Exploring Synergy between Classic Mutagens and Antibiotics To Examine Mechanisms of Synergy and Antibiotic Action

Abstract

We used classical mutagens in Gram-negative Escherichia coli to study synergies with different classes of antibiotics, test models of antibiotic mechanisms of action, and examine the basis of synergy. We used 4-nitroquinoline 1-oxide (4NQO), zebularine (ZEB), 5-azacytidine (5AZ), 2-aminopurine (2AP), and 5-bromodeoxyuridine (5BrdU) as mutagens (with bactericidal potency of 4NQO > ZEB > 5AZ > 2AP > 5BrdU) and vancomycin (VAN), ciprofloxacin (CPR), trimethoprim (TMP), gentamicin (GEN), tetracycline (TET), erythromycin (ERY), and chloramphenicol (CHL) as antibiotics. We detected the strongest synergies with 4NQO, an agent that oxidizes guanines and ultimately results in double-strand breaks when paired with the bactericidal antibiotics VAN, TMP, CPR, and GEN, but no synergies with the bacteriostatic antibiotics TET, ERY, and CHL. Each of the other mutagens displays synergies with the bactericidal antibiotics to various degrees that reflect their potencies, as well as with some of the other mutagens. The results support recent models showing that bactericidal antibiotics kill bacteria principally by ultimately generating more double-strand breaks than can be repaired. We discuss the synergies seen here and elsewhere as representing dose effects of not the proximal target damage but rather the ultimate resulting double-strand breaks. We also used the results of pairwise tests to place the classic mutagens into functional antibacterial categories within a previously defined drug interaction network.

INTRODUCTION

New strategies are needed to combat the rise of multidrug-resistant pathogens (1, 2). One avenue of research takes advantage of the synergy between antibiotics in combination (e.g., reference 3; see also a review in reference 4). Previously, we used the synergy between different antibiotics to potentiate the low concentration of vancomycin that is able to enter Gram-negative cells (5). The outer membrane of these cells normally acts as a barrier to vancomycin and many other drugs (6, 7). A more comprehensive understanding of the basis of synergy between certain pairwise combinations of antibiotics is important for developing this approach more thoroughly. Recently, Kohanski and coworkers (8) proposed that bactericidal antibiotics kill cells in part by generating hydroxyl radicals, causing DNA damage that leads to double-strand breaks (9). Recent works by Dwyer et al. (10) and Belenky et al. (11) strongly support this idea. In this study, we used a new strategy to examine both the mechanism of synergy and the mechanism of action of antibiotics by quantifying interactions between classic mutagens and different classes of commonly used antibiotics. In particular, we focused on mutagens that are strongly bactericidal via known mechanisms, generating double-strand breaks. The use of synergistic relationships allows us to look at the mechanism of antibiotic killing through a different lens and shows that strongly bactericidal mutagens, particularly 4-nitroquinoline 1-oxide (4NQO), are highly synergistic with bactericidal antibiotics. The results suggest that this synergy may be due to what would be equivalent to a dose effect of the ultimate lethal lesion resulting from both types of agents, namely, double-strand breaks. Thus, our study provides support for the idea that double-strand breaks play a significant role in the mechanism of killing of some antibiotics. Moreover, we can place the mutagens into the drug interaction network defined by Yeh and coworkers (3) based on their pairwise interactions with traditional antibiotics. The network approach aids in the understanding of the cellular targets of drugs and their mechanisms of action.

MATERIALS AND METHODS

Escherichia coli strains.

The RecA- and RecB-deficient E. coli strains used here are from the Keio collection, described in Baba et al. (12), made from the starting strain BW25113 (11). This starting strain (lacI^q rrnB_T14 ΔlacZ_WJ16 hsdR514 ΔaraBAD_AH33 ΔrhaBAD_LD78) was used as the wild type (WT) in the experiments reported here, unless otherwise stated. Each mutant carries a complete deletion of the respective gene, with a kan insert in place of the gene.

Medium.

The medium (12) used was LB (10 g of tryptone, 5 g of yeast extract, and 10 g of NaCl per liter).

