All individual chemicals were purchased from Sigma Aldrich (St. Louis, MO, United States) except for sn-glycero-3-Phosphocholine (Cayman Chemical, Ann Arbor, MI, United States). Wolfe’s Vitamin supplement (MD-VS™) and Wolfe’s Trace Mineral supplement (MD-TMS™) were purchased from ATCC (Manassas, VA, United States). NLDM was prepared by adding the 64 metabolites to a base medium composed of 1x Wolfe’s mineral and 1x Wolfe’s vitamin solutions, potassium phosphate, sodium phosphate, calcium chloride, magnesium sulfate, and ammonium chloride (Supplementary Table 1). The pH of NLDM was adjusted to 7.0–7.2 and was sterilized using a 0.22 μm PES filter. A recipe for preparation of NLDM can be found in Supplementary Table 2.

NLDM was compared to R2A medium at 1x concentration (Tecknova, Hollister CA) for their ability to support the growth of a broad range of ORFRC isolates, each in duplicate or triplicate. For growth analysis, the 110 phylogenetically diverse isolates (Supplementary Table 3) were revived in 5 mL liquid R2A medium from frozen glycerol stocks or from single colonies from frozen glycerol stocks streaked onto R2A plates. Aliquots from overnight cultures were washed twice with PBS (Sigma Aldrich) by centrifugation at 7,000 × g and diluted with the test medium, either R2A or NLDM, to an OD₆₀₀ of 0.1 prior to inoculation into 96-well plates (Corning CLS3370, Sigma Aldrich). For each fresh medium, 36 μL of washed starter culture (or 36 μL of sterile medium for uninoculated media controls) was added to 144 μL test medium, either R2A or NLDM, (for a final OD₆₀₀ of 0.02) and plates were incubated under aerobic conditions for 24–48 h at 27°C, and shaken at 250 rpm. Growth data were collected by measuring OD₆₀₀ on microplate readers at 15 min intervals. Growth was defined as an increase of 0.05 or greater from the first time point (max OD₆₀₀–initial OD₆₀₀) after subtraction of the uninoculated media control. The significance between growth in NLDM vs. R2A for all isolates was analyzed with Python pingouin 0.3.11 pairwise T-tests corrected p-values including the interaction between isolate and medium and taxonomic class and medium in a separate analysis using Benjamini/Hochberg false discovery rate correction. A phylogenetic tree was constructed from isolate 16S rRNA gene sequences. The16S rRNA gene of the isolates were PCR amplified using primers 27F and 1492R and Sanger sequencing was performed directly on the PCR products. The 16S rRNA gene sequences were manually end-trimmed to remove ambiguities. The shortest trimmed sequence from each phylogentic family was used to recruit guide sequences from the SILVA LTP and RDP databases. A total of 98 guide sequences were added to the global alignment along with 13 “References” sequences composed of 16S rRNA gene sequences of well-studied and characterized isolates to help orient the reader (marked as “REF” in Figure 1; see Supplementary Information for complete list of FASTA sequences). The sequence alignment was performed in MAFFT (version 7) using the L-INS-i iterative refinement method (Katoh and Standley, 2013). The tree was prepared using the FastTree 2 software package with the Generalized Time-Reversible (GTR) model and optimized Gamma20 likelihood with the “pseudocounts” flag for uneven sequence length (Price et al., 2010; Supplementary Figure 1). For clarity, the resulting tree was pruned of guide sequences within TreeGraph 2 prior to visualization which was imported and displayed using the ITOL software (Letunic and Bork, 2021).

Triplicate 1.2 mL cultures (including medium controls) of 30 different isolates were cultured in NLDM at 30°C using 24 well plates (same inoculation technique described in the previous section). A single time point was collected from individual wells, each at 24 h. A culture fraction of 1 mL was centrifuged at 7,000 × g for 5 min at 4°C and 0.5 mL of the supernatant was collected. The supernatants were then frozen at –80°C, lyophilized to dryness and resuspended in 250 μL methanol containing internal standards (Supplementary Table 4). The resuspended samples were filtered through 0.2 μm modified nylon membrane centrifugal filters (Pall Corporation, Port Washington, NY, United States) for 2 min at 5,000 × g and analyzed as described in the Liquid Chromatography Tandem Mass Spectrometry (LC-MS/MS). section.

