Division of labor and collective functionality in Escherichia coli under acid stress

The acid stress response is an important factor influencing the transmission of intestinal microbes such as the enterobacterium Escherichia coli. E. coli activates three inducible acid resistance systems - the glutamate decarboxylase, arginine decarboxylase, and lysine decarboxylase systems to counteract acid stress. Each system relies on the activity of a proton-consuming reaction catalyzed by a specific amino acid decarboxylase and a corresponding antiporter. Activation of these three systems is tightly regulated by a sophisticated interplay of membrane-integrated and soluble regulators. Using a fluorescent triple reporter strain, we quantitatively illuminated the cellular individuality during activation of each of the three acid resistance (AR) systems under consecutively increasing acid stress. Our studies highlight the advantages of E. coli in possessing three AR systems that enable division of labor in the population, which ensures survival over a wide range of low pH values.

T he acquisition of acid resistance (AR) is an important property of Escherichia coli and many other neutralophilic bacteria that enables survival in acidic environments, such as the human gastrointestinal tract or acidic soils 1,2 . Although the cytoplasmic membrane is impermeable for protons, some protons enter the cytoplasm through protein channels, transient water chains, or damaged membranes 3,4 . Bacteria are generally able to maintain a fairly constant internal pH when grown in a wide range of media at different external pH values 5,6 . Despite this, pH homeostasis varies among individual bacterial cells as reported amongst others for E. coli and Bacillus subtilis 7 . Interestingly, persister cells of E. coli display a lower intracellular pH allowing survival after antibiotic treatment 8 . Nevertheless, after exposure of E. coli to an external pH of 5.8, the pH of its cytoplasm drops transiently; however, the pH rapidly returns to a neutral level due to the intrinsic buffering capacity of the cytoplasm or alterations in the flux of ions 9 . A lower external pH such as 4.4 causing a lower intracellular pH of around 6.0 has adverse effects on all macromolecules of a cell, which might result in lowered enzyme activity, acid-induced protein unfolding, membrane damage, and DNA damage.
Besides its passive mechanisms, E. coli possesses several inducible AR systems to counteract acidic environments 1,3 . The major systems are the glutamate decarboxylase (Gad) system, arginine decarboxylase (Adi) system, and lysine decarboxylase (Cad) system (AR2, AR3, and AR4, respectively) 10-12 . AR1 does not require an amino acid and is regulated by the alternative σfactor (RpoS) and the cAMP receptor protein 10,13 . The ornithine decarboxylase system (AR5) plays only a minor role in E. coli MG1655, yet has a more important role in avian pathogenic E. coli 14 . The core components of each of the three major AR systems are a proton-consuming amino acid decarboxylase and a cognate antiporter to excrete the decarboxylated (more alkaline) compounds in exchange with the corresponding extracellular amino acid. In this way, both the intracellular and extracellular pH values can be increased 3,11,15,16 . Each system is active at different external pH values and growth phases, and induction is regulated by a sophisticated interplay of membrane-integrated and soluble regulators.
Activation of the Gad System occurs during the transition of an E. coli culture to stationary phase and during exponential growth in acidified media. Furthermore, this system is essential for cell survival at an extremely low pH of 2.5 10,[17][18][19] . The Gad system employs GadA and GadB, two pyridoxalphosphate-dependent decarboxylases that catalyze the proton-consuming decarboxylation of L-glutamate to generate γ-aminobutyrate (GABA), and GadC, the cognate antiporter that performs the import of L-glutamate and the export of GABA 17,20 . It is important to note that the activity of GadC is pH-dependent, and the antiporter preferentially exchanges protonated glutamate (Glu 0 ) in exchange for protonated GABA (GABA + ) at an external pH of 3.0 and lower [21][22][23] . The induction of the gad genes is rather complex. The main transcriptional regulator is GadE, whose expression is regulated by the transcription factors EvgA, GadW, GadX, and YdeO as well as σ-factor RpoS. RpoS is activated and stabilized in response to different conditions, including the stationary phase, various stress treatments, and for example by the signal transduction cascade proceeding from PhoQ/PhoP to RssB and IraM 24,25 . Furthermore, the EvgS/EvgA histidine kinase/response regulator system is a primary detector of mild acidic environments (pH 5) and additionally regulates the activation of the Gad system through a cascade of EvgA-YdeO-GadE regulators 26,27 (Fig. 1). In addition, RcsB is a critical partner of GadE and the binding of both regulators as a heterodimer to the GAD box activates gadA transcription [28][29][30] .
The Adi system is responsible for the conversion of arginine to agmatine under the consumption of one proton (Fig. 1). The main components are the transcriptional activator AdiY, a member of the AraC-family; the arginine decarboxylase AdiA; and the arginine/agmatine antiporter AdiC. The genes encoding the Adi system have an unusual genomic arrangement in E. coli, as the gene of adiY is located downstream of adiA and upstream of adiC. The Adi system is maximally induced under the conditions of acidic pH (pH 4.4), anaerobiosis, and a rich medium 31 . The activity of AdiC is regulated in response to acidic pH and remains fully active at a pH of ≤6.0 32 . The expression of adiY seems to be indirectly influenced by the transcriptional repressor CsiR, as its overexpression results in the repression of adiY and adiA 16 . Under anaerobic conditions, the adiY mRNA can be base-paired by the small RNA SgrS resulting in posttranscriptional downregulation 33 .
The Cad system, which is activated when E. coli is exposed to pH 5.8 in the presence of lysine, uses external lysine, which is converted to cadaverine by the lysine decarboxylase CadA under the consumption of one proton. The lysine/cadaverine antiporter CadB imports lysine and excretes the more alkaline cadaverine, thereby elevating the external pH 34 . Expression of the cadBA operon is activated by the membrane-integrated one-component regulator CadC, which is a representative of the ToxR-family 35 (Fig. 1). The pH-sensory function and the feedback inhibition by cadaverine could be assigned to distinct amino acids within the periplasmic sensory domain of CadC 36,37 . The availability of external lysine is transduced to CadC via the co-sensor and inhibitor LysP, which is a lysine-specific transporter 38,39 . Recently, we demonstrated that not only the copy number of CadC affects the dual-stress response 40 , but also the noise of the target protein abundance. Using fluorophore-tagged CadB, a heterogeneous output in the single cells of E. coli occurred under mild acidic stress. Moreover, an increase in the CadC copy number was correlated with decreased heterogeneous behavior 41 .
