummy variable” that differentiates between free-living bacteria (0) and attached bacteria (1). Arrows indicate the direction of increasing values of each phytoplankton variable, with the length of the arrows correlated with the percentage of community composition data accounted for by that variable (“Type” refers to bacterial community type, i.e., attached or free-living). Sample abbreviations are as follows: MW, MD, and ML are samples taken on 2 May 2000 from stations Wolves, Deadmans Harbour, and Lime Kiln, respectively; JW, JD, JL, JP, and JB are samples taken 31 July 2000 from stations Wolves, Deadmans Harbour, Lime Kiln, Passamaquoddy Bay, and Bocabec Bay; SW, SD, SL, SP, and SBr are samples taken 25 Sept 2000 from stations Wolves, Deadmans Harbour, Lime Kiln, Passamaquoddy Bay, and Brandy Cove, respectively.

Full size image

This approach enabled us to effectively remove variation due to bacterial community type (i.e., attached or free-living) and analyze remaining variation that was due to other factors, such as phytoplankton community characteristics. Ordination axes two, three, and four, with eigenvalues of 0.57, 0.29, and 0.14, respectively, were not correlated with the community type dummy variable (with correlation coefficients with community type variable ≤ |0.023|), but were instead derived from variability remaining in the data set after community type (i.e., ordination axis one) had been taken into consideration (Fig. 6B). CCA ordination axis two was moderately correlated with the relative abundance of diatoms in the phytoplankton community (correlation coefficient of −0.71) and the total number of phytoplankton cells L−1 (−0.57). Axis three was moderately correlated with phytoplankton species diversity (Shannon’s diversity index) (−0.71) and the total number of phytoplankton cells L−1 (0.55). It is evident from the diagram of ordination axis two vs. axis three (Fig. 6B) that the samples from the FLB community cluster closer to the origin than those from the AB community, with the latter exhibiting a greater dispersion in ordination space. Thus, phytoplankton community characteristics such as total number of cells L−1, species diversity, and the percentage of diatoms accounted for more variation in the AB than in the FLB community.

Phylogenetic Diversity of Bacterial Communities

Phylogenetic analyses placed 16S rRNA gene fragments from the attached and free-living bacterial communities into two major phylogenetic groups, the Roseobacter group of the alpha Proteobacteria, and the Cytophaga– Flexibacter– Bacteroides (CFB) group (Table 1). Sequences AFB-1 and AFB-2 were found at all stations on all sampling dates, in both the free-living and attached bacterial communities. Both sequences were members of the Roseobacter group, with AFB-1 most closely related to Sulfitobacter mediterraneus str. CH-B427 (98% similarity) and AFB-2 most closely related to Roseobacter litoralis (96.4% similarity). An additional sequence obtained from the FLB community samples, FB-1, was extremely closely related to Roseobacter sp. str. Shippagan (99.8% similarity). The AB community also contained several additional Roseobacter group sequences (Table 1, Fig. 7). Several Cytophaga group sequences were retrieved from the attached bacterial community, although they did not exhibit very high similarities to their closest relatives (91.3–94%, Table 1).

Table 1 Closest relatives of retrieved 16S rRNA gene sequences
Full size table
Figure 7
figure 7

16S rRNA phylogenetic tree of sequences obtained in this study (bold) and members of the Roseobacter litoralis subgroup, constructed using a maximum likelihood method. Bootstrap values (out of 100 trees) are shown adjacent to nodes and were generated using maximum parsimony trees.

Full size image

Discussion

Although many studies have shown close links between bulk parameters of bacteria and phytoplankton communities in coastal marine environments, little is known about how these communities interact at the species composition level. We used detailed morphological species composition data of the phytoplankton community [44] and molecular techniques to analyze the free-living and attached bacterial communities in the Bay of Fundy to determine whether relationships between bacteria and phytoplankton community dynamics existed. Ordination analyses of phytoplankton community data and bacterial community DGGE profiles enabled us to analyze trends that would otherwise be difficult to detect because of sample complexity. Our results indicate that phytoplankton community dynamics were more closely correlated with the attached than the free-living bacteria. These results indicate that specific associations between phytoplankton and the bacteria adhered to them may be important in controlling the species composition dynamics of these communities.

