Resource partitioning of phytoplankton metabolites that support bacterial heterotrophy

The communities of bacteria that assemble around marine microphytoplankton are predictably dominated by Rhodobacterales, Flavobacteriales, and families within the Gammaproteobacteria. Yet whether this consistent ecological pattern reflects the result of resource-based niche partitioning or resource competition requires better knowledge of the metabolites linking microbial autotrophs and heterotrophs in the surface ocean. We characterized molecules targeted for uptake by three heterotrophic bacteria individually co-cultured with a marine diatom using two strategies that vetted the exometabolite pool for biological relevance by means of bacterial activity assays: expression of diagnostic genes and net drawdown of exometabolites, the latter detected with mass spectrometry and nuclear magnetic resonance using novel sample preparation approaches. Of the more than 36 organic molecules with evidence of bacterial uptake, 53% contained nitrogen (including nucleosides and amino acids), 11% were organic sulfur compounds (including dihydroxypropanesulfonate and dimethysulfoniopropionate), and 28% were components of polysaccharides (including chrysolaminarin, chitin, and alginate). Overlap in phytoplankton-derived metabolite use by bacteria in the absence of competition was low, and only guanosine, proline, and N-acetyl-d-glucosamine were predicted to be used by all three. Exometabolite uptake pattern points to a key role for ecological resource partitioning in the assembly marine bacterial communities transforming recent photosynthate.

2 concentration 0.75X; Life Technologies, Waltham., MA, USA). Samples were analyzed on an Agilent Quanteon flow cytometer (Acea, Biosciences Inc, San Diego CA). Flouorescence was detected with a 405 nm laser using a 530/30 bandpass filter for SYBR Green (bacteria) a 695/40 bandpass filter for chlorophyll a. There was no bacterial contamination of axenic cultures based on scattergrams from flow cytometry and aliquots from axenic cultures spread onto YTSS plates.

RNA-seq Analysis
Filters were cut into pieces using sterile technique and incubated at room temperature for 1 h in Quality control was performed on 207 million 50-bp reads using the FASTX toolkit, imposing a minimum quality score of 20 over 80% of read length. Reads aligning to an in-house rRNA database were removed (blastn, score cutoff > 50). Remaining reads were mapped to the R. pomeroyi, Stenotrophomonas sp. SKA14, or P. dokdonensis MED152 genomes (Bowtie 2; (2) and counted (HTSeq; (3) , conserving strand information and removing reads that mapped to more than one location. Genes with differential expression were determined with DESeq2 (4) . Annotations of the bacterial genomes were updated by comparison to closely-related genomes in the IMG (5) and NCBI (6,7) databases, and to a database of TnSeq derived annotations (8) . The dbCAN web resource was used for identification of carbohydrate-active enzyme annotations, taking into account results of HMMs, peptide pattern recognitions, and protein alignments (9) . Reciprocal best hit analysis was carried out against genomes of well-annotated Gammaproteobacteria (Stenotrophomonas maltophila K279a), and Flavobacteriia (Formosa sp. Hel_1_33_133, Bacteriodetes ovatus ATCC8433, Gramella forsetii KT0803, and Zobellia galactanivorans DsijT).

Mass Spectrometry and NMR Analysis
Chemical analysis was conducted on filtered spent media from the co-cultures and axenic T.
pseudonana culture and on uninoculated L1 as the medium blank. For mass spectrometry analysis, 8, 24, and 48 h co-culture spent media were analyzed. Metabolites were derivatized with benzoyl chloride (Widner et al. in prep) by modification of methods from Oehlke et al. (10) and Wong et al. (11), extracted using a solid phase resin (Agilent, Bond Elut PPL), and analyzed using ultra high performance liquid chromatography coupled with electrospray ionization and tandem mass spectrometry (UHPLC-ESI-MSMS) with modifications to Kido Soule et al. (12). Unless otherwise noted, all samples and reagents for mass spectrometry analysis were stored and transferred with acid-washed (10% HCl), combusted (450 o C for 5 hours) glassware, and all solvents were Thermo Fisher Optima Grade. Standards were obtained from Sigma-Aldrich at the highest purity available. Filtered (0.2 µm) samples were divided into two 1 mL aliquots and one was spiked with a standard mix (100 ng/mL final concentration; Table S4) to correct for matrix effects. Each sample was combined with 300 µL sodium carbonate, 30 µL sodium hydroxide, and 200 µL working reagent (5% benzoyl chloride in acetone V:V). After vortexing for 5 min, samples were acidified with concentrated phosphoric acid to pH 2-3 (15 µL). The acetone was evaporated using a vacufuge (Eppendorf), and an equivalent volume of Milli-Q water was added to replace the acetone.
