Eight species of marine phytoplankton were selected for this study to investigate variability in the composition of organic matter released by unicellular photoautotrophs at various taxonomic levels (Table 1). Individual strains were chosen on the criteria that a published sequenced genome is, or will soon be available, and that axenic strains were capable of growth to high cell densities (ca. 10⁷ cells/ml) in natural seawater-based media. All media was prepared by the National Center for Marine Algae and Microbiota (NCMA) using filtered, autoclaved surface water collected from the Sargasso Sea. Media amendments were added aseptically after sterilization. Triplicate 1 L batch cultures of each strain were prepared alongside triplicate 1 L controls containing all media amendments, but no cell additions. Cells were pre-conditioned in 60–400 ml of the appropriate medium prior to inoculation in 1 L.

Thalassiosira pseudonana (CCMP str. 1335), Thalassiosira rotula (CCMP str. 1647), Phaeodactylum tricornutum (CCMP str. 632), and both Synechococcus strains (WH8102 and WH7803) were grown at the NCMA in L1 medium prepared according to existing protocols (Guillard and Hargraves, 1993). Diatom strains were grown at 20°C and Synechococcus strains at 24°C. All five strains were grown on a 13/11 light/dark cycle with gentle periodic mixing. Synechococcus str. WH7803 was grown under 10–20 μmol photons/m²/s illumination, while the remaining strains were grown at 100–120 μmol photons/m²/s. Diatom cells were enumerated using a Palmer-Maloney counting chamber at 40X magnification on a Zeiss light microscope. Synechococcus cells were enumerated via epifluorescence microscopy. Three Prochlorococcus strains (MED4, MIT9301, and MIT9313) were grown at 22°C in Pro99 medium prepared by the NCMA according to existing protocols (Moore et al., 2002, 2007) with the addition of 5 mM sodium bicarbonate. All three strains were grown on a 14/10 light/dark cycle. Prochlorococcus str. MIT9313 was grown under low light conditions at ca. 20 μmol photons/m²/s, while MIT9301 and MED4 were grown at ca. 40 μmol photons/m²/s. Prochlorococcus cultures were monitored for growth using total fluorescence and for changes in pH, which were mitigated by additions of 2 mM HEPES buffer and 5 mM sodium bicarbonate 6 days after inoculation (sensu Moore et al., 2007). Samples for direct cell counts by flow cytometry were fixed with 0.125% final concentration of grade I glutaraldehyde (Sigma) and stored at –80°C prior to analysis. Fixed samples were diluted in 0.2 μm filtered seawater and Prochlorococcus were enumerated using an Influx Cell Sorter (BD Biosciences) as previously described (Olson et al., 1985; Cavender-Bares et al., 1999).

Samples for cell enumeration and particulate and dissolved organic carbon (POC and DOC, respectively) quantification were taken at the onset of stationary phase growth to evaluate organic carbon production by each strain. An exception was that DOC data for Prochlorococcus strains was not acquired due to the addition of HEPES buffer during growth to mitigate pH changes (sensu Moore et al., 2007). All organic carbon samples were processed using combusted (450°C for 8 h) glassware. Duplicate samples for POC analysis were taken by vacuum filtering 25 ml of sample onto a combusted 25 mm 0.7 μm glass fiber filter (Whatman GF/F). Filters were then placed inside a combusted glass petri dish, wrapped in foil, and immediately frozen. A blank filter was also prepared at each sampling point. Filters were later thawed and dried (60°C; overnight) before encapsulation into 9 × 10 mm tin capsules. Measurements for POC quantification were performed by the University of California Davis Stable Isotope Facility. POC filtrates were transferred into combusted glass vials, acidified with 150 μl of a 25% phosphoric acid solution and measured for DOC using high temperature catalytic oxidation on a Shimadzu TOC-V_CSH with platinized aluminum catalyst. Sample concentrations were determined alongside potassium hydrogen phthalate standards and consensus reference materials provided by the DOC-CRM program (www.rsmas.miami.edu/groups/biogeochem/CRM.html).

