Code + Data for COG Identification

To identify clusters of orthologous genes (COGs) that correlate with nutrient limitation in the modern ocean, we examined the Ocean Microbial Reference Catalog v2 (OM-RGC.v2) from the Tara Oceans Project. The OM-RGC.v2 includes relative gene abundances of all COGs (n = 4,787) in 139 Tara Oceans metagenomic samples, along with metadata information including phosphate, oxygen, and nitrate/nitrite concentrations. (Nitrate/nitrite values were reported together for OM-RGC v2.) Iron concentrations for Tara Oceans samples were not available and were thus estimated using the PISCES2 model based on iron concentration model predictions for Tara Oceans sampling locations as described in Table S1 of Caputi et al., 2019. Iron concentrations were predicted for surface and the deep chlorophyll maximum (DCM) only; iron concentrations for samples from the mesopelagic zone were not available under the PISCES2 model. All other metadata for Tara Oceans samples were directly obtained from Salazar et al., 2019.Estimation of correlations between COGs and metadata information was performed using regression models. Compound poisson linear models were fitted in bulk using the MaAsLin2 software package (v. 1.18.0). Separate models were fit for each COG to analyze the effect of metadata variables on individual COG abundances. While the main focus was to investigate correlation with nutrient abundance, environmental metadata was included in the model to control for as many potential confounding effects as the data allowed. The following predictors were included in the final model (based on variables available from the Tara Oceans dataset): the size fraction at which the sample was taken, mean temperature, depth, salinity, mean oxygen concentration, PO4 concentration, NO2 + NO3 concentration, iron concentration, and absolute latitude. Of these, the following predictors were log-transformed to allow greater model fit: depth, PO4 concentration, NO2 + NO3 concentration. To the same end, the iron concentration was transformed by taking the square root, and the absolute value of the latitude was taken. Otherwise, no transformations or normalization was performed. No abundance cutoff was applied, but COGs present in less than one-third of the Tara Oceans samples were discarded in order to ensure that the COGs identified by the statistical model were meaningful.