The impact of adaptation on the persistence of a balanced polymorphism was explored using the lactose operon of Escherichia coli as a model system. Competition in chemostats for two substitutable resources, methylgalactoside and lactulose, generates stabilizing frequency-dependent selection when two different naturally isolated lac operons (TD2 and TD10) are used. The fate of this balanced polymorphism was tracked over evolutionary time by monitoring the frequency of fhuA−, a linked neutral genetic marker that confers resistance to the bacteriophage T5. In four out of nine chemostats the lac polymorphism persisted for 400–600 generations when the experiments were terminated. In the other five chemostats the fhuA polymorphism, and consequently the lac operon polymorphism, was lost between 86 and 219 generations. Four of 13 chemostat cultures monomorphic for the lac operon retained the neutral fhuA polymorphism for 450–550 generations until they were terminated; the remainder became monomorphic at fhuA between 63 and 303 generations. Specialists on each galactoside were isolated from chemostats that maintained the fhuA polymorphism, whether polymorphic or monomorphic at the lac operon. Strains isolated from three of four chemostats in which the lac polymorphism was preserved had switched their galactoside preference. Most of the chemostats where the fhuA polymorphism was lost also contained specialists. These results demonstrate that the initial polymorphism at lac was of little consequence to the outcome of long-term adaptive evolution. Instead, the fitnesses of evolved strains were dominated by mutations arising elsewhere in the genome, a fact confirmed by showing that operons isolated from their evolved backgrounds were alone unable to explain the presence of both specialists. Our results suggest that, once stabilized, ecological specialization prevented selective sweeps through the entire population, thereby promoting the maintenance of linked neutral polymorphisms.
LABORATORY microbial populations offer a promising approach to investigate the origin and persistence of adaptive polymorphisms. Transient polymorphisms arise as swarms of independently derived mutant alleles of similar phenotypic effect sweep through large populations (Notley-McRobb and Ferenci 2000). Balanced polymorphisms develop spontaneously in response to new niches created by the activities of the organisms themselves: Escherichia coli routinely, although not invariably, evolves a complex multitrophic ecosystem of commensalistic cross-feeding when starved for glucose (Helling et al. 1987; Rosenzweig et al. 1994; Treves et al. 1998), and Pseudomonas fluorescens invariably diverges into three ecotypes, each adapted to specific microenvironments formed in static test tube microcosms (Rainey and Travisano 1998).
We studied a balanced polymorphism at the lactose operon of E. coli that arises when differential consumption of galactosides generates stabilizing frequency-dependent selection in chemostats (Lunzer et al. 2002). TD2, a strain carrying a deregulated K12 lac operon favored during competition for lactulose, and TD10, a co-isogenic strain carrying a deregulated ECOR16 operon (Ochman and Selander 1984) favored during competition for methylgalactoside, form a balanced polymorphism on the mixed sugars when 12
The fitness of TD10 with respect to TD2 varies between a weighted arithmetic mean and a weighted harmonic mean of the fitnesses on pure lactulose and on pure methylgalactoside . The weights, lu and mg = 1 − lu, are the proportions of the total sugar in the fresh medium supplied to the chemostat that are lactulose and methylgalactoside. Selection protects the polymorphism if TD10 is favored when rare and disfavored when common . Figure 1A reveals that, despite the very strong selection pressures on the pure sugars, these conditions are satisfied only within a narrow zone on the resource axis.
In this article we investigate the persistence of this polymorphism to further adaptation and determine whether similar polymorphisms arise spontaneously in monomorphic cultures. Neither outcome is guaranteed. Rare, unconditionally advantageous mutations tend to destabilize the polymorphism by either eliminating (Figure 1B) or displacing (Figure 1C) the narrow zone of coexistence away from the resource supply (held fixed at a ratio of 28:72 methylgalactoside:lactulose). Although mutations producing fitness trade-offs broaden the zone (Figure 1D), there is no guarantee that it remains close to the resource supply. Indeed, because rates of adaptation on different sugars can be very different (Travisano 1997) evolution will likely displace the zone of coexistence away from the resource supply. The spontaneous appearance of a balanced polymorphism requires that fitness trade-offs generate a zone of coexistence that lies across the resource supply. A priori, this would seem improbable when large selection coefficients on the pure galactosides produce only a narrow zone of coexistence on the resource axis.
