Introduction

The composition of phytoplankton blooms, and the factors that control them, are both important aspects of global primary production. Several theories have been proposed to describe the physical–biological interactions controlling the timing of vernal (spring) blooms in seasonal seas for the phytoplankton community as a whole. One of these, Sverdrup's Critical Depth Hypothesis, predicts that phytoplankton blooms begin when the mixed layer shoals to the depth where net phytoplankton growth exceeds net losses (Sverdrup, 1953). In contrast, the Dilution-Recoupling Hypothesis predicts initiation of blooms in mid-winter, placing relatively more weight on dilution of grazing pressure because of deep mixing (DM) rather than the enhancement of phytoplankton growth rates from nutrient delivery (Behrenfeld, 2010).

The Bermuda Atlantic Time-Series Study (BATS), initiated in 1988, is one of the lengthiest oceanographic time-series studies, and includes complete records of biogeochemical and biological properties of the water column, such as nutrients, cell counts and phytoplankton pigments (Steinberg et al., 2001; Lomas et al., 2010), as well as extensive physical and meteorological measurements (Michaels and Knap, 1996). Furthermore, several studies have addressed prokaryotic population dynamics at BATS (Morris et al., 2005; Carlson et al., 2009; Treusch et al., 2009). Annual convective mixing at BATS results in the entrainment of nutrients into the photic zone initiating an annual spring phytoplankton bloom that results in carbon export (Steinberg et al., 2001). These and other systemic biological responses to nutrient variations are well described (see Michaels and Knap, 1996; Steinberg et al., 2001 and citations within). Patterns of primary production (Menzel and Ryther, 1960; Lohrenz et al., 1992; Michaels and Knap, 1996; Steinberg et al., 2001) and the temporal variability of specific phytoplankton groups have been studied extensively at BATS with a variety of approaches, including radiotracers (Brzezinski and Nelson, 1995; Nelson and Brzezinski, 1997), photosynthetic pigment proxies (Bidigare et al., 1990; Goericke, 1998; Steinberg et al., 2001; Lomas and Bates, 2004), flow cytometric measurements of phytoplankton abundance (Olson et al., 1990; Li et al., 1992; DuRand et al., 2001) and growth (Li et al., 1992; Worden and Binder, 2003a, 2003b; Cuvelier et al., 2010). Most notably, a recent study at BATS showed a >50% increase in chlorophyll a (Chl a) data between 1996 and 2007 (Lomas et al., 2010).

In subtropical and tropical open-ocean environments, phytoplankton are typically dominated numerically by marine cyanobacteria, specifically Prochlorococcus and Synechococcus. However, although lower in abundance, eukaryotic taxa can be important in these systems. For example, photosynthetic picoeukaryotes, eukaryotic phytoplankton that are <2–3 μm in size, have been shown to be responsible for 68% of the primary production in the North Atlantic (Li et al., 1993). Their large contribution to production is due to both their size, which is greater than their cyanobacterial counterparts and their high growth rates (Worden et al., 2004). Thus, although photosynthetic picoeukaryotes contributed a maximum of 10% to the ultraphytoplankton (Synechococcus, Prochlorococcus and eukaryotic phytoplankton <3 μm) cell counts in a North Atlantic study (Li, 1995), they contributed significantly more to primary production (Li, 1994). In Sargasso Sea surface waters, uncultured pico-prymnesiophytes have been shown to contribute roughly equally to primary production as Prochlorococcus at the surface (Cuvelier et al., 2010). This group of uncultured eukaryotic phytoplankton is responsible for 25% of total phytoplankton primary production in other North Atlantic regions (Jardillier et al., 2010). Another geographically widespread group of picophytoplankton belongs to the order Mamiellales within the prasinophytes. Although pigment-based methods have addressed prasinophyte dynamics as a whole, it has become clear that specific taxa within the Mamiellales have considerable differences in spatial distributions (Foulon et al., 2008; Demir-Hilton et al., 2011). Moreover, it appears that these taxa are sensitive to environmental change and one, Micromonas, has increased numerically in the arctic, where larger phytoplankton taxa have decreased (Li et al., 2009). To date, few studies have quantified contributions of these pico-prasinophytes in the open-ocean, hence there is no baseline established against which future changes can be assessed.

