Rediversification following ecotype isolation reveals hidden adaptive potential

SUMMARY

Microbial communities play a critical role in ecological processes, and their diversity is key to their functioning. However, little is known about if communities can regenerate ecological diversity following ecotype removal or extinction, and how the rediversified communities would compare to the original ones. Here we show that simple two-ecotype communities from the E. coli Long Term Evolution Experiment (LTEE) consistently rediversified into two ecotypes following the isolation of one of the ecotypes, coexisting via negative frequency-dependent selection. Communities separated by more than 30,000 generations of evolutionary time rediversify in similar ways. The rediversified ecotype appears to share a number of growth traits with the ecotype it replaces. However, the rediversified community is also different compared to the original community in ways relevant to the mechanism of ecotype coexistence, for example in stationary phase response and survival. We found substantial variation in the transcriptional states between the two original ecotypes, whereas the differences within the rediversified community were comparatively smaller, but with unique patterns of differential expression. Our results suggest that evolution may leave room for alternative diversification processes even in a maximally reduced community of only two strains. We hypothesize that the presence of alternative evolutionary pathways may be even more pronounced in communities of many species, where there are even more potential niches, highlighting an important role for perturbations, such as species removal, in evolving ecological communities.

eTOC:

Ascensao et al. explore the capacity of simple microbial communities to regenerate ecological diversity after ecotype removal. Rediversified communities show major shifts in ecological dynamics and interactions. This study highlights how community structure is shaped by adaptive pathways that only become accessible after a community perturbation.

Introduction

Ecological diversification refers to the evolution of a population or community of organisms to occupy distinct ecological niches or habitats within an ecosystem¹. Such diversification can manifest through various mechanisms, including the evolution of unique physical or behavioral traits that enable individuals to utilize diverse resources or withstand varied environmental conditions^2–4. The propensity for ecological diversification in a community is influenced by factors like environmental conditions, prevailing biodiversity, and the interactions among the species present^4–7. The potential for diversification can be modulated by the presence of unoccupied niches, or “ecological opportunities”^5,8. These opportunities may diminish as diversity increases and niches become occupied. However, existing communities can also generate new niches, facilitating the introduction of novel ecotypes. An example is cross-feeding, where species produce metabolites that pave the way for the rise of new ecotypes by forming exploitable niches^9–13.

Microbial communities offer a valuable model for investigating the intertwined evolutionary and ecological processes driving diversification due to their rapid reproductive and evolutionary rates^11,13–21. In both natural settings^22–25 and experimental systems, swift ecological diversification in microbial communities has been documented, typically propelled by mechanisms such as cross-feeding^26–30, resource partitioning^11,31–34, spatial niche differentiation^15,35–37, and potentially other ecological trade-offs^38,39. Interactions within microbial communities can either inhibit^17,18, promote^11,13, or have mixed impacts⁴⁰ on diversification. The enduring coexistence of a new ecotype with its immediate ancestor is not assured and may hinge on community characteristics, such as metabolic trade-offs^12,41. In experimental contexts, ecological differentiation of a diversified ecotype is often indicated when an ecotype’s fitness inversely correlates with its frequency, i.e. displaying negative frequency-dependent fitness effects. Stable coexistence between the diversified ecotype and its ancestor is implied if the former can invade when rare but not when abundant.

Even when ecotypes can stably coexist, it does not guarantee that they will coexist indefinitely or at all locations. Ecotypes can migrate to new territories, potentially without other community members, or some ecotypes within the community may spontaneously go extinct (e.g. due to demographic stochasticity or environmental fluctuations). In either case, the community becomes perturbed, losing one or more members and potentially leaving ecological niches unfilled. Environmental disturbances that cause ecotype loss are prevalent across diverse types of microbial ecosystems, including aquatic, soil, and human-associated environments^42–48. Oftentimes, local ecotype/species extinction is not benign–loss of microbial taxa has been associated with deterioration of ecosystem functioning in natural systems^49,50. It has long been noted among biologists that newly isolated species, and species extinctions, can open up ecological opportunities and lead to rapid diversification events^51–54.

Theoretical models suggest that perturbed communities may respond with a combination of ecological and evolutionary changes^55–58. These evolutionary changes may include both directional and diversifying selection⁵⁸, with newly evolved variants either replacing existing community members or coexisting alongside them. However, it remains unclear which communities have the potential to rediversify. When rediversification does occur after ecotype removal, there are two possible scenarios: (i) the perturbed community rediversifies and eventually returns to a state similar to the original community before the disturbance, or (ii) the perturbed community rediversifies and forms a community that is qualitatively different from the original one.

Here, we investigate the aforementioned questions surrounding rediversification using a minimal microbial model community of only two, naturally diversified E. coli strains. Specifically, we employ two strains derived from the E. coli Long-Term Evolution Experiment (LTEE), which was started by Dr. Richard Lenski and has been running for over 30 years or more than 70,000 generations⁵⁹. An initially isogenic strain of E. coli was split into 12 replicate populations and propagated through daily dilutions in glucose minimal media (DM25). At the outset of the LTEE around 6.5k generations, it was found that one lineage, ara-2, spontaneously diversified into two lineages–S and L–that coexist via negative frequency dependence⁶⁰. The ecotypes were named for the sizes of their colonies on certain agar plates, either small (S) or large (L). The S and L lineages inhabit distinct temporal and metabolic niches in the LTEE environment. During exponential phase, L grows more quickly on glucose, while S specializes in stationary phase survival and utilizes acetate, a byproduct of overflow metabolism^27,61. Since their diversification, the lineages have persisted and evolved over time, exhibiting genetic, transcriptional, and metabolic divergence^27,60–66. The LTEE-derived communities are ideal for our plan to investigate the possibility and potential patterns of rediversification over evolutionary time. We can revive the S-L community at 6.5k generations to probe rediversification right after emergence of the community, and compare with rediversification at later stages of the evolution experiment.

We found that when we isolated the S ecotype under certain conditions, it would spontaneously rediversify, giving rise to a new big colony ecotype S_B, even if we used S clones separated by more than 30,000 generations of evolutionary time. The new ecotype, S_B, displays hallmarks of ecological differentiation, including negative frequency-dependent fitness effects when in coculture with its ancestral S clone. We dissected the new, rediversified community, and found that while S_B shares a number of traits with both L and S, it also behaves in entirely new ways. Our findings suggest that even in a maximally reduced community of only two strains, evolution may leave room for alternative diversification processes, suggesting a hidden adaptive potential only revealed by ecotype removal. This raises the possibility that perturbations, such as ecotype removal, could play an important role in evolving ecological communities by creating opportunities for alternative evolutionary pathways.

Results

S can quickly diversify into a new ecotype

The ability of the S ecotype to emerge and coexist with the L ecotype in the LTEE has been attributed to its proficiency in scavenging acetate released from overflow metabolism during glycolysis, as well as its ability to survive and thrive during stationary phase^27,61. It has been proposed that the L-S and similar polymorphisms may arise because of a fundamental, hard-to-break trade-off between glucose and acetate growth rates in E. coli^12,67. Based on these explanations, one may suspect, that after removing either L or S in the two-ecotype community, the community may eventually rediversify and will eventually approach a two-ecotype community similar to the original L-S community.

We performed a simple experiment where we cultured an S clone isolated around 6.5k generations, immediately after the ara-2 lineage diversified into S and L, in glucose minimal media (DM25) for approximately 60 generations (9 days), with 12 biological replicates. To visualize colony morphologies of the resulting cultures, we plated the cultures on tetrazolium arabinose (TA) agar plates. Surprisingly, 2 of the independent cultures displayed a mixture of large and small colonies (Figure 1A).

Figure 1. Emergence of the stably heritable S_B morph and frequency-dependent fitness effects.

(A) Big colony morphs can arise in S cultures derived from 3 different LTEE timepoints, separated by more than 30,000 generations of evolution (6.5k clones are shown here as an example). When both small and big colonies are isolated and propagated in liquid DM25 culture for about 30 generations, then plated on TA agar plates, we see that the colony size is heritable. (B-D) Reciprocal invasion experiments, measuring relative fitness of clones when they are in the minority of the population (approximately 1%) or in the majority (approximately 99%). Each point represents a biological replicate, horizontal lines represent mean across all points. Competitions between (B) S_B clones and S, (C) S and L, and (D) S_B and L. We generally see negative frequency dependent fitness effects across all strains and competitions. (E) Triple competition between S_B, S, and L, where L and S_B are near their equilibrium frequencies and S_B in the minority (around 1%). See also Figures S1, S2, and S3.

