Environmental structure impacts microbial composition and secondary metabolism

Determining the drivers of microbial community assembly is a central theme of microbial ecology, and chemical ecologists seek to characterize how secondary metabolites mediate these assembly patterns. Environmental structure affects how communities assemble and what metabolic pathways aid in that assembly. Here, we bridged these two perspectives by addressing the chemical drivers of community assembly within a spatially structured landscape with varying oxygen availability. We hypothesized that structured environments would favor higher microbial diversity and metabolite diversity. We anticipated that the production of a compound would be more advantageous in a structured environment (less mixing) compared to an unstructured environment (more mixing), where the molecule would have a diminished local effect. We observed this to be partially true in our experiments: structured environments had similar microbial diversity compared to unstructured environments but differed significantly in the metabolites produced. We also found that structured environments selected for communities with higher evenness, rather than communities with higher richness. This supports the idea that when characterizing the drivers of community assembly, it matters less about who is there and more about what they are doing. Overall, these data contribute to a growing effort to approach microbial community assembly with interdisciplinary tools and perspectives.

Zymo Quick DNA manufacturer protocol), transferred to Zymo bashing bead tubes (Zymo Research Corp., Irvine, CA, USA), homogenized for 45 seconds using a hand-held reciprocating saw with custom attachment (RYOBI, Anderson, SC, USA), and stored at -20 °C until DNA extraction. Lastly, 103 mL of ethyl acetate was added to the remaining volume of the pooled samples (~103 mL) and mixed to generate a 1:1 ratio of culture and solvent for an overnight, organic extraction at room temperature. Medium controls for each time point, described above, were pooled, sampled for DNA extraction, and prepared for ethyl acetate organic extraction as described above for the cultures to be used as blanks for both sequencing and metabolomic data.

HPLC-ESI Mass Spectrometry
A 5 μL volume at 11 mg/mL was used for injection on an Agilent 1290 HPLC system (Agilent, Santa Clara, CA, USA) with a Waters ACQUITY UPLC BEH column (ODS-18; 2.1 x 100 mm; 1.7 μm particle size, Waters, Milford, MA, USA). Mobile phase A was comprised of 0.1% formic acid in Optima LC/MS water (FisherScientific, Waltham, MA, USA), while mobile phase B was comprised of 0.1% formic acid in 100% HPLC-grade acetonitrile. A binary gradient at 0.4 mL/min flow rate was used under the following steps: 90% solvent A and 10% solvent B from 0 to 1 min, linear gradient to 20% solvent B from 1-2 min, linear gradient to 80% solvent B from 2-16 min, linear gradient to 100% solvent B from 16-18 min, and 100% solvent B from 20 -21 min, which was maintained for an additional 2 min before the next sample injection. Eluent from the column was run on an Agilent 6545 accurate mass Q-TOF mass spectrometer with an electrospray ionization source operating in the positive mode (Agilent, Santa Clara, CA, USA).
Nitrogen was used as a nebulizing gas (40 lbs/in 2 ) and as a drying gas (325 °C; 10 L/min flow rate) (Bartley et al., 2013). Fragmentor voltage was 180 V, skimmer voltage was 45 V, and capillary voltage was 4,000 V. Untargeted MS/MS was used with mass-dependent collision energy ramp from 20-35 V using ultra high purity nitrogen. Data was collected with Mass Hunter Acquisition software (B.08.00).

Molecular Networking and Spectral Library Search in GNPS
Raw data obtained from the LCMS/MS instrument was converted to .mzXML files using msConvert within ProteoWizard v3 (Chambers et al., 2012). The mass spectrometry data were first processed with MZmine2 v2.51 (Pluskal et al., 2010 (Nothias et al., 2020).
In GNPS the data were analyzed using FBMN workflow (Nothias et al., 2020). The data were filtered by removing all MS/MS fragment ions within +/-17 Da of the precursor m/z. MS/MS spectra were filtered by choosing the top 6 fragment ions in the +/-50 Da window. The precursor ion mass tolerance was set to 0.02 Da and the MS/MS fragment ion tolerance to 0.02 Da. A molecular network was created with edges (filtered via cosine score > 0.7 and > 4 matched peaks). Also, edges between nodes were kept only if each of the nodes appeared in each other's respective top 10 most similar nodes. Finally, the maximum size of a molecular family was set to 100. The analogue search mode searched against MS/MS spectra with a maximum difference of 100 Da in the precursor ion value. The library spectra were filtered as the input data was filtered, as discussed above, cosine score above 0.7 and at least 4 matched peaks.

Microbial and Metabolite Richness Correlations
To investigate the relationship between microbial and metabolite alpha diversity measures, we performed a Spearman's rank-based correlation comparing the metabolite richness to each alpha diversity measure using the r packages Hmisc and corrplot (Frank E Harrell Jr, 2020;Wei & Simko, 2017)

Supplemental Results
Day 0 beta diversity Differences in community structure were primarily driven by inoculum size (PERMANOVA F= 18.97, r 2 = 0.18, p=0.001; PERMDISP F=0.71, p=0.599). Both cultivation condition and day significantly contributed to community structure. However, within-group variance was high and could affect the significance contribution to community structure. Day 0 samples were tightly clustered, indicating high similarity between Day 0 samples, which was expected (Supplemental Figure 2). Because Day 0 samples represented the starting community that had not undergone any selection for which we were testing and would thus skew the distance matrix, they were removed for subsequent analyses. Like the sequence data, Day 0 metabolites clustered tightly and away from the rest of the data set. Because these metabolites represent what was in the inoculum before any selection would take place based on our hypothesis, these samples were removed from subsequent analysis (Supplemental Figure 3).

Microbial and Metabolite Richness Correlations
We initially hypothesized that structured environments would have higher microbial richness as a function of environmental gradients creating diversified niche-availability and would also have higher richness of metabolites as a function of higher richness of community members. To investigate this relationship, we performed a Spearman's correlation on the richness and evenness of the microbial community and the metabolomes. We observed that metabolite richness had no significant correlations with microbial alpha diversity measures (breakaway/microbial richness: r 2 = -0.02, p= 0.88; microbial evenness: r 2 = 0.23, p= 0.16, metabolite evenness : r 2 = -0.04, p = 0.83) (Supplemental Figure 4). This suggests that it may not be the number of organisms in a community that matter for metabolite diversity, but rather who they are and the strategies they employ during community assembly, like pseudomonads and the fleet of metabolites they produce, compared to Aeromonas sp., that quickly outgrew other populations.