Bacterial interspecies interactions modulate pH-mediated antibiotic tolerance

Predicting antibiotic efficacy within microbial communities remains highly challenging. Interspecies interactions can impact antibiotic activity through many mechanisms, including alterations to bacterial physiology. Here, we studied synthetic communities constructed from the core members of the fruit fly gut microbiota. Co-culturing of Lactobacillus plantarum with Acetobacter species altered its tolerance to the transcriptional inhibitor rifampin. By measuring key metabolites and environmental pH, we determined that Acetobacter species counter the acidification driven by L. plantarum production of lactate. Shifts in pH were sufficient to modulate L. plantarum tolerance to rifampin and the translational inhibitor erythromycin. A reduction in lag time exiting stationary phase was linked to L. plantarum tolerance to rifampicin, opposite to a previously identified mode of tolerance to ampicillin in E. coli. This mechanistic understanding of the coupling among interspecies interactions, environmental pH, and antibiotic tolerance enables future predictions of growth and the effects of antibiotics in more complex communities.


Introduction
Decades of investigations have described detailed and precise molecular mechanisms of antibiotic action using model organisms such as Escherichia coli grown in monoculture. However, unlike in laboratory monocultures, the vast majority of bacteria live in diverse communities. In environments such as the human intestine, antibiotics impact communities in many ways, ranging from the loss of diversity (Dethlefsen et al., 2008;Jernberg et al., 2007) to the evolution of multidrug-resistant pathogens (Santajit and Indrawattana, 2016). Hence, there is a pressing need for new frameworks that predict how antibiotics affect bacterial communities in ways that cannot be predicted from simple monocultures.
Synthetic microbial communities provide the opportunity to perform controlled experiments that uncover mechanistic principles of microbial ecology (Momeni et al., 2011;Zuk and Travisano, 2018). Such communities have been used to demonstrate the evolution and consequences of microbial cooperation (Harcombe, 2010;Momeni et al., 2013), reveal the role of interactions on spatial patterning (Momeni et al., 2013), and to link physiology and metabolism to mutualistic relationships (Ponomarova et al., 2017). Interspecies interactions can occur through specific mechanisms involving members of a community (e.g. cross-feeding or competition for resources) and through global environmental variables modified by bacterial activity. An example of the latter is pH, which has recently been shown to drive community dynamics in an artificial laboratory system of decomposition bacteria (Ratzke and Gore, 2018).
A natural system with low diversity provides a consortium of species with a common evolutionary and ecological history from which to build synthetic communities (Ponomarova et al., 2017). The gut microbiota of Drosophila melanogaster fruit flies is such a low diversity assemblage in which all members can be cultured in vitro (Obadia et al., 2017), making it amenable to the systematic dissection of bacterial interactions. This community consists of~5 species predominantly from the Lactobacillus and Acetobacter genera (Wong et al., 2011) ( Figure 1A). The metabolic lifestyles of species from these genera in isolation have been largely uncovered. Lactobacillus spp. produce lactic acid (Makarova et al., 2006), while Acetobacter spp. produce acetic acid and are distinguished by their ability to oxidize lactate to carbon dioxide and water (Yamada et al., 1997).
Bacteria can survive antibiotics through (i) resistance mutations, which counteract the antibiotic mechanism and increase the minimum inhibitory concentration (MIC); (ii) tolerance, whereby the entire population enters an altered physiological state that prolongs survivability without changing the MIC of the antibiotic, leading to an increase in the time required to kill a given fraction of the population; (iii) heteroresistance, whereby a subset of the population has a higher MIC and grows at concentrations that would otherwise kill the population; and (iv) persistence, whereby a subset of the population survives treatment for a longer period (Balaban et al., 2019;Brauner et al., 2016). Members of multispecies communities, such as biofilms and models of urinary tract infections, can display altered sensitivity to antibiotics in the community context (Adamowicz et al., 2018;de Vos et al., 2017;Nicoloff and Andersson, 2016;Sanchez-Vizuete et al., 2015). Previous studies have used synthetic communities to uncover the interplay between interspecies interactions and antibiotic efficacy; for example, the exoproducts of Pseudomonas aeruginosa affect the survival of Staphylococcus aureus through changes in antibiotic uptake, cell-wall integrity, and intracellular ATP pools (Radlinski et al., 2017). In genetically modified communities, intracellular antibiotic degradation affords cross-species protection against chloramphenicol (Sorg et al., 2016). Additionally, metabolic dependencies within synthetic communities can lower the viability of bacteria when antibiotics eliminate providers of essential metabolites, leading to an apparent change in the MICof the dependent species (Adamowicz et al., 2018). However, we still lack understanding of how metabolic interactions between bacteria affect the physiological processes targeted by antibiotics and the resulting balance between growth inhibition (bacteriostatic activity) and death (bactericidal activity). For example, the intimate relationship between bacterial metabolism and environmental pH could also lead to changes in antibiotic efficacy, as previously shown in monocultures (Aagaard et al., 1991;Argemi et al., 2013;Kamberi et al., 1999;Karslake et al., 2016;Yang et al., 2014). The interplay of all these processes in complex communities will provide new ways to combat pathogen survival and resistance evolution, particularly in cases involving tolerance, an important and understudied aspect of antibiotic susceptibility that can be elicited by diverse mechanisms and can facilitate the evolution of resistance (Levin-Reisman et al., 2017).
In this study, we interrogated how interspecies interactions affect growth, pH, and antibiotic susceptibilities. We used high-throughput assays to measure and compare these parameters in monocultures and co-cultures. We found that Lactobacillus plantarum (Lp) exhibited antibiotic tolerance (delay in death Brauner et al., 2016) in the presence of Acetobacter species. Lactate accumulation by Lp in monocultures acidified the media, inhibiting growth during stationary phase. Acetobactermediated lactate consumption released this inhibition by increasing pH, leading to a shorter Lp lag while exiting stationary phase. This reduced lag exiting stationary phase was correlated with the antibiotic tolerance of Lp that we observed. We determined that changes in pH elicited by Acetobacter activity were sufficient to modulate tolerance of Lp to both rifampin and erythromycin. Taken together, our findings indicate that simple changes to the environment can drive complex physiological behaviors and antibiotic responses within bacterial communities.   Figure 1. Interspecies interactions in synthetic communities derived from the fruit fly gut microbiome induce rifampin tolerance. (A) Synthetic community construction scheme. Relative abundances of the dominant species in the D. melanogaster gut microbiome determined from 16S rRNA sequencing. All reads mapped to two Lactobacillus species: L. plantarum (Lp) and L. brevis (Lb), and to three Acetobacter species: A. pasteurianus (Ap), A. tropicalis (At), and A. aceti (Aa). Values are mean ± standard deviation (S.D.), n = 18. Mean and S.D. were weighed by the total number of reads for each fly. Strains were isolated in agar plates and grown in liquid culture. Synthetic communities were then built by co-culturing individual species in liquid culture. (B) When grown with Ap, Lp survived after 24 h at rifampin concentrations above the MIC. Viable cell plating counts of Lp after growth in rifampin for 24 h normalized to the counts at the start of the experiment (t = 0, 3 Â 10 5 and 4 Â 10 5 CFU/mL in monoculture and co-culture, respectively). Dashed line indicates where the population at t = 24 h would have the same size as at t = 0. Error bars are S.D. for each condition, n = 3. p-values are from a Student's two-sided t-test of the difference of the co-culture from the monoculture (*: p<4Â10 À3 , **: p<8Â10 À4 , ***: p<8Â10 À5 , equivalent to *: p<0.05/n, **: p<0.01/n, ***: p<1Â10 À3 /n, where n = 12 is the number of comparisons; NS: not significant). (C) Protection of Lp at supra-MIC concentrations of rifampin is elicited by all Acetobacter species tested. CFU/mL of Lp grown in monoculture (Lp) or in co-culture with Ap, At, A. orientalis (Ao), A. indonesiensis (Ai), and Aa, and then treated with 20 mg/mL rifampin for 24 h normalized to counts at the start of the experiment (t = 0). Error bars are S.D. for each condition, n = 3. p-values are from a Student's two-sided t-test of the difference from the monoculture (*: p<0.01, equivalent to *: p<0.05/n, where n = 5 is the number of comparisons; NS: not significant). (D) Ap-mediated survival of Lp at rifampin concentrations above the MIC is history-dependent, requiring co-culturing before exposure as compared with mixing. CFU/mL of Lp grown in monoculture, in coculture with Ap (Co.), or mixed with Ap without subsequent growth in the absence of antibiotic (Mix), and treated with 20 mg/mL rifampin for 24 h Figure 1 continued on next page

