Introduction

Bacterial infections are a major global cause of death, and this burden is intensified by the ability of bacterial pathogens to evade killing by antibiotics. Antibiotic resistance – the ability of a bacterium to proliferate in the presence of high concentrations of antibiotics – contributed to an estimated 4.95 million deaths in 20191. However, bacteria have evolved other strategies to evade antibiotic action that are thought to contribute to high rates of antibiotic treatment failure in the absence of resistance2,3,4,5,6. Antibiotic tolerance describes the ability of an entire bacterial population to survive longer exposures to lethal concentrations of bactericidal antibiotics than a susceptible strain, without exhibiting an elevated minimum inhibitory concentration (MIC)7. Tolerance is commonly quantified by the metric minimum duration for killing, or MDK; the MDK99, for example, is the antibiotic exposure time required to kill 99% of the population. In contrast, antibiotic persistence refers to tolerance that applies only to a small bacterial subpopulation; the majority of the population dies at the same rate as a susceptible strain, but a subpopulation emerges that dies at a slower rate8. In the case of both tolerance and persistence, incomplete killing of the bacterial population leads to persistent, relapsing, and recurrent infections. Collectively, resistance, tolerance, and persistence pose a formidable and multifaceted threat to our ability to successfully treat bacterial infections.

As scientists look for new antibiotic targets and treatment approaches, it becomes increasingly important to understand the molecular mechanisms underlying these three antibiotic evasion strategies. The mechanisms behind resistance are largely well understood and often involve the expression of one or more resistance proteins, such as an antibiotic-inactivating enzyme, a resistant form of a target protein, or an efflux pump9. However, the regulation of expression of these proteins can be complex and impacted by the growth and metabolism of the bacterium10. Growth and metabolism are also key to the molecular mechanisms behind tolerance and persistence. A common physiological basis that applies to most cases of genotypic and phenotypic tolerance is slow growth and reduced metabolism2. Most clinically relevant bactericidal antibiotics target active metabolic processes; therefore, slowing these processes down leads to slower bacterial killing11,12. In the case of phenotypic tolerance, stresses and nutrient limitations encountered in the host, for example, can limit the growth of the bacterium and induce tolerance13,14. Similarly, mutations that confer genotypic tolerance most commonly occur in metabolic pathway enzymes and reduce or modulate their activity2. Persistence can also be phenotypically induced or genotypic, and there is considerable overlap between the triggers and pathways between tolerance and persistence15,16,17,18.

Among the diverse strategies and mechanisms employed by pathogenic bacteria to evade antibiotic killing, modulation of purine metabolism has recently emerged as a common (and clinically relevant) contributor to tolerance, persistence and resistance19,20,21,22,23,24,25,26. Interestingly, purine synthesis has already been strongly implicated as a contributor to bacterial virulence (reviewed elsewhere27,28). Here, we review the evidence for a relationship between cellular purine levels and antibiotic efficacy, derived from both in vitro and in vivo data. While examples are provided from a range of pathogenic and non-pathogenic species, the vast majority of data available and referred to in this review are derived from the “model” organisms Escherichia coli and Staphylococcus aureus. We begin by providing a brief overview of purine synthesis before reviewing in detail the purine synthesis-associated mutations implicated in tolerance and persistence, and finally, the impact of purine metabolism on resistance.