Growth conditions.

Unless otherwise stated, all genetic methods are as described by Miller (13). Overnight cultures containing different concentrations of a given antibiotic were seeded with approximately 1 × 10³ cells by inoculating 2-ml cultures with 50 μl of a 10⁻⁴ dilution of a 4-h culture with a density of 2 × 10⁸ cells/ml . After 18 h of incubation at 37°C on a rotor at 50 rpm, the optical density at 600 nm (OD₆₀₀) was measured. Graphs of these data display the percent growth versus that in LB.

Determination of single-drug concentrations.

Overnight cultures containing a range of concentrations of a given antibiotic (usually starting from the reported MIC in 2-fold intervals) were prepared using the methods described above. Subinhibitory concentrations were typically chosen to be those that yielded 50 to 95% growth compared to that of cultures without any antibiotics.

Drug interaction assay.

Cell cultures were prepared with the same method as described above, using LB medium supplemented with no drugs, with each drug individually, and with the two drugs together at subinhibitory concentrations. Experiments using each set of single antibiotics and combinations were carried out at the same time. The experiments were done multiple times on different days. Bar graphs were used to compare the effects of the paired drugs with those of the corresponding single drugs at the same dose and with the control grown in LB only.

Classification of drug interactions.

We used the same classifications detailed in a previous work (3). Additivity was defined as W_xy = W_xW_y, where W_x is the proportion of growth relative to the control with no drug using drug X, W_y is the proportion of growth with drug Y, and W_xy is the proportion of growth with the drugs combined. For example, if drug X has a residual growth of 0.6 of the control with no drug, and drug Y has a residual growth of 0.7, the additive expectation of the two drugs together would be 0.42. There is a range around 0.42 that would still be considered additive, e.g., 0.43. We discuss below how to calculate this range. Anything above this range would be antagonistic, and anything below this range would be synergistic. More formally, deviation from additivity is defined by ε̃, which is calculated from the formulas below. When ε̃ falls within the range of −1 to −0.5, we classified it as synergistic; when ε̃ is between −0.5 and 0.5, we classified it as additive; and when ε̃ falls between 0.5 and 2, we classified it as antagonistic.

$${\stackrel{\sim }{W}}_{xy}=\min \hspace{0.17em}[W_x,W_y]\hspace{0.17em}\text{for}\hspace{0.17em}{W}_{xy}>W_x{W}_y,\hspace{0.17em}\text{and}\hspace{0.17em}\text{is}\hspace{0.17em}0\hspace{0.17em}\text{otherwise}$$

$$\text{if}\hspace{0.17em}{W}_{xy}>\min \hspace{0.17em}[W_x,W_y],\hspace{0.17em}\text{then}$$

$$\stackrel{\sim }{\epsilon }=\left[{}^{(W_{xy}\hspace{0.17em}-\hspace{0.17em}\min \hspace{0.17em}[W_x,\hspace{0.17em}{W}_y\left]\right)}{/}_{(1\hspace{0.17em}-\hspace{0.17em}\min \hspace{0.17em}[W_x,\hspace{0.17em}{W}_y\left]\right)}\right]+1$$

A pair was labeled inconclusive if the results from multiple experiments were inconsistent.

Antibiotics.

Tetracycline (TET), erythromycin (ERY), chloramphenicol (CHL), ciprofloxacin (CPR), gentamicin (GEN), trimethoprim (TMP), vancomycin (VAN), 4-nitroquinoline 1-oxide (4NQO), 2-aminopurine (2AP), 5-azacytidine (5AZ), and 5-bromodeoxyuridine (5BrdU) were purchased from Sigma (St. Louis, MO). Zebularine (ZEB) was a gift from Victor Marquez.

RESULTS

We hypothesize that if a principal mode of killing was indeed due to the generation of double-strand breaks, one would expect that synergy between bactericidal drugs of different functional classes can result from each drug contributing damage that results in double-strand breaks. We can test this by using compounds that cause double-strand breaks via known mechanisms and thus act as antimicrobials. These compounds, shown in Fig. 1 and 2, include commonly used mutagens. Compounds ii to v (defined below) are base analogs. Note that even though each of these mutagens has a primary target or effect that results in base mispairing, there are secondary effects that can play a role in their action, such as saturating repair systems (e.g., the mismatch repair system), inducing or partially inducing stress response systems, or altering the pools of deoxynucleoside triphosphates (dNTPs).

