We used six whole-population samples and six clones from generation 50,000 of the LTEE as the founders for our new evolution experiment (Supplementary material 1). The populations are those named Ara–1, Ara–4, Ara–5, Ara–6, Ara+2, and Ara+5, and from those same populations, we used the designated “A” clones that were previously isolated. The whole-population samples and clones were stored at −80 °C, where they have remained viable and available for future studies. Two of the six populations (Ara–1 and Ara–4) evolved hypermutability, while the other four retained the low ancestral mutation rate. Before starting our evolution experiment, we re-isolated clones from the freezer stocks for the six A clones on Davis minimal (DM) agar plates supplemented with 4 mg/ml glucose to ensure the genetic homogeneity of the clonal ancestors. Both the re-isolated clones and 120 μl of each whole-population sample were inoculated into 50-ml Erlenmeyer flasks containing 9.9 ml of DM medium supplemented with 1,000 μg/ml glucose. These cultures were incubated for 24 h in a shaking incubator at 37 °C and 120 rpm. They were then frozen at −80 °C with glycerol as a cryoprotectant, in order to generate and preserve samples of the precise ancestral stocks we used for our evolution experiment.

The new evolution experiment itself was begun as follows. On day –2, we inoculated 0.1 ml of each ancestral stock into 9.9 ml of DM medium supplemented with 2,000 μg/ml glucose. We incubated these cultures for 24 h in the same conditions as described above. On day −1, we diluted a portion of each culture 100-fold into isotonic saline solution (8.5 g/l sodium chloride), then transferred 0.1 ml of the diluted culture into 9.9 ml of DM medium supplemented with 25 μg/ml glucose (the same medium as used in the LTEE), and incubated the cultures for 24 h. On day 0, we took 2 ml from each of the 6 clonal cultures, mixed them well in a flask, and made a starter mix for the MC treatment. We made a similar mix for the MP treatment. We then transferred 0.1 ml of each culture into 9.9 ml of DM medium with 150 μg/ml D-serine (DS150) in an 18 × 150 mm test tube, vortexed the culture, and then incubated the cultures for 24 h in a standing incubator at 37 °C. We prepared 3 biological replicates from each of the 6 clonal and population cultures, making a total of 18 evolving populations in the SC and SP treatments (Figure 1). In those treatments, six sets of three populations shared the same initial genetic background (SC treatment) or the same initial genetic diversity (SP treatment). We also started 18 populations from the clonal starter mix for the MC treatment, and 18 populations from the population starter mix for the MP treatment (Figure 1). The 18 populations in the MC treatment share the same set of initial genetic backgrounds, and the populations in the MP treatment share their initial genetic diversity, although very rare alleles might have been distributed unevenly, by chance, among the replicates of these treatments at the start of the evolution experiment.

Each SC population was derived from a single colony and hence from one haploid cell. When we say the SC populations had no initial variation that was precisely true at the moment the colony began to grow. Of course, as a colony grows, some mutations invariably occur. Consider a large colony of ∼10⁹ cells. Excluding the lineages that became hypermutable, E. coli in the LTEE have a point-mutation rate of ∼10^–10 per bp (Wielgoss et al., 2011) and a genome length of ∼5 × 10⁶ bp (Jeong et al., 2009). Thus, one expects ∼5 × 10⁵ mutations to occur during the growth of that colony. Although this number is large, several points should be kept in mind. (i) The first point mutation will typically occur only when the colony reaches ∼10³ cells, most mutations happen in the last few cell divisions in a growing colony, and at the end, the great majority (∼99.9%) of cells still have no mutations. (ii) Metagenomic analyses of the LTEE reveal substantial genetic diversity within evolving populations (Barrick and Lenski, 2009; Good et al., 2017), whereas no meaningful variation is seen when clones are sequenced with comparable coverage (Barrick and Lenski, 2009). (iii) Clones from different LTEE populations at 50,000 generations differ in all cases by more than 100 mutations (Tenaillon et al., 2016). (iv) Useful measures of genetic variation reflect not only the number of genotypes but also their relative abundance. This point is critical for understanding adaptive evolution because a population’s rate of improvement depends on its genetic variation in fitness, which is higher when competing genotypes are equally abundant than when one type dominates and others are rare (Fisher, 1930; Lenski et al., 1991). By mixing several clones or populations equally, as we do in the MC and MP treatments, the resulting variation is maximal. Thus, we can say unequivocally that populations in the MP treatment have the most initial variation, those in the SC treatment have the least variation, and populations in the MC and SP treatments have intermediate levels of initial variation. It is immaterial to our results whether some mutations in SC populations occurred during the growth of a colony just prior to the start of our evolution experiment or during the experiment proper.