The de novo and salvage purine synthesis pathways

Purines are heterocyclic aromatic compounds consisting of two nitrogen-containing rings. Purine nucleotides, such as AMP, ATP, GMP, and GTP, are made up of a purine base linked to a pentose sugar and one or more phosphate groups. These nucleotides play an essential role in nucleic acid synthesis, energy and amino acid metabolism, and cell signalling. Purine synthesis in bacteria occurs via two pathways: the de novo pathway and the salvage pathway (Fig. 1). The de novo pathway commences with the conversion of ribose-5-phosphate (R5P) and ATP into phosphoribosyl pyrophosphate (PRPP) and AMP, catalysed by the enzyme PRPP synthetase (Prs)29,30. PRPP is central to the de novo purine pathway, but it also feeds into the synthesis of histidine, tryptophan, NAD+, and pyrimidines. Estimates suggest that the purine and pyrimidine synthesis pathways each consume 30–40% of the cellular PRPP pool, while histidine and tryptophan synthesis each use 10–15% and only 1–2% feeds into NAD+ synthesis30,31. In the de novo purine pathway, PRPP is converted into inosine 5-monophosphate (IMP) via a 10-step/11-enzyme pathway, commencing with PurF27,32,33,34. IMP acts as a split point for the pathway; from this point, IMP can feed into the formation of either AMP/ADP/ATP or GMP/GDP/GTP27,32,33,34. Conversion of IMP into AMP is catalysed by PurA and PurB, while GMP is synthesised from IMP by GuaA and GuaB27,32,33,34. Purine nucleobases and nucleosides can also be recycled into nucleotides via the purine salvage pathway24,35. Adenine is recycled into AMP by the enzyme Apt, xanthine is recycled into XMP via the enzyme Xpt, and hypoxanthine and guanine are both recycled by HprT/Hpt into IMP and GMP, respectively24. Each one of these enzymes utilises PRPP as a substrate in the reaction, highlighting the centrality of PRPP in purine biosynthesis35.

Fig. 1: The de novo and salvage purine pathways in bacteria.
figure 1

The core de novo purine synthesis pathway, from ribose-5-phosphate to AMP and GMP via inosine monophosphate, is shown in detail with the enzyme(s) that catalyse each step shown above. The enzymes involved are largely common between the model organisms Escherichia coli (EC) and Staphylococcus aureus (SA), but superscript letters indicate additional enzymes that are species-specific. Steps in the purine salvage pathway that use PRPP are shaded in blue. Chemical structures are shown above the name for key precursors, intermediates, and products. Ribose-5-phosphate (R5P), phosphoribosyl pyrophosphate (PRPP), inosine monophosphate (IMP).

Full size image

Purine synthesis is transcriptionally controlled by the regulator PurR29,36,37. In many bacteria, including E. coli and S. aureus, PurR acts as a transcriptional repressor36,37; however, PurR acts as a transcriptional activator in some other species29. In E. coli, PurR binds to the promoters of the pur and related operons/genes (e.g., guaBA) in response to excess levels of purines in the cell; both guanine and hypoxanthine act as corepressors that bind to PurR and promote its DNA-binding activity38. PRPP also binds to PurR in low-GC Gram-positive species, but it blocks its binding to DNA, resulting in derepression of purine biosynthesis29. While it will not be reviewed in detail here, it should be noted that purine synthesis and its regulation are also intimately linked to the synthesis of purine-derived second messengers. In turn, these messengers play a role in antibiotic efficacy. Purine synthesis is both transcriptionally and post-translationally regulated by the “alarmones” guanosine tetra- and pentaphosphate (collectively known as [p]ppGpp)39,40. (p)ppGpp, the signalling molecule of the stringent response, binds directly to PurR in members of the Firmicutes and increases its affinity for promoter DNA39. In a range of bacterial species, (p)ppGpp also binds allosterically to many purine synthesis pathway enzymes and inhibits their activity, including PurF, GuaB, and HprT40. Partial activation of the stringent response (i.e., elevated [p]ppGpp) is known to occur in clinical isolates and contribute to antibiotic tolerance and resistance expression41,42. Finally, a second purine-derived messenger, cyclic-di-AMP (c-di-AMP), has been associated with vancomycin susceptibility in S. aureus, and its synthesis is interconnected with both purine and (p)ppGpp levels43 (see footnote of Table 1 for details of the cellular target of antibiotics referred to in this review).