Interspecies interactions induce tolerance to rifampin in a synthetic community
To determine the composition of the gut microbiota in our laboratory fruit flies, we performed 16S rRNA sequencing from 18 individual dissected guts (Materials and methods). We identified five species belonging to seven unique operational taxonomic units (OTUs) by clustering the sequences at 99% identity: L. plantarum (Lp), L. brevis (Lb), Acetobacter pasteurianus (Ap), A. tropicalis (At), and A. aceti (Aa) ( Figure 1A). We then isolated the species in culture (Supplementary file 1a) and determined the antibiotic sensitivities of the four major fly gut inhabitants (Lp, Lb, Ap, and At; Figure 1A) in vitro grown in Man, Rogosa, and Sharpe (MRS) medium.
We tested 10 antibiotics representing a wide variety of classes using plate-based growth assays (Materials and methods). For many drugs, some of the isolates were resistant (detectable growth) at least up to the highest concentrations tested. Rifampin was the only drug for which all four species exhibited sensitivity (Supplementary file 1b) and it is bactericidal (Walsh and Wencewicz, 2016), hence there is the opportunity to study survival as well as sensitivity.
Since Lb grew significantly more slowly and had a much longer lag phase than Lp (Figure 1-figure supplement 1), we focused on Lp and its interactions with the Acetobacter species, particularly Ap, which is more abundant in the fly gut than the other Acetobacter species ( Figure 1A). We grew Lp and Ap separately for 48 h in test tubes, combined them in test tubes at an optical density at 600 nm (henceforth OD) of 0.02 each, and co-cultured them in MRS for 48 hr. We then diluted this coculture and 48 h monocultures of Lp and Ap into fresh MRS at~5 Â 10 5 colony-forming units/mL (CFU/mL) in 96-well plates over a range of rifampin concentrations. To determine whether the coculture still contained both species, we measured the percentage of survival and the fraction of each species at various rifampin concentrations by plating on selective media and exploiting species-specific colony morphologies (Supplementary file 1a). In the co-culture, Ap died off at a similar concentration of rifampin as during growth in a monoculture (1.25 mg/mL, Figure 1-figure supplement 2A). For Lp, the MIC was similar in co-culture as in monoculture (1.25 mg/mL, Figure 1B), but at concentrations above the MIC, significantly more Lp cells survived in co-culture versus monoculture ( Figure 1B). This effect could not be explained by small differences in the initial inoculum, as increasing cell densities up to 100-fold did not change the MIC or survival of Lp in monoculture (Figure 1figure supplement 2B,C). Because of the change in Lp survival, we focused herein on this phenotype.
To determine whether Lp's increased survival was specific to co-culturing with Ap, we co-cultured Lp with each of the Acetobacter species, including a wild fly isolate of A. indonesiensis (Ai), and lab fly isolates of A. orientalis (Ao) and Aa, the fifth member of the microbiota of our flies ( Figure 1A). We then diluted each co-culture to an initial Lp cell density of~5 Â 10 5 CFU/mL into fresh MRS with 20 mg/mL rifampin (16X MIC) and let the cells grow for 24 hr. Co-culturing with any of the Acetobacter species increased survival by approximately one order of magnitude ( Figure 1C). To determine whether this increased survival requires co-culturing prior to rifampin treatment (rather than the presence of Acetobacter species being sufficient), we grew Lp and Ap separately for 48 h and normalized to counts at the start of the experiment (t = 0). Error bars are S.D. for each condition, n = 3. p-vValues are from a Student's two-sided t-test of the difference from the monoculture (**: p<5Â10 À3 , equivalent to **: p<0.01/n, where n = 2 is the number of comparisons; NS: not significant). (E) The time to killing of Lp under rifampin treatment is extended in the presence of an Acetobacter. CFU/mL of Lp grown in monoculture and co-cultured with Ap, and treated with 50 mg/mL rifampin, normalized to counts at the start of the experiment (t = 0). Error bars are S.D. for each condition, n = 3. p-values are from a Student's two-sided t-test of the difference from the monoculture at the corresponding timepoint (**: p<2Â10 À3 , equivalent to **: p<0.01/n, where n = 5 is the number of comparisons; † : measurement below limit of detection; NS: not significant). Values off the graph were below the limit of detection of the assay. The online version of this article includes the following figure supplement(s) for figure 1:  mixed and diluted them at the time of addition of 20 mg/mL rifampin. After 24 h of incubation, the number of CFU/mL was significantly lower in mixed culture than in co-culture ( Figure 1D), indicating a history dependence to increased survival.
To determine whether co-culturing slows killing by the drug, we examined the survival of Lp over time at a high drug concentration (50 mg/mL, 40X MIC). Similar to the experiments above, we compared Lp CFU/mL in a monoculture with that in a co-culture with Ap. In monoculture, Lp rapidly died, with CFU/mL becoming undetectable within 18 hr; by contrast, Lp survived >30 h after co-culturing ( Figure 1E). To test whether the increased time to death was due to a disruption in the balance of growth and death, we performed single-cell imaging of mono-and co-cultures on MRS agarose with 50 mg/mL rifampin. We observed no growth in either mono-or co-cultures over 3 h of imaging (n = 568 and n = 236 cells, respectively; Figure 1-figure supplement 2D), indicating that survivability is due to protection from death, as opposed to increased growth. Because the MIC of Lp was unchanged in co-cultures compared with monocultures (no resistance), and the delay in killing was observed in the bulk population in co-cultures, these results indicate that co-culturing Lp with Ap induces tolerance of Lp to rifampin (Balaban et al., 2019;Brauner et al., 2016).

