Oscillatory dynamics in a bacterial cross-protection mutualism

Significance

Enzymatic deactivation of antibiotics is a cooperative behavior that can allow resistant cells to protect sensitive cells from antibiotics. The prevalence of this mechanism of antibiotic resistance in clinical isolates and in soil bacteria makes it important both clinically and ecologically. Here, we demonstrate experimentally that two populations of resistant bacteria can form a cross-protection mutualism in a two-drug environment, allowing the coculture to survive high antibiotic concentrations at which neither of the two strains can survive alone. Moreover, we find that daily growth-dilution cycles result in large oscillations in the relative abundances of the two strains, thus demonstrating that mutualisms can display striking dynamical behavior.

Abstract

Cooperation between microbes can enable microbial communities to survive in harsh environments. Enzymatic deactivation of antibiotics, a common mechanism of antibiotic resistance in bacteria, is a cooperative behavior that can allow resistant cells to protect sensitive cells from antibiotics. Understanding how bacterial populations survive antibiotic exposure is important both clinically and ecologically, yet the implications of cooperative antibiotic deactivation on the population and evolutionary dynamics remain poorly understood, particularly in the presence of more than one antibiotic. Here, we show that two Escherichia coli strains can form an effective cross-protection mutualism, protecting each other in the presence of two antibiotics (ampicillin and chloramphenicol) so that the coculture can survive in antibiotic concentrations that inhibit growth of either strain alone. Moreover, we find that daily dilutions of the coculture lead to large oscillations in the relative abundance of the two strains, with the ratio of abundances varying by nearly four orders of magnitude over the course of the 3-day period of the oscillation. At modest antibiotic concentrations, the mutualistic behavior enables long-term survival of the oscillating populations; however, at higher antibiotic concentrations, the oscillations destabilize the population, eventually leading to collapse. The two strains form a successful cross-protection mutualism without a period of coevolution, suggesting that similar mutualisms may arise during antibiotic treatment and in natural environments such as the soil.

Elucidating the roles of cooperative and competitive ecological interactions is critical to understanding community formation, function, and stability. Cooperative interactions can occur when one individual or group provides nutrition, protection, or mobility to another (1, 2), and whether an interaction is cooperative or competitive can depend on the environment (3). Reciprocal cooperative interactions, or mutualisms, arise when two partners cooperate to increase each other’s fitness (1, 3–7). The ecological roles of mutualisms include stabilizing community structure and enabling survival in challenging environments (2, 5, 8, 9).

Understanding microbial mutualisms will help clarify the important role they play in diverse natural communities. So far, the majority of studies of microbial mutualisms have focused on cross-feeding, the mutually beneficial exchange of metabolites (10–15). Studies involving cross-feeding amino acid auxotrophs (10–12) found that the division of labor within the mutualism can increase the fitness of cross-feeding cocultures compared with a self-sufficient strain (10). The greatest gains seem to emerge when amino acids are costly to produce (11). These mutualisms can even form immediately when complementary strains encounter each other and without a period of coevolution, although allowing time for adaptation may increase their resilience (12).

Another important microbial interaction involves the protection of one microbe by another, potentially enabling the formation of cross-protection mutualisms. Protective interactions are especially interesting in the context of antibiotic resistance, where they can affect the success of antibiotic treatment and influence the evolution of antibiotic resistance (16–20). For example, the well-known inoculum effect refers to the observation that inhibiting bacterial infections composed of more cells can require disproportionally more antibiotic (21). Mechanisms that allow bacteria to help each other survive antibiotic exposure include the formation of protective structures like biofilms, coordination of a group response (via quorum sensing), or production of enzymes that modify and deactivate the target antibiotic (16, 17).