We transferred the 72 populations (18 populations × 4 treatments) in 9.9 ml of fresh DS150 medium in test tubes daily, following the same 100-fold dilution protocol for 300 days. In this environment, the populations reach a stationary-phase density of ∼5 × 10⁷ cells/ml and a total size of ∼5 × 10⁸ cells. The bottleneck population size after the 100-fold dilutions is thus ∼5 × 10⁶ cells. These values are essentially the same as those for the glucose-limited LTEE populations. We froze samples of each population at −80 °C with glycerol as a cryoprotectant every 15 days through day 165, and then every 15 or 30 days through day 300. We also froze the remaining volume of each culture from day 0.

During the evolution experiment, we diluted and spread cells from each population on tetrazolium arabinose (TA) indicator agar plates every 15 days to check for possible cross-contamination among the populations in the SC and SP treatments, where each population derived from either an Ara^– or Ara⁺ lineage. We did not find any evidence of contamination during the 300 days of our evolution experiment. The populations in the MC and MP treatments had lineages with both marker states at the start, and we tracked the marker ratio in those populations for evidence of changing ratios, which would indicate fitness differences among the heterogenous founders and their descendants in these populations. To that end, we plated samples from the populations in the MC and MP treatments every other day until day 15, then every three days until day 45, and finally every five days until day 300. There was one interruption in the experiment on day 75. When we restarted the populations from the frozen samples, we plated all of them for the first three days to check whether freezing and reviving the samples altered the relative abundance of the marker states in the MC and MP populations with mixed ancestry. We did not see any substantial changes in the marker ratios, indicating that these steps did not substantially perturb the evolution experiment. Moreover, these procedures were applied to the populations in all four treatments, and thus they would not systematically bias the outcome.

We isolated clones from each population at generations 500 and 2,000 (i.e., days 75 and 300, Supplementary material 1) on DM agar plates with 900 μg/ml D-serine, and we re-streaked the clones on TA plates to confirm their Ara marker state. The clones were chosen at random, except that each clone had the numerically dominant marker state for its source population at these time points for the MC and MP treatments. We then isolated Ara⁺ mutants of several Ara^– clones from generation 500 to identify potential common competitors with intermediate fitness relative to other clones from generation 0 to 2,000. Using a single pair of common competitors (isogenic except for the Ara marker state) for the fitness assays simplifies procedures and inferences, and having intermediate fitness allows accurate estimates across a wide range of fitness values. We chose MI2228 and an Ara⁺ revertant MI2339 as the common competitors for the main set of fitness assays (Supplementary material 1). MI2228 and MI2339 have equal fitness in DS150 medium, which indicates that the Ara⁺ mutation is selectively neutral in that environment.

On day −2 of the assays, we transferred 0.1 ml from each competitor’s freezer stock into 9.9 ml of Lysogeny Broth (LB) in a 50-ml Erlenmeyer flask, and we incubated the cultures overnight at 37 °C and 120 rpm. On day −1, we diluted each culture 100-fold in saline solution, transferred 0.1 ml into 9.9 ml of DS150 medium in a test tube, vortexed it, and then incubated the cultures for 24 h in a standing incubator at 37 °C. This day served as the conditioning step to ensure that competitors were acclimated to the environment where they would compete, and where the experimental populations had evolved. The rest of the procedure is the same as described elsewhere for the LTEE (Lenski et al., 1991; Wiser et al., 2013), except for the medium and culture vessel. In brief, an evolved clone always competed against the common competitor with the opposite marker state. We transferred 0.05 ml of each competitor’s acclimated culture into 9.9 ml of DS150, and vortexed the new culture to mix the two competitors. We immediately took a sample, diluted it in saline solution, and plated cells on TA agar. The cultures were incubated for 24 h, at which time we again sampled the cultures and plated cells on TA agar. The resulting red (Ara^–) and white (Ara⁺) colonies were counted after the plates were incubated for a day at 37 °C. We calculated each competitor’s realized growth rate as the log-transformed ratio of its final and initial densities. We then computed the fitness of the strain of interest relative to the common competitor as the ratio of their growth rates during the competition.