The ON/OFF behavior of the E. coli population in activating the Cad system prompted us to ask whether the "Cad-OFF" cells activate the Gad and/or Adi systems. Here, we used a fluorescent triple reporter to study the induction of the three inducible AR systems, Gad, Adi, and Cad, in individual E. coli cells under consecutively increasing acid stress. We herein report extensive heterogeneity and division of labor in the acid stress response of individual E. coli cells. For the first time, we present a model that explains the cellular individuality that occurs during activation of each of the three AR systems in the context of ensuring AR for the whole population over a wide range of acidic pH values.

Results
Heterogenous activation of the three inducible AR systems by acid stress in E. coli. In our previous studies, we demonstrated that heterogenous activation of the Cad system results in~70% ON and 30% OFF cells 41 . Here, we constructed a three-color reporter strain E. coli (gadC:eGFP-adiC:mCerulean-cadB:mCherry) to study whether the "Cad-OFF" cells activate the Gad and/or the Adi systems and whether these two systems are activated heterogeneously. In this strain, each of the antiporter genes (gadC, adiC, and cadB) is fused with a different fluorophore gene coding for eGFP, mCerulean, or mCherry, respectively (Fig. 1). These fusions were chromosomally integrated to avoid a copy number effect, and the functionality of the hybrid proteins was confirmed ( Supplementary  Fig. 1).
First, the three-color reporter strain was exposed to consecutively increasing acidic conditions (pH 7.6, pH 5.8, and pH 4.4) in a well-mixed environment (Fig. 2a). Then, at the indicated time points, cells were imaged, and the fluorescence intensities were quantified (Fig. 2b).
GadC:eGFP showed the highest production during the stationary phase and under the acidic conditions of pH 5.8 and pH 4.4 ( Fig. 2b-d, t 300 ). In the stationary phase at physiological pH, GadC:eGFP was produced two times less than under acidic conditions; however, its heterogenous distribution increased, as indicated by noise values of 0.11 (defined as the standard deviation divided by the mean of log-transformed intensity values 42 ) (t 300 ) ( Fig. 2c; left panel). When cells were exposed to mild acid stress (pH 5.8), GadC:eGFP was homogenously distributed among individual cells in the E. coli population and exhibited low noise values of 0.04-0.08 (Fig. 2b, c; left panel).
The Cad system is activated by mild acidic stress in the presence of lysine; therefore, CadB:mCherry became visible after 2.5 h of growth in a minimal medium supplemented with lysine at pH 5.8. At this time point, CadB:mCherry was heterogeneously distributed, as reflected by a high noise value of 0.37 ( Fig. 2b-d), which is in perfect agreement with previous experiments of CadB tagged with eGFP 41 . The distribution of CadB:mCherry remained heterogeneous at pH 4.4 in complex medium, but the mean value increased 1.5-fold ( Fig. 2c; right panel, t 300 ).
The Adi system is only activated under stronger acidic conditions, such as pH 4.4 in a complex medium with a noise value of 0.12 (Fig. 2b, c; middle panel).
Overall, the distribution of CadB:mCherry among single E. coli cells under acid stress was not a symmetric Gaussian-like distribution but rather an asymmetric right-skewed distribution (Fig. 2d), as up to 83% of the population produced CadB:m-Cherry at varying high levels, and the remaining cells were in the OFF state, which was in agreement with previous findings using CadB:eGFP 41 . The distribution of GadC:eGFP and AdiC:mCerulean follows a symmetric Gaussian-like distribution (Fig. 2d), which indicated that individual cells produce different amounts of GadC:eGFP and AdiC:mCerulean. However, almost all cells produced varying amounts of GadC:eGFP (99%), whereas only half of the population (47.7%) produced AdiC:mCerulean at pH 4.4 ( Fig. 2d; right panel). As a control, non-tagged E. coli MG1655 cells were examined under the same conditions; these cells exhibited low background fluorescence and extremely low noise values (Supplementary Table 1).
To exclude pH-effects on the fluorophores, each of the three fluorophores (eGFP, mCherry, and mCerulean) was fused to the antiporter gene gadC, because GadC is produced under all tested external pH values. Neither the different fluorophores nor the external pH affected the output or noise of the fluorescent hybrids (Supplementary Fig. 2; Supplementary Table 1).
Overall, the three inducible AR systems were heterogeneously activated with a different degree of noise. The Cad system showed the highest heterogeneous distribution (pH 5.8 t 150 ) of all three systems, whereas the Gad and Adi systems showed comparable noise values (0.11 and 0.12, respectively), but under different conditions. The Gad system was the most heterogenous during the stationary phase, whereas the Adi system was activated heterogeneously by stronger acid stress (pH 4.4).
Simultaneous activation of the AR systems in single E. coli cells. Since the three AR systems showed a heterogenous output, although each had a different strength of heterogeneity, we examined their simultaneous activation within individual E. coli cells under exposure to consecutively increasing acid stress (Fig. 2a).