The production of resistance enzymes is a common mechanism of antibiotic resistance (22) and can be considered cooperative because resistant cells can protect sensitive cells by reducing the concentration of active antibiotics in the environment, allowing their sensitive neighbors to survive an otherwise lethal exposure (19, 23–28). Multiple examples of such protective interactions have been observed in clinically relevant strains where pathogenic, but otherwise sensitive, strains were protected against antibiotics by neighboring microbes (23, 29, 30). It is conceivable that the cooperative nature of antibiotic deactivation could allow a bacterial community composed of multiple resistant bacterial strains to survive in a multidrug environment, even if some strains were not individually resistant to each drug present. In this work, we investigate whether two bacterial populations can form a cross-protection mutualism in a two-drug environment and probe the dynamical properties of this mutualism.

Results

To explore the possibility of a microbial cross-protection mutualism, we developed a model system that consists of two strains of Escherichia coli, each of which is resistant to a different antibiotic (Materials and Methods). Each strain produces an enzyme that deactivates one of the two antibiotics in the environment, thereby potentially allowing the other strain to survive in the presence of the antibiotic to which it is sensitive. The first strain is ampicillin-resistant as a result of a plasmid carrying a gene encoding a β-lactamase enzyme, which deactivates ampicillin (Fig. 1A). This enzymatic deactivation occurs both in the periplasmic space of the cell and in the extracellular medium (28, 31) (SI Appendix, Fig. S1), thus leading to immediate benefits to the sensitive partner (19, 32). The second strain is chloramphenicol-resistant as a result of a plasmid carrying a gene encoding the chloramphenicol acetyltransferase type I enzyme, which deactivates the antibiotic chloramphenicol (33, 34) (Fig. 1A). Although this enzymatic deactivation of chloramphenicol occurs inside the cell (28) (SI Appendix, Fig. S1), diffusion of chloramphenicol between the media and the interior of resistant cells may cause the extracellular concentration of the antibiotic to decrease enough to allow the growth of sensitive cells (28). These plasmids do not conjugate, and we did not observe horizontal gene transfer in our experiments (SI Appendix, Fig. S2). Given that each strain has the potential to provide protection to the other strain, we hypothesized that the two strains could help each other survive in an environment containing both antibiotics (Fig. 1A).

Fig. 1.

Two strains of resistant bacteria can form a mutualism in a multidrug environment by protecting each other from antibiotics. (A) In an environment containing the antibiotics ampicillin and chloramphenicol, a mutualism forms between bacteria producing a β-lactamase enzyme (which protects against ampicillin) and bacteria producing a chloramphenicol acetyltransferase enzyme (which protects against chloramphenicol). (B) In our serial dilution experiments, we periodically diluted microbial cultures into fresh growth media, replenishing the supply of nutrients and antibiotics. We determined the size of each subpopulation by combining spectrophotometry measurements of the total culture density together with flow cytometry measurements of the relative abundances of each subpopulation. (C) A coculture of the two strains can survive above the concentrations at which the individual strains survive alone, indicating that the two populations form an obligatory mutualism. The black lines indicate the region beyond which only the coculture can survive. (D) An ampicillin-resistant monoculture can survive in high concentrations of ampicillin but cannot survive alone in chloramphenicol concentrations above 2.2 μg/mL. (E) Similarly, a chloramphenicol-resistant monoculture can survive in high concentrations of chloramphenicol but cannot survive alone in ampicillin concentrations above 2 μg/mL. The antibiotic concentrations in D and E are the same as in C. The populations shown in C–E were subject to five daily dilution cycles at 100×.

The Two Strains Form a Cross-Protection Mutualism.

To probe the extent to which the partnership between the two strains enables survival, we cocultured the strains in the presence of varying concentrations of the antibiotics ampicillin and chloramphenicol (Fig. 1B and Materials and Methods). We found that the cocultured strains were able to survive across a wide range of antibiotic concentrations for the duration of a 10-day experiment when subject to daily 100× dilutions into fresh media (Fig. 1C). This range of antibiotic concentrations included concentrations at which neither of the two strains could survive alone (Fig. 1 D and E and SI Appendix, Figs. S3 and S4). In fact, survival of the coculture extended to concentrations fourfold larger than the inhibitory concentrations of the sensitive strains (black lines in Fig. 1 C–E), and we did not observe a shift in the inhibitory concentration of either strain over the course of the experiments. These two antibiotic resistant strains therefore form an effective obligatory mutualism.