Metabolites in the NLDM were chromatographically separated using hydrophilic interaction liquid chromatography (HILIC) and detected by high resolution tandem mass spectrometry. Analyses were performed using an InfinityLab Poroshell 120 HILIC-Z column (Agilent, Santa Clare, CA, United States) on an Agilent 1,290 stack connected to a Q-Exactive Hybrid Quadrupole-Orbitrap Mass Spectrometer (Thermo Fisher Scientific, Waltham, MA, United States) equipped with a Heated Electrospray Ionization (HESI-II) source probe. Separation, ionization, fragmentation and data acquisition parameters are specified in Supplementary Table 4. Briefly, metabolites were separated by gradient elution followed by MS1 and data dependent (top 2 most abundant MS1 ions not previously fragmented in last 7 s) MS2 collection; targeted data analysis was performed by comparison of sample peaks to a library of analytical standards analyzed under the same conditions. Three parameters were compared: matching m/z, retention time and fragmentation spectra using Metabolite Atlas¹ (Bowen and Northen, 2010; Yao et al., 2015). Metabolite background signals detected in the extraction blanks were subtracted from the experimental sample peak heights/areas. Metabolite peak heights and peak areas were normalized by setting the maximum peak height or peak area detected in NLDM uninoculated control samples to 100%. Normalized metabolite feature peak heights or peak areas were used for relative abundance comparisons. Hierarchical clustering analysis with a Bray–Curtis dissimilarity matrix was performed with the Python 3.7.4 Seaborn 0.9.0 package. The significance between control medium and isolate metabolic profiles was analyzed with the Python Scipy version 1.6.3 Ranksums test.

Spearman correlation coefficients between metabolite fold-change patterns were calculated with Pengouin version 0.3.11 pairwise correlations and Bonferroni correction to the p-values. Exometabolomic profile distances used are the Euclidean pairwise distances of the isolate metabolic profiles relative to uninoculated media (log 2 normalized). Corresponding phylogenetic distances were generated by first aligning trimmed 16S rRNA gene sequences from the isolates in the MAFFT software package (version 7) using the L-INS-i iterative refinement method and followed by the generation of a distance matrix using the FastTree 2 software package using the Generalized Time-Reversible (GTR) model and optimized Gamma20 likelihood (Price et al., 2010; Katoh and Standley, 2013). The distance matrix generated for the phylogenetic tree (e.g., distance between isolates on the phylogentic tree) were used on the x-axis. Raw data and targeted outputs have been deposited in the JGI Genome Portal under FD ID 1354404 (Nordberg et al., 2014).

Feature-Based Molecular Networking (FBMN) was performed using MZmine 2 and GNPS. A molecular network was created with the Feature-Based Molecular Networking (FBMN) workflow on GNPS² (Wang et al., 2016; Nothias et al., 2020). The mass spectrometry data were first processed with MZMINE2 (see Supplementary Xml File for detailed parameters) and the results were exported to GNPS for FBMN analysis. The data was filtered by removing all MS/MS fragment ions within ± 17 Da of the precursor m/z. MS/MS spectra were window filtered by choosing only the top 6 fragment ions in the ± 50 Da window throughout the spectrum. The precursor ion mass tolerance was set to 0.01 Da and the MS/MS fragment ion tolerance to 0.02 Da. A molecular network was then created where edges were filtered to have a cosine score above 0.70 and more than 3 matched peaks. Further, edges between two nodes were kept in the network if and only if each of the nodes appeared in each others respective top 10 most similar nodes. Finally, the maximum size of a molecular family was set to 0, and the lowest scoring edges were removed from molecular families until the molecular family size was below this threshold. The spectra in the network were then searched against GNPS spectral libraries (Horai et al., 2010; Wang et al., 2016). The library spectra were filtered in the same manner as the input data. All matches kept between network spectra and library spectra were required to have a score above 0.4 and at least 3 matched peaks. The molecular networks were visualized using Cytoscape software (Shannon et al., 2003). Features observed in 3 or less (3 out 18 = 16.7%) NLDM uninoculated control samples were colored green to yellow in the networks based on the ratio (0.01–1) of the number of isolate spent medium samples the feature was detected in divided by the total number of isolate spent medium samples. Only features with 10× higher peak heights than in extraction blanks were selected for analysis.

Except for glucose and pyruvic acid, R2A is a complex and undefined metabolite mixture, primarily based on yeast extract, casamino acids, soluble starch, and proteose peptone (Reasoner and Geldreich, 1985). To identify the small molecules from these complex components, we previously analyzed R2A medium using LC-MS/MS (Supplementary Table 5; Jenkins et al., 2017; Kosina et al., 2018). Through these efforts, a number of metabolites present in R2A were identified. These included most of the standard amino acids, all 5 standard nucleobases, and 3 standard ribonucleosides. Based on these findings, we included all 20 standard amino acids, the 5 standard nucleobases, and the 5 standard ribonucleosides in NLDM. In addition, the amino acid citrulline, the nucleobases and ribonucleosides xanthine, hypoxanthine, inosine and xanthosine were included in NLDM based on their presence in R2A.