Rappé et al., (1995) identified 16S ribosomal RNA (rRNA) genes derived from eukaryotic plastids in environmental clone libraries and found considerable diversity. Subsequently, a number of other studies have shown extensive diversity within marine eukaryotic phytoplankton communities using 16S rRNA gene clone libraries (Rappé et al., 1997; Brown and Bowman, 2001; Wilmotte et al., 2002; Fuller et al., 2006; Jardillier et al., 2010). These results indicate that there are also many as-of-yet uncultured phytoplankton taxa, similar to findings from environmental 18S rDNA clone libraries (Moon-van der Staay et al., 2001; Not et al., 2007; Viprey et al., 2008; Shi et al., 2009; Cuvelier et al., 2010). Although the diversity of eukaryotic phytoplankton is now appreciated, there is a paucity of data resolving the temporal and spatial distributions of these organisms in open-ocean plankton communities.

In this study, we investigate phytoplankton population dynamics at BATS with a large terminal restriction fragment length polymorphism (T-RFLP) data set previously used to study bacterial plankton community structure (Carlson et al., 2009; Treusch et al., 2009). Plastid 16S rDNA clones were sequenced to identify phytoplankton T-RFLP patterns and quantitative PCR (qPCR) measurements were used to confirm observations of prasinophyte blooms. The data illustrate complex seasonal patterns in phytoplankton turnover that appear to be driven by seasonally changing environmental conditions, and show that an amalgam of spring bloom models is needed to describe the net behavior of phytoplankton in the northwestern Sargasso Sea.

Materials and methods

Sample collection, nucleic acid extraction, community profiling and clone libraries, de-noising, alignment, triplet determination and species identification

Samples and methods described previously are unchanged (Supplementary Table S1, Treusch et al., 2009) and are detailed in brief in supplementary materials. In addition to previously sequenced clones, seven new clones predicted to be of plastid origin based on the observed triplets were sequenced and deposited under the accession numbers JF968594 to JF968600. All newly identified plastid clones were from a clone library constructed from a sample collected in March 2000 at 160 m (BATS138–160).

Enumeration of Mamiellales taxa by qPCR

Samples from a selection of 2003 BATS cruises were analyzed using qPCR. Three qPCR primer-probe sets were used for detection of Mamiellales, specifically Bathycoccus, Micromonas and Ostreococcus, as described in Demir-Hilton et al. (2011). Briefly, field samples were tested for inhibition and triplicate reactions were performed in 25 μl using 1 μl of 1 ng template, 12.5 μl of Taqman Universal Master Mix (Applied Biosystems Foster City, CA, USA) with 2.5 μl of forward and reverse primers (900 nM final concentration) as well as 2.5 μl of probe (250 nM final concentration). Thermal cycling conditions for qPCR were 10 min at 95 °C (initial denaturation), followed by 45 cycles of 15 s at 95 °C and 1 min at 60 °C using an ABI7500 (Applied Biosystems). Data were collected during the annealing phase. Each run included triplicated no-template controls and 10-fold serial dilutions of plasmids containing the targeted 18S rDNA genes, composing the standard curves for each of the respective primer-probe sets. To calculate copy numbers per reaction, linear regression of Ct values against log scale copy numbers of standards were used where threshold and baseline values were calculated for each reaction using the ABI7500 software (Foster City, CA, USA). 18S rDNA copy numbers per reaction were then converted to 18S rDNA copies ml−1 based on the use of 1 ng template per reaction, which composed a known fraction of the total DNA extracted from the known volume of seawater filtered at each sampling. Extraction efficiency was assumed to be 100%. One to two copies of the rRNA operon are found in each of the three publicly available Ostreococcus genome sequences. Thus, the Ostreococcus 18S rDNA copies ml−1 values calculated here likely are either equivalent to cells ml−1 or twice the number of cells ml−1. For Bathycoccus, two copies of the rRNA operon are present in the nuclear genome (Derelle and Moreau, personal communication) and for Micromonas 2 (CCMP1545) to 3 (RCC299) are present although genome assembly issues are an issue in this region for Micromonas CCMP1545, which may have three copies. The operon copy numbers can again be used to estimate cells ml−1 for these taxa.