After eliminating contamination possibilities by sequencing several diagnostic genetic loci, we examined whether the large colony phenotype was heritable. We isolated several large and small colonies and propagated them in DM25 for around 30 generations (5 days). The phenotype appeared to be stably heritable for all selected colonies. To avoid prematurely associating the larger colony phenotype with the L type, we referred to the emerging type in our experiments as S_B, due to its large (big) colonies and its ancestor S.

To gain insights into the robustness of the observed rediversification after isolation of S over evolutionary timescales, we isolated S from later generations, spanning more than 30,000 generations of evolution. We repeated the same experiment with S clones from 17k and 40k generations with 24 independent cultures each; however, we did not see any noticeable emergence of big colonies after 60 generations (screened about 200 colonies per plate). It is unclear why we did not see any big colonies; one possible explanation may be that the rate at which S morphs transition to S_B morphs may be low enough that we would need to have many more replicate cultures to observe rediversification (as in the 6.5k S clones). We previously noticed that 6.5k S_B clones (labeled 1 and 6) grew much better in LB liquid media compared to S clones (potentially accounting for their bigger colonies sizes on similar agar plates). Thus, we sought to see if we could enrich for the appearance of S_B by growing 6.5k, 17k, and 40k S clones in LB liquid culture. Under these growing conditions, we indeed saw that S_B colonies appeared rapidly, within 1–3 days, in nearly all of the independent S cultures across the three LTEE timepoints (Figure S3). We attributed this to the higher fitness of S_B in LB, relative to S. The new S_B clones were again stably heritable for at least 30 generations.

The big colony phenotype S_B bears at least a superficial resemblance to L, which begs the question: do S_B and S represent genuinely different ecotypes, occupying different ecological niches, with the potential to coexist with each other? To answer this question, we performed reciprocal invasion experiments, where we mixed S and S_B clones at high and low frequencies, and tracked how their frequencies change via flow cytometry over the course of three growth cycles (see Methods), to estimate their relative, frequency-dependent fitness effects (Figure 1B). While relative frequencies of LTEE strains are typically measured by colony counting, we found significant bias (Figure S1G–H) in frequency measurements of S/L when measured via colony forming units (CFUs). In contrast, we see that flow cytometry provides unbiased frequency measurements (Figure S1A–F). We thus chose to use flow cytometry for all further measurements instead of CFUs, owing to its minimal bias and reduced measurement noise (Figure S1). The introduction of genomically-integrated fluorescent proteins does not have a measurable impact on fitness (Figure S2C).

We found that most S_B clones had significant negative frequency-dependent fitness differences when in competition with their parental S clone, a hallmark of ecological differentiation (p < 0.05 for all clones except 6.5k S_B 3). These data suggest that many of the S_B clones can coexist with S, because relative fitness is greater than 0 at low frequencies and less than 0 at high frequencies. However, it is not clear if this is the case for all of the isolated S_B clones, as some have a relative fitness near or less than 0 at low frequencies. This may be because the aforementioned S_B clones either genuinely do not coexist with S, or perhaps they coexist at a frequency around or lower than the one where we took the measurements.

The frequency-dependent fitness differences between S_B and S were similar in magnitude to the fitness differences between L and S (Figure 1C), which were all significant at p < 0.01. We also competed S_B against L (Figure 1D), and again found significant frequency-dependent fitness differences for most clones (p < 0.01 for all clones except 6.5k S_B 3 and 17k S_B 1). However, if at least some S_B clones can invade both S and L when rare, why has the S_B morph not appeared in the ara-2 population of the LTEE, where L and S have been coexisting and coevolving for tens of thousands of generations? We hypothesized that S_B could not invade an already “full” community, and could only have the chance to invade when one of the ecotypes is removed. We performed a triple competition experiment, with L and S near their equilibrium frequency and S_B in the minority (Figure 1E). We found that most S_B clones had a significantly lower fitness compared to when it was in the minority with either S or L alone (p < 0.05 except for 6.5k S_B clones 2 and 3 compared to when competed against S alone, and 17k S_B 1 6.5k S_B 3 when competed against L alone).

While we have shown that S_B spontaneously emerges from a monoclonal population of S and occupies a distinct ecological niche, it is not yet clear how S_B compares to S and L. In particular, we want to understand if S_B simply fills the same niche that L had occupied before removal, making it somewhat functionally equivalent to L. The negative frequency-dependent selection between L and S_B suggests that they must be different to some degree, but it is still unclear if S_B represents a sort of intermediate between S and L, or if it shows novel traits. In the following, we will show that while S_B resembles L in some of its growth properties, it also shows clear differences that are critical for its coexistence with S.

Within-cycle growth dynamics of cocultures

To better understand how ecological differentiation arises in the S_B-S and L-S systems, we measured the within-cycle growth dynamics of S_B, L, and S in coculture with each other via flow cytometry. The LTEE environment is a seasonal one^27,68–every 24 hours, cultures are transferred 1:100 into fresh glucose minimal media. The populations spend the first part of the day in exponential phase; the remaining time, more than 2/3 of the day, is spent transitioning out of exponential phase and in stationary phase. It has been previously shown that L and S occupy different temporal niches from one another, where L specializes on exponential growth on glucose, and S specializes on stationary phase survival and growth on acetate. Thus, it is natural to ask how temporal variations in growth are similar or different in the S_B-S system.

To perform the experiments, we propagated S, S_B, and L separately in monocultures for two days, before mixing S with S_B and S with L, both at high and low frequencies. We mixed strains with their partners from the same LTEE generation. For simplicity, we only used S_B clone 1 for all experiments and LTEE generations. We propagated the cocultures for one more cycle to allow the populations to physiologically adapt to the new environment. At the end of the 24 hour cycle, we took a flow cytometry measurement of the culture, then split the cultures into biological replicates and diluted the cocultures 1:100 into fresh media. Afterwards, we took flow cytometry measurements from the cocultures approximately every hour for about eight hours, then we took one last measurement at the end of the 24 hour cycle (Figures 2, S4). We chose this design because the fastest dynamics occur during and right after exponential phase–the first 8 hours–while dynamics in stationary phase are much slower. The cultures were grown in a 37°C shaking water bath. We corrected the cell counts measured in flow cytometry by the total dilution rate.

Figure 2. Growth dynamics of cocultures over the course of one twenty-four hour growth cycle.

Measurements were taken approximately every hour via flow cytometry for the first eight hours after transfer into new media. An additional measurement was taken approximately 24 hours after the start of the cycle. Mixed S_B 1 with S along with L with S, all from 6.5k generations, where ecotypes were mixed both in the majority and minority of the population. Different lines represent biological replicates. (A-B) Frequency dynamics of S against S_B and against L. (C-D) Total cell count dynamics, separated by each strain in the cocultures. (E-F) Empirically measured growth rates over time for each strain in the cocultures, calculated as the slope of log-transformed abundance between adjacent timepoints, using the second timepoint as the x-axis location. Insets show growth rates during stationary phase, from around 8 to 24 hours, on the y-axis–presented to provide a more fine-grained view of the slow changes in abundance during stationary phase. Error bars represent standard errors. See also Figures S4 and S5.

We initially focus on the dynamics of strains from 6.5k generations (Figure 2). Overall, it is immediately clear that there are larger differences in dynamics in the L-S cocultures compared to the S_B-S cocultures. When S is in both the majority and minority, L has a long, two hour lag time, while S starts growing much more quickly (Figure 2C–D), causing a large upward spike in S frequency. We fit a generalized logistic model to the growth curves to more precisely extract the lag times (Figure S5), and we see that 6.5k L has a longer lag time than S (p < 0.01, in both cases). When S is cocultured with S_B, we do not see any noticeable lag time; however, when S is in the minority, S “wakes up” more quickly than S_B (p = 7·10⁻⁸), leading to a small spike in S frequency at the beginning of the time course. We see similar patterns in the cocultures from 17k and 40k generations–both L and S_B appear to have growth rates very close to 0 at the beginning of the timecourse, but S consistently has a larger initial growth rate (Figure S4). The initially faster growth of S only occurs when S is in the minority for both L and S_B strains (p < 0.02, across all comparisons); there is no longer a noticeable difference when S is in the majority (p > 0.1, across all comparisons).

When 6.5k L starts growing, it has a significantly larger growth rate than S, pushing the frequency of S back down. The magnitude of this growth rate difference is similar regardless of the relative frequency of the ecotypes (Figure 2E–F). In contrast, the differences between S_B and S are much smaller. At both starting frequencies, S_B may have a small growth rate advantage compared to S early in exponential phase, then S appears to grow faster in late exponential phase.