The correlation between GadC:eGFP and CadB:mCherry was analyzed at the time point of the highest degree of heterogeneity of the Cad system (pH 5.8 t 150 ; Fig. 3a); however, no correlation occurred, thus indicating that there was a lack of dependency Fig. 1 The regulatory network of inducible amino acid decarboxylase-antiporter systems in E. coli three-color strain. Regulated induction of the Gad (blue symbols), Adi (purple symbols), and Cad (yellow symbols) systems requires a network of membrane-integrated pH-sensors, namely EvgS and CadC; interconnected transcription factors; and the sRNA GadY. The glutamate decarboxylases GadA and GadB convert glutamate (Glu) into γ-aminobutyrate (GABA), which is excreted by GadC. The arginine decarboxylase AdiA converts arginine (Arg) into agmatine (Agm), which is excreted by AdiC. The lysine decarboxylase CadA converts lysine (Lys) into cadaverine (Cad), which is excreted by CadB. Adapted from 3,15,24,25,47,74 . The stimuli leading to an induction of each of the three AR systems is indicated below each system. Each fluorescent hybrid of the three antiporter is indicated with a star symbol: GadC:eGFP (green star), AdiC:mCerulean (blue star), and CadB:mCherry (red star) according to the hybrid with the fluorophore eGFP, mCerulean, and mCherry, respectively. Fig. 2 Heterogenous activation of the three inducible AR systems in E. coli in response to acid stress. a Schematic representation of the experimental setup: at the exponential phase (t 0 ), the three-color reporter strain E. coli gadC:eGFP-adiC:mCerulean-cadB:mCherry was shifted from the MM at pH 7.6 to MM at pH 5.8 supplemented with 10 mM lysine, followed by a second shift to the LB medium at pH 4.4. In parallel, half of the culture was incubated under the original pH condition. b Fluorescent microscopic images of the three-color reporter strain E. coli gadC:eGFP-adiC:mCerulean-cadB:mCherry, cultivated as presented in (a), were taken at the indicated time points. Representative fluorescence overlay images are shown. Scale bar, 5 µm. c Quantified noise and mean RF were calculated for 1000 cells per condition and time point of the cultivated three-color reporter strain E. coli gadC:eGFP-adiC:mCerulean-cadB:mCherry, as presented in (a). Noise (standard deviation/mean of log-transformed values) is presented by the size of the dots (the higher the noise, the larger the size of the dot), and the average of the RF is presented with a color code for each fluorescent hybrid, GadC:eGFP (left panel), AdiC:mCerulean (middle panel), and CadB:mCherry (right panel). RF was quantified by using the MicrobeJ plugin of the ImageJ software of the fluorescent microscopic images. d Histogram presentation of the nRF quantified for 1000 cells per fluorescent hybrid, GadC:eGFP, AdiC:mCerulean, and CadB:mCherry, grown at t 300 in MM pH 7.6 (left panel), MM at pH 5.8 supplemented with 10 mM lysine (middle panel), and LB medium at pH 4.4 (right panel), respectively. A comparison of the frequencies of CadB:mCherry with AdiC:mCerulean and with GadC:eGFP using the Chi-square test showed a p-value < 0.0001 for each time point. LB lysogeny broth, MM minimal medium, nRF normalized relative fluorescence intensity, RF relative fluorescence intensity. between these two systems (Pearson's correlation coefficient, r = 0.11). During the stationary phase under strong acid stress (pH 4.4 t 300 ), when all three systems were activated, GadC:eGFP and AdiC:mCerulean showed a positive correlation (r = 0.45; green dots); however, no correlation occurred between CadB:m-Cherry and GadC:eGFP or AdiC:mCerulean (r = 0.10 and r = −0.07, respectively) (Fig. 3b). In summary, in stationary phase under strong acid stress, almost all cells activated the Gad system; however, simultaneous activation of the Cad and Adi systems did not occur.
Phylogenetic distribution of the components of the three AR systems within the bacterial kingdom. To identify connecting regulators between the three AR systems and to understand the segregation of the Adi and the Cad systems in E. coli, we used a bioinformatic approach to investigate the presence and distribution of regulatory components of the three AR systems within the bacterial kingdom. We used the antiporters GadC, AdiC, and CadB as the basis for the construction of phylogenetic trees and focused on discovering a potential co-occurrence of their specific regulators ( Fig. 1; Supplementary Table 2).
The antiporter CadB of the Cad system was found mainly in γproteobacteria. 857 homologs were identified belonging to Enterobacteriaceae (55.8%), Vibrionaceae (20.0%), Aeromonadaceae (7.0%), Hafniaceae (5.6%), and Yersiniaceae (4.7%). CadB mainly co-occurs with CadA and CadC homologs. Of the species that have a CadB homolog, 94.3% also possess a CadA homolog of the cognate lysine decarboxylase of the Cad system, and 95.5% also have a homolog of the cognate pH-sensor CadC (Fig. 4c). In total, 573 homologs of the lysine-permease LysP were identified, 66.9% of which were found in species containing a CadB homolog. As previously reported, LysP predominantly co-occurs with the Cad system within the Enterobacteriaceae family (82.3%) and is mostly absent in Vibrionaceae ( Fig. 4c; Supplementary Table 2) 41 . CsiR homologs are less coupled with CadB compared to other components of the Cad system. However, similar to LysP, CsiR is mainly found in Enterobacteriaceae and Hafniaceae, such as Escherichia (49.1%), Klebsiella (23.4%), Salmonella (14.6%), and Hafnia (4.1%) species. In summary, the regulatory components specific to E. coli of the three AR systems that were studied were predominantly conserved in Enterobacteriaceae. The Gad system was the most widespread AR system, whereas the Adi and Cad systems were restricted to γ-proteobacteria. Interestingly, the decarboxylases, together with their cognate antiporters, were more widely distributed than other regulatory components. The central regulators of the Cad system and the Adi system were more specific to the Enterobacteriaceae family.
Mechanisms of heterogenous activation of the three AR systems. To gain insight into the molecular mechanisms of heterogeneous activation of the AR systems in E. coli, we analyzed the importance of different transcriptional regulators. We focused mainly on the Adi and Cad systems, as these systems exhibited a higher degree of heterogeneity. First, we assessed the role of the regulator CsiR, which co-occurs with the components of the Gad, Adi, and Cad systems, mainly in Enterobacteriaceae (Fig. 4), and is supposed to modulate AR in E. coli 16 . However, artificially increased CsiR levels affected only the Adi system and not the Gad or Cad systems (Supplementary Table 1). Higher intracellular CsiR levels caused a 2.4-fold decrease in the amount of AdiC:mCerulean but did not affect the heterogenous distribution. The measured noise value was 0.10 under these conditions and is consistent with the wild type (Supplementary Table 1). These results are in good agreement with the suggested repressing impact of CsiR on the Adi system according to ChiP-Seq analysis 16 .
Second, we already know that the native low CadC copy number generates heterogenous activation of the Cad system, as an increase in the CadC copy number correlates with a decrease in heterogeneous distribution of the Cad system 41 . In addition, CadC itself is distributed heterogeneously 43 . Because of the mutually exclusive activation of the Cad and Adi systems (Fig. 3b), we tested whether CadC also has an influence on the Adi system. When we artificially elevated the copy number of CadC, we found four-times lower mean fluorescence of AdiC:mCherry (Fig. 5a). Moreover, the heterogenous distribution of AdiC:mCherry was much lower under this condition (Fig. 5a). It should be noted that the higher level of CadC decreased the amount of GadC:mCerulean by only two-fold in cells exposed to pH 5.8, and the already less heterogenous distribution of the Gad system was not affected (noise value of 0.05, which is comparable to that of the wild type) (Supplementary Table 1). In agreement with previous observations, the higher level of CadC increased the production of CadB:eGFP while reducing its heterogeneous distribution (Supplementary Table 1).