Environmental data

Pigment and primary production data were downloaded from the BATS webpage (http://bats.bios.edu/), adjusted similarly to the T-RFLP data to the month of DM, and plotted as a composite year (for details, see Carlson et al., 2009; Treusch et al., 2009).

Pearson correlations were determined for environmental measurements, unadjusted to the month of DM, over the duration of the data set. Average plastid signals were obtained by integrating over the entire depth profile (surface to 300 m). Summed nitrate plus nitrite and dissolved phosphate were integrated to the mixed layer depth. Daylight was calculated as the average over each month at the latitude of the BATS site, 32°N.

Results

Overview of the spring bloom in the northwestern Sargasso Sea

Interannual variability in the magnitude and timing of the spring phytoplankton bloom at BATS is determined largely by the timing of maximal convective mixing. Two previous studies, based on the same suite of BATS phytoplankton data from 1990 to 1997 (Steinberg et al., 2001) and additional data extending to 2007 (Lomas et al., 2010) have shown that the bloom can begin as early as January and as late as April. These studies show that the calendar month is irrelevant to the timing of the bloom and that rather the month of DM serves as the best indicator of the bloom onset. Therefore, we aligned data from all years according to their respective month of DM to display inter-relatable annual patterns (Figures 1, 3, 4, 5 and 6 and Supplementary Table S2) (Carlson et al., 2009; Treusch et al., 2009).

Figure 1
figure 1

Annual composite of average spatio-temporal distributions of: (a) Chl a; (b) the sum of all identified plastid T-RFLP fragments, plotted as a fraction of the summed T-RFLP signal from all amplified 16S rRNA molecules; (c) cyanobacteria (Synechococcus and Prochlorococcus, reproduced from Treusch et al., 2009); (d) 4,6-diamidino-2-phenylindole (DAPI) cell counts. Chlorophyll and DAPI cell counts were taken from the BATS website (http://bats.bios.edu). The white dashed lines indicate the average mixed layer depth that was based on a variable sigma-t (potential density) criterion (Sprintall and Tomczak, 1992). MLD was determined as the depth where sigma-t is equal to sea surface sigma-t plus an increment in sigma-t equivalent to a 0.2° C temperature decrease. Monthly data collected over 10 years was adjusted according to the month of DM and collapsed into a representative composite year (Carlson et al., 2009; Treusch et al., 2009).

Full size image

Chl a concentrations follow predictable annual patterns during the northwestern Sargasso Sea spring bloom (Figure 1a) (Steinberg et al., 2001; Lomas et al., 2010). Over the period of this data set, Chl a standing stock, integrated over the photic zone (0–140 m), increased with the onset of mixing and peaks 2 months following the DM events. Surface chlorophyll concentrations then declined and the deep chlorophyll maximum (DCM) developed between 80 and 120 m during the period of summer stratification. In the upper 5 m, maximum primary production and Chl a concentrations coincided with the month of DM (Supplementary Figure S1).

The summed signal from plastid genes contributed up to 10% to the total integrated 16S rRNA gene T-RFLP signal during the spring bloom (Figure 1b). In contrast to Chl a concentrations, which peaked in spring 2 months after DM events, the peak integrated T-RFLP plastid signal coincided with DM events (Figure 1b). Thereafter, the T-RFLP plastid signal decreased and was strongest at the DCM. The apparent difference between patterns of plastid abundance and Chl a can be attributed to a variety of factors, particularly variation in plastid gene copy number and cell chlorophyll content. Patterns in cyanobacterial distributions (Prochlorococcus and Synechococcus combined) were essentially the converse of those observed for eukaryotic plastids; cyanobacteria increased as stratification developed and reached a maximum during the period of greatest stratification (Figure 1c). These trends in phytoplankton distributions are similar to those reported in DuRand et al. (2001), based on Chl a and flow cytometry measurements. The total bacterial count demonstrated some seasonality (Carlson et al., 1996; Steinberg et al., 2001), but was constrained within a range of ca 1 × 108–6 × 108 cells l−1 throughout the year (Figure 1d).