All cocultures that form a successful mutualism (Fig. 1C) must survive multiple growth-dilution cycles, reaching a high population density day after day (SI Appendix, Figs. S4 and S5). Although cocultures grown at high antibiotic concentrations may survive one growth-dilution cycle, they collapse shortly thereafter, unable to survive subsequent growth-dilution cycles (SI Appendix, Figs. S4 and S5). Thus, initial growth of a coculture does not necessarily mean that the mutualism is stable: To survive indefinitely over many growth-dilution cycles, the shifts that occur in the relative abundances of the two strains over the course of a day must be sustainable.

Growth-Dilution Cycles Give Rise to Strong Oscillatory Dynamics.

To better understand the underlying subpopulation dynamics of this mutualism, we tracked the subpopulation densities of the two strains over many growth-dilution cycles. Because the antibiotic resistance plasmids also encoded fluorescent proteins, we could use flow cytometry to measure the relative abundances (ratio of the subpopulation sizes) of the two strains (Materials and Methods and SI Appendix, Fig. S6). Strikingly, the population dynamics of the coculture in the region where there is an obligatory mutualism revealed oscillations in the abundances of the two partner strains (Fig. 2), although the total population density at the end of each cycle remained relatively steady. The oscillations in the relative abundances typically had a period of 3 days and spanned four orders of magnitude (Fig. 3A). Because these oscillations occurred with a period (3 days) longer than the period of the daily dilution (1 day), they were not a trivial consequence of the daily growth-dilution cycle.

Fig. 2.

The subpopulation sizes of the two mutualists oscillate over a broad range of antibiotic concentrations, both inside and outside the region where the mutualism is obligatory. Many of the mutualisms settled into apparent period three oscillations, which had a period (3 days) longer than the period between successive exposures to antibiotics (1 day). These oscillations were substantial in magnitude with the relative abundances of the two mutualists changing by as much as 10⁴-fold. The red region of each subplot represents the size of the ampicillin-resistant subpopulation, and the blue region represents the size of the chloramphenicol-resistant subpopulation. The green line delineates the range of antibiotic concentrations above which neither strain can survive alone. In this experiment, the cocultures were diluted by 100× every 24 h into fresh media supplemented with antibiotics.

Fig. 3.

Oscillations in the relative abundances of the two mutualists are driven by the periodic dilution of the coculture into fresh media containing antibiotics. (A) Under periodic exposure to antibiotics, the relative abundances of the two mutualists in the coculture oscillate by as much as 10⁴-fold. (B) Under (pseudo)continuous exposure to antibiotics, the relative abundances of the two mutualists converge to an equilibrium ratio, and we do not see any sign of the large amplitude oscillations present in A. To transition from the periodic regime (A) to the (pseudo)continuous regime (B), we decreased the time between consecutive dilutions from ΔT = 24 h to ΔT = 1 h and the dilution strength from 100× to 1.2×. The death rate due to dilution, ln(dilution strength)/ΔT, is equivalent in both regimes. The detection limits on our flow cytometer were on the order of N_amp/N_chlor = 10^±3. In these experiments, the concentrations of ampicillin and chloramphenicol in the fresh media were 10 μg/mL and 5.1 μg/mL, respectively. See Fig. 2 and SI Appendix, Fig. S10 for the population sizes corresponding to these trajectories.