Third, AdiY also belongs to low copy number regulators with approximately 11-37 copies per cell 44 . Based on our previous experience with CadC 41 , we analyzed the influence of the copy number of AdiY on heterogenous activation of the Adi system in pH 4.4-stressed cells. We increased the copy number by placing adiY under the control of the P BAD promoter on a plasmid. The artificially increased copy number of AdiY resulted in a two-fold increase in the mean fluorescence and thus a higher amount of AdiC:mCerulean, which is not surprising, since AdiY is the main transcriptional activator of adiC. Furthermore, the heterogeneous distribution of AdiC:mCerulean was strongly reduced (Fig. 5b). However, E. coli cells producing a higher copy number of AdiY had a growth disadvantage under acid stress compared with wildtype cells (Fig. 5c). A similar effect on growth was previously demonstrated for E. coli cells producing a higher CadC copy number 41 (Supplementary Fig. 3).
The effect of CadC on the Adi system could be direct or indirect. A direct effect, i.e., that CadC binds upstream of adiC as a transcriptional regulator, can be excluded, as we did not find a second binding site for CadC in E. coli 43,45,46 . Therefore, we hypothesized that an indirect effect was caused by a CadC- Fig. 4 Phylogenetic trees of the E. coli antiporters GadC, AdiC, and CadB and co-occurring regulators of the three AR systems within the bacterial kingdom. a The protein sequences of 1112 of E. coli GadC homologs were aligned, and a phylogenetic tree was generated, which is shown as a circular cladogram. The branches of the tree were colored according to the family of the organisms containing a GadC homolog. The presence of different regulatory components CsiR, EvgS, GadE, GadW, GadX, GadY, and YdeO and the decarboxylases GadA and GadB (GadA/B) is indicated by solid colors in the respective ring layer. Translucent colors represent components that were not detected. b The protein sequences of 756 E. coli AdiC homologs were aligned, and a phylogenetic tree was generated, which is shown as a circular cladogram. The branches of the tree are colored according to the family of organisms containing an AdiC homolog. The presence of the different regulatory components CsiR and AdiY as well as the decarboxylase AdiA is indicated by solid colors in the respective ring layer. If these components were not found, the colors are translucent. c The protein sequences of 857 E. coli CadB homologs were aligned, and a phylogenetic tree was generated, which is shown as a circular cladogram. The branches of the tree are colored according to the family of organisms containing a CadB homolog. The presence of the different regulatory components CsiR, CadC, and LysP as well as the decarboxylase CadA are indicated by solid colors in the respective ring layer. Translucent colors represent components that were not detected. Identifiers of all GadC, AdiC, and CadB homologs with an indication of the presence or absence of their main regulators are summarized in Supplementary Data 1.
induced increase in H + -consuming lysine decarboxylase CadA, thus conferring moderate AR to all cells (homogeneous response; Fig. 6a). To test this hypothesis, we deleted cadA from our reporter strain (ΔcadA gadC:eGFP-adiC:mCerulean-cadB:m-Cherry). Deletion of cadA reduced the heterogenous distribution of AdiC:mCerulean and strongly increased the mean fluorescence of AdiC:mCerulean (by 11-fold), thus suggesting that many more cells produced AdiC:mCerulean compared with the wild type (Fig. 6a). Deletion of cadA did not affect the heterogenous distribution of either GadC:eGFP or CadB:mCherry in pH-4.4stressed cells. Only the overall mean level of CadB:mCherry was increased under this condition, which was due to the lack of negative feedback inhibition of cadaverine on CadC (Supplementary Table 1) 34,36 .
To further support these results, we analyzed AdiY-driven pHdependent activation of the adiA promoter at the population level. In wild-type cells, adiA expression began when cells were exposed to an acidic environment of pH < 4.8 (Fig. 6b). The increased number of CadC molecules did not change the pHdependent induction profile but caused downregulation of the adiA promoter activity. In contrast, in the cadA mutant, the promoter activity of adiA not only increased, but the onset of induction shifted to a higher external pH such as 5.8 (Fig. 6b). These data suggest that activation of the Adi system is influenced by the intracellular pH, as a manipulation of the cytosolic proton concentration by either deletion of cadA or overproduction of cadA (triggered by increased CadC copy numbers) leads to increased activation or downregulation, respectively, of the Adi system in E. coli.

Discussion
To counteract acidic environments, many bacteria possess inducible AR systems that rely on H + -consuming amino acid decarboxylases and their corresponding antiporters. The latter function as importers for corresponding amino acids and exporters for decarboxylated products. The number and complexity of the three inducible Gad, Adi, and Cad AR systems vary among bacteria and reflect an adaptation to the needs of their individual natural habitat 47 . Our phylogenetic analysis of the distribution of antiporters GadC, AdiC, and CadB revealed that the Gad system is the most widely distributed AR system, whereas the Adi system and Cad system are restricted to γ-proteobacteria (Fig. 4). Consistent with previous reports, Salmonella has Adi and Cad systems, whereas Shigella has the Gad system 47 . In Shigella, we also identified the Adi system, and in some species, the Cad system, i.e., in S. boydii and S. flexneri (Supplementary Table 2, Supplementary Data 1). However, not only does the number of  Table 1). RF was quantified using the MicrobeJ plugin of the ImageJ software of the fluorescent microscopic images for 1000 cells per condition. nRF normalized RF values. c E. coli MG1655 wild type transformed with the plasmid pBAD24-adiY or with empty pBAD24 were grown in minimal medium at pH 5.8 supplemented with lysine and then shifted to LB medium of the indicated pH values. Growth (OD 600 ) was determined every 10 min in microtiter plates with continuous shaking. All experiments were performed three times (n = 3), and error bars represent standard deviation of the means. LB lysogeny broth.
AR systems possessed by bacteria differ, but the complexity of their regulation varies (Fig. 4). The most sophisticated regulatory network of AR systems is found in E. coli. Its close relatives E. albertii have almost all components as well but lack the EvgSA two-component system and YdeO.
For the first time, we investigated the activation of the three major AR systems in E. coli simultaneously at the single-cell level under consecutive increasing acid stress. By using the three-color reporter strain E. coli gadC:eGFP-adiC:mCerulean-cadB:mCherry, we confirmed that each of the three AR systems is specialized for a certain strength of acid stress under certain environmental conditions. The Gad system was activated under mild acid strength and during the stationary phase. The Cad system was induced at pH 5.8 and required the presence of lysine. The Adi system was induced at pH 4.4 and required the lysogeny broth medium, which contains tryptone and yeast extract. However, activation of the three systems differs greatly from cell to cell (Figs. 2 and 7). The Cad system showed the highest heterogeneous distribution of all three systems, when activated at pH 5.8 in the presence of lysine. Under almost all conditions, all cells of the population activate the Gad system but to varying degrees. The Adi system is activated heterogeneously under strong acid stress (pH 4.4) (Fig. 2).