In contrast to the dynamics in lag and exponential phase, the later “stationary phase” dynamics are highly dependent on which ecotype is in the majority. While most conditions show non-zero growth rates after about eight hours of growth, we still refer to this period as stationary phase, because the growth rates are small. When S is cocultured with L, S grows better than L under both conditions, but the absolute growth rates differ between the conditions (insets in Figure 2E–F). When S is in the minority with L, both S and L have net positive growth in stationary phase, although it is higher for S (p = 9·10⁻⁴), potentially pointing to the favorable conditions of L-dominated stationary phase and the putatively large amount of excreted acetate available for exploitation. In contrast, when S is in the majority with L, S has a smaller, albeit still positive, net growth rate, while L has a net negative growth rate in stationary phase (p = 5·10⁻⁴). Concordantly, these patterns suggest that S-dominated stationary phase is much less hospitable to both S and L.

We see different stationary phase patterns when S_B and S are in coculture, where S_B now performs consistently better than S (insets in Figure 2E–F). The ecotype growth rates are significantly different (p = 0.016) when S_B is in the majority with S–S_B has a moderately positive net growth rate, while S has essentially a net 0 growth rate in stationary phase. Then when S_B is in the minority, both S_B and S have net negative growth rates, but S declines more than S_B, although the difference is non-significant (p = 0.16). If S_B were more similar to L, i.e. an exponential phase specialist that secretes a substantial amount of acetate, we would have expected that S_B-S and L-S cocultures would have similar behavior in stationary phase. Instead, S_B appears to have enhanced survival in stationary phase, and decreases the survival prospects of S, perhaps because of the reduced availability of acetate. Thus, while S_B does not have a significant advantage over S in exponential phase, like L has, it compensates with a clear advantage over S in stationary phase, essential for coexistence of S_B with S.

The results show differences in stationary phase behavior across generations, as well as several conserved features (Figure S4). Similar to the 6.5k strains, when 17k S are in the minority with L, S has a large positive growth rate during stationary phase, while L does not grow. However, when S is in the majority with L, its growth rate is comparable to that of L. The 40k S and L strains show different patterns, where L generally has a higher stationary phase growth rate. However, this appears to be offset by a large growth advantage of S right at the end of exponential phase/beginning of stationary phase; this growth advantage is much larger when S is in the minority compared to when it is in the majority. This indicates that the growth advantage of 40k S has shifted earlier, potentially because it has adapted to consumed the acetate secreted by L much more quickly.

Again, the stationary phase behavior when 17k and 40k S and S_B are grown in coculture is noticeably distinct from the behavior of L-S coculture. Similarly, 17k S also does not grow well in S_B-dominated stationary phase. And 17k S_B actually has a large positive stationary phase growth rate when S is setting the environment, suggesting that S_B has more to gain from stationary phase when it is in the minority compared to vice versa. The picture shifts again with the 40k strains–S_B benefits very little from being in stationary phase, but in contrast, S grows well in stationary phase, especially when dominated by S_B. This is quite different from the behavior of 40k S-L cocultures, albeit in a different direction than the strains from the earlier generations. Thus, in 40k cultures, it appears that S_B-S cocultures act more like L-S cocultures from earlier generations, where S_B is the clear exponential phase specialist and S is the stationary phase specialist.

Together, these results show that growth traits of L-S cocultures change over evolutionary time, and S_B-S cocultures are similar in important ways (e.g. initial growth rates), but also show departures from the original community (e.g. stationary phase behavior) that reveal how the ecological dynamics have shifted with the new, rediversified ecotype.

Growth traits in novel environments

While we have shown that S_B has distinct growth traits when in coculture with S in the evolutionary condition, does S_B also behave differently compared to S in novel environments that neither have been in contact with before? If S and S_B mostly behave similarly in novel environments, then perhaps the underlying change between the two morphs is targeted only towards traits relevant to the mechanism of ecological differentiation. Other newly diversified ecotypes have previously shown targeted changes to niche adaptation, such as acetate-specialist E. coli ecotypes that evolve due to specific mutations in the main acetate-scavenging gene, acs^29,30. On the other hand, if pleiotropic effects are widespread, then the underlying metabolic/physiological shift in S_B may involve global, rather than targeted changes.

To this end, we competed S_B clones against S clones for each LTEE timepoint in the same minimal media base as the evolutionary condition (DM), supplemented with different carbon sources (Figure 3). For comparison, we also competed S against L clones for each LTEE timepoint in each of the conditions. We chose four different carbon sources that support growth of S, S_B, and L clones from all timepoints and that enter into central metabolism at different points⁶⁹, potentially allowing us to gain insight into global changes in physiology and metabolism. After growing cocultures together for two days in DM25, we diluted them 1:100 in each different media. We kept the cultures in exponential phase, and took two ecotype frequency measurements via flow cytometry: one right before transfer into the new media, and one at the end of exponential phase. As usual, relative fitness was computed as the change in logit frequency.

Figure 3. Competition of S_B and L against S in novel environments.

(A-D) Red and blue points represent the relative fitness of S in competition with S_B and L clones from the same LTEE timepoint, respectively, where different symbols represent different clones. Competitions performed in exponential phase in the same media base (DM) supplemented with different carbon sources: (A) 200mg/L acetate, (B) 1mg/mL casamino acids, (C) 20mM pyruvate, (D) 20mM glycerol.(E) Principal components analysis, using relative fitness in each environment as features. Percentages in parentheses represent percent variance explained by each principal component. See also Table S4.

We see that for most S_B clones, across most conditions, S_B is noticeably non-neutral relative to S (FDR-corrected two-sided t-tests; Table S4). Consistent with previous experiments⁶³, we see that L is also usually non-neutral relative to S across the different carbon sources. The relative fitness of S_B and L clones varies considerably across timepoints and carbon sources. There is not a clear relationship between the fitness of S_B (relative to S) and the fitness of L (relative to S) from the same timepoint, also visible in the PCA representation of the data (Figure 3E). In the PCA, it appears that S_B clones within timepoints generally cluster together (but not completely), not with the L clones from their timepoint; however, leveraging a modified permutational multivariate analysis of variance (PERMANOVA) test (see Methods), we could not reject the null hypothesis that S_B clones within a timepoint cluster together more than the L clone within the timepoint (p > 0.15 for all timepoints).

Again, there is some variation between different S_B clones. For example, the 17k S_B 1 clone behaves noticeably differently compared to the 17k S_B 2 and S_B 3 clones especially in the pyruvate (p < 0.001, both clones) and glycerol (p < 0.02, both clones) conditions, while the three clones cluster together in the acetate condition. The 17k S_B 2 and S_B 3 clones also appear to cluster together, away from the 17k S_B 1 clone in the PCA plot (Figure 3E). The 17k S_B 1 clone also behaved differently compared to the other two in the reciprocal invasion experiment against 17k L, where 17k S_B 1 did not show noticeable frequency-dependence (Figure 1D). The 6.5k S_B 3 and 40k S_B 1 clones also cluster away from the other two clones within their timepoint. The conditions where these “outlier” clones diverge from the other clones varies between timepoints–6.5k S_B 3 is different when grown in in pyruvate and casamino acids (p < 0.03, all comparisons), and 40k S_B 1 is primarily different in the acetate condition (p < 0.03, all comparisons).

Across all three timepoints, we see that S is better at growth in acetate compared with L. The evidence for this is stronger for strains from 6.5k generations (p = 5·10⁻⁴) and 17k generations (p = 0.017), compared to those from 40k generations (p = 0.093). This is consistent with prior findings⁶¹, and the notion that S represents a consistent acetate-scavenging specialist over evolutionary time. In contrast, the behavior of S_B in acetate is more variable, both across time points and between different S_B clones. Two of three 6.5k S_B clones have a fitness disadvantage in acetate (p < 0.03, in both cases) relative to S (albeit less pronounced compared to L), whereas at least one S_B clone from both 17k and 40k generations have a fitness advantage in acetate (p < 0.02, in both cases).

We found the fitness disadvantage of 6.5k S_B in acetate puzzling, because it appears to generally perform better in stationary phase compared to S (Figure 2). At least in the S-L community, the advantage of S in stationary phase appears to be mostly driven by decreased lag time to acetate growth⁷⁰, increased acetate growth rate⁶¹, and increased ability to scavenge dead cells²⁷. We began by testing if S_B has a smaller glucose-to-acetate lag time compared to S, which could explain its stationary phase advantage. By growing S_B, S, and L strains on media containing both glucose and acetate, we could observe how they transition from glucose to acetate growth (Figure 4A). Contrary to expectations, S_B actually has a longer lag time than S, more similar to that of L, which should give it a disadvantage in stationary phase. So we moved on to testing if S_B could scavenge cell debris more efficiently than S. To do so, we competed mixtures of S_B and S, and L and S, together in heat-killed cultures, resuspended in blank media to eliminate the presence of extracellular metabolites (Figure 4B). Consistent with prior data²⁷, we see that S is generally more efficient than L at growing and surviving in an environment of dead cells, albeit with a frequency-dependent effect. In contrast, S_B is able to scavenge dead cells better than S, which may explain its ability to perform better in stationary phase.