Primary energy sources in R2A are the sugar glucose and the glucose polymer starch; the latter is too large for small molecule LC-MS/MS detection. As a result, we included glucose plus two additional sugars detected in R2A, the dihexose trehalose and the sugar alcohol myo-inositol. In addition, to increase substrate diversity and assess additional metabolic pathways, we also included the pentose xylose, the dihexose sucrose, and the amino sugars N-acetyl-glucosamine and n-acetylmuramic acid, which are commonly found in soils (Glaser et al., 2004; Gunina and Kuzyakov, 2015; Sasse et al., 2019; Ni et al., 2020). Pyruvic acid is another major defined energy source in R2A medium. To capture organic acids as potential energy sources for bacteria pyruvic acid was included along with 5 other common organic acids detected in R2A medium and/or in soil, namely lactic acid, malic acid, citric acid, succinic acid and α-ketoglutaric acid.

Other metabolites were selected for NLDM based on the analysis of R2A medium and/or soil. In total 64 metabolites were selected to be included in NLDM. We decided to include spermidine even though it was not detected in R2A or soil because polyamines are “essential” cofactors and because another polyamine, A-acetylputrescine, has been detected in R2A medium (Xavier et al., 2017).

After the formulation of NLDM, we checked if the selected metabolites in the environment/medium would be reduced after inoculation with the microbes. To do this, we analyzed the decreased abundance of the selected metabolites by microbes using existing exometabolomic data collected in Web of Microbes (Kosina et al., 2018). A metabolite was deemed decreased if the metabolite was significantly lower in the presence of a microbe compared to the uninoculated control. Out of the 64 metabolites present in NLDM, 54 were in the Web of Microbes database. All but 4 metabolites, α-ketoglutaric acid, cysteine, cytidine and uridine, were decreased by at least 1 microbe (Supplementary Table 6).

The quantitative formulation of the 64 selected metabolites was based on the amount of organic carbon (C) and nitrogen (N) in R2A (Kim et al., 2019). We divided all 64 metabolites in NLDM into 4 different groups: sugars, organic acids, amino acids, and other metabolites (Supplementary Table 1). Metabolites within each group were assigned fixed equimolar concentrations and the total organic C and N was calculated (Table 1). This yielded a C:N ratio of 9:1, which is similar to R2A (8:1) and the soil microbial biomass ratio (9:1) [based on the atomic C:N:P ratios in the soil microbial biomass (60:7:1)] (Cleveland and Liptzin, 2007). As for the salts, NLDM contains 5 mM phosphate, 1 mM ammonium, 2 mM sodium, 7 mM potassium, 1 mM magnesium, 1 mM sulfur, 1 mM calcium, and 2 mM chloride (Supplementary Table 1). NLDM is supplemented with 1x Wolfe’s vitamins and 1x Wolfe’s minerals (Atlas, 2004).

We selected 110 bacterial isolates from the same ORFRC field site for this study and compared the growth of the isolates in both NLDM and R2A. Notably, most isolations (35%) were made using R2A (Supplementary Table 3; Thorgersen et al., 2015; Hemme et al., 2016; Liu et al., 2018; Price et al., 2018; Carlson et al., 2019; Vuono et al., 2019; Lui et al., 2021). It was found that only 2 out of the 110 isolates did not display significant growth in NLDM, whereas 6 out of the 110 isolates did not display significant growth in R2A medium (Figure 1 and Supplementary Tables 3, 7A–C). The overall growth (as measured as the highest observed OD₆₀₀ minus starting OD₆₀₀) was similar for R2A and NLDM, except for beta- and gamma proteobacteria which displayed significantly better growth in NLDM than on R2A (Supplementary Figure 2). The lag phases and growth rates were similar for R2A and NLDM, except for betaproteobacteria which displayed significantly shorter lag phase in NLDM than on R2A (Supplementary Figures 3, 4 and Supplementary Table 8).

We next used the NLDM to investigate the substrate preferences of 30 isolates using LC-MS/MS. Using hydrophilic interaction liquid chromatography (HILIC) LC-MS/MS, all metabolites were detected in the full medium formulation (Supplementary Table 9). All 64 metabolites from the NLDM medium were utilized by at least 1 isolate after 24 h (compared to medium control using ranksum, P < 0.05). Pseudomonas FW300-N2F2 decreased the abundance of the most metabolites (63 significantly decreased in abundance relative to media control) whereas Cupravidius GW822-FHT05A05 resulted in the fewest metabolites decreased in abundance (34) (Figure 2A). Some NLDM metabolites were excreted by the isolates, as their abundance significantly increased. Feature-based molecular networking indicated that the isolates excreted metabolites not present in NLDM (Supplementary Figure 5). Interestingly, we observed clear differences in metabolite depletion between isolates and between metabolite classes (Figure 2A). Specifically, hierarchical clustering revealed that metabolite utilization is conserved with phylogeny. Correlation analysis of metabolite utilization revealed that isolates belonging to the same family and genus have significantly similar metabolite usage profiles (P < 0.05) (Figure 2B).