We used the generation 0 stocks multiple times for estimating initial fitness levels. We have only 12 generation 0 stocks because we used the same six clones for the three replicates of each clone in the SC treatment, and the same six whole-population samples for the three replicates of each population in the SP treatment. The populations in the MC and MP treatments were derived from their respective starter mixes. We cannot measure the fitness of samples that contain both Ara^– and Ara⁺ cells using our method, which relies on a common competitor with the opposite marker state. Therefore, we used the same six clonal stocks at generation 0 for both the SC and MC treatments, and the same six population stocks at generation 0 for both the SP and MP treatments, and for all three replicates.

We also ran a second set of competition assays using the LTEE ancestors, REL606, and REL607, as common competitors. For these assays only, we used a 1:4 starting ratio at day 0, instead of the 1:1 starting ratio described above, because of the substantially lower fitness of the LTEE ancestors in comparison to the common competitors used above. Specifically, we began each competition assay by mixing 0.08 ml of REL606 or REL607 and 0.02 ml of the strain of interest in the test tube containing the DS150 medium. The assay conditions and the calculations of relative fitness were otherwise the same.

All of our statistical analyses were performed using the referenced tests in R version 4.2.0. The analysis scripts and underlying data have been deposited as indicated in the Data availability statement.

To assess the effect of the initial within-population diversity on adaptation to the new environment, we measured the relative fitness of the evolved bacteria by competing them against the common competitor strains. We cannot measure the fitness of the entire evolved populations using our method, however, because that method requires mixing the evolved bacteria with the common competitor strain bearing the alternative Ara marker, and some populations in the MC and MP treatments had descendants of lineages with both marker states. Therefore, we isolated random clones at generations 500 and 2,000 as representatives of each population, and we measured their fitness. For generation 0, we used the stocks of the founder clones and populations that we froze immediately after the start of the evolution experiment. We used the six clone stocks that we had used to found populations in the SC and MC treatments as the generation 0 samples for those treatments, and we used the six whole-population stocks used to found populations in the SP and MP treatments as the generation 0 samples for those treatments. As a consequence, the generation 0 samples for the SC and MC treatments are technically identical, as are those for the SP and MP treatments.

Figure 2 shows the trajectories of the ln-transformed relative fitness values for the four treatments. As a reminder, the replicate populations in the SC and SP treatments had six different founding backgrounds. In contrast, the replicate populations in the MC and MP treatments originated from the same starter mix of six clones or six whole populations, respectively, and thus the replicates in those treatments shared the same founding backgrounds and diversity. The rate of increase in relative fitness clearly slowed over time in the D-serine environment (Figure 2). That deceleration is similar to what was seen during the first 2,000 generations in the glucose-limited environment of the LTEE (Lenski et al., 1991), and it is indicative of diminishing-returns epistasis (Wiser et al., 2013).

Most importantly for our aims and questions, we found no significant difference in fitness among the four treatments at either generation 500 or 2,000 (p = 0.2300 and p = 0.7213, respectively; one-way ANOVA, Supplementary Table 1). The absence of meaningful differences among the treatments in the rate and extent of their adaptation was surprising to us, given the different levels of within-population genetic diversity at the beginning of the experiment. One possible explanation for the negative results with respect to differences in the final fitness values is that the initial variation present in treatments SP, MC, and MP did not include alleles that were sufficiently beneficial in the novel environment relative to new mutations. In other words, the populations in all four treatments ultimately depended on new mutations for adaptation to the novel D-serine medium, regardless of the different levels of initial genetic diversity. In the sections that follow, we present and examine additional data that helps to explain this result.

