Purine and pyrimidine synthesis differently affect the strength of the inoculum effect for aminoglycoside and β-lactam antibiotics

Abstract

The inoculum effect has been observed for nearly all antibiotics and bacterial species. However, explanations accounting for its occurrence and strength are lacking. We previously found that growth productivity, which captures the relationship between [ATP] and growth, can account for the strength of the inoculum effect for bactericidal antibiotics. However, the molecular pathway(s) underlying this relationship, and therefore determining the inoculum effect, remain undiscovered. We show that nucleotide synthesis can determine the relationship between [ATP] and growth, and thus the strength of inoculum effect in an antibiotic class-dependent manner. Specifically, and separate from activity through the tricarboxylic acid cycle, we find that transcriptional activity of genes involved in purine and pyrimidine synthesis can predict the strength of the inoculum effect for β-lactam and aminoglycosides antibiotics, respectively. Our work highlights the antibiotic class-specific effect of purine and pyrimidine synthesis on the severity of the inoculum effect and paves the way for intervention strategies to reduce the inoculum effect in the clinic.

Introduction

Antibiotic resistance poses a significant threat to global public health(1). With the antibiotic pipeline largely drying up(2), and recent projections indicating that upwards of 10 million annual deaths from antibiotic resistance will occur by the year 2050(3), we must understand the mechanisms by which bacteria tolerate and resist antibiotic treatment. While the majority of previous work has focused on how a bacterium resists antibiotics(4), there is a growing appreciation that bacteria can resist antibiotics as a collective. For example, collective degradation of β-lactams antibiotics by β-lactamases can enhance the resistance of the entire population(5), facilitate the growth of non-resistant bacteria(6), and can enhance horizontal gene transfer(7). Bacteria can make shared use of secreted extrapolymeric substances during biofilm formation, which increases antibiotic tolerance(8). The collective swarming action of bacteria can also increase antibiotic tolerance(9). Finally, bacteria can use altruistic cell death to enhance antibiotic resistance(10). Overall, understanding the mechanisms by which bacteria tolerate and resist antibiotics as a collective will lead to the development of novel ways to disrupt such cooperative behaviors, which will prolong the use of our existing antibiotics.

It has been well-established that bacteria can resist and tolerate antibiotics as a collective by increasing their population density(11). For a given concentration of antibiotic, if the density of the bacterial population is sufficiently high, the bacteria will tolerate the antibiotic and grow. Otherwise, if the density of the bacterial population is sufficiently low, the bacteria are susceptible to the antibiotic and die. This phenomenon, called the inoculum effect (IE), has been observed for multiple antibiotics and bacterial species(11–20), and can occur in the absence of canonical antibiotic resistance mechanisms(15). IE has been shown to reduce antibiotic efficacy in animal models(13, 21) and has been suspected in antibiotic treatment failure in the clinic(22–31). Moreover, antibiotic tolerance owing to IE has been postulated to drive the evolution of additional resistance mechanisms(32, 33), which further renders antibiotics ineffective. Accordingly, it is critical to understand the mechanisms that allow IE to arise to increase the efficacy and prolong the usefulness of existing antibiotics.

Multiple mechanisms to explain IE have been proposed. Some proposed mechanisms are antibiotic-specific. These include collective degradation of β-lactam antibiotics by β-lactamase producing bacteria(34, 35), degradation of ribosomes following treatment with aminoglycosides(15), and differential growth rates during quinolone treatment(18). Other proposed mechanisms are more general and cover multiple antibiotics. These include a decrease in the ratio of antibiotic to antibiotic target(35), and differences in the length of lag phase owing to changes in initial density(36). More recently, we showed that interactions between bacterial metabolism and growth rate can determine the strength of IE for bactericidal antibiotics(37). Specifically, we found that, for a given growth environment, the strength of IE is dependent on the change in [ATP] relative to the change in growth rate, a relationship that we call growth productivity. Increasing growth productivity reduced the strength of IE; if growth productivity was sufficiently high, IE was abolished. While previous work has found that increasing growth rate or bacterial metabolism potentiates antibiotic lethality(38, 39), we found that IE is best explained by the relationship between [ATP] and growth rate.

Recent work has highlighted the importance of identifying the cellular pathways involved in determining antibiotic efficacy. For example, the SOS response pathway(40) and the stringent response(41) have both been implicated in affecting antibiotic resistance and tolerance. Multiple studies have implicated metabolic activity through the tricarboxylic acid cycle as a determining factor in antibiotic lethality(42–44). More recently, it has been suggested that adenine limitation as determined through nucleotide synthesis pathways during antibiotic treatment can potentiate antibiotic efficacy(45). The identification and subsequent study of these pathways have led to the discovery of novel genes and pathways under selection during antibiotic treatment(46) and have spurred interest in the formulation of novel antibiotic adjuvants(47). However, it remains unclear as to which pathway(s) are involved in determining the relationship between ATP and growth rate in the context of IE. Accordingly, we sought to identify this pathway the discovery of which may lead to the identification of novel drug targets or new approaches to treating recalcitrant high-density infections.

Results

Flux balance analysis coupled with Optknock identifies the superpathway of histidine, purine, and pyrimidine biosynthesis as a determinant of the inoculum effect.

Our previous work demonstrated that flux balance analysis (FBA) could predict the qualitative trends between the strength of IE and the relationship between ATP and growth rate, the latter of which was determined by computing the ratio of ATP synthesis (ATPsyn) to steady-state biomass production, which served as both the objective function and a measure of growth(37). To identify pathways involved in determining ATP production and growth rate in Escherichia coli, we performed FBA with Optknock(48). Optknock individually removes genes from a genomic model by eliminating the lower bound flux value. This approach is the computational equivalent of screening a library of knockout strains but has the additional benefit of being able to assess the effect of removing both non-essential and essential genes. Using a set of parameters that capture the composition of M9 medium (Supplementary Table 1), we sequentially removed all 1516 genes from the E. coli whole genome model. Next, we determined how each gene deletion impacted ATPsyn/biomass, a metric that approximates growth productivity. Considering only genes whose removal increased ATPsyn/biomass above wildtype levels, we assigned pathways to each gene and calculated the frequency at which each pathway occurred in the dataset. We found that the superpathway of histidine, purine, and pyrimidine biosynthesis occurred with the greatest frequency (Fig. 1A, fold enrichment = 3.7, enrichment false discovery rate (FDR) < 0.0001). Several additional pathways involved in nucleotide synthesis were also found in this analysis including the superpathway of purine de novo biosynthesis II (fold enrichment = 4.9, enrichment FDR < 0.0001), the superpathway of pyrimidine deoxyribonucleotides de novo biosynthesis (fold enrichment = 3.5, enrichment FDR = 0.005) and the superpathway of pyrimidine ribonucleotide de novo biosynthesis (fold enrichment = 4.6, enrichment FDR = 0.0045).

Figure 1: Flux balance analysis with Optknock implicates pathways involved in nucleotide synthesis in having the most frequent effect on ATPsyn and biomass.

A) The frequency of pathways in which genes removed by OptKnock increases the ratio of ATPsyn/biomass above wildtype. Each gene removed was assigned to a corresponding pathway as determined using EcoCyc. The percentage of the top 20 pathways is shown. Blue text indicates pathways involved in nucleotide synthesis. Parameters for FBA in Supplementary Table 1.

B) Average ATPsyn/biomass values for the top 20 pathways with the greatest ATPsyn/biomass values. Purple text indicates pathways involved in nucleotide synthesis.

We then averaged ATPsyn/biomass across all genes assigned to a given pathway. For example, the independent removal of 17 different genes from the superpathway of histidine, purine, and pyrimidine biosynthesis was predicted to increase ATPsyn/biomass relative to wildtype. In our analysis, we averaged the ATPsyn/biomass values for these 17 genes and compared this averaged value to all other averaged ATPsyn/biomass values assigned to other pathways. While the greatest average ATPsyn/biomass value belonged to genes in the folate polyglutamylation pathway, we found that genes in the superpathway of histidine, purine, and pyrimidine biosynthesis, the superpathway of pyrimidine deoxyribonucleotides de novo biosynthesis, and the superpathway of pyrimidine deoxyribonucleotides de novo biosynthesis (E. coli) were all represented in the top 20. Together, this computational analysis implicated the production of nucleotides in influencing ATPsyn/biomass, which we explored further.