Integrated plastid abundance (0–300 m) was highly correlated with mixed layer depth, and integrated nitrate and nitrite, but not with integrated dissolved phosphate or with daylight (Table 1). Dissolved phosphate concentrations are low and frequently undetectable at BATS, making other measures, such as mixed layer depth, a better measure of the delivery of nutrients to mixed layer phytoplankton populations.

Table 1 Pearson correlations of integrated plastid values and environmental variables
Full size table

Identification of plastid peaks in T-RFLP data

Group-level identification of plastid T-RFLP fragments was based on studies of plastid 16S rRNA genes cloned and sequenced from the BATS site. Phylogenetic reconstruction identified these clones as being derived from the Prymnesiophyceae, Cryptophyceae, Prasinophyceae, Pelagophyceae, Chrysophyceae and Bacillariophyceae, with the latter three groups belonging to the heterokonts (stramenopiles). The phylogenetic distribution of these clones was similar to that seen in an 18S rRNA gene clone library study conducted at BATS (Not et al., 2007).

Among heterokont clones, BATS138-160-34 was closely related (99.1% identity) to the diatom Thalassiosira pseudonana, and another clone (BATS138-160-16) was most closely related to Rhizosolenia sp., while a third (SAR04M78) was only distantly related to its nearest neighbors in the Bacillariophyceae (Figure 2a). Data shown in Figures 3d, 5e and 6f are a composite of the triplet signals from Thalassiosira pseudonana and Rhizosolenia sp. These taxa were detected on multiple occasions (Supplementary Figure S3). It is possible their abundance was underestimated because of unusually high genetic variability in the Bacillariophyceae at the sites cleaved by the restriction enzymes we used.

Figure 2
figure 2

(a) Neighbor-joining distance phylogenetic reconstruction of plastid 16S rRNA genes derived from eukaryotes based on 499 informative positions. Several different lineages were detected, specifically prymnesiophytes (SAR04M02, SAR04M22, SAR04M70 and SAR04M79), diatoms (BATS138-160-16 and BATS138-160-34), pelagophytes (BATS138-160-6), chrysophytes (SAR04M64) and cryptophytes (BATS138-160-3). Bootstrap values over 50% of 1000 replicates are shown. Predicted fragment sizes are indicated in brackets (BsuRI/Bsh1236I/MspI). (b) Phylogenetic reconstruction of plastid 16S rRNA genes recovered at BATS belonging to pico-prasinophytes within the order Mamiellales. The neighbor-joining distance tree was calculated from 787 informative positions. Prochlorococcus marinus str. MIT9313 served as the outgroup. Bootstrap values over 50% of 1000 replicates are indicated. Sequences from a 40 m sample collected in March 2004 (Carlson et al., 2009) were also included and are labeled SAR04M plus clone number in the phylogenetic trees.

Full size image
Figure 3
figure 3

Annual composite of average spatio-temporal distributions of phytoplankton taxa from 12 years of data, adjusted to the month of the maximum mixed layer depth (DM). Shown is the percent abundance of 16S rRNA genes for each group, relative to the total T-RFLP signal: (a) prymnesiophytes/OM21; (b) dinophytes/cryptophytes; (c) pelagophytes/chrysophytes; (d) diatoms. Note differences in z axis scales.

Full size image

BATS138-160-6 and SAR04M64 shared nearly the same T-RFLP triplet but were phylogenetically placed in the Pelagophyceae and the Chrysophyceae, respectively (Figure 2a), leading us to report a combined signal for pelagophytes and chrysophytes. Dinoflagellates have plastids that appear to have originated from cryptophytes (Keeling, 2010) and could not be resolved from cryptophytes by T-RFLP analysis; therefore, the signal from these two groups was also combined (Figures 3b, 5c and 6d).