To see why this cross-protection mutualism might exhibit oscillations, we consider the dynamics of the two bacterial populations and the two antibiotics over the course of one growth-dilution cycle. Suppose that initially one of the strains is significantly more abundant than the other. This first strain rapidly deactivates its target antibiotic, allowing the second strain, its mutualistic partner, to start growing. The first strain can only start growing once the second strain is abundant enough to deactivate the second antibiotic, which has so far inhibited growth of the first strain. However, at this point, it is already too late for the first strain to catch up with its partner, which is far more abundant. Hence, the mutualism exhibits oscillations because the strain that is more abundant at the beginning of a growth cycle might end up less abundant at the end of that cycle. Overall, this logic explains how growth-dilution cycles can give rise to oscillatory dynamics. Of course, many of the oscillations that we observe experimentally seem to have a period of 3 days (rather than 2 days); this period arises because of asymmetries in how the two strains grow in the presence of and deactivate the antibiotics.

A simple mechanistic model incorporating the basic time dynamics of the two strains and the two antibiotics can indeed yield an obligatory mutualism with period three oscillations (SI Appendix, Fig. S7). Although the strains used in our experiments have slightly different growth rates in the absence of antibiotics, our model suggests that this difference is not necessary for the oscillations to occur (SI Appendix, Fig. S7). Consistent with our intuitive understanding of the population dynamics, this model suggests that the oscillations arise because of the daily dilutions (SI Appendix, Fig. S8), and a simulated bifurcation diagram displays limit cycle dynamics of varying oscillation periods interspersed with regions of chaos (SI Appendix, Fig. S9). In particular, our model predicts that in a continuous culture experiment in which the antibiotics are continuously added (and cells continuously removed), there will be no oscillations in the population abundances (SI Appendix, Fig. S8), and instead the ratio between the two strains should approach a stable equilibrium.

To test this prediction, we performed a (pseudo)continuous experiment by diluting the cocultures every hour by a small amount (1.2×) into fresh medium supplemented with ampicillin and chloramphenicol (10 μg/mL and 5.1 μg/mL, respectively), mimicking a chemostat operating at a fixed dilution rate; the dilution strength and cycle time in these experiments were chosen so that the effective daily dilution rate would be equivalent to that in our daily dilution experiments. We found that with frequent dilution the coculture was able to form a stable mutualism, with the strains reaching a stable equilibrium ratio without oscillations: Over the timescale of the experiment, cocultures starting at different initial relative abundances converge toward a stable equilibrium ratio (Fig. 3 and SI Appendix, Figs. S10 and S11). Moreover, a stable equilibrium was even observed for higher dilution rates (SI Appendix, Fig. S12), and we could also remove the oscillations by reducing the dilution factor in daily dilution experiments (SI Appendix, Figs. S13 and S14). Generally, when periodic forcing due to the growth-dilution cycles was weak, oscillations did not emerge over the timescale for potential oscillations (the time between subsequent growth-dilution cycles). These results argue that the growth-dilution cycles drive the oscillations we observe between the two strains in the mutualism.

A major question concerning the observation of oscillations is whether they are robust, transient, or neutrally stable. Robust oscillations typically converge to a limit cycle oscillation with a characteristic period and amplitude, whereas transient oscillations decay to an equilibrium. In contrast, in neutrally stable oscillations, both the amplitude and period may depend on the initial conditions (e.g., the classic Lotka–Volterra model of predator–prey dynamics displays neutral oscillations; ref. 35).

To probe the nature of the oscillations observed in our experiments, we initiated cocultures at a wide range of starting fractions. Consistent with our oscillations being a true limit cycle, we found that a wide range of starting conditions all led to apparent period three oscillations with similar amplitudes (Fig. 4A). Although this oscillation pattern resembles a period three limit cycle, the combination of experimental noise and time required to converge to the limit cycle make it difficult to unambiguously distinguish period three from longer periods (for instance, a period six limit cycle could resemble a period three limit cycle). These stable, limit cycle oscillations are present in a range of antibiotic concentrations, both inside and outside the region where the mutualism is obligatory (Fig. 2 and SI Appendix, Figs. S15 and S16).