Phenotypic variations can be beneficial for the E. coli population under acid stress. Our results provide a model of how the three inducible AR systems overlap and generate a division of labor and functional cooperation in the population (Fig. 7). All cells of the E. coli population individually adapt to mild acid stress by activating the Gad system to varying degrees, and the glutamate decarboxylases GadA and GadB might initially utilize intracellularly available glutamate under the consumption of protons. As described below, the antiporter GadC becomes more important under extreme acid stress. It should also be noted here that the decarboxylase GadB also undergoes a pH-dependent conformational change and exhibits an activity optimum at low pH 48 . Furthermore, glutamate is the most abundant intracellular metabolite, with a concentration of 100 mM, and its concentration is much higher compared to that of lysine (0.41 mM) and arginine (0.57 mM) 49 . Under stronger acid stress (pH 5.8 to pH 4.4), some cells in the population activate the Cad system, whereas others activate the Adi system. These two subpopulations each contribute to acid stress relief by secreting the more alkaline cadaverine and agmatine, respectively (Fig. 7). These two polyamines are considered common goods and contribute to an increase of the extracellular pH, which benefits the whole population. This behavior can be considered as an example of true division of labor in bacteria, as the metabolic burden of the individual cells to produce both the Adi and Cad systems is prevented, but nevertheless, the whole population benefits from the elevation of external pH by secretion of cadaverine and agmatine (Fig. 7). The fact that homogenous production of the components of the Adi system in all cells would be a burden for Fig. 6 Manipulation of intracellular stress resistance by either eliminating (ΔcadA) or increasing (+cadC) the activity of the Cad system affects the degree of heterogeneity and pH-dependent induction of the Adi system. a Normalized RF values of cells expressing AdiC:mCerulean in the E. coli wild type background (blue) or the cadA mutant (gray). Cells were monitored after growth at pH 4.4 in LB medium at t 300 . On top of the histogram, the cells are shown in a schematic overview, with blue cells representing the AdiC:mCerulean ON state and yellow dots representing CadA. Below the histogram, the noise (standard deviation/mean of log-transformed values) is presented by the size of the dots (the higher the noise, the larger the size of the dot), and the average RF is represented by a blue color (the more intense the blue, the higher the average relative fluorescence intensity). RF was quantified using the MicrobeJ plugin of the ImageJ software of the fluorescent microscopic images for 1000 cells per condition. The calculated mean RF and noise values are summarized in Supplementary Table 1. nRF normalized RF values. b E. coli MG1655 wild type and the cadA mutant (each transformed with the plasmid pBBR1-MCS5-P adiA -lux) were grown in MM at pH 5.8 supplemented with 10 mM lysine and then shifted to LB medium of the indicated pH values. Luminescence and growth were determined every 10 min in microtiter plates with a Tecan Infinite F500 system (Tecan, Crailsheim, Germany). Data are reported as relative light units (RLUs) in counts per second per milliliter per OD 600 , and maximal RLU at 1.9 h is shown. All experiments were performed three times (n = 3), and error bars represent standard deviation of the means. Growth over 10 h is presented in Supplementary Fig. 3 at selected pH values. LB lysogeny broth, MM minimal medium, RF relative fluorescence intensity.
the population is clearly shown by the reduced growth rate under this condition (Fig. 6c), which was similar for a homogenous production of the Cad system 41 (Supplementary Fig. 3).
We were then interested in the molecular mechanisms that not only lead to this phenotypic heterogeneity but that are also heritable. CadC and AdiY are low copy number proteins; therefore, they are stochastically distributed among the cells of the population 44 . Here and in our previous studies 41 , we showed that an increase of their copy number was sufficient to shift the population into homogeneous behavior. Cells that activated the Cad system are protected against acid stress due to the H + -consuming activity of the CadA decarboxylase. It should be noted that CadA is produced on average with 60,000 copies per cell, which accounts for at least 2% of all cytoplasmic proteins in CadA-producing cells 50 . In cells lacking the Cad system, the intracellular pH decreases under strong external acid stress, such as pH 4.4. Our results have shown that any manipulation of the activity of the Cad system, such as its elimination (ΔcadA) or increase (+cadC), affected the degree of heterogeneity and pHdependent induction of the Adi system, which suggests that the regulator of the Adi system is an intracellular pH sensor that senses a decrease of cytoplasmic pH.
Induction of the Gad system of E. coli is controlled by an unusually high number of transcription factors, the stationary sigma factor, and a small RNA GadY (Figs. 1 and 4a). Moreover, its activation is strongly influenced by the medium composition and growth phase 10,17 . The transcription factors GadX and GadW were already associated with higher noise levels of their target genes of unstressed cells 42 and potentially contribute to the heterogenous distribution of GadC:eGFP (Fig. 2b). In addition, the stationary sigma factor RpoS is known generate extensive transcriptional heterogeneity [51][52][53] . Thus, the Gad system integrates multiple extracellular and intracellular inputs, which results in quantitative differences and continuous variation of phenotypic traits in the individual cells of the population. The noisy activation of the gadBC promoter enables prediction of the survival of single cells 54 . Usually, high levels of noise are found for genes that are involved in either the energy metabolism of carbon sources or adaptation to stress, like osmotic pressure, temperature extremes, starvation response, pH response, and mechanical, nutritional, or oxidative stress. Indeed, the Gad system is a major system under extreme acid stress (pH ≤ 2.5) 10 . The antiporter GadC preferentially transports protonated glutamate (Glu + ) in exchange for protonated GABA (GABA + ) at an external pH of 3.0 and lower [21][22][23] . This transport activity is crucial for protecting cells against incoming protons. It should be noted that at extremely low pH values, such as pH 2.5, only <0.01% of a non-preadapted population survives 27 , and therefore, for maintenance of the population, it is extremely important that all E. coli cells produce the Gad system, albeit with different levels.