Co-culturing leads to growth of Lp in stationary phase
To investigate possible environmental factors that could be linked to Lp tolerance to rifampin when co-cultured with Acetobacter species ( Figure 1C,E), we first inquired whether the total amount of growth of the co-culture was larger or smaller than expected from the yield of the monocultures. We grew Lp and each of the Acetobacter species separately for 48 hr, diluted the monocultures to OD = 0.04, combined the Lp monoculture 1:1 with each Acetobacter monoculture, and grew the cocultures for 48 h in a plate reader. We then diluted the saturated cultures 1:30 in phosphate-buffered saline (PBS) to accurately measure the final OD, and computed an interaction score based on an additive model: where OD co and OD A are the final ODs of the co-culture and the Acetobacter monoculture, respectively. With this metric, a>0 indicates synergy and a<0indicates antagonism. Ap showed a strong (positive) interaction with Lp, whereas At, Ai, and Aa had an interaction score closer to 0 ( We then determined the total carrying capacity of each species in the co-cultures by counting CFUs. The Lp CFU/mL values for 48-h co-cultures with Ap, At, and Ai were higher than the Lp monoculture ( Figure 2A); Ap showed the strongest effect. Aa and Ao did not significantly increase Lp CFU/mL ( To determine when the additional Lp growth took place, we monitored CFU/mL values for Lp and Ap in co-culture throughout a 48 h time course starting from an initial combined cell density of~5 Â 10 5 CFU/mL. Initially, Lp accounted for the bulk of the growth in the co-culture ( Figure 2B). Interestingly, Ap in liquid monoculture showed little to no growth in most replicates after 40 h ( Figure 2B); by contrast, in co-culture Ap started to grow after~20 h and reached saturation by~40 h ( Figure 2B), indicating that Ap also benefited from growth as a co-culture. Thus, a mutualism exists between Lp and Ap driven by growth during stationary phase.
Lactate metabolism leads to changes in pH in stationary phase cocultures Interestingly, after 30 hr, Lp displayed a significant (~2X) increase in CFU/mL in the Lp-Ap co-culture that did not occur in the monoculture ( Figure 2B), indicating that the increase in final yield occurs late in stationary phase. Since this increase occurs generally ( Figure 2B), we hypothesized that Lp has a common metabolic interaction with each of the Acetobacter species. An obvious candidate is cross-feeding, since Lp produces lactate and the Acetobacter species consume it. We measured lactate levels in the supernatants of Lp monocultures and co-cultures of Lp with each of the Acetobacter species individually, after 48 h of growth. As expected, the Lp monoculture accumulated Land D-lactate to high levels (>100 mM; Figure 2C). All co-cultures had significantly lower concentrations of both isomers than the monoculture (<2 mM, Figure 2C). The Lp-Ao co-culture harbored higher levels of L-lactate than any other co-culture and Lp-Aa had the highest concentration of L-lactate of the co-cultures ( Figure 2C). Lp-Ap, Lp-Ao, and Lp-Aa co-cultures all accumulated lactate to >10 mM by 20 h (Figure 2-figure supplement 2A). Taken together, these data suggest that Lp † † NS NS Figure 2. Lp growth during stationary phase in Acetobacter co-cultures is associated with an increase in pH and a decrease in lactate concentration. (A) Co-culturing Lp with Ap, At, or Ai resulted in increased Lp cell density after 48 hr. Co-culturing with Ao or Aa did not significantly increase Lp cell density by 48 hr. Error bars are standard deviation (S.D.) for each condition, n = 3. p-values are from a Student's two-sided t-test of the difference from the monoculture (*: p<0.01, **: p<2Â10 À3 , equivalent to *: p<0.05/n, **: p<0.01/n, where n = 5 is the number of comparisons; NS: not significant). (B) Co-culturing Lp with Ap resulted in higher Lp cell density in stationary phase, as well as faster growth and shorter lag for Ap. Shaded regions indicate S. D., n = 3. Inset: zoom-in on region inside dashed box highlighting increase in carrying capacity in co-culture. (C) L-and D-lactate accumulated to much higher levels in Lp monocultures than in co-cultures after 48 h of growth, suggesting that Acetobacter spp. consumed Lp-produced lactate. Lactate concentration was measured enzymatically from culture supernatants at 48 hr. Error bars are S.D. for each condition, n = 3. p-values are from a Student's two-sided t-test of the difference from the monoculture (**: p<2Â10 À3 , ***: p<2Â10 À4 , equivalent to **: p<0.01/n, ***: p<1Â10 À3 /n, where n = 5 is the number of comparisons; † : measurement below limit of detection). (D) The increase in Lp cell density in stationary phase is associated with an Acetobacter-dependent increase in pH early in stationary phase. pH was measured with the pH-sensitive dye 2',7-bis-(2-carboxyethyl)À5-(and-6)carboxyfluorescein over time (Materials and methods). Shaded regions indicate S.D., n = 3. Data points were taken every 8.3 min. The online version of this article includes the following figure supplement(s) for figure 2:   metabolism leads to an initial accumulation of lactate and that the Acetobacter species consume it, although Aa and Ao are less efficient at consuming L-lactate than the other species.
Since lactate is a short-chain fatty acid with a pK a of 3.86, we suspected that lactate production would affect the pH of the culture. We first monitored the pH dynamics of monocultures of Lp and of each Acetobacter species using the pH-sensitive fluorophore 2',7-bis-(2-carboxyethyl)À5-(and-6)carboxyfluorescein (BCECF) (James-Kracke, 1992). In the Lp monoculture, pH decreased from pH = 6.75 to below four during growth ( Figure 2D); more precisely, we measured a final supernatant pH = 3.77 using a pH meter (Figure 2-figure supplement 2A). For monocultures of all Acetobacter species except Ap, the medium first acidified down to pH~5, and then increased back to pH = 6-7 ( Figure 2D).
To test whether Acetobacter species reverse the pH decrease due to the accumulation of lactate produced by Lp, we measured the pH of co-cultures over time using BCECF. Co-cultures with Ap, At, and Ai followed similar trajectories in which the pH mimicked that of the Lp monoculture for the first 20 hr, after which the pH increased up to a final value of~7 ( Figure 2D). The Lp-Aa co-culture experienced a~10 h delay in the pH increase, while the co-culture with Ao showed only a slight pH increase by 48 h ( Figure 2D). The slight pH increase in the Lp-Ao co-culture is consistent with lower L-lactate consumption by Ao ( Figure 2C). Using a pH meter for validation, we measured final pH values of 5.9, 5.8, 4.6, 5.4, and 4.8 in co-cultures with Ap, At, Ao, Ai, and Aa, respectively (Figure 2figure supplement 2B). Thus, lactate metabolism dictates dramatic shifts in environmental pH that are related to physiological changes in antibiotic tolerance ( Figure 1C).
Given the strong acidification of the medium in Lp monoculture but not in Lp-Ap co-culture ( Figure 2D), we hypothesized that intracellular pH decreases in monoculture and increases in co-culture. To measure intracellular pH, we transformed our Lp strain with a plasmid expressing pHluorin (a GFP variant that acts as a ratiometric pH sensor Martinez et al., 2012) under the control of a strong constitutive promoter (Rud et al., 2006). The two absorbance peaks, which we measured at 405 and 475 nm, are sensitive to pH and the ratio of the emission (at 509 nm) at these two excitation wavelengths can be used to estimate intracellular pH. We grew this strain in monoculture and in coculture with Ap and measured fluorescence over time in a plate reader. Because of the high autofluorescence of the medium at 405 nm (data not shown), we could only track changes in fluorescence at an excitation wavelength of 475 nm. We observed an initial increase in signal as the Lp cells started to proliferate ( To verify that the decrease in fluorescence in monoculture was due to a drop in intracellular pH, as opposed to a decrease in protein concentration, we sampled cells after 48 h of growth, centrifuged them, resuspended them in PBS to measure pHluorin signal at both its excitation wavelengths, and measured fluorescence within 1 min of resuspension. The pHluorin signal ratio was significantly higher in Lp-Ap co-culture than in an Lp monoculture (Figure 2-figure supplement 3B). Taken together, these data indicate that the intracellular pH of Lp cells is significantly lower in monoculture than in co-culture with Acetobacter species.