Fig. 4.

In the presence of daily 100× dilutions, the two mutualists undergo stable limit cycle oscillations. (A) We found that the relative abundance of the ampicillin-resistant subpopulation with respect to the chloramphenicol-resistant subpopulation settled into period three oscillations, even when test cultures started with different subpopulation abundances. The population size of these limit cycle trajectories remained close to the carrying capacity. (B) If the population size of each mutualist is sufficiently large, then the two mutualists settle into stable oscillations (colored trajectories); otherwise, the mutualism collapses (gray trajectory). (C) A discrete-time framework of an obligatory mutualism featuring a “healthy state” characterized by a limit cycle (white region) and a “collapsed state” (gray region). A boundary (called a separatrix) separates the sets of subpopulation compositions that converge to each state. (A and B) Experiments were carried out in an environment inside the region of obligatory mutualism (100× dilution strength, 24-h dilution cycle, 10 μg/mL ampicillin, and 5.1 μg/mL chloramphenicol). The blue trajectories present in A and B are the same time series. Open circles indicate the initial population composition for each trajectory.

Oscillations Can Destabilize the Mutualism.

Presumably sufficiently large densities of each partner are required for the mutualism to survive. To probe the subpopulation densities necessary for establishing a mutualism at a given set of antibiotic concentrations, we measured trajectories of cocultures starting at a broad range of initial conditions, including extreme relative subpopulation abundances and very low cell densities. When the initial population size of either mutualist was too small, the mutualism could not converge to the limit cycle and ultimately collapsed (Fig. 4B).

Based on our measurements of the population dynamics, we propose a simple time-discrete framework to provide a qualitative explanation of the population dynamics of the mutualism in the presence of periodic antibiotic exposure (Fig. 4C). In this framework, depending on the initial population composition, the mutualism either converges to a limit cycle or collapses. The boundary that separates the set of trajectories that map to each fate is called the separatrix. A deteriorating environment (either via increased antibiotic concentrations or via increased dilution rates) could bring the limit cycle and the separatrix closer together, increasing the likelihood that the oscillations would push the population composition of the mutualism too far to one side.

To test the intuition provided by this framework, we mapped the separatrix in a range of different chloramphenicol concentrations (SI Appendix, Figs. S17 and S18). Because stochasticity causes populations of similar compositions to have different fates, thus blurring the separatrix, we took a probabilistic approach in determining whether a coculture with a given relative abundance of the two mutualists would collapse (Fig. 5 and SI Appendix, Figs. S19 and S20). We found that at relatively low chloramphenicol concentrations, cocultures can successfully recover from the extreme changes in the relative abundance of the two mutualists brought on by the oscillations. Upon increasing the concentration of chloramphenicol, trajectories passed closer to the separatrix and became more erratic, losing their characteristic period three oscillation (Fig. 5 and SI Appendix, Figs. S21–S23). At high chloramphenicol concentrations, the probability of collapse increases rapidly when the relative abundance of ampicillin-resistant cells falls below approximately 1:100 (Fig. 5 and SI Appendix, Figs. S19 and S20). Thus, one extreme of the oscillation is more susceptible to collapse: When the population of ampicillin-resistant strain in the mutualism becomes too small, it can no longer protect the chloramphenicol-resistant strain, leading to insufficient growth on the next day and ultimately to the collapse of the coculture. It is interesting to note that, even at the higher chloramphenicol concentrations, there exists an unstable fixed point that in the absence of noise would allow for indefinite survival (SI Appendix, Fig. S24); in that sense, the presence of oscillations decreases the range of antibiotic concentrations in which the coculture can survive.

Fig. 5.