We tracked the relative abundance of the two Ara marker states in all treatments during the evolution experiment (Figure 3 and Supplementary Figure 2). The populations in the SC and SP treatments began with a single marker state; in these populations, checking the marker states allowed us to check for cross-contamination, which we did not see. The populations in the MC and MP treatments began with a mix of the two marker states. By tracking the relative abundance of the two states in those populations, we could observe the effects of both initial fitness variation linked to the markers and later beneficial mutations that gave rise to selective sweeps. The MC and MP treatments started with equal culture volumes of four Ara^– lineages and two Ara⁺ lineages; therefore, the log-transformed ratios of Ara^– to Ara⁺ cells were initially > 0 for all of the populations in those treatments (Figure 3 and Supplementary Figure 2).

We observed strikingly similar marker trajectories among the 18 replicate populations in the MC and MP treatments, especially during the first ∼100 generations (Figure 3 and Supplementary Figure 2). Despite the initially greater number of Ara^– lineages, cells derived from one or more Ara⁺ lineages increased in relative abundance in all 36 populations. By 30–50 generations, the Ara⁺ cells were numerically dominant in all 18 MP populations and in most MC populations as well (Figure 3). These initial “bursts” imply that one or more of the Ara⁺ clones and populations initially present in the MC and MP treatments were substantially more fit than the Ara^– clones and populations. We will return to this point in the next section.

By generation 100, all 18 populations in the MC treatment, and most of the MP populations, had reversed course, with descendants of one or more Ara^– lineages rising sharply in abundance relative to the Ara⁺ descendants (Figure 3). The Ara^– descendants remained numerically dominant through the first 500 generations in all 18 MC populations (Figure 3, top), and they evidently fixed in all 18 cases by 2,000 generations (Supplementary Figure 2, top). By contrast, the later marker-ratio trajectories of the MP populations were much more variable. Descendants of Ara^– founders were usually more abundant through the first 500 generations, but with tremendous dispersion between the trajectories (Figure 3, bottom). By 2,000 generations, most MP populations had also evidently fixed one of the marker states, but with several fixations in each direction (Supplementary Figure 2, bottom).

The marker-ratio trajectories also show that bursts leading to the early rise of cells derived from one or more Ara⁺ lineages were much steeper for the populations in the MP treatment than for those in the MC treatment. While the initial ratios were virtually identical, at generation 47 (day 7), the mean log₂ ratios were −0.825 and −7.004 for the MC and MP treatments, respectively, even excluding two MP populations without any Ara^– cells among the hundreds of cells sampled. In fact, all 18 MP populations had a much lower ratio than any of the 18 MC populations, a highly significant difference (p < 0.0001; two-tailed Welch’s t-test). We chose day 7 for this comparison because that is when the MC treatment showed the lowest average log ratio, although several adjacent days show a similarly stark difference between these two treatments.

In summary, we observed strikingly similar marker trajectories among the replicate populations in the MC and MP treatments in the early generations of our evolution experiment. Given the inevitable genetic linkage in asexual populations, this pattern implies that the metagenomes of the populations also evolved in parallel during this early phase. Moreover, this parallelism indicates that selection acted on shared genetic variation present in these populations at the start of experiment (i.e., identical by descent). It is reminiscent of the repeatability observed in previous evolution experiments with other organisms that were also founded by populations with shared initial variation (Burke et al., 2014).

We examined the relative fitness values of the six founding populations and the six founding clones to better understand the similar early marker trajectories seen among the replicate populations in the MP and MC treatments, as well as the difference between those treatments in the slope of those early trajectories (Figure 3). For these analyses, we use the same data as the generation 0 data that underlie the grand means for each treatment in the fitness trajectories (Figure 2).