Supplementation with nitrogenous bases alters [ATP], growth rate, and log[ATP]/growth rate.

To start, we considered a minimal network that describes the synthesis of both nucleotide classes and salvage of nitrogenous bases (Fig. 2A). Nucleotide synthesis occurs as two separate pathways, purine and pyrimidine synthesis. Pyrimidine synthesis begins with glutamate and 5-phospho-α-D-ribose 1-diphosphate (prpp), which, after several enzymatic reactions, is converted into UTP. UTP is used to create CTP. Purine synthesis begins with ribose-5-phosphate (R-5-P), which is converted to inosine monophosphate (IMP). IMP serves as the substrate for the stepwise synthesis of both GTP and ATP. prpp serves as a link between de novo purine and pyrimidine synthesis as it is used as a substrate for both pathways, thus allowing both pathways to autoregulate (49). A reduction in activity in one pathway (e.g., purine synthesis) leads to the accumulation of prpp, which can subsequently be used by the other pathway (e.g., pyrimidine synthesis), thus increasing its activity(45). Nitrogenous bases can be salvaged from the surrounding environment, can be used for nucleotide synthesis, and have been shown to alter purine and pyrimidine synthesis(45, 50). Imported adenine is converted to either AMP or IMP, both of which are involved in purine synthesis(51). Cytosine is imported and converted into uracil; both salvaged uracil and cytosine-derived uracil are converted to UMP via pyrimidine nucleobases salvage pathways(51). Previous work has also shown that altering nucleotide synthesis can alter bacterial metabolism, including [ATP], growth rate, and antibiotic lethality(45, 50), all of which are important in determining IE(37). Accordingly, we chose to perturb nucleotide synthesis, growth rate, and [ATP] by providing nitrogenous bases in the growth medium. Additional details on both nucleotide synthesis and salvage pathways can be found in the Supplementary Results.

Figure 2: Supplementing the growth medium with nitrogenous bases alters growth rate, [ATP], and the ratio of log[ATP]/growth rate.

A) A minimal network of pyrimidine and purine synthesis. gln = glutamate, prpp = 5-phospho-α-D-ribose 1-diphosphate, 5-R-P = ribose-5-phosphate, IMP = inosine monophosphate. PyrC and PurK are shown to facilitate understanding of data presented in Fig. 5.

B) Average growth rate of E. coli grown in M9 medium supplemented with adenine (top left), cytosine (top right), thymine (bottom left), or uracil (bottom right). Average plotted from 5 different percentages of casamino acids, each consisting of ≥ 4 biological replicates. Error bars = standard error from the mean (SEM). * = different than control (0 mM nitrogenous bases, P < 0.005 (two-tailed t-test); all P values for this figure in Supplementary Table 2. Residual values for growth curve fitting in Supplementary Table 3. Growth curves in Supplementary Fig. 1. Raw data in Supplementary Fig. 2.

C) Average [ATP] of E. coli grown in M9 medium supplemented with adenine (top left), cytosine (top right), thymine (bottom left), or uracil (bottom right). Average plotted from 5 different percentages of casamino acids, each consisting of 4 biological replicates. Raw data in Supplementary Fig. 3. Errors bars = SEM. * = different than control (0 mM nitrogenous bases, P < 0.005 (two-tailed t-test).

D) log[ATP]/growth rate for E. coli grown in M9 medium supplemented with adenine (top left), cytosine (top right), thymine (bottom left), or uracil (bottom right). Data from panels B and C. Errors bars = SEM.

To quantify the effect of exogenously supplied nitrogenous bases on growth rate and ATP production, we grew E. coli in M9 medium with different concentrations of cytosine, adenine, thymine, and uracil. Guanine could not be used as it could not be dissolved at 1 mM in M9 medium. Cultures that lacked the addition of nitrogenous bases in the growth medium served as the control. Consistent with our previous work(37), we provided different percentages of casamino acids in the M9 medium, which served as a nitrogen source. Varying casamino acids allowed us to study IE over a range of [ATP] and growth rates for a given growth environment, which was defined by the concentration and type of nitrogenous base provided. We chose to use nitrogenous bases supplied at concentrations between 1–10 mM as these concentrations are sufficient to perturb nucleotide synthesis(45).

To quantify the maximum growth rate, we performed high-resolution measurements of bacterial density (OD₆₀₀) for 10 hours in a microplate reader(37). We then fit the resulting growth curves to a logistic equation whereupon we extracted the maximum growth rate. We found that the addition of 5 mM or 10 mM of adenine and thymine significantly reduced growth rate as compared to the control (Fig. 2B). The addition of 1 mM cytosine or 1 mM thymine increased growth rate relative to the control. Otherwise, no other significant changes in growth rate were noted. Next, we quantified the effect of nitrogenous base supplementation on [ATP]. We measured [ATP] normalized by cell density using an approach that allows measurement of bacterial metabolism separate from growth(37, 38). Measurement of [ATP] under this condition is strongly correlated to other measures of metabolism, including [NAD⁺]/[NADH] and oxygen consumption rate(37, 38). The addition of either adenine or thymine at all concentrations tested significantly increased [ATP] relative to the control. With adenine supplementation, [ATP] appeared biphasic, where the greatest levels observed occurred with 5 mM adenine (Fig. 2C). With thymine supplementation, we observed that as the concentration of thymine increased, [ATP] also increased. When the medium was supplemented with either cytosine or uracil, a significant decrease in [ATP] was observed at 1 mM and 5 mM. A decrease in [ATP] was also observed with 10 mM uracil.

Finally, we determined the ratio of log[ATP]/growth rate, which served as the experimental measure of ATPsyn/biomass. We found that when either uracil or cytosine was supplied in the growth medium, log[ATP]/growth rate was largely consistent with the control (Fig. 2D). However, with the addition of either 5 mM or 10 mM adenine and thymine, log[ATP]/growth rate was greater than the control by 2–3 fold. A small increase in this ratio was also noted when 1 mM adenine was supplied in the growth medium. Overall, supplementation of adenine and thymine, but not cytosine or uracil, altered [ATP], growth rate, and log[ATP]/growth rate.

Nitrogenous base driven increases in [ATP]/growth rate decrease the strength of the inoculum effect.

To assess the impact that manipulating log[ATP]/growth rate through nucleotide synthesis had on IE, we measured the MIC of the aminoglycoside streptomycin in E. coli. MICs were measured for two populations initiated from high and low density. Importantly, the high-density populations (2.03 × 10⁶ CFU/mL +/− 4.75 × 10⁵) were above the standard density used to measure MICs in the clinic (5.5 × 10⁵ CFU/mL) whereas the low-density population (1.67 × 10⁵ CFU/mL +/− 5.42 × 10⁴) were below this density. MICs were assessed in increasing concentrations of streptomycin in M9 medium with and without nitrogenous bases (control). We could not achieve reliable growth at all percentages of casamino acids when 10 mM adenine was included in the medium; MICs in this condition could not be quantified. In general, we found that as the percentage of casamino acids increased in the growth medium, the MIC of both high- and low-density populations increased (Fig. 3A). Consistent with previous work(37), we calculated the strength of the inoculum effect (ΔMIC) by determining the average difference in MIC between the high and low-density populations. We found that when either adenine or thymine was included in the growth medium, ΔMIC decreased as compared to the control (Fig. 3B); the reduction in ΔMIC was significant when 5 mM adenine, 5 mM thymine, or 10 mM thymine was included in the growth medium. Interestingly, supplementation with 10 mM thymine yielded a ΔMIC value that was not statistically different than zero, indicating that IE was abolished. Finally, supplementation with 1 mM adenine, 1mM thymine, cytosine, or uracil did not result in a significant change in ΔMIC.