We can only speculate to what extent this model might hold up in a natural epitope of E. coli, such as digestive system of humans or animals. Before E. coli arrives in the intestine, it is exposed to an extreme acid stress with HCl in the stomach that this bacterium can only survive with the help of the Gad system. The mean pH in the proximal small intestine is 6.6 and increases in the terminal ileum to 7.5. Then, there is a sharp fall in pH to 6.4 in the caecum, and an increase in colon with a final value of 7.0 55 . Depending on the availability of the amino acids lysine and arginine, the Cad and Adi systems might be switched on under these conditions. The latter system is increasingly activated due to the anaerobic conditions 31 . Both systems will still be heterogeneously distributed due to the low copy number of the main regulators CadC and AdiY. Fermentation products, such as acetate, also activate the Gad system 56 . The extent to which microcolonies of E. coli exist in the colon, in which common goods such as cadaverine and agmatine can change the micromilieu, is also unclear, as is the extent to which the host plays a role in acid stress. In fact, the levels of putrescine, a conversion product of agmatine, and cadaverine were found to be 15-and 3fold, increased, respectively, in the inflamed mouse intestine mono-colonized with E. coli compared to an equally colonized healthy mouse cohort 57 .
Considered together, our results provide a model for the advantage of E. coli to have three AR systems that allow for division of labor in the population and ensure its survival over a wide range of low pH values, thus making this bacterium highly acid-stress resistant.

Methods
Bacteria and growth conditions. Bacterial strains and plasmids used in this study are listed in Table 1. E. coli strains were cultivated in LB medium (10 g/liter NaCl, 10 g/l tryptone, 5 g/l yeast extract) or in Kim-Epstein (KE) minimal medium 58 adjusted to pH 5.8 (MM pH 5.8) or pH 7.6 (MM pH 7.6), using the corresponding phosphate buffer. E. coli strains were incubated aerobically in a rotary shaker at 37°C. KE medium was always supplemented with 0.2% (w/v) glucose. When indicated lysine was added to a final concentration of 10 mM.
If necessary, media were supplemented with 100 µg/ml ampicillin, or 50 µg/ml kanamycin sulfate. To allow the growth of the conjugation strain E. coli WM3064, meso-diamino-pimelic acid (DAP) was added to a final concentration of 300 µM.
Construction of plasmids. Molecular methods were carried out according to standard protocols or according to the manufacturer's instructions. Kits for the isolation of plasmids and the purification of PCR products were purchased from Süd-Laborbedarf (SLG; Gauting, Germany). Enzymes and HiFi DNA Assembly Master Mix were purchased from New England BioLabs (Frankfurt, Germany).
To construct different fluorescent fusions with gadC, adiC and cadB, to be inserted in-frame chromosomally, the corresponding flanking regions (600 bp upstream and downstream) of gadC, adiC or cadB were amplified by PCR using MG1655 genomic DNA as template. Plasmid pET-mCherry-cadC 43 was used as Fig. 7 Model of the heterogenous activation of three inducible acid resistance systems and division of labor in the E. coli population. Fluorescent microscopic images of the three-color reporter strain E. coli gadC:eGFP-adiC:mCerulean-cadB:mCherry at pH 4.4 in a complex medium at t 300 . All cells of the E. coli population individually adapt to acid stress by activating the Gad system to varying degrees (green fluorescent cells) due to multiple extracellular and intracellular inputs. These cells utilize primarily intracellularly available glutamate under consumption of protons to increase their internal pH. Under stronger acid stress (pH 5.8 to pH 4.4), some cells in the population activate the Cad system (red fluorescent cells), while others activate the Adi system (blue fluorescent cells). These cells secrete (as indicated by the rings) the more alkaline products cadaverine and agmatine, respectively, thereby contributing to acid-stress relief with an increase of the extracellular pH, which benefits the whole population. In addition, their internal pH is elevated by consuming protons due to the conversion of lysine to cadaverine and arginine to agmatine, respectively.
Functional hybrid proteins of GadC with each of the three different fluorophores was only achieved by using a shorter version of GadC (amino acids 1-470), in which the C-terminal C-plug is removed. Truncation of this C-plug does not affect the transport activity, but shifts the pH-optimum to a higher pH 21 .
To construct a marker-less in-frame deletion of cadA in E. coli MG1655 gadC:eGFP-adiC:mCerulean-cadB:mCherry, the suicide plasmid pNPTS138-R6KT-ΔcadA was generated. Briefly, flanking regions (1000 bp upstream and downstream) of cadA were amplified by PCR using MG1655 gadC:eGFP-adiC:mCerulean-cadB:mCherry genomic DNA as template. After purification of the different fragments, the DNA fragments were assembled via Gibson assembly 60 into by PCR  linearized pNPTS138-R6KT plasmid, resulting in the plasmid pNPTS138-R6KT-ΔcadA. Correct insertion was verified by sequence analysis using primer pNTPS_Seq_fwd. For construction of the reporter plasmid pBBR1-MCS5-P adiY -lux, 200 bp of the region upstream of adiA was amplified by PCR using primers (PadiA_XbaI_fwd and PadiA_XmaI_rev) and MG1655 genomic DNA as template, and cloned into plasmid pBBR1-MCS5-TT-RBS-lux using restriction sites XbaI and XmaI. Correct insertion was verified by colony PCR and sequencing.
For construction of N-terminal His 6 -tagged AdiY and His 6 -tagged CsiR, adiY and csiR were amplified by PCR using MG1655 genomic DNA as template. The codons for the His 6 -tag were introduced by using a respective forward primer. After purification of the different fragments, the DNA fragments were assembled via Gibson assembly 60 into SmaI-digested pBAD24 plasmid, resulting in the plasmids pBAD24-His 6 -adiY and pBAD24-His 6 -csiR. Correct insertion was verified by colony PCR and sequencing.
Construction of chromosomally integrated fluorescent fusions and deletion strains. The genes encoding the three different fluorophores, egfp, mCherry and mCerulean, are separately C-terminally fused to the genes gadC, adiC and cadB encoding the antiporter of the three AR systems. Expression of each fusion is under the control of the respective native promoter.
To generate the three-color E. coli MG1655 gadC:mCerulean-adiC:mCherry-cadB:egfp strain, at first the suicide plasmid pNPTS138-R6KT-adiC:mCherry was introduced into E. coli MG1655 cadB:egfp by conjugative mating using E. coli WM3064 as a donor in LB medium containing DAP. Single-crossover integration mutants were selected on LB plates containing kanamycin but lacking DAP. Single colonies were grown over a day without antibiotics and plated onto LB plates containing 10% (w/v) sucrose but lacking NaCl to select for plasmid excision. Kanamycin-sensitive colonies were checked for targeted deletion by colony PCR using primers bracketing the site of the insertion. Insertion of mCherry was verified by colony PCR and sequencing resulting in the strain E. coli MG1655 adiC:mCherry-cadB:egfp. In the next step tagging of gadC with mCerulean was achieved using the suicide plasmid pNPTS138-R6KT-gadC:mCerulean. The plasmid pNPTS138-R6KT-gadC:mCerulean was introduced into E. coli MG1655 adiC:mCherry-cadB:egfp by conjugative mating using E. coli WM3064 as a donor in LB medium containing DAP as described above. Insertion of mCerulean was verified by colony PCR and sequencing, resulting in the strain E. coli MG1655 gadC:mCerulean-adiC:mCherry-cadB:egfp.