Low pH inhibits the growth of Lp and extends lag phase
The growth rates of lactic acid bacteria such as Lp are known to be affected by pH (Stecka and Grzybowsii, 2000). Since Ap growth causes a large increase in the extracellular pH of an Lp-Ap coculture, we sought to determine the general dependence of Lp growth properties on pH. We diluted a 48 h culture of Lp cells grown in MRS at starting pH = 6.75 to a starting OD = 0.02 in MRS adjusted to starting pH ranging from 3 to 8 (Materials and methods). We then measured growth and BCECF fluorescence (Figure 3-figure supplement 1A,B). For lower starting pH values, the carrying capacity was lower ( Figure 3A) and varied over a large OD range from <0.02 to>2. For all starting pH values, Lp cells reduced the pH to a common final value of~3.7 ( Figure 3A). The bulk growth rate approached zero ( Figure 3B) as the pH approached its final value, explaining the differences in yield. Interestingly, the maximum growth rate was also pH-dependent ( Figure 3C), with the highest growth rate at starting pH = 7.
Given these findings, we hypothesized that the inhibition of growth in stationary phase of an Lp monoculture is due to the decreased intracellular and extracellular pH, and that Ap releases Lp's growth inhibition by raising intracellular and extracellular pH. To test this hypothesis, we inoculated a 48 h culture of Lp to an initial OD = 0.02 into the supernatant of a 48 h Lp culture at pH = 3.77 or manually adjusted to pH = 7. Almost no growth took place in supernatant starting from pH = 3.77, but we observed substantial growth (~20 fold increase in OD) of the bulk culture in supernatant raised to pH = 7 ( Figure 3D). As expected, the maximal growth rate was lower than in fresh MRS, presumably due to the partial depletion of nutrients ( Figure 3D); addition of glucose to the conditioned medium supported faster growth, but only starting from neutral pH (Figure 3-figure supplement 1C). We also hypothesized that the accumulation of lactate by Lp would allow growth of Ap in Lp-conditioned medium even at low starting pH. When we diluted a saturated Ap culture in Maximal growth rate in MRS increases with increasing initial pH. Error bars are S.D., n = 4. (D) Increasing the pH of a saturated, spent Lp culture from 3.77 to 7 allows growth, although not as much as fresh MRS. Error bars are S.D., n = 3. Data points were taken every 8.3 min. (E) Increasing the pH of an Lp monoculture at t = 30 h from 4 to 7 to mimic the pH increase in Lp-Ap co-culture (top) leads to a shorter lag phase (bottom). Lag time was calculated by fitting growth curves to the Gompertz equation. Error bars are S.D., n = 3. p-values are from a Student's two-sided t-test of the difference from the control (***: p<5Â10 À4 , equivalent to ***: p<1Â10 À3 /n, where n = 2 is the number of comparisons; NS: not significant). The online version of this article includes the following figure supplement(s) for figure 3: Lp-conditioned medium generated as above, Ap achieved an even higher OD than in fresh MRS ( Figure 3-figure supplement 1D).
These findings suggested that Ap's effects on Lp growth might be due primarily to the pH changes that Ap initiates because Acetobacter species can grow at low pH and consume lactate. Thus, we first increased the pH of an Lp monoculture to 7 after 30 h of growth, which is when the pH increased most rapidly in the Lp-Ap co-culture ( Figure 3E). We then incubated the cells for an additional 18 hr. We did not observe a significant increase in CFUs/mL from this pH-adjusted culture versus controls that simply grew for 48 h or were subjected to all washes required for pH adjustment and then returned to the same supernatant ( Figure 3-figure supplement 1E), possibly due to the high cell density and the transient nature of pH neutralization. We then assessed if increasing the pH at 30 h resulted in a decrease in the duration of lag phase by diluting the monoculture to OD = 0.0375 into fresh MRS after an additional 18 h of growth. Indeed, the lag phase was shorter for the pH-adjusted cells compared with a control culture ( Figure 3E). Thus, pH is a driver of the growth advantages of Lp in lag phase even in the absence of a net increase in cell number.
Co-culturing Lp with Acetobacter species reduces lag time Canonical antibiotic tolerance in E. coli results from a decrease in growth rate or an increase in lag phase that protects cells through metabolic inactivity (Brauner et al., 2016). To measure growth rate and lag phase, we co-cultured Lp with each of the Acetobacter species individually for 48 hr, diluted the culture to a common OD of 0.0375, and monitored growth in a plate reader. The maximum growth rate was the same for the Lp monoculture and co-cultures with Ap, At, and Ai, and slightly higher for co-cultures with Ao and Aa (Figure 4-figure supplement 1). We previously observed for Lp monocultures when shifting the pH that the stimulation of growth in stationary phase was connected with a shorter lag phase ( Figure 3E). In agreement with these data, there was a significant decrease in bulk lag time for all the Acetobacter co-cultures ( Figure 4A,B). Surprisingly, this shorter lag phase was linked to Lp tolerance to rifampin, opposite to the longer phase linked to E. coli tolerance to ampicillin (Brauner et al., 2016). Ap, At, and Ai co-cultures had the largest lag decreases. The Aa and Ao co-cultures had smaller, although still significant, decreases ( Figure 4A, B); interestingly, Aa and Ao were also less efficient at consuming lactate than the other Acetobacter species ( Figure 2C). These data indicate that interspecies interactions can change the physiology of community members, and that differences across the Acetobacter species constitute an opportunity to probe the underlying cause of the lag phenotype.
As with Lp tolerance to antibiotics ( Figure 1B), the shortened lag phase of the Lp-Ap co-culture depended on the inoculum coming from a co-culture. When we simply mixed independent 48 h cultures of Lp and Ap, the resulting bulk culture had the same lag time as an Lp monoculture ( Figure 4C,D). To determine which of the two species was responsible for the decrease in lag, we performed time-lapse microscopy (Materials and methods) to monitor the initiation of growth at the single-cell level ( Figure 4E). Lp and Ap were clearly distinguishable based on morphology (Figure 4-figure supplement 2): Lp cells are longer (2.46 ± 0.78 mm vs. 1.66 ± 0.38 mm) and thinner (0.72 ± 0.12 mm vs. 0.90 ± 0.09 mm) than Ap cells. Therefore, we used the aspect ratio (length/width; 3.41 ± 0.91 for Lp and 1.87 ± 0.46 for Ap) to distinguish single cells from each species in co-culture. We validated this strategy on co-cultures of fluorescently tagged strains of the same two species and measured a 10% error rate in classification (Figure 4-figure supplement 2D-F). In co-culture, most Lp cells were observed to have grown by 1 h after spotting, but in Lp monoculture, few cells were growing even after 2 h ( Figure 4E,F). Ap cells did not grow during the time of imaging ( Figure 4F), indicating that the reduced lag time was due to Lp's growth, in agreement with CFU/ mL measurements from liquid cultures ( Figure 2B).