Sequences from two genera in the Mamiellales were recovered in the 16S rRNA gene clones. These two Micromonas sequences (BATS138-160-20 and SAR04M54) and an Ostreococcus sequence (BATS138-160-35) corroborated in silico predictions (Figure 2b, Supplementary Figure S2). Throughout the manuscript, we use the term ‘pico-prasinophyte’ to refer to the T-RFLP signal from Micromonas, Ostreococcus and likely Bathycoccus, which we did not recover in the clone libraries, but has been reported previously at Hydrostation S (Anderson et al., 1996) and was resolved using qPCR (see below).

During the assignment of T-RFLP triplets to plastid lineages it became obvious that triplets determined for plastid 16S rRNA gene clones do not necessarily represent the potential diversity within a broad phytoplankton group. In silico predictions of triplets for sequences from the SILVA database (Pruesse et al., 2007) confirmed this trend. The Mamiellales were an exception, showing relatively higher conservation in T-RFLP predictions across the order. Therefore, apart from the pico-prasinophytes, our phylogenetic assignments based on the patterns detected in clones from the BATS site may under-represent overall contributions from some phytoplankton lineages and instead provide insights to particular taxa within these broad groups. This depends on the diversity within each lineage and the extent to which the T-RFLP pattern represented all members. We took a robust approach to data interpretation by relying on T-RFLP patterns from restricted cloned plastid genes for peak identification and also considering the data in the context of all available plastid 16S rRNA gene sequences. The electrophoretic mobility of clones was directly compared with T-RFLP peaks in environmental samples, reducing errors that might result from differences between observed and predicted fragment mobility.

Succession in eukaryotic phytoplankton after DM

Prymnesiophyte blooms developed strongly early in the winter period as the mixed layer began to deepen reaching as high as 1.8–4.8% of the 16S rRNA signal in the surface 80 m 1 month before the timing of maximal convective mixing (Figures 3a, 5b and 6c). The relative contribution of prymnesiophytes remained elevated (0.6–3.8% of the 16S rRNA signal) as the water column stratified, and persisted throughout the euphotic zone for several months post DM.

The relative abundance of dinophytes/cryptophytes increased during the vernal period, however, they remained a minor contributor to the overall phytoplankton population (<0.5%; Figures 3b, 5c and 6d) during the period of mixing, and were only occasionally detected during the summer (Figures 3b, 5c and 6d). Similarly, pelagophyte/chrysophytes bloomed in the first month after DM although contributing more substantially to the total 16S rRNA signal, ranging from 0.4% to 2.7% (Figures 3c, 5d and 6e). The averaged pelagophyte/chrysophyte signal also extended deeper into the water column, ranging from 0.9% to 2.7% of the 16S rRNA signal at 160 m during month +1. In the upper 100 m of the water column, their relative contribution remained approximately 1% until the onset of stratification, and persisted at 0.5–1% between 80 and 120 m throughout the summer.

Thalassiosira pseudonana and Rhizosolenia sp. like taxa were detected on multiple occasions, but not every year (Supplementary Figure S3). Their average contribution peaked following DM events (Figures 3d, 5e and 6f). Other time-series studies (Steinberg et al., 2001; Lomas et al., 2010) have concluded that diatoms appear episodically at BATS and may be becoming increasingly rare at this site. The period most heavily represented in our data set—1998–2003—corresponds to a period in which diatom abundances were shown to decline at BATS (Lomas et al., 2010). In addition, previous results have shown that when diatoms bloom in the northwestern Sargasso Sea the bloom is ephemeral lasting only days (Krause et al., 2009; Lomas et al., 2009).