FIG 1.

Molecular structure of 4-nitroquinoline 1-oxide.

FIG 2.

Molecular structures of zebularine, 5-azacytidine, 2-aminopurine, and 5-bromodeoxyuridine.

Mutagen-antibiotic pairs.

We used the following mutagens: (i) 4-nitroquinoline 1-oxide (4NQO) (see Fig. 1), a compound that forms adducts to guanine that oxidize to 8-oxodeoxyguanosine (8-oxodG) and that also can be metabolized to generate anion radicals (13, 14); (ii) zebularine (ZEB) (Fig. 2), a cytidine analog lacking the amino group that when incorporated into DNA forms covalent complexes with deoxynucleotide methyltransferases (15) (however, its toxicity emanates from mismatch repair correction that leads to strand breaks [16]); (iii) 5-azacytidine (5AZ) (Fig. 2), another cytidine analog, which blocks replication after making complexes with deoxynucleotide methyltransferases, leading to double-strand breaks and cell death (17–20); (iv) 2-aminopurine (2AP) (Fig. 2), which is significantly less toxic than ZEB and 5AZ but ultimately leads to excessive mismatch repair excision and some resulting double-strand breaks (it is very toxic, however, in dam strain backgrounds [21]); and (v) 5-bromodeoxyuridine (5BrdU) (Fig. 2), which is barely toxic, even in recA strains (22), but is a potent base analog mutagen (e.g., 23) (the presumption is that a very small number of double-strand breaks are generated by 5BrdU).

Figure 3 shows the increased sensitivity of recA and recB strains to the first four of these compounds by sequential spotting, together with results for the bactericidal antibiotics used here. RecA is involved in the repair of both single- and double-strand breaks, while RecB is involved only in recombination and recombinational repair of double-strand breaks. Increased sensitivity of recB strains is an indication of a failure to repair double-strand breaks.

FIG 3.

Sequential spotting tests for sensitivities to different mutagens and antibiotics for wild type and recA and recB mutants. Serial dilutions of cells were printed onto plates containing each agent and incubated at 37°C for 18 h. The number of cells spotted ranged from 0.5 × 10⁶ to 0.5 × 10¹.

We examined pairwise combinations of the five mutagens with seven different antibiotics for synergistic effects on cellular growth inhibition, using assays we have described previously (8; see also Materials and Methods). Three of these antibiotics are bactericidal (VAN, CPR, and GEN), three are bacteriostatic (ERY, CHL, and TET), and one, TMP, is sometimes bacteriostatic and sometimes bactericidal. We aimed for concentrations that would allow partial growth, ideally 60% to 90% of the bacterial growth in no-drug environments. Table 1 shows the concentration ranges for each antibiotic and mutagen.

TABLE 1.

Mutagens and antibiotics used in the study, with dosage and primary targets

Mutagen or antibiotic	Abbreviation	Dose range (μg/ml)	Primary target(s)
4-Nitroquinoline 1-oxide	4NQO	0.38–1.47	Guanine residues
Zebularine	ZEB	5	Cytidine analog
2-Aminopurine	2AP	500–700	Adenine analog
5-Azacytidine	5AZ	20	Cytidine analog
5-Bromodeoxyuridine	5BrdU	150–300	Deoxyuridine analog
Chloramphenicol	CHL	0.5	Protein synthesis, 50S
Ciprofloxacin	CPR	0.012–0.013	DNA gyrase
Erythromycin	ERY	50–150	Protein synthesis, 50S
Gentamicin	GEN	0.4	DNA
Tetracycline	TET	0.2–0.25	Protein synthesis, 30S
Trimethoprim	TMP	0.115–0.15	Folic acid biosynthesis
Vancomycin	VAN	12.5–150	Cell wall synthesis

The antibiotics were chosen as representatives of the main groups of antibiotics, differentiated by their mechanism of action, used in two prior studies (3, 5). Figure 4 displays the results in the format employed by Yeh and coworkers (3). Here, percent residual growth versus growth in LB without antibiotic is plotted for each single antibiotic and for each pair of antibiotics. This study characterized the interactions as additive, suppressive, antagonistic, or synergistic (see Materials and Methods for a fuller explanation, e.g., synergistic effects are those that are significantly greater than simple additive effects). The background color of each graph in Fig. 4 designates the form of epistasis (see the legend to Fig. 4).