This article describes three sets of long-term chemostat experiments and a number of short-term experiments that were designed to explore the dynamics of ecological specialization. The first set of experiments involved 9 chemostats that contained both lac operons, TD2, which is selected for on lactulose, and TD10, which is selected for on methylgalactoside (Lunzer et al. 2002), in a mixture of 28% methylgalactoside and 72% lactulose. The second set of experiments involved 13 chemostats where either TD2 or TD10 was used with the same mixture of sugars to determine if the long-term evolutionary dynamics are independent of the lac operon polymorphism. The third set of experiments involved 20 chemostats where either operon was used with either sugar to determine the evolutionary consequences of eliminating environmental heterogeneity.
MATERIALS AND METHODS
Minimal medium is Davis salts [minimal Davis, MD: 7 g K2HPO4, 2 g KH2PO4, 1 g (NH4)2SO4, 0.5 g trisodium citrate in 1 liter of distilled deionized water with 1 ml of 1 m MgSO4·7H2O and 0.5 ml of 1% thiamine added after autoclaving] supplemented with 2 g/liter sugar (from a 20% wt/vol stock sterilized by passage through a 0.2-μm filter). Rich medium is Luria broth (LB: 5 g yeast extract, 10 g tryptone, 10 g NaCl in 1 liter of distilled deionized water with 2.5 mm CaCl2 added after autoclaving). Soft agar contains 8 g/liter Bacto agar and the hard agar used for plates contains 15 g/liter. Difco (Detroit) MacConkey lactose agar medium was prepared according to the manufacturer's instructions.
Strains and alleles:
As on previous occasions (Dykhuizen and Davies 1980; Dean et al. 1986, 1988; Dykhuizen et al. 1987; Dean 1989, 1995; Silva and Dykhuizen 1993; Dykhuizen and Dean 1994; Lunzer et al. 2002), strain DD320 (a K12 wild-type strain except for a small deletion spanning the lac operon) served as the genetic background for chemostat competition experiments. Strains were constructed using the generalized transducing bacteriophage P1(cml clr100) (Miller 1972). Strain TD2 carries a constitutive (lacI−) but otherwise wild-type K12 operon (Dean et al. 1986). Strain TD10 carries a constitutive (lacI−) derivative (Dean 1989) of the operon from strain ECOR 16, isolated from leopard scat (Ochman and Selander 1984). The lac operons of TD2 and TD10 were moved into a fresh isolate of DD320 to ensure that the genetic backgrounds of the competitors were as similar as possible.
Following transduction, strains were isolated on MD-lactose plates and purified on Bacto MacConkey-lactose agar plates. Spontaneous mutants resistant to the bacteriophage T5 (fhuA−) were isolated from soft LB agar overlays supplemented with 2.5 mm CaCl2 in the presence of excess T5 phage and purified by streaking for single colonies on LB plates. All strains were stored at −80° in 16% glycerol.
The chemostats used and the basic methodology of selection experiments can be found in Dykhuizen (1993). The medium for all chemostat experiments was MD, with methylgalactoside or lactulose, the sole sources of carbon and energy, added to a final concentration of 0.1 g/liter. This is sufficiently low to ensure that growth at steady state is limited by their availability. A mixture of 0.072 g/liter lactulose and 0.028 g/liter methylgalactoside was used for dual nutrient limitation, the proportions having previously been determined to allow a balanced polymorphism to be established between TD10 and TD2 (Lunzer et al. 2002). Isopropyl-β-d-thiogalactopyranoside (IPTG) was added at a concentration of 10−5 m to ensure full induction of the operons.
A peristaltic pump delivers fresh medium into the chemostat growth chamber at a dilution rate D = 0.33 hr−1 (10 ml/hr into a 30-ml growth chamber). Spent medium and cells exit through an overflow siphon. Filtered humidified air, entering the base of the growth chamber, mixes and aerates the culture before escaping through a vent. Temperature is controlled by immersing the chemostat growth chamber in a 37° water bath.
Chemostats were inoculated with pairs of strains, one sensitive to the phage T5 and the other resistant. Each was inoculated into separate side arm flasks containing MD with 2 g/liter lactulose and 10−5 m IPTG and grown to a Klett density of ∼150. Two-milliliter inoculates for the chemostats were prepared by mixing pure cultures (density ∼5 × 108 cells/ml) in the appropriate proportions as determined by Klett readings.