Upon entering stationary phase, cells were removed by centrifugation at either 878 RCF (diatom cultures) or 10,751 RCF (cyanobacteria cultures) for 15 min, followed by gentle filtration (0.1 μm; Whatman Polycap 36 TC capsule filter). Filtrates were then stored briefly in the dark at 4°C until solid-phase extraction (SPE). Material obtained via SPE constitutes 20–60% of the total DOC in marine DOM (Dittmar et al., 2008), encompassing a large amount of organic substrate available to heterotrophic bacterioplankton. Media controls were processed and stored alongside the culture samples. Filtrates were acidified to pH 2–3 by addition of trace metal grade hydrochloric acid and organic matter was extracted onto ISOLUTE C18(EC) SPE columns (0.5 g, Biotage) at a rate of 1 ml/min. SPE columns were preconditioned with 5 ml HPLC-grade methanol followed by 10 ml ultrapure water. After sampling, mineral salts were washed from the columns with acidified ultrapure water (pH 2–3) at a flow rate of 1 ml/min. Organic matter was recovered by gravity elution using 10 column volumes of acidified HPLC-grade methanol (pH 2–3). Samples were concentrated to a small volume by rotary evaporation, and then taken to dryness under filtered, high purity nitrogen. Samples were resuspended in 156 μl of methanol, and stored briefly in combusted amber vials at 4°C in the dark prior to chemical analysis. A 3 μl subsample of each was placed onto a combusted 25 mm, 0.7 μm glass fiber filter (Whatman GF/F), and submitted for POC analysis to quantify the organic carbon recovered by SPE.

Chromatographic and spectrometric analyses were performed using an Agilent 1200 series high-performance liquid chromatograph coupled to an Agilent 6130 (single quadrupole) mass spectrometer. Organic extracts were separated on a ZORBAX SB-C18 column (Agilent; 3.5 μm 4.6 × 150 mm) eluted at 0.8 ml min⁻¹ using a linear gradient (% solvent A, % solvent B, minutes): 100, 0, 0; 20, 80, 31.25; 0, 100, 43.75; 0, 100, 64, where solvent A is aqueous formic acid (0.1%) and solvent B is methanolic formic acid (0.1%). Mass spectrometry was performed using an atmospheric electrospray ionization (ESI) source. Drying gas was set at 11.5 l min⁻¹ and 300°C, the nebulizer was at 60 psig, and capillary voltage was set to 4000 V. Data was obtained in the positive ion mode from 100–2000 m/z with a 4.0 fragmentor, 150 threshold, and 0.1 step size. A tune solution containing five standard compounds (117–2122 m/z) was used to determine mass accuracy of the instrument prior to analysis. Mass accuracy in this size range was found to always be within a tolerance of 0.2 Da.

Mass spectral data was analyzed using MZmine 2 molecular profiling software (Pluskal et al., 2010). Ions with signal intensity at least 5-fold greater than the maximum noise level were identified using a centroid mass detector. Chromatograms were then built from the raw data using a retention time tolerance of ±5 s, and a mass tolerance of ±0.3 m/z. Alignments were made to correct for small shifts in retention times between samples and individual peaks were identified by setting a minimum acceptable intensity (5-fold greater than the noise level) and duration (5 s) to remove noise within each chromatogram, and also by searching for local minima within each chromatogram. Adduct peaks ([M+Na]⁺, [M+2Na-H]⁺, [M+H₂PO₄]⁺, [M+HSO₄]⁺) and isotopic peaks were identified and removed within a retention time tolerance of ±1.42 s, and a mass tolerance of ±0.2 m/z. Additional adduct peaks for Prochlorococcus ([M+NH₄]⁺) and both Synechococcus and diatoms ([M+K]⁺) were identified and removed due to the presence of these chemical species in the media types used to grow these strains. Constraints were placed on the intensity ratio of identified isotope pairs based on the minimum ((M–1)/30) and maximum ((M–3)/14) number of carbon atoms for a given mass. Sample feature lists were then aligned and gaps were filled in using a secondary threshold of 3-fold greater than the noise to identify common peaks that fell just below the initial strict threshold. A feature was defined as a unique m/z at a specific retention time, thus multiple features could share a specific m/z or retention time, but never both. While the possibility of distinct metabolites sharing identical retention times and m/z values within the tolerance levels mentioned above cannot be ruled out, the use of two independent parameters (polarity and mass) to designate DOM_P features reduces the likelihood of falsely identifying common features.

Biological triplicates of each strain were first compared against triplicate media controls to distinguish material produced by the organism from any background material, including DOM present in the seawater medium used to cultivate each strain. Features were considered to be associated with a particular organism only if they were present in all three biological replicates of that strain and absent in all three replicate controls of the appropriate media type. Removal of features present in sterile media controls reduced the possibility of differences arising due to variations in growth media (i.e., Pro99 medium vs. L1 medium). Features were also removed if they were present in any blank samples, including triplicate instrument blank injections of pure methanol and triplicate processing blanks created by rinsing and eluting pre-cleaned resins without any prior sample loading. Feature lists for each organism were then aligned to identify both common and unique features among the eight strains tested. Aligned feature lists were also used to investigate intensity level differences among features common to multiple samples.