To generate the three-color E. coli MG1655 gadC:mCherry-adiC:mCerulean-cadB:egfp strain, at first the suicide plasmid pNPTS138-R6KT-adiC:mCerulean was introduced into E. coli MG1655 cadB:egfp by conjugative mating using E. coli WM3064 as a donor in LB medium containing DAP as described above. Insertion of mCerulean was verified by colony PCR and sequencing, resulting in the strain E. coli MG1655 adiC:mCerulean-cadB:egfp. In the next step tagging of gadC with mCherry was achieved using the suicide plasmid pNPTS138-R6KT-gadC:mCherry. The plasmid pNPTS138-R6KT-gadC:mCherry was introduced into E. coli MG1655 adiC:mCerulean-cadB:egfp by conjugative mating using E. coli WM3064 as a donor in LB medium containing DAP as described above. Insertion of mCherry was verified by colony PCR and sequencing resulting in the strain E. coli MG1655 gadC:mCherry-adiC:mCerulean-cadB:egfp.
To generate the two-color E. coli MG1655 gadC:mCerulean-cadB:mCherry strain, at first the suicide plasmid pNPTS138-R6KT-cadB:mCherry was introduced into E. coli MG1655 by conjugative mating using E. coli WM3064 as a donor in LB medium containing DAP as described above. Insertion of mCherry was verified by colony PCR and sequencing, resulting in the strain E. coli MG1655 cadB:mCherry. In the next step tagging of gadC with mCerulean in this strain was achieved using the suicide plasmid pNPTS138-R6KT-gadC:mCerulean. The plasmid pNPTS138-R6KT-gadC:mCerulean was introduced into E. coli MG1655 cadB:mCherry by conjugative mating using E. coli WM3064 as a donor in LB medium containing DAP as described above. Insertion of mCerulean was verified by colony PCR and sequencing resulting in the strain E. coli MG1655 gadC:mCerulean-cadB:mCherry.
To generate the three-color E. coli MG1655 gadC:egfp-adiC:mCerulean-cadB:mCherry strain, at first the suicide plasmid pNPTS138-R6KT-adiC:mCerulean was introduced into E. coli MG1655 cadB:mCherry by conjugative mating using E. coli WM3064 as a donor in LB medium containing DAP as described above. Insertion of mCerulean was verified by colony PCR and sequencing, resulting in the strain E. coli MG1655 adiC:mCerulean-cadB:mCherry. In the next step tagging of gadC with egfp was achieved using the suicide plasmid pNPTS138-R6KT-gadC:egfp. The plasmid pNPTS138-R6KT-gadC:egfp was introduced into E. coli MG1655 adiC:mCerulean-cadB:mCherry by conjugative mating using E. coli WM3064 as a donor in LB medium containing DAP as described above. Insertion of egfp was verified by colony PCR and sequencing, resulting in the strain E. coli MG1655 gadC:egfp-adiC:mCerulean-cadB:mCherry.
In vivo fluorescence microscopy and data analysis. To analyze the spatial temporal localization of the different fluorescent hybrids in E. coli, overnight cultures were prepared in KE medium pH 7.6 and aerobically cultivated at 37°C. The overnight cultures were used to inoculate day cultures (OD 600 of 0.1) in fresh medium at pH 7.6. At an OD 600 of 0.5 (t 0 ), half of the culture was gently centrifuged and resuspended in KE medium pH 5.8 + lysine as the first pH shift. The rest of the culture continued growing in KE medium pH 7.6 at 37°C. Then the cultures were cultivated aerobically at 37°C for 2.5 h (t 150 ), and half of the culture was gently centrifuged and resuspended in LB medium pH 4.4 as the second pH shift. The rest of the culture continued growing in KE medium pH 5.8 + lysine at 37°C (t 300 ). At the beginning of the experiment (t 0 ) and at every pH shift (t 150 and t 300 ), 2 µl of the culture was spotted on 1% (w/v) agarose pads (prepared with the respective media), placed onto microscope slides and covered with a coverslip. Subsequently, images were taken on a Leica DMi8 inverted microscope equipped with a Leica DFC365 FX camera (Wetzlar, Germany). An excitation wavelength of 546 nm and a 605-nm emission filter with a 75-nm bandwidth was used for mCherry fluorescence with an exposure of 500 ms, gain 5, and 100% intensity for mCherry-tagged strains. An excitation wavelength of 485 nm and a 510-nm emission filter with a 75-nm bandwidth was used for eGFP fluorescence with an exposure of 500 ms, gain 5, and 100% intensity for eGFP-tagged strains. An excitation wavelength of 436 nm and 480-nm emission filter with a 40-nm bandwidth was used for mCerulean fluorescence with an exposure of 500 ms, gain 5, and 100% intensity for mCerulean-tagged strains.
To analyze the influence of an increased copy number of CsiR, the E. coli strains gadC:mCerulean-cadB:mCherry and adiC:mCerulean-cadB:mCherry were transformed with plasmid pBAD-His 6 -csiR by electroporation. 0.1% (w/v) L-arabinose was added throughout the experiment and cells were cultivated as described above.
To analyze the influence of an increased copy number of AdiY, the E. coli strain adiC:mCerulean-cadB:mCherry was transformed with plasmid pBAD-His 6 -adiY by electroporation. 0.1% (w/v) L-arabinose was added throughout the experiment and cells were cultivated as described above.
To analyze the influence of an increased copy number of CadC, the E. coli strains gadC:mCerulean-cadB:mCherry and adiC:mCherry-cadB:eGFP were transformed with plasmid pBAD-His 6 -adiY by electroporation. 0.1% (w/v) L-arabinose was added throughout the experiment and cells were cultivated as described above.
To quantify relative fluorescent intensities (RF) representing fluorescent hybrids of single cells, phase contrast and fluorescent images were analyzed using the ImageJ 61 plugin MicrobeJ 62 . Default settings of MicrobeJ was used for cell segmentation (Fit shape, rod-shaped bacteria) apart from the following settings: area: 0.1-max µm 2 ; length: 1.2-5 µm; width: 0.1-1 µm; curvature 0.−0.15 and angularity 0.−0.25 for E. coli cells. In total >1000 cells were quantified per strain and condition and time point. The background of the agarose pad was subtracted from each cell per field of view.