Figure 3: Ratio of log[ATP]/growth rate predicts the strength of inoculum effect (ΔMIC) for streptomycin and carbenicillin, but not ciprofloxacin.

A) MIC of initial high (dark blue) and low (light blue) density E. coli populations grown in M9 medium supplemented with different nitrogenous bases and with streptomycin (strp). SEM from ≥ 5 biological replicates. Raw data in Supplementary Fig. 4.

B) Average ΔMIC for E. coli populations grown in M9 medium as in panel A. * different than no nitrogenous base control (P ≤ 0.036, two-tailed t-test); ** not different than zero (P = 0.104, one-tailed t-test). Error bars = SEM.

C) ΔMIC of strp as a function of log[ATP]/growth rate. R² and P value from a linear regression. Weighted least squares (WLS) regression (R² = 0.98, P < 0.0001); Deming regression (P < 0.0001). Error bars = SEM. MIC data from panel A; log[ATP]/growth from Fig. 2D.

D) ΔMIC of carbenicillin (carb) as a function of log[ATP]/growth rate. R² and P value from a linear regression. WLS regression (R² = 0.82, P = 0.002); Deming regression (P = 0.0006). Error bars = SEM. * different than no nitrogenous base control (P ≤ 0.03, two-tailed t-test); ** not different than zero (P ≥ 0.071, one-tailed t-test). MIC data from Supplementary Fig. 5; log[ATP]/growth from Fig. 2D.

E) ΔMIC of ciprofloxacin (cipro) as a function of log[ATP]/growth rate. R² and P value from a linear regression. WLS regression (R² = 0.15, P = 0.34); Deming regression (P = 0.28). * different than no nitrogenous base control (P = 0.009, two-tailed t-test); ** not different than zero (P = 0.054, one-tailed t-test, 1 mM thymine within data cluster). Error bars = SEM. MIC data from Supplementary Fig. 6; log[ATP]/growth from Fig. 2D.

Next, we performed a regression analysis between ΔMIC and log[ATP]/growth rate. We found a strong and significant linear relationship between ΔMIC of streptomycin and log[ATP]/growth rate; as log[ATP]/growth rate increased, ΔMIC decreased (Fig. 3C). To test the generality of the relationship, we challenged E. coli with two additional antibiotics, carbenicillin (β-lactams) and ciprofloxacin (fluoroquinolones). Consistent with streptomycin, we found a strong and significant relationship between the ΔMIC of carbenicillin and log[ATP]/growth rate (Fig. 3D). We did not find a strong or significant relationship with ΔMIC of ciprofloxacin (Fig. 3E). The relationship between log[ATP]/growth rate and ΔMIC of streptomycin and carbenicillin holds when a higher initial density (2.00 × 10⁷ CFU/mL +/− 4.50 × 10⁶) of E. coli is used (Supplementary Fig. 7). Overall, we found that supplementation with nitrogenous bases could alter log[ATP]/growth rate, which determined ΔMIC for streptomycin and carbenicillin, but not ciprofloxacin.

To test alternative hypotheses that could explain these trends in ΔMIC, we found that concentrations of adenine (1 mM) and thymine (0.9 mM) that altered either growth rate or [ATP], but not both, did not change ΔMIC of streptomycin (Supplementary Fig. 8). Importantly, these nitrogenous bases reduced ΔMIC when provided in higher concentrations. We next performed regression analysis between ΔMIC for each antibiotic and log[ATP] or growth rate (Supplementary Fig. 8). Similar to our previous work, we found significant relationships between [ATP] and growth rate when independently correlated with ΔMIC(37). However, in general, the R² values of these relationships were most often less than that of ΔMIC plotted as a function of log[ATP]/growth. Our findings cannot be explained by other forms of density-dependent antibiotic resistance including quorum sensing (luxS) or efflux pumps (mdtA) as knockout strains lacking these activities continued to show IE. Moreover, supplementation with thymine, but not cytosine, reduced ΔMIC in these knockout strains (Supplementary Fig. 9). While previous work(15, 37) has suggested that changes to the ratio of antibiotic to antibiotic target can explain IE, we did not find a strong nor significant relationship between ΔMIC for streptomycin and the concentration of rRNA, which serves as a measure of ribosome concentration (Supplementary Fig. 9, Supplementary Methods and Results). As ribosomes are the target of streptomycin, it is unlikely that a change in the number of antibiotic targets (ribosomes) can account for IE in our system. For a more detailed explanation of alternative hypotheses, see the Supplementary Results.

Exogenous nitrogenous bases supplementation affects the strength of the inoculum effect in multiple bacterial species.

To test the ability of nitrogenous base supplementation to alter log[ATP]/growth rate and ΔMIC in additional bacterial species, we grew Pseudomonas aeruginosa in M9 medium supplemented with nitrogenous bases and measured growth rate and [ATP] as above. We found that supplementation with nitrogenous bases could insignificantly alter growth rate compared to the control (Fig. 4A). Supplementation with nitrogenous bases could also alter ATP production. However only supplementation with 5 mM adenine significantly reduced [ATP] relative to the control (Fig. 4B). Together, these trends altered log[ATP]/growth rate and, consistent with E. coli, the two greatest values were observed when either 10 mM thymine or 5 mM adenine was supplemented in the medium (Fig. 4C). Next, we determined ΔMIC for streptomycin and carbenicillin. We found that supplementation with either 5 mM adenine or 10 mM thymine reduced the ΔMIC of streptomycin and carbenicillin. However, only the reduction in ΔMIC for thymine was significant relative to the control (Fig. 4D). Similar to our findings in E. coli, we found a strong and significant relationship between ΔMIC for both antibiotics and log[ATP]/growth rate; as log[ATP]/growth rate increased, ΔMIC decreased (Fig. 4E and F). However, when either log[ATP] and growth rate alone were plotted as a function of ΔMIC, the relationship was not consistently strong or significant (Supplementary Fig. 10). Overall, these findings were consistent with E. coli suggesting that the relationship between log[ATP]/growth rate and ΔMIC as determined by nitrogenous base supplementation can be found in additional Gram-negatives.

Figure 4: Nitrogenous base driven changes in [ATP]/growth rate determine ΔMIC in P. aeruginosa.

A) Average growth rate of P. aeruginosa grown in M9 medium supplemented with nitrogenous bases at the concentration indicated. Data from three percentages of casamino acids (0.1, 0.5, and 1%), each consisting of ≥ 3 biological replicates. Errors bars = SEM. Growth curves and raw data in Supplementary Fig. 10. Residual values for growth curve fitting in Supplementary Table 4. All P values for the figure in Supplementary Table 5.

B) Average [ATP] of P. aeruginosa grown in M9 medium supplemented with nitrogenous bases. Average from 3 percentages of casamino acids, each consisting of 3 biological replicates. Raw data in Supplementary Fig. 10. Errors bars = SEM. * = different than the no nitrogenous base control (P = 0.032, two-tailed t-test).

C) log[ATP]/growth rate for P. aeruginosa. Data from panels A and B. Errors bars = SEM.

D) Average ΔMIC for P. aeruginosa populations grown in M9 medium. Top: ΔMIC of streptomycin (strp). Bottom: ΔMIC of carbenicillin (carb). Error bars = SEM. Average from ≥ 4 biological replicates. Raw data in Supplementary Fig. 10. * = different than the no nitrogenous base control (P ≤ 0.041, one-tailed t-test). Percentage of casamino acids and concentration of nitrogenous bases as in panel A.

E) ΔMIC of strp as a function of log[ATP]/growth rate. R² and P value from a linear regression. Error bars = SEM. MIC data from panel D; log[ATP]/growth from panel C. WLS: R²= 0.71, P = 0.073. Deming regression: P = 0.0272.

F) ΔMIC of carb as a function of log[ATP]/growth rate. R² and P value from a linear regression. Error bars = SEM. MIC data from panel D; log[ATP]/growth from panel C. WLS: R²= 0.85, P = 0.026. Deming regression: P = 0.031.