Prasinophytes (Figures 4a, 5a and 6b) contributed maximally during the month of DM, on average contributing >5% to the 16S rRNA signal during this period of water column physical instability. Spring bloom pico-prasinophyte populations varied considerably over the years of the study, from below detection in some years to a maximum of 13.4%. They declined as the surface layer stratified and their signal nearly disappeared once the system was strongly stratified. To evaluate the robustness of the T-RFLP data, we also employed qPCR primer-probe sets to analyze individual genera within the Mamiellales for the final year of the study. qPCR results for Bathycoccus, Micromonas and Ostreococcus showed the same trends as the T-RFLP data for the three cruises investigated (Figure 4). The qPCR results also allowed us to differentiate between the combined T-RFLP pico-prasinophyte signal and revealed that Ostreococcus was the numerically dominant Mamiellales in the samples investigated (Figure 4b). Notably, although abundant in samples from relatively shallow depths (40 m, Figure 4b), only clade OII Ostreococcus were detected (Figure 2b, BATS138-160-35), which likely represents an open-ocean clade (Demir-Hilton et al., 2011) but was formerly hypothesized to be a deep-adapted clade (Rodriguez et al., 2005).

Figure 4
figure 4

Spatio-temporal distributions of prasinophytes at BATS. (a) Average spatio-temporal distributions of prasinophyte 16S rRNA genes, relative to the total T-RFLP signal. (b) 18S rDNA copies ml−1 enumerated by qPCR for the pico-prasinophytes Micromonas, Ostreococcus and Bathycoccus in depth profiles from the final complete year analyzed using T-RFLP methods (2003). Note differences in y axis scales between top and bottom panels.

Full size image
Figure 5
figure 5

Average distribution of phytoplankton taxa with depth, calculated from the T-RFLP data shown in Figures 3 and 4: (a) prasinophytes (see Figure 2a); (b) prymnesiophytes/OM21 clade; (c) dinophytes/cryptophytes; (d) pelagophyte/chrysophyte clades; (e) combined signal of diatom clones BATS138-160-16 and BATS138-160-34.

Full size image
Figure 6
figure 6

Average seasonal change in relative abundance of phytoplankton taxa, integrated over depth (0–160 m): (a) modeled clear sky PAR; (b) prasinophytes; (c) prymnesiophytes/OM21 clade; (d) dinophytes/cryptophytes; (e) pelagophytes/chrysophytes and (f) bacillariophyte clones BATS138-160-16 and BATS138-160-34 (diatoms). Note that for scaling, the data were normalized to the maximum abundance of the relative fluorescence observed for each taxon.

Full size image

Patterns in phytoplankton vertical distributions

The lineages detected in our plastid data differed in their vertical distributions (Figure 5). Pico-prasinophytes and prymnesiophytes have similar average maximum abundance peaks, between 80 and 120 m for the former and 120 m for the latter. However, the dinophyte/cryptophyte and pelagophyte/chrysophyte signals were more uniformly distributed throughout the photic zone and extended deeper in the water column. Interestingly, diatoms were not detected in the surface sample (Figure 5e). Plastid T-RFLP signals were also sometimes detected in the mesopelagic, where light should be insufficient for photosynthetic growth, suggesting export from the photic zone (Figures 5a, c–e).

Discussion

16S rRNA gene T-RFLP analysis has previously been shown to deliver reliable, reproducible results for the analysis of large sample sets (reviewed in Nocker et al., 2007). In some cases, it has been shown to be more sensitive than pigment analyses (Stachowski-Haberkorn et al., 2009), although clearly it complements such analyses and is subject to different biases. Of these the most significant is likely to be variation in gene copy number between different phytoplankton species; there are two copies of the 16S rRNA gene in many plastid genomes, but many phytoplankton species contain more than one plastid, and plastids can contain many genome copies (Ersland et al., 1981). For this and other reasons, T-RFLP data provide only a relative indication of cellular abundances. The fact that qPCR analysis of pico-prasinophyte genera at BATS corresponded well with the T-RFLP data provides one example confirming the reliability of T-RFLP for measurements of relative microbial abundance (Figure 4). Many of the interesting observations that emerged from this T-RFLP analysis are based on the broad taxonomic range and comparative power of this technique, including the early bloom onset and seasonal succession of taxa, and the depth differentials observed between taxa (Figure 6).