Mean and standard deviation (std) of the relative fluorescence (RF) were quantified using MicrobeJ and background was subtracted. Moreover, the RF of the GFP channel was corrected for the overlapping signal of the CFP channel via linear regression with the experimentally determined proportionality factor of 0.456. The noise was defined as std/mean of log-transformed values per sample. Logtransformed values were used because the coefficient of variation, as a measure of noise, is based on the assumption that the underlying data are normally distributed. However, not all data were normally distributed under all conditions, therefore the log-transformed values were used (log(x + 1)) as previously published for CadB:eGFP 41 . In addition, this helped to overcome the problem that the noise becomes higher with lower gene expression, which is often observed 42 .
In order to compare fluorescence intensity values of the different fluorophores, the RF values were normalized (nRF) according to the highest values, as each of the three fluorophores has a different brightness (eGFP: 33; mCerulean: 16; mCherry: 15) 63 .
Statistical analysis and presentation were performed using GraphPad Prism 9.1.0 and R 4.0.3 64 .
Functionality of fluorescent antiporter hybrids. The functionality of the Cad-B:eGFP hybrid was previously confirmed using a liquid-based colorimetric assay with a pH indicator 41 . To test the functionality of the fluorescent hybrid proteins GadC:eGFP and AdiC:mCerulean, an acid survival assay was performed as described previously 10,41 with the following modifications. E. coli strains were cultivated in LB medium (pH 7.6) to an OD 600 of 0.6-0.8, then the cultures were adjusted to an OD 600 of 0.5 and resuspended in LB medium with a pH of 3.0 or 4.4. The low pH challenge was conducted at 37°C and samples were collected immediately after resuspension (t = 0) and then hourly for 3 h. Samples were serially diluted and plated onto LB agar plates to assess the number of colonies surviving the acid challenge. As controls, the parental E. coli MG1655 and the gadC and adiC deletion mutants were used.
Measurement of adiA promoter activity in vivo. In vivo promoter activity of adiA was probed with a luminescence-based reporter (P adiA -luxCDABE). The influence of the Cad system and the AdiY copy number was tested in cells exposed to different external pH values in LB medium using luminescence as readout. The strains E. coli MG1655 carrying pBBR1-MCS5-P adiA -lux, E. coli MG1655 ΔcadA carrying pBBR1-MCS5-P adiA -lux, E. coli MG1655 carrying pBBR1-MCS5-P adiA -lux co-transformed with pBAD-cadC were incubated in KE medium pH 7.6 supplemented with the respective antibiotics overnight. The overnight cultures were inoculated to an OD 600 of 0.1 in fresh KE medium pH 7.6, aerobically cultivated until exponential phase and then shifted to KE medium pH 5.8 + 10 mM lysine and cultivated for another 2.5 h at 37°C (comparable to the experimental setup described in Fig. 2a). In the next step, the cultures were shifted into a 96-well plate and aerobically cultivated at 37°C in LB medium at different pH values supplemented with the respective antibiotics. Bioluminescence and growth were determined every 10 min in the microtiter plates with a Tecan Infinite F500 system (Tecan, Crailsheim, Germany). Data are reported as relative light units (RLU) in counts per second per milliliter per OD 600 .
To analyze the effect of an increased copy number of CadC or AdiY, 0.1% (w/v) L-arabinose was added during the growth at pH 5.8 + 10 mM lysine and at pH 4.4. Thereby the CadC copy number is elevated to about 100 CadC molecules per cell, whereas the E. coli MG1655 wild type strain contains only ≤4 CadC molecules per cell 40 .
Alignment and construction of phylogenetic trees. To identify non-redundant orthologues of the components of the three AR systems in relation to each antiporter GadC, AdiC and CadB, a Protein BLAST search of the NCBI RefSeq protein database 66 using the full-length sequence of either GadC (XasA), AdiC and CadB from E. coli MG1655 as the query sequence was carried out. Different expect thresholds were used for the Protein BLAST search (e < 10 −100 for GadC, e < 10 −120 for AdiC and CadB), in order to include all homologous yet functional similar proteins (June 2020). A tolerance of 10% of the amino acid length was set as default parameters.
Three alignments and phylogenetic trees were constructed, each based on the antiporter GadC, AdiC or CadB with the different components, and visualized as metadata. Therefore, a pairwise alignment of 1112 sequences for GadC, of 755 sequences for AdiC and of 857 sequences for CadB was done with a progressive algorithm from the software CLC Main Workbench 20.0.3 (CLC Bio Qiagen, Hilden, Germany) using the following parameters: gap open cost 10, gap extension cost 1, high accuracy 67 . The results served as the basis for the construction of three phylogenetic trees by the software's high-accuracy, distance-based neighbor-joining algorithm (100 bootstrap replicates and the Jukes-Cantor distance correction as default parameters). The branch lengths, therefore, represent the degree of evolutionary divergence between any two nodes in the tree.
We screened the organisms containing a AdiC orthologue for the presence of the AdiA, AdiY and CsiR by searching for orthologues of E. coli MG1655 with NCBI Protein BLAST (e < 10 −170 for AdiA, e < 10 −100 for AdiY, e < 10 −50 for CsiR).
We screened the organisms containing an CadB orthologue for the presence of the CadA, CadC, LysP and CsiR by searching for orthologues of E. coli MG1655 with NCBI Protein BLAST (e < 10 −170 for CadA, e < 10 −120 for LysP, e < 10 −50 for CsiR). For the phylogenetic distribution of CadC, the data from our previous publication 41 were used.
All data for the construction of these phylogenetic trees are available in Supplementary Data 1.
Statistics and reproducibility. All experiments were repeated multiple (three replicates unless stated) times to ensure reproducibility. Statistics were performed using either a two-tailed t-test, one-way ANOVA, or a two-tailed Pearson correlation coefficients comparison using GraphPad Prism 9.1.0. All graphs were plotted using the GraphPad Prism 9.1.0 and R 4.0.3 64 . For multi-well plate assays, replicates were also contained within the plate to determine well to well variability.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
Supplementary Data 1 contains a list of all GadC, AdiC, and CadB homologs identified by a local alignment search based on the full-length sequence of E. coli GadC, AdiC, and CadB underlying Fig. 4. All source data underlying the graphs and charts presented in