The large oscillations in the relative abundance of the two strains (N_amp/N_chl) destabilize the mutualism in harsher environments, causing the population to collapse. At low chloramphenicol concentrations (7.6 μg/mL), cocultures can successfully recover from the extreme changes in the relative abundance caused by the oscillations. However, at higher chloramphenicol concentrations (17.1 and 38.4 μg/mL), the probability of collapse (portrayed by the gradient in the background of each subplot) increases rapidly when the relative abundance of the ampicillin-resistant population falls below 1:100, causing the oscillations to destabilize the mutualism. One extreme of the oscillation is more susceptible to collapse: When the population of ampicillin-resistant strain in the mutualism becomes too small, it can no longer protect the chloramphenicol-resistant strain from ampicillin, ultimately leading to the collapse of the coculture. Red circles denote the last time point when the overall size of the population remained above the detection limit (see SI Appendix, Fig. S21 for the time series of the population sizes of the shown trajectories). In these experiments, the media contained 10 μg/mL ampicillin and the specified amount of chloramphenicol.

Invasion Can Disrupt the Oscillatory Dynamics.

Up to now, we have shown that the growth-dilution cycles can produce robust oscillations in our cross-protection mutualism, and explored the consequences of these oscillations on the stability of the mutualism. However, because natural populations interact not only with the environment, but also with one another, we sought to understand whether the observed oscillations can persist robustly in the face of invasion by other populations. An important source of potential invaders arises from within the mutualism itself (8). A mutation or horizontal gene transfer event could create a mutant strain resistant to both antibiotics. Because this double-resistant strain does not require protection from the mutualistic strains to survive, it may be able to invade the mutualism and destabilize the limit cycle.

Indeed, after we added a small number of double-resistant cells (Fig. 6 and Materials and Methods) to the mutualistic cocultures on the sixth growth cycle, the double-resistant strain quickly displaced the ampicillin-resistant population, presumably because the cost of the chloramphenicol resistance plasmid is less than the growth inhibition induced by the chloramphenicol in the environment. The double-resistant strain and the chloramphenicol-resistant strain remain, with the double-resistant cells protecting the chloramphenicol-resistant cells from ampicillin, analogous to the situation described in ref. 19 where ampicillin-resistant cells protect sensitive cells. In this situation when there is no longer a cross-protection mutualism, there are no oscillations between the double-resistant and chloramphenicol-resistant strains (Fig. 6).

Fig. 6.

A double-resistant strain can invade the mutualism and cause the oscillations to vanish, illustrating that the existence of oscillations depends on how resistance is allocated in the microbial population. (A) A double-resistant strain (carrying both resistance plasmids) does not require protection from the mutualistic strains to survive. (B) In the absence of the invader, the ampicillin- and chloramphenicol-resistant subpopulation oscillate. (C) After introducing the double-resistant strain at the beginning of the seventh cycle (dashed gray line) to a replicate culture of B, the double-resistant invader established in the microbial population, outcompeting the ampicillin-resistant subpopulation and removing the oscillations. (B and C) The concentrations of ampicillin and chloramphenicol were 10 μg/mL and 7.5 μg/mL respectively.

Discussion

In this work, we have shown that a pair of antibiotic-deactivating E. coli strains can successfully form an obligatory mutualism in a multidrug environment. Because the capacity to form the mutualism only depends on the production of resistance enzymes, the mutualism does not require a period of coevolution as long as the enzymes are readily expressed. The cooperative nature of antibiotic deactivation could allow a sensitive strain (SI Appendix, Fig. S25 and Materials and Methods) to use the two mutualists for protection. Indeed, we find that sensitive cells naturally emerge in our experiments as a result of plasmid loss; however, these cells only manage to attain a small portion of the total population size and do not disrupt the mutualism (SI Appendix, Figs. S26 and S27). A key contribution to the robustness of the mutualism is that resistant cells preferentially protect themselves against the antibiotic (19).