Long-term evolution experiments:
Long-term evolution experiments were conducted between pairs of strains: the first carries one lac operon and was sensitive to the bacteriophage T5 while the second carries a second lac operon and was resistant. The progress of competition was monitored by following the frequency of T5 resistance over time. The proportion of the population resistant to T5 was estimated daily by counting the number of colonies formed after plating the appropriately diluted samples in soft LB agar overlays. The colonies on the four plates with no T5 provide a total census, while those on four other plates with excess T5 phage provide the resistant count. Colonies are counted using an automated colony counter (Protocol Synoptics). Reference samples were taken daily and stored at −80° in 16% glycerol. The cultures were transferred to fresh chemostats every 6 days to prevent wall growth from dominating the selection dynamics. Experiments were terminated after the T5-resistant strain remained either >99% or <1% for >3 days, or when interest waned, whichever came first. Each experiment was numbered. Since each chemostat is numbered, when multiple chemostats were run at the same time, each run was named by the experiment number and the number of the initial chemostat, e.g., 329.10.
Short-term competition experiments:
Short-term competition experiments were used to determine fitnesses and reconstruct the equilibria seen in the long-term experiments according to standard methods (Dykhuizen and Hartl 1980; Dykhuizen and Dean 1994). The resistant strain was inoculated at low frequency (<10%) and the progress of the competition monitored by determining the proportion (R) of the population resistant to T5 twice daily. The selection coefficient, s, was estimated as the slope of a plot of the Loge ratio of population densities (R and 1 − R) against time in t (measured in D−1 generations), 3and where R0 and 1 − R0 are the population densities at time zero. Estimates of s and its standard error were obtained by least-squares linear regression (Snedecor and Cochran 1967). The fitness of a rare strain relative to a common strain (w) is simply 4
Dynamics of adaptation:
Lac polymorphic chemostats:
Long-term chemostat evolution experiments were established with TD2 and TD10 competing for a 72:28 mixture of lactulose:methylgalactoside and with one of the strains carrying a mutation in fhuA. The latter confers resistance to the bacteriophage T5 that serves as a selectively neutral genetic marker (Dykhuizen and Hartl 1980). The fate of the original lac polymorphism is then followed simply by monitoring changes in the frequency of T5 resistance, there being no genetic recombination in this experimental system (Levin 1981). A culture fixed for an allele at fhuA must also be fixed at lac for one or the other operon. Our operational criterion for loss of the polymorphism is that the frequency of T5 resistance remains >99% or <1% in three consecutive daily samples (>24 generations).
Following inoculation of a chemostat fed with a 72:28 mixture of lactulose:methylgalactoside, the frequency of T5 resistance in a mixed culture will converge on the predicted equilibrium frequency (62% if linked to the TD10 operon; 38% if linked to the TD2 operon; Lunzer et al. 2002) where it would remain indefinitely but for the appearance of advantageous mutations. Those having the same, or very similar, fitness effects and arising at approximately the same time in both strains will have no detectable effect on the frequency of T5 resistance. Those staggered in time will cause the frequency to spike before it returns to equilibrium. Rare advantageous mutations, or combinations thereof, will cause the frequency to shift permanently away from the original equilibrium. Marked changes in the frequency of T5 resistance seen during long-term adaptation in chemostats (Figure 2) are indicative of the accumulation of rare advantageous mutations. In some instances the fhuA polymorphism, and hence the lac polymorphism, is lost. In others the polymorphisms persist with fluctuating ratios for many hundreds of generations.
Lac monomorphic chemostats:
The impact of the lac polymorphism on the persistence of the fhuA polymorphism was investigated using cultures monomorphic at lac. The dynamics of T5 resistance are qualitatively similar to those in the presence of the lac polymorphism, with some fhuA polymorphisms persisting for many hundreds of generations while others are rapidly lost, usually within 200 generations (Table 1). The increased frequency of fhuA monomorphism is not significant (P = 0.41 by a one-tailed Fisher's exact test). Regardless of the presence or absence of the lac polymorphism those cultures retaining the fhuA polymorphisms are seemingly well separated from the monomorphic cultures by gaps of 140–180 generations without extinction. The dynamics of T5 resistance provide no evidence to suggest that a balanced polymorphism at lac influences the evolutionary fate of the linked selectively neutral fhuA polymorphism.