To provide a rigorous analysis of the similarity between all cultures, an aligned feature list containing all the features detected in every replicate was created and converted into a binary matrix where each column represented an individual culture and each row a feature present in at least one culture. Multiplying this matrix by its transpose created a second matrix where each element was the number of features common to each pairwise culture comparison. The matrix was then normalized by the total number of features present in each pairwise comparison. This normalized similarity matrix was used to quantify the similarity of feature lists and their parent strains, and to generate a cluster diagram as a visual representation of DOM_P similarity among strains. The number of features shared by any two cultures was divided by the sum of features present in those cultures to generate a percent similarity value for all pairwise sample combinations.

HPLC-ESI-MS detected 577, 673, and 463 features in samples from strains MIT9313, MIT9301, and MED4, respectively. Purity broth tests revealed subsequent heterotrophic contamination in one of the MED4 replicate cultures; therefore this sample was removed from all analyses (see N/A in Table 1). Approximately 40% of the features associated with each Prochlorococcus strain were detected in all replicates of that strain (Table 1). MIT9313 replicates had the least agreement of all the strains in this study (Figure 1A). Prochlorococcus-derived DOM_P generally consisted of small, non-polar material, with the majority of features detected between 200–500 m/z and 20–50 min, when the mobile phase composition was between 51 and 100% methanol (Figure 2).

Of the total Prochlorococcus-derived features detected in this study, 13% were produced by all three strains. High-light adapted strains MIT9301 and MED4 were the most similar Prochlorococcus strains, sharing 34% of their features. Low-light adapted strain MIT9313 shared approximately 22% of its features with both high-light adapted strains (Figure 3A). Unique features (as much as 52% of the total) were identified in all three strains. Comparing all replicates as individual samples revealed that biological replicates were generally more alike (27–67% similar) than cultures of different strains (12–41% similar), although both MED4 samples (MED4B and MED4C) had greater similarity to replicates of MIT9301 than to each other (Table 3).

HPLC-ESI-MS detected 773 and 434 features in samples from strains WH8102 and WH7803, respectively. Approximately one half to three quarters of the features associated with each Synechococcus strain were detected in all three replicates of that strain (Table 1). The agreement among biological replicates of WH7803 was representative of the average agreement among replicates for all of the strains tested (Figure 1B). The majority of Synechococcus-derived features were detected between 150–500 m/z and 12–50 min, when the mobile phase composition was between 31 and 100% methanol (Figure 2). Of the total Synechococcus-derived features detected in this study, 14% were found associated with both strains. Unique features (as much as 83% of the total) were identified in both strains (Figure 3B). Comparing all replicates as individual samples revealed that all biological replicates were more alike (42–72% similar) than cultures of different strains (13–21% similar) (Table 3).

Among the features consistently detected in Prochlorococcus and Synechococcus samples, 14% were found associated with all five cyanobacteria strains. The most similar strains between the two groups were Prochlorococcus str. MIT9301 and Synechococcus str. WH8102, sharing 20% of their features. All Prochlorococcus strains shared more features with each other than with either Synechococcus strain. Synechococcus str. WH7803 shared more features with the other Synechococcus strain tested (WH8102) than with any of the Prochlorococcus strains; however the reverse was not true. Synechococcus str. WH8102 had the most overlap with Prochlorococcus str. MIT9301, sharing 20%, followed by MIT9313 (15%) and then WH7803 (14%). Low-light adapted Synechococcus str. WH7803 was more similar to the low-light adapted Prochlorococcus str. MIT9313 than it was to either of the high-light adapted Prochlorococcus strains. The percent of features shared between all cyanobacteria strain combinations are summarized in Table 4.

Comparing all replicates as individual samples confirmed that Synechococcus str. WH8102 and Prochlorococcus str. MIT9301 were the most similar Synechococcus and Prochlorococcus strains (15–23% similar) with respect to the DOM features analyzed, and also revealed that WH7803 and the high-light adapted Prochlorococcus strains (MIT9301 and MED4) were the least similar cyanobacteria tested (6–14% similar). In general, all of the Prochlorococcus strains were more similar to WH8102 (11–23% similar) than WH7803 (6–14% similar) (Table 3).

