Ecological interactions are crucial to the structure and function of biological communities, but we lack a causal understanding of the forces shaping their emergence during evolutionary diversification. Here we provide a conceptual framework linking different modes of diversification (e.g. ecological diversification), which depend on environmental characteristics, to the evolution of particular interactions (e.g. resource partitioning) in asexual lineages. We tested the framework by examining the net interactions in communities of Pseudomonas aeruginosa produced via experimental evolution in nutritionally simple (SIM) or complex (COM) environments by contrasting the productivity and competitive fitness of whole evolved communities relative to their component isolates. As expected, we found that nutritional complexity drove the evolution of communities with net positive interactions whereas SIM communities had similar performance as their component isolates. A follow-up experiment revealed that high fitness in two COM communities was driven by rare variants (frequency <0.1%) that antagonized PA14, the ancestral strain and common competitor used in fitness assays. Our study suggests that the evolution of de novo ecological interactions in asexual lineages is predictable at a broad scale from environmental conditions. Further, our work demonstrates that rare variants can disproportionately impact the function of relatively simple microbial communities.

Evolution experiment and generation of isolate library and isolate mixes

The evolution experiment that produced the evolved communities used in this study is described in detail in Schick and Kassen (2018). Briefly, 60 replicate lines of P. aeruginosa strain PA14 were independently propagated in either COM or SIM environments. COM environments consisted of synthetic cystic fibrosis media (SCFM), which is a defined media composed of the essential chemical elements of sputum from cystic fibrosis patients (Palmer et al. 2007). The SIM environment consisted of M9 media (7.78 g/L Na₂HPO₄, 3 g/L KH₂PO₄, 0.5 g/L NaCl, 1 g/L NH₄Cl, 493 mg/L MgSO₄, 14.7 mg/L CaCl₂) supplemented with 4 g/L of glucose as the sole carbon source. Half of the founding strains in each environment contained a chromosomally inserted genetic marker, lacZ, such that they could be distinguished from unmarked strains on agar plates supplemented with x-galactosidase (X-Gal). Populations were propagated in 1.5 mL batch cultures at 37°C with daily serial transfers of 1% of the previous day’s culture into fresh media in unshaken 24-well cell culture plates (Costar® 24-well Clear TC-treated Multiple Well Plates), resulting in ~6.6 generations occurring each day. Microcosms were left unshaken to promote spatial heterogeneity, which may in turn promote diversification. Endpoint communities from each line were stored in 20% glycerol at –80°C after 32 days of experimental evolution (~220 generations). Unless otherwise stated, bacterial cultures in this study were grown in unshaken 1.5 mL microcosms at 37°C in 24-well culture plates.

We generated working stocks of 8 randomly selected lines from each environment by reviving the endpoint lines from Schick and Kassen (2018) in Lysogeny Broth (LB) microcosms and allowing them to grow for 24 hrs before storing them in in glycerol at –80°C. We generated a library of isolates from each line by first plating a sample from the frozen working stock onto an agar plate (SIM media + 1.3% agar (w/v); Oxoid Bacteriological Agar). We then randomly selected 22 colonies from each plate that we grew individually in LB before storing the resulting cultures separately in glycerol at –80°C. We also generated mixes of all 22 isolates from each of the 8 examined COM lines (henceforth referred to as “isolate mixes”) by growing all 22 isolates from each line in individual LB microcosms for 24 hrs and mixing the resultant cultures in equal proportions before storing the mixes in glycerol at –80°C.