Figure 4. Traits affecting coexistence.

(A) Growth dynamics of 6.5k strains in glucose/acetate media. All strains were grown in monoculture, and prior to approximately 6 hours, they are growing exponentially on glucose. Afterwards, a diauxic shift if visible in all growth curves, where the strains begin to switch to using acetate. Error bars are standard deviations, over 8 biological replicates for each culture. Inset represents lag time from glucose to acetate growth; error bars are 95% CIs. (B) Survival and growth of strains on heat-killed cells provides evidence of differential ability to grow and survive on cell debris. After 24 hours of growth, we thoroughly washed a 6.5k S culture (to remove any extracellular metabolites), resuspended the culture in blank media, and heat-killed the culture. We then washed cultures of 6.5k strains L 1, S_B 1, and S 1, mixed the strains, took flow cytometry measurements, and transferred the cultures 1:1000 to the heat-killed culture. We then allowed the cultures to incubate for 16 hours before taking another measurement.

Together, this is another sign that S_B is occupying a genuinely different ecological niche compared to L, which may be shifting over evolutionary time.

Transcriptional differences between ecotypes

Given the strong heritability of the S_B phenotype, and multiple traits that differ with respect to S, we reasoned that the S_B phenotype may have an underlying genetic cause. Thus, we performed whole-genome shotgun sequencing of several S and S_B clones with both short-read sequencing (Illumina) and long-read sequencing (Nanopore) (see Methods). After reference-based assembly, we saw that all S_B clones had several mutations relative to their ancestor, and all S clones from the same LTEE generation also has several mutations relative to each other. The mutations were a mix of synonymous and non-synonymous point mutations, insertions and deletions, and several large genomic rearrangements (Table S3). However, none of the mutations differentiated S and S_B–there were no consistent mutations in specific genes or operons. The large number of mutations separating S_B clones from their S ancestor is not surprising; the ara-2 lineage fixed a hypermutator allele before the S and L lineages split, such that the germline mutation rate is about 100x higher than that of the LTEE ancestor⁷¹. This makes it likely that many of the mutations are likely (nearly) neutral hitchhikers, or otherwise were not affected by selection. We attempted to determine if there was any parallelism on the level of KEGG annotations instead of genes, but again, we did not detect anything. Thus, because of the combination of the high mutational background and lack of detectable genetic parallelism, we cannot determine if the S_B phenotype has a genetic cause, or what the causative mutation(s) would be. If the S_B phenotype is caused by some genetic change, it is likely that many different mutations are able to cause a similar phenotype.

To further understand the underlying causes of the S_B phenotype, we turned to measuring transcriptional differences between L, S, and S_B from 6.5k generations using RNA-Seq. We chose to focus on 6.5k strains because this is the LTEE timepoint immediately after the S and L lineages diversified, allowing us to focus on the “minimal” differences between S and L, rather than after extensive evolution and divergence. We cultured two biological replicates of two independent clones of each L, S, and S_B from 6.5k generations in glucose minimal media, and collected samples in mid-exponential phase (see Methods), in line with previous, similar transcriptomic measurements^63,72. For a broad overview of the data, we first performed a principal components analysis, using (normalized, transformed) expression for each gene as the features (Figure 5A). We see that the first principal component already captures more than half of the variance between samples, which primarily serves to separate the L clones from the S and S_B clones. The S clones appear to cluster together strongly, with the S_B clones flanking them. Hierarchical clustering also reveals that the S clones cluster together, with the S_B 2 clone as the next most similar, and the S_B 1 clone as the outer-most member of the cluster (Figure S6A). This suggests that there are more differences between S and S_B than there are between the two S clones, but there are stronger differences comparing both S and S_B with the L clones. The same picture emerges if we look at the distribution of log₂ fold expression changes between different ecotypes (Figures 5B, S6D). Comparing S and S_B with L, there are many genes with a large range of expression changes, both increasing and decreasing in expression. In contrast, there are generally smaller differences between the two S_B clones and S. Again, there are larger and more differences between S_B 1 and S, compared with S_B 2 and S, suggesting variability between the two S_B clones.

Figure 5. Results from RNA-Seq of L, S, and S_B clones from 6.5k generations.

(A) Principal components analysis of RNA-Seq data, after processing. Samples with the same name represent biological replicates of the same clone; the 1 and 2 labels are to indicate which clone the samples come from. (B) Distributions of log₂ fold changes in gene expression across all genes, comparing different strains to each other. (C-F) Results of a KEGG gene set enrichment analysis to identify pathways with coordinated changes in gene expression between ecotypes, where (C-D) is comparing S_B to S and (E-F) is comparing L to S. Only pathways that are called as significant at p < 0.05 after an FDR correction are included; points are colored by FDR-corrected log₁₀ p-value. Pathways are ordered by normalized enrichment score, which is roughly a measure of the extent to which pathway-associated genes are overrepresented at the top or bottom of the entire list of genes, ranked by fold expression change. The size of the points is proportional to the “gene ratio”, which is the ratio of core enrichment genes to the total number of genes in the pathway, i.e. the fraction of genes in the pathway that show differential expression. See also Table S3 and Figure S6.

Given that there are noticeable differences between S_B and S, we next sought to understand what those differences represent. Are there identifiable pathways with coordinated expression changes? How do they compare with the differences between L and S? To this end, we performed gene set enrichment analyses to identify differentially expressed KEGG (Kyoto Encyclopedia of Genes and Genomes) pathways⁷³. We first compared S_B to S and L to S, and only look at pathways that are significantly enriched at p < 0.05 after a multiple-testing correction (Figure 5C–F). We see that there are a number of pathways significantly down-regulated in S_B compared to S, and only one pathway significantly up-regulated (Figure 5C–D). Most of the downregulated pathways are related to different aspects of amino acid metabolism. We also separately compared S_B 1 and S_B 2 against S to better understand the variability between the two clones (Figure S6C). As expected, most of the terms identified in the pooled analysis (e.g. ribosomal proteins, amino acid metabolism terms) appeared as the top terms when we analyzed the clones separately, albeit at a lower significance level than when the data from the two clones are pooled together. There are potentially a handful of differences in enriched pathways between the two clones. For example, terms related with O-antigen biosynthesis (e.g. Biosynthesis of nucleotide sugars, O-Antigen nucleotide sugar biosynthesis) may be upregulated in S_B 1, but not S_B 2. The differentially expressed pathways between L and S are mostly different, there are no terms related to amino acid biosynthesis, and many terms related to lipid metabolism and O-antigen biosynthesis (Figure 5E–F). Differentially expressed pathways in L tend to not be differentially expressed in S_B, and vice versa (Figure S6B).

There are two pathways that are enriched in both comparisons: flagellar assembly and ribosomal proteins. The changes to flagellar assembly expression is in the opposite direction for S_B and L, where it is up-regulated in S_B but down-regulated in L, suggesting that gene expression for this pathway is ordered L < S < S_B. In contrast, expression of ribosomal proteins is down-regulated in both L and S_B, perhaps indicating some degree of parallelism involving a fundamental aspect of cell physiology between the two ecotypes. But overall, with the exception of the down-regulation of ribosomal proteins, it appears that the transcriptional changes that differentiate S_B and L from S are quite distinct.

Our RNA-Seq data-set was restricted to monoculture mid-exponential phase culture, so we sought to elucidate gene expression changes throughout the growth cycle, while varying community composition, by leveraging the fluorescent reporters we introduced into the ecotypes, all inserted into the same genomic location. Fluorescence intensity is commonly used to measure protein concentration and promoter activity^74–77. The genes encoding the fluorescent proteins are under the control of a constitutive σ⁷⁰ promoter, giving a read-out of σ⁷⁰ activity. The net activity of σ⁷⁰ is a useful measure of the global transcriptional state because it is highly responsive to changes in environmental conditions and growth phases. It plays a crucial role in coordinating the expression of genes in response to environmental stresses, nutrient availability, and other external stimuli^78–81. By monitoring σ⁷⁰ activity, we hope to gain insights into how the cell responds and adapts to different conditions, providing a snapshot of the global transcriptional state.