Chemostats with a single sugar:
The impact of environmental heterogeneity on the persistence of the fhuA polymorphism was investigated using cultures limited by a single nutrient, either lactulose or methylgalactoside. These experiments required using cultures monomorphic at lac because the outcome of pure scramble competition for a single limiting galactoside is rapid fixation of TD2 on lactulose and of TD10 on methylgalactoside (Lunzer et al. 2002). In all cases, single-nutrient limitation results in loss of the fhuA polymorphism within 320 generations. That not one of 20 cultures retains the fhuA polymorphism is significant with P = 0.003 by a one-tailed Fisher's exact test when compared to 8 of 22 cultures with mixed sugars. Hence, environmental heterogeneity in the form of dual nutrient limitation allows a selectively neutral polymorphism to persist far longer than limitation by a single nutrient does.
Characterizing evolved strains:
Evolution of cross-feeding:
Evolution of complex polymorphisms with multitrophic commensalistic cross-feeding often appears during adaptation to a glucose-limited environment (Helling et al. 1987; Rosenzweig et al. 1994; Treves et al. 1998). This type of evolution does not seem to play a major role in our system. First, the tiny colonies commonly associated with the evolution of acetate and glycerol cross-feeding have never been seen among the thousands of plated samples from our chemostat cultures. Second, linear fits to frequency data, obtained using purified chemostat isolates competing for single limiting nutrients, provide no evidence for the frequency-dependent selection characteristic of cross-feeding and other commensalistic interactions (Figure 3). Third, dual nutrient limitation generates frequency-dependent selection that shepherds purified isolates toward the original chemostat equilibrium (Figure 3), indicating that other mutants, if present in the original culture, are either similar in phenotype or rare. Fourth, fitnesses among multiple isolates from polymorphic cultures produced similar fitnesses and similar equilibria (Figure 3). Fifth, seven competitions between multiple isolates of the same strain (T5S isolates) from the same chemostat sample (chemostat 326.14B, Table 3B) provided no evidence of selection, and in the one instance where selection was observed the isolate was unambiguously deleterious on the mixed galactosides (chemostat 326.14B in Table 2).
In one mixed-sugar chemostat experiment (326.7) the frequency of the T5-resistant strain remained between 0.1 and 0.5%, rather than plummeting to <0.001% as is usually seen when cultures go monomorphic. Both initial strains contained the same lactose operon from TD10. Single colonies of the T5R (DD2269) and the T5S (DD2268) were isolated. The T5R strain was selected against at a rate of ∼0.18/generation on each of the pure sugars but maintained an equilibrium of ∼1% on the mixture of sugars. This suggests that the low-frequency T5R strain may be involved in cross-feeding and that the T5S strain is a generalist. Future experiments will address this issue.
The evolution of specialists:
One of the questions that motivated this set of experiments involves the probability of evolving generalists compared to specialists. Cultures retaining the fhuA polymorphism for ≥400 generations are expected to contain specialists. Table 3B contains the results from the four chemostats where the fhuA and lac polymorphism was retained. One T5R and one T5S strain was isolated from each chemostat, and their relative fitnesses were determined on pure lactulose and on pure methylgalactoside. Without exception, one strain was fitter on lactulose and the other fitter on methylgalactoside. The equilibrium values at the end of the long-term cultures and the equilibrium values reestablished from isolates are all very similar, confirming that two specialist populations dominate the equilibrium dynamics.
The fitnesses of strains isolated from chemostat 329.10 after 610 generations of adaptation to mixed galactosides (Table 3B) are, relative to each other, similar to those at the start (Table 3A). It is not the case that little adaptive evolution has occurred over the course of this experiment. Repeated dramatic changes in the frequency of T5 resistance (Figure 2) stand as testimony to a great deal of underlying adaptive evolution. What has not occurred in this experiment is any systematic resource partitioning or character displacement. Adaptive specialization is not an inevitable outcome of adaptation to mixed resources.
While not inevitable, adaptive specialization often occurs and can occur when an increase in fitness on one galactoside incurs a cost in fitness on the second. The examples listed in Table 3B are especially dramatic. Three of four long-term chemostats retaining the original polymorphism produced TD2 evolvants with high fitness on methylgalactoside instead of lactulose. These TD2 lineages switched specialization from lactulose to methylgalactoside. Because the original lac polymorphism was maintained in these chemostats, high fitness on lactulose became associated with TD10 lineages. Reciprocal exchanges in fitness illustrate dramatically the latent potential for specialization. They also demonstrate that the long-term outcome of adaptation is difficult to predict on the basis of current strain fitness.