Table 1 Purine biosynthesis pathway mutations/disruptions are associated with tolerance and/or increased antibiotic killing
Full size table

The complex role of purine metabolism in antibiotic tolerance and persistence

The relationship between cellular ATP, tolerance, and persistence

PRPP and purine nucleotides are essential to many aspects of bacterial metabolism, including the synthesis of DNA, RNA, amino acids, and NAD+. Additionally, they play a role in cell signalling and stress responses. However, ATP synthesis also has a central role in energy homeostasis. There is now considerable evidence to suggest that antibiotic lethality is linked to cellular ATP level and that bactericidal antibiotics induce cell death by accelerating respiration11,44,45. These studies have demonstrated an inverse correlation between cellular ATP level and bacterial survival in the presence of bactericidal antibiotics in diverse bacterial species. While antibiotic exposure appears to induce an initial increase in ATP level45, several studies have associated lower basal ATP levels with antibiotic tolerance and persistence. Zheng et al. evolved E. coli under ampicillin selection in vitro and found that a tolerant strain exhibited downregulation of pathways involved in aerobic respiration and lower intracellular ATP content compared with the ancestral strain46. This phenotype was due to the deletion of a sodium-proton antiporter, but how this deletion impacts ATP metabolism remains unclear. In both E. coli and S. aureus, inhibition of ATP synthesis using arsenate has been shown to induce persister formation and increase antibiotic survival47,48. Conversely, increasing the cellular ATP level by adding glucose to the growth medium led to increased killing of S. aureus by ciprofloxacin48. Both studies concluded that a decrease in ATP would decrease the activity of ATP-dependent antibiotic targets, leading to reduced bacterial killing. However, two studies have demonstrated that ATP depletion also promotes the formation of protein aggresomes in E. coli and Vibrio splendidus, and these aggresomes are an indicator of bacterial dormancy and tolerance49,50. While the correlation between low ATP and tolerance has been demonstrated in a number of different bacterial species and under different conditions, it should be noted that tolerance can be induced by changes in growth independent of ATP level51.

The emerging relationship between ATP and antibiotic killing has also been noted in more clinically relevant scenarios. Rowe et al. reported that S. aureus residing inside macrophages display antibiotic tolerance as a consequence of the action of reactive oxygen species (ROS), which damage TCA cycle enzymes, leading to reduced respiration and lower cellular ATP14. The ability of ROS to induce tolerance was also observed in vivo (although respiration and ATP were not measured). Finally, the clinical significance of basal ATP level has been investigated among clinical isolates of S. aureus20. Four sets of paired isolates associated with persistent or resolving bacteraemia were tested for ATP level and killing by vancomycin in vitro. The four isolates associated with persistent infections had consistently lower ATP levels and higher antibiotic survival than the isolates from resolving infections. Curiously, despite having lower ATP levels, the persistent isolates demonstrated faster growth rates and higher expression of three pur genes than their paired resolving isolates. As noted in the introduction and discussed more in the following sections, tolerance is generally associated with slow growth2.

Mutations in Prs (PRPP synthase) confer tolerance

ATP represents the end point of one branch of the de novo purine synthesis pathway that begins with the synthesis of PRPP. As PRPP is a central metabolite that feeds into multiple biosynthetic pathways, and antibiotic efficacy is inherently linked to bacterial metabolism, it is unsurprising that mutations in PRPP synthetase have been identified in tolerant strains of both E. coli and S. aureus, including a clinical isolate. In general, two types of tolerance have been reported: tolerance by lag, and tolerance by slow growth (longer doubling time in exponential phase)52. Mutations in PRPP synthetase, or Prs, appear to confer tolerance by extending the lag phase of the bacterium. In two seminal tolerance studies from the Balaban group52,53, in vitro evolution of E. coli in the presence of ampicillin led to tolerant strains bearing three different mutations in Prs (Table 1). These studies found that tolerance evolved prior to the development of resistance and any increase in MIC (which is in keeping with the definition of tolerance). All three Prs mutants exhibited lag phases that were considerably longer than the ancestral strain. Furthermore, the MDK99 value was reported for one mutant and found to be >10 times longer than the ancestral strain52. In vitro evolution of S. aureus under daptomycin selection also resulted in a tolerant mutant bearing a Prs mutation54; however, a mutation in another gene was also present in this strain, so causation between the Prs mutation and the tolerance phenotype cannot be made. The growth phenotype of this strain was also not investigated.