Mass spectrometry samples were analyzed on a Vanquish UHPLC system [Waters Acquity HSS T3 column (2.1 x 100 mm, 1.8 μm) with a Vanguard pre-column at 40 o C] coupled via heated electrospray ionization (H-ESI) to an ultrahigh resolution tribrid mass spectrometer, the Orbitrap Fusion Lumos (Thermo Fisher Scientific). The column was eluted with mobile phase (A) 0.1% formic acid in water and (B) 0.1% formic acid in acetonitrile at a flow rate of 0.5 mL min -1 . The gradient conditions were 0-0.5 min (1% B), 2 min (10% B), 2-5 min (10% B), 7 min (25% B), 7-9 min (25% B), 12.5 min (50% B), 13 min (95% 4 B), 13-14.5 min (95% B). The system was then returned to 1% B for re-equilibration prior to the next injection (total gradient time 16 min). Separate injections of 5 μL each were made for positive and negative ion modes. The electrospray voltages were 3600 V (positive) and 2600 V (negative), and the source gases were 55 (sheath) and 20 (auxiliary), and 1 (sweep). The capillary temperature was 350 o C and the vaporizer temperature was 400 o C. The Lumos was operated in full-scan MS mode with datadependent tandem mass spectrometry, guided by a list of user-defined parent ions with a retention time window for each (full MS/ddMS2 with inclusion list). In this targeted quantitation approach, data are collected in full-scan mode in the Orbitrap analyzer (resolution 60,000 fwhm, at m/z 200). When any of the parent ions on the inclusion list are detected within the retention time window, MS/MS scans are automatically acquired (resolution 7,500 fwhm, at m/z) and then the instrument resumes full-scan operation. Data obtained in full-scan mode were used for detection and quantification, while MS/MS data were used for identification and confirmation. Full-scan MS data were collected between 170-1000 m/z, and the automatic gain control (AGC) setting was 4e5 with a max injection time of 50 msec. MS/MS data were collected using higher energy collisional dissociation (HCD) with collision energy at 35% and intensity threshold at 2e4. The MS/MS AGC target setting was 5e4 with a max injection time of 22 msec. Parent ions were isolated within the quadrupole at an isolation width of 1 m/z. All data were collected in profile mode. See Table S4 for retention times and parent and product ion values.
Standards were prepared in duplicate in L1 medium from 0.5 to 500 ng/mL. Sample and standard peaks were integrated using Skyline (13,14), and standard curves were calculated using at least 5 standards. Standard curve R squared values were required to be greater than 0.92 for metabolite quantification (Table S4). In preliminary experiments (Widner et al. in prep), we determined that the instrument response could be enhanced or suppressed in some spent culture media relative to the standards in L1 medium. The analytical recoveries were calculated as the difference between the calculated concentrations of the spiked sample (sample concentration + 100 ng/mL spike) and the unspiked sample divided by the spike concentration (100 ng/mL). The concentration of each sample was corrected for matrix effects by multiplying the raw concentration by the average analytical recovery for that culture media type. Because of these corrections, we are reporting 'adjusted concentration' for metabolites.
Mass spectrometry metabolites were evaluated statistically in Matlab by comparing adjusted sample concentrations (from all time points) by treatment using a one-way ANOVA (α = 0.05) and post-5 hoc Dunnett's test to compare each co-culture to the axenic T. pseudonana. Outliers were defined as values that exceeded 3 scaled median absolute deviations and were excluded from statistical analysis.