DGGE Analysis of Bacterial Communities

The potential biases inherent in PCR-DGGE methods used are well documented and have been discussed previously [12, 37]. In order to avoid PCR bias as much as possible, we used low-stringency annealing temperatures with a minimum number of amplification cycles. In addition, several studies have suggested minimal PCR bias using primers that target the same region that we amplified, with effective detection of community dominants [22, 41, 43]. We chose to use both DGGE band presence and relative intensity (normalized to total band intensity for a given lane, or community) as input for ordination analyses. Our intention was not to infer that relative band intensity reflected the relative abundance of a given phylotype, but instead to take into account qualitative differences between communities. Several studies have shown that DGGE patterns, including relative intensity of DGGE bands, are reproducible and therefore appear to reflect differences in template DNA composition [7, 43, 45]. Using CCA ordination analysis and artificial data sets, Muylaert et al. [34] recently showed that potential biases associated with relative band intensity data, such as preferential DNA extraction and/or PCR amplification, did not obscure CCA results. They also found that including relative band intensity information yielded better correlations between CCA explanatory variables and bacterial community composition data [34]. Given that there is no method currently available to comprehensively determine microbial community structure, DGGE remains an effective tool to meet this study’s objective of analyzing relative temporal and spatial changes in dominant community members.

DGGE profiles revealed that there were few community dominants that were well adapted to ecological conditions in the Bay of Fundy, although it is likely that many more species comprised minor community members that were not detectable by DGGE [47]. Two phylotypes were found in all FLB and AB samples. Schauer et al. [45] used DGGE to analyze the FLB community in northwestern Mediterranean coastal samples and also found several dominant members that were detected year-round. However, seasonal variability in abiotic conditions is more dramatic in the BOF than the Mediterranean, and it was therefore somewhat surprising that there were several community members that remained dominant from February to September.

As described in the Methods section, FLB and AB were operationally defined as those organisms collected by sequential filtration onto 5-μm diameter pore size filters and 0.2-μm diameter pore size filters, respectively, after prefiltration with 100-μm pore size filters. This approach has been used to separate FLB and AB communities previously [8, 32, 45] and enabled rapid processing of samples at sea. Visual assessment of filters after sample processing indicated that most phytoplankton were collected on the 5-μm pore-size filters, which were greenish brown, while 0.2-μm pore-size filters were white or cream-colored. However, it is possible that some FLB were collected on the 5-μm pore-size filter, as pores may have clogged, or, alternatively, that AB were dislodged from particles during filtration and subsequently collected on the 0.2-μm pore size filter, thereby leading to imperfect separation of the two community types. Thus, differences in community types that were observed should be considered an underestimate of true differences in FLB and AB communities.

As in other studies comparing AB and FLB communities in aquatic environments, DGGE profiles (Figs. 3, 4) and sequence analysis (Fig. 7 and Table 1) indicated differences between the two community types. The AB community exhibited greater diversity (paired two-tailed t-test, P = 0.006) and spatial variability than the FLB community, especially in July and September, when absolute abundances of phytoplankton were highest and most variable (Fig. 2) and the species richness of the phytoplankton community was highest [44]. The finding that AB diversity was greater than FLB diversity was in contrast with bacterioplankton communities in offshore Mediterranean waters, in which FLB exhibited much higher diversity than AB [1], but similar to the increased diversity observed in the AB community during the development of a diatom bloom in a nutrient-enriched mesocosm [42].

Links between Community Dynamics of Attached Bacteria and Phytoplankton

Ordination analysis of the AB community suggested that shifts in AB composition were correlated with changes in the phytoplankton community. For example, separate CA ordinations of the phytoplankton and AB communities show several analogous clusters of sites (Fig. 5). CA is an indirect gradient analysis method in which community composition data alone are used to infer ordination axes. Thus, samples that cluster in ordination space have similar community compositions whereas those that are dispersed are less similar. Using symbols that represent different phytoplankton community clusters, Fig. 5 shows that the same clusters are evident in the AB community, with the exception of one AB sample (September Lime Kiln). In contrast, these clusters are not clearly separated in the FLB community, in which most July and September samples formed one group and the May Lime Kiln sample is distant from other May samples (Fig. 5), suggesting that FLB community dynamics were not linked to those of the phytoplankton.