Succession in eukaryotic phytoplankton

We observed blooms of the pico-prasinophytes Ostreococcus and Micromonas that have not previously been reported in the open-ocean (Figures 4a and b). These taxa have reduced genomes and are distributed throughout tropical and temperate regions of the oceans; Micromonas and Bathycoccus are also found in Arctic systems (Lovejoy et al., 2007; Worden et al., 2009). Little quantitative data exists for this suite of taxa in open-ocean settings. The blooms observed here occurred during and just after the deep winter mixing period (months 0 (DM) and +1) with a decline in contributions through the remainder of the year. Thus the distribution of Ostreococcus and Micromonas appears to be related to factors associated with DM rather than day length. This supports the hypothesis that the primary determinant of interannual variability in these populations is nutrient input caused by deep convective mixing. Our data are consistent with the observations of Lomas and Bates (2004) who attributed 2–8% of integrated Chl a stocks at BATS to prasinophytes during spring, based on high-performance liquid chromatography measurements of prasinoxanthin. However, in contrast to our T-RFLP and qPCR results, Goericke (1998) found that prasinophytes were not abundant between March 1986 and February 1987 based on high-performance liquid chromatography data. Our qPCR analyses (Figure 4b) indicated there were up to 5478±164 Ostreococcus 18S rDNA copies ml−1, likely relating to ∼2739 or 5478 cells ml−1, depending on whether one or two copies of the operon are present in the genome, respectively. The discrepancy between this study and that of Goericke (1998) is likely due to the fact that in the latter samples were not collected between March and June (with March roughly corresponding to a common time for the DM event). Low prasinophyte contributions at other times of the year reported by Goericke are consistent with our observations, as were other results, such as relatively low overall contributions of dinophytes/cryptophytes and diatoms. Overall, the periods of highest pico-prasinophyte contributions (Figure 4) track periods during which picocyanobacteria as a whole had the lowest contributions to the overall 16S rRNA gene signal (Figure 1c).

Our study revealed that prymnesiophytes consistently form an intense bloom early in the year. In contrast to the pico-prasinophytes, the bloom was before the DM event. Moreover, their contributions remained high throughout spring mixing, into early summer as the water column stratified (that is, +4, Figure 3a) and their contributions remained substantial at the DCM for the remainder of the year (Figures 5b and 6c). Given that the clones used to identify the prymnesiophyte T-RFLP signal were broadly distributed across this group, these trends do not preclude a shift from one prymnesiophyte taxon (for example, coccolithophores; which previously have been reported for wintertime in this region) to another (Figure 2a). High-performance liquid chromatography studies indicate that most of the prymnesiophyte biomass is in the picoplankton size fraction (for example, Goericke, 1998). More recently, molecular approaches have shown that pico-prymnesiophytes comprise ∼25% of the picophytoplankton biomass in the Atlantic, with increasing contributions at higher latitudes (Cuvelier et al., 2010; Jardillier et al., 2010). In the Sargasso Sea, they have been shown to contribute ∼22% of picophytoplankton carbon at the surface and DCM (Cuvelier et al., 2010). Both Jardillier et al. (2010) and Cuvelier et al. (2010) also showed that the pico-prymnesiophytes responsible for much of this biomass and production were from uncultured taxa. The 16S rDNA sequences retrieved here also appeared to be from uncultured prymnesiophytes (Figure 2a).