These frequent switches in fitness suggest that mutations leading to specialization are happening randomly in cells and then this leads to the prediction that cultures monomorphic at lac will also evolve specialists. Isolating sensitive and resistant strains from long-term cultures that retain the fhuA polymorphism allows evolved specialists to be revealed (Table 3C). Analysis of one of the four remaining polymorphic fhuA cultures revealed the most highly specialized strains yet in a chemostat culture that were initially monomorphic for the TD10 lac operon.
Chemostat cultures growing on mixed galactosides may lose the fhuA polymorphism for either of two reasons. First, a rare but unambiguously advantageous mutation arises in one strain and sweeps to fixation. Second, a mutation causing a switch in resource specialization allows one strain to sweep away its competitor. The latter process gives rise to a cryptic polymorphism in an ostensibly monomorphic culture. The two outcomes are readily distinguished, however, for only in the second are there specialists of differing fitness.
Evolved specialists were purified from two TD10 cultures fixed for T5 sensitivity as follows. From each long-term culture, a single sample was used to inoculate two fresh chemostats, one containing lactulose and the other methylgalactoside. After 16 generations of enrichment (sufficient to produce a fivefold increase in the frequency of a rare strain favored by an s = 0.1) samples were streaked on plates, single colonies were isolated, and fhuA mutants were selected. Paired descendants, each isolated from the different enrichments of the same chemostat sample with one sensitive to T5 and the other resistant derivative, were then placed in competition for single galactosides. In all cases investigated, fitness estimates reveal that each descendant was fitter on the galactoside upon which it had been enriched, and that an equilibrium could be established on the 72:28 galactoside mix (Table 3D, Figure 4). That these cryptic specialists are not merely artifacts of the method of isolation was suggested by showing they could not be isolated from long-term cultures evolved on single galactosides.
If we assume that the initial mutations causing specialization are mutations of large effect and can happen randomly, then one expects them to happen in different fhuA backgrounds at a frequency related to the initial frequency of the two strains when the long-term chemostats were established. This ranged between 50:50 and 60:40. Thus one expects the frequency of chemostats where the fhuA polymorphism is retained to be between 2(0.5)(0.5) and 2(0.4)(0.6) or between 0.5 and 0.48. We obtained a frequency of 4/9 when the strains were polymorphic at the lac operon, suggesting all these cultures contained specialists. When the strains were monomorphic at lac, 4/13 strains retained the fhuA polymorphism. Again, this suggests that most cultures contained specialists, although the lower frequency is also compatible with the evolution of a few generalists.
Evolution at lac:
The lac operons of four evolvants were isolated from adaptations elsewhere in their genomes by transduction into the ancestral genetic background of DD320 (Table 4). Competition experiments reveal that after 335 generations of adaptive evolution the operon of strain 2261R (derived from the methylgalactoside specialist TD10R, but now a lactulose specialist; Table 3B) confers superior fitness to the operon of strain 2262 (derived from the lactulose specialist TD2, but now a methylgalactoside specialist; Table 3B) on both galactosides and on the 72:28 mixture. Hence, at least one mutation has occurred at lac that confers a higher fitness on both lactulose and methylgalactoside. This change cannot explain the evolution of the specialists. Changes elsewhere in the genome must be important. By contrast, 471 generations of adaptive evolution produced only a small adaptive difference in the TD10 lac operon of DD2266 (the methylgalactoside specialist, Table 3C) over the TD10 lac operon of DD2267R (the lactulose specialist, Table 3C). So in one pair the advantageous mutation was in the lactulose specialist and in the other pair it was in the methylgalactoside specialist. These results show that mutations conferring specialization arise elsewhere in the genome. It should be noted that only one copy of a duplication at lac need be recombined following generalized P1 transduction. Thus there could be specialization at the lac operon by duplication. But even if this has happened, duplications alone are insufficient to generate the observed increase in resource partioning. Because of this result, we decided to discontinue this approach and to investigate genetic change in these strains using genomic methods. The results from this study will be reported in a future article.