T. pseudonana (CCMP1335) and T. rotula (CCMP1647) were chosen to represent different species of the Thalassiosira genus. HPLC-ESI-MS detected 973 and 1065 features in samples from T. pseudonana and T. rotula, respectively. Approximately one half to five sixths of the features associated with each Thalassiosira strain were detected in all three replicates of that strain (Table 1). The majority of Thalassiosira-derived features were detected between 150–700 m/z and 20–45 min, when the mobile phase composition was between 51 and 100% methanol (Figure 2). Of the total Thalassiosira-derived features detected in this study, 11% were found associated with both strains. Unique features (as much as 82% of the total) were identified in both strains (Figure 3C). Comparing all replicates as individual samples revealed that all biological replicates were more alike (47–71% similar) than cultures of different strains (12–15% similar) (Table 3).

HPLC-ESI-MS detected 1190 features in P. tricornutum samples. Approximately three fourths of the features associated with each P. tricornutum culture were detected in all three replicates (Table 1). P. tricornutum replicates had the most agreement of all the strains in this study (Figure 1C). Phaeodactylum-derived DOM_P generally consisted of larger, more polar material when compared to cyanobacteria DOM_P, with the majority of features detected between 150–1000 m/z and 12–45 min, when the mobile phase composition was between 31 and 100% methanol (Figure 2). Comparing all replicates as individual samples revealed that the biological replicates were more alike (66–73% similar) than cultures of any other strain (2–10% similar) (Table 3).

In general, DOM_P composition was more diverse among the diatom strains than among the cyanobacteria. Among the features consistently detected in the diatom strains tested, only 2% were found associated with all three strains. T. pseudonana and T. rotula were the most similar diatoms, sharing 11% of their features. T. rotula and P. tricornutum shared 6% of their features, while T. pseudonana and P. tricornutum shared 5% (Figure 3C). A large number of unique features (as much as 87% of the total) were detected in all three diatom strains. T. pseudonana and T. rotula were both more similar to each other than to any of the cyanobacteria strains tested and P. tricornutum was more similar to both Thalassiosira diatoms than to any cyanobacteria strain. Comparing all replicates as individual samples revealed that biological replicates were more alike (47–73% similar) than cultures of different strains (7–15% similar), and confirmed that T. pseudonana and T. rotula are the most similar diatoms (12–15% similar), followed by T. pseudonana and P. tricornutum (7–10% similar), and T. rotula and P. tricornutum (7–9% similar) (Table 3).

The 2032 DOM_P features detected in this study were organized into three groups: those detected in all Prochlorococcus strains, those detected in all Synechococcus strains, and those detected in all diatom strains. Of the 122 features that can be classified this way, only 5% (6 features) were common to all three groups. Prochlorococcus and Synechococcus were the most similar groups, sharing 14% of their features, followed by diatoms and Prochlorococcus sharing 10%, and finally diatoms and Synechococcus sharing 8% (Figure 3D). All three groups contained a large quantity of features that were unique to that group; the Synechococcus group had the most with 54 unique features, or 79% of its total. Combining the Prochlorococcus and Synechococcus groups into one cyanobacteria group and comparing this with the diatom group reduced the number of total classifiable features to 39. The same 6 features (now 15% of the total) mentioned above were common to both groups, while 8 features (21%) were unique to the cyanobacteria group and 25 features (64%) were unique to the diatom group.

Comparing all replicates as individual samples revealed that all five cyanobacteria strains were more similar to each other (12–41% similar) than to any diatom strain tested (2–12% similar). Comparisons between cyanobacteria and diatom strains revealed Prochlorococcus str. MIT9301 and T. rotula, and Synechococcus str. WH8102 and T. rotula to be the most similar strains between these two groups, both sharing 8% of their features. Prochlorococcus str. MED4 and P. tricornutum were the least similar strains, sharing only 1% of their features. Comparing all replicates as individual samples confirmed that all five cyanobacteria strains were most like T. rotula (4–12% similar), followed by T. pseudonana (3–8% similar) and finally P. tricornutum (2–6% similar). T. pseudonana and P. tricornutum were both more similar to all diatom strains than to any cyanobacteria tested. While T. rotula was most like T. pseudonana (11% similar), it had more features in common with Prochlorococcus str. MIT9301 (8% similar) and Synechococcus str. WH8102 (8% similar) than with P. tricornutum (6% similar). All three diatom strains were more like Synechococcus str. WH8102 (4–12% similar) than any other cyanobacteria strain (2–10% similar) and more like Prochlorococcus str. MIT9301 (3–10% similar) than any other Prochlorococcus strain (2–7% similar). Similarity values are summarized in Table 3, while the percent of features shared between strains are shown in Table 4, alongside their rRNA gene sequence distances. Unique features were detected in all eight phytoplankton strains that were not found in any other strain. P. tricornutum was the most distinct strain with 587 unique features (49% of its total), while Prochlorococcus str. MED4 was the least distinct with 30 unique features (6% of its total). DOM_P composition was most similar at the clade-level (14–34% of features in common) and least similar at the domain-level (1–8% of features in common).