To this end, we performed an experiment where we mixed 6.5k S 1 with either 6.5k L 1 or S_B 1 at high and low frequencies, after three days of growth in monoculture, and measured their frequencies over time and population-averaged fluorescence intensity at the end of each cycle (Figure 6A–D). We see large, consistent, community composition-dependent shifts in fluorescence intensity for several conditions upon introduction to the community context, especially noticeable when S is in the minority with L, and when S_B is in the minority with S. Specifically, in both cases, the fluorescence intensity sharply increases after one day of coculture growth, a shift that is maintained for the duration of the experiment. The parallel changes in putative σ⁷⁰ activity across S_B and S reveal that both actively change their transcriptional programs in response to community composition. To a less noticeable degree, we see that the fluorescence intensity of L is lower when it is in the minority with S, again pointing to a degree of composition-dependence.

Figure 6. Dynamics of σ⁷⁰ promoter activity revealed by fluorescence intensity measurements.

(A-D) We performed an experiment where we mixed 6.5k S 1 and either 6.5k L 1 or S_B 1 at high and low frequencies at day 0 after three days of growth in monoculture, and measured (A-B) frequencies and (C-D) YFP/BFP fluorescence intensities via flow cytometry. The YFP and BFP genes are under the control of a constitutive σ⁷⁰ promoter. We propagated the strains in coculture thereafter. We see large, consistent, community composition-dependent shifts in fluorescence intensity for several conditions upon introduction to the community context, especially noticeable when S is in the minority with L, and when S_B is in the minority with S. (E-F) We quantified the within-cycle fluorescence dynamics (same dataset as shown in Figure 2), where we see substantial changes over the course of the growth cycle.

We measured the fluorescence dynamics over the course of the growth cycle to investigate how the difference arise (Figure 6E–F). We see substantial, parallel changes throughout the cycle, with fluorescence intensity dropping during exponential phase, before increasing again during stationary phase. These data reveal that increased fluorescence intensity at the end of the growth cycle seen when S and S_B are in the minority with L and S respectively accumulates during stationary phase, with the difference mostly disappearing during exponential phase. Together, these data provide evidence that S_B has a transcriptional reaction to the presence of S in the majority that is parallel to how S reacts to the presence of L in the majority.

Discussion

Our study explores the capacity of an evolved microbial community to quickly regenerate ecological diversity following the removal of an ecotype. Our results suggest that even in the case of a community composed of only two strains in a minimal environment, evolution can leave room for alternative diversification processes.

The rediversified ecotype, S_B, demonstrates the robustness of microbial communities to perturbations by sharing several growth traits with the ecotype it replaces, L. For instance, both S_B and L exhibit slower initial growth or longer lag times compared to S across all LTEE timepoints, which may be involved in a trade-off allowing for higher exponential growth rates, as observed in other systems⁸². However, differences between the rediversified and original communities suggest that the mechanism of ecotype coexistence has shifted. Notably, we observe variations in stationary phase responses, survival, and ability to scavenge dead cells, as well as distinct patterns of gene expression. At least in the 6.5k strains, coexistence between S and S_B appears primarily driven by a tradeoff between glucose growth in late exponential phase (where S does better), and the ability to survive in stationary phase, owing to the advantage in scavenging dead cells (where S_B does better). Together, these findings indicate that ecological rediversification in the S-L system may be influenced by a combination of constraints and opportunities. While some traits may evolve nearly deterministically due to strong ecological or physiological constraints, other trait values may more unconstrained. The interplay between contingency and determinism mirrors patterns observed in various other evolving systems, including the LTEE^83–85. Dissecting why some traits are more evolutionarily constrained during diversification compared to others could be a fruitful avenue for future investigation.

We attempted to determine a potential genetic origin of the S_B phenotype. However, we did not find any consistent mutations shared between the independent S_B clones, relative to their S ancestor. Thus, the S_B phenotype likely either has a large target size, such that many different mutations can cause the same phenotype^86,87, or it is caused by a non-genetic heritable change. Despite the fact that we did not find any shared mutations, the transcriptional changes of two S_B clones were targeted to the same handful of pathways, predominantly related to amino acid metabolism. This points to parallelism among independent S_B clones, at least on the transcriptional level, if not on the genetic level. The downregulation of amino acid biosynthesis may be related to the scavenging lifestyle of S_B, where its amino acid needs can be met by consuming dead cells instead of through de novo synthesis. Additionally, while the differentially expressed pathways in S_B and L relative to S were generally different, we saw decreased expression of ribosomal proteins in both ecotypes. The fraction of the proteome devoted to ribosomes is known to control many growth traits in bacteria^88,89, so the similar changes in L and S_B may help to explain the handful of observed similarities in growth traits. One might expect that ribosome expression should be lower in S, due to its slower exponential growth rate^90,91; so the fact that this is not the case may suggest that S_B and L are both allocating their proteome not just to optimize exponential growth rate, but also other growth traits as well.

While we saw that S could rediversify following isolation, we did not see any obvious ecological or phenotypic diversification when L was isolated. There may be several reasons for this. (i) S may have some amount of physiological/ genetic/ metabolic plasticity that allows it to diversify that L lacks. (ii) Diversification of L may happen slowly or rarely, or more quickly only under certain environmental conditions. (iii) Perhaps L can rapidly diversify, but cryptically, where no phenotypic changes are obvious without more extensive phenotyping. It is certainly the case that we would not have found S_B without the obvious changes in colony size. It could be that rediversification is much more common than currently appreciated, but simply not detected. Sequencing technologies, including metagenomic⁶⁵ and DNA barcoding-based methods⁹², could help to better reveal the full extent of rediversification across microbial communities. In fact, through metagenomic sequencing, we now know that ecological diversification is much more common in the LTEE than previously thought⁶⁵.

Our study contributes to the understanding of the ecological consequences of ecotype removal or extinction, which often occurs in natural microbial communities due to sudden environmental shifts^42–48. The ability of these communities to recover their diversity after such disruptions might be key to maintaining their functions and stability over time. Contrary to the notion that evolutionary processes are too slow to influence ecological recovery, our findings underscore the importance of evolution in the rebound of communities after a disturbance. We used simplified two-ecotype communities, which are ideal for such studies because they are well-documented and amenable to experimental manipulation^27,60–66. The methods we developed to investigate rediversification in this simple model could serve as a framework for understanding this process in more complex ecosystems. The presence of alternative eco-evolutionary pathways, even in a maximally reduced community of only two strains, hints at more complex dynamics in richer ecosystems. Ultimately, our work sheds light on the resilience of microbial communities, their ability to recover ecological diversity, and their adaptability to environmental changes. Future research on the processes that control these dynamics is essential for a comprehensive understanding of microbial community function and stability, especially in the face of environmental shifts.

Resource availability

Further information regarding the manuscript and requests for reagents may be directed to, and will be fulfilled by the lead contact, Oskar Hallatschek (ohallats@berkeley.edu).

Materials availability

All newly constructed strains and plasmids presented in this paper are available upon request.