Criterion for invasion:
Assuming scramble competition remains the only mechanism by which competing strains interact, a new mutant can successfully invade a stable polymorphism either by displacing both residents or by displacing the resident of similar specialty (appendix). In the first case the balanced polymorphism between TD2 and TD10 is lost and the culture becomes monomorphic at fhuA. In the second case a new polymorphism is established between the mutant and whichever resident strain has the complementary resource preference. The culture will again become monomorphic at fhuA if the invading mutant has switched resource preference (because now both specialists of the new polymorphism are derived from the same ancestor). The culture will retain the fhuA polymorphism if the invading mutant retains the original resource preference so that it displaces its immediate ancestor. Preserving the fhuA polymorphism in the face of resource switching requires at least two mutations, one in each competing strain.
We predicted that a protected polymorphism of specialists adapting to a mixture of galactosides will rapidly become imbalanced, that all ecological diversity would be lost, and that a single phenotype would come to dominate each culture (Lunzer et al. 2002). This prediction is wrong. In every case investigated, except possibly one, a chemostat fed by two galactosides is inhabited by two specialists.
Evolution of specialists:
It is not the case that the original specialists (TD2 and TD10) are never lost. Five of nine cultures initially polymorphic at lac were rendered monomorphic. Loss of the original lac polymorphism might arise by either of two mechanisms: (1) a mutation of high fitness arises in only one strain and sweeps to fixation or (2) a mutation causes a switch in specialization sufficiently advantageous to drive the competitor, but not the mutant's immediate ancestor, to extinction. In either event the original lac polymorphism is lost, but only the first produces a monomorphic culture. The second mechanism produces a new polymorphism of specialists both derived from the same parental strain. The two hypotheses can be distinguished using a selection scheme designed to isolate cryptic specialists present in cultures that fixed the T5-sensitive strain. In all chemostat cultures tested, two distinct specialists were found. The evidence supports the second hypothesis.
Finding cryptic specialists in cultures fixed for one or the other lac operon demonstrates that the resource preferences represented by the original polymorphism are of no discernible consequence to the outcome of long-term adaptation. It also illustrates the rapidity with which resource partitioning can evolve and strongly suggests that the process involves no more than a few mutations of large effect. Adaptation need not proceed gradually.
That mutations of large effect can dominate the initial adaptive process is best illustrated by the phenomenon of resource switching, wherein two specialists swap resource preferences (Figure 5). Theory (appendix) predicts that invasion by a single mutant of switched resource preference will render the culture monomorphic at fhuA. To preserve the fhuA polymorphism both strains must switch preference more or less contemporaneously. The theoretical prediction that mutations causing switched resource preference must occur in both strains is assured in the absence of additional evolved frequency-dependent interactions (Figure 5). Furthermore, resource switching is not an especially rare phenomenon—it occurred in three of four cultures retaining the lac polymorphism (chemostat experiments 329.7, 303LT, and 324.14B in Table 3B).
The same types of specialists are selected even when the lac operon was initially monomorphic. However, in the results we presented evidence that one chemostat culture (326.7) seemed to contain a generalist and a very low-frequency strain probably involved in cross-feeding. Since the frequency of fhuA monomorphic chemostats is a little higher than expected when the lac operon is initially monomorphic, it is possible that a generalist has arisen, at least initially, in a few of these chemostat cultures.
Our discovery of resource switching contradicts the oft-illustrated metaphor of adaptation as an incremental peak-climbing process mapped onto a mountainous fitness landscape. The underlying assumption that adaptation proceeds by many mutations, each of infinitesimally small effect, is clearly incorrect in this instance. Real mutations produce discrete phenotypic effects, and mutations producing large discrete phenotypic effects readily permit strains to leap across maladaptive valleys from one adaptive peak to another. Envisioning adaptation as an incremental peak-climbing process is misleading whenever the phenotypic effects of mutations are not scaled appropriately to the ruggedness of the landscape. The caveats of this experiment are that the sugars used are not sugars to which E. coli is adapted and there is no evidence that the differences in the operons of TD2 and TD10 are caused by anything but chance. However, our experiments serve as a model for the initial stages of adaptation in novel environments. Longer chemostat runs could simulate the fine tuning of adaptation.
Mechanisms of adaptation:
Three mechanisms (Elena and Lenski 2003) capable of producing specialists are: (1) mutation accumulation, where neutral mutations accumulated in one environment prove deleterious in the second, (2) independent specialization, where beneficial mutations accumulated in one environment prove selectively neutral in the second, and (3) antagonistic pleiotropy, where beneficial mutations accumulated in one environment prove deleterious in the second. Note that these mechanisms are not necessarily mutually exclusive and that each may play, to a greater or lesser extent, a role in the evolution of specialists. Thus far, efforts to distinguish these mechanisms in strains subjected to long-term adaptation have proven difficult (MacLean and Bell 2002; Elena and Lenski 2003).