A recent study has begun to provide some much-needed insight into the molecular basis of tolerance conferred by mutations in Prs, as well as identifying its clinical relevance. COL is a clinical isolate and commonly used “model” strain of methicillin-resistant S. aureus (MRSA) that bears an E112K mutation in Prs (Table 1). Like other Prs mutants, COL exhibits an extended lag phase and tolerance to daptomycin and ciprofloxacin in vitro compared with other S. aureus strains19. The MDK99 values for COL were approximately two-fold higher than any other strain. The Prs mutation was found to account for the majority of the slow growth and tolerance phenotype of COL, as reversal of the mutation produced a strain with a lag time similar to that of the closely related, but non-tolerant, strain Newman. This strain also exhibited a rate of antibiotic killing that was significantly increased compared with wild-type COL. Conversely, introduction of the E112K mutation into Newman conferred a lag time that was significantly longer than that of both wild-type Newman and COL, and a degree of tolerance that was similar to, or even greater than, that of wild-type COL. Comparison of the in vitro activity of wild-type Prs and the E112K mutant revealed that the mutation greatly reduces the PRPP synthetase activity of the enzyme. This presumably limits the pool of PRPP available for downstream pathways. Interestingly, transcriptomic analysis of the two isogenic Prs strain pairs (Newman and COL) showed that the Prs mutation broadly leads to upregulation of most of the de novo purine synthesis pathway genes and downregulation of genes involved in the histidine, tryptophan, pyrimidine, and salvage purine pathways. While this study provides the first mechanistic insights into Prs-mediated tolerance, it remains unclear how and why decreased Prs activity leads to an extended lag phase and whether this translates into antibiotic tolerance and treatment failures in vivo. Furthermore, the clinical prevalence of Prs mutations requires investigation; the aforementioned study identified three novel Prs mutations among the genomes of a collection of clinical S. aureus isolates19.

Pur gene mutations/disruptions have varying effects on antibiotic killing

Following synthesis by Prs, PRPP is fluxed into a number of different pathways. PRPP that flows into the de novo purine pathway is converted into inosine monophosphate (IMP) – the branchpoint for ATP and GTP synthesis – via an 11-enzyme cascade (Fig. 1). Point mutations have not been reported in any of these 11 enzymes among in vitro-evolved tolerant strains and, until recently, only pur gene deletions/disruptions had been studied in terms of tolerance (see Table 1 and below for details). However, a recent screen of E. coli strains with single point mutations in metabolic genes identified two pur gene mutants that exhibit tolerance (Table 1)26. These mutations in PurA and PurM conferred a slow growth rate and subsequent tolerance to carbenicillin, but, intriguingly, they also conferred low-level resistance to carbenicillin (2-fold increase in MIC; see the “Other associations between purine synthesis and resistance” section for further discussion of these mutations in relation to resistance). Therefore, tolerance was detected by comparing the rate of killing of wild-type and mutant strains when exposed to carbenicillin at 2× their respective MICs. Metabolomic analysis of these mutants implied that the mutations negatively impact enzymatic activity, as ATP and GTP levels were reduced and the substrates for PurA and PurM, respectively, were increased. Finally, this study was expanded to a collection of E. coli clinical isolates. Screening for restricted growth in minimal medium led to the identification of a PurK point mutation in a urinary tract isolate. This strain was highly resistant to carbenicillin (presumed to be due to expression of a β-lactamase), so tolerance to carbenicillin could not be measured directly. However, it exhibited low levels of AMP, ADP, and ATP compared with a paired isolate, and a combination of carbenicillin and the β-lactamase inhibitor sulbactam was only able to kill this mutant when exogenous adenine was added. Therefore, the authors concluded that the PurK mutation and purine limitation led to carbenicillin tolerance26.