Knockout strain analysis reveals opposing effects of purine and pyrimidine synthesis on the strength of the inoculum effect

To provide direct support for the role of purine and pyrimidine synthesis in determining ΔMIC, we used FBA with OptKnock to identify genes involved in nucleotide synthesis whose removal would alter ATPsyn and biomass; together, this would alter log[ATP]/growth rate. While our FBA and OptKnock analysis identified multiple genes whose removal would increase ATP/growth rate, we focused on two genes, purK, and pyrC, both of which lack mammalian homologs and are involved in purine and pyrimidine synthesis, respectively(52) (Fig. 5A, see Fig. 2A for position in a network). Our computational analysis predicts that the removal of pyrC would increase ATP synthesis (ATPsyn) while reducing growth rate (biomass) relative to wildtype (Fig. 5B and C). Conversely, the removal of purK would decrease both ATP synthesis and growth rate relative to wildtype (Fig. 5B and C). To test these predictions, we measured growth rate and [ATP] as above. While growth of E. coli lacking these genes was achieved in LB medium, growth was not consistently observed in traditional M9 medium (Supplementary Fig. 11). This was consistent with FBA and Optknock predictions that predicted biomass values of 0 when these genes were removed. However, growth of these strains could be restored in M9 medium if low equimolar concentrations (1 μM, 4 μM, and 7 μM) of all five nitrogenous bases were provided. Consistent with our model predictions, we found that removal of pyrC (ΔpyrC) resulted in elevated [ATP]. Conversely, [ATP] in a strain lacking purK (ΔpurK) was lower than the wildtype (Fig. 5D). We also found that both ΔpyrC and ΔpurK had reduced growth rates relative to the wildtype (Fig. 5E). As predicted computationally, these changes led to greater log[ATP]/growth rate ratios as compared to wildtype (Fig. 5F).

Figure 5: Knockout strains reveal antibiotic-specific effects of purine and pyrimidine synthesis on ΔMIC for streptomycin and carbenicillin.

A) Simulation: Genes identified by FBA and OptKnock that are involved in nucleotide synthesis and that increase ATPsyn/biomass above wildtype (wt). Dark blue bars indicate pyrC and purK. FBA parameters in Supplementary Table 1. Positions of PyrC and PurK in nucleotide synthesis pathway shown in Fig. 2.

B) Simulations: FBA predicted effects on ATP synthesis (ATPsyn) for wt, ΔpyrC, and ΔpurK.

C) Simulation: FBA predicted effects on biomass for wt, ΔpyrC, and ΔpurK.

D) Removal of pyrC increases [ATP], while removal of purK decreases [ATP], relative to wildtype. Error bars = SEM. Average from 3 biological replicates. * indicates different than wildtype (P < 0.01, two-tailed t-test). For D and E, raw data in Supplementary Fig. 11 and all P values in Supplementary Table 6.

E) Removal of pyrC and purK decreases growth rate relative to wildtype. Error bars = SEM. Average from ≥ 5 biological replicates. * indicates different than control (P < 0.001, two-tailed t-test). Residuals for curve fitting in Supplementary Table 7.

F) Removal of pyrC and purK increases log[ATP]/growth rate relative to wildtype. Data from panels D and E.

G) ΔMIC of streptomycin (strp). Error bars = SEM. Average plotted from ≥ 5 biological replicates. * = different than wildtype (two-tailed t-test, P = 0.038), ** not different than zero (one-tailed t-test, P = 0.211). For G and H, raw data in Supplementary Fig. 12.

H) ΔMIC of carbenicillin (carb). Error bars = SEM. Average plotted from ≥ 4 biological replicates. * = different than wildtype (P < 0.025, two-tailed t-test). ** not different than zero (one-tailed t-test, P = 0.211).

We then quantified the ΔMIC of streptomycin and carbenicillin for both strains in addition to the wildtype strain. ΔMIC of streptomycin and carbenicillin decreased significantly and was not different than zero for ΔpyrC relative to the wildtype (Fig. 5G and H). Conversely, for ΔpurK, ΔMIC was no different than wildtype when challenged with streptomycin. However, when challenged with carbenicillin, ΔMIC increased relative to the wildtype. Interestingly, these trends in ΔMIC did not follow trends in log[ATP]/growth rate. While the increase in log[ATP]/growth rate in ΔpyrC was consistent with the reduction in ΔMIC for both antibiotics, the increase in log[ATP]/growth in ΔpurK did not alter (streptomycin) or increase (carbenicillin) ΔMIC. This data suggests that genetic inhibition of purine synthesis (ΔpurK) protects bacteria by increasing (carbenicillin), or not altering (streptomycin), ΔMIC. On the other hand, genetic inhibition of pyrimidine synthesis (ΔpyrC) decreases ΔMIC, effectively abolishing IE for both antibiotics and is consistent with the observed increase in log[ATP]/growth rate.

Chemical perturbation of purine and pyrimidine synthesis pathways alters ΔMIC.

Our genetic analysis implied non-equivalent effects of purine and pyrimidine synthesis in determining ΔMIC. It also implied that the relationship between ΔMIC and log[ATP]/growth could be disrupted through direct genetic perturbation of both pathways. However, the removal of key enzymes in both nucleotide synthesis pathways would alter regulation in and between both pathways by removing key regulatory steps, which could also alter the energetic requirements of both pathways. In addition to autoregulation(50), negative feedback plays a key role in regulating activity within each pathway (Fig. 6A). During purine synthesis, accumulation of AMP represses PurF, which attenuates purine synthesis(53). However, accumulation of IMP, a purine synthesis intermediate from which AMP is produced, potentiates the activity of pyrimidine synthesis. UTP produced by pyrimidine synthesis attenuates the activity of several upstream enzymes, including PyrH(54) and PyrE(55). UMP also inhibits the first steps of pyrimidine synthesis(56). To preserve the regulation both within and between both nucleotide synthesis pathways, we sought to chemically manipulate nucleotide synthesis, allowing greater insight into the role of purine and pyrimidine synthesis in determining ΔMIC.

Figure 6: Chemical manipulation of nucleotide synthesis alters ΔMIC in an antibiotic-specific manner.

A) A minimal network of pyrimidine and purine synthesis, and salvage (blue) showing regulatory steps. red = 6-MP, purple = IMP = inosine monophosphate, gln = glutamate, prpp = 5-phospho-α-D-ribose 1-diphosphate, 5-R-P = ribose-5-phosphate.

B) Average growth rate (top), [ATP] (middle), and log[ATP]/growth rate ratio (bottom) of E. coli grown in M9 medium supplemented with 5 mM adenine and 10 mM cytosine in the presence of 6-MP. Errors bars = SEM. * = different than control (no 6-MP, no nitrogenous bases, P ≤ 0.031 (two-tailed t-test). For panels B-D, raw data in Supplementary Fig. 14. For panels B and E , residuals for curve fitting in Supplementary Table 8. P values in Supplementary Table 9.

C) ΔMIC of E. coli grown in M9 medium supplemented with 6-MP, nitrogenous bases and streptomycin (strp). * = difference in ΔMIC relative to control (no 6-MP, no nitrogenous bases, P ≤ 0.009, two-tailed t-test). ** = not different than zero (P = 0.31, one-tailed t-test). SEM from ≥ 5 biological replicates.

D) ΔMIC of E. coli grown in M9 medium supplemented with 6-MP, nitrogenous bases, and carbenicillin (carb). * = difference in ΔMIC as compared to control (no 6-MP, no nitrogenous bases P = 0.006, two-tailed t-test). SEM from ≥ 5 biological replicates.

E) Average growth rate (top), average [ATP] (middle), and log[ATP]/growth rate ratio (bottom) of E. coli grown in M9 medium supplemented with IMP, nitrogenous bases, and 0.1% casamino acids. Errors bars = SEM. * = different than control (no IMP, no nitrogenous bases, P ≤ 0.026, two-tailed t-test). For panels E-G, raw data in Supplementary Fig. 15.

F) ΔMIC of E. coli populations grown in M9 medium supplemented with IMP, nitrogenous bases, and strp. * = difference in ΔMIC as compared to control (no IMP, no nitrogenous bases, P = 0.009, two-tailed t-test). SEM from ≥ 5 biological replicates.