The persistent survival over many dilution cycles that we observed suggests that this obligatory mutualism can survive indefinitely over a broad range of antibiotic concentrations. However, we note that while studying the stability of the mutualism over time, we found that some cocultures only survive transiently in the presence of the antibiotics, eventually collapsing despite reaching a high population density during earlier dilution cycles (Fig. 2 and SI Appendix, Fig. S5). This finding is reminiscent of a previously published report exploring the possibility of a bacterial mutualism between an ampicillin-resistant strain and a kanamycin-resistant strain, where the authors found that the coculture was viable only for the first day of growth (36), presumably because kanamycin deactivation was not sufficiently cooperative. This distinction highlights the need to explore conditions in which mutualisms are stable over the long term, which necessitates understanding the underlying population dynamics.

Ecologists have been long fascinated by the mechanisms that can give rise to oscillations in population abundances (37–40). Especially intriguing are cases where population oscillations have a different period than seasonal variations and, thus, cannot be easily explained by changes in the environment (41–43). For instance, the predator–prey oscillations of the Canada lynx and the snowshoe hare have a period of approximately 10 years and are thought to arise principally from the dynamics of predation rather than from seasonal variations in some abiotic factor (e.g., temperature) (42). In contrast, historical data on the dynamics of the measles virus exhibit strong seasonal dependence because major outbreaks typically occurred in the winter (43); however, the time between outbreaks (2 years) was not equal to the period of the seasons (1 year) (43). In a study of larval, pupal, and adult beetles, experimental control over periodically imposed adult mortality and recruitment rates resulted in cyclical population dynamics (44), although again the period of population oscillations was longer than the periodic forcing in the experiment. Interestingly, even in constant environments, the presence of oscillations can depend on the parameter regime: The population dynamics in a rotifer-algae (predator–prey) system could be modulated in the laboratory by changing the chemostat dilution rate and nutrient inflow rate, shifting dynamics from coexistence at an equilibrium ratio to oscillations (45).

Consumer–resource interactions (e.g., predator–prey or host–parasite) often exhibit oscillatory dynamics in both static and periodically varying environments, as evidenced by the large number of known examples in which this happens; however, mutualisms are typically thought to stabilize population abundances over time. Previous observations of sustained oscillations in mutualisms have been limited mostly to theoretical studies (6, 7, 46–48). Here, we have shown that growth-dilution cycles in the environment can give rise to oscillations in the relative abundances of the two mutualists, with an oscillation period longer than the time between consecutive growth-dilution cycles. This observation of oscillations in a cross-protection mutualism leads one to speculate whether periodic forcing can induce oscillations in other types of mutualisms as well. Based on preliminary modeling (SI Appendix), we hypothesize that the ability of periodic forcing to induce oscillations may be more characteristic of cross-protection mutualisms than of cross-feeding mutualisms. Consistent with this idea, a number of studies of laboratory cross-feeding mutualisms observed that the relative abundance of the two mutualists converged to a fixed point (12, 15, 49). More broadly, metabolic codependence (although not in the form of a cross-feeding mutualism) can give rise to oscillations in the growth of peripheral and interior cells in a biofilm (50): The oscillations increase the community’s resilience to environmental stress by resolving the conflict of how the peripheral cells both protect and starve the interior cells. In all, these studies highlight the importance of understanding how the balance between cooperative and competitive interactions influences ecological dynamics in different environments.

Given the abundance of antibiotic deactivating enzymes found in both soil microbes (51–53) and in clinical pathogens (22, 54–56), we suspect that cross-protection mutualisms, or even more complex interaction networks (26), may frequently arise in natural bacterial populations and that such mutualisms may often form between different species of bacteria. Our model also suggests that cell death due to antibiotic exposure is not necessary to form a mutualism (nor even oscillations, see SI Appendix), suggesting that mutualisms could form with different combinations of bacteriostatic or bactericidal antibiotics (28). In addition, the ability of the two strains to protect each other and survive in a multidrug environment may buy time for a multidrug resistant strain to evolve. Because many plasmids carrying genes conferring resistance are readily spread via horizontal gene transfer (54, 57, 58), the type of cooperative interactions explored here may be able to evolve rapidly. Although we did not observe horizontal gene transfer during our experiments (SI Appendix, Fig. S2), the results of our double-resistant invasion experiment suggest that even if a double-resistant strain were to arise, the tradeoff between producing a resistance enzyme and overall fitness can prevent the double-resistant strain from fixing. Overall, this study adds to the growing evidence that protective interactions arising from cooperative deactivation of antibiotics can play a significant role in shaping the structure of bacterial communities and regulating the spread of antibiotic resistance genes.