FIG 4.

Systematic measurements of pairwise interactions between all combinations of agents X and Y. Each bar graph represents one experiment performed on 1 day for that drug pair. Within each panel, the bars, from left to right, represent the median growth rates of four replicates for cultures with no drugs, drug X only, drug Y only, and the combination of the two drugs X and Y (as shown in inset). The growth rates are represented as the percentages of the no-drug control. The error bars represent the ranges of replicate measurements for each experiment. The background color of each graph designates the form of epistasis, in which red is strong synergy (ε̃_max < −0.5), pink is weak synergy (−0.5 < ε̃_max < −0.25), white is additive (−0.25 < ε̃_max < 0.5 and −0.5 < ε̃_min < 0.25), dark green is strong antagonistic buffering (0.5 < ε̃_min < 1.15), light green is weak antagonistic buffering (0.25 < ε̃_min < 0.5), blue is antagonistic suppression (ε̃_min > 1.15), and gray is an inconclusive result. See Table 1 for a list of antibiotics used, their abbreviations, and the range of concentrations tested for each antibiotic.

It is evident in Fig. 4 that the strongest bactericidal agent among the classical mutagens, 4NQO, separated the antibiotics based on their bactericidal properties. Namely, strong synergies are found with VAN, CPR, GEN, and TMP but not with CHL, ERY, and TET. In fact, various levels of suppression or antagonism are seen with the pairings of 4NQO and CHL, ERY, or TET. 4NQO causes DNA damage that leads to double-strand breaks (13, 14, 20; also see below); therefore, this result implies that VAN, CPR, GEN, and TMP lead to double-strand breaks that combine with the 4NQO-caused breaks to yield too large a load for the cellular repair systems (see Discussion). The mutagen ZEB is strongly synergistic with two of the bactericidal antibiotics, weakly synergistic with a third, and additive with a fourth (GEN), but it also shows weak synergy with CHL and ERY. An unexpected result is the strong synergy of ZEB with TET, an antibiotic that fails to show strong synergy with any of 21 antibiotics tested with it in pairs (3, 5). The three remaining mutagens, 5AZ, 2AP, and 5BrdU, all of which are weaker bactericidal agents than ZEB or 4NQO, gave mixed responses when paired with this set of antibiotics. With respect to displaying synergies against these antibiotics, they are clearly much weaker than 4NQO and ZEB.

Mutagen-mutagen pairs.

On the right side of Fig. 4, we can see the results of pairwise combinations of mutagens with each other. Interestingly, each agent, like other antibiotics, has a distinct pattern of interaction. Again, 4NQO has the most synergistic interactions in pairings with other mutagens. However, a striking exception is the antagonistic buffering exhibited by the 4NQO-5AZ pair. Given the strong synergies 4NQO displays with the bactericidal antibiotics and the other three mutagens, including 5BrdU (the weakest agent with regard to killing), the suppression seen with 5AZ is remarkable.

Classification of mutagens based on pairwise interactions.

We applied the methodology used by Yeh and coworkers (3) to place the mutagenic antibacterial agents within the context of groupings based on their mechanism of action, using the data from pairwise tests with representative antibiotics (Fig. 4). This resulted in the interaction network shown in Fig. 5. However, as one adds more data, this picture changes and comes into better focus, as shown in Fig. 6. Here, the expanded interaction network is shown using the results from this study, together with the data from the study by Yeh and coworkers (3) and our recent results with VAN (8). Note that ZEB and VAN fit perfectly into the same grouping and 4NQO fits into the aminoglycoside grouping. In Fig. 6, the red lines correspond only to strong synergies (ε̃_max < −0.5), but the green lines include the suppressive (ε̃_min > 1.15) and buffering (0.25 < ε̃_min < 1.15) interactions. Additive interactions (−0.25 < ε̃_max < 0.5 and −0.5 < ε̃_min < 0.25) were not included in this figure, as they represent no interaction between antibiotics.