Data and code availability

All raw genomic and transcriptomic data has been deposited at the NCBI Sequence Read Archive (SRA), accession numbers are listed in the key resources table. Code and processed data are available at https://github.com/joaoascensao/Rediversification. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Key resources table.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Bacterial and virus strains
Various LTEE-derived E. coli strains	This paper	See Table S1
Biological samples
Chemicals, peptides, and recombinant proteins
RNAse-free DNAse	Invitrogen	Catalogue number: AM2222
Ready-Lyse Lysozyme Solution	Lucigen	Catalogue number: R1804M
Critical commercial assays
Monarch Total RNA Miniprep Kit	New England BioLabs	Catalogue number: T2010S
Stranded Total RNA Prep, Ligation with Ribo-Zero Plus	Illumina	Catalogue number: 20040529
DNeasy Blood and Tissue Kit	Qiagen	Catalogue number: 69504
NEBNext DNA Library Prep kit	New England BioLabs	Catalogue number: E7645
Nanopore ligation sequencing kit	Oxford Nanopore	Catalogue number: SQK-LSK109
Deposited data
All DNA/RNA sequencing data	This paper	NCBI BioProject accession: PRJNA970313
REL606 genome	Genbank	NCBI RefSeq assembly: GCF_000017985.1
Processed data	This paper	https://github.com/joaoascensao/Rediversification
Experimental models: Cell lines
Experimental models: Organisms/strains
Oligonucleotides
Oligonucleotides for genotyping and cloning	This paper	See Table S2
Recombinant DNA
pBad-EBFP2 (plasmid with eBFP2 gene)	\cite{Ai2007ExplorationProteins}	Addgene catalog number: 14891
pMRE-Tn7-133 (miniTn7 plasmid to insert sYFP2 into genome at attTn7)	\cite{Schlechter2018ChromaticBacteria}	Addgene catalog number: 118551
pJA17 (miniTn7 plasmid to insert eBFP2 with strong promoter into genome at attTn7)	Designed in this paper; Backbone from pMRE-Tn7-133; eBFP2 from pBad-EBFP2; BBa_J23119 promoter \cite{Part:BBaParts.igem.org}; RBS designed in silico \cite{Reis2020AnModels}	N/A
pJA18 (miniTn7 plasmid to insert sYFP2 with strong promoter into genome at attTn7)	Designed in this paper; Backbone from pMRE-Tn7-133; BBa_J23119 promoter \cite{Part:BBaParts.igem.org}; RBS designed in silico \cite{Reis2020AnModels}	N/A
Software and algorithms
Code	This paper	https://github.com/joaoascensao/Rediversification
flowcytometrytools (v0.4.5)	\cite{yurtsev_flowcytometrytools_2015}	https://eyurtsev.github.io/FlowCytometryTools/index.html
minimap2 (v2.26)	\cite{Li2018Minimap2:Sequences}	https://github.com/lh3/minimap2
Sniffles (v2.2)	\cite{Sedlazeck2018AccurateSequencing}	https://github.com/fritzsedlazeck/Sniffles
HISAT2 (v2.2.0)	\cite{Kim2019Graph-basedHISAT-genotype}	https://daehwankimlab.github.io/hisat2/
featureCounts (v2.0.1)	\cite{Liao2014FeatureCounts:Features}	https://subread.sourceforge.net/featureCounts.html
DESeq2 (v1.38.3)	\cite{Love2014ModeratedDESeq2}	https://bioconductor.org/packages/release/bioc/html/DESeq2.html
clusterProfiler (v4.6.2)	\cite{Wu2021ClusterProfilerData}	https://bioconductor.org/packages/release/bioc/html/clusterProfiler.html
Other
Attune Flow Cytometer (2017 model)	ThermoFisher	N/A
SpectraMax 190 (shaking plate reader)	Molecular Devices	N/A
Pippin Prep	Sage Science	N/A
Femto pulse system	Aligent	N/A

Experimental model and subject details

Most of the experiments presented here were performed in Davis Minimal Media (DM) base [5.36 g/L potassium phosphate (dibasic), 2g/L potassium phosphate (monobasic), 1g/L ammonium sulfate, 0.5g/L sodium citrate, 0.01% Magnesium sulfate, 0.0002% Thiamine HCl]. The media used in the LTEE and the competitions shown in Figures 1 and 2 is DM25, that is DM supplement with 25mg/L glucose. The strains used in this work were all isolated from the E. coli Long Term Evolution Experiment (LTEE), and listed in Table S1.

For competition experiments, generally we first inoculated the strain into 1mL LB + 0.2% glucose + 20mM pyruvate (which we found prevented the emergence of the S_B while allowing for robust growth). After overnight growth, we washed the culture 3 times in DM0 (DM without a carbon source added) by centrifuging it at 2500×g for 3 minutes, aspirating the supernatant, and resuspending in DM0. We transferred the washed culture 1:1000 into DM25 in a glass tube. If a strain was isolated directly from a colony, we would instead directly resuspend the colony in DM25. Generally, we grew 1mL cultures in a glass 96 well plate (Thomas Scientific 6977B05). We then grew the culture for 24 hours at 37°C in a shaking incubator. The next day, we transferred all the cultures 1:100 again into 1mL DM25. After another 24 hours of growth under the same conditions, we would mix selected cultures at desired frequencies, then transfer the mixture 1:100 to DM25. After another 24 hours of growth under the same conditions, we would transfer the culture 1:100 to a desired media and start taking flow cytometry measurements–in the competitions of Figures 1 and 2, the media is DM25, for the competitions of Figure 3, the media is DM supplemented with 200mg/L acetate, 1mg/mL casamino acids, 20mM pyruvate, or 20mM glycerol. For the competitions of Figure 1, we took measurements for 3–4 total days, doing 1:100 serial transfers every 24 hours in DM25; for Figure 2 we took measurements approximately every hour for 8 hours, then another measurement at 24 hours; for Figure 3 we took a second measurement after 8 hours, when the cultures were still in exponential phase. For growth in glucose/acetate media, we grew all strains in DM + 250mg/L acetate + 250mg/L glucose, after three cycles of growth on DM25, measuring OD600 absorbance in a shaking plate reader (SpectraMax 190; Molecular Devices) over the course of 24 hours. For the competitions in the heat-killed cultures, we grew 6.5k S 1, L 1, and S_B 1 cultures overnight in DM2000. We washed all the cultures in DM0 3x, as described above, to eliminate the presence of extracellular metabolites. We then heat-killed a portion of the S 1 culture by incubating it at 70°C for 45 minutes. We then mixed 6.5k S 1, L 1, and S_B 1 cultures appropriately, took a flow cytometry measurement, and resuspended them 1:1000 in 1mL of the heat-killed culture. We allowed the cultures to grow at 37°C for 16 hours (the approximate length of stationary phase), and then took another measurement.

Method details

Integration of fluorescent proteins

We sought to use flow cytometry to quantify ecotype abundances, which would necessitate that we could differentiate the strains via fluorescence. We decided to integrate fluorescent proteins into a neutral genomic location of our various strains rather than using plasmids, because plasmids can carry a significant metabolic burden, and it is often necessary to add antibiotics to the media to select against plasmid loss. We used a system based on that of Schlechter et al.⁹³ to integrate fluorescent proteins with miniTn7, a transposon that inserts cargo at a putatively neutral intergenic site downstream of glmS. Briefly, the system works by mating the recipient strain-of-interest with a donor strain, harboring a plasmid with the miniTn7 proteins, an ampicillin-resistance gene, a temperature-dependent origin of replication, and the cargo flanked by the left and right Tn7 recognition sites. In this case, the cargo consists of a fluorescent protein, under the control of a broad host-range promoter, and a chloramphenicol resistance gene, for selection of integration.

Our protocol for integration proceeded as follows. First, we grew the donor strain with the desired plasmid in LB + 100μg/mL carbenicillin + 10μg/mL chloramphenicol at 30°C shaken, overnight. We also grew the recipient strain overnight in DM2000 media at 37°C, directly from glycerol stock. The next day, we washed the donor culture by centrifuging it at 2500×g for 3 minutes, aspirating the supernatant, and resuspending in DM0. We then measured the optical density (OD) of both cultures, and mixed about 1 OD·mL of each culture on a 20mL LB/agar plate supplemented with 0.2% glucose + 20mM pyruvate. The cultures were allowed to grow into a lawn overnight at 30°C, allowing the donor strain to conjugate with the recipient. Afterwards, we scraped up the lawn and resuspended it in 3mL DM0. We washed the resuspended culture 3 times, as previously described, and then streaked out the culture on a DM2000 + 10μg/mL chloramphenicol + agar plate, then allowing the plates to incubate overnight at 37°C. This step simultaneously selects against the presence of the donor (the donor is a proline auxotroph), against the Tn7 plasmid (it has a temperature-sensitive origin of replication), and for integration of the Tn7 cargo (via the chloramphenicol resistance gene). After two days of growth, we restreaked a number of colonies that appeared on DM2000/agar plates for isolation. We then tested for integration of the Tn7 cargo by amplifying and sanger sequencing the junction between the genome and the fluorescent protein insertion (see Table S2 for oligonucleotide sequences), and by looking for fluorescence via fluorescence microscopy. We confirmed that the plasmid was not present in the colony by testing resistance against carbenicillin. We ensured that the colony was not the donor or a contaminant by checking colony morphologies on tetrazolium -maltose (TM), -arabinose (TA), and -xylose (TX) agar plates. We further confirmed identity by sanger sequencing the arcA and aspS loci of the clones we moved forward with (see Table S2 for oligonucleotide sequences).

We found that the fluorescence provided by the plasmids designed in Schlechter et al.⁹³ were insufficiently strong for our purposes. We also needed two different fluorescent proteins with non-overlapping fluorescence profiles so that we could distinguish the two in our flow cytometer. We decided to use the fluorescent proteins sYFP2⁹⁴ and eBFP2⁹⁵ because they share the same ancestor and are highly homologous, and are thus likely to have the same or similar physiological effects on their host, and they have sufficiently different fluorescence profiles that are compatible with our flow cytometer. Thus, we sought to increase the expression levels of the fluorescent proteins, and add in BFP, by constructing new plasmids. We chose to use the strong, constitutive σ⁷⁰ BBa_J23119 promoter⁹⁶ and a ribosome binding site (RBS) designed in silico with the Salis lab “RBS calculator”⁹⁷, placing them immediately upstream of the fluorescent protein sequences. We used Gibson assembly to construct the plasmids by ordering compatible oligonucleotides with the promoter and RBS sequences on them, and then using the backbone of pMRE-Tn7–133 from Schlechter et al.⁹³ and the eBFP2 gene from pBad-EBFP2⁹⁵ for the BFP plasmid. Final plasmid sequences were confirmed via sanger sequencing.