Growth status and pH are drivers of antibiotic tolerance
Since pH changes shortened lag phase ( Figure 3E) and changes in lag time were linked to antibiotic tolerance (Figures 1 and 4), we tested whether shortening lag phase was sufficient to induce tolerance. Increased time in starvation leads to changes in physiology, including an increase in the duration of lag phase . To determine the relationship between time spent in stationary phase and lag time for Lp, we grew Lp cultures for varying amounts of time in stationary phase then used these cultures as inocula for growth in a plate reader. Incubating the Lp monocultures for more than 48 h resulted in a dramatic increase in the duration of lag phase, while reducing the culturing time shortened lag phase ( Figure 5A).
Because lag phase in co-culture is slightly shorter than that of a 24 h monoculture ( Figure 5A), we decided to match the lag time of a co-culture by growing a monoculture for 20 h from an initial   Lp as a monoculture for a shorter time leads to higher cell survival. Viable cell plating counts of Lp after growth in 20 mg/mL rifampin for 24 h normalized to the counts at the start of the experiment (t = 0). Error bars are S.D. for each condition, n = 3. p-values are from a Student's two-sided ttest of the difference between the cultures (***: p<1Â10 À3 ). (C) Neutralization of pH in stationary phase in Lp monocultures is sufficient to induce tolerance. Increasing the pH of an Lp monoculture at t = 30 h or t = 40 h to 7 to mimic the pH increase in co-cultures of Lp with Acetobacter species Figure 5 continued on next page OD = 0.02. We then measured Lp survival in 20 mg/mL rifampin after 24 h in cultures diluted from a 48-hr-old or a 20-hr-old culture. Culturing for 20 h resulted in a significant increase in survival ( Figure 5B). We next tested whether shortening lag phase by changing the pH in stationary phase also yielded increased rifampin tolerance of Lp in monoculture. We increased the pH of an Lp monoculture to 7 after 30 and 40 hr, and grew cells for an additional 18 h and 8 hr, respectively. We then measured the change in CFU/mL upon treatment with 50 mg/mL rifampin for 24 hr. The upshift in pH at t = 30 h or 40 h resulted in increased tolerance relative to the unshifted monoculture ( Figure 5C). To test whether pH changes can also decrease tolerance, we grew co-cultures of Lp and Ap for a total of 48 hr, decreasing the pH to 3.7 at t = 30 h or 40 hr. In both cases, the viability after 24 h of rifampin exposure was significantly reduced relative to an untreated monoculture ( Figure 5D). Thus, pH can affect tolerance both positively and negatively.
In co-culture with Lp, Ap raised the pH earlier than did Aa, while Ao only raised the pH very slightly ( Figure 2D). We hypothesized that due to these distinct pH dynamics, rifampin would also have different Lp killing dynamics in these co-cultures. We grew co-cultures of these Acetobacter species with Lp as previously, and then treated the co-cultures with 50 mg/mL rifampin. While all cocultures had extended survival relative to Lp monoculture, the killing dynamics of Lp were indeed distinct, with Aa inducing the highest tolerance ( Figure 5E). To determine the extent to which these dynamics can be explained by the time at which each species raises the pH, we measured CFU/mL at various time points after rifampin treatment for Lp monocultures grown for 48 h whose pH was raised to pH seven at t = 30 h or 40 hr, mimicking the early and late increases in pH for Ap and Aa co-cultures, respectively. The shift to pH 7 at 40 h induced higher rifampin tolerance than the shift at 30 h ( Figure 5F), consistent with the increased tolerance of the Lp-Aa co-culture ( Figure 5E). Moreover, shifting the pH to 4.5 at t = 30 hr, to mimic the slight increase caused by Ao, was also sufficient to increase tolerance comparable to pH neutralization at t = 30 h ( Figure 5F), consistent with the similar killing dynamics of the Ap and Ao co-cultures ( Figure 5E). All pH shifts induced higher tolerance compared to control cultures that underwent the same protocol but whose pH was maintained ( Figure 5F). Taken together, these experiments establish that pH changes drive changes in both the exit from stationary phase and tolerance of Lp to rifampin.

Growth status and pH also modulate tolerance to a ribosome-targeting antibiotic
The robust relationships among changes in pH, lag time, and rifampin tolerance prompted us to explore how changes in pH and lag time affect survival to other antibiotics. We decided to use the Figure 5 continued (upper panel) increased cell survival (CFU/mL normalized to counts at the start of the experiment, t = 0) after treatment with 20 mg/mL rifampin for 24 h (lower panel). A 48-h-old culture with no changes in pH was used as a control (Ctrl.). Error bars are S.D. for each condition, n = 3. p-values are from a Student's two-sided t-test of the difference between the cultures (**: p<5Â10 À3 , equivalent to **: p<0.01/n, where n = 2 is the number of comparisons). (D) Acidification of Lp co-cultures with Ap during the exponential-to-stationary phase transition or in late stationary phase sensitizes Lp to rifampin. Decreasing the pH of an Lp co-culture with Ap at t = 30 h or t = 40 h to 3.7 to mimic the pH of an Lp monoculture (upper panel) increased survival (CFU/mL normalized to counts at the start of the experiment, t = 0) after treatment with 20 mg/mL rifampin for 24 h (lower panel). Error bars are S.D. for each condition, n = 3. p-values are from a Student's two-sided t-test of the difference between the cultures (***: p<5Â10 À4 , equivalent to ***: p<1Â10 À3 , where n = 2 is the number of comparisons). † : values below the limit of detection. (E) The dynamics of killing in Lp co-culture with Acetobacter species differs quantitatively according to species and from Lp monoculture (blue), indicating that the Acetobacter species induce rifampin tolerance to different degrees. CFU/mL of Lp in monoculture and in co-culture with Acetobacter species, and treated with 50 mg/mL rifampin, normalized to counts at the start of the experiment (t = 0). Error bars are S.D. for each condition, n = 3. (F) The timing of the pH change in Acetobacter co-culture predicts the extent of protection against 50 mg/mL rifampin. Neutralization of pH in Lp monocultures at 40 h of growth (to mimic Lp+Aa cocultures) elicits longer protection against rifampin than neutralization at 30 hr. A small increase in pH (from 3.85 to 4.5) at 30 h (to mimic Lp+Ao cocultures) provides protection comparable to complete neutralization. CFU/mL were normalized to counts at the start of the experiment (t = 0). Error bars are S.D. for each condition, n = 3. (G) Lp survival to erythromycin is~10 times higher after 24 h of treatment with erythromycin at supraMIC concentrations for 48-h-old monocultures of Lp than for 20-h-old Lp monocultures. Viable cell plating counts of Lp after growth in erythromycin for 24 h normalized to cell counts at the start of the experiment (t = 0). Error bars are S.D., n = 3. p-values are from a Student's two-sided t-test of the difference between the two samples at a given time point (*: p<4Â10 À3 , **: p<8Â10 À4 equivalent to *: p<0.05/n, **: p<0.01/n, where n = 12 is the number of comparisons; NS: not significant). (H) Shifting the pH of an Lp monoculture at 30 h to 4 or 7, followed by 18 h of growth before treatment with 2 mg/mL erythromycin, mimics the survival dynamics of a 48-h-old or 20-h-old culture in stationary phase, respectively. CFU/mL of Lp monocultures were normalized to counts at the start of the experiment (t = 0). Error bars are S.D. for each condition, n = 3. ribosome-targeting macrolide erythromycin because it is bacteriostatic (in contrast to rifampin) and Lp is sensitive to it (Supplementary file 1b). We treated Lp monocultures grown for 20 h or 48 h with increasing concentrations of erythromycin for 24 h at a starting cell density of~5 Â 10 5 CFU/ mL. Opposite to our observations with rifampin ( Figure 5B), a 48 h Lp monoculture displayed tolerance to erythromycin, while a 20 h culture did not ( Figure 5G). While both cultures had the same MIC in erythromycin (0.078 mg/mL, Figure 5G), at concentrations above the MIC, the 48 h culture showed no changes in CFU/mL after 24 h of erythromycin treatment; the 20 h culture had a reduction of~10 fold in CFU/mL ( Figure 5G). This result suggests that Lp cells are typically tolerant to erythromycin but killing is accelerated under conditions that make them tolerant to rifampin.
To determine whether antibiotic tolerance underlies the survival of Lp to erythromycin as well as to rifampin, we diluted 20-and 48 h cultures to a starting density of~5 Â 10 5 CFU/mL, exposed them to a high concentration of erythromycin (2 mg/mL, 25X MIC), and monitored CFU/mL over time. Cells from a 20 h culture died significantly more rapidly than cells from a 48-hr-old culture ( Figure 5H), indicating that the differences in survival ( Figure 5G) are explained by erythromycin tolerance. Further, increasing the pH of an Lp monoculture at 30 h and then exposing it after an additional 18 h of growth to 2 mg/mL erythromycin in fresh medium had an increase in the rate of killing that was similar to that achieved with a 20 h culture ( Figure 5H). These results highlight that the effects of changing the growth status of a culture on antibiotic response are not limited to rifampin and-as in the case of erythromycin-can be opposite.