G) ΔMIC of E. coli grown in M9 medium supplemented with IMP, nitrogenous bases, and carb. * = difference in ΔMIC as compared to control (no IMP, no nitrogenous bases, P ≤ 0.027, two-tailed t-test). SEM from ≥ 5 biological replicates.

H) ΔMIC of strp as a function of log[ATP]/growth rate. R² and P value from a linear regression. Weighted least squares (WLS) regression (R² = 0.81, P = 0.036). Deming regression: P = 0.042. Error bars = SEM. MIC data from panels C and F; log[ATP]/growth from panels B and E.

I) ΔMIC of carb as a function of log[ATP]/growth rate. R² and P value from a linear regression. WLS regression (R² = 0.69, P = 0.042). Deming regression: P = 0.21. Error bars = SEM. MIC data from panels D and G; log[ATP]/growth from panels B and E.

To inhibit purine synthesis, we grew E. coli in the presence of a sub-lethal concentration of 6-mercaptopurine (6-MP), which inhibits PurF(57). We focused on completing our experiments in one percentage of casamino acids, 0.1%, which was the intermediate concentration used previously (Fig. 3) and had ΔMICs that were not different than the average ΔMIC for all percentages of casamino acids measured (Supplementary Fig. 13). We focused on using 5 mM adenine and 10 mM cytosine as they represented the highest concentrations of nitrogenous bases that did, or did not, affect log[ATP]/growth rate and ΔMIC (Fig. 3). Moreover, unlike thymine, salvage directly influences activity each nucleotide synthesis pathway(58). We found that 6-MP alone had no appreciable effect on growth rate, but reduced [ATP] relative to the control (E. coli without 6-MP and nitrogenous bases); together, this resulted in a lower log[ATP]/growth rate as compared to the control (Fig. 6B). Supplementation with adenine and 6-MP reduced growth rate and increased [ATP], which increased log[ATP]/growth rate (Fig. 6B). Finally, supplementation with cytosine and 6-MP did not alter growth rate, but reduced [ATP], which led to a decrease in log[ATP]/growth rate (Fig. 6B). These perturbations had largely opposed effects on ΔMIC for both antibiotics. 6-MP alone significantly reduced ΔMIC of streptomycin (Fig. 6C) but not of carbenicillin (Fig. 6D). Both 6-MP and adenine reduced ΔMIC of streptomycin such that it was not different than zero, but significantly increased ΔMIC of carbenicillin. Finally, supplementation with both 6-MP and cytosine did not result in a change in ΔMIC for both antibiotics.

Next, we grew E. coli in the presence of the purine synthesis pathway intermediate inosine-5-monophosphate (IMP). Owing to the regulatory steps in purine synthesis, providing IMP would itself not directly repress purine synthesis; repression instead would be provided by accumulated AMP produced later in the pathway(53) (Fig. 6A). IMP also increases pyrimidine synthesis(59). Supplementation with IMP increased growth relative to the control (no IMP, no nitrogenous bases) but had no significant effect on [ATP]; together, this resulted in a small reduction in log[ATP]/growth (Fig. 6E). IMP and adenine decreased growth rate and increased [ATP], which resulted in an increase in log[ATP]/growth rate (Fig. 6E). Finally, IMP and cytosine significantly increased growth rate and reduced [ATP]; together, this reduced log[ATP]/growth rate (Fig. 6E). Consistent with our findings using 6-MP, we found that the use of IMP had opposing effects on ΔMIC for streptomycin and carbenicillin. Treatment with IMP alone left the ΔMIC of streptomycin unchanged (Fig. 6F) but significantly increased the ΔMIC of carbenicillin (Fig. 6G). Both IMP and adenine reduced the ΔMIC of streptomycin, while that of carbenicillin increased significantly. Finally, both IMP and cytosine left ΔMIC for streptomycin unchanged but increased it for carbenicillin.

Interestingly, supplementation of the growth medium with IMP or 6-MP reduced or abolished the relationship between ΔMIC and log[ATP]/growth rate and was thus generally consistent with our findings using the ΔpyrC and ΔpurK knockout strains. While this relationship remained significant for streptomycin, (Fig. 6H) it was weakened as compared to when only nitrogenous bases were provided in the growth medium (Fig. 3). For carbenicillin, the relationship between ΔMIC and log[ATP]/growth rate was not strong nor significant (Fig. 6I). Overall, we found that chemical perturbations to nucleotide synthesis had largely opposing effects on ΔMIC for streptomycin and carbenicillin; a decrease in ΔMIC for streptomycin often led to an increase in ΔMIC for carbenicillin.

Relative flux through purine and pyrimidine synthesis can account for opposing trends in ΔMIC

To gain insight into why manipulating purine and pyrimidine synthesis using inhibitors led to opposing effects on ΔMIC for streptomycin and carbenicillin, we used FBA to simulate the effects of the experimental perturbations above on purine synthesis and pyrimidine synthesis. To verify these FBA predictions, we used reporter strains that contained low copy plasmids with promoters reporting activity of purine (purM-gfp) and pyrimidine (pyrC-gfp) synthesis (see Fig. 6A positions of purM and pyrC in each pathway). Importantly, pathway activity can be accurately measured using these reporter strains as many of the regulatory steps in nucleotide synthesis occur at the transcriptional level(60–64). Together, this would allow us to determine how each of the perturbations above differentially impacted activity through purine and pyrimidine synthesis.

We first simulated the effect of adding only adenine or cytosine to the growth medium. FBA predicts that adenine reduces purine synthesis activity while pyrimidine synthesis activity increases (Fig. 7A, top panels). Conversely, the addition of cytosine left purine synthesis relatively unchanged while decreasing pyrimidine synthesis. These predictions were largely consistent with the transcriptional activity of pyrC and purM as determined by our reporter strains (Fig. 7A, bottom panels). An exception to this was transcriptional activity reported by the purM reporter strain which showed a significant, albeit small, reduction in transcriptional activity when cytosine was provided in the medium. This discrepancy may reflect the fact that the FBA model does not contain all reactions present in the cell and assumes the system has reached steady state. Nevertheless, our FBA predictions largely matched the transcriptional activity of the reporter strains and these findings were consistent with literature (45, 49).

Figure 7: FBA-predicted changes in transcriptional activity of genes involved in purine and pyrimidine synthesis account for the opposing trends in ΔMIC for streptomycin and carbenicillin.

A) Top: FBA predicted activity through purine synthesis (top left) and pyrimidine synthesis (top right) supplemented with adenine and cytosine. For panels A-C, flux activity normalized to no base control. Sensitivity analysis in Supplementary Fig. 16. Near identical predictions occur when multiple reactions in purine and pyrimidine synthesis are considered (Supplementary Fig. 16). Bottom: Transcriptional activity of purM (left) and pyrC (right) reporter strains. For all experimental data, SEM from 5 biological replicates. * = different from no nitrogenous base (no bases.) control (P ≤ 0.019, two-tailed t-test). Adenine = 5 mM, cytosine = 10 mM. For panels A-C, GFP/OD₆₀₀ normalized to no base control. Transcriptional activity measured at 24 hours shows similar trends (Supplementary Fig. 17).

B) Top: FBA predicted activity through purine synthesis (left) and pyrimidine (right) synthesis supplemented with 6-MP, adenine, and cytosine. Bottom: Transcriptional activity of purM (left) and pyrC (right) reporter strains. * = P ≤ 0.035.

C) Top: FBA predicted activity through purine synthesis (left) and pyrimidine (right) synthesis supplemented with IMP, adenine, and cytosine. Bottom: Transcriptional activity of purM (left) and pyrC (right) reporter strains. * = P ≤ 0.017.

D) ΔMIC of streptomycin (strp) as a function of purM (left) and pyrC (right) reporter transcriptional activity. R² and P value from a linear regression. WLS regression (purM: R² < 0.01, P = 0.93; pyrC: R² = 0.83, P =0.032); Deming regression (purM - P = 0.89; pyrC - P = 0.007). Error bars = SEM. ΔMIC from Fig 6. For panels D and E, trends are consistent when reporter activity is measured at 24 hours (Supplementary Fig. 17).