Flow cytometry

For all population measurements taken with flow cytometry, we used the ThermoFisher Attune Flow Cytometer (2017 model) at the UC Berkeley QB3 Cell and Tissue Analysis Facility (CTAF). For every measurement, we loaded the samples into a round bottom 96 well plate, for use with the autosampler. Typically we diluted the samples 1:5 in DM0, but we changed the dilution rate over the course of the 8 hour within-cycle timecourse. We set the flow cytometer to perform one washing and mixing cycle before each measurement, and ran 50μL of bleach through the autosampler in between each measurement to ensure that there was no cross-contamination between wells. We used the “VL1” channel to detect eBFP2 fluorescence, which uses a 405nm laser and a 440/50nm bandpass emission filter. We used the “BL1” channel to detect sYFP2 fluorescence, which uses a 488nm laser and a 530/30nm bandpass emission filter. For the triple competitions shown in Figure 1E, we used a BFP-tagged S, a YFP-tagged S_B, and a non -fluorescent L strain. To estimate the frequency of L, we added 5 μM of Syto62 red fluorescent dye (ThermoFisher S11344) to the sample immediately before measurement. We used the “RL1” channel to detect Syto62 fluorescence, which uses a 637nm laser and a 670/14nm bandpass emission filter. We always used a sample flow rate of 25μL/min.

We use the package flowcytometrytools (v0.4.5)⁹⁸ to analyze the flow cytometry data. We first perform a hyperlog transform⁹⁹ and then created threshold gates to sufficiently separate the “noise cloud” (nonfluorescent particles present even when running blank media) from particles with clear fluorescence. We noticed that in addition to seeing single positive BFP⁺ and YFP⁺ particles, we also see some particles called as fluorescent in both channels (Figure S2A). We observed that the proportion of double positive events decreased as a function of fluid flow rate and dilution rate (Figure S2B), suggesting that sometimes multiple cells end up in front of the flow cytometry laser at the same time, and are counted as one event. Thus, we sought to correct for this effect. We assume that the probability of a cell ending up in front of the laser is constant per unit time, and uncorrelated in time, i.e. that it is a Poisson process. Thus, for any given window of time, the probability of observing some number of events is distributed as a Poisson distribution. So under this model, the observed BFP or YFP “clouds” will consist of single cells, double cells, triple cells, and so on. Similarly, there are many combinations of BFP/YFP cells that can end up in the double positive cloud. So, in order to get the expectation of the observed frequencies, we add up the contributions of singlets, doublets, triplets, etc by considering the probability of $n$ cells passing in front of the laser together times the probability of all $n$ cells being the same color,

$$f_i^{obs}=\sum _{n=1}^{\infty }p\left(n\text{cells}\right)f_i^n$$ (1)

where $i\in \{1,2\}$. As previously mentioned, $p\left(n\mathrm{c}\mathrm{e}\mathrm{l}\mathrm{l}\mathrm{s}\right)$ will follow a Poisson distribution, but as we do not observe the case when zero cells pass in front of the laser, we will use a zero-truncated Poisson.

$$f_i^{obs}=\sum _{n=1}^{\infty }\frac{{\lambda }^n}{n!\left(e^{\lambda }-1\right)}{f}_i^n=\frac{e^{\lambda f_i}-1}{e^{\lambda }-1}$$ (2)

Where $\lambda$ is the average number of cells per event. We have two equations (for $f_1^{obs}$ and $f_2^{obs}$) and two unknowns ($\lambda$ and $f_1$), so we can solve for the real frequencies, which we solve for via numerical root-solving. The total cell count $N$ also must be corrected, where $N_{\mathit{corrected}}=N_{\mathit{observed}}\lambda e^{\lambda }/\left(e^{\lambda }-1\right)$. The post-correction frequencies appear to be well-reflective of frequencies measured with colony counting (Figure S1). The primary reason why we chose to use a mathematical correction rather than diluting the samples to the point where $\lambda \approx 0$ was for time and efficiency. We found that in order to reduce the number of mixed events to near zero, we would have to run a much larger volume through the flow cytometer, which takes much more time. This is especially problematic for the growth curve experiments in Figure 2, where the dynamics are quite rapid, and long times spent in the flow cytometer would likely distort the data.

Quantification and statistical analysis

Once we obtained estimates of strain frequencies and total abundances, we can calculate several downstream metrics of the dynamics. Throughout the manuscript, we estimate the fitness effect of a strain $s$ from the dynamics of ecotype frequencies, $f\left(t\right)$, using the model

$$\mathrm{l}\mathrm{o}\mathrm{g}\mathrm{i}\mathrm{t}f\left(t\right)=st+\mathrm{l}\mathrm{o}\mathrm{g}\mathrm{i}\mathrm{t}{f}_0+{\epsilon }_t.$$ (3)

We measure time $t$ in units of 24-hour growth cycles, and thus fitness effects are measured in units of 1/cycle. We fit the model using ordinary least squares, jointly estimating $s$ and the intercept $f_0$. We calculated p-values for differences in fitness effects as a function of initial ecotype frequency (Figure 1B–D) using a standard two-sample $t$-test, then performed a standard Benjamini-Hochberg FDR correction, pooled across all of the comparisons.

Once we obtained fitness data to measure growth traits in novel environments (Figure 3), we were first interested in testing how often fitness effects are significantly non-neutral relative to $S$ (i.e. deviate from 0). Thus, we performed standard two-sided, one-sample $t$-tests on all environment-strain conditions, then corrected the p-values with a Benjamini-Hochberg FDR correction (Table S4). To test if S_B clones (within a timepoint) had significantly different fitness effects in each condition, we used two-sided, two-sample t-tests, then again corrected with a Benjamini-Hochberg FDR correction. Then, we were interested in testing the hypothesis that the S_B clones within a timepoint cluster together more strongly than with the L clone within the timepoint. Thus, we turned to using a slightly modified version of permutational multivariate analysis of variance (PERMANOVA)¹⁰⁰, to better reflect the structure of our data, where we have multiple biological replicates for each measurement. To calculate the F statistic, we first computed the mean relative fitness effect of each strain in each environment, across biological replicates, $\bar{s}$. We used total Euclidean distance between mean fitness effects as the metric, i.e. the total squared distance is the sum over all environments of the squared difference in fitness between two strains. We only computed the distance for the four strains (S_B clones 1–3, L clone) within a timepoint. As previously described¹⁰⁰, and without modification, we used the total sum-of-squares and the within groups sum-of-squares to compute the F statistic, using the distance metric. Then, we needed to estimate a null distribution to compare the F statistic. If we were to estimate the null distribution with the standard method, we would only get 3 values (treating either S_B clone 1, 2, or 3 as the outgroup), which is not sufficient to construct a null distribution. However, the estimated mean fitness effects are calculated from a finite number of biological replicates; with resampling, the mean will change. Thus, we use a parametric resampling scheme to model the variability from sampling. We resampled each mean fitness effect 10,000 times using a Student’s $t$ distribution, $s_{rs}=\bar{s}-t\sqrt{\mathrm{v}\mathrm{a}\mathrm{r}s/n}$, where $s_{rs}$ is the resampled fitness effect, $\bar{s}$ is the empirical mean fitness effect, vars is the empirical (unbiased) variance across biological replicates, $n$ is the number of replicates, and $t\sim t_{n-1}$ is drawn from a t-distribution with $n-1$ degrees of freedom. We get similar results if we resample using a gaussian distribution instead of a t distribution. We then treat either S_B clone 1, 2, or 3 as the outgroup (instead of L), and repeat the multivariate ANOVA procedure to get F statistics for each resampled sample. We concatenate the samples from cases where each of the S_B clones is treated as the outgroup, so that the final size of the null distribution is 30,000 samples. We calculate the p-value as the fraction of values in the null distribution that are larger than the F statistic.