Discussion
Our measurements of growth behavior in synthetic communities motivated by the natural context of the fly gut indicate that interspecies interactions impact both growth and the effect of antibiotics on individual species. The pH-based mechanism underlying the tolerance of Lp induced by Acetobacter species is intrinsically connected to the metabolic capacity of each species, and hence is likely to be generally relevant in more complex communities. Moreover, these findings could have important implications for human health, for example in the context of Lactobacillus-dominated vaginal microbiotas (Boris and Barbés, 2000), hence their generality should be tested broadly in other contexts.
In this study, we observed a novel form of antibiotic tolerance. Tolerance has been defined as increased time to killing of the population as a whole (Brauner et al., 2017), as opposed to resistance (a change in the MIC), or persistence or heterotolerance whereby a subpopulation of bacteria displays increased time to killing (Balaban et al., 2019;Brauner et al., 2016). Lp cells in monoculture rapidly died, while Lp cells in co-cultures showed close to 100% survival for extended periods ( Figures 1E and 5E). Furthermore, the time and magnitude of pH change affected the kinetics of killing but not the size of the surviving population ( Figure 5E and F). These behaviors contrast with the stereotypical rapid decrease in viability, with a small subpopulation of surviving cells, observed in persistence. At high doses and late time points, cell counts were below our limit of detection and therefore we cannot rule out that, in addition to tolerance, pH changes in stationary phase could also modulate persistence. Tolerance to beta lactams such as ampicillin has been observed in E. coli cultures that exhibit slow growth or a long lag phase (Brauner et al., 2016), and E. coli mutants with longer lag phases can be selected through experimental evolution to match the time of treatment (Fridman et al., 2014;Levin-Reisman et al., 2017). Based on these previous studies, we were surprised to find the opposite effect with rifampin on Lp: cultures with a shorter lag phase exhibited increased tolerance (Figures 1 and 4). Moreover, although tolerance to erythromycin was associated with a longer lag phase ( Figures 3E and 5A,H), killing retardation was at least an order of magnitude longer than the change in lag time ( Figure 5A,H), indicating that tolerance is not determined by an elongation of lag phase alone, in contrast to the effects of ampicillin on E. coli (Fridman et al., 2014). The opposite behaviors of Lp to rifampin and erythromycin upon pH changes suggest that tolerance is more complex than previously thought, and should motivate future investigations into the physiological and genetic basis of tolerance.
Several genetic factors that increase time to killing have been identified in E. coli, including toxinantitoxin modules such as hipBA (Rotem et al., 2010) that induce the stringent response and thus cause transient growth arrest. In Lp co-culture with Acetobacter species, metabolic interactions alter the physiological state of Lp during late stationary phase by changing the environmental pH (Figure 2). The stringent response is required to survive acid shock in Helicobacter pylori (Mouery et al., 2006) but not in Enterococcus faecalis (Abranches et al., 2009), which is in the same taxonomic order as Lp. In the case of Lp, whether the stringent response could be a major factor in the increased tolerance to rifampin is unclear due to the surprising connection with decreased lag.
The pH in stationary phase can affect many factors, such as the chemistry of extracellular metabolites and macromolecules as well as the surface of the cell (Peters et al., 2016). Importantly, our assays of antibiotic sensitivities were all performed at a starting pH of 7. Nonetheless, shifts in extracellular pH can lead to buffered drops in cytoplasmic pH (Choi and Groisman, 2016;Olsen et al., 2002) that can be regulated (Chakraborty et al., 2015) or result from internalization of low-pK a species such as short-chain fatty acids (Ricke, 2003). Such changes could lead to protonation of macromolecules involved in adsorption or changes in the proton motive force (Krulwich et al., 2011). How these factors affect non-polycationic antibiotics such as rifampin remains to be determined; neither of the ionizable functional groups of rifampin (pK a s 1.7 and 7.9 Maggi et al., 1966) nor erythromycin (pK a 8.88 McFarland et al., 1997) have pK a s in the pH range achieved in our cultures (Figure 2, Figure 3-figure supplement 1B). Protonation changes in target macromolecules could also lead to protection against antibiotics, although we would expect a subsequent change in MIC, contrary to our findings (Figures 1 and 5). Intracellular acidification by the short-chain fatty acid propionate has been proposed to lengthen lag phase in Salmonella in vitro and in the mouse gut (Jacobson et al., 2018), consistent with our finding that lag time ( Figure 4)  Changes in intra-and extracellular pH have been shown to lead to transcriptional responses that provide cross-protection against antibiotics (Komora et al., 2017;McMahon et al., 2007;O'Sullivan and Condon, 1997), suggesting that the killing retardation due to a pH increase in stationary phase may result from a complex regulatory process. One major factor influencing the Lactobacillus-Acetobacter interaction is that these organisms form a recurrent community and may therefore have evolved to sense and benefit from each other's presence. Further experiments are needed to uncover the molecular mechanisms that link growth state and susceptibility to antibiotics in Lactobacillus species, other non-model organisms, and microbial communities. In addition, although we consistently observed related shifts in lag phase and tolerance (Figures 3 and 5), it remains to be established whether lag time and tolerance are causally linked or coupled to some global variable. The opposite connections with rifampin and erythromycin tolerance underscore the complexity of the link between growth and antibiotic action. Changes in growth may lead to changes in the levels and activity of these antibiotics' molecular targets. Moreover, the molecular mechanisms that lead to death downstream of the antibiotic target could be a function of the growth state of the cell. Previous work has shown that bacterial interactions can elicit changes in antibiotic sensitivity by changing cellular physiology or interfering with antibiotic action directly or indirectly (Adamowicz et al., 2018;Radlinski et al., 2017;Sorg et al., 2016).
In principle, a myriad of intra-and extracellular variables are subject to the composition and dynamics of the ecosystems that bacteria inhabit, and microbial communities within animal hosts can elicit changes in environmental variables both locally and globally. Specifically, the microaerobic and anaerobic microenvironments of the fly (Obadia et al., 2017) and human gastrointestinal tracts enable the growth of short chain fatty acid producers. Some of these short chain fatty acids, like butyrate, have been shown to play an important role on host physiology and health (Nicholson et al., 2012). The consequences of the accumulation of these short chain fatty acids and other small molecules on microenvironments, as well as their effect on bacterial physiology and antibiotic treatment efficacy in vivo, have yet to be systematically explored. Our findings linking short chain fatty acid metabolism, growth, and antibiotic action in commensal microbes from the fruit fly gut opens the door to studying these phenomena in a model organism. While the human gut microbiome comprises hundreds of bacterial species, the simplicity of the Drosophila gut microbiota (Wong et al., 2011), the genetic tractability of Drosophila, and the fact that~65% of human disease-causing genes have homologs in the Drosophila genome (Ugur et al., 2016), make the fruit fly a powerful model for host-microbiome interactions (Douglas, 2018). Our results also emphasize the need to probe the action of antibiotics -as well as other drugs that are thought not to target microbial growth (Maier et al., 2018) -in complex and varied conditions (Beppler et al., 2017). Furthermore, our findings highlight the utility of studying growth physiology in co-cultures in the absence of antibiotics for uncovering novel mechanisms of community-encoded protection against antibiotics.