The DCM

The persistence of the chlorophyll signal in the DCM throughout the summer (Figure 1a) reflects both a shift in the distribution of phototrophs to greater depths as surface nutrients become depleted and the influence of photoadaptation, which causes an offset between the DCM and the region of highest phytoplankton density and water column productivity. At BATS, the DCM phytoplankton community is numerically dominated by picocyanobacteria (Figure 1c), specifically Prochlorococcus (for example, DuRand et al., 2001). As discussed above the cellular biomass of eukaryotic phytoplankton taxa are greater than that of the picocyanobacteria, hence even when fairly low in abundance their biomass contributions are likely still significant. Like Steinberg et al. (2001), we observed prymnesiophytes in the region of the DCM during periods of intense water column stratification. We also observed pelagophytes in the DCM, but their overall contributions appeared low, about 1% of T-RFLP signal (Figures 3c, 5d and 6e). Steinberg et al. (2001) also observed pelagophytes in the DCM but concluded they were among the most abundant eukaryotic phytoplankton at BATS. Although these data are in contrast to our findings, the dates of their study do not overlap significantly with ours. Hence, it is possible this discrepancy reflects ecosystem level changes, as reported for diatoms (Lomas et al., 2010). The observation of pelagophytes and other taxa in the summer DCM might have implications for understanding the genesis of the spring bloom; populations that endure the summer in the relative refuge of the DCM could seed blooms as the mixed layer deepens during the winter. But, at present, not enough is known about phytoplankton diversity to be certain that the taxa that occur in the region of the DCM are genetically and phenotypically identical to those that constitute winter and spring bloom populations.

Spring bloom models

Behrenfeld (2010) and Sverdrup (1953) provided a relatively broad theoretical basis for understanding how physical, chemical and biological factors might interact to control phytoplankton bloom dynamics. Others have modified these concepts by considering the impact of variations between ocean regions. For example, Siegel et al. (2002) proposed that there are two distinct regimes for spring bloom dynamics in the North Atlantic Ocean. North of 40°N, the pattern follows Sverdrup's (1953) hypothesis. South of 40°N, where mixing is reduced and depth of the euphotic zone is often >100 m, it was proposed that nutrients, rather than light, are the limiting factor, causing the spring bloom to be initiated early. According to this low(er) latitude model, nutrients entrained into the euphotic zone during convective mixing trigger a phytoplankton bloom in advance of mixed layer shoaling. Thus BATS, located at 32°N, was considered to be outside the region where Sverdrup's hypothesis applies.

Our observations show clearly that mixing drives the timing of the earliest blooms of phytoplankton taxa. This conclusion is supported by correlations (or lack thereof) between integrated stocks of N and P, mixed layer depth, day length, and the integrated plastid signal (Table 1). The low correlation between the integrated plastid signal and dissolved phosphate is unsurprising because phosphate is rapidly reduced to undetectable levels after its delivery into the euphotic zone by mixing events. These data fit predictions of Behrenfeld's Dilution-Recoupling Hypothesis, which emphasizes the relaxation of predation caused by the dilution of predators and prey as the mixed layer deepens. But the data are also consistent with Siegel's hypothesis (2002) that the mixing of nutrients into the photic zone initiates the spring bloom in the northwestern Sargasso Sea. The data we present do not distinguish between these possibilities. However, the data do illustrate details of the diversification of phytoplankton into niche space at BATS and contribute to the perspective that phytoplankton turnover is driven by changing physical and biogeochemical conditions over the course of the annual spring bloom in the northwestern Sargasso Sea. As Siegel and coworkers highlight, laboratory measurements have ascribed Ic values ranging from 0.1 to 0.8 mol photons m−2 day−1, to different phytoplankton species, which could dramatically alter growth rates during winter mixing when incident photosynthetically active radiation (PAR) is low (Figure 6a) and cells spend more time deeper in the water column as a result of advection (Siegel et al., 2002). Other factors, such as adaptations for predator or virus evasion, half saturation constants for nutrients, and differing temperature optima, might also have important roles in determining the temporal distribution of phytoplankton blooms. For example, pico-prymnesiophyte adaptations to low nutrients inferred from population level genome sequence data, are consistent with their presence during stratified conditions observed here (Cuvelier et al., 2010). Our results provide insight into the partitioning of the niche space by phytoplankton taxa at BATS. These insights should facilitate development and testing of mechanistic hypotheses about factors controlling these taxa, and may lead to improved predictions of phytoplankton responses to changes in ocean conditions.