Molecular weight and polarity distributions of DOM_P were analyzed for additional connections between DOM_P composition and phytoplankton phylogeny. While all eight strains produced low-molecular-weight (≤1466 m/z) material over a broad polarity spectrum, cyanobacteria-derived DOM_P was generally comprised of smaller, less polar features when compared to diatom-derived DOM_P (Figure 2). P. tricornutum cultures in particular produced large polar DOM, while T. rotula produced smaller DOM that more closely resembled cyanobacterial profiles. The low-light adapted Synechococcus str. WH7803 produced slightly larger DOM, resembling centric diatom profiles.

While much of the observed variation in DOM_P composition appeared related to producer phylogeny, the data also suggested connections between the composition of DOM_P and the locations from which these strains were originally isolated (Table 5). Among the diatoms tested, the DOM_P composition of P. tricornutum was found to be more similar to that of T. pseudonana, than to that of T. rotula. Both P. tricornutum and T. pseudonana were isolated from the North Atlantic and under lower temperature conditions than T. rotula, which was isolated from the Mediterranean Sea. We also found that DOM_P derived from Prochlorococcus str. MED4 was more compositionally similar to that of T. rotula (the only other strain isolated from the Mediterranean Sea) than to either of the other diatom strains tested. Additionally, DOM_P derived from Prochlorococcus str. MIT9313 as well both of the Synechococcus strains tested was more similar to that of Prochlorococcus str. MIT9301 than to that of Prochlorococcus str. MED4. MIT9301, MIT9313, and both Synechococcus strains were all isolated from the Sargasso Sea and MIT9313 and MIT9301 were both isolated from deeper depths than MED4. Prochlorococcus str. MIT9301 and Synechococcus str. WH8102 had the highest correlation in DOM_P composition between the two cyanobacteria groups. These two strains were originally isolated from the lowest latitudes and most pelagic locations of all the strains tested in this study (Table 5).

Although strictly quantitative assessments within a sample are not feasible due to uncertainties in ionization efficiency in ESI, if we assume that matrix effects are minimal and that abundance scales linearly with signal intensity, then the intensity level of features detected in multiple samples offers semi-quantitative information regarding common features associated with different phytoplankton strains. Comparing a particular strain with its respective medium control can identify material present in the seawater-based medium that were also produced by the organism as well as highlight material consumed by phytoplankton during growth.

Scatter plots of signal intensity provide a visual representation of semi-quantitative differences among features common to multiple samples (Figure 4). Intensity comparisons above and below a threshold of 4-fold were chosen because the majority of common features (93% on average) among all replicate samples and blanks were within a 4-fold intensity difference (Figure 4A). A representative comparison of common features found in a culture sample (Prochlorococcus str. 9301 replicate C) and its respective media control (Pro99 medium replicate C) demonstrates that while the majority of features (95% on average) are within a 4-fold intensity difference, there was material present in the medium produced (dots above the upper 4:1 line) and consumed (dots below the lower 4:1 line) during phytoplankton growth (Figure 4B). Pairwise comparisons of different strains exhibited greater intensity differences than among replicates, and the degree of intensity differences between strains varied widely. On average, 91% of common features among different Prochlorococcus strains were within a 4-fold intensity difference (Figure 4C). An average of 82% of common features among the different Synechococcus strains were within a 4-fold intensity difference, while 77% of common features were within a 4-fold intensity difference among the different diatom strains. On average, 84% of features common to Prochlorococcus and Synechococcus were within a 4-fold intensity difference, while an average of 75 and 68% of common features were within a 4-fold intensity difference when comparing Prochlorococcus to diatoms and Synechococcus to diatoms, respectively (Figures 4D–F).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.