Fruit fly stocks and gut microbiome sequencing
Wolbachia-free Drosophila melanogaster Canton-S (BL64349) flies were obtained from the Bloomington Drosophila Stock Center, and were reared and maintained as previously described in Obadia et al. (2017). To determine the bacterial strains present in our flies, we performed culture-independent 16S amplicon sequencing targeting the V4 region on an Illumina MiSeq. Individual flies were CO 2 -anesthetized, surface-sterilized by washing with 70% ethanol and sterile PBS six times each. Flies were dissected under a stereo microscope and their guts were placed in 2 mL screw cap microtubes containing 200 mL of 0.1 mm sterile zirconia-silicate beads (BioSpec Products 11079101z) and 350 mL of sterile lysis buffer (10 mM Tris-HCl, pH 8, 25 mM NaCl, 1 mM EDTA, 20 mg/mL lysozyme). Samples were homogenized by bead beating at maximum speed (Mini-Beadbeater, BioSpec Products) for 1 min. Proteinase K was added at 400 mg/mL and samples were incubated for 1 h at 37˚C. Sample were then centrifuged (3000 Â g for 3 min) and 300 mL of the nucleic acids-containing supernatant were transferred to 1.7 mL microtubes. Genomic DNA from samples was cleaned up through a DNA Clean and Concentrator-5 column (Zymo Research D4014). Using the protocol described in Fadrosh et al. (2014) for library preparation and sequencing, we sequenced the gut contents of 18 individual flies, three flies each from six independent vials. Paired-end 250-base pair sequencing generated >10,000 reads per sample. Reads were filtered using PrinSeq as in Koch et al. (2016). The reads were then clustered into operational taxonomic units (OTUs) at 99% identity and assigned taxonomy using LOTUS (Hildebrand et al., 2014) with the following parameters: [-threads 60 -refDB SLV -highmem 1 -id 0.99 p miseq -useBestBlastHitOnly 1 -derepMin 3:10,10:3 -simBasedTaxo 1 CL 3]. Redundant strain identities were collapsed into single OTUs. Common reagent contaminant strains were then removed (Salter et al., 2014). After filtering, only five unique species were identified ( Figure 1A). We isolated these species in culture and verified the taxonomic identity of our isolates using Sanger sequencing of the complete 16S rRNA gene (Gould et al., 2018). At 97% OTU clustering, only three species were found: Acetobacter sp., Lactobacillus plantarum, and Lactobacillus brevis. When less stringent FASTQ quality filtering was used, trace amounts (~0.01%) of two mammalian gut strains were identified: Blautia sp. and Bacteroides sp. Because these OTUs were eliminated by more stringent quality filtering, we speculate that they may have resulted from barcode bleed-through on the MiSeq flowcell. Sequencing data is available at the NCBI website under BioProject accession number PRJNA530819.
Frozen stocks were streaked onto MRS agar plates (1.5% agar, Difco agar, granulated, BD 214530) and single colonies were picked to start cultures. Colonies were inoculated into 3 mL MRS in glass test tubes and grown for 48 h at 30˚C with constant shaking. Unless otherwise noted, the saturated cultures were diluted to OD = 0.02 and grown in 3 mL MRS in glass test tubes for 48 h at 30˚C with constant shaking before the start of the experiment. For co-cultures, we grew Lp and each of the Acetobacter species separately for 48 h from colonies as above, diluted the monocultures to OD = 0.04, combined the Lp monoculture 1:1 with each Acetobacter monoculture, and grew them in 3 mL MRS in glass test tubes for 48 h at 30˚C with constant shaking before the start of the experi- To count CFUs in cultures, aliquots were diluted serially in PBS. For cultures treated with high concentrations of antibiotics, cells were centrifuged for 1.5 min at 8000 x g and resuspended in 1X PBS pH 7.4 (Gibco 70011044) after removing the supernatants. PBS-diluted cultures were plated on MRS and MYPL because Lactobacillus species grow faster than Acetobacter species on MRS and vice versa on MYPL. Colony morphology and color enable differentiation of Lactobacillus from Acetobacter species.

Conditioned media
Conditioned media were obtained by centrifuging cultures at 4500 x g for 5 min and filtering the supernatant with a 0.22 mm polyethersulfone filter (Millex-GP SLGP033RS) to remove cells.

MIC estimations
To estimate the sensitivity of each species to various antibiotics, colonies were inoculated into MRS and grown for 48 h at 30˚C with constant shaking. Cultures were diluted to an OD of 0.001 for Lp, Lb, and At, and 0.01 for Ap. Diluted cultures (195 mL) were transferred to 96-well plates containing 5 mL of antibiotics at 40X the indicated concentration. Antibiotics used were ampicillin (ampicillin sodium salt, MP Biomedicals 02194526, Lot R25707, stock at 100 mg/mL in milliQ H 2 O), streptomycin (streptomycin sulfate salt, Sigma S9137 Lot SLBN3225V, stock at 50 mg/mL in milliQ H 2 O), chloramphenicol (Calbiochem 220551, Lot D00083225, stock at 50 mg/mL in ethanol), tetracycline (tetracycline hydrochloride, MP Biomedicals 02103011, Lot 2297K, stock at 25 mg/mL in dimethyl sulfoxide (DMSO)), erythromycin (Sigma E5389-1G, Lot WXBC4044V, stock at 64 mg/mL in methanol), ciprofloxacin (Sigma-Aldrich 17850, Lot 116M4062CV, stock at 1.2 mg/mL in DMSO), trimethoprim (Alfa Aesar J63053-03, Lot T16A009, stock at 2 mg/mL in DMSO), spectinomycin (spectinomycin hydrochloride, Sigma-Aldrich PHR1426-500MG, Lot LRAA9208, stock at 50 mg/mL in milliQ H 2 O), rifampin (Sigma R3501-5G, Lot SLBP9440V, stock at 50 mg/mL in DMSO), and vancomycin (vancomycin hydrochloride, Sigma-Aldrich PHR1732À4 Â 250 MG, Lot LRAB3620, stock at 200 mg/mL in DMSO:H 2 O 1:1). Antibiotics were diluted serially in 2-fold increments into MRS. Cultures were grown for 24 h at 30˚C with constant shaking and absorbance was measured in an Epoch2 plate reader (BioTek Instruments) at 600 nm. The MIC was estimated as the minimum concentration of antibiotic with absorbance within two standard deviations of media controls.
For experiments in Figure 1B and Figure 1-figure supplement 2A-C, mono-and co-cultures were diluted to an OD of 0.001 (final cell density~5 Â 10 5 CFU/mL) and transferred to 96-well plates containing 5 mL of rifampin at 40X working concentration. Cultures were grown for 24 h and then serially diluted in 5-fold increments in PBS, and 3 mL of the dilutions were spotted onto MRS and MYPL rectangular plates using a semi-automated high-throughput pipetting system (BenchSmart 96, Mettler Toledo). Plates were incubated at 30˚C until colonies were visible for quantification of viability.