We calculated empirical growth rates (Figures 2, S4) over the course of the 24 hour cycle using the total ecotype abundance data, $n\left(t\right)$. For each pair of adjacent time points in a cycle, we used ordinary least squares to extract estimates of the growth rate $\left(r\right)$ and its standard error, using the model $\mathrm{l}\mathrm{o}\mathrm{g}n\left(t\right)=rt+n_0+{\epsilon }_t$. Growth rate estimates were compared using Wald’s test, and p-values were corrected via a Benjamini-Hochberg FDR correction.

We fit generalized logistic models (Richard’s curves) to the 24 hours growth curve abundance data to extract estimates of lag times for each strain and condition (Figures 4A, S5)¹⁰¹. For the OD600 data in Figure 4A, we used the timecourse after the end of glucose exponential phase. Denoting the fitted abundance as $\hat{n}\left(t\right)$, we use the following form for the generalized logistic model,

$$\hat{n}(t)=a{\left(1+T{e}^{-k\left(t-t_m\right)}\right)}^{-1/T}+b,$$ (4)

We jointly infer the parameters $a,T,b,k,t_m$ by using inverse-variance weighted least squares, i.e. $\mathrm{m}\mathrm{i}\mathrm{n}{\sum }_i{\sum }_j\left(n_j\left(t_i\right)-\right.{\left.{\hat{n}}_j\left(t_i\right)\right)}^2/v_{i,j}$, where $i$ labels time points and $j$ labels biological replicates. We use $v_{i,j}$ as the variance of the measurement error, which is taken to be the variance of a Poisson random variable. We implement the global minimization by using the differential evolution optimization algorithm implemented in scipy¹⁰². We used standard bootstrapping to estimate the sampling distributions of the lag times, $L$, with 1000 resamples. We removed outliers from the bootstrapped distributions by first robustly estimating the standard deviation, $\hat{\sigma }$, of the distribution via the median absolute deviation (MAD), $\hat{\sigma }=\mathrm{m}\mathrm{e}\mathrm{d}|L-\mathrm{m}\mathrm{e}\mathrm{d}L|/0.67449$. Then we counted a bootstrapped replicate as an outlier, and discarded it, if it was more than 3 standard deviations away from the median, $3\hat{\sigma }>|L-\mathrm{m}\mathrm{e}\mathrm{d}L|$. We then used the bootstrapped distributions to compute confidence intervals and p-values.

Whole genome sequencing

To perform short-read sequencing of S_B and S clones (see SI), we first grew the clones overnight in 1mL of DM2000, then pelleted the cultures and extracted genomic DNA with the DNeasy Blood and Tissue Kit (Qiagen 69504). We prepared the sample libraries with NEBNext DNA Library Prep kit for Illumina according to the manufacturer’s protocol (New England Biolabs E7645). We sequenced the samples with the Illumina 4000 HiSeq 150PE. We used breseq (v0.33.2)¹⁰³ to compare raw reads to the REL606 genome¹⁰⁴ (GenBank: CP000819.1) and to the S ancestor of each S_B, and then call genetic variants. Read coverage was around 100x across the genome, for all samples. We used default parameters for the breseq pipeline, which uses a bayesian model to call single nucleotide polymorphisms, incorporating information from the FASTQ phred quality score from each read¹⁰³.

We attempted to determine if there was any parallelism between 6.5k S_B clone mutations on the level of KEGG annotations, focusing on nonsynonymous and indel mutations. We excluded 6.5k S_B clone 4 from the analysis, as it is a sister to clone 2. We first compiled all KEGG annotations of all genes with nonsynonymous and indel mutations across the 6.5k S_B clones, and computed how many times a gene mapped to a given annotation. For each annotation that appeared, only one gene in our set mapped to it. We then expanded our list to include genes immediately adjacent to intergenic mutations, as well as pseudogenes affected by mutations. In the expanded gene set, we see two genes each map to three different annotations (carbon metabolism, exopolysaccharide biosynthesis, sulfur metabolism). We had planned to implement the multiplicity test to detect parallelism presented in Good et al. (2017)⁶⁵, however, they recommend focusing on set items with 3 or more hits to avoid false positives from low counts. We thus do not believe that there is any parallelism between mutations on the level of KEGG annotations.

To perform long-read sequencing of S_B and S clones (see SI), we again grew the clones overnight in 1mL of DM2000, then pelleted the cultures. High-molecular weight DNA extraction was performed via a standard phenol-chloroform extraction and isopropanol precipitation. Distribution of DNA fragment sizes were obtained using the Agilent Femto Pulse System. Fragment size selection was performed using Pippin Prep (Sage Science). The samples were prepared for sequencing with the Nanopore ligation sequencing kit (Oxford Nanopore, SQK-LSK109). The libraries were then sequenced on an Oxford Nanopore MinION. We used minimap2 (v2.26)¹⁰⁵ and sniffles (v2.2)¹⁰⁶ with default parameters to detect structural variants.

RNA sequencing

6.5k S and L clones 1 and 2 were isolated from REL11555 and REL11556 respectively; 6.5k S_B clones 1 and 2 were the same clones as previously described. Cultures of 6.5k S, S_B, and L clones 1 and 2 were started directly from glycerol stock into 1ml LB + 2g/L glucose + 20mM pyruvate, as a pre-culture. We started two independent cultures for each clone as biological replicates. After overnight growth, the cultures were washed by centrifuging the cultures at 2500×g for 3 minutes, aspirating the supernatant, and resuspending in DM0, repeated three times. Then, the cultures were diluted 1 : 10⁻⁴ into 1mL fresh DM media supplemented with 4g/L glucose, in glass tubes. After approximately four hours of growth at 37°C, the cultures were again diluted 1 : 50 in 1mL of the same media in glass tubes. The cultures were grown shaken at 37°C. The cultures were grown to mid-exponential phase, i.e. until OD ∼ 0.4, then the entire culture was immediately centrifuged at 2500×g for 3 minutes to pellet. Immediately after centrifugation, we resuspended the pellets in 25μL TES buffer (10 mM Tris-HCl [pH 7.5], 1 mM EDTA, and 100 mM NaCl) and then lysed the pelleted cultures with 250U/μL lysozyme (Ready-Lyse Lysozyme Solution; Lucigen R1804M) at room temperature for 5 minutes. For all subsequent steps, we used Monarch Total RNA Miniprep Kit (New England BioLabs T2010S) according to the standard given protocol for gram-negative bacteria. Samples were eluted in 30μl nuclease-free water, and stored at −80°C. The concentration and purity of all RNA samples was quantified using Qubit.

RNAse-free DNAse (Invitrogen AM2222) was used to treat the samples for DNA removal. The library preparation was conducted using Illumina’s Stranded Total RNA Prep Ligation with Ribo-Zero Plus kit and 10bp IDT for Illumina indices. Subsequently, the samples were sequenced using NextSeq2000, resulting in 2×51bp reads. The process of demultiplexing, quality control, and adapter trimming was carried out using bcl-convert (v3.9.3) and bcl2fast (v2.20.0.445) (both are proprietary Illumina software for the conversion of bcl files to basecalls). HISAT2 (v2.2.0)¹⁰⁷ was used for read mapping. Reads were mapped to the REL606 genome¹⁰⁴ (GenBank: CP000819.1). The read quantification was performed using the functionality of featureCounts (v2.0.1) in Subread¹⁰⁸. All of the above steps in the pipeline were performed with default parameters, the last two steps also were run with -very-sensitive and -Q 20 tags, respectively. All sequencing and pre-processing steps were performed by SeqCenter, LLC.

After pre-processing, we obtained a matrix of read counts for each gene for each sample. With this table, we used DESeq2(v1.38.3)¹⁰⁹ to compute fold change in expression between strains and variance-stabilized relative expression values for each gene across samples (blindly with respect to the design matrix), all with default parameters. We used the variance-stabilized relative expression values for the principal components analysis (PCA). We used the ashr method (v2.2)¹¹⁰ with default parameters to shrink and regularize the log₂ fold changes in expression. We computed log₂ fold change in expression between samples in two ways, (1) treating the S_B clones as one “strain”, and (2) treating the S_B clones as separate, so that we get different fold changes in expression for each clone. Otherwise, for S and L, we pooled data across the two clones and biological replicates when computing fold change in expression. We used the package clusterProfiler (v4.6.2)¹¹¹ to perform the KEGG gene set enrichment analysis (GSEA)¹¹². We used the previously computed log₂ fold change in expression as the metric to pre-sort the list of genes. We used the gseKEGG method along with the parameters organism=“ebr”, nPerm=1000000, minGSSize=3, maxGSSize=800, eps=1e-20 to perform the analysis.

Highlights:

• Demonstrates rapid microbial diversity regeneration after ecotype removal.

• Unveils changes in ecological dynamics and community interactions.

• Identifies key growth and survival traits in rediversified ecotypes.