In order to explore the relationships between the phytoplankton and bacterial communities further, we used a direct gradient analysis method (CCA) with phytoplankton community characteristics as explanatory variables (Fig. 6). In contrast with CA, CCA relies on explanatory variables to derive ordination axes. Both AB and FLB communities were included in the same ordination in order to enable a direct comparison of the two community types within the same ordination space. As explained above, this was accomplished by using a community-type dummy variable. This approach eliminated the problem of comparing the dispersion patterns of communities in separate ordinations, each with abstract axes that have arbitrary units. As expected, the first ordination axis was strongly correlated with the community type variable (correlation coefficient of |0.984|) and showed clear separation of the AB and FLB communities (Fig. 6A). The remaining ordination axes can be used to examine variation remaining after community type is taken into consideration. In both CCA diagrams, the samples collected in May form clusters that are distinct from other samples. This result reflects the fact that DGGE profiles of the May bacterial communities were distinct (Figs. 3, 4), as well as the fact that phytoplankton abundances were dramatically lower in May than at other sampling times (Fig. 2). In Fig. 6B, the July Wolves and July Deadmans Harbour AB communities formed outliers, apparently as a result of their distinct bacterial DGGE profiles (Fig. 4), high densities of phytoplankton cells (N) (Fig. 2), and relatively low phytoplankton diversity (H). It is evident from the diagram of ordination axis 2 vs axis 3 (Fig. 6B) that the samples from the FLB community cluster closer to the origin than those from the AB community, with the latter exhibiting a greater dispersion in ordination space. Thus, phytoplankton community characteristics such as total number of cells L−1, species diversity, and the percentage of diatoms account for more of the variation in the AB than in the FLB community.

Both CA and CCA results indicate an association between phytoplankton and attached bacterial community dynamics that would be expected if specific bacteria–phytoplankton interactions occurred. Marine bacteria phylogenetically related to those found in this study (see below) are known to be colonizers of particulate matter [6] and are likely to be utilizing organic compounds provided by phytoplankton cells. AB have several potential advantages over FLB, including greater access to nutrients, efficient use of extracellular enzymes, and shelter from predation [6]. Although AB have been found to be less abundant than FLB, representing as little as 10% of the total community [32], they generally have much greater activity per cell and have been shown to account for up to 92% of total bacterial productivity during phytoplankton blooms [46]. AB have been shown to have high ectohydrolase activity and are likely to actively release organic carbon from the phytoplankton cells they colonize [32, 46]. Organic substrates provided by phytoplankton cells include cell wall components (e.g., chitin and cellulose), secreted mucilage, photosynthates, and amino acids. In addition, some phytoplankton have been shown to produce antimicrobial compounds, thereby inhibiting bacterial growth or limiting it to resistant bacteria. As the availability and qualitative nature of organic compounds in the phycosphere varies among phytoplankton species, the bacteria utilizing them are also likely to vary. Several studies have provided circumstantial evidence for such associations. For example, Smith et al. [46] found bacterial colonization of one species of Chaetoceros but absence of bacterial colonization of another Chaetoceros species during a diatom bloom. Hold et al. [23] found different bacterial assemblages associated with different dinoflagellate species, suggesting species-specific interactions. Similarly, Fukami et al. [15] described species-specific interactions between Flavobacterium sp. str. 5 N-3 and Gymnodinium nagasakiense. To our knowledge, this study is the first to describe correlations between phytoplankton and attached bacterial community composition in nature.

Significance of the Roseobacter Group

DGGE profiles and sequence analysis revealed that members of the Roseobacter subgroup within the Rhodobacter group of the α-Proteobacteria were found in both the FLB and AB communities at all stations sampled and at all sampling times, from 23 February to 25 September 2000. Two sequences in particular, AFB-1 and AFB-2 were present in all samples, with additional Roseobacter group sequences detected in the FLB (FB-1) and the AB community (AB-5, AB-6, AB-7, AB-8, and AB-9). These results confirm previous reports of the importance of this group in coastal bacterial communities [6, 10, 11, 21]. Members of the Roseobacter group are thought to be numerically dominant [20] and to be primary colonizers of surfaces in coastal marine environments [6]. The Roseobacter group has also been shown to include members that have close associations with marine algae. For example, Roseobacter group species have been isolated from galls of marine red algae [3] and the phycosphere of the toxic dinoflagellate Prorocentrum lima [27]. This group has also been indicated as a predominant community member after the peak of a diatom bloom in nutrient-enriched marine mesocosms [40]. The persistent dominance of the Roseobacter group observed here indicates that it is well adapted to the nutrient-rich, productive waters of the Bay of Fundy.

In conclusion, our results show that shifts in the community composition of eukaryotic phytoplankton and bacteria attached to them are correlated, suggesting that the dynamics of these two communities are linked. In contrast, patterns of community composition of free-living bacteria were not correlated with those of the phytoplankton. To our knowledge, this is the first study showing links between phytoplankton-attached bacterial community dynamics in a natural coastal environment. The results indicate that specific interactions between phytoplankton and attached bacteria may occur and that such interactions could be important in controlling the composition of both communities.