In our experimental design the contribution of mutation accumulation to the evolution of specialization is prevented precisely because mutations selectively neutral for one galactoside and deleterious for the second are immediately purged by selection. Of the other two mechanisms, independent specialization predicts the existence of generalists, double mutants carrying alleles each beneficial toward each galactoside. Individual advantageous mutations would first occur in different cells and these lineages would be swept toward fixation, temporarily creating specialists. Then the other mutation would occur in one of the specialists and intense selection should rapidly sweep this generalist to fixation (Dykhuizen and Davies 1980). Yet most cultures had only specialists present. Furthermore, the region where polymorphisms are protected by frequency-dependent selection represents only 18% of that available given the observed fitnesses (Figure 6). That no generalists could be isolated from the seven populations analyzed given an 82% target region would seem sufficient to reject the hypothesis of independent specialization. By contrast, antagonistic pleiotropy provides a ready explanation for the negative correlation between fitness on lactulose and fitness on methylgalactoside, the switching of specialization from one resource to the other, and the absence of generalists (Figure 6).
Our experimental design, like others, does not distinguish between two types of antagonistic pleiotropy. The first is a direct trade-off that increases fitness on one resource and decreases fitness on the other. The second is an indirect trade-off that increases fitness to a greater extent on one resource than on the other. This evolvant is a generalist, in the sense that it is fitter than either ancestor on either resource, and is expected to sweep to fixation. However, a selective sweep is avoided if, at the same time, the competing strain increases its fitness to a greater extent on the second resource than on the first. The intensified selection between the evolvants on the pure galactosides suggests increased specialization, while the ability of each evolvant to displace either progenitor on either galactoside suggests the evolution of generalists. Unfortunately, the appearance of increased generalism in competitions between evolvants and their progenitors is compromised by the likely appearance of background mutations having nothing to do with resource partitioning per se and everything to do with adaptation to the chemostat environment in general. The only reliable means to determine whether increased specialization is caused by a direct or an indirect trade-off is to identify all mutations in the evolvants and determine their phenotypes.
The importance of the lac operon:
Transducing evolved operons into DD320 allows adaptation at lac to be assessed independently of other mutations accumulated in the genetic background. The fitnesses of the DD2261R operon with respect to DD2262 (an evolved TD10 operon with respect to an evolved TD2 operon) are quite different from those in the original ancestral strains (Table 4). This provides direct evidence that mutations in (or around) lac contribute to adaptation. The gains in fitness by the TD10 operon on both galactosides are characteristic of the evolution of a generalist. Not only is there no evidence of specialization at lac, but also alone the fitnesses conferred by the operons are unable to explain their maintenance as a balanced polymorphism. Nor can the de novo appearance, in a culture initially monomorphic for the TD10 operon, of specialists DD2266 and DD2267R be explained by the weak selection associated with their lac operons. Other loci must also determine resource preference.
Published examples of trade-offs following adaptation at lac include: (1) constitutive expression that becomes deleterious during competition for glucose (Novick 1958; Horiuchi et al. 1962, 1963), (2) constitutive expression that becomes strongly deleterious when the concentration of lactose is suddenly raised to excess (Hartl and Dykhuizen 1978; Wilson et al. 1981; Ghazi et al. 1983), and (3) increased permeability of OmpF porins to lactose, which renders cells more susceptible to antibiotics (Zhang and Ferenci 1999). We anticipate finding similar trade-offs, possibly at the same loci, once our evolved strains are exposed to novel environments. We also expect, on the basis of the results of Notley-McRobb and Ferenci (2000) who studied adaptation in E. coli to a glucose-limited environment, that specialists on methylgalactoside will commonly carry loss-of-function mutations at mglD/O that render constitutive a sugar transport system capable of efficiently transporting methylgalactoside. It is conceivable that the biochemical basis for the observed trade-off lies in an interaction between the mgl and lac transport systems when both are overexpressed.