Cellular productivity assay

We revived evolved communities and isolates by transferring a small volume of frozen stock from each of the stored cultures into either LB or SIM microcosms. We used LB instead of SCFM to represent the complex environment for this assay and the Competitive fitness assay (below) for its ease of preparation. Like SCFM, LB is a complex medium with many carbon and nitrogen sources. After 24 hrs of growth, we transferred 1% of the mixture into a fresh microcosm containing the same media for another 24-hr growth period. We then diluted the cultures in SIM media by a factor of 10⁶ and plated 100 μL from each on SIM agar plates to ensure they would produce a countable number of colonies. After 48 hrs of growth at 37°C, we counted the number of colonies on each plate, which we used as our measure of cellular productivity. We measured the productivity of all 22 isolates from each line in each environment without replication. Isolates were measured without replication because we chose to maximize the diversity of isolate measured instead of making repeated measurements on individual isolates. We measured the productivity of each community in each environment with 8-fold replication.

Three productivity measurements in LB were lost due to contamination or dilution errors. Two were isolates from different COM lines and one was an isolate from a SIM line.

Competitive fitness assay

As above, we revived evolved communities and isolates as well as the ancestral strain from frozen stocks in fresh LB or SIM microcosms and allowed them to grow for 24 hrs. We then generated co-cultures by mixing an equal volume of each evolved strain with the oppositely marked ancestral strain, which acted as a common competitor in this experiment (lacZ marked ancestral strain for unmarked evolved strains and vice-versa), adding 50 μL of the evolved and ancestral bacterial into 900 μL of the same media in which they were revived. Next, we diluted these mixtures into additional 1 mL microcosms containing the same media at a 5-fold dilution factor and allowed these co-cultures to grow for a 24-hr period, resulting in ~5.6 generations of growth in the mix. Before and after the co-culture period, we diluted and plated a 100 μL sample of each mixed culture on SIM agar plates with X-Gal. We recorded the number of marked and unmarked colonies at both time points (lacZ-marked colonies appear blue on agar plates containing X-Gal), and used these counts to calculate fitness using the following formula:

 s = (ln(n1f/n1i)-ln(n2f/n2i))/No. Generations         (1)

Where s represents the selection coefficient, n1_f and n1_i represent the number of colonies of the focal bacteria at the final and initial timepoint, respectively, and n2_f and n2_i represent the number of colonies of the competitor bacteria at the final and initial timepoint, respectively. The selection coefficient reflects the natural logarithm of the geometric growth rate of the ratio between focal bacteria and the competitor per generation. If the selection coefficient is 1, this reflects an e-fold increase in the ratio between focal and competitor bacteria each generation.

For certain COM community replicates, zero colonies from the common competitor were present following co-culture (out of hundreds of colonies), meaning we could not estimate competitive fitness using the above formula and had to remove those replicates from our analyses. This may have led us to underestimate the competitive fitness of these communities, as the fitness of certain replicates exceeded our limit of detection. This occurred three times in each environment across three different COM lines (6.1, 6.2, and 6.15). 39 total replicates were also lost due to contamination or dilution errors: 38 in the COM environment (16 from COM isolates, 9 from COM communities, 9 from MIN isolates, and 4 from MIN communities) and one in the SIM environment (one COM community).

We measured the fitness of 22 isolates in LB and 11 isolates in SIM from each of the 16 lines without replication. We measured the fitness of communities with 8-fold replication in LB and 4-fold replication in SIM. We used fewer replicates in SIM because this environment tended to yield less variable fitness values.

To investigate whether the diversity captured in our 22 isolate samples were representative of the diversity contained in six focal COM lines (6.1, 6.2, 6.8, 6.12, 6.15, 6.17), we also compared the fitness of communities and isolate mixes from each line in LB with 4-fold replication. One COM community replicate was lost due to exclusion of the common competitor (line 6.15). Three replicates were lost due to contamination or dilution errors, two from on COM community (line 6.1) and one from a COM isolate (6.15).