E) ΔMIC of carbenicillin (carb) as a function of purM (left) and pyrC (right) transcriptional activity (left panel). R² and P value from a linear regression. WLS regression (purM: R² = 0.85, P =0.009; pyrC: R² = 0.32, P =0.24); Deming regression (purM - P = 0.007; pyrC P = 0.89). Error bars = SEM. MIC data from Fig 6.

FBA predicts that the addition of 6-MP reduces purine synthesis while having little to no effect on pyrimidine synthesis (Fig. 7B, top). The addition of both 6-MP and adenine further reduces purine synthesis concomitant with an increase in pyrimidine synthesis. Supplementation with both 6-MP and cytosine is predicted to reduce both purine and pyrimidine synthesis. As above, these predictions were generally consistent with activity from the reporter strains (Fig. 7B, bottom). A significant reduction in purM promoter activity, but not pyrC promoter activity, was observed with 6-MP, thus confirming the inhibitory effect of 6-MP on purine synthesis(57). Supplementation with both 6-MP and adenine significantly reduced purM promoter activity while increasing pyrC promoter activity. Finally, supplementation with cytosine and 6-MP reduced both purM and pyrC reporter activity. Returning to our FBA, we found that supplementation with IMP, IMP with adenine, and IMP with cytosine results in a reduction in activity through purine synthesis (Fig. 7C, top), which was confirmed using the purM reporter strain (Fig. 7C, bottom). Supplementation with IMP and both IMP and adenine were predicted to increase the activity of pyrimidine synthesis. Conversely, supplementation with both IMP and cytosine was predicted to decrease pyrimidine synthesis. While this general trend was confirmed using the pyrC reporter, we did observe a significant reduction in transcriptional activity in the pyrC promoter strain when IMP alone was provided in the growth medium. Nevertheless, our findings indicate that the perturbation performed above altered the transcriptional activity of genes involved in purine and pyrimidine synthesis, which were largely consistent with flux activity as determined using FBA.

To determine if differences in the transcriptional activity of purine and pyrimidine synthesis could account for the opposing changes in ΔMIC, we performed a linear regression between the transcriptional activity of purine and pyrimidine synthesis and ΔMIC for both antibiotics. We found that while there was no significant relationship between ΔMIC for streptomycin and transcriptional activity of purine synthesis, there was a strong and significant relationship with pyrimidine transcriptional activity; as transcriptional activity of pyrC increased, ΔMIC of streptomycin decreased (Fig. 7D). These trends were also consistent with kanamycin, implying that this relationship is also found in additional aminoglycosides (Supplementary Fig. 18). Interestingly, we found the opposite for ΔMIC of carbenicillin (Fig. 7E). Here, the relationship between the transcriptional activity of purine synthesis was strongly and significantly correlated to ΔMIC of carbenicillin; as the transcriptional activity of purM increased, ΔMIC of carbenicillin decreased. Conversely, there was not a strong nor significant relationship between the transcriptional activity of pyrimidine synthesis and the ΔMIC of carbenicillin. We did not find a significant nor strong relationship between transcriptional activity in the tricarboxylic acid cycle (TCA), as measured using a succinate dehydrogenase reporter, and ΔMIC of all antibiotics measured in this study (Supplementary Fig. 19). This suggests that the opposing trends in ΔMIC cannot be explained by transcriptional activity in the TCA. Taken together, our findings suggest that activity through pyrimidine synthesis can account for ΔMIC in representative aminoglycosides, whereas activity through purine synthesis can account for ΔMIC of representative β-lactams.

Discussion

Herein, we have provided evidence that purine and pyrimidine synthesis can impact the strength of IE. Initially, and consistent with our previous work, we found that log[ATP]/growth rate could predict the strength of IE for carbenicillin (β-lactam) and streptomycin (aminoglycoside), but not ciprofloxacin (fluoroquinolone). Using knockout strains and chemical modifiers of purine and pyrimidine synthesis, we showed that the strength of IE could be decoupled from log[ATP]/growth rate. Instead, our data suggests that activity through purine and pyrimidine synthesis can account for the strength of IE but is dependent upon antibiotic class. Transcriptional activity in pyrimidine synthesis predicts ΔMIC of aminoglycosides, whereas transcriptional activity in purine synthesis predicts ΔMIC in β-lactams. Taken together, our results implicate nucleotide synthesis as being a core pathway determining the strength of IE.

Recent work has highlighted the importance that nucleotide synthesis, or nucleotides themselves, has in determining antibiotic efficacy. For example, it has long been established that trimethoprim interferes with the activity of dihydrofolate reductase, which negatively impacts nucleic acid synthesis in addition to other pathways(65). Trimethoprim has also been observed to alter the concentration of intracellular adenine, which impacts its lethality in a growth environment-dependent manner(66). In E. coli, the oxidation of guanine to 8-oxo-guanine during exposure to quinolones and β-lactams results in double-stranded breaks, which leads to cell death(67). When Mycobacteria reach the stationary phase, the oxidation of dCTP as a result of antibiotic exposure can contribute to DNA damage and cell death (68). More recently, adenine limitation has been proposed to increase ATP production through central carbon metabolism. This potentiates the production of reactive oxygen species and additional toxic byproducts, enhancing antibiotic lethality(45). Along this line, the exogenous addition of thymine to bacterial cultures was shown to increase respiration concomitant with an increase in the lethality of ciprofloxacin(69). Exogenously supplied adenine has been shown to increase antibiotic activity against persister cells, independent of growth(70). Given the multiple roles that nucleotides have in determining antibiotic lethality, alterations in their synthesis can convey antibiotic tolerance and resistance. For example, mutations in genes involved in purine synthesis were associated with increased resistance to rifampicin in S. aureus (71), while mutations to both purine and pyrimidine synthesis confer resistance in E. coli (72). The presence of AMP at the infection site and during antibiotic treatment has been shown to decrease antibiotic lethality (73). Finally, overexpression of genes that deplete cofactor pools required for nucleotide synthesis increases antibiotic tolerance(74). Our research adds to our growing understanding of the role that nucleotides and their synthesis play in determining antibiotic efficacy. To our knowledge, this represents the first instance where nucleotide synthesis has been implicated in determining IE.

Our previous work identified that the relationship between ATP production and growth rate, which we called growth productivity, can account for IE across multiple antibiotic classes(37). In this previous study, we used different carbon sources to alter growth productivity; carbon sources that increased this metric consistently reduced the strength of IE(37). In the present study, we found that increasing log[ATP]/growth rate using exogenously supplied nitrogenous bases reduced ΔMIC, implicating nucleotide synthesis as a determining factor in IE. It is important to consider how our previous findings fit within the context of the current study. Briefly, our previous mechanism relied on the intersection of bacterial metabolism (ATP), growth rate, and initial density to explain IE. Bacterial populations initiated from high density have a short period of log phase growth where ATP production is greatest before entering the stationary phase, where ATP synthesis slows. As antibiotic lethality is dependent upon both growth rate and bacterial metabolism, the time over which both are high is relatively small for a high-density population. Thus, they are inherently more tolerant of antibiotics. Conversely, populations initiated from lower density spend considerably more time in log phase, where both growth rate and metabolism are high, and are thus more susceptible to antibiotics. Within the scope of the current work, as bacteria enter the stationary phase, a reduction in DNA replication(75) and nucleotide pools(76) has been reported previously. Owing to the high energetic cost of producing nucleotides(77), it is not unreasonable to suggest that the decrease in [ATP] owing to entry into the stationary phase is a result of a reduction in the need to synthesize nucleotides for DNA replication and additional biochemical reactions. Thus, during the stationary phase, reductions in DNA synthesis coincide with reductions in both ATP production and growth, which together affect antibiotic tolerance. Conversely, DNA synthesis and nucleotide pools are increased during the exponential phase where bacteria are most sensitive to antibiotics. As DNA synthesis and nucleotide synthesis generally correlate to growth phase and ATP abundance, the role that purine and pyrimidine synthesis appear to play in determining IE fits within our previously proposed mechanism. Interestingly, the lack of correlation between ΔMIC of ciprofloxacin and log[ATP]/growth rate provides additional support to the role of nucleotide synthesis in determining IE (Fig. 3). Ciprofloxacin targets DNA gyrase and causes death, in part, due to the formation of single and double-stranded breaks during DNA synthesis(78). However, during the stationary phase, DNA gyrase is tasked with restoring DNA supercoiling(79) and not necessarily DNA replication. Coupled with a reduction of ATP during the stationary phase, the activity of this enzyme is limited(18). Accordingly, even when perturbing ATP, growth rate, and the synthesis of nucleotides using exogenously supplied nitrogenous bases, if the molecular target of the antibiotic is inactive (e.g., DNA gyrase), such perturbations are unlikely to affect antibiotic lethality. This finding is consistent with our previous work showing that the relationship between ΔMIC of ciprofloxacin can be insensitive to changes in growth productivity(37).