FIG 5.

Classification of the antibiotic network into monochromatically interacting classes of drugs based on mechanisms of action. Red lines represent strong synergistic interactions between groups (ε̃_max < −0.5), green lines represent antagonistic buffering between groups (0.25 < ε̃_min < 1.15), and blue lines represent antagonistic suppression between groups (ε̃_min > 1.15).

FIG 6.

Expanded interaction network using the results in this study together with the data of Yeh and coworkers (3) and our recent results with VAN (5). Note that while 5AZ and 2AP still cluster by themselves in two separate groups and ZEB clusters with VAN, as seen in the previous interaction network (Fig. 5) with fewer data points, we now see 4NQO clustering with the aminoglycosides, and 5BrdU, which previously clustered with both the folic acid biosynthetic inhibitor group (TMP) and the DNA gyrase inhibitor group, now clusters with DNA gyrase inhibitors only (LOM, CPR, and NAL). The comparison of these two figures shows us both the power and the limitations of interactive networks, in terms of how data are needed to yield the most accurate picture of the mechanisms of action. AMK, amikacin; AMP, ampicillin; CLI, clindamycin; DOX, doxycycline hyclate; FOX, cefoxitin; FUS, fusidic acid; LOM, lomefloxacin; NAL, nalidixic acid; NIT, nitrofurantoin; PIP, piperacillin; SLF, sulfamonomethoxine; SPR, spiramycin; SPX, spectinomycin; TOB, tobramycin.

DISCUSSION

Yeh and coworkers (3) examined pairwise interactions of 21 antibiotics and generated a drug interaction network, classifying antibiotics based on whether they demonstrated synergy, antagonism, or suppression of other antibiotics (see reference 5). Synergy between drug pairs allows one to design multidrug therapies (e.g., for a review, see reference 4). What is the mechanism of synergy? An examination of the synergies displayed by 22 antibiotics (3, 5) reveals synergies between the vast majority of pairs of bactericidal antibiotics that are within the same class (e.g., aminoglycosides). These synergies likely are due to a straightforward dose effect (24). Thus, adding the doses of two different aminoglycosides, for example, is equivalent to doubling the dose of either drug alone. Sometimes, synergy can be explained by two drugs operating on different steps in the same pathway, such as trimethoprim and sulfanilamide (25). However, there are strong synergies evident among drug pairs involving antibiotics of different categories (3, 5). The majority of these involve antibiotics that are bactericidal (e.g., aminoglycosides, β-lactams, fluoroquinolones, and vancomycin). The classical explanation for the synergy between β-lactams and aminoglycosides is that the inhibition of cell wall synthesis by β-lactams increases the permeability and thus the efficacy of aminoglycosides (e.g., see reference 4). However, Kohanski and coworkers (8) showed that bactericidal antibiotics can also act via a pathway that generates hydroxyl radicals that lead to double-strand breaks (9), even though their initial (proximal) target may be different. More-recent results from Dwyer et al. (10) and Belenky et al. (11) provide extensive experimental data that support this and that corroborate the work presented here. We hypothesize that these interclass synergies (e.g., fluoroquinolone–β-lactam pairs; see reference 3) result from a dose effect of the ultimate double-strand breaks that finally exceed the repair capacity of the cell. To test this, we used antibacterial compounds that are highly mutagenic and thus are not used for clinical treatment. However, they cause lethality to various degrees by generating double-strand breaks via different known mechanisms (13–20). One can see the increased sensitivity of recA and recB strains to the most bactericidal of these mutagens (4NQO, 5AZ, and ZEB; see Fig. 3). The strongest lethal effects are exerted by 4NQO, which oxidizes DNA to 8-oxodG, causes the generation of anionic radicals, and leads to double-strand breaks (13, 14). If a buildup of double-strand breaks that exceeds the cellular repair capacity is the cause of synergy for bactericidal antibiotics, 4NQO should be synergistic with bactericidal drugs but not bacteriostatic antibiotics. This is exactly what we found, as shown in Fig. 3. On the other hand, 4NQO displays antagonistic buffering with CHL and displays strong suppression with TET.