Based on the relationship between decreased Prs activity and tolerance, and the E. coli PurA, PurK, and PurM point mutants just described, we could reasonably conclude that reduced purine synthesis leads to tolerance (Fig. 2). However, results reported for pur gene deletions or disruptions are not always in agreement with this conclusion (Table 1). Using a combination of biochemical screening and machine learning, Yang et al. preliminarily identified purine synthesis as a contributor to antibiotic lethality55. They then experimentally characterised four E. coli pur gene knockout strains: purD, purE, purK, and purM. All four deletion strains exhibited tolerance to ampicillin and/or ciprofloxacin during exponential growth, while three of the mutant strains exhibited faster killing by gentamicin than the wild type, suggesting that the relationship between purine synthesis and tolerance may be antibiotic-specific. A purM-disrupted mutant of S. aureus also exhibited increased killing by gentamicin (and rifampicin) compared with the wild type during the stationary phase56. This PurM mutant was one of four pur gene mutants found to exhibit poor survival when exposed to rifampicin in a transposon mutagenesis library screen. The S. aureus PurB mutant was also characterised in more detail and, like the PurM mutant, exhibited no growth defect but was more rapidly killed by gentamicin and rifampicin than the wild type; however, killing was very slow overall due to the use of stationary phase cultures. The same authors later went on to characterise a PurN deletion mutant of S. aureus22. Interestingly, they observed no impact of the mutation on the killing of stationary phase cultures by ampicillin and levofloxacin, but growing cultures of the mutant were killed more rapidly than the wild type.

Fig. 2: Schematic summary of the impact of purine synthesis alterations on antibiotic tolerance, persistence, and resistance.
figure 2

Purine synthesis gene variations are shown either as gene deletions/disruptions (Δ) or point mutations (Mut), with the two italicised letters before the protein name indicating the bacterial species (Ec = E. coli, Sa = S. aureus, Et = Edwardsiella tarda, Vs = Vibrio splendidus, Se = Salmonella enterica, Ms = Mycobacterium smegmatis, Ng = Neisseria gonorrhoeae, Bs = Bacillus subtilis). Other abbreviations indicate resistant bacteria: methicillin-resistant S. aureus (MRSA), vancomycin-intermediate S. aureus (VISA), extended-spectrum beta-lactamase E. coli (ESBL-Ec), carbapenem-resistant Acinetobacter baumannii (CRAB), daptomycin non-susceptible S. aureus (SaDAPNS). Boxes indicate key purine metabolite/purine-derived messenger conditions that have been implicated in various aspects of antibiotic efficacy. Arrow colouring indicates the antibiotic efficacy state that the variation leads to. See Fig. 1 for details of which step in the purine synthesis pathway each enzyme catalyses.

Full size image

One pur gene has been investigated for its impact on antibiotic killing both in vitro and in vivo. A PurF deletion mutant of S. aureus has been extensively studied, beginning with the observation that it exhibits a slower growth rate than the wild type (as did purM and purN knockouts)20. Intriguingly, and seemingly incongruent with its slow growth rate, late-exponential phase cultures of the PurF mutant exhibited a greater degree of killing by vancomycin in vitro than the wild type (although the difference in survival between the mutant and wild type was small)20. Follow-up experiments revealed that the PurF mutant bound more vancomycin and was more rapidly lysed by the antibiotic than the wild type43. This increased interaction with vancomycin is proposed to be due to a lower level of cell wall teichoic acid in the mutant, as a result of lower c-di-AMP. The hypersensitivity of the PurF mutant to killing by vancomycin was also observed in vivo, whereby bacterial burden following vancomycin treatment was lower in mice infected with the PurF mutant than those infected with the wildtype20. Whether this relationship between PurF and antibiotic killing is specific to vancomycin remains to be tested.

