Dataset: NMR-based polar metabolomics data

Spectra were imported into MATLAB, version 7.12.0, and preprocessing was performed in NMRLab (6) and ProMetab 3.3 (7). Spectra were aligned to the chemical shift of the internal standard (TMSP) at 0.00 ppm, manually phased, and baseline corrected. Spectral regions around TMSP (−0.5 to 0.5 ppm), water (4.6–5.0 ppm for A. glacialis; 4.0–4.9 ppm for T. pseudonana), and residual methanol (3.3–3.5 ppm) were removed before noise filtering. For the A. glacialis experiment only in which a small number of K. brevis cells escaped from the dialysis tubing, we removed signals caused by K. brevis cell contamination (identified by their presence in K. brevis blank culture extracts), to not confound intracellular K. brevis and T. pseudonana metabolites. These regions included the following: 1.41–1.42, 2.65– 2.67, 2.93–2.95, 2.98–3.02, and 4.15–4.40 ppm. Spectra were then binned (0.005 ppm), probabilistic quotient normalized (8) to account for slight differential dilution among samples, generalized log (glog) transformed to account for variance among signals within each sample reducing bias toward highly concentrated metabolites, and mean centered. The lambda values for glog optimization were obtained using a set of five quality control extracts generated with the above methods from a single large batch culture of each diatom species (9). The λ value for T. pseudonana was 9.8068 × 10−8, whereas a λ value of 5.7289 × 10−9 was used for A. glacialis samples.