Given the consistent marker-ratio trajectories toward the Ara⁺ marker state, we expect to see that one or both of the Ara⁺ founders had the highest fitness. Also, given that the early trend toward the Ara⁺ state was much faster in the MP treatment than in the MC treatment, we expect that fitness differential to be greater among the whole-population founders than among the clonal founders. Figure 4 shows the relative fitness of the founding populations (panel A) and clones (panel B). In each panel, we have arranged the founders from the lowest to highest relative fitness.

Focusing first on the whole-population data (Figure 4A), we see that both of the Ara⁺ founders have higher mean fitness than any of the Ara^– founders in the DS150 environment. An ANOVA confirms that there is significant variation in fitness among the founders (p < 0.0001, Supplementary Table 2, top), and Tukey’s test confirms that the Ara+5 whole-population founders are significantly more fit than any of the Ara^– founders. These results thus support our expectation from the marker trajectories that one or both of the Ara⁺ founders had the highest fitness.

When we look at the corresponding data for the clonal founders, we see a more ambiguous pattern (Figure 4B). The relative fitness levels of the clones are more similar; four clones (two Ara⁺ and Ara^–) are virtually identical to one another and slightly higher than two others (both Ara^–). An ANOVA confirms that there is significant variation in fitness among the clone founders (p = 0.0004, Supplementary Table 2, bottom), while Tukey’s test finds no significant difference in fitness among the several most fit founder clones.

Based on the ANOVAs, we estimated the among-founder variance components, V_A, for fitness in these two treatments (Sokal and Rohlf, 1995). That founding variation is what would fuel the earliest response to selection in the evolution experiment before new mutations have had enough time to become relevant. As expected, the estimated variance in fitness among the whole-population founders (V_A = 0.0052) is much greater than among the clonal founders (V_A = 0.0009).

We also performed an additional set of competition assays to estimate the fitness of the founders of our evolution experiment relative to a different pair of common competitors. In this case, the six founders of whole populations and clones competed against the marked ancestors of the LTEE (Supplementary Figure 3). The founders generally had higher fitness relative to the LTEE ancestors than relative to the common competitors used in our other assays. Therefore, we used a 1:4 starting ratio of the founders relative to the LTEE ancestors, instead of the 1:1 starting ratio used in the other competitions (Materials and methods). Otherwise, the assay conditions and calculations of relative fitness as the ratio of realized growth rates were the same. We also arranged and analyzed these data as before.

These additional data also support one of our two expectations based on the marker trajectories, namely, that one Ara⁺ founder had higher fitness than any of the Ara^– founders. In this case, we see that Ara+5 has the highest mean fitness among both the whole population (Supplementary Figure 3A) and clonal (Supplementary Figure 3B) founders. The results of the Tukey tests confirm that Ara+5 had significantly higher fitness than all other whole-population founders and higher fitness than all but one clonal founder. The ANOVAs indicate significant variation in fitness among both the whole-population (Supplementary Table 3, top) and clonal (Supplementary Table 3, bottom) founders. However, the variation in fitness is not greater among the whole-population founders than among the clonal founders. The estimated among-founder variance component for fitness for the whole-population founders (V_A = 0.0274) is essentially identical to the variance among the clonal founders (V_A = 0.0269).

Across the four sets of competitions (founder clones and whole populations, against two pairs of common competitors), we find that the founders derived from LTEE population Ara+5 had the highest fitness in three of these sets (Figure 4A and Supplementary Figures 3A,B), while they were tied for the highest fitness in one set (Figure 4B). These results clearly imply that the early trends toward the Ara⁺ state in the marker-ratio trajectories in the MC (Figure 3, top) and especially the MP (Figure 3, bottom) treatments were caused by the initial fitness advantage that the Ara+5 founders had in the new DS150 environment. By contrast, the subsequent reversals in most trajectories are presumably associated with new mutations that arose during our evolution experiment. In theory, very gradual and uniform reversals could occur even without new mutations if the single most fit founder had a different marker state than the marker state with the higher average fitness across its constituent lineages. However, this hypothetical scenario is clearly not the case for the MP treatment, nor can it explain the variation in the time and strength of the reversals in the MP and MC treatments, as shown in Figure 3. A deeper understanding of the reversals will require future genomic analyses, as we explain in the Discussion section.