To suggest that all changes in the frequency of T5 resistance during the course of long-term adaptation occur in only a handful of genes controlling galactoside uptake and metabolism, especially when the remainders of the genomes of 3 billion cells in the chemostat populations are immediately available for mutational improvement, seems highly implausible. Indeed, increased adaptive specialization is not an inevitable outcome of adaptation to mixed resources as the 610 generations of experiment 336/329.10 show (Table 3). Mutations, perhaps at different loci but of similar selective advantage, arise continuously in both strains so that the polymorphisms at lac and fhuA are perpetuated even as the genetic backgrounds churn adaptively. An occasional mutation of exceptionally high adaptive value, or a chance spacing in time between otherwise frequent mutations of similar fitness, may permit one or another strain to be swept toward fixation. Nevertheless, we do not see this over the time span of these experiments.
The prediction that the lac polymorphism would be lost was based on two assumptions: that advantageous mutations are rare and that they would likely shift the narrow zone of coexistence away from the resource supply. However, it appears that advantageous mutations are common and so closely spaced in time that adaptive sweeps through the entire culture are rare. This, and the trade-off between lactulose and methylgalactoside specializations, contributes to perpetuating linked polymorphisms at lac and fhuA. The last selective sweeps through entire cultures occur between 250 and 300 generations. Chemostats emerging polymorphic at fhuA from this period remain polymorphic for up to 400 generations more. Perhaps as a consequence of further specialization selective sweeps through the entire culture are now more difficult to achieve. This suggests that the populations are in the process of, or perhaps have completed, transiting from a region of latent potential for ecological specialization to a region of enforced ecological specialization. If true, selective sweeps are now restricted to the clonal ecotypes in which the fitter mutations arise. At this stage one might see the more typical picture of incremental increases in fitness as in hill climbing. Only recombination into another ecotype will allow a fitter allele to spread further. Restricted recombination might help promote the preservation of linked neutral polymorphisms among natural populations of E. coli.
At equilibrium TD2 and TD10 grow at the same rate, D, the dilution rate of the chemostat, A1where μi is the growth rate of strain i, Yj is the yield coefficient with resource j, the ∑ij are a collection of terms describing galactoside uptake and catabolism, and LU and MG are the concentrations of lactulose and methylgalactoside at equilibrium (Lunzer et al. 2002). Solving yields A2and A3since fitness on a single resource is given by (Lunzer et al. 2002). Substituting these into the growth equation for a new mutant (mut), A4yields A5and A6which are equivalent. The criterion for successful invasion by the mutant of an established TD2-TD10 polymorphism is simply A7
Because the number of strains coexisting in a balanced polymorphism cannot exceed the number of limiting resources when the only interaction is scramble competition (Stewart and Levin 1973; Tilman 1982; Grover 1997), a successful invader must drive either one or both residents to extinction. Extinction of resident i is assured if wmuti.A > 1 and wmuti.H > 1 since a balanced polymorphism cannot form between it and the invading mutant. Resident j will form a polymorphism with the new mutant only if wmutj.A > 1 and wmutj.H < 1.
Additional criteria are necessary to predict the outcome of competition when the fitnesses allow all three pairwise polymorphisms to exist (e.g., balanced polymorphisms are formed between the pairs with , with , and with ). TD10 and TD2 form a balanced polymorphism with w102.MG > 1 and w102.LU < 1. From Equation A5, a mutant will invade if A8
The condition for TD10's survival is given by its ability to invade a polymorphism established between the mutant and TD2. This can be written out by inspection, A9and rearranged to show that it is identical to Equation A8 when wmut2.MG < 1 and wmut2.LU > 1. Evidently, TD10 will survive when the successful invader is a lactulose specialist.
The condition for TD2 to survive is A10
Noting and allows Equation A10 to be rearranged, A11
Equation A11 contradicts Equation A6 and TD2 will be excluded when the successful invader is a lactulose specialist. By a similar argument it can be shown that when the successful invader is a methylgalactoside specialist (wmut2.MG > 1 and wmut2.LU < 1), TD2 survives while TD10 is excluded. We conclude that, in the absence of other interactions, successful invaders exclude residents of similar specialty.
We thank Stacy Satornino for excellent technical assistance and Ben Kerr, Dusty Brisson, and Dan Stoebel for valuable criticism. This work was supported by National Institutes of Health grants to A.M.D. and D.E.D.
Communicating editor: J. Bergelson
- Received October 24, 2003.
- Accepted April 20, 2004.
- Genetics Society of America