For NMR analysis, 5 ml of 48 h co-culture spent medium were lyophilized, homogenized dry (MP Biomedicals FastPrep-96) through 3 x 30 s cycles at 1800 rpm using 5 x 3.5 mm glass beads, reconstituted in 200 L dimethyl-sulfoxide-d6 (DMSO-d6) with 50 M sodium trimethylsilylpropanesulfonate-d6 (DSS-d6, Cambridge Isotope Laboratories, Inc.) and re-homogenized through 3 x 30 s cycles at 1800 rpm. Following centrifugation for 10 min at 22 o C and 5000 x g, the supernatant was centrifuged for 15 min at 5000 x g and added in 40 L aliquots to 1.7 mm NMR tubes in a 96 tube rack. Two-dimensional HSQC-NMR spectra were collected (Bruker 800 MHz NEO with 1.7 mm cryoprobe) with an automatic and refrigerated (279 K) sample changer (SampleJet) using IconNMR (V5.1). Before NMR data acquisition, samples were preincubated for 5 min in the probe for temperature equilibration at 300 K. 2D data were collected using acquisition parameters modified from a hsqcetgpsisp2.2 pulse program (TopSpin V4.0.6). Spectra were acquired in Echo-Antiecho acquisition mode in 3 h 58 min with presaturation of residual water using 64 scans and 32 dummy scans. The indirect 13 C (f1) dimension had a spectral width of 90.0027 ppm, 128 data points, and an offset of 45 ppm. In the direct 1 H dimension (f2) had a spectral width of 13.0255 ppm, 4166 data points, and an offset of 3.691 ppm.
Spectra were processed in MNOVA by applying a 90 o sine square apodization, 4K zero-filling, and Fourier-transformation. Transformed spectra were auto-phased, baseline-corrected and referenced along f1 and f2 to DSS-d6 (0.0, 0.0 ppm). All peaks above the noise were manually integrated. These peaks were divided into four sub-regions for visualization ( Figure S2) and analysis using MATLAB. Raw data, peak lists, and analysis scripts are available on MetaboLights under accession MTBLS1544.
The peak integral values were imported into a MATLAB workflow to normalize, scale, and analyze spectral features. The peak integral table was normalized using built-in probabilistic quotient normalization (PQN), mean-centering, and univariate scaling functions. The built-in MATLAB principal components analysis (PCA) function was applied to feature integrals across co-culture conditions, and a loadings plot along principal component 1 was generated to determine which peaks accounted for the largest proportion of variance among samples. The loadings plot informed our decision for which peaks to analyze further, and all peaks between the covariance threshold of -0.5 and 0.5 were omitted from 6 further analysis as they accounted for the least between-sample variation. From these data, P-values, false discovery rates (FDR), and q-values were calculated using MATLAB built-in functions.   Table S1. RNA-seq data for R. pomeroyi DSS-3. Log2 fold-differences that are negative are enriched in the co-culture, differences that are positive are enriched in the glucose control. Table S2. RNA-seq data for Stenotrophomonas sp. SKA14. Log2 fold-differences that are negative are enriched in the co-culture, differences that are positive are enriched in the glucose control. Table S3. RNA-seq data for P. dokdonensis MED152. Log2 fold-differences that are negative are enriched in the co-culture, differences that are positive are enriched in the glucose control. Table S4. Targeted metabolites that could be quantified by mass spectrometry in culture media with a calibration curve R 2 > 0.92. Each parent ion was fragmented into one or more product ions (Product 1, Product 2). N/A = absence of a product ion; # bz = number of benzoyl groups added to the target compound in the derivatization process; # Standards = number of standards in the calibration curve.   Table S4. Targeted metabolites that could be quantified by mass spectrometry in culture media with a calibration curve R2 > 0.92. Each parent ion was fragmented into one or more product ions (Product 1, Product 2), where "N/A" indicates the absence of a product ion. "# bz" is the number of benzoyl groups added to the target compound in the derivatization process, and "Number of Standards" is the number of standards in the calibration curve.