Plate reader growth curves
Cultures were grown from single colonies for 48 h in MRS at 30˚C with constant shaking. Then, cultures were diluted to a final OD of 0.02 and 200 mL of the dilutions were transferred to clear-bottom transparent 96-well plates. Plates were sealed with transparent film pierced with a laser cutter to have~0.5 mm holes to allow aeration in each well. Absorbance was measured at 600 nm in an Epoch2 plate reader (BioTek Instruments). Plates were shaken between readings with linear and orbital modes for 145 s each.
Growth rates and lag times were quantified using MATLAB (Mathworks, R2008a). The natural logarithm of OD was smoothed with a mean filter with window size of 5 timepoints for each condition over time, and the smoothed data were used to calculate the instantaneous growth rate d(ln(OD))/ dt. The smoothed ln(OD) curve was fit to the Gompertz equation (Zwietering et al., 1990) to determine lag time and maximum growth rate.

pH measurements
Culture pH was measured using the dual-excitation ratiometric pH indicator 2',7-bis-(2carboxyethyl)À5-(and-6)-carboxyfluorescein, mixed isomers (BCECF, Invitrogen B1151, Lot 1831845), which has a pK a of~6.98. A stock solution of 1 mg/mL BCECF in DMSO (Fisher BioReagents BP231, Lot 165487) was diluted 1000-fold into MRS to a final concentration of 1 mg/mL. Cells were grown in a Synergy H1 plate reader (BioTek Instruments) following the procedure described above. In addition to absorbance, fluorescence was measured every cycle using monochromators at excitation (nm)/emission (nm) wavelength combinations 440/535 and 490/535. After subtracting the fluorescence of wells containing cells without the indicator, the ratio of the signals excited at 490 nm and 440 nm was used to calculate the culture pH using a calibration curve of MRS set to various pH values.
Culture pH after 48 h of growth was directly measured with a pH meter (sympHony, VWR) equipped with a pH combination electrode (Fisherbrand accumet 13-610-104A).

Changes in pH during growth
To change the pH of monocultures and co-cultures in stationary phase, we obtained conditioned medium at 30 h or 40 h as described above and set the pH to the desired values. We then centrifuged 2 mL of a replicate culture for 3 min at 8000 x g, removed the supernatant, and resuspended cells in 1 mL of the corresponding medium to wash the cells. The suspension was centrifuged a second time and the pellets were resuspended in 2 mL of the corresponding medium.

Time-lapse and fluorescence microscopy
Cells were imaged on a Nikon Eclipse Ti-E inverted fluorescence microscope with a 100X (NA 1.40) oil-immersion objective. Images were collected on a DU897 electron multiplying charged couple device camera (Andor) using mManager v. 1.4 (Edelstein et al., 2010). Cells were maintained at 30˚C during imaging with an active-control environmental chamber (Haison).
Cultures grown for 48 h were diluted 100-fold into PBS and 2 mL were spotted onto a 1% (w/v) agarose MRS pad. After drying at room temperature, the pads were covered with a cover slip, sealed with a mixture of equal portions of Vaseline, lanolin, and paraffin, and transferred to the microscope. Images were taken every 2 min using mManager v. 1.4.
To quantify the morphology of cells using fluorescent strains, co-cultures were diluted 100-fold into PBS and 2 mL were spotted onto a 1% (w/v) agarose PBS pad. After drying, the pads were covered with a cover slip and transferred to the microscope. Images were acquired at room temperature using mManager v. 1.4.

Single-cell tracking and analysis
Images were segmented and cells were tracked using the software SuperSegger v. 3 (Stylianidou et al., 2016). Further analysis of single-cell growth was performed using MATLAB. Cells with length >6 mm were removed from further analysis due to issues with segmentation. Length traces were smoothed using a mean filter of window size 5. Cells were classified as Lp or Ap if 90% of their traces were above (Lp) or below (Ap) a log 10 (length-to-width ratio) of 0.375. Traces with more than 15 timepoints were used for further analysis. Elongation rates d(ln L)/dt were calculated for each cell and the mean and standard error were computed for each time point.

pHluorin measurements
Cells were grown following the procedure described above. The Lp pHluorin strain was grown in MRS containing 10 mg/mL chloramphenicol for the first 48 h of growth. In addition to absorbance, fluorescence was measured every cycle using monochromators at excitation (nm)/emission (nm) wavelength combinations 405/509 and 475/509. Because the signal from excitation wavelength 405 nm was indistinguishable from signal from medium (data not shown), we also measured pHluorin signal at both excitation/emission wavelength combinations for cells in PBS. Cultures (48-h-old, 250 mL) were centrifuged at 10,000 x g for 1 min and resuspended in 1X PBS. Aliquots (200 mL) were transferred to a 96-well plate and fluorescence was measured using monochromators at excitation (nm)/ emission (nm) wavelength combinations 405/509 and 475/509 within 1 min of resuspension in a Synergy H1 plate reader (BioTek Instruments).

Lactate measurements
Colonies of Lp and Acetobacter species were inoculated into 3 mL MRS and grown for 48 h at 30˚C with constant shaking. Saturated cultures were diluted to OD 0.02, mixed 1:1, and grown at 30˚C with constant shaking. After mixing for 20 h and 48 hr, a 700 mL aliquot was transferred to a microcentrifuge tube and centrifuged at 10,000 x g for 4 min. Supernatant (600 mL) was transferred to a new tube and centrifuged at 10,000 x g for 4 min. Supernatant (500 mL) was transferred to a new tube and kept on ice for not longer than 1 hr, until lactate was measured.
L-and D-lactate concentrations were measured using the EnzyChrom L-(BioAssay Systems ECLC-100, Lots BH06A30 and BI07A09) and D-lactate (BioAssay Systems EDLC-100, Lots BH0420 and BI09A07) Assay Kits. Samples were diluted 10-and 100-fold in water, and absorbance was measured according to the manufacturer's instructions in a plate reader (Tecan M200). We also included controls without lactate dehydrogenase to account for endogenous activity in the supernatants.

Statistical analyses
To determine significance of differences, we performed pairwise Student's two-sided t-tests throughout. To decrease Type I error, we performed Bonferroni corrections for each experiment. Significant differences are denoted in the figures: *: p<0.05/n, **: p<0.01/n, ***: p<0.001/n, where n is the number of comparisons.
The number of technical replicates are indicated in the figure legends. In all cases, technical replicates originated from a single bacterial colony, which was grown for 48 h in liquid media. The saturated culture was then split into technical replicates. Biological replicates correspond to experiments starting from a different bacterial colony, freshly streaked from frozen stocks. Biological replicates of the main conclusions of this study are present in multiple figures: co-culturing of Lp with Ap protects Lp from rifampin killing ( Figure 1B,C,D); the protective effect is due to a change in killing kinetics ( Figure 1E and Figure 5E); co-culturing of Lp with Ap leads to an increase in pH in stationary phase ( Figure 2D and  Figure 5A).