Materials and Methods

Strains.

All strains are derived from Escherichia coli DH5α, which is sensitive to both ampicillin and chloramphenicol (SI Appendix, Fig. S25). The chloramphenicol-resistant strain is an E. coli DH5α strain transformed with the pBbS5c-RFP plasmid (59). The plasmid encodes a gene for chloramphenicol acetyltransferase (type I) enzyme and a gene for monomeric red fluorescent protein. It has a pSC101 origin of replication. pbBS5c-RFP was obtained from Jay Keasling (University of California, Berkeley, CA) via Addgene (plasmid 35284) (59). The ampicillin-resistant strain is an E. coli DH5α strain transformed with a plasmid encoding a gene for the β-lactamase enzyme (TEM-1) and a gene for enhanced yellow fluorescent protein (EYFP). The plasmid was assembled using the 2011 BioBrick distribution kit (60). We combined a constitutive promoter (J23116) with a sequence encoding a ribosome binding site, EYFP, and two stop codons (E0430). This construct was cloned into a vector containing the BioBrick pSB6A1 backbone. The double-resistant strain is an E. coli DH5α strain transformed with both plasmids.

Experiments.

Initial single cultures of our strains were grown for 24 h in culture tubes (3 or 5 mL) in LB supplemented with antibiotic for selection (50 μg/mL ampicillin and 25 μg/mL chloramphenicol for the ampicillin-resistant strain and the chloramphenicol-resistant strain, respectively) at 37 °C and shaken at 300 rpm (0.3 × g). The following day, 200 μL of cocultures of the two strains were grown at varying initial population fractions in LB without antibiotics. Subsequently, serial dilution experiments were done in well-mixed batch culture with a culture volume of 200 μL. Every cycle, the culture was diluted by a fixed amount into fresh LB medium supplemented with the antibiotics ampicillin and chloramphenicol. Except for where noted otherwise, each cycle lasted 24 h and cultures were diluted by 100×. Cultures were shaken at 500 rpm at a temperature of 37 °C. Growth medium was prepared by using BD’s DifcoTM LB Broth (Miller) (catalog no. 244620). Ampicillin stock was prepared by dissolving ampicillin sodium salt (Sigma-Aldrich catalog no. A9518) in LB at a concentration of 50 mg/mL. The solution was filter sterilized, stored frozen at −80 °C, and thawed before use. Chloramphenicol stock was prepared by dissolving chloramphenicol powder (Sigma-Aldrich catalog no. C0378) in 200 proof pure ethanol (KOPTEC) at a concentration of 25 mg/mL. This solution was filter sterilized and stored at −20 °C . Prepared 96-well plates of media supplemented with antibiotics were stored at −80 °C and thawed 1 d before inoculation at 5 °C.

Measurement and Data Analysis.

At the end of each growth cycle, we took spectrophotometric (Thermo Scientific Varioskan Flash at 600 nm) and flow cytometry (Miltenyi Biotec MACSQuant VYB) measurements of the cultures to determine subpopulation sizes. Relative abundances were confirmed by plating (SI Appendix, Fig. S6). Data analysis was performed using a combination of Mathematica, matplotlib, and IPython. Flow cytometry data were analyzed using the python package FlowCytometryTools (61). Data are available upon request.