Finally, the impact of the deletion of PurR, the repressor of the purine synthesis pathway, on vancomycin killing of S. aureus in vivo has recently been investigated21. The authors confirmed that disruption of purR leads to derepression of purine synthesis, evidenced by increased expression of purF and a higher intracellular level of the pathway intermediate AICAR (Fig. 1). In a mouse model of infective endocarditis, the purR mutant was highly tolerant to killing by vancomycin compared with the wild type. This in vivo effect of increased purine synthesis on vancomycin efficacy is opposite to, and therefore consistent with, the hypersensitivity of the purF deletion mutant20. Interestingly, a purR point mutation has been identified in a slow-growing and tolerant clinical isolate of S. aureus associated with persistent bacteraemia57. The expression of the purine synthesis pathway and the mechanistic basis of tolerance in this PurR mutant have yet to be investigated.

Overall, the collective data derived from pur gene mutants are complex and, at times, incongruent (Fig. 2). Since each of the anabolic enzymes discussed above is required for the de novo purine pathway (Fig. 1), we could reasonably expect that disruption of any one may lead to a reduction in cellular purine levels and increased tolerance to antibiotic killing11,26,47,48. However, the data described above suggest that the relationship between purine levels and antibiotic killing may be opposite between E. coli and S. aureus, as well as there being antibiotic- and growth phase-specific effects. Therefore, the overall impact of pur gene disruption on antibiotic killing requires further investigation with consistent methodology.

Purine metabolism and antibiotic resistance

Impact of purine synthesis on resistance in S. aureus

While tolerance and persistence are emerging as important contributors to clinical antibiotic treatment failures, antibiotic resistance remains a global and ever-increasing threat to effective antibiotic therapy. MRSA is a major contributor to the antibiotic resistance problem, and considerable effort has gone into understanding the processes behind resistance expression in this pathogen, with a view to potentially reversing it. β-lactam resistance expression in MRSA is typically low-level and heterogeneous; the population, as a whole, exhibits a relatively low MIC, but subpopulations of cells exhibit very high levels of resistance. However, some isolates (like COL) exhibit high-level, homogeneous resistance58. Purine synthesis has been strongly implicated in the expression of high-level β-lactam resistance in MRSA24,25,59. Sequencing of highly resistant colonies within a heterogeneous population revealed a suite of mutations among purine synthesis genes, including prs, guaA, guaB, and hprT59 (the reported prs mutation differs from that identified in COL). More recently, 11 pur gene MRSA knockouts were shown to exhibit a 2-fold higher oxacillin MIC than the wild type, while the PurR knockout (i.e., derepression of pur expression) displayed a 2-fold decrease in MIC24. Higher oxacillin MICs were also exhibited by two HprT point mutants and deletion mutants of putative transporters involved in the salvage pathway. Finally, it was demonstrated that exogenous guanosine and xanthosine can resensitise MRSA to β-lactams, while adenosine causes antagonism. Taken together, these studies suggest that reduced purine synthesis/uptake is necessary for high-level resistance. Mechanistically, while previous work suggested that point mutations in prs, guaA, and hprT supported high-level resistance by increasing the expression of PBP2A (the non-native penicillin-binding protein that confers methicillin resistance)59, a number of studies have now shown that changes in β-lactam resistance can occur independent of changes in PBP2A level24,25,60. Instead, it has been proposed that the purine-derived second messenger c-di-AMP is required for high-level resistance expression24,25. Elevated (p)ppGpp has also been implicated previously58,59. From a clinical perspective, four small colony variant MRSA isolates associated with osteomyelitis were recently found to carry mutations in hprT25. Characterisation of one of these strains revealed a slow growth phenotype, high-level resistance to oxacillin and elevated c-di-AMP, as well as decreased GMP (the product of HprT; Fig. 1).