Initially, we found that nitrogenous base driven changes in log[ATP]/growth rate could account for ΔMIC for streptomycin and carbenicillin (Fig. 3). However when purine and pyrimidine synthesis was directly perturbed using gene deletion (Fig. 5) or chemical inhibitors (Fig. 6), the relationship between log[ATP]/growth and ΔMIC were significantly weakened or no longer significant. We also did not find significant relationships between ΔMIC and transcription activity of succinate dehydrogenase of the TCA, which is often used as a measure of bacterial metabolism in light of antibiotic lethality(45). Instead, we found that transcriptional activity of purine and pyrimidine synthesis could better explain ΔMIC for β-lactams and aminoglycosides, respectively. How is it possible that log[ATP]/growth rate can account for ΔMIC when only nitrogenous bases are provided in the growth medium but not when inhibitors of nucleotide synthesis are used? One possibility is that applying exogenous nitrogenous bases affects biochemical processes outside of bacterial metabolism and growth rate which could impact antibiotic tolerance. For example, supplementation of growth medium with uracil has been shown to impact the production of macromolecules required for biofilm formation(80) and the expression of genes involved in quorum sensing(81), both of which can influence antibiotic tolerance(82, 83), metabolism, and growth(84, 85). However, with a more targeted approach of gene deletion and the inclusion of chemicals that specifically target nucleotide synthesis, the relationship between ΔMIC and either purine or pyrimidine synthesis comes to light and effectively weakens the relationship between log[ATP]/growth rate and ΔMIC. Importantly, we found that the removal of pyrC could reduce ΔMIC for both carbenicillin and streptomycin (Fig. 5), which conflicts with our findings showing that pyrimidine transcriptional activity does not correlate to ΔMIC of carbenicillin (Fig. 7). One reason for this is that removal of pyrC would disrupt downstream regulatory steps in pyrimidine synthesis, and its crosstalk with purine synthesis, which could impact the relationship between transcriptional activity of both nucleotide synthesis pathways and ΔMIC. Interestingly, the opposing effects of purine and pyrimidine synthesis on ΔMIC of β-lactam and aminoglycoside are consistent with previous work showing the effect of these pathways on antibiotic lethality. Inhibition of purine synthesis through gene deletion or the addition of chemical inhibitors decreases lethality for β-lactams while increasing it for aminoglycosides. Similarly, increasing purine synthesis through exogenously provided activators potentiates lethality for β-lactams but protects bacteria against aminoglycosides. While increased bacterial metabolism owing to adenine limitation was suggested to account for these trends(45), our work suggests that pathway activity alone can account for IE. The reason(s) for this remains unclear. Altering activity of purine and pyrimidine synthesis may decouple cell growth and proliferation by altering nucleotide pool balance, which has been shown to impact sensitivity to chemotherapeutic drugs in cancer cells(86). It may also impact the creation of purine alarmones that have roles in regulating the activity of antibiotic targets, including ribosome(87) and cell wall synthesis(88), and can generally increase antibiotic tolerance(89, 90). Why altering purine and pyrimidine synthesis impacts IE in an antibiotic class-specific manner should be explored in the future.

Even though case reports have highlighted the increased morbidity and mortality that can occur owing to IE, we currently lack approaches to increase the sensitivity of high-density bacterial populations to antibiotics. While increasing the concentration of antibiotics administered has shown promise in eliminating high-density infections in animal models, such an approach in the clinic may lead to off-target toxicities, which can lead to patient harm(91). Accordingly, targeted approaches to treat IE in the clinic are required. Interestingly, our data has shown that removing pyrC, a gene that does not have a mammalian homolog(52), through genetic manipulation can reduce IE for both carbenicillin and streptomycin. pyrC encodes a dihydroorotase enzyme involved in the synthesis of pyrimidines in E. coli and additional pathogens. Recent work in Acinetobacter baumanii has demonstrated that dihydroorotases may represent a novel class of antibiotic targets(92). Specifically, two triazolopyrimidine/imidazopyrimidine analogs that were initially developed as anti-malarial compounds showed both in vitro and in vivo efficacy against A. baumanii. Similar compounds that target the pyrC protein may be able to reduce IE. This approach would not only increase the efficacy of existing β-lactams and aminoglycosides against high-density infections but could also reduce the evolution of antibiotic resistance, which has been associated with IE when high-density populations are treated with antibiotics at sub-MIC levels.

Materials and Methods

Experimental design, strains and growth conditions

We used E. coli strain BW25113 (F- Δ(araD-araB)567 ΔlacZ4787:rrnB-3 λ- rph-1 Δ(rhaD-rhaB)568 hsdR514) for the majority of our experiments. P. aeruginosa PA14 (BEI Resources) was used where indicated. E. coli knockout strains from the Keio collection (93) were acquired from Horizon Discovery (Boyertown, PA). Previously created reporter strains that contain plasmid-borne (SC101 origin of replication, kanamycin resistance)(94) copies of E. coli promoters driving the expression of gfpmut2 were acquired from Horizon Discovery. Plasmids from these strains were isolated using a ZymoPURE plasmid miniprep kit (Zymo Research Corporation, Irvine, CA) and transformed into BW25113 using a Mix and Go Transformation kit (Zymo Research) as per the manufacturer’s recommended protocol. Each experiment was initiated from an overnight culture, which was created from single colonies of bacteria grown on lysogeny broth (LB) agar medium (MP Biomedicals, Solon OH). Colonies were inoculated into 3 mL of LB liquid medium contained in 15 mL culture tubes (Genesee Scientific, Morrisville, NC). Cultures were shaken for 24 hours at 250 RPM and 37°C, which allows bacteria to achieve stationary phase (37). Experiments were performed in modified M9 medium [1X M9 salts (48 mM Na₂HPO₄, 22 mM KH₂PO₄, 862 mM NaCl, 19 mM NH₄Cl), 0.5% thiamine (Alfa Aesar, Ward Hill, MA), 2 mM MgSO₄, 0.1 mM CaCl₂, 0.04% glucose] with various percentages of casamino acids (0.01%, 0.05%, 0.1%, 0.5%, and 1%, as indicated). Where indicated, the growth medium was supplemented with adenine (Alfa Aesar), cytosine (Acros Organics, Geel, Belgium), guanine (Thermofisher, Waltham, MA), thymine (Thermofisher), or uracil (Sigma Aldrich, St. Louis, MO) at the concentration indicated. We were unable to dissolve guanine at a final concentration of 1 mM or greater in M9 medium under the growth conditions used throughout the study. Knockout strains were grown in M9 medium supplemented with 1% casamino acids and with equimolar concentrations of all five nitrogenous bases at concentrations of 1, 4, and 7 μM. 6-MP (Alfa Aesar) and IMP (Thermofisher) were supplemented where indicated at final concentrations of 0.05 μg/mL and 1 mM, respectively.