Dilution-diversity experiment

We designed this method based on similar experiments with natural microbial communities (Wertz et al. 2006, 2007; van Elsas et al. 2012) to probe the relationship between diversity and fitness in two COM communities (6.1 and 6.2). After reviving these communities in LB microcosms as above, we ran separate 10-fold dilution series on these cultures using LB and generating a dilution gradient that varied in density from ~3×10⁹ to ~3×10² CFU/mL. We then took 150 μL samples from each of the final six wells of each dilution series, which varied in density from ~3×10⁷ to ~3×10² CFU/mL. As such, the densest sample contained ~4.5x10⁶ cells and the least dense sample contained ~45 cells. These density estimates are based on the cellular density of COM communities following 24 hrs of growth in LB (see Supplementary Figure 1A). We then allowed the community samples to grow for 24 hrs in LB before storing them in glycerol at –80°C. Although the density of community samples varied across the dilution gradient, we expect that this growth period allowed them to reach similar densities before storage.

For each of the two focal lines, we produced community samples varying in diversity from three replicate dilution series initiated independently from frozen stocks. Each replicate dilution series produced six community samples varying over a wide range of diversity, since each was produced by a progressively smaller bottleneck of the original community. We measured the competitive fitness of each community sample in LB as described above (see Competitive fitness assay) in duplicate, producing six total fitness measurements for each community at each dilution level. This allowed us to determine at what bottleneck size the competitive fitness of community samples changed from community-like (which is to say relatively high) to isolate mix-like (which is to say relatively low).

Five replicates were lost from line 6.1 because the common competitor was completely excluded after the co-culture period. One replicate was lost from line 6.2 due to a dilution error.

Sterile spent media assay

We performed a sterile spent media assay to investigate whether RVs conferred high fitness in two focal COM lines (6.1 and 6.2) by secreting material into the media that facilitated other evolved genotypes and/or inhibited the ancestral strain. This assay was performed on community samples of both focal lines from the Dilution-diversity experiment. Specifically, we focused on two community samples from each line that were taken from adjacent dilution levels in which the less dilute sample contained the RV (RV⁺) and the more dilute sample was RV^-, as inferred by their fitness values in the Dilution-diversity experiment. We grew the RV⁺ and RV^- samples from frozen stocks in LB for 6 hrs before centrifuging them in Eppendorf tubes at 4000 rpm for 10 min. A 6-hr growth period was used to ensure the community samples would be in the exponential phase of growth when their spent media was collected. Following centrifugation, we transferred supernatant into fresh Eppendorf tubes using sterile 3 mL syringes (BD) and 0.22 μm disposable syringe filters (Millex-GP), removing any remaining cells.

As we prepared the sterile spent media, we grew two evolved isolates from each focal line as well as PA14 for 24 hrs in LB from frozen stocks. We then adjusted the optical density (OD₆₀₀) of these cultures to 0.1 by diluting them in LB at dilution factors proportional to their density readings following the 24-hr growth period. Next, we added the density-adjusted cultures into one of three growth treatments with 5-fold replication: RV⁺ supernatant, RV^- supernatant, or a control treatment with no supernatant (2 focal lines × 3 growth treatments × 3 focal lines × 5 replicates = 90 total growth curve replicates). The RV⁺ and RV^- supernatant treatments consisted of 120 μL of 1.5× LB, 60 μL supernatant (either RV+ or RV-), and 20 μL of the density-adjusted bacterial cultures. The no supernatant control consisted of 180 μL of 1× LB and 20 μL of the density-adjusted bacterial cultures. 1.5× LB was used in the former two treatments to ensure an equal density of nutrients was supplied from the growth media across all treatments. Once these ingredients were mixed in separate wells of a 96-well plate (Corning 96-well Polypropylene Microplate), the plate was placed in a spectrophotometer (BioTek PowerWave XS2 Microplate Spectrophotometer) which measured the optical density (OD₆₀₀) of each culture every 10 min over a 12-hr period. The plate reader kept the cultures at 37°C and shook the plate prior to each density reading to minimize the development of biofilms that may have interfered with density readings.

Four total growth curve replicates were lost due to contamination, one from line 6.2 isolate 1, two from 6.2 isolate 2, and one from PA14, all grown in RV⁺ spent media from line 6.2.

Microsoft Excel and R are sufficient. Excel files may be converted to CSVs, which can be read by many open source softwares.