In contrast to the clear and consistent relationship between reduced purine synthesis and high-level β-lactam resistance in S. aureus, the possible relationship between the vancomycin-intermediate S. aureus (VISA) phenotype and purine synthesis has been historically blurry. Early studies suggested that an upregulation of purine synthesis at the level of both transcription and translation was associated with a ≥4-fold higher MIC, and a PurR-inactivating mutation was identified in a strain with a high MIC61,62. However, a follow-up study revealed that PurR inactivation was insufficient to confer reduced susceptibility, and the addition of exogenous purines didn’t affect vancomycin killing of a susceptible strain63. More recently, a metabolomic comparison of a pair of vancomycin-sensitive and VISA isolates from the same patient supported the findings from the original study by identifying increased purine synthesis in the VISA isolates64. Furthermore, the addition of a purine synthesis inhibitor increased the rate of vancomycin killing of the VISA strain. Interestingly, increased purine synthesis has also been reported for daptomycin non-susceptible mutants of S. aureus65. Finally, a study of “slow VISA” strains (that exhibit slow growth and intermediate vancomycin MICs) derived from a heterogeneous VISA clinical isolate identified four Prs mutations among five isolates when compared with the parental strain66. However, these strains were not characterised further, and it is not known whether/how these mutations impact the activity of Prs and downstream purine synthesis. Overall, it appears that while reduced purine synthesis promotes the expression of high-level β-lactam resistance in S. aureus, the opposite may be true for vancomycin-intermediate resistance and daptomycin non-susceptibility.

Other associations between purine synthesis and resistance

As purine synthesis is a major component of central metabolism, it’s unsurprisingly that there are a number of other, diverse ways in which purine metabolism can impact resistance (and vice versa). This relationship has also led to studies focused on targeting purine synthesis as a means to resensitise resistant bacteria. In a recent screening of a library of E. coli mutants bearing point mutations in essential genes, mutations in purA, purD, purF, purL, and purM were frequently associated with low-level resistance to carbenicillin, but not gentamicin26. More detailed characterisation of some of these mutants revealed that they had reduced cellular levels of ATP and GTP, and that their resistance could be reversed by the addition of exogenous adenine. Intriguingly, two of these mutants studied in more detail were also found to exhibit slow growth and tolerance (as discussed above), suggesting some cross-activation of these two antibiotic evasion strategies. While tolerance mutations are known to precede and promote the development of resistance in vitro53,57, the mutations responsible are typically distinct2.

Other relationships between purine metabolism and resistance have been drawn from clinical isolates. A recent metabolomic comparison identified downregulation of purine synthesis and lower ATP, ADP, and AMP levels as characteristic features of carbapenem-resistant Acinetobacter baumannii (CRAB) compared with susceptible strains67. Interestingly, they showed that the addition of exogenous AMP or ATP could drastically increase the killing of multiple CRAB strains by meropenem, and this synergistic effect also applied in a mouse model of peritoneal infection. A study of ceftazidime-resistant Edwardsiella tarda also reported downregulation of purine synthesis in resistant vs susceptible strains68. In both cases, the downregulation in purine synthesis was concluded to be the result of lower activity of the pyruvate cycle, which fluxes to purine metabolism via glutamine. A study of aminoglycoside resistance in mycobacteria also observed decreased AMP, ADP, and GMP in a resistant mutant69, while downregulation of purine synthesis (specifically purK) has been associated with ciprofloxacin resistance in Neisseria gonorrhoeae70. Finally, glutamine has been shown to resensitise resistant E. coli to ampicillin by stimulating purine synthesis71. In contrast, enrichment of purine synthesis has been associated with resistance in two metabolomic and proteomic studies of resistant E. coli72,73. In general, multiple studies suggest that the addition of exogenous purines has the potential to resensitise a range of resistant pathogens, for example, by increasing the proton motive force required for uptake and/or stimulating metabolism via the TCA cycle24,26,67,69,71,74. A similar sensitisation has also been reported for tolerant and persister cells26,75,76. It should be noted, however, that the effect of adding purines is not always positive; addition of xanthine reduces killing of mycobacteria by gentamicin69, while addition of ATP, adenosine, or guanosine reduces trimethoprim activity against Bacillus subtilis77.