A systems analysis of pairwise interactions placed 4NQO in the same group as aminoglycosides in the drug interaction network (Fig. 6). This is fascinating, because it shows that the bactericidal properties of a mutagenic agent allow us to classify it as we would any typical antimicrobial drug. At this point, it is worth noting that a number of compounds that are used as mutagens in basic research are also employed as chemotherapeutic agents. Thus, bleomycin and cisplatin are used as mutagens in studies in bacteria and higher cells (e.g., references 26 and 27) yet are also used as effective agents against testicular cancer (28, 29). Both 5AZ and ZEB, which were employed in this study (Fig. 4), are used in chemotherapy as demethylating agents to reactivate silenced tumor suppressor genes (30, 31). Interestingly, ZEB fits into the same drug class as VAN in the drug interaction network (Fig. 6), while 5AZ constitutes its own group at this stage. 2AP also defines a new group, but 5BrdU groups with the quinolones and fluoroquinolones.

With regard to drug interactions, ZEB shows strong suppression with two of the four bactericidal drugs, weak suppression with a third, and additivity with a fourth. However, what makes ZEB unique is that it shows not only weak synergy with ERY and CHL but strong suppression with TET. This is extraordinary, since none of the 21 antibiotics previously paired with TET showed strong suppression with TET. Future studies will need to unravel the mechanism of this interaction. Because ZEB toxicity is reduced in a mismatch repair-deficient strain (16), we tested whether strong synergies would also be reduced. This is indeed the case for the tested pair of ZEB and VAN, which displays a strong synergy in the wild-type strain that is completely eliminated in a mismatch repair-deficient background (data not shown). The weaker bactericidal agents show differing patterns with the antibiotics that allow us to place all of them in different groups (Fig. 6). 5BrdU, although a strong mutagen, is the weakest bactericidal mutagen of those used here, with concentrations of 1,000 μg/ml barely affecting viability (22). The fact that it shows even weak synergy with any of the drugs (VAN) is extraordinary.

We can also look at the effects of pairing the five mutagens with each other. These results (Fig. 4) again show that 4NQO is the strongest with respect to displaying synergy for bactericidal effects. Even the pairing of 4NQO and 5BrdU displays strong synergy, in accordance with the idea that the number of double-strand breaks resulting from 4NQO is sufficient to combine with the very small number of such breaks generated by 5BrdU to result in synergistic effects. An unexpected result is that the pairing of 5AZ with 4NQO results in weak antagonistic buffering. Future studies will be aimed at deciphering the nature of this effect.

The exact nature of the DNA damage from the generated hydroxyl radical pathway of conventional antibiotics (8, 9) that causes double-strand breaks is an interesting question. The results with 4NQO shown here (Fig. 4) might appear to support the idea that the main lesion responsible is 8-oxodG, and particularly from the oxidized precursor dGTP, with the combined action of the MutY and MutM proteins generating double-strand breaks (9). This is based in part on increased resistance to bactericidal antibiotics in strains overproducing the MutT protein, which hydrolyzes oxidized dGTP, and in strains lacking both MutY and MutM (9). However, the picture appears to be more complicated, as MutT-deficient strains are not more sensitive to antibiotics (our unpublished data), and even MutY-MutM double-deficient strains still show 99% killing by bactericidal antibiotics (compared with 99.9% killing of the wild type [9]). Additional experiments will need to clarify the exact cause of DNA strand breaks (32, 33).

Funding Statement

P.Y. was supported by the Hellman Fellows Fund and a UCLA Faculty Career Development Award. This work was partly supported by a Faculty Research Grant from the UCLA Academic Senate.