Flux balance analysis

Flux balance analysis was performed using the COBRA toolbox v.3.0 (95) coupled with the iML1515 genome-scale model of E. coli metabolism (96). This version of the model contains 1516 genes, 1877 metabolites, and 2712 reactions. All upper and lower bounds were kept the same as in the core model. However, the following changes were made to capture the composition of M9 medium. Lower and upper bound exchange values for ions found in M9 medium (K⁺, Mg²⁺, Na⁺, NH₄⁺, Cl⁻, P_i, SO₄²⁻, and Ca²⁺) were set at −1000 and 1000, respectively. The consumption rates of thiamine, glucose, oxygen, and nitrogenous bases were estimated from previously published literature. Parameter justification can be found in the Supplementary Results. Critical lower and upper bound values are in Supplementary Table 1. For simulations using OptKnock, we sequentially removed all 1516 genes in the model and considered only genes that increased ATPsyn/biomass above wildtype (where no genes are removed) in our analysis. To identify enzymes involved in nucleotide synthesis that increase ATPsyn/biomass, we considered only genes that increased ATPsyn/biomass and that were assigned to one of the following pathways as reported in Ecocyc: superpathway of histidine, purine, and pyrimidine biosynthesis, superpathway of purine nucleotides de novo biosynthesis II, superpathway of guanosine nucleotides de novo biosynthesis II, superpathway of pyrimidine deoxyribonucleotides de novo biosynthesis, pyrimidine deoxyribonucleotides de novo biosynthesis I, and pyrimidine deoxyribonucleotides de novo biosynthesis II. To report on flux through purine and pyrimidine synthesis, we simulated the flux activity of the PRAIS and DHORTS reactions, which match the reactions catalyzed by PurM and PyrC, respectively. To simulate the addition of 6-MP, the upper bound value of the GLUPRT, which captures the reaction catalyzed by PurM, was reduced. To simulate the effect of IMP addition, we assigned a lower bound value to the IMP exchange reaction (Ex_IMP_e). See the Supplementary Results for additional information on simulations performed with FBA.

Growth rate

To determine growth rate, overnight cultures were washed once in dH₂O. 1 μL of this culture was diluted into 200 μL of M9 medium contained in an untreated 96 well plate (Genesee Scientific). The medium was overlaid with 70 μL of mineral oil. The plate was placed in a prewarmed (37°C) microplate reader (Victor X4, Perkin Elmer, Waltham, MA) and optical density (OD) at 600 nm (OD₆₀₀) was recorded every 10 minutes for 10 hours. To determine growth rate, the OD₆₀₀ of cell-free medium was subtracted from all measurements. Maximum growth rate was determined by fitting growth curves to a logistic equation(97) in a custom MATLAB (R2022a, MathWorks Inc., Natick, MA) program as described previously(37). To achieve high-quality fits between the logistic function and growth curves of ΔpyrC and ΔpurK, we added an extra layer of smoothing by using a median filter with a window size of 2×2 to disregard the outliers in the data. In this case, median filtering works better than other filtering methods such as Gaussian filtering and average filtering as a median filter sorts out the neighboring values in the window matrix, chooses the median value, and eliminates the extreme values. Unlike moving average filtering this method preserves the original data better than other techniques as it uses an existing value in the window instead of numerical averages. Additional details about curve fitting are provided in the Supplementary Methods.

ATP measurements

ATP measurements occurred as previously described(37, 38). Briefly, an overnight culture was diluted 1/40 into 40 mL of M9 medium with 0.04% glucose but without casamino acids housed in a 50 mL conical tube (Genesee Scientific). The culture was placed in a prewarmed (37°C) shaker incubator and was subsequently shaken at 250 RPM for 2 hours. The bacteria were concentrated 2-fold in freshly diluted (3:1) M9 medium (containing both 0.04% glucose and casamino acids (0.01–1%)) using centrifugation. 100 μL was subsequently placed in the wells of an opaque walled 96 well microplate (Costar 3370, Corning, Kennebunk, ME), which was overlaid with two Breathe Easy sealing membranes (Sigma Aldrich). The microplate was shaken at 250 RPM at 37°C for 1 hour whereupon OD₆₀₀ reached ~ 0.1. The BacTiter-Glo assay (Promega, Madison, WI) was used to measure the concentration of ATP according to the manufacturer’s recommendations. ATP (luminescence) and OD₆₀₀ were measured in a Victor X4 microplate reader (Perkin Elmer). We used pure ATP (adenosine 5’-triphosphate disodium salt hydrate, Sigma Aldrich) to create a standard curve to quantify the concentration of ATP normalized by cell density (OD₆₀₀)(37). When computing [ATP]/growth rate, we log transformed ATP so that the order of magnitude was closer to that of growth rate (0.1–1/hr) and to match our previous calculation of growth productivity(37).

MIC assays

200 μL of M9 medium containing casamino acids, nitrogenous bases (where indicated), and antibiotics (where indicated) was placed in the wells of an untreated 96 well microplate (Genesee Scientific). An overnight culture was washed once in dH₂O and diluted either 1000-fold (high density) or 10000-fold (low density). 4 μL of these dilutions were inoculated in the growth medium in the 96 well microplate, which was subsequently overlaid with two Breathe Easy sealing membranes (Sigma). Cultures were shaken for 24 hours at 250 RPM and at 37°C whereupon the Breathe Easy sealing membranes were removed and cell density was measured using OD₆₀₀ in a Victor NIVO microplate reader (Perkin Elmer). OD₆₀₀ of cell-free medium was removed from all cell density values. Any condition where growth did not exceed 0.01 was set to zero, as performed previously (Supplementary Results). Concentration gradients used for each antibiotic are found in the Supplementary Methods. The general qualitative trends in OD₆₀₀ are consistent with colony-forming units, and 24 hours is sufficient for MIC assays to reach the steady state as previously shown(37).

Reporter assays

Reporter strains were grown overnight with 50 μg/mL of kanamycin. They were then washed twice in dH₂O and diluted 200-fold into M9 medium containing 0.1% CAA, which was contained in a black-walled 96 well plate (Corning 3340). The medium contained no nitrogenous bases, 5 mM adenine, or 10 mM cytosine. The plate was overlaid with two Breathe Easy sealing membranes (Sigma) and shaken at 37°C and 250RPM in a darkened incubator. GFP (excite: OD₄₈₅, emit: OD₅₃₅) and OD₆₀₀ were measured in a Victor X4 microplate reader (Perkin Elmer) after 18 hours and 24 hours of growth. GFP and OD₆₀₀ values from cell-free medium were subtracted from all measurements and GFP was normalized by OD₆₀₀. When performing regression analysis between ΔMIC and GFP/OD₆₀₀, we log transformed GFP/OD₆₀₀ so that it was the same order of magnitude as ΔMIC (1–25 μg/mL).

Statistical analysis

The statistical test performed is indicated in the text or figure legend. T-tests (unpaired, unequal variance) were performed in Microsoft Excel (Redmond, WA). Linear and Deming regressions were performed in GraphPad Prism (version 9.3.1, GraphPad, San Diego, CA). For Deming regressions, the average standard deviation in ΔMIC values was used for the error for the y-axis. The average standard deviation for log[ATP]/growth was used for the x-axis. Deming regressions do not report R² values. Weighted least squares (WLS) regression was performed in JMP Pro 16 (SAS Institute Inc., Cary, NC). We used the inverse variance of ΔMIC values for error on the y-axis. An expansion of errors procedure was used when combining errors from multiple measurements (e.g., the standard error of log[ATP]/growth accounts for both the standard error from log(ATP) and from growth rate). ShinyGo(98) (version 0.80) was used to determine fold enrichment and enrichment FDR. Genes whose removal increased ATPsyn/biomass above that of wildtype were examined for enrichment against all genes contained within the IML1515 model. We used an FDR cutoff of 0.05, genes were annotated using the Curated.EcoCyc database and the ecoli_eg_gene ensembl/STRING-db database (captures the E. coli K12 – MG1655 genome) was used.

Data and materials availability:

All data are available in the main text or the supplementary materials. Raw data and modeling code to be deposited in the Dryad Digital Repository upon acceptance of the manuscript.