Conclusions

Bacterial purine metabolism is a complex and central metabolic pathway that contributes to the synthesis of DNA, RNA, amino acids, cofactors, and second messengers. It has previously been identified as an important component of bacterial virulence (reviewed elsewhere27,28), but there is now growing in vitro, in vivo and clinical evidence to suggest that it contributes to antibiotic efficacy. Figure 2 summarises the different relationships reported between purine levels and antibiotic evasion strategies. Broadly (and logically), we can conclude that disruption of purine synthesis leads to low ATP (and GTP), which confers antibiotic tolerance, while stimulation of purine synthesis and high ATP are associated with increased bacterial killing. These relationships are in agreement with the demonstrated correlations between growth rate, metabolism, and antibiotic killing rate11,12,44. There is considerable and diverse evidence to support this conclusion for E. coli, including from pur gene deletions and point mutants, a clinical PurK mutant, and chemical inhibition/stimulation of purine synthesis (Table 1 and Fig. 2)26,55. However, the situation for S. aureus is more complex. Data for four pur gene knockouts (which presumably disrupt the synthesis of ATP) show that these mutants are more readily killed by antibiotics than the wild type20,22.56. Consistent with these data, deletion of the transcriptional repressor, PurR, leads to increased purine synthesis and tolerance21. Interestingly, while the tolerance-conferring Prs mutation identified in S. aureus COL negatively impacts PRPP activity in vitro and, therefore, would be expected to reduce downstream purine synthesis, transcriptomic analysis of the mutant suggests that most genes in the core purine synthesis pathway are actually upregulated19. Therefore, metabolomic analysis of S. aureus pur gene and Prs mutants is required to elucidate whether the changes reported in antibiotic killing are due to increased or decreased purine synthesis. Alternatively, other cellular effects of the mutations could be responsible for the observed phenotypes, such as changing levels of purine-derived second messengers. S. aureus and E. coli utilise different purine-derived second messengers, with S. aureus producing c-di-AMP and E. coli synthesising cyclic-di-GMP (c-di-GMP)78. They also produce (p)ppGpp, which is linked to antibiotic efficacy41, but the downstream signalling pathways and effects of (p)ppGpp in these two species are quite different79. Ultimately, a more detailed molecular and cellular characterisation of purine synthesis mutants of different bacterial species is required to fully unravel the relationship between purine levels and antibiotic killing.

The relationship between purine synthesis and resistance is similarly complex and variable depending on the antibiotic and species. For example, decreased purine synthesis is associated with high expression of β-lactam resistance in MRSA24 and endogenous carbapenem resistance in E. coli26, but increased purine synthesis has been observed in VISA and ESBL-containing E. coli strains64,72. In cases where changes in purine levels have been noted between susceptible and resistant isolates, it is still unclear whether dysregulation of purine synthesis is a consequence or contributor to the resistance phenotype. Overall, it is clear that attempts to modulate purine levels (either up or down) as a means to resensitise tolerant, persistent, and/or resistant bacteria28, or as an anti-virulence strategy27, require careful consideration of the wider impacts. Purine analogues are frequently used in other fields of medicine, and there is growing interest in their use against bacteria27,80. The idea of metabolic stimulation is also gaining attention as a therapeutic strategy to combat antibiotic inefficacy (recently reviewed elsewhere81). Based on the studies reviewed here, we conclude that there are many outstanding questions regarding the multifaceted, pathogen-specific, and antibiotic-specific relationship(s) between purine metabolism and antibiotic efficacy, and further research is required before we can confidently